From aca5f9aeea8da8857235347ed1363ccda5460cbb Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Sun, 23 Apr 2023 21:48:25 +0100 Subject: [PATCH] [CostModel][X86] getMemoryOpCost - increase cost of sub-32-bit vector load/stores For 8-bit/16-bit vector loads/stores we scalarize and transfer to/from the vector unit, or use the (usually slow) PINSR/PEXTR instructions. Fixes #59867 --- llvm/lib/Target/X86/X86TargetTransformInfo.cpp | 4 + .../CostModel/X86/interleaved-load-i8-stride-3.ll | 8 +- .../CostModel/X86/interleaved-load-i8-stride-5.ll | 6 +- .../CostModel/X86/interleaved-load-i8-stride-7.ll | 6 +- .../CostModel/X86/interleaved-store-i8-stride-3.ll | 8 +- .../CostModel/X86/interleaved-store-i8-stride-5.ll | 8 +- .../CostModel/X86/interleaved-store-i8-stride-7.ll | 8 +- llvm/test/Analysis/CostModel/X86/load_store.ll | 4008 ++++++++++---------- llvm/test/Analysis/CostModel/X86/shuffle-load.ll | 12 +- llvm/test/Analysis/CostModel/X86/vector-extract.ll | 52 +- .../CostModel/X86/vector-insert-inseltpoison.ll | 92 +- .../Analysis/CostModel/X86/vector-insert-value.ll | 92 +- llvm/test/Analysis/CostModel/X86/vector-insert.ll | 92 +- .../test/Transforms/SLPVectorizer/X86/bool-mask.ll | 140 +- llvm/test/Transforms/SLPVectorizer/X86/pr59867.ll | 15 +- 15 files changed, 2246 insertions(+), 2305 deletions(-) diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp index 158023f..784c803 100644 --- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp @@ -4861,8 +4861,12 @@ InstructionCost X86TTIImpl::getMemoryOpCost(unsigned Opcode, Type *Src, // This isn't exactly right. We're using slow unaligned 32-byte accesses // as a proxy for a double-pumped AVX memory interface such as on // Sandybridge. + // Sub-32-bit loads/stores will be slower either with PINSR*/PEXTR* or + // will be scalarized. if (CurrOpSizeBytes == 32 && ST->isUnalignedMem32Slow()) Cost += 2; + else if (CurrOpSizeBytes < 4) + Cost += 2; else Cost += 1; diff --git a/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-3.ll b/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-3.ll index 92972c4..6f8c305 100644 --- a/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-3.ll +++ b/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-3.ll @@ -15,14 +15,14 @@ target triple = "x86_64-unknown-linux-gnu" define void @test() { ; SSE2-LABEL: 'test' ; SSE2: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; SSE2: LV: Found an estimated cost of 23 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; SSE2: LV: Found an estimated cost of 24 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 50 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 93 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 189 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 ; ; AVX1-LABEL: 'test' ; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX1: LV: Found an estimated cost of 15 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX1: LV: Found an estimated cost of 16 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 27 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 52 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 99 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 @@ -30,7 +30,7 @@ define void @test() { ; ; AVX2-LABEL: 'test' ; AVX2: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX2: LV: Found an estimated cost of 6 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX2: LV: Found an estimated cost of 7 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX2: LV: Found an estimated cost of 6 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX2: LV: Found an estimated cost of 9 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX2: LV: Found an estimated cost of 13 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 @@ -38,7 +38,7 @@ define void @test() { ; ; AVX512DQ-LABEL: 'test' ; AVX512DQ: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX512DQ: LV: Found an estimated cost of 6 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX512DQ: LV: Found an estimated cost of 7 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 6 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 9 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 14 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 diff --git a/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-5.ll b/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-5.ll index febfeea..53df030 100644 --- a/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-5.ll +++ b/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-5.ll @@ -15,14 +15,14 @@ target triple = "x86_64-unknown-linux-gnu" define void @test() { ; SSE2-LABEL: 'test' ; SSE2: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; SSE2: LV: Found an estimated cost of 37 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; SSE2: LV: Found an estimated cost of 38 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 75 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 155 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 315 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 ; ; AVX1-LABEL: 'test' ; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX1: LV: Found an estimated cost of 23 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX1: LV: Found an estimated cost of 24 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 45 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 83 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 165 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 @@ -38,7 +38,7 @@ define void @test() { ; ; AVX512DQ-LABEL: 'test' ; AVX512DQ: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX512DQ: LV: Found an estimated cost of 23 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX512DQ: LV: Found an estimated cost of 24 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 45 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 85 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 165 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 diff --git a/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-7.ll b/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-7.ll index e8db1b2..15f1e02 100644 --- a/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-7.ll +++ b/llvm/test/Analysis/CostModel/X86/interleaved-load-i8-stride-7.ll @@ -15,14 +15,14 @@ target triple = "x86_64-unknown-linux-gnu" define void @test() { ; SSE2-LABEL: 'test' ; SSE2: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; SSE2: LV: Found an estimated cost of 56 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; SSE2: LV: Found an estimated cost of 57 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 110 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 217 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; SSE2: LV: Found an estimated cost of 441 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 ; ; AVX1-LABEL: 'test' ; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX1: LV: Found an estimated cost of 33 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX1: LV: Found an estimated cost of 34 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 62 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 118 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX1: LV: Found an estimated cost of 231 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 @@ -38,7 +38,7 @@ define void @test() { ; ; AVX512DQ-LABEL: 'test' ; AVX512DQ: LV: Found an estimated cost of 1 for VF 1 For instruction: %v0 = load i8, ptr %in0, align 1 -; AVX512DQ: LV: Found an estimated cost of 33 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 +; AVX512DQ: LV: Found an estimated cost of 34 for VF 2 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 62 for VF 4 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 120 for VF 8 For instruction: %v0 = load i8, ptr %in0, align 1 ; AVX512DQ: LV: Found an estimated cost of 233 for VF 16 For instruction: %v0 = load i8, ptr %in0, align 1 diff --git a/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-3.ll b/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-3.ll index e729e62..2cfd81a 100644 --- a/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-3.ll +++ b/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-3.ll @@ -15,14 +15,14 @@ target triple = "x86_64-unknown-linux-gnu" define void @test() { ; SSE2-LABEL: 'test' ; SSE2: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v2, ptr %out2, align 1 -; SSE2: LV: Found an estimated cost of 25 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 +; SSE2: LV: Found an estimated cost of 26 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 ; SSE2: LV: Found an estimated cost of 52 for VF 4 For instruction: store i8 %v2, ptr %out2, align 1 ; SSE2: LV: Found an estimated cost of 101 for VF 8 For instruction: store i8 %v2, ptr %out2, align 1 ; SSE2: LV: Found an estimated cost of 204 for VF 16 For instruction: store i8 %v2, ptr %out2, align 1 ; ; AVX1-LABEL: 'test' ; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v2, ptr %out2, align 1 -; AVX1: LV: Found an estimated cost of 15 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 +; AVX1: LV: Found an estimated cost of 16 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX1: LV: Found an estimated cost of 27 for VF 4 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX1: LV: Found an estimated cost of 53 for VF 8 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX1: LV: Found an estimated cost of 100 for VF 16 For instruction: store i8 %v2, ptr %out2, align 1 @@ -30,7 +30,7 @@ define void @test() { ; ; AVX2-LABEL: 'test' ; AVX2: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v2, ptr %out2, align 1 -; AVX2: LV: Found an estimated cost of 7 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 +; AVX2: LV: Found an estimated cost of 8 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX2: LV: Found an estimated cost of 7 for VF 4 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX2: LV: Found an estimated cost of 9 for VF 8 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX2: LV: Found an estimated cost of 13 for VF 16 For instruction: store i8 %v2, ptr %out2, align 1 @@ -38,7 +38,7 @@ define void @test() { ; ; AVX512DQ-LABEL: 'test' ; AVX512DQ: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v2, ptr %out2, align 1 -; AVX512DQ: LV: Found an estimated cost of 7 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 +; AVX512DQ: LV: Found an estimated cost of 8 for VF 2 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX512DQ: LV: Found an estimated cost of 7 for VF 4 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX512DQ: LV: Found an estimated cost of 9 for VF 8 For instruction: store i8 %v2, ptr %out2, align 1 ; AVX512DQ: LV: Found an estimated cost of 14 for VF 16 For instruction: store i8 %v2, ptr %out2, align 1 diff --git a/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-5.ll b/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-5.ll index 2e50c41..eef594d 100644 --- a/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-5.ll +++ b/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-5.ll @@ -15,14 +15,14 @@ target triple = "x86_64-unknown-linux-gnu" define void @test() { ; SSE2-LABEL: 'test' ; SSE2: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v4, ptr %out4, align 1 -; SSE2: LV: Found an estimated cost of 43 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 +; SSE2: LV: Found an estimated cost of 44 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 ; SSE2: LV: Found an estimated cost of 87 for VF 4 For instruction: store i8 %v4, ptr %out4, align 1 ; SSE2: LV: Found an estimated cost of 178 for VF 8 For instruction: store i8 %v4, ptr %out4, align 1 ; SSE2: LV: Found an estimated cost of 360 for VF 16 For instruction: store i8 %v4, ptr %out4, align 1 ; ; AVX1-LABEL: 'test' ; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v4, ptr %out4, align 1 -; AVX1: LV: Found an estimated cost of 23 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 +; AVX1: LV: Found an estimated cost of 24 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX1: LV: Found an estimated cost of 46 for VF 4 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX1: LV: Found an estimated cost of 84 for VF 8 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX1: LV: Found an estimated cost of 166 for VF 16 For instruction: store i8 %v4, ptr %out4, align 1 @@ -30,7 +30,7 @@ define void @test() { ; ; AVX2-LABEL: 'test' ; AVX2: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v4, ptr %out4, align 1 -; AVX2: LV: Found an estimated cost of 23 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 +; AVX2: LV: Found an estimated cost of 24 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX2: LV: Found an estimated cost of 46 for VF 4 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX2: LV: Found an estimated cost of 84 for VF 8 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX2: LV: Found an estimated cost of 166 for VF 16 For instruction: store i8 %v4, ptr %out4, align 1 @@ -38,7 +38,7 @@ define void @test() { ; ; AVX512DQ-LABEL: 'test' ; AVX512DQ: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v4, ptr %out4, align 1 -; AVX512DQ: LV: Found an estimated cost of 23 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 +; AVX512DQ: LV: Found an estimated cost of 24 for VF 2 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX512DQ: LV: Found an estimated cost of 46 for VF 4 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX512DQ: LV: Found an estimated cost of 87 for VF 8 For instruction: store i8 %v4, ptr %out4, align 1 ; AVX512DQ: LV: Found an estimated cost of 166 for VF 16 For instruction: store i8 %v4, ptr %out4, align 1 diff --git a/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-7.ll b/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-7.ll index e4bbad1..5c7a6a8 100644 --- a/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-7.ll +++ b/llvm/test/Analysis/CostModel/X86/interleaved-store-i8-stride-7.ll @@ -15,14 +15,14 @@ target triple = "x86_64-unknown-linux-gnu" define void @test() { ; SSE2-LABEL: 'test' ; SSE2: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v6, ptr %out6, align 1 -; SSE2: LV: Found an estimated cost of 56 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 +; SSE2: LV: Found an estimated cost of 57 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 ; SSE2: LV: Found an estimated cost of 112 for VF 4 For instruction: store i8 %v6, ptr %out6, align 1 ; SSE2: LV: Found an estimated cost of 225 for VF 8 For instruction: store i8 %v6, ptr %out6, align 1 ; SSE2: LV: Found an estimated cost of 456 for VF 16 For instruction: store i8 %v6, ptr %out6, align 1 ; ; AVX1-LABEL: 'test' ; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v6, ptr %out6, align 1 -; AVX1: LV: Found an estimated cost of 33 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 +; AVX1: LV: Found an estimated cost of 34 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX1: LV: Found an estimated cost of 63 for VF 4 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX1: LV: Found an estimated cost of 119 for VF 8 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX1: LV: Found an estimated cost of 232 for VF 16 For instruction: store i8 %v6, ptr %out6, align 1 @@ -30,7 +30,7 @@ define void @test() { ; ; AVX2-LABEL: 'test' ; AVX2: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v6, ptr %out6, align 1 -; AVX2: LV: Found an estimated cost of 33 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 +; AVX2: LV: Found an estimated cost of 34 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX2: LV: Found an estimated cost of 63 for VF 4 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX2: LV: Found an estimated cost of 119 for VF 8 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX2: LV: Found an estimated cost of 232 for VF 16 For instruction: store i8 %v6, ptr %out6, align 1 @@ -38,7 +38,7 @@ define void @test() { ; ; AVX512DQ-LABEL: 'test' ; AVX512DQ: LV: Found an estimated cost of 1 for VF 1 For instruction: store i8 %v6, ptr %out6, align 1 -; AVX512DQ: LV: Found an estimated cost of 33 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 +; AVX512DQ: LV: Found an estimated cost of 34 for VF 2 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX512DQ: LV: Found an estimated cost of 63 for VF 4 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX512DQ: LV: Found an estimated cost of 121 for VF 8 For instruction: store i8 %v6, ptr %out6, align 1 ; AVX512DQ: LV: Found an estimated cost of 234 for VF 16 For instruction: store i8 %v6, ptr %out6, align 1 diff --git a/llvm/test/Analysis/CostModel/X86/load_store.ll b/llvm/test/Analysis/CostModel/X86/load_store.ll index 9183cb1..ef36bd7 100644 --- a/llvm/test/Analysis/CostModel/X86/load_store.ll +++ b/llvm/test/Analysis/CostModel/X86/load_store.ll @@ -30,10 +30,10 @@ define i32 @stores_align4(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <256 x i1> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <32 x i8> undef, ptr undef, align 4 @@ -81,10 +81,10 @@ define i32 @stores_align4(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <256 x i1> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <32 x i8> undef, ptr undef, align 4 @@ -132,10 +132,10 @@ define i32 @stores_align4(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <256 x i1> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <32 x i8> undef, ptr undef, align 4 @@ -183,11 +183,11 @@ define i32 @stores_align4(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <1 x i1> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <256 x i1> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <32 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i16> undef, ptr undef, align 4 @@ -324,151 +324,151 @@ define i32 @stores_partial_align4(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <9 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <10 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <12 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <23 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i16> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i16> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <3 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <5 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <7 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <11 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <12 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <13 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <13 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <14 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <15 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <19 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <23 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <29 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <31 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <35 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <37 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <39 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <43 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <43 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <45 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <45 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <46 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: store <47 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <51 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <55 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <55 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <59 x i8> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <12 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <14 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <15 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <19 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <21 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <22 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <23 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <43 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <45 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <46 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <47 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <49 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <50 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <51 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <52 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <53 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <54 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <55 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <56 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <59 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <61 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <62 x i1> undef, ptr undef, align 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <63 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <61 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: store <63 x i8> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <9 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <10 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <12 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <23 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <24 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <42 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <44 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <47 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <57 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <58 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <60 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i1> undef, ptr undef, align 4 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <63 x i1> undef, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE41-LABEL: 'stores_partial_align4' @@ -510,151 +510,151 @@ define i32 @stores_partial_align4(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <9 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <10 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <12 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <14 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i16> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i16> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <23 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <35 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <35 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <37 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <39 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <39 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <43 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <47 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <45 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <51 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <51 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <53 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <55 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <55 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i8> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i8> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <19 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <21 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <22 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <23 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <27 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <29 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <30 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <31 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <43 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <45 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <46 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <47 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <49 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <50 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <51 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <52 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <53 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <54 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <55 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <56 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <57 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <58 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <59 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <60 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <61 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <62 x i1> undef, ptr undef, align 4 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <63 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <23 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <24 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <25 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <26 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <28 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <31 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <47 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <58 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <60 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i1> undef, ptr undef, align 4 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <63 x i1> undef, ptr undef, align 4 ; SSE41-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'stores_partial_align4' @@ -696,151 +696,151 @@ define i32 @stores_partial_align4(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <23 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i16> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i16> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <19 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <21 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <22 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <23 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <35 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <31 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <37 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <38 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <39 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <37 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <43 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <47 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <47 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <51 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <55 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <54 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <55 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i8> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i8> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <19 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <21 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <22 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <23 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <43 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <45 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <46 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <47 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <51 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <53 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <54 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <55 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <59 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <61 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <62 x i1> undef, ptr undef, align 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <63 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <20 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <23 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <24 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <47 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <51 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <52 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <53 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <55 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <56 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <57 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <58 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <60 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i1> undef, ptr undef, align 4 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <63 x i1> undef, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'stores_partial_align4' @@ -882,151 +882,151 @@ define i32 @stores_partial_align4(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x float> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <19 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <25 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i16> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i16> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <19 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <21 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <22 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <23 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <31 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <33 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <34 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <35 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <36 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <37 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <38 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <39 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <43 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <47 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <45 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <51 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <49 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <50 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <51 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <52 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <54 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <55 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <53 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <54 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <55 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <58 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <59 x i8> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <63 x i8> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <19 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <21 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <22 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <23 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <43 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <45 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <46 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <47 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <59 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <61 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <62 x i1> undef, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <63 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 13 for instruction: store <63 x i8> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <20 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <23 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <24 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <35 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <36 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <37 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <38 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <39 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <40 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <42 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <44 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <47 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <48 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <49 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <50 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <52 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <55 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <56 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <57 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <58 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <60 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i1> undef, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i1> undef, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; store <1 x i64> undef, ptr undef, align 4 @@ -1276,10 +1276,10 @@ define i32 @stores_align1(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <256 x i1> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <32 x i8> undef, ptr undef, align 1 @@ -1327,10 +1327,10 @@ define i32 @stores_align1(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <256 x i1> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <32 x i8> undef, ptr undef, align 1 @@ -1378,10 +1378,10 @@ define i32 @stores_align1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <256 x i1> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <32 x i8> undef, ptr undef, align 1 @@ -1429,11 +1429,11 @@ define i32 @stores_align1(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <1 x i1> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <256 x i1> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <32 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i16> undef, ptr undef, align 1 @@ -1570,151 +1570,151 @@ define i32 @stores_partial_align1(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <9 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <10 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <12 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <23 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i16> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i16> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <3 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <5 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <7 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <11 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <12 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <13 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <13 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <14 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <15 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <19 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <23 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <29 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <31 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <35 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <37 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <39 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <43 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <43 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <45 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <45 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <46 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: store <47 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <51 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <55 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <55 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <59 x i8> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <12 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <14 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <15 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <19 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <21 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <22 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <23 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <43 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <45 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <46 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <47 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <49 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <50 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <51 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <52 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <53 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <54 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <55 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <56 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <59 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <61 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <62 x i1> undef, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <63 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <61 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: store <63 x i8> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <9 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <10 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <12 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <23 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <24 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <42 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <44 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <47 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <57 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <58 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <60 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i1> undef, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <63 x i1> undef, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE41-LABEL: 'stores_partial_align1' @@ -1756,151 +1756,151 @@ define i32 @stores_partial_align1(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <9 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <10 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <12 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <14 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i16> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i16> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <23 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <35 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <35 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <37 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <39 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <39 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <43 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <47 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <45 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <51 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <51 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <53 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <55 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <55 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i8> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i8> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <19 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <21 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <22 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <23 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <27 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <29 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <30 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <31 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <43 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <45 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <46 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <47 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <49 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <50 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <51 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <52 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <53 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <54 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <55 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <56 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <57 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <58 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <59 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <60 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <61 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <62 x i1> undef, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <63 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <23 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <24 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <25 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <26 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <28 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <31 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <47 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <58 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <60 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i1> undef, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <63 x i1> undef, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'stores_partial_align1' @@ -1942,151 +1942,151 @@ define i32 @stores_partial_align1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <23 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i16> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i16> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <19 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <21 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <22 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <23 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <35 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <31 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <37 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <38 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <39 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <37 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <43 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <47 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <47 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <51 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <55 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <54 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <55 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i8> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i8> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <19 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <21 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <22 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <23 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <43 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <45 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <46 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <47 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <51 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <53 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <54 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <55 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <59 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <61 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <62 x i1> undef, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <63 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <20 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <23 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <24 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <47 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <51 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <52 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <53 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <55 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <56 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <57 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <58 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <60 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i1> undef, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <63 x i1> undef, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'stores_partial_align1' @@ -2128,151 +2128,151 @@ define i32 @stores_partial_align1(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x float> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <19 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <25 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i16> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i16> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <19 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <21 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <22 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <23 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <31 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <33 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <34 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <35 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <36 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <37 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <38 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <39 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <43 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <47 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <45 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <51 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <49 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <50 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <51 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <52 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <54 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <55 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <53 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <54 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <55 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <58 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <59 x i8> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <63 x i8> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <19 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <21 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <22 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <23 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <43 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <45 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <46 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <47 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <59 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <61 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <62 x i1> undef, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <63 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 13 for instruction: store <63 x i8> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <20 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <23 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <24 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <35 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <36 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <37 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <38 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <39 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <40 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <42 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <44 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <47 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <48 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <49 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <50 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <52 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <55 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <56 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <57 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <58 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <60 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i1> undef, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i1> undef, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; store <1 x i64> undef, ptr undef, align 1 @@ -2522,10 +2522,10 @@ define i32 @stores_align64(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <256 x i1> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <32 x i8> undef, ptr undef, align 64 @@ -2573,10 +2573,10 @@ define i32 @stores_align64(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <256 x i1> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <32 x i8> undef, ptr undef, align 64 @@ -2624,10 +2624,10 @@ define i32 @stores_align64(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <256 x i1> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <32 x i8> undef, ptr undef, align 64 @@ -2675,11 +2675,11 @@ define i32 @stores_align64(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x float> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x double> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x ptr> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <16 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <8 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <4 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <1 x i1> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <256 x i1> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <32 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <16 x i16> undef, ptr undef, align 64 @@ -2816,151 +2816,151 @@ define i32 @stores_partial_align64(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <9 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <10 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <12 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <23 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i16> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i16> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <3 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <5 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <7 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <11 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <12 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <13 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <13 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <14 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <15 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <19 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <23 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <29 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <31 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <35 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <37 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <39 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <43 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <43 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <45 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <45 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <46 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: store <47 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <51 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <55 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <55 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <59 x i8> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <12 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <14 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <15 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <19 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <21 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <22 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <23 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <43 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <45 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <46 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <47 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <49 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <50 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <51 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <52 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <53 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <54 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <55 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <56 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <59 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <61 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <62 x i1> undef, ptr undef, align 64 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <63 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <61 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: store <63 x i8> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <9 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <10 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <12 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <23 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <24 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <42 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <44 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <47 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <57 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <58 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <60 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i1> undef, ptr undef, align 64 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <63 x i1> undef, ptr undef, align 64 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE41-LABEL: 'stores_partial_align64' @@ -3002,151 +3002,151 @@ define i32 @stores_partial_align64(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <9 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <10 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <11 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <12 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <13 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <14 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i16> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i16> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <23 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <27 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <35 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <35 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <37 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <39 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <39 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <43 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <47 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <45 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <51 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <51 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <53 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <55 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <55 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i8> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i8> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <19 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <21 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <22 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <23 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <25 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <26 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <27 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <29 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <30 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <31 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <43 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <45 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <46 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <47 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <49 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <50 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <51 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <52 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <53 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <54 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <55 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <56 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <57 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <58 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <59 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <60 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <61 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <62 x i1> undef, ptr undef, align 64 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <63 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <23 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <24 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <25 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <26 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <28 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <31 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <47 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <58 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <60 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i1> undef, ptr undef, align 64 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <63 x i1> undef, ptr undef, align 64 ; SSE41-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'stores_partial_align64' @@ -3188,151 +3188,151 @@ define i32 @stores_partial_align64(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <15 x float> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <17 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <18 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <19 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <20 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <21 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <22 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <23 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <24 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <28 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <29 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <30 x i16> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i16> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <19 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <21 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <22 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <23 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <35 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <31 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <37 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <38 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <39 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <37 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <38 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <43 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <47 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <47 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <51 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <55 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <54 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <55 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <59 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i8> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <61 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <62 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i8> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <19 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <21 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <22 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <23 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <33 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <34 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <35 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <36 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <37 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <38 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <39 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <40 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <41 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <42 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <43 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <44 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <45 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <46 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <47 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <48 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <49 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <50 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <51 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <52 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <53 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <54 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <55 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <56 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <57 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <58 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <59 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <60 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <61 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <62 x i1> undef, ptr undef, align 64 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <63 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <63 x i8> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <20 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <23 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <24 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <43 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <44 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <45 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <46 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <47 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <51 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <52 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <53 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <54 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <55 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <56 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <57 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <58 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <60 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i1> undef, ptr undef, align 64 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <63 x i1> undef, ptr undef, align 64 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'stores_partial_align64' @@ -3374,151 +3374,151 @@ define i32 @stores_partial_align64(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x float> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <3 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <7 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <11 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <15 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <19 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <23 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <25 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i16> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i16> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <31 x i16> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <1 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <2 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <3 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <2 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <3 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <4 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <5 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <6 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <7 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <5 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <6 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <7 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <8 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <9 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <10 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <11 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <15 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <17 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <18 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <19 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <13 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <14 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <15 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <19 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <20 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <21 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <22 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <23 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <21 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <22 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <23 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <27 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <25 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <26 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <27 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <29 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <30 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <31 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <35 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <31 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <33 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <34 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <35 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <36 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <37 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <38 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <39 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <37 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <38 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <39 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <41 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <42 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <43 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <47 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <45 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <46 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <47 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <49 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <50 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <51 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <49 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <50 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <51 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: store <52 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <53 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <54 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <55 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <53 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <54 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: store <55 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <59 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <57 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <58 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <59 x i8> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <61 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: store <62 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: store <63 x i8> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <3 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <5 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <6 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: store <7 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <9 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <10 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <11 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <12 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <13 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <14 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <15 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <17 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <18 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <19 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <20 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <21 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <22 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <23 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <24 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <25 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <26 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <27 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <28 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <29 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <30 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <31 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <33 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <34 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <35 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <36 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <37 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <38 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <39 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <40 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <41 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <42 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <43 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <44 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <45 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <46 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <47 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: store <48 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <49 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <50 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <51 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <52 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <53 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <54 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <55 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <56 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <57 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <58 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <59 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <60 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <61 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <62 x i1> undef, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <63 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 13 for instruction: store <63 x i8> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <3 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <5 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <6 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: store <7 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <9 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <10 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <11 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <12 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <13 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <14 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <15 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <17 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <18 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <19 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <20 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <21 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <22 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <23 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: store <24 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <25 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <26 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <27 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <28 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <29 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <30 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: store <31 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <33 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <34 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <35 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <36 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <37 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <38 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <39 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <40 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <41 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <42 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <43 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <44 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <45 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <46 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <47 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: store <48 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <49 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <50 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <51 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <52 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <53 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <54 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <55 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: store <56 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <57 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <58 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <59 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <60 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <61 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <62 x i1> undef, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: store <63 x i1> undef, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; store <1 x i64> undef, ptr undef, align 64 @@ -3921,11 +3921,11 @@ define i32 @loads_align4(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = load <4 x float>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = load <2 x double>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = load <2 x ptr>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = load <16 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %22 = load <8 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %23 = load <4 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %24 = load <2 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %25 = load <1 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = load <16 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = load <8 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = load <4 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = load <2 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %25 = load <1 x i1>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %26 = load <256 x i1>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %27 = load <32 x i8>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %28 = load <16 x i16>, ptr undef, align 4 @@ -4708,17 +4708,17 @@ define i32 @loads_partial_align4(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %124 = load <61 x i8>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %125 = load <62 x i8>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %126 = load <63 x i8>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %127 = load <3 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %128 = load <5 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %129 = load <6 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %130 = load <7 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %131 = load <9 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %132 = load <10 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %133 = load <11 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %134 = load <12 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %135 = load <13 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %136 = load <14 x i1>, ptr undef, align 4 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %137 = load <15 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %127 = load <3 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %128 = load <5 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %129 = load <6 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %130 = load <7 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %131 = load <9 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %132 = load <10 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %133 = load <11 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %134 = load <12 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %135 = load <13 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %136 = load <14 x i1>, ptr undef, align 4 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %137 = load <15 x i1>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %138 = load <17 x i1>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %139 = load <18 x i1>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %140 = load <19 x i1>, ptr undef, align 4 @@ -5014,10 +5014,10 @@ define i32 @loads_align1(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = load <4 x float>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = load <2 x double>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = load <2 x ptr>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = load <16 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %22 = load <8 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %23 = load <4 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %24 = load <2 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = load <16 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = load <8 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = load <4 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = load <2 x i1>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %25 = load <1 x i1>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %26 = load <256 x i1>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %27 = load <32 x i8>, ptr undef, align 1 @@ -5065,10 +5065,10 @@ define i32 @loads_align1(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = load <4 x float>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = load <2 x double>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = load <2 x ptr>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = load <16 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %22 = load <8 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %23 = load <4 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %24 = load <2 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = load <16 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = load <8 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = load <4 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = load <2 x i1>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %25 = load <1 x i1>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %26 = load <256 x i1>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %27 = load <32 x i8>, ptr undef, align 1 @@ -5116,10 +5116,10 @@ define i32 @loads_align1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = load <4 x float>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = load <2 x double>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = load <2 x ptr>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = load <16 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %22 = load <8 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %23 = load <4 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %24 = load <2 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = load <16 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = load <8 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = load <4 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = load <2 x i1>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %25 = load <1 x i1>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %26 = load <256 x i1>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %27 = load <32 x i8>, ptr undef, align 1 @@ -5167,11 +5167,11 @@ define i32 @loads_align1(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = load <4 x float>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = load <2 x double>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = load <2 x ptr>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = load <16 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %22 = load <8 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %23 = load <4 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %24 = load <2 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %25 = load <1 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = load <16 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = load <8 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = load <4 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = load <2 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %25 = load <1 x i1>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %26 = load <256 x i1>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %27 = load <32 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %28 = load <16 x i16>, ptr undef, align 1 @@ -5308,151 +5308,151 @@ define i32 @loads_partial_align1(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %36 = load <15 x float>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %37 = load <1 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %38 = load <2 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %39 = load <3 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %39 = load <3 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %40 = load <4 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %41 = load <5 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %41 = load <5 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %42 = load <6 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %43 = load <7 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %44 = load <9 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %43 = load <7 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %44 = load <9 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %45 = load <10 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %46 = load <11 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %46 = load <11 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %47 = load <12 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %48 = load <13 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %48 = load <13 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %49 = load <14 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %50 = load <15 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %51 = load <17 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %50 = load <15 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %51 = load <17 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %52 = load <18 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %53 = load <19 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %53 = load <19 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %54 = load <20 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %55 = load <21 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %55 = load <21 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %56 = load <22 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %57 = load <23 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %57 = load <23 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %58 = load <24 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %59 = load <25 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %59 = load <25 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %60 = load <26 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %61 = load <27 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %61 = load <27 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %62 = load <28 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %63 = load <29 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %63 = load <29 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %64 = load <30 x i16>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %65 = load <31 x i16>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %65 = load <31 x i16>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %66 = load <1 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %67 = load <2 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %68 = load <3 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %67 = load <2 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %68 = load <3 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %69 = load <4 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %70 = load <5 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %71 = load <6 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %72 = load <7 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %70 = load <5 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %71 = load <6 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %72 = load <7 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %73 = load <8 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %74 = load <9 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %75 = load <10 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %76 = load <11 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %74 = load <9 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %75 = load <10 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %76 = load <11 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %77 = load <12 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %78 = load <13 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %79 = load <14 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %80 = load <15 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %81 = load <17 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %82 = load <18 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %83 = load <19 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %78 = load <13 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %79 = load <14 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %80 = load <15 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %81 = load <17 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %82 = load <18 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %83 = load <19 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %84 = load <20 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %85 = load <21 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %86 = load <22 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %87 = load <23 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %85 = load <21 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %86 = load <22 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %87 = load <23 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %88 = load <24 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %89 = load <25 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %90 = load <26 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %91 = load <27 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %89 = load <25 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %90 = load <26 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %91 = load <27 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %92 = load <28 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %93 = load <29 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %94 = load <30 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %95 = load <31 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %96 = load <33 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %97 = load <34 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %98 = load <35 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %93 = load <29 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %94 = load <30 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %95 = load <31 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %96 = load <33 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %97 = load <34 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %98 = load <35 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %99 = load <36 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %100 = load <37 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %101 = load <38 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %102 = load <39 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %100 = load <37 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %101 = load <38 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %102 = load <39 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %103 = load <40 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %104 = load <41 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %105 = load <42 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %106 = load <43 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %104 = load <41 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %105 = load <42 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %106 = load <43 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %107 = load <44 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %108 = load <45 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %109 = load <46 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %110 = load <47 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %108 = load <45 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %109 = load <46 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %110 = load <47 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %111 = load <48 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %112 = load <49 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %113 = load <50 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %114 = load <51 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %112 = load <49 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %113 = load <50 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %114 = load <51 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %115 = load <52 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %116 = load <53 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %117 = load <54 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %118 = load <55 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %116 = load <53 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %117 = load <54 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %118 = load <55 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %119 = load <56 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %120 = load <57 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %121 = load <58 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %122 = load <59 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %120 = load <57 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %121 = load <58 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %122 = load <59 x i8>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %123 = load <60 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %124 = load <61 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %125 = load <62 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %126 = load <63 x i8>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %127 = load <3 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %128 = load <5 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %129 = load <6 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %130 = load <7 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %131 = load <9 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %132 = load <10 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %133 = load <11 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %134 = load <12 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %135 = load <13 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %136 = load <14 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %137 = load <15 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %138 = load <17 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %139 = load <18 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %140 = load <19 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %141 = load <20 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %142 = load <21 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %143 = load <22 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %144 = load <23 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %145 = load <24 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %146 = load <25 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %147 = load <26 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %148 = load <27 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %149 = load <28 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %150 = load <29 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %151 = load <30 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %152 = load <31 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %153 = load <33 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %154 = load <34 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %155 = load <35 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %156 = load <36 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %157 = load <37 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %158 = load <38 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %159 = load <39 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %160 = load <40 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %161 = load <41 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %162 = load <42 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %163 = load <43 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %164 = load <44 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %165 = load <45 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %166 = load <46 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %167 = load <47 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %168 = load <48 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %169 = load <49 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %170 = load <50 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %171 = load <51 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %172 = load <52 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %173 = load <53 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %174 = load <54 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %175 = load <55 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %176 = load <56 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %177 = load <57 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %178 = load <58 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %179 = load <59 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %180 = load <60 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %181 = load <61 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %182 = load <62 x i1>, ptr undef, align 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %183 = load <63 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %124 = load <61 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %125 = load <62 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %126 = load <63 x i8>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %127 = load <3 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %128 = load <5 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %129 = load <6 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %130 = load <7 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %131 = load <9 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %132 = load <10 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %133 = load <11 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %134 = load <12 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %135 = load <13 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %136 = load <14 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %137 = load <15 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %138 = load <17 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %139 = load <18 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %140 = load <19 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %141 = load <20 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %142 = load <21 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %143 = load <22 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %144 = load <23 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %145 = load <24 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %146 = load <25 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %147 = load <26 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %148 = load <27 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %149 = load <28 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %150 = load <29 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %151 = load <30 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %152 = load <31 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %153 = load <33 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %154 = load <34 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %155 = load <35 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %156 = load <36 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %157 = load <37 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %158 = load <38 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %159 = load <39 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %160 = load <40 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %161 = load <41 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %162 = load <42 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %163 = load <43 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %164 = load <44 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %165 = load <45 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %166 = load <46 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %167 = load <47 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %168 = load <48 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %169 = load <49 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %170 = load <50 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %171 = load <51 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %172 = load <52 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %173 = load <53 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %174 = load <54 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %175 = load <55 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %176 = load <56 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %177 = load <57 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %178 = load <58 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %179 = load <59 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %180 = load <60 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %181 = load <61 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %182 = load <62 x i1>, ptr undef, align 1 +; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %183 = load <63 x i1>, ptr undef, align 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE41-LABEL: 'loads_partial_align1' @@ -5494,151 +5494,151 @@ define i32 @loads_partial_align1(i32 %arg) { ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %36 = load <15 x float>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %37 = load <1 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %38 = load <2 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %39 = load <3 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %39 = load <3 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %40 = load <4 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %41 = load <5 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %41 = load <5 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %42 = load <6 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %43 = load <7 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %44 = load <9 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %43 = load <7 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %44 = load <9 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %45 = load <10 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %46 = load <11 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %46 = load <11 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %47 = load <12 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %48 = load <13 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %48 = load <13 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %49 = load <14 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %50 = load <15 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %51 = load <17 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %50 = load <15 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %51 = load <17 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %52 = load <18 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %53 = load <19 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %53 = load <19 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %54 = load <20 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %55 = load <21 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %55 = load <21 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %56 = load <22 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %57 = load <23 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %57 = load <23 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %58 = load <24 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %59 = load <25 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %59 = load <25 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %60 = load <26 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %61 = load <27 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %61 = load <27 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %62 = load <28 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %63 = load <29 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %63 = load <29 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %64 = load <30 x i16>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %65 = load <31 x i16>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %65 = load <31 x i16>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %66 = load <1 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %67 = load <2 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %68 = load <3 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %67 = load <2 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %68 = load <3 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %69 = load <4 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %70 = load <5 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %71 = load <6 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %72 = load <7 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %70 = load <5 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %71 = load <6 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %72 = load <7 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %73 = load <8 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %74 = load <9 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %75 = load <10 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %76 = load <11 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %74 = load <9 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %75 = load <10 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %76 = load <11 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %77 = load <12 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %78 = load <13 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %79 = load <14 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %80 = load <15 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %81 = load <17 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %82 = load <18 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %83 = load <19 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %78 = load <13 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %79 = load <14 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %80 = load <15 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %81 = load <17 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %82 = load <18 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %83 = load <19 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %84 = load <20 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %85 = load <21 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %86 = load <22 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %87 = load <23 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %85 = load <21 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %86 = load <22 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %87 = load <23 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %88 = load <24 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %89 = load <25 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %90 = load <26 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %91 = load <27 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %89 = load <25 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %90 = load <26 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %91 = load <27 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %92 = load <28 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %93 = load <29 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %94 = load <30 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %95 = load <31 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %96 = load <33 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %97 = load <34 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %98 = load <35 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %93 = load <29 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %94 = load <30 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %95 = load <31 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %96 = load <33 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %97 = load <34 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %98 = load <35 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %99 = load <36 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %100 = load <37 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %101 = load <38 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %102 = load <39 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %100 = load <37 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %101 = load <38 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %102 = load <39 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %103 = load <40 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %104 = load <41 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %105 = load <42 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %106 = load <43 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %104 = load <41 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %105 = load <42 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %106 = load <43 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %107 = load <44 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %108 = load <45 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %109 = load <46 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %110 = load <47 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %108 = load <45 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %109 = load <46 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %110 = load <47 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %111 = load <48 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %112 = load <49 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %113 = load <50 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %114 = load <51 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %112 = load <49 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %113 = load <50 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %114 = load <51 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %115 = load <52 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %116 = load <53 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %117 = load <54 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %118 = load <55 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %116 = load <53 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %117 = load <54 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %118 = load <55 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %119 = load <56 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %120 = load <57 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %121 = load <58 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %122 = load <59 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %120 = load <57 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %121 = load <58 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %122 = load <59 x i8>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %123 = load <60 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %124 = load <61 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %125 = load <62 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %126 = load <63 x i8>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %127 = load <3 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %128 = load <5 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %129 = load <6 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %130 = load <7 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %131 = load <9 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %132 = load <10 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %133 = load <11 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %134 = load <12 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %135 = load <13 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %136 = load <14 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %137 = load <15 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %138 = load <17 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %139 = load <18 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %140 = load <19 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %141 = load <20 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %142 = load <21 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %143 = load <22 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %144 = load <23 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %145 = load <24 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %146 = load <25 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %147 = load <26 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %148 = load <27 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %149 = load <28 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %150 = load <29 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %151 = load <30 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %152 = load <31 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %153 = load <33 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %154 = load <34 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %155 = load <35 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %156 = load <36 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %157 = load <37 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %158 = load <38 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %159 = load <39 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %160 = load <40 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %161 = load <41 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %162 = load <42 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %163 = load <43 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %164 = load <44 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %165 = load <45 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %166 = load <46 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %167 = load <47 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %168 = load <48 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %169 = load <49 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %170 = load <50 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %171 = load <51 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %172 = load <52 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %173 = load <53 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %174 = load <54 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %175 = load <55 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %176 = load <56 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %177 = load <57 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %178 = load <58 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %179 = load <59 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %180 = load <60 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %181 = load <61 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %182 = load <62 x i1>, ptr undef, align 1 -; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %183 = load <63 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %124 = load <61 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %125 = load <62 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %126 = load <63 x i8>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %127 = load <3 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %128 = load <5 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %129 = load <6 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %130 = load <7 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %131 = load <9 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %132 = load <10 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %133 = load <11 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %134 = load <12 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %135 = load <13 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %136 = load <14 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %137 = load <15 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %138 = load <17 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %139 = load <18 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %140 = load <19 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %141 = load <20 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %142 = load <21 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %143 = load <22 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %144 = load <23 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %145 = load <24 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %146 = load <25 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %147 = load <26 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %148 = load <27 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %149 = load <28 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %150 = load <29 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %151 = load <30 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %152 = load <31 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %153 = load <33 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %154 = load <34 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %155 = load <35 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %156 = load <36 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %157 = load <37 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %158 = load <38 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %159 = load <39 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %160 = load <40 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %161 = load <41 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %162 = load <42 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %163 = load <43 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %164 = load <44 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %165 = load <45 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %166 = load <46 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %167 = load <47 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %168 = load <48 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %169 = load <49 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %170 = load <50 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %171 = load <51 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %172 = load <52 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %173 = load <53 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %174 = load <54 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %175 = load <55 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %176 = load <56 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %177 = load <57 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %178 = load <58 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %179 = load <59 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %180 = load <60 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %181 = load <61 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %182 = load <62 x i1>, ptr undef, align 1 +; SSE41-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %183 = load <63 x i1>, ptr undef, align 1 ; SSE41-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'loads_partial_align1' @@ -5680,151 +5680,151 @@ define i32 @loads_partial_align1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %36 = load <15 x float>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %37 = load <1 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %38 = load <2 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %39 = load <3 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %39 = load <3 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %40 = load <4 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %41 = load <5 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %41 = load <5 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %42 = load <6 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %43 = load <7 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %44 = load <9 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %43 = load <7 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %44 = load <9 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %45 = load <10 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %46 = load <11 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %46 = load <11 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %47 = load <12 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %48 = load <13 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %48 = load <13 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %49 = load <14 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %50 = load <15 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %51 = load <17 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %50 = load <15 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %51 = load <17 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %52 = load <18 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %53 = load <19 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %53 = load <19 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %54 = load <20 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %55 = load <21 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %55 = load <21 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %56 = load <22 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %57 = load <23 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %57 = load <23 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %58 = load <24 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %59 = load <25 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %59 = load <25 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %60 = load <26 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %61 = load <27 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %61 = load <27 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %62 = load <28 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %63 = load <29 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %63 = load <29 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %64 = load <30 x i16>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %65 = load <31 x i16>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %65 = load <31 x i16>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %66 = load <1 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %67 = load <2 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %68 = load <3 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %67 = load <2 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %68 = load <3 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %69 = load <4 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %70 = load <5 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %71 = load <6 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %72 = load <7 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %70 = load <5 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %71 = load <6 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %72 = load <7 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %73 = load <8 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %74 = load <9 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %75 = load <10 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %76 = load <11 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %74 = load <9 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %75 = load <10 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %76 = load <11 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %77 = load <12 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %78 = load <13 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %79 = load <14 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %80 = load <15 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %81 = load <17 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %82 = load <18 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %83 = load <19 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %78 = load <13 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %79 = load <14 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %80 = load <15 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %81 = load <17 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %82 = load <18 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %83 = load <19 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %84 = load <20 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %85 = load <21 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %86 = load <22 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %87 = load <23 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %85 = load <21 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %86 = load <22 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %87 = load <23 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %88 = load <24 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %89 = load <25 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %90 = load <26 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %91 = load <27 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %89 = load <25 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %90 = load <26 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %91 = load <27 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %92 = load <28 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %93 = load <29 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %94 = load <30 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %95 = load <31 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %96 = load <33 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %97 = load <34 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %98 = load <35 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %93 = load <29 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %94 = load <30 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %95 = load <31 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %96 = load <33 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %97 = load <34 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %98 = load <35 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %99 = load <36 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %100 = load <37 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %101 = load <38 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %102 = load <39 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %100 = load <37 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %101 = load <38 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %102 = load <39 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %103 = load <40 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %104 = load <41 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %105 = load <42 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %106 = load <43 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %104 = load <41 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %105 = load <42 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %106 = load <43 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %107 = load <44 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %108 = load <45 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %109 = load <46 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %110 = load <47 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %108 = load <45 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %109 = load <46 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %110 = load <47 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %111 = load <48 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %112 = load <49 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %113 = load <50 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %114 = load <51 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %112 = load <49 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %113 = load <50 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %114 = load <51 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %115 = load <52 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %116 = load <53 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %117 = load <54 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %118 = load <55 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %116 = load <53 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %117 = load <54 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %118 = load <55 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %119 = load <56 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %120 = load <57 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %121 = load <58 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %122 = load <59 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %120 = load <57 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %121 = load <58 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %122 = load <59 x i8>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %123 = load <60 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %124 = load <61 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %125 = load <62 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %126 = load <63 x i8>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %127 = load <3 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %128 = load <5 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %129 = load <6 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %130 = load <7 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %131 = load <9 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %132 = load <10 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %133 = load <11 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %134 = load <12 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %135 = load <13 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %136 = load <14 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %137 = load <15 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %138 = load <17 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %139 = load <18 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %140 = load <19 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %141 = load <20 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %142 = load <21 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %143 = load <22 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %144 = load <23 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %145 = load <24 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %146 = load <25 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %147 = load <26 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %148 = load <27 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %149 = load <28 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %150 = load <29 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %151 = load <30 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %152 = load <31 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %153 = load <33 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %154 = load <34 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %155 = load <35 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %156 = load <36 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %157 = load <37 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %158 = load <38 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %159 = load <39 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %160 = load <40 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %161 = load <41 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %162 = load <42 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %163 = load <43 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %164 = load <44 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %165 = load <45 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %166 = load <46 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %167 = load <47 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %168 = load <48 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %169 = load <49 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %170 = load <50 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %171 = load <51 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %172 = load <52 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %173 = load <53 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %174 = load <54 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %175 = load <55 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %176 = load <56 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %177 = load <57 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %178 = load <58 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %179 = load <59 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %180 = load <60 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %181 = load <61 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %182 = load <62 x i1>, ptr undef, align 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %183 = load <63 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %124 = load <61 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %125 = load <62 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %126 = load <63 x i8>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %127 = load <3 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %128 = load <5 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %129 = load <6 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %130 = load <7 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %131 = load <9 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %132 = load <10 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %133 = load <11 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %134 = load <12 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %135 = load <13 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %136 = load <14 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %137 = load <15 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %138 = load <17 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %139 = load <18 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %140 = load <19 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %141 = load <20 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %142 = load <21 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %143 = load <22 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %144 = load <23 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %145 = load <24 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %146 = load <25 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %147 = load <26 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %148 = load <27 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %149 = load <28 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %150 = load <29 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %151 = load <30 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %152 = load <31 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %153 = load <33 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %154 = load <34 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %155 = load <35 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %156 = load <36 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %157 = load <37 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %158 = load <38 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %159 = load <39 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %160 = load <40 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %161 = load <41 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %162 = load <42 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %163 = load <43 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %164 = load <44 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %165 = load <45 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %166 = load <46 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %167 = load <47 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %168 = load <48 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %169 = load <49 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %170 = load <50 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %171 = load <51 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %172 = load <52 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %173 = load <53 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %174 = load <54 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %175 = load <55 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %176 = load <56 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %177 = load <57 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %178 = load <58 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %179 = load <59 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %180 = load <60 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %181 = load <61 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %182 = load <62 x i1>, ptr undef, align 1 +; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %183 = load <63 x i1>, ptr undef, align 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'loads_partial_align1' @@ -5866,151 +5866,151 @@ define i32 @loads_partial_align1(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %36 = load <15 x float>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %37 = load <1 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %38 = load <2 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %39 = load <3 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %39 = load <3 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %40 = load <4 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %41 = load <5 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %41 = load <5 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %42 = load <6 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %43 = load <7 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %44 = load <9 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %43 = load <7 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %44 = load <9 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %45 = load <10 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %46 = load <11 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %46 = load <11 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %47 = load <12 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %48 = load <13 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %48 = load <13 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %49 = load <14 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %50 = load <15 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %51 = load <17 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %50 = load <15 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %51 = load <17 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %52 = load <18 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %53 = load <19 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %53 = load <19 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %54 = load <20 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %55 = load <21 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %55 = load <21 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %56 = load <22 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %57 = load <23 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %57 = load <23 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %58 = load <24 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %59 = load <25 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %59 = load <25 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %60 = load <26 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %61 = load <27 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %61 = load <27 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %62 = load <28 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %63 = load <29 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %63 = load <29 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %64 = load <30 x i16>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %65 = load <31 x i16>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %65 = load <31 x i16>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %66 = load <1 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %67 = load <2 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %68 = load <3 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %67 = load <2 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %68 = load <3 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %69 = load <4 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %70 = load <5 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %71 = load <6 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %72 = load <7 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %70 = load <5 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %71 = load <6 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %72 = load <7 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %73 = load <8 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %74 = load <9 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %75 = load <10 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %76 = load <11 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %74 = load <9 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %75 = load <10 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %76 = load <11 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %77 = load <12 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %78 = load <13 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %79 = load <14 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %80 = load <15 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %81 = load <17 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %82 = load <18 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %83 = load <19 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %78 = load <13 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %79 = load <14 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %80 = load <15 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %81 = load <17 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %82 = load <18 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %83 = load <19 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %84 = load <20 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %85 = load <21 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %86 = load <22 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %87 = load <23 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %85 = load <21 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %86 = load <22 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %87 = load <23 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %88 = load <24 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %89 = load <25 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %90 = load <26 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %91 = load <27 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %89 = load <25 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %90 = load <26 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %91 = load <27 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %92 = load <28 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %93 = load <29 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %94 = load <30 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %95 = load <31 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %96 = load <33 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %97 = load <34 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %98 = load <35 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %93 = load <29 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %94 = load <30 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %95 = load <31 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %96 = load <33 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %97 = load <34 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %98 = load <35 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %99 = load <36 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %100 = load <37 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %101 = load <38 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %102 = load <39 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %100 = load <37 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %101 = load <38 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %102 = load <39 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %103 = load <40 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %104 = load <41 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %105 = load <42 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %106 = load <43 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %104 = load <41 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %105 = load <42 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %106 = load <43 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %107 = load <44 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %108 = load <45 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %109 = load <46 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %110 = load <47 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %108 = load <45 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %109 = load <46 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %110 = load <47 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %111 = load <48 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %112 = load <49 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %113 = load <50 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %114 = load <51 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %112 = load <49 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %113 = load <50 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %114 = load <51 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %115 = load <52 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %116 = load <53 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %117 = load <54 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %118 = load <55 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %116 = load <53 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %117 = load <54 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %118 = load <55 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %119 = load <56 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %120 = load <57 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %121 = load <58 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %122 = load <59 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %120 = load <57 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %121 = load <58 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %122 = load <59 x i8>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %123 = load <60 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %124 = load <61 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %125 = load <62 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %126 = load <63 x i8>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %127 = load <3 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %128 = load <5 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %129 = load <6 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %130 = load <7 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %131 = load <9 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %132 = load <10 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %133 = load <11 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %134 = load <12 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %135 = load <13 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %136 = load <14 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %137 = load <15 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %138 = load <17 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %139 = load <18 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %140 = load <19 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %141 = load <20 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %142 = load <21 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %143 = load <22 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %144 = load <23 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %145 = load <24 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %146 = load <25 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %147 = load <26 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %148 = load <27 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %149 = load <28 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %150 = load <29 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %151 = load <30 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %152 = load <31 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %153 = load <33 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %154 = load <34 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %155 = load <35 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %156 = load <36 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %157 = load <37 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %158 = load <38 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %159 = load <39 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %160 = load <40 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %161 = load <41 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %162 = load <42 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %163 = load <43 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %164 = load <44 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %165 = load <45 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %166 = load <46 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %167 = load <47 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %168 = load <48 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %169 = load <49 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %170 = load <50 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %171 = load <51 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %172 = load <52 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %173 = load <53 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %174 = load <54 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %175 = load <55 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %176 = load <56 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %177 = load <57 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %178 = load <58 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %179 = load <59 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %180 = load <60 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %181 = load <61 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %182 = load <62 x i1>, ptr undef, align 1 -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %183 = load <63 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %124 = load <61 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %125 = load <62 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %126 = load <63 x i8>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %127 = load <3 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %128 = load <5 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %129 = load <6 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %130 = load <7 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %131 = load <9 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %132 = load <10 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %133 = load <11 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %134 = load <12 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %135 = load <13 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %136 = load <14 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %137 = load <15 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %138 = load <17 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %139 = load <18 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %140 = load <19 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %141 = load <20 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %142 = load <21 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %143 = load <22 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %144 = load <23 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %145 = load <24 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %146 = load <25 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %147 = load <26 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %148 = load <27 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %149 = load <28 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %150 = load <29 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %151 = load <30 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %152 = load <31 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %153 = load <33 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %154 = load <34 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %155 = load <35 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %156 = load <36 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %157 = load <37 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %158 = load <38 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %159 = load <39 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %160 = load <40 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %161 = load <41 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %162 = load <42 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %163 = load <43 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %164 = load <44 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %165 = load <45 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %166 = load <46 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %167 = load <47 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %168 = load <48 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %169 = load <49 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %170 = load <50 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %171 = load <51 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %172 = load <52 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %173 = load <53 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %174 = load <54 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %175 = load <55 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %176 = load <56 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %177 = load <57 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %178 = load <58 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %179 = load <59 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %180 = load <60 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %181 = load <61 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %182 = load <62 x i1>, ptr undef, align 1 +; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %183 = load <63 x i1>, ptr undef, align 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; load <1 x i64>, ptr undef, align 1 @@ -6413,11 +6413,11 @@ define i32 @loads_align64(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = load <4 x float>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = load <2 x double>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = load <2 x ptr>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = load <16 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %22 = load <8 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %23 = load <4 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %24 = load <2 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %25 = load <1 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %21 = load <16 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %22 = load <8 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %23 = load <4 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %24 = load <2 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %25 = load <1 x i1>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %26 = load <256 x i1>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %27 = load <32 x i8>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %28 = load <16 x i16>, ptr undef, align 64 @@ -7200,17 +7200,17 @@ define i32 @loads_partial_align64(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %124 = load <61 x i8>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %125 = load <62 x i8>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %126 = load <63 x i8>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %127 = load <3 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %128 = load <5 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %129 = load <6 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %130 = load <7 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %131 = load <9 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %132 = load <10 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %133 = load <11 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %134 = load <12 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %135 = load <13 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %136 = load <14 x i1>, ptr undef, align 64 -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %137 = load <15 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %127 = load <3 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %128 = load <5 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %129 = load <6 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %130 = load <7 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %131 = load <9 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %132 = load <10 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %133 = load <11 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %134 = load <12 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %135 = load <13 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %136 = load <14 x i1>, ptr undef, align 64 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %137 = load <15 x i1>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %138 = load <17 x i1>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %139 = load <18 x i1>, ptr undef, align 64 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %140 = load <19 x i1>, ptr undef, align 64 diff --git a/llvm/test/Analysis/CostModel/X86/shuffle-load.ll b/llvm/test/Analysis/CostModel/X86/shuffle-load.ll index 8046333..12d0545 100644 --- a/llvm/test/Analysis/CostModel/X86/shuffle-load.ll +++ b/llvm/test/Analysis/CostModel/X86/shuffle-load.ll @@ -25,7 +25,7 @@ define void @shuffle_load() { ; SSE-LABEL: 'shuffle_load' -; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 +; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sf_2xi8 = shufflevector <2 x i8> %ld_2xi8, <2 x i8> undef, <2 x i32> zeroinitializer ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_4xi8 = load <4 x i8>, ptr undef, align 4 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %sf_4xi8 = shufflevector <4 x i8> %ld_4xi8, <4 x i8> undef, <4 x i32> zeroinitializer @@ -88,7 +88,7 @@ define void @shuffle_load() { ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; ; SSE2-LABEL: 'shuffle_load' -; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 +; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sf_2xi8 = shufflevector <2 x i8> %ld_2xi8, <2 x i8> undef, <2 x i32> zeroinitializer ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_4xi8 = load <4 x i8>, ptr undef, align 4 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %sf_4xi8 = shufflevector <4 x i8> %ld_4xi8, <4 x i8> undef, <4 x i32> zeroinitializer @@ -151,7 +151,7 @@ define void @shuffle_load() { ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; ; SSE3-LABEL: 'shuffle_load' -; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 +; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sf_2xi8 = shufflevector <2 x i8> %ld_2xi8, <2 x i8> undef, <2 x i32> zeroinitializer ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_4xi8 = load <4 x i8>, ptr undef, align 4 ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %sf_4xi8 = shufflevector <4 x i8> %ld_4xi8, <4 x i8> undef, <4 x i32> zeroinitializer @@ -214,7 +214,7 @@ define void @shuffle_load() { ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; ; AVX-LABEL: 'shuffle_load' -; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sf_2xi8 = shufflevector <2 x i8> %ld_2xi8, <2 x i8> undef, <2 x i32> zeroinitializer ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_4xi8 = load <4 x i8>, ptr undef, align 4 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %sf_4xi8 = shufflevector <4 x i8> %ld_4xi8, <4 x i8> undef, <4 x i32> zeroinitializer @@ -277,7 +277,7 @@ define void @shuffle_load() { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; ; AVX2-LABEL: 'shuffle_load' -; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 +; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sf_2xi8 = shufflevector <2 x i8> %ld_2xi8, <2 x i8> undef, <2 x i32> zeroinitializer ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_4xi8 = load <4 x i8>, ptr undef, align 4 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %sf_4xi8 = shufflevector <4 x i8> %ld_4xi8, <4 x i8> undef, <4 x i32> zeroinitializer @@ -340,7 +340,7 @@ define void @shuffle_load() { ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; ; AVX512-LABEL: 'shuffle_load' -; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %ld_2xi8 = load <2 x i8>, ptr undef, align 2 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %sf_2xi8 = shufflevector <2 x i8> %ld_2xi8, <2 x i8> undef, <2 x i32> zeroinitializer ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ld_4xi8 = load <4 x i8>, ptr undef, align 4 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %sf_4xi8 = shufflevector <4 x i8> %ld_4xi8, <4 x i8> undef, <4 x i32> zeroinitializer diff --git a/llvm/test/Analysis/CostModel/X86/vector-extract.ll b/llvm/test/Analysis/CostModel/X86/vector-extract.ll index 036deb3..b5eefbd 100644 --- a/llvm/test/Analysis/CostModel/X86/vector-extract.ll +++ b/llvm/test/Analysis/CostModel/X86/vector-extract.ll @@ -693,7 +693,7 @@ define i32 @extract_i16(i32 %arg) { define i32 @extract_i8(i32 %arg) { ; SSE2-LABEL: 'extract_i8' -; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -726,7 +726,7 @@ define i32 @extract_i8(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'extract_i8' -; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -759,7 +759,7 @@ define i32 @extract_i8(i32 %arg) { ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'extract_i8' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -792,7 +792,7 @@ define i32 @extract_i8(i32 %arg) { ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'extract_i8' -; SSE4-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; SSE4-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -825,7 +825,7 @@ define i32 @extract_i8(i32 %arg) { ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'extract_i8' -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -858,7 +858,7 @@ define i32 @extract_i8(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'extract_i8' -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -891,7 +891,7 @@ define i32 @extract_i8(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'extract_i8' -; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; SLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -924,7 +924,7 @@ define i32 @extract_i8(i32 %arg) { ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'extract_i8' -; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = extractelement <2 x i8> undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = extractelement <2 x i8> undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_1 = extractelement <2 x i8> undef, i32 1 ; GLM-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i8_a = extractelement <4 x i8> undef, i32 %arg @@ -997,16 +997,16 @@ define i32 @extract_i8(i32 %arg) { define i32 @extract_i1(i32 %arg) { ; SSE-LABEL: 'extract_i1' -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = extractelement <2 x i1> undef, i32 0 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = extractelement <2 x i1> undef, i32 1 -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = extractelement <4 x i1> undef, i32 0 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = extractelement <4 x i1> undef, i32 2 -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = extractelement <8 x i1> undef, i32 0 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = extractelement <8 x i1> undef, i32 4 -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg +; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = extractelement <16 x i1> undef, i32 0 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = extractelement <16 x i1> undef, i32 8 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = extractelement <16 x i1> undef, i32 15 @@ -1030,16 +1030,16 @@ define i32 @extract_i1(i32 %arg) { ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'extract_i1' -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = extractelement <2 x i1> undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = extractelement <2 x i1> undef, i32 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = extractelement <4 x i1> undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = extractelement <4 x i1> undef, i32 2 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = extractelement <8 x i1> undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = extractelement <8 x i1> undef, i32 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = extractelement <16 x i1> undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = extractelement <16 x i1> undef, i32 8 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = extractelement <16 x i1> undef, i32 15 @@ -1063,27 +1063,27 @@ define i32 @extract_i1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512F-LABEL: 'extract_i1' -; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = extractelement <2 x i1> undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = extractelement <2 x i1> undef, i32 1 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = extractelement <4 x i1> undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = extractelement <4 x i1> undef, i32 2 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = extractelement <8 x i1> undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = extractelement <8 x i1> undef, i32 4 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = extractelement <16 x i1> undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = extractelement <16 x i1> undef, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = extractelement <16 x i1> undef, i32 15 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = extractelement <32 x i1> undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v32i1_a = extractelement <32 x i1> undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = extractelement <32 x i1> undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = extractelement <32 x i1> undef, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = extractelement <32 x i1> undef, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = extractelement <32 x i1> undef, i32 15 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = extractelement <32 x i1> undef, i32 24 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = extractelement <32 x i1> undef, i32 31 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v64i1_a = extractelement <64 x i1> undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v64i1_a = extractelement <64 x i1> undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = extractelement <64 x i1> undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = extractelement <64 x i1> undef, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = extractelement <64 x i1> undef, i32 8 @@ -1096,16 +1096,16 @@ define i32 @extract_i1(i32 %arg) { ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512BW-LABEL: 'extract_i1' -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = extractelement <2 x i1> undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = extractelement <2 x i1> undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = extractelement <2 x i1> undef, i32 1 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = extractelement <4 x i1> undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = extractelement <4 x i1> undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = extractelement <4 x i1> undef, i32 2 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = extractelement <8 x i1> undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = extractelement <8 x i1> undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = extractelement <8 x i1> undef, i32 4 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = extractelement <16 x i1> undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = extractelement <16 x i1> undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = extractelement <16 x i1> undef, i32 8 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = extractelement <16 x i1> undef, i32 15 diff --git a/llvm/test/Analysis/CostModel/X86/vector-insert-inseltpoison.ll b/llvm/test/Analysis/CostModel/X86/vector-insert-inseltpoison.ll index e6a4de6..9caaae0 100644 --- a/llvm/test/Analysis/CostModel/X86/vector-insert-inseltpoison.ll +++ b/llvm/test/Analysis/CostModel/X86/vector-insert-inseltpoison.ll @@ -663,7 +663,7 @@ define i32 @insert_i16(i32 %arg) { define i32 @insert_i8(i32 %arg) { ; SSE2-LABEL: 'insert_i8' -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -696,7 +696,7 @@ define i32 @insert_i8(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'insert_i8' -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -729,7 +729,7 @@ define i32 @insert_i8(i32 %arg) { ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'insert_i8' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -762,7 +762,7 @@ define i32 @insert_i8(i32 %arg) { ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'insert_i8' -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -795,7 +795,7 @@ define i32 @insert_i8(i32 %arg) { ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'insert_i8' -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -828,7 +828,7 @@ define i32 @insert_i8(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'insert_i8' -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -861,7 +861,7 @@ define i32 @insert_i8(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'insert_i8' -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -894,7 +894,7 @@ define i32 @insert_i8(i32 %arg) { ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'insert_i8' -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> poison, i8 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> poison, i8 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> poison, i8 undef, i32 1 ; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> poison, i8 undef, i32 %arg @@ -967,16 +967,16 @@ define i32 @insert_i8(i32 %arg) { define i32 @insert_i1(i32 %arg) { ; SSE2-LABEL: 'insert_i1' -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1000,16 +1000,16 @@ define i32 @insert_i1(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'insert_i1' -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1033,16 +1033,16 @@ define i32 @insert_i1(i32 %arg) { ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'insert_i1' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1066,16 +1066,16 @@ define i32 @insert_i1(i32 %arg) { ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'insert_i1' -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1099,16 +1099,16 @@ define i32 @insert_i1(i32 %arg) { ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'insert_i1' -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1132,27 +1132,27 @@ define i32 @insert_i1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512F-LABEL: 'insert_i1' -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v32i1_a = insertelement <32 x i1> poison, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v32i1_a = insertelement <32 x i1> poison, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> poison, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> poison, i1 undef, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> poison, i1 undef, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> poison, i1 undef, i32 15 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> poison, i1 undef, i32 24 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> poison, i1 undef, i32 31 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v64i1_a = insertelement <64 x i1> poison, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v64i1_a = insertelement <64 x i1> poison, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> poison, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> poison, i1 undef, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> poison, i1 undef, i32 8 @@ -1165,16 +1165,16 @@ define i32 @insert_i1(i32 %arg) { ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512BW-LABEL: 'insert_i1' -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1198,16 +1198,16 @@ define i32 @insert_i1(i32 %arg) { ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'insert_i1' -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 @@ -1231,16 +1231,16 @@ define i32 @insert_i1(i32 %arg) { ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'insert_i1' -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> poison, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> poison, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> poison, i1 undef, i32 1 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> poison, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> poison, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> poison, i1 undef, i32 2 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> poison, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> poison, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> poison, i1 undef, i32 4 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> poison, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> poison, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> poison, i1 undef, i32 8 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> poison, i1 undef, i32 15 diff --git a/llvm/test/Analysis/CostModel/X86/vector-insert-value.ll b/llvm/test/Analysis/CostModel/X86/vector-insert-value.ll index 80613b6..2524976 100644 --- a/llvm/test/Analysis/CostModel/X86/vector-insert-value.ll +++ b/llvm/test/Analysis/CostModel/X86/vector-insert-value.ll @@ -663,7 +663,7 @@ define i32 @insert_i16(i32 %arg, i16 %val, <2 x i16> %src32, <4 x i16> %src64, < define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512) { ; SSE2-LABEL: 'insert_i8' -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -696,7 +696,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'insert_i8' -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -729,7 +729,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'insert_i8' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -762,7 +762,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'insert_i8' -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -795,7 +795,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'insert_i8' -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -828,7 +828,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'insert_i8' -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -861,7 +861,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'insert_i8' -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -894,7 +894,7 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'insert_i8' -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1 ; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg @@ -967,16 +967,16 @@ define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1> %src8, <16 x i1> %src16, <32 x i1> %src32, <64 x i1> %src64) { ; SSE2-LABEL: 'insert_i1' -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1000,16 +1000,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'insert_i1' -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1033,16 +1033,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'insert_i1' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1066,16 +1066,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'insert_i1' -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1099,16 +1099,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'insert_i1' -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1132,27 +1132,27 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512F-LABEL: 'insert_i1' -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8 @@ -1165,16 +1165,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512BW-LABEL: 'insert_i1' -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1198,16 +1198,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'insert_i1' -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 @@ -1231,16 +1231,16 @@ define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1 ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'insert_i1' -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15 diff --git a/llvm/test/Analysis/CostModel/X86/vector-insert.ll b/llvm/test/Analysis/CostModel/X86/vector-insert.ll index fd40da9..1d2e1f2 100644 --- a/llvm/test/Analysis/CostModel/X86/vector-insert.ll +++ b/llvm/test/Analysis/CostModel/X86/vector-insert.ll @@ -663,7 +663,7 @@ define i32 @insert_i16(i32 %arg) { define i32 @insert_i8(i32 %arg) { ; SSE2-LABEL: 'insert_i8' -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -696,7 +696,7 @@ define i32 @insert_i8(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'insert_i8' -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -729,7 +729,7 @@ define i32 @insert_i8(i32 %arg) { ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'insert_i8' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -762,7 +762,7 @@ define i32 @insert_i8(i32 %arg) { ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'insert_i8' -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -795,7 +795,7 @@ define i32 @insert_i8(i32 %arg) { ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'insert_i8' -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -828,7 +828,7 @@ define i32 @insert_i8(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'insert_i8' -; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -861,7 +861,7 @@ define i32 @insert_i8(i32 %arg) { ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'insert_i8' -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -894,7 +894,7 @@ define i32 @insert_i8(i32 %arg) { ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'insert_i8' -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> undef, i8 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> undef, i8 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> undef, i8 undef, i32 1 ; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> undef, i8 undef, i32 %arg @@ -967,16 +967,16 @@ define i32 @insert_i8(i32 %arg) { define i32 @insert_i1(i32 %arg) { ; SSE2-LABEL: 'insert_i1' -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1000,16 +1000,16 @@ define i32 @insert_i1(i32 %arg) { ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE3-LABEL: 'insert_i1' -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; SSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; SSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; SSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; SSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1033,16 +1033,16 @@ define i32 @insert_i1(i32 %arg) { ; SSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSSE3-LABEL: 'insert_i1' -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1066,16 +1066,16 @@ define i32 @insert_i1(i32 %arg) { ; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SSE4-LABEL: 'insert_i1' -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; SSE4-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; SSE4-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; SSE4-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1099,16 +1099,16 @@ define i32 @insert_i1(i32 %arg) { ; SSE4-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX-LABEL: 'insert_i1' -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1132,27 +1132,27 @@ define i32 @insert_i1(i32 %arg) { ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512F-LABEL: 'insert_i1' -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v32i1_a = insertelement <32 x i1> undef, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v32i1_a = insertelement <32 x i1> undef, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> undef, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> undef, i1 undef, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> undef, i1 undef, i32 8 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> undef, i1 undef, i32 15 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> undef, i1 undef, i32 24 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> undef, i1 undef, i32 31 -; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v64i1_a = insertelement <64 x i1> undef, i1 undef, i32 %arg +; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v64i1_a = insertelement <64 x i1> undef, i1 undef, i32 %arg ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> undef, i1 undef, i32 0 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> undef, i1 undef, i32 7 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> undef, i1 undef, i32 8 @@ -1165,16 +1165,16 @@ define i32 @insert_i1(i32 %arg) { ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512BW-LABEL: 'insert_i1' -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1198,16 +1198,16 @@ define i32 @insert_i1(i32 %arg) { ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; SLM-LABEL: 'insert_i1' -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; SLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 @@ -1231,16 +1231,16 @@ define i32 @insert_i1(i32 %arg) { ; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; GLM-LABEL: 'insert_i1' -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> undef, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> undef, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> undef, i1 undef, i32 1 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> undef, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> undef, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> undef, i1 undef, i32 2 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> undef, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> undef, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> undef, i1 undef, i32 4 -; GLM-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg +; GLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> undef, i1 undef, i32 %arg ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> undef, i1 undef, i32 0 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> undef, i1 undef, i32 8 ; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> undef, i1 undef, i32 15 diff --git a/llvm/test/Transforms/SLPVectorizer/X86/bool-mask.ll b/llvm/test/Transforms/SLPVectorizer/X86/bool-mask.ll index edac08c..cb44d05 100644 --- a/llvm/test/Transforms/SLPVectorizer/X86/bool-mask.ll +++ b/llvm/test/Transforms/SLPVectorizer/X86/bool-mask.ll @@ -1,8 +1,8 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py -; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64 -S | FileCheck %s --check-prefixes=SSE,SSE2 -; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64-v2 -S | FileCheck %s --check-prefixes=SSE,SSE4 -; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64-v3 -S | FileCheck %s --check-prefixes=AVX -; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64-v4 -S | FileCheck %s --check-prefixes=AVX512 +; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64 -S | FileCheck %s --check-prefixes=CHECK,SSE,SSE2 +; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64-v2 -S | FileCheck %s --check-prefixes=CHECK,SSE,SSE4 +; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64-v3 -S | FileCheck %s --check-prefixes=CHECK,AVX +; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown -mcpu=x86-64-v4 -S | FileCheck %s --check-prefixes=CHECK,AVX512 ; // PR42652 ; unsigned long bitmask_16xi8(const char *src) { @@ -15,105 +15,39 @@ ; } define i64 @bitmask_16xi8(ptr nocapture noundef readonly %src) { -; SSE-LABEL: @bitmask_16xi8( -; SSE-NEXT: entry: -; SSE-NEXT: [[TMP0:%.*]] = load i8, ptr [[SRC:%.*]], align 1 -; SSE-NEXT: [[TOBOOL_NOT:%.*]] = icmp ne i8 [[TMP0]], 0 -; SSE-NEXT: [[OR:%.*]] = zext i1 [[TOBOOL_NOT]] to i64 -; SSE-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 1 -; SSE-NEXT: [[TMP1:%.*]] = load <8 x i8>, ptr [[ARRAYIDX_1]], align 1 -; SSE-NEXT: [[TMP2:%.*]] = icmp eq <8 x i8> [[TMP1]], zeroinitializer -; SSE-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i64> zeroinitializer, <8 x i64> -; SSE-NEXT: [[ARRAYIDX_9:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 9 -; SSE-NEXT: [[TMP4:%.*]] = load <4 x i8>, ptr [[ARRAYIDX_9]], align 1 -; SSE-NEXT: [[TMP5:%.*]] = icmp eq <4 x i8> [[TMP4]], zeroinitializer -; SSE-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i64> zeroinitializer, <4 x i64> -; SSE-NEXT: [[ARRAYIDX_13:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 13 -; SSE-NEXT: [[TMP7:%.*]] = load i8, ptr [[ARRAYIDX_13]], align 1 -; SSE-NEXT: [[TOBOOL_NOT_13:%.*]] = icmp eq i8 [[TMP7]], 0 -; SSE-NEXT: [[OR_13:%.*]] = select i1 [[TOBOOL_NOT_13]], i64 0, i64 8192 -; SSE-NEXT: [[ARRAYIDX_14:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 14 -; SSE-NEXT: [[TMP8:%.*]] = load i8, ptr [[ARRAYIDX_14]], align 1 -; SSE-NEXT: [[TOBOOL_NOT_14:%.*]] = icmp eq i8 [[TMP8]], 0 -; SSE-NEXT: [[OR_14:%.*]] = select i1 [[TOBOOL_NOT_14]], i64 0, i64 16384 -; SSE-NEXT: [[ARRAYIDX_15:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 15 -; SSE-NEXT: [[TMP9:%.*]] = load i8, ptr [[ARRAYIDX_15]], align 1 -; SSE-NEXT: [[TOBOOL_NOT_15:%.*]] = icmp eq i8 [[TMP9]], 0 -; SSE-NEXT: [[OR_15:%.*]] = select i1 [[TOBOOL_NOT_15]], i64 0, i64 32768 -; SSE-NEXT: [[TMP10:%.*]] = call i64 @llvm.vector.reduce.or.v8i64(<8 x i64> [[TMP3]]) -; SSE-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.or.v4i64(<4 x i64> [[TMP6]]) -; SSE-NEXT: [[OP_RDX:%.*]] = or i64 [[TMP10]], [[TMP11]] -; SSE-NEXT: [[OP_RDX1:%.*]] = or i64 [[OP_RDX]], [[OR_13]] -; SSE-NEXT: [[OP_RDX2:%.*]] = or i64 [[OR_14]], [[OR_15]] -; SSE-NEXT: [[OP_RDX3:%.*]] = or i64 [[OP_RDX1]], [[OP_RDX2]] -; SSE-NEXT: [[OP_RDX4:%.*]] = or i64 [[OP_RDX3]], [[OR]] -; SSE-NEXT: ret i64 [[OP_RDX4]] -; -; AVX-LABEL: @bitmask_16xi8( -; AVX-NEXT: entry: -; AVX-NEXT: [[TMP0:%.*]] = load i8, ptr [[SRC:%.*]], align 1 -; AVX-NEXT: [[TOBOOL_NOT:%.*]] = icmp ne i8 [[TMP0]], 0 -; AVX-NEXT: [[OR:%.*]] = zext i1 [[TOBOOL_NOT]] to i64 -; AVX-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 1 -; AVX-NEXT: [[TMP1:%.*]] = load <8 x i8>, ptr [[ARRAYIDX_1]], align 1 -; AVX-NEXT: [[TMP2:%.*]] = icmp eq <8 x i8> [[TMP1]], zeroinitializer -; AVX-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i64> zeroinitializer, <8 x i64> -; AVX-NEXT: [[ARRAYIDX_9:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 9 -; AVX-NEXT: [[TMP4:%.*]] = load <4 x i8>, ptr [[ARRAYIDX_9]], align 1 -; AVX-NEXT: [[TMP5:%.*]] = icmp eq <4 x i8> [[TMP4]], zeroinitializer -; AVX-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i64> zeroinitializer, <4 x i64> -; AVX-NEXT: [[ARRAYIDX_13:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 13 -; AVX-NEXT: [[TMP7:%.*]] = load i8, ptr [[ARRAYIDX_13]], align 1 -; AVX-NEXT: [[TOBOOL_NOT_13:%.*]] = icmp eq i8 [[TMP7]], 0 -; AVX-NEXT: [[OR_13:%.*]] = select i1 [[TOBOOL_NOT_13]], i64 0, i64 8192 -; AVX-NEXT: [[ARRAYIDX_14:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 14 -; AVX-NEXT: [[TMP8:%.*]] = load i8, ptr [[ARRAYIDX_14]], align 1 -; AVX-NEXT: [[TOBOOL_NOT_14:%.*]] = icmp eq i8 [[TMP8]], 0 -; AVX-NEXT: [[OR_14:%.*]] = select i1 [[TOBOOL_NOT_14]], i64 0, i64 16384 -; AVX-NEXT: [[ARRAYIDX_15:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 15 -; AVX-NEXT: [[TMP9:%.*]] = load i8, ptr [[ARRAYIDX_15]], align 1 -; AVX-NEXT: [[TOBOOL_NOT_15:%.*]] = icmp eq i8 [[TMP9]], 0 -; AVX-NEXT: [[OR_15:%.*]] = select i1 [[TOBOOL_NOT_15]], i64 0, i64 32768 -; AVX-NEXT: [[TMP10:%.*]] = call i64 @llvm.vector.reduce.or.v8i64(<8 x i64> [[TMP3]]) -; AVX-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.or.v4i64(<4 x i64> [[TMP6]]) -; AVX-NEXT: [[OP_RDX:%.*]] = or i64 [[TMP10]], [[TMP11]] -; AVX-NEXT: [[OP_RDX1:%.*]] = or i64 [[OP_RDX]], [[OR_13]] -; AVX-NEXT: [[OP_RDX2:%.*]] = or i64 [[OR_14]], [[OR_15]] -; AVX-NEXT: [[OP_RDX3:%.*]] = or i64 [[OP_RDX1]], [[OP_RDX2]] -; AVX-NEXT: [[OP_RDX4:%.*]] = or i64 [[OP_RDX3]], [[OR]] -; AVX-NEXT: ret i64 [[OP_RDX4]] -; -; AVX512-LABEL: @bitmask_16xi8( -; AVX512-NEXT: entry: -; AVX512-NEXT: [[TMP0:%.*]] = load i8, ptr [[SRC:%.*]], align 1 -; AVX512-NEXT: [[TOBOOL_NOT:%.*]] = icmp ne i8 [[TMP0]], 0 -; AVX512-NEXT: [[OR:%.*]] = zext i1 [[TOBOOL_NOT]] to i64 -; AVX512-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 1 -; AVX512-NEXT: [[TMP1:%.*]] = load <8 x i8>, ptr [[ARRAYIDX_1]], align 1 -; AVX512-NEXT: [[TMP2:%.*]] = icmp eq <8 x i8> [[TMP1]], zeroinitializer -; AVX512-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i64> zeroinitializer, <8 x i64> -; AVX512-NEXT: [[ARRAYIDX_9:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 9 -; AVX512-NEXT: [[TMP4:%.*]] = load <4 x i8>, ptr [[ARRAYIDX_9]], align 1 -; AVX512-NEXT: [[TMP5:%.*]] = icmp eq <4 x i8> [[TMP4]], zeroinitializer -; AVX512-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i64> zeroinitializer, <4 x i64> -; AVX512-NEXT: [[ARRAYIDX_13:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 13 -; AVX512-NEXT: [[TMP7:%.*]] = load i8, ptr [[ARRAYIDX_13]], align 1 -; AVX512-NEXT: [[TOBOOL_NOT_13:%.*]] = icmp eq i8 [[TMP7]], 0 -; AVX512-NEXT: [[OR_13:%.*]] = select i1 [[TOBOOL_NOT_13]], i64 0, i64 8192 -; AVX512-NEXT: [[ARRAYIDX_14:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 14 -; AVX512-NEXT: [[TMP8:%.*]] = load <2 x i8>, ptr [[ARRAYIDX_14]], align 1 -; AVX512-NEXT: [[TMP9:%.*]] = icmp eq <2 x i8> [[TMP8]], zeroinitializer -; AVX512-NEXT: [[TMP10:%.*]] = select <2 x i1> [[TMP9]], <2 x i64> zeroinitializer, <2 x i64> -; AVX512-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.or.v8i64(<8 x i64> [[TMP3]]) -; AVX512-NEXT: [[TMP12:%.*]] = call i64 @llvm.vector.reduce.or.v4i64(<4 x i64> [[TMP6]]) -; AVX512-NEXT: [[OP_RDX:%.*]] = or i64 [[TMP11]], [[TMP12]] -; AVX512-NEXT: [[OP_RDX1:%.*]] = or i64 [[OP_RDX]], [[OR_13]] -; AVX512-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP10]], i32 0 -; AVX512-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP10]], i32 1 -; AVX512-NEXT: [[OP_RDX2:%.*]] = or i64 [[TMP13]], [[TMP14]] -; AVX512-NEXT: [[OP_RDX3:%.*]] = or i64 [[OP_RDX1]], [[OP_RDX2]] -; AVX512-NEXT: [[OP_RDX4:%.*]] = or i64 [[OP_RDX3]], [[OR]] -; AVX512-NEXT: ret i64 [[OP_RDX4]] +; CHECK-LABEL: @bitmask_16xi8( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[SRC:%.*]], align 1 +; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp ne i8 [[TMP0]], 0 +; CHECK-NEXT: [[OR:%.*]] = zext i1 [[TOBOOL_NOT]] to i64 +; CHECK-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 1 +; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i8>, ptr [[ARRAYIDX_1]], align 1 +; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <8 x i8> [[TMP1]], zeroinitializer +; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i64> zeroinitializer, <8 x i64> +; CHECK-NEXT: [[ARRAYIDX_9:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 9 +; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i8>, ptr [[ARRAYIDX_9]], align 1 +; CHECK-NEXT: [[TMP5:%.*]] = icmp eq <4 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[TMP5]], <4 x i64> zeroinitializer, <4 x i64> +; CHECK-NEXT: [[ARRAYIDX_13:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 13 +; CHECK-NEXT: [[TMP7:%.*]] = load i8, ptr [[ARRAYIDX_13]], align 1 +; CHECK-NEXT: [[TOBOOL_NOT_13:%.*]] = icmp eq i8 [[TMP7]], 0 +; CHECK-NEXT: [[OR_13:%.*]] = select i1 [[TOBOOL_NOT_13]], i64 0, i64 8192 +; CHECK-NEXT: [[ARRAYIDX_14:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 14 +; CHECK-NEXT: [[TMP8:%.*]] = load i8, ptr [[ARRAYIDX_14]], align 1 +; CHECK-NEXT: [[TOBOOL_NOT_14:%.*]] = icmp eq i8 [[TMP8]], 0 +; CHECK-NEXT: [[OR_14:%.*]] = select i1 [[TOBOOL_NOT_14]], i64 0, i64 16384 +; CHECK-NEXT: [[ARRAYIDX_15:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 15 +; CHECK-NEXT: [[TMP9:%.*]] = load i8, ptr [[ARRAYIDX_15]], align 1 +; CHECK-NEXT: [[TOBOOL_NOT_15:%.*]] = icmp eq i8 [[TMP9]], 0 +; CHECK-NEXT: [[OR_15:%.*]] = select i1 [[TOBOOL_NOT_15]], i64 0, i64 32768 +; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.vector.reduce.or.v8i64(<8 x i64> [[TMP3]]) +; CHECK-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.or.v4i64(<4 x i64> [[TMP6]]) +; CHECK-NEXT: [[OP_RDX:%.*]] = or i64 [[TMP10]], [[TMP11]] +; CHECK-NEXT: [[OP_RDX1:%.*]] = or i64 [[OP_RDX]], [[OR_13]] +; CHECK-NEXT: [[OP_RDX2:%.*]] = or i64 [[OR_14]], [[OR_15]] +; CHECK-NEXT: [[OP_RDX3:%.*]] = or i64 [[OP_RDX1]], [[OP_RDX2]] +; CHECK-NEXT: [[OP_RDX4:%.*]] = or i64 [[OP_RDX3]], [[OR]] +; CHECK-NEXT: ret i64 [[OP_RDX4]] ; entry: %0 = load i8, ptr %src, align 1 diff --git a/llvm/test/Transforms/SLPVectorizer/X86/pr59867.ll b/llvm/test/Transforms/SLPVectorizer/X86/pr59867.ll index 25d267f..0148cd5 100644 --- a/llvm/test/Transforms/SLPVectorizer/X86/pr59867.ll +++ b/llvm/test/Transforms/SLPVectorizer/X86/pr59867.ll @@ -6,12 +6,15 @@ define i1 @PR59867(ptr %s1, ptr %s2) { ; CHECK-LABEL: @PR59867( -; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i8>, ptr [[S1:%.*]], align 1 -; CHECK-NEXT: [[TMP2:%.*]] = load <2 x i8>, ptr [[S2:%.*]], align 1 -; CHECK-NEXT: [[TMP3:%.*]] = icmp eq <2 x i8> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i1> [[TMP3]], i32 0 -; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i1> [[TMP3]], i32 1 -; CHECK-NEXT: [[RES:%.*]] = select i1 [[TMP4]], i1 [[TMP5]], i1 false +; CHECK-NEXT: [[V1_1:%.*]] = load i8, ptr [[S1:%.*]], align 1 +; CHECK-NEXT: [[V2_1:%.*]] = load i8, ptr [[S2:%.*]], align 1 +; CHECK-NEXT: [[C1:%.*]] = icmp eq i8 [[V1_1]], [[V2_1]] +; CHECK-NEXT: [[S1_2:%.*]] = getelementptr inbounds i8, ptr [[S1]], i64 1 +; CHECK-NEXT: [[V1_2:%.*]] = load i8, ptr [[S1_2]], align 1 +; CHECK-NEXT: [[S2_2:%.*]] = getelementptr inbounds i8, ptr [[S2]], i64 1 +; CHECK-NEXT: [[V2_2:%.*]] = load i8, ptr [[S2_2]], align 1 +; CHECK-NEXT: [[C2:%.*]] = icmp eq i8 [[V1_2]], [[V2_2]] +; CHECK-NEXT: [[RES:%.*]] = select i1 [[C1]], i1 [[C2]], i1 false ; CHECK-NEXT: ret i1 [[RES]] ; %v1.1 = load i8, ptr %s1, align 1 -- 2.7.4