[CostModel][X86] i8/i16 sitofp/uitofp are sext/zext to i32 for sitofp

Provide a generic fallback that extends sub-i32 scalars before using the existing sitofp instructions.

These numbers can be tweaked for specific sse levels, but we should get the default handling in place first.

We get the extension for free for non-vector loads.

diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 3fd1970..cf767fc 100644 (file)
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -2298,6 +2298,23 @@ InstructionCost X86TTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
                                                    LTDest.second, LTSrc.second))
       return AdjustCost(std::max(LTSrc.first, LTDest.first) * Entry->Cost);
 
+  // Fallback, for i8/i16 sitofp/uitofp cases we need to extend to i32 for
+  // sitofp.
+  if ((ISD == ISD::SINT_TO_FP || ISD == ISD::UINT_TO_FP) &&
+      1 < Src->getScalarSizeInBits() && Src->getScalarSizeInBits() < 32) {
+    Type *ExtSrc = Src->getWithNewBitWidth(32);
+    unsigned ExtOpc =
+        (ISD == ISD::SINT_TO_FP) ? Instruction::SExt : Instruction::ZExt;
+
+    // For scalar loads the extend would be free.
+    InstructionCost ExtCost = 0;
+    if (!(Src->isIntegerTy() && I && isa<LoadInst>(I->getOperand(0))))
+      ExtCost = getCastInstrCost(ExtOpc, ExtSrc, Src, CCH, CostKind);
+
+    return ExtCost + getCastInstrCost(Instruction::SIToFP, Dst, ExtSrc,
+                                      TTI::CastContextHint::None, CostKind);
+  }
+
   return AdjustCost(
       BaseT::getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I));
 }

diff --git a/llvm/test/Analysis/CostModel/X86/sitofp.ll b/llvm/test/Analysis/CostModel/X86/sitofp.ll
index b979f65..69af3a2 100644 (file)
--- a/llvm/test/Analysis/CostModel/X86/sitofp.ll
+++ b/llvm/test/Analysis/CostModel/X86/sitofp.ll
@@ -1,32 +1,46 @@
 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx  | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512DQ
 ;
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=slm | FileCheck %s --check-prefixes=SSE,SSE42
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=slm | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=SSE42
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1
 
 define i32 @sitofp_i8_double() {
-; SSE-LABEL: 'sitofp_i8_double'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'sitofp_i8_double'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'sitofp_i8_double'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sitofp_i8_double'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'sitofp_i8_double'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'sitofp_i8_double'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'sitofp_i8_double'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = sitofp i8 undef to double
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = sitofp <2 x i8> undef to <2 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f64 = sitofp <4 x i8> undef to <4 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v8i8_v8f64 = sitofp <8 x i8> undef to <8 x double>
@@ -40,22 +54,36 @@ define i32 @sitofp_i8_double() {
 }
 
 define i32 @sitofp_i16_double() {
-; SSE-LABEL: 'sitofp_i16_double'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'sitofp_i16_double'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'sitofp_i16_double'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sitofp_i16_double'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'sitofp_i16_double'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'sitofp_i16_double'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'sitofp_i16_double'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = sitofp i16 undef to double
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = sitofp <2 x i16> undef to <2 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f64 = sitofp <4 x i16> undef to <4 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v8i16_v8f64 = sitofp <8 x i16> undef to <8 x double>
@@ -155,24 +183,40 @@ define i32 @sitofp_i64_double() {
 }
 
 define i32 @sitofp_i8_float() {
-; SSE-LABEL: 'sitofp_i8_float'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
-; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v16i8_v16f32 = sitofp <16 x i8> undef to <16 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'sitofp_i8_float'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %cvt_v16i8_v16f32 = sitofp <16 x i8> undef to <16 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'sitofp_i8_float'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v16i8_v16f32 = sitofp <16 x i8> undef to <16 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sitofp_i8_float'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v16i8_v16f32 = sitofp <16 x i8> undef to <16 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'sitofp_i8_float'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v16i8_v16f32 = sitofp <16 x i8> undef to <16 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'sitofp_i8_float'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v16i8_v16f32 = sitofp <16 x i8> undef to <16 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'sitofp_i8_float'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = sitofp i8 undef to float
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i8_v2f32 = sitofp <2 x i8> undef to <2 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i8_v4f32 = sitofp <4 x i8> undef to <4 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = sitofp <8 x i8> undef to <8 x float>
@@ -188,24 +232,40 @@ define i32 @sitofp_i8_float() {
 }
 
 define i32 @sitofp_i16_float() {
-; SSE-LABEL: 'sitofp_i16_float'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
-; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v16i16_v16f32 = sitofp <16 x i16> undef to <16 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'sitofp_i16_float'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %cvt_v16i16_v16f32 = sitofp <16 x i16> undef to <16 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'sitofp_i16_float'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v16i16_v16f32 = sitofp <16 x i16> undef to <16 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sitofp_i16_float'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %cvt_v16i16_v16f32 = sitofp <16 x i16> undef to <16 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'sitofp_i16_float'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v16i16_v16f32 = sitofp <16 x i16> undef to <16 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'sitofp_i16_float'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v16i16_v16f32 = sitofp <16 x i16> undef to <16 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'sitofp_i16_float'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = sitofp i16 undef to float
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v2i16_v2f32 = sitofp <2 x i16> undef to <2 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %cvt_v4i16_v4f32 = sitofp <4 x i16> undef to <4 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = sitofp <8 x i16> undef to <8 x float>

diff --git a/llvm/test/Analysis/CostModel/X86/uitofp.ll b/llvm/test/Analysis/CostModel/X86/uitofp.ll
index 159649c..e7dedf7 100644 (file)
--- a/llvm/test/Analysis/CostModel/X86/uitofp.ll
+++ b/llvm/test/Analysis/CostModel/X86/uitofp.ll
@@ -1,32 +1,46 @@
 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx  | FileCheck %s --check-prefixes=AVX,AVX1
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512DQ
 ;
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=slm | FileCheck %s --check-prefixes=SSE
-; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=SSE
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=slm | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=SSE42
 ; RUN: opt < %s -mtriple=x86_64-apple-darwin -cost-model -analyze -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1
 
 define i32 @uitofp_i8_double() {
-; SSE-LABEL: 'uitofp_i8_double'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'uitofp_i8_double'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'uitofp_i8_double'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'uitofp_i8_double'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'uitofp_i8_double'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'uitofp_i8_double'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'uitofp_i8_double'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f64 = uitofp i8 undef to double
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v2i8_v2f64 = uitofp <2 x i8> undef to <2 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f64 = uitofp <4 x i8> undef to <4 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v8i8_v8f64 = uitofp <8 x i8> undef to <8 x double>
@@ -40,22 +54,36 @@ define i32 @uitofp_i8_double() {
 }
 
 define i32 @uitofp_i16_double() {
-; SSE-LABEL: 'uitofp_i16_double'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'uitofp_i16_double'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'uitofp_i16_double'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'uitofp_i16_double'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'uitofp_i16_double'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'uitofp_i16_double'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'uitofp_i16_double'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f64 = uitofp i16 undef to double
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v2i16_v2f64 = uitofp <2 x i16> undef to <2 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f64 = uitofp <4 x i16> undef to <4 x double>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v8i16_v8f64 = uitofp <8 x i16> undef to <8 x double>
@@ -155,24 +183,40 @@ define i32 @uitofp_i64_double() {
 }
 
 define i32 @uitofp_i8_float() {
-; SSE-LABEL: 'uitofp_i8_float'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v16i8_v16f32 = uitofp <16 x i8> undef to <16 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'uitofp_i8_float'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %cvt_v16i8_v16f32 = uitofp <16 x i8> undef to <16 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'uitofp_i8_float'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v16i8_v16f32 = uitofp <16 x i8> undef to <16 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'uitofp_i8_float'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v16i8_v16f32 = uitofp <16 x i8> undef to <16 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'uitofp_i8_float'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v16i8_v16f32 = uitofp <16 x i8> undef to <16 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'uitofp_i8_float'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %cvt_v16i8_v16f32 = uitofp <16 x i8> undef to <16 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'uitofp_i8_float'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i8_f32 = uitofp i8 undef to float
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i8_v2f32 = uitofp <2 x i8> undef to <2 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i8_v4f32 = uitofp <4 x i8> undef to <4 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v8i8_v8f32 = uitofp <8 x i8> undef to <8 x float>
@@ -188,24 +232,40 @@ define i32 @uitofp_i8_float() {
 }
 
 define i32 @uitofp_i16_float() {
-; SSE-LABEL: 'uitofp_i16_float'
-; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v16i16_v16f32 = uitofp <16 x i16> undef to <16 x float>
-; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
-;
-; AVX-LABEL: 'uitofp_i16_float'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
-; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %cvt_v16i16_v16f32 = uitofp <16 x i16> undef to <16 x float>
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+; SSE2-LABEL: 'uitofp_i16_float'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v16i16_v16f32 = uitofp <16 x i16> undef to <16 x float>
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'uitofp_i16_float'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %cvt_v16i16_v16f32 = uitofp <16 x i16> undef to <16 x float>
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'uitofp_i16_float'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %cvt_v16i16_v16f32 = uitofp <16 x i16> undef to <16 x float>
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'uitofp_i16_float'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %cvt_v16i16_v16f32 = uitofp <16 x i16> undef to <16 x float>
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
 ;
 ; AVX512-LABEL: 'uitofp_i16_float'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
+; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_i16_f32 = uitofp i16 undef to float
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %cvt_v2i16_v2f32 = uitofp <2 x i16> undef to <2 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v4i16_v4f32 = uitofp <4 x i16> undef to <4 x float>
 ; AVX512-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %cvt_v8i16_v8f32 = uitofp <8 x i16> undef to <8 x float>

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast-inseltpoison.ll b/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast-inseltpoison.ll
index c5977a7..44d0f17 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast-inseltpoison.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast-inseltpoison.ll
@@ -1,10 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefixes=CHECK,SLM
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefixes=CHECK,AVX
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefixes=CHECK,AVX
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefixes=CHECK,AVX512
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefixes=CHECK,AVX512
+; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
 
 define <8 x float> @sitofp_uitofp(<8 x i32> %a) {
 ; CHECK-LABEL: @sitofp_uitofp(
@@ -197,81 +197,24 @@ define <8 x float> @sitofp_4i32_8i16(<4 x i32> %a, <8 x i16> %b) {
 
 ; Inspired by PR38154
 define <8 x float> @sitofp_uitofp_4i32_8i16_16i8(<4 x i32> %a, <8 x i16> %b, <16 x i8> %c) {
-; SSE-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; SSE-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; SSE-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; SSE-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; SSE-NEXT:    [[TMP4:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> undef, <2 x i32> <i32 0, i32 1>
-; SSE-NEXT:    [[TMP5:%.*]] = sitofp <2 x i16> [[TMP4]] to <2 x float>
-; SSE-NEXT:    [[TMP6:%.*]] = uitofp <2 x i16> [[TMP4]] to <2 x float>
-; SSE-NEXT:    [[TMP7:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP6]], <2 x i32> <i32 0, i32 3>
-; SSE-NEXT:    [[TMP8:%.*]] = shufflevector <16 x i8> [[C:%.*]], <16 x i8> undef, <2 x i32> <i32 0, i32 1>
-; SSE-NEXT:    [[TMP9:%.*]] = sitofp <2 x i8> [[TMP8]] to <2 x float>
-; SSE-NEXT:    [[TMP10:%.*]] = uitofp <2 x i8> [[TMP8]] to <2 x float>
-; SSE-NEXT:    [[TMP11:%.*]] = shufflevector <2 x float> [[TMP9]], <2 x float> [[TMP10]], <2 x i32> <i32 0, i32 3>
-; SSE-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; SSE-NEXT:    [[TMP12:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SSE-NEXT:    [[R53:%.*]] = shufflevector <8 x float> [[R31]], <8 x float> [[TMP12]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 undef, i32 undef>
-; SSE-NEXT:    [[TMP13:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SSE-NEXT:    [[R72:%.*]] = shufflevector <8 x float> [[R53]], <8 x float> [[TMP13]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
-; SSE-NEXT:    ret <8 x float> [[R72]]
-;
-; SLM-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; SLM-NEXT:    [[B0:%.*]] = extractelement <8 x i16> [[B:%.*]], i32 0
-; SLM-NEXT:    [[B1:%.*]] = extractelement <8 x i16> [[B]], i32 1
-; SLM-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; SLM-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; SLM-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; SLM-NEXT:    [[AB4:%.*]] = sitofp i16 [[B0]] to float
-; SLM-NEXT:    [[AB5:%.*]] = uitofp i16 [[B1]] to float
-; SLM-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i8> [[C:%.*]], <16 x i8> undef, <2 x i32> <i32 0, i32 1>
-; SLM-NEXT:    [[TMP5:%.*]] = sitofp <2 x i8> [[TMP4]] to <2 x float>
-; SLM-NEXT:    [[TMP6:%.*]] = uitofp <2 x i8> [[TMP4]] to <2 x float>
-; SLM-NEXT:    [[TMP7:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP6]], <2 x i32> <i32 0, i32 3>
-; SLM-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; SLM-NEXT:    [[R4:%.*]] = insertelement <8 x float> [[R31]], float [[AB4]], i32 4
-; SLM-NEXT:    [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 5
-; SLM-NEXT:    [[TMP8:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SLM-NEXT:    [[R72:%.*]] = shufflevector <8 x float> [[R5]], <8 x float> [[TMP8]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
-; SLM-NEXT:    ret <8 x float> [[R72]]
-;
-; AVX-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; AVX-NEXT:    [[B0:%.*]] = extractelement <8 x i16> [[B:%.*]], i32 0
-; AVX-NEXT:    [[B1:%.*]] = extractelement <8 x i16> [[B]], i32 1
-; AVX-NEXT:    [[C0:%.*]] = extractelement <16 x i8> [[C:%.*]], i32 0
-; AVX-NEXT:    [[C1:%.*]] = extractelement <16 x i8> [[C]], i32 1
-; AVX-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; AVX-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; AVX-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; AVX-NEXT:    [[AB4:%.*]] = sitofp i16 [[B0]] to float
-; AVX-NEXT:    [[AB5:%.*]] = uitofp i16 [[B1]] to float
-; AVX-NEXT:    [[AB6:%.*]] = sitofp i8 [[C0]] to float
-; AVX-NEXT:    [[AB7:%.*]] = uitofp i8 [[C1]] to float
-; AVX-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT:    [[R4:%.*]] = insertelement <8 x float> [[R31]], float [[AB4]], i32 4
-; AVX-NEXT:    [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 5
-; AVX-NEXT:    [[R6:%.*]] = insertelement <8 x float> [[R5]], float [[AB6]], i32 6
-; AVX-NEXT:    [[R7:%.*]] = insertelement <8 x float> [[R6]], float [[AB7]], i32 7
-; AVX-NEXT:    ret <8 x float> [[R7]]
-;
-; AVX512-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; AVX512-NEXT:    [[B0:%.*]] = extractelement <8 x i16> [[B:%.*]], i32 0
-; AVX512-NEXT:    [[B1:%.*]] = extractelement <8 x i16> [[B]], i32 1
-; AVX512-NEXT:    [[C0:%.*]] = extractelement <16 x i8> [[C:%.*]], i32 0
-; AVX512-NEXT:    [[C1:%.*]] = extractelement <16 x i8> [[C]], i32 1
-; AVX512-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; AVX512-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; AVX512-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; AVX512-NEXT:    [[AB4:%.*]] = sitofp i16 [[B0]] to float
-; AVX512-NEXT:    [[AB5:%.*]] = uitofp i16 [[B1]] to float
-; AVX512-NEXT:    [[AB6:%.*]] = sitofp i8 [[C0]] to float
-; AVX512-NEXT:    [[AB7:%.*]] = uitofp i8 [[C1]] to float
-; AVX512-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX512-NEXT:    [[R4:%.*]] = insertelement <8 x float> [[R31]], float [[AB4]], i32 4
-; AVX512-NEXT:    [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 5
-; AVX512-NEXT:    [[R6:%.*]] = insertelement <8 x float> [[R5]], float [[AB6]], i32 6
-; AVX512-NEXT:    [[R7:%.*]] = insertelement <8 x float> [[R6]], float [[AB7]], i32 7
-; AVX512-NEXT:    ret <8 x float> [[R7]]
+; CHECK-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
+; CHECK-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
+; CHECK-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> undef, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP5:%.*]] = sitofp <2 x i16> [[TMP4]] to <2 x float>
+; CHECK-NEXT:    [[TMP6:%.*]] = uitofp <2 x i16> [[TMP4]] to <2 x float>
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP6]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <16 x i8> [[C:%.*]], <16 x i8> undef, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP9:%.*]] = sitofp <2 x i8> [[TMP8]] to <2 x float>
+; CHECK-NEXT:    [[TMP10:%.*]] = uitofp <2 x i8> [[TMP8]] to <2 x float>
+; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <2 x float> [[TMP9]], <2 x float> [[TMP10]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[R53:%.*]] = shufflevector <8 x float> [[R31]], <8 x float> [[TMP12]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP13:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[R72:%.*]] = shufflevector <8 x float> [[R53]], <8 x float> [[TMP13]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
+; CHECK-NEXT:    ret <8 x float> [[R72]]
 ;
   %a0 = extractelement <4 x i32> %a, i32 0
   %a1 = extractelement <4 x i32> %a, i32 1

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast.ll b/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast.ll
index 72940f9..5b537ac 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/alternate-cast.ll
@@ -1,10 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefix=CHECK --check-prefix=SSE
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefix=CHECK --check-prefix=SLM
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512
+; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=slm -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=knl -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skx -basic-aa -slp-vectorizer -instcombine -S | FileCheck %s
 
 define <8 x float> @sitofp_uitofp(<8 x i32> %a) {
 ; CHECK-LABEL: @sitofp_uitofp(
@@ -197,81 +197,24 @@ define <8 x float> @sitofp_4i32_8i16(<4 x i32> %a, <8 x i16> %b) {
 
 ; Inspired by PR38154
 define <8 x float> @sitofp_uitofp_4i32_8i16_16i8(<4 x i32> %a, <8 x i16> %b, <16 x i8> %c) {
-; SSE-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; SSE-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; SSE-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; SSE-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; SSE-NEXT:    [[TMP4:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> undef, <2 x i32> <i32 0, i32 1>
-; SSE-NEXT:    [[TMP5:%.*]] = sitofp <2 x i16> [[TMP4]] to <2 x float>
-; SSE-NEXT:    [[TMP6:%.*]] = uitofp <2 x i16> [[TMP4]] to <2 x float>
-; SSE-NEXT:    [[TMP7:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP6]], <2 x i32> <i32 0, i32 3>
-; SSE-NEXT:    [[TMP8:%.*]] = shufflevector <16 x i8> [[C:%.*]], <16 x i8> undef, <2 x i32> <i32 0, i32 1>
-; SSE-NEXT:    [[TMP9:%.*]] = sitofp <2 x i8> [[TMP8]] to <2 x float>
-; SSE-NEXT:    [[TMP10:%.*]] = uitofp <2 x i8> [[TMP8]] to <2 x float>
-; SSE-NEXT:    [[TMP11:%.*]] = shufflevector <2 x float> [[TMP9]], <2 x float> [[TMP10]], <2 x i32> <i32 0, i32 3>
-; SSE-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; SSE-NEXT:    [[TMP12:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SSE-NEXT:    [[R53:%.*]] = shufflevector <8 x float> [[R31]], <8 x float> [[TMP12]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 undef, i32 undef>
-; SSE-NEXT:    [[TMP13:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SSE-NEXT:    [[R72:%.*]] = shufflevector <8 x float> [[R53]], <8 x float> [[TMP13]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
-; SSE-NEXT:    ret <8 x float> [[R72]]
-;
-; SLM-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; SLM-NEXT:    [[B0:%.*]] = extractelement <8 x i16> [[B:%.*]], i32 0
-; SLM-NEXT:    [[B1:%.*]] = extractelement <8 x i16> [[B]], i32 1
-; SLM-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; SLM-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; SLM-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; SLM-NEXT:    [[AB4:%.*]] = sitofp i16 [[B0]] to float
-; SLM-NEXT:    [[AB5:%.*]] = uitofp i16 [[B1]] to float
-; SLM-NEXT:    [[TMP4:%.*]] = shufflevector <16 x i8> [[C:%.*]], <16 x i8> undef, <2 x i32> <i32 0, i32 1>
-; SLM-NEXT:    [[TMP5:%.*]] = sitofp <2 x i8> [[TMP4]] to <2 x float>
-; SLM-NEXT:    [[TMP6:%.*]] = uitofp <2 x i8> [[TMP4]] to <2 x float>
-; SLM-NEXT:    [[TMP7:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP6]], <2 x i32> <i32 0, i32 3>
-; SLM-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; SLM-NEXT:    [[R4:%.*]] = insertelement <8 x float> [[R31]], float [[AB4]], i32 4
-; SLM-NEXT:    [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 5
-; SLM-NEXT:    [[TMP8:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
-; SLM-NEXT:    [[R72:%.*]] = shufflevector <8 x float> [[R5]], <8 x float> [[TMP8]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
-; SLM-NEXT:    ret <8 x float> [[R72]]
-;
-; AVX-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; AVX-NEXT:    [[B0:%.*]] = extractelement <8 x i16> [[B:%.*]], i32 0
-; AVX-NEXT:    [[B1:%.*]] = extractelement <8 x i16> [[B]], i32 1
-; AVX-NEXT:    [[C0:%.*]] = extractelement <16 x i8> [[C:%.*]], i32 0
-; AVX-NEXT:    [[C1:%.*]] = extractelement <16 x i8> [[C]], i32 1
-; AVX-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; AVX-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; AVX-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; AVX-NEXT:    [[AB4:%.*]] = sitofp i16 [[B0]] to float
-; AVX-NEXT:    [[AB5:%.*]] = uitofp i16 [[B1]] to float
-; AVX-NEXT:    [[AB6:%.*]] = sitofp i8 [[C0]] to float
-; AVX-NEXT:    [[AB7:%.*]] = uitofp i8 [[C1]] to float
-; AVX-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX-NEXT:    [[R4:%.*]] = insertelement <8 x float> [[R31]], float [[AB4]], i32 4
-; AVX-NEXT:    [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 5
-; AVX-NEXT:    [[R6:%.*]] = insertelement <8 x float> [[R5]], float [[AB6]], i32 6
-; AVX-NEXT:    [[R7:%.*]] = insertelement <8 x float> [[R6]], float [[AB7]], i32 7
-; AVX-NEXT:    ret <8 x float> [[R7]]
-;
-; AVX512-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
-; AVX512-NEXT:    [[B0:%.*]] = extractelement <8 x i16> [[B:%.*]], i32 0
-; AVX512-NEXT:    [[B1:%.*]] = extractelement <8 x i16> [[B]], i32 1
-; AVX512-NEXT:    [[C0:%.*]] = extractelement <16 x i8> [[C:%.*]], i32 0
-; AVX512-NEXT:    [[C1:%.*]] = extractelement <16 x i8> [[C]], i32 1
-; AVX512-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
-; AVX512-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
-; AVX512-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
-; AVX512-NEXT:    [[AB4:%.*]] = sitofp i16 [[B0]] to float
-; AVX512-NEXT:    [[AB5:%.*]] = uitofp i16 [[B1]] to float
-; AVX512-NEXT:    [[AB6:%.*]] = sitofp i8 [[C0]] to float
-; AVX512-NEXT:    [[AB7:%.*]] = uitofp i8 [[C1]] to float
-; AVX512-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
-; AVX512-NEXT:    [[R4:%.*]] = insertelement <8 x float> [[R31]], float [[AB4]], i32 4
-; AVX512-NEXT:    [[R5:%.*]] = insertelement <8 x float> [[R4]], float [[AB5]], i32 5
-; AVX512-NEXT:    [[R6:%.*]] = insertelement <8 x float> [[R5]], float [[AB6]], i32 6
-; AVX512-NEXT:    [[R7:%.*]] = insertelement <8 x float> [[R6]], float [[AB7]], i32 7
-; AVX512-NEXT:    ret <8 x float> [[R7]]
+; CHECK-LABEL: @sitofp_uitofp_4i32_8i16_16i8(
+; CHECK-NEXT:    [[TMP1:%.*]] = sitofp <4 x i32> [[A:%.*]] to <4 x float>
+; CHECK-NEXT:    [[TMP2:%.*]] = uitofp <4 x i32> [[A]] to <4 x float>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <8 x i16> [[B:%.*]], <8 x i16> undef, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP5:%.*]] = sitofp <2 x i16> [[TMP4]] to <2 x float>
+; CHECK-NEXT:    [[TMP6:%.*]] = uitofp <2 x i16> [[TMP4]] to <2 x float>
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP6]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT:    [[TMP8:%.*]] = shufflevector <16 x i8> [[C:%.*]], <16 x i8> undef, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP9:%.*]] = sitofp <2 x i8> [[TMP8]] to <2 x float>
+; CHECK-NEXT:    [[TMP10:%.*]] = uitofp <2 x i8> [[TMP8]] to <2 x float>
+; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <2 x float> [[TMP9]], <2 x float> [[TMP10]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT:    [[R31:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[R53:%.*]] = shufflevector <8 x float> [[R31]], <8 x float> [[TMP12]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP13:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> poison, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[R72:%.*]] = shufflevector <8 x float> [[R53]], <8 x float> [[TMP13]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
+; CHECK-NEXT:    ret <8 x float> [[R72]]
 ;
   %a0 = extractelement <4 x i32> %a, i32 0
   %a1 = extractelement <4 x i32> %a, i32 1

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/sitofp-inseltpoison.ll b/llvm/test/Transforms/SLPVectorizer/X86/sitofp-inseltpoison.ll
index b230a21..c0b1bae 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/sitofp-inseltpoison.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/sitofp-inseltpoison.ll
@@ -321,14 +321,20 @@ define void @sitofp_8i32_8f64() #0 {
 }
 
 define void @sitofp_2i16_2f64() #0 {
-; CHECK-LABEL: @sitofp_2i16_2f64(
-; CHECK-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; CHECK-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; CHECK-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
-; CHECK-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
-; CHECK-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; CHECK-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; CHECK-NEXT:    ret void
+; SSE-LABEL: @sitofp_2i16_2f64(
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = sitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    ret void
+;
+; AVX-LABEL: @sitofp_2i16_2f64(
+; AVX-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
+; AVX-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
+; AVX-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
+; AVX-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
+; AVX-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX-NEXT:    ret void
 ;
   %ld0 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
   %ld1 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
@@ -341,18 +347,12 @@ define void @sitofp_2i16_2f64() #0 {
 
 define void @sitofp_4i16_4f64() #0 {
 ; SSE-LABEL: @sitofp_4i16_4f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; SSE-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2), align 4
-; SSE-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 3), align 2
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i16 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i16 [[LD3]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP3:%.*]] = sitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP4:%.*]] = sitofp <2 x i16> [[TMP2]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @sitofp_4i16_4f64(
@@ -378,30 +378,18 @@ define void @sitofp_4i16_4f64() #0 {
 
 define void @sitofp_8i16_8f64() #0 {
 ; SSE-LABEL: @sitofp_8i16_8f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; SSE-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2), align 4
-; SSE-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 3), align 2
-; SSE-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 4), align 8
-; SSE-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 5), align 2
-; SSE-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 6), align 4
-; SSE-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 7), align 2
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i16 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i16 [[LD3]] to double
-; SSE-NEXT:    [[CVT4:%.*]] = sitofp i16 [[LD4]] to double
-; SSE-NEXT:    [[CVT5:%.*]] = sitofp i16 [[LD5]] to double
-; SSE-NEXT:    [[CVT6:%.*]] = sitofp i16 [[LD6]] to double
-; SSE-NEXT:    [[CVT7:%.*]] = sitofp i16 [[LD7]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
-; SSE-NEXT:    store double [[CVT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 32
-; SSE-NEXT:    store double [[CVT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
-; SSE-NEXT:    store double [[CVT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 16
-; SSE-NEXT:    store double [[CVT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP3:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 4) to <2 x i16>*), align 8
+; SSE-NEXT:    [[TMP4:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 6) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP5:%.*]] = sitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP6:%.*]] = sitofp <2 x i16> [[TMP2]] to <2 x double>
+; SSE-NEXT:    [[TMP7:%.*]] = sitofp <2 x i16> [[TMP3]] to <2 x double>
+; SSE-NEXT:    [[TMP8:%.*]] = sitofp <2 x i16> [[TMP4]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
+; SSE-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 32
+; SSE-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX256-LABEL: @sitofp_8i16_8f64(
@@ -447,14 +435,20 @@ define void @sitofp_8i16_8f64() #0 {
 }
 
 define void @sitofp_2i8_2f64() #0 {
-; CHECK-LABEL: @sitofp_2i8_2f64(
-; CHECK-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; CHECK-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; CHECK-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
-; CHECK-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
-; CHECK-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; CHECK-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; CHECK-NEXT:    ret void
+; SSE-LABEL: @sitofp_2i8_2f64(
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = sitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    ret void
+;
+; AVX-LABEL: @sitofp_2i8_2f64(
+; AVX-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
+; AVX-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
+; AVX-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
+; AVX-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
+; AVX-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX-NEXT:    ret void
 ;
   %ld0 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
   %ld1 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
@@ -467,18 +461,12 @@ define void @sitofp_2i8_2f64() #0 {
 
 define void @sitofp_4i8_4f64() #0 {
 ; SSE-LABEL: @sitofp_4i8_4f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2), align 2
-; SSE-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 3), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i8 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i8 [[LD3]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = sitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP4:%.*]] = sitofp <2 x i8> [[TMP2]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @sitofp_4i8_4f64(
@@ -504,30 +492,18 @@ define void @sitofp_4i8_4f64() #0 {
 
 define void @sitofp_8i8_8f64() #0 {
 ; SSE-LABEL: @sitofp_8i8_8f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2), align 2
-; SSE-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 3), align 1
-; SSE-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 4), align 4
-; SSE-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 5), align 1
-; SSE-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 6), align 2
-; SSE-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 7), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i8 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i8 [[LD3]] to double
-; SSE-NEXT:    [[CVT4:%.*]] = sitofp i8 [[LD4]] to double
-; SSE-NEXT:    [[CVT5:%.*]] = sitofp i8 [[LD5]] to double
-; SSE-NEXT:    [[CVT6:%.*]] = sitofp i8 [[LD6]] to double
-; SSE-NEXT:    [[CVT7:%.*]] = sitofp i8 [[LD7]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
-; SSE-NEXT:    store double [[CVT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 32
-; SSE-NEXT:    store double [[CVT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
-; SSE-NEXT:    store double [[CVT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 16
-; SSE-NEXT:    store double [[CVT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 4) to <2 x i8>*), align 4
+; SSE-NEXT:    [[TMP4:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 6) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP5:%.*]] = sitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP6:%.*]] = sitofp <2 x i8> [[TMP2]] to <2 x double>
+; SSE-NEXT:    [[TMP7:%.*]] = sitofp <2 x i8> [[TMP3]] to <2 x double>
+; SSE-NEXT:    [[TMP8:%.*]] = sitofp <2 x i8> [[TMP4]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
+; SSE-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 32
+; SSE-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX256-LABEL: @sitofp_8i8_8f64(

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/sitofp.ll b/llvm/test/Transforms/SLPVectorizer/X86/sitofp.ll
index ffa821a..f0cb3e5 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/sitofp.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/sitofp.ll
@@ -321,14 +321,20 @@ define void @sitofp_8i32_8f64() #0 {
 }
 
 define void @sitofp_2i16_2f64() #0 {
-; CHECK-LABEL: @sitofp_2i16_2f64(
-; CHECK-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; CHECK-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; CHECK-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
-; CHECK-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
-; CHECK-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; CHECK-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; CHECK-NEXT:    ret void
+; SSE-LABEL: @sitofp_2i16_2f64(
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = sitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    ret void
+;
+; AVX-LABEL: @sitofp_2i16_2f64(
+; AVX-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
+; AVX-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
+; AVX-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
+; AVX-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
+; AVX-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX-NEXT:    ret void
 ;
   %ld0 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
   %ld1 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
@@ -341,18 +347,12 @@ define void @sitofp_2i16_2f64() #0 {
 
 define void @sitofp_4i16_4f64() #0 {
 ; SSE-LABEL: @sitofp_4i16_4f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; SSE-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2), align 4
-; SSE-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 3), align 2
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i16 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i16 [[LD3]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP3:%.*]] = sitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP4:%.*]] = sitofp <2 x i16> [[TMP2]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @sitofp_4i16_4f64(
@@ -378,30 +378,18 @@ define void @sitofp_4i16_4f64() #0 {
 
 define void @sitofp_8i16_8f64() #0 {
 ; SSE-LABEL: @sitofp_8i16_8f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; SSE-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2), align 4
-; SSE-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 3), align 2
-; SSE-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 4), align 8
-; SSE-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 5), align 2
-; SSE-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 6), align 4
-; SSE-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 7), align 2
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i16 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i16 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i16 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i16 [[LD3]] to double
-; SSE-NEXT:    [[CVT4:%.*]] = sitofp i16 [[LD4]] to double
-; SSE-NEXT:    [[CVT5:%.*]] = sitofp i16 [[LD5]] to double
-; SSE-NEXT:    [[CVT6:%.*]] = sitofp i16 [[LD6]] to double
-; SSE-NEXT:    [[CVT7:%.*]] = sitofp i16 [[LD7]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
-; SSE-NEXT:    store double [[CVT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 32
-; SSE-NEXT:    store double [[CVT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
-; SSE-NEXT:    store double [[CVT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 16
-; SSE-NEXT:    store double [[CVT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP3:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 4) to <2 x i16>*), align 8
+; SSE-NEXT:    [[TMP4:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 6) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP5:%.*]] = sitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP6:%.*]] = sitofp <2 x i16> [[TMP2]] to <2 x double>
+; SSE-NEXT:    [[TMP7:%.*]] = sitofp <2 x i16> [[TMP3]] to <2 x double>
+; SSE-NEXT:    [[TMP8:%.*]] = sitofp <2 x i16> [[TMP4]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
+; SSE-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 32
+; SSE-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX256-LABEL: @sitofp_8i16_8f64(
@@ -447,14 +435,20 @@ define void @sitofp_8i16_8f64() #0 {
 }
 
 define void @sitofp_2i8_2f64() #0 {
-; CHECK-LABEL: @sitofp_2i8_2f64(
-; CHECK-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; CHECK-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; CHECK-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
-; CHECK-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
-; CHECK-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; CHECK-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; CHECK-NEXT:    ret void
+; SSE-LABEL: @sitofp_2i8_2f64(
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = sitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    ret void
+;
+; AVX-LABEL: @sitofp_2i8_2f64(
+; AVX-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
+; AVX-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
+; AVX-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
+; AVX-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
+; AVX-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX-NEXT:    ret void
 ;
   %ld0 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
   %ld1 = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
@@ -467,18 +461,12 @@ define void @sitofp_2i8_2f64() #0 {
 
 define void @sitofp_4i8_4f64() #0 {
 ; SSE-LABEL: @sitofp_4i8_4f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2), align 2
-; SSE-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 3), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i8 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i8 [[LD3]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = sitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP4:%.*]] = sitofp <2 x i8> [[TMP2]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @sitofp_4i8_4f64(
@@ -504,30 +492,18 @@ define void @sitofp_4i8_4f64() #0 {
 
 define void @sitofp_8i8_8f64() #0 {
 ; SSE-LABEL: @sitofp_8i8_8f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2), align 2
-; SSE-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 3), align 1
-; SSE-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 4), align 4
-; SSE-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 5), align 1
-; SSE-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 6), align 2
-; SSE-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 7), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = sitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = sitofp i8 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = sitofp i8 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = sitofp i8 [[LD3]] to double
-; SSE-NEXT:    [[CVT4:%.*]] = sitofp i8 [[LD4]] to double
-; SSE-NEXT:    [[CVT5:%.*]] = sitofp i8 [[LD5]] to double
-; SSE-NEXT:    [[CVT6:%.*]] = sitofp i8 [[LD6]] to double
-; SSE-NEXT:    [[CVT7:%.*]] = sitofp i8 [[LD7]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
-; SSE-NEXT:    store double [[CVT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 32
-; SSE-NEXT:    store double [[CVT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
-; SSE-NEXT:    store double [[CVT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 16
-; SSE-NEXT:    store double [[CVT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 4) to <2 x i8>*), align 4
+; SSE-NEXT:    [[TMP4:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 6) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP5:%.*]] = sitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP6:%.*]] = sitofp <2 x i8> [[TMP2]] to <2 x double>
+; SSE-NEXT:    [[TMP7:%.*]] = sitofp <2 x i8> [[TMP3]] to <2 x double>
+; SSE-NEXT:    [[TMP8:%.*]] = sitofp <2 x i8> [[TMP4]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
+; SSE-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 32
+; SSE-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX256-LABEL: @sitofp_8i8_8f64(

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/uitofp.ll b/llvm/test/Transforms/SLPVectorizer/X86/uitofp.ll
index b5d10d9..67feab3 100644 (file)
--- a/llvm/test/Transforms/SLPVectorizer/X86/uitofp.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/uitofp.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s -mtriple=x86_64-unknown -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX256NODQ,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=bdver1 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX256NODQ,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX256NODQ,AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=bdver1 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX2
 ; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=-prefer-256-bit -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX512
 ; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=+prefer-256-bit -basic-aa -slp-vectorizer -S | FileCheck %s --check-prefixes=CHECK,AVX,AVX256,AVX256DQ
 
@@ -258,14 +258,20 @@ define void @uitofp_8i32_8f64() #0 {
 }
 
 define void @uitofp_2i16_2f64() #0 {
-; CHECK-LABEL: @uitofp_2i16_2f64(
-; CHECK-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; CHECK-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; CHECK-NEXT:    [[CVT0:%.*]] = uitofp i16 [[LD0]] to double
-; CHECK-NEXT:    [[CVT1:%.*]] = uitofp i16 [[LD1]] to double
-; CHECK-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; CHECK-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; CHECK-NEXT:    ret void
+; SSE-LABEL: @uitofp_2i16_2f64(
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = uitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    ret void
+;
+; AVX-LABEL: @uitofp_2i16_2f64(
+; AVX-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
+; AVX-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
+; AVX-NEXT:    [[CVT0:%.*]] = uitofp i16 [[LD0]] to double
+; AVX-NEXT:    [[CVT1:%.*]] = uitofp i16 [[LD1]] to double
+; AVX-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX-NEXT:    ret void
 ;
   %ld0 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
   %ld1 = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
@@ -278,18 +284,12 @@ define void @uitofp_2i16_2f64() #0 {
 
 define void @uitofp_4i16_4f64() #0 {
 ; SSE-LABEL: @uitofp_4i16_4f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; SSE-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2), align 4
-; SSE-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 3), align 2
-; SSE-NEXT:    [[CVT0:%.*]] = uitofp i16 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = uitofp i16 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = uitofp i16 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = uitofp i16 [[LD3]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP3:%.*]] = uitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP4:%.*]] = uitofp <2 x i16> [[TMP2]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @uitofp_4i16_4f64(
@@ -315,30 +315,18 @@ define void @uitofp_4i16_4f64() #0 {
 
 define void @uitofp_8i16_8f64() #0 {
 ; SSE-LABEL: @uitofp_8i16_8f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 1), align 2
-; SSE-NEXT:    [[LD2:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2), align 4
-; SSE-NEXT:    [[LD3:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 3), align 2
-; SSE-NEXT:    [[LD4:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 4), align 8
-; SSE-NEXT:    [[LD5:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 5), align 2
-; SSE-NEXT:    [[LD6:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 6), align 4
-; SSE-NEXT:    [[LD7:%.*]] = load i16, i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 7), align 2
-; SSE-NEXT:    [[CVT0:%.*]] = uitofp i16 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = uitofp i16 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = uitofp i16 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = uitofp i16 [[LD3]] to double
-; SSE-NEXT:    [[CVT4:%.*]] = uitofp i16 [[LD4]] to double
-; SSE-NEXT:    [[CVT5:%.*]] = uitofp i16 [[LD5]] to double
-; SSE-NEXT:    [[CVT6:%.*]] = uitofp i16 [[LD6]] to double
-; SSE-NEXT:    [[CVT7:%.*]] = uitofp i16 [[LD7]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
-; SSE-NEXT:    store double [[CVT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 32
-; SSE-NEXT:    store double [[CVT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
-; SSE-NEXT:    store double [[CVT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 16
-; SSE-NEXT:    store double [[CVT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i16>, <2 x i16>* bitcast ([32 x i16]* @src16 to <2 x i16>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 2) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP3:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 4) to <2 x i16>*), align 8
+; SSE-NEXT:    [[TMP4:%.*]] = load <2 x i16>, <2 x i16>* bitcast (i16* getelementptr inbounds ([32 x i16], [32 x i16]* @src16, i32 0, i64 6) to <2 x i16>*), align 4
+; SSE-NEXT:    [[TMP5:%.*]] = uitofp <2 x i16> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP6:%.*]] = uitofp <2 x i16> [[TMP2]] to <2 x double>
+; SSE-NEXT:    [[TMP7:%.*]] = uitofp <2 x i16> [[TMP3]] to <2 x double>
+; SSE-NEXT:    [[TMP8:%.*]] = uitofp <2 x i16> [[TMP4]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
+; SSE-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 32
+; SSE-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX256-LABEL: @uitofp_8i16_8f64(
@@ -385,22 +373,28 @@ define void @uitofp_8i16_8f64() #0 {
 
 define void @uitofp_2i8_2f64() #0 {
 ; SSE-LABEL: @uitofp_2i8_2f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = uitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = uitofp i8 [[LD1]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = uitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP2]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
 ; SSE-NEXT:    ret void
 ;
-; AVX256NODQ-LABEL: @uitofp_2i8_2f64(
-; AVX256NODQ-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; AVX256NODQ-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; AVX256NODQ-NEXT:    [[CVT0:%.*]] = uitofp i8 [[LD0]] to double
-; AVX256NODQ-NEXT:    [[CVT1:%.*]] = uitofp i8 [[LD1]] to double
-; AVX256NODQ-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; AVX256NODQ-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; AVX256NODQ-NEXT:    ret void
+; AVX1-LABEL: @uitofp_2i8_2f64(
+; AVX1-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
+; AVX1-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
+; AVX1-NEXT:    [[CVT0:%.*]] = uitofp i8 [[LD0]] to double
+; AVX1-NEXT:    [[CVT1:%.*]] = uitofp i8 [[LD1]] to double
+; AVX1-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX1-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX1-NEXT:    ret void
+;
+; AVX2-LABEL: @uitofp_2i8_2f64(
+; AVX2-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
+; AVX2-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
+; AVX2-NEXT:    [[CVT0:%.*]] = uitofp i8 [[LD0]] to double
+; AVX2-NEXT:    [[CVT1:%.*]] = uitofp i8 [[LD1]] to double
+; AVX2-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
+; AVX2-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
+; AVX2-NEXT:    ret void
 ;
 ; AVX512-LABEL: @uitofp_2i8_2f64(
 ; AVX512-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
@@ -425,18 +419,12 @@ define void @uitofp_2i8_2f64() #0 {
 
 define void @uitofp_4i8_4f64() #0 {
 ; SSE-LABEL: @uitofp_4i8_4f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2), align 2
-; SSE-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 3), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = uitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = uitofp i8 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = uitofp i8 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = uitofp i8 [[LD3]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = uitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP4:%.*]] = uitofp <2 x i8> [[TMP2]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP3]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP4]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX-LABEL: @uitofp_4i8_4f64(
@@ -462,30 +450,18 @@ define void @uitofp_4i8_4f64() #0 {
 
 define void @uitofp_8i8_8f64() #0 {
 ; SSE-LABEL: @uitofp_8i8_8f64(
-; SSE-NEXT:    [[LD0:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 0), align 64
-; SSE-NEXT:    [[LD1:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 1), align 1
-; SSE-NEXT:    [[LD2:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2), align 2
-; SSE-NEXT:    [[LD3:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 3), align 1
-; SSE-NEXT:    [[LD4:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 4), align 4
-; SSE-NEXT:    [[LD5:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 5), align 1
-; SSE-NEXT:    [[LD6:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 6), align 2
-; SSE-NEXT:    [[LD7:%.*]] = load i8, i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 7), align 1
-; SSE-NEXT:    [[CVT0:%.*]] = uitofp i8 [[LD0]] to double
-; SSE-NEXT:    [[CVT1:%.*]] = uitofp i8 [[LD1]] to double
-; SSE-NEXT:    [[CVT2:%.*]] = uitofp i8 [[LD2]] to double
-; SSE-NEXT:    [[CVT3:%.*]] = uitofp i8 [[LD3]] to double
-; SSE-NEXT:    [[CVT4:%.*]] = uitofp i8 [[LD4]] to double
-; SSE-NEXT:    [[CVT5:%.*]] = uitofp i8 [[LD5]] to double
-; SSE-NEXT:    [[CVT6:%.*]] = uitofp i8 [[LD6]] to double
-; SSE-NEXT:    [[CVT7:%.*]] = uitofp i8 [[LD7]] to double
-; SSE-NEXT:    store double [[CVT0]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 0), align 64
-; SSE-NEXT:    store double [[CVT1]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 1), align 8
-; SSE-NEXT:    store double [[CVT2]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2), align 16
-; SSE-NEXT:    store double [[CVT3]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 3), align 8
-; SSE-NEXT:    store double [[CVT4]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4), align 32
-; SSE-NEXT:    store double [[CVT5]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 5), align 8
-; SSE-NEXT:    store double [[CVT6]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6), align 16
-; SSE-NEXT:    store double [[CVT7]], double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 7), align 8
+; SSE-NEXT:    [[TMP1:%.*]] = load <2 x i8>, <2 x i8>* bitcast ([64 x i8]* @src8 to <2 x i8>*), align 64
+; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 2) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP3:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 4) to <2 x i8>*), align 4
+; SSE-NEXT:    [[TMP4:%.*]] = load <2 x i8>, <2 x i8>* bitcast (i8* getelementptr inbounds ([64 x i8], [64 x i8]* @src8, i32 0, i64 6) to <2 x i8>*), align 2
+; SSE-NEXT:    [[TMP5:%.*]] = uitofp <2 x i8> [[TMP1]] to <2 x double>
+; SSE-NEXT:    [[TMP6:%.*]] = uitofp <2 x i8> [[TMP2]] to <2 x double>
+; SSE-NEXT:    [[TMP7:%.*]] = uitofp <2 x i8> [[TMP3]] to <2 x double>
+; SSE-NEXT:    [[TMP8:%.*]] = uitofp <2 x i8> [[TMP4]] to <2 x double>
+; SSE-NEXT:    store <2 x double> [[TMP5]], <2 x double>* bitcast ([8 x double]* @dst64 to <2 x double>*), align 64
+; SSE-NEXT:    store <2 x double> [[TMP6]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 2) to <2 x double>*), align 16
+; SSE-NEXT:    store <2 x double> [[TMP7]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 4) to <2 x double>*), align 32
+; SSE-NEXT:    store <2 x double> [[TMP8]], <2 x double>* bitcast (double* getelementptr inbounds ([8 x double], [8 x double]* @dst64, i32 0, i64 6) to <2 x double>*), align 16
 ; SSE-NEXT:    ret void
 ;
 ; AVX256-LABEL: @uitofp_8i8_8f64(