[CostModel][X86] Add more thorough BSWAP test coverage

Replace the existing tests with more complete coverage for all cost model kinds

diff --git a/llvm/test/Analysis/CostModel/X86/bswap-codesize.ll b/llvm/test/Analysis/CostModel/X86/bswap-codesize.ll
new file mode 100644 (file)
index 0000000..2af8fc4
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/bswap-codesize.ll
@@ -0,0 +1,221 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
+
+;
+; bswap(X)
+;
+
+define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
+; SSE2-LABEL: 'cost_bswap_i64'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i64'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i64'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i64'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i64'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i64'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i64'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i64'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I64    = call i64 @llvm.bswap.i64(i64 %a64)
+  %V2I64  = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+  %V4I64  = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+  %V8I64  = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+  ret void
+}
+
+define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
+; SSE2-LABEL: 'cost_bswap_i32'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i32'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i32'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i32'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i32'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i32'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i32'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i32'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I32   = call i32 @llvm.bswap.i32(i32 %a32)
+  %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+  %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+  %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+  ret void
+}
+
+define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
+; SSE2-LABEL: 'cost_bswap_i16'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i16'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i16'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i16'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i16'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i16'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i16'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i16'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 102 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I16    = call i16 @llvm.bswap.i16(i16 %a16)
+  %V8I16  = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+  %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+  %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+  ret void
+}
+
+declare i64 @llvm.bswap.i64(i64)
+declare i32 @llvm.bswap.i32(i32)
+declare i16 @llvm.bswap.i16(i16)
+
+declare <2 x i64>  @llvm.bswap.v2i64(<2 x i64>)
+declare <4 x i32>  @llvm.bswap.v4i32(<4 x i32>)
+declare <8 x i16>  @llvm.bswap.v8i16(<8 x i16>)
+
+declare <4 x i64>  @llvm.bswap.v4i64(<4 x i64>)
+declare <8 x i32>  @llvm.bswap.v8i32(<8 x i32>)
+declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
+
+declare <8 x i64>  @llvm.bswap.v8i64(<8 x i64>)
+declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
+declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)

diff --git a/llvm/test/Analysis/CostModel/X86/bswap-latency.ll b/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
new file mode 100644 (file)
index 0000000..00c36d8
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
@@ -0,0 +1,221 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
+
+;
+; bswap(X)
+;
+
+define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
+; SSE2-LABEL: 'cost_bswap_i64'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i64'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i64'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i64'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i64'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i64'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i64'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i64'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I64    = call i64 @llvm.bswap.i64(i64 %a64)
+  %V2I64  = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+  %V4I64  = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+  %V8I64  = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+  ret void
+}
+
+define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
+; SSE2-LABEL: 'cost_bswap_i32'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i32'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i32'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i32'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i32'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i32'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i32'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i32'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I32   = call i32 @llvm.bswap.i32(i32 %a32)
+  %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+  %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+  %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+  ret void
+}
+
+define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
+; SSE2-LABEL: 'cost_bswap_i16'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i16'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i16'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i16'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 388 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i16'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 388 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i16'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i16'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i16'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I16    = call i16 @llvm.bswap.i16(i16 %a16)
+  %V8I16  = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+  %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+  %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+  ret void
+}
+
+declare i64 @llvm.bswap.i64(i64)
+declare i32 @llvm.bswap.i32(i32)
+declare i16 @llvm.bswap.i16(i16)
+
+declare <2 x i64>  @llvm.bswap.v2i64(<2 x i64>)
+declare <4 x i32>  @llvm.bswap.v4i32(<4 x i32>)
+declare <8 x i16>  @llvm.bswap.v8i16(<8 x i16>)
+
+declare <4 x i64>  @llvm.bswap.v4i64(<4 x i64>)
+declare <8 x i32>  @llvm.bswap.v8i32(<8 x i32>)
+declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
+
+declare <8 x i64>  @llvm.bswap.v8i64(<8 x i64>)
+declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
+declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)

diff --git a/llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll
new file mode 100644 (file)
index 0000000..052483b
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/bswap-sizelatency.ll
@@ -0,0 +1,221 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
+
+;
+; bswap(X)
+;
+
+define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
+; SSE2-LABEL: 'cost_bswap_i64'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i64'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i64'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i64'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i64'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i64'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i64'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i64'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I64    = call i64 @llvm.bswap.i64(i64 %a64)
+  %V2I64  = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+  %V4I64  = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+  %V8I64  = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+  ret void
+}
+
+define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
+; SSE2-LABEL: 'cost_bswap_i32'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i32'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i32'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i32'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i32'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 52 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i32'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i32'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i32'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 26 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 54 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I32   = call i32 @llvm.bswap.i32(i32 %a32)
+  %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+  %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+  %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+  ret void
+}
+
+define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
+; SSE2-LABEL: 'cost_bswap_i16'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i16'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i16'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 192 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 384 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i16'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 388 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i16'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 388 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i16'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i16'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i16'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 96 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 194 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 390 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
+;
+  %I16    = call i16 @llvm.bswap.i16(i16 %a16)
+  %V8I16  = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+  %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+  %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+  ret void
+}
+
+declare i64 @llvm.bswap.i64(i64)
+declare i32 @llvm.bswap.i32(i32)
+declare i16 @llvm.bswap.i16(i16)
+
+declare <2 x i64>  @llvm.bswap.v2i64(<2 x i64>)
+declare <4 x i32>  @llvm.bswap.v4i32(<4 x i32>)
+declare <8 x i16>  @llvm.bswap.v8i16(<8 x i16>)
+
+declare <4 x i64>  @llvm.bswap.v4i64(<4 x i64>)
+declare <8 x i32>  @llvm.bswap.v8i32(<8 x i32>)
+declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
+
+declare <8 x i64>  @llvm.bswap.v8i64(<8 x i64>)
+declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
+declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)

diff --git a/llvm/test/Analysis/CostModel/X86/bswap-vec.ll b/llvm/test/Analysis/CostModel/X86/bswap-vec.ll
deleted file mode 100644 (file)
index b153ac7..0000000
--- a/llvm/test/Analysis/CostModel/X86/bswap-vec.ll
+++ /dev/null
@@ -1,236 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 | FileCheck %s -check-prefixes=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s -check-prefixes=SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx | FileCheck %s -check-prefixes=AVX,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s -check-prefixes=AVX,AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+xop,+avx | FileCheck %s -check-prefixes=AVX,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+xop,+avx2 | FileCheck %s -check-prefixes=AVX,AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s -check-prefixes=AVX,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512bw | FileCheck %s -check-prefixes=AVX,AVX512,AVX512BW
-
-; Verify the cost of vector bswap instructions.
-
-declare <2 x i64> @llvm.bswap.v2i64(<2 x i64>)
-declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>)
-declare <8 x i16> @llvm.bswap.v8i16(<8 x i16>)
-
-declare <4 x i64> @llvm.bswap.v4i64(<4 x i64>)
-declare <8 x i32> @llvm.bswap.v8i32(<8 x i32>)
-declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
-
-declare <8 x i64> @llvm.bswap.v8i64(<8 x i64>)
-declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
-declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)
-
-define <2 x i64> @var_bswap_v2i64(<2 x i64> %a) {
-; SSE2-LABEL: 'var_bswap_v2i64'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %bswap = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v2i64'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %bswap
-;
-; AVX-LABEL: 'var_bswap_v2i64'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %bswap
-;
-  %bswap = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a)
-  ret <2 x i64> %bswap
-}
-
-define <4 x i64> @var_bswap_v4i64(<4 x i64> %a) {
-; SSE2-LABEL: 'var_bswap_v4i64'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v4i64'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bswap
-;
-; AVX1-LABEL: 'var_bswap_v4i64'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bswap
-;
-; AVX2-LABEL: 'var_bswap_v4i64'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bswap
-;
-; AVX512-LABEL: 'var_bswap_v4i64'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bswap
-;
-  %bswap = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a)
-  ret <4 x i64> %bswap
-}
-
-define <8 x i64> @var_bswap_v8i64(<8 x i64> %a) {
-; SSE2-LABEL: 'var_bswap_v8i64'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v8i64'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bswap
-;
-; AVX1-LABEL: 'var_bswap_v8i64'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bswap
-;
-; AVX2-LABEL: 'var_bswap_v8i64'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bswap
-;
-; AVX512F-LABEL: 'var_bswap_v8i64'
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bswap
-;
-; AVX512BW-LABEL: 'var_bswap_v8i64'
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bswap
-;
-  %bswap = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a)
-  ret <8 x i64> %bswap
-}
-
-define <4 x i32> @var_bswap_v4i32(<4 x i32> %a) {
-; SSE2-LABEL: 'var_bswap_v4i32'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %bswap = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v4i32'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %bswap
-;
-; AVX-LABEL: 'var_bswap_v4i32'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %bswap
-;
-  %bswap = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a)
-  ret <4 x i32> %bswap
-}
-
-define <8 x i32> @var_bswap_v8i32(<8 x i32> %a) {
-; SSE2-LABEL: 'var_bswap_v8i32'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v8i32'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bswap
-;
-; AVX1-LABEL: 'var_bswap_v8i32'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bswap
-;
-; AVX2-LABEL: 'var_bswap_v8i32'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bswap
-;
-; AVX512-LABEL: 'var_bswap_v8i32'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bswap
-;
-  %bswap = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a)
-  ret <8 x i32> %bswap
-}
-
-define <16 x i32> @var_bswap_v16i32(<16 x i32> %a) {
-; SSE2-LABEL: 'var_bswap_v16i32'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v16i32'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bswap
-;
-; AVX1-LABEL: 'var_bswap_v16i32'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bswap
-;
-; AVX2-LABEL: 'var_bswap_v16i32'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bswap
-;
-; AVX512F-LABEL: 'var_bswap_v16i32'
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bswap
-;
-; AVX512BW-LABEL: 'var_bswap_v16i32'
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bswap
-;
-  %bswap = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a)
-  ret <16 x i32> %bswap
-}
-
-define <8 x i16> @var_bswap_v8i16(<8 x i16> %a) {
-; SSE2-LABEL: 'var_bswap_v8i16'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %bswap = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v8i16'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bswap
-;
-; AVX-LABEL: 'var_bswap_v8i16'
-; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bswap
-;
-  %bswap = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a)
-  ret <8 x i16> %bswap
-}
-
-define <16 x i16> @var_bswap_v16i16(<16 x i16> %a) {
-; SSE2-LABEL: 'var_bswap_v16i16'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v16i16'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bswap
-;
-; AVX1-LABEL: 'var_bswap_v16i16'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bswap
-;
-; AVX2-LABEL: 'var_bswap_v16i16'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bswap
-;
-; AVX512-LABEL: 'var_bswap_v16i16'
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
-; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bswap
-;
-  %bswap = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a)
-  ret <16 x i16> %bswap
-}
-
-define <32 x i16> @var_bswap_v32i16(<32 x i16> %a) {
-; SSE2-LABEL: 'var_bswap_v32i16'
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bswap
-;
-; SSE42-LABEL: 'var_bswap_v32i16'
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bswap
-;
-; AVX1-LABEL: 'var_bswap_v32i16'
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bswap
-;
-; AVX2-LABEL: 'var_bswap_v32i16'
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bswap
-;
-; AVX512F-LABEL: 'var_bswap_v32i16'
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bswap
-;
-; AVX512BW-LABEL: 'var_bswap_v32i16'
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bswap
-;
-  %bswap = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a)
-  ret <32 x i16> %bswap
-}

diff --git a/llvm/test/Analysis/CostModel/X86/bswap.ll b/llvm/test/Analysis/CostModel/X86/bswap.ll
index 54afc1b1944a28c70fcae131350ffa1e56cc1787..3458a1d7bdb2db9939f690020566610e34297bee 100644 (file)
--- a/llvm/test/Analysis/CostModel/X86/bswap.ll
+++ b/llvm/test/Analysis/CostModel/X86/bswap.ll
@@ -1,56 +1,221 @@
 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output | FileCheck %s --check-prefixes=ALL,X64
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+movbe | FileCheck %s --check-prefixes=ALL,X64
-; RUN: opt < %s -mtriple=i686-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output | FileCheck %s --check-prefixes=ALL,X86
-; RUN: opt < %s -mtriple=i686-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+movbe | FileCheck %s --check-prefixes=ALL,X86
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s --check-prefixes=SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512vl,+avx512dq | FileCheck %s --check-prefixes=AVX512DQ
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512vl,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
 
-declare i16 @llvm.bswap.i16(i16)
-declare i32 @llvm.bswap.i32(i32)
-declare i64 @llvm.bswap.i64(i64)
-declare i128 @llvm.bswap.i128(i128)
-
-; Verify the cost of scalar bswap instructions.
-
-define i16 @var_bswap_i16(i16 %a) {
-; ALL-LABEL: 'var_bswap_i16'
-; ALL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call i16 @llvm.bswap.i16(i16 %a)
-; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i16 %bswap
 ;
-  %bswap = call i16 @llvm.bswap.i16(i16 %a)
-  ret i16 %bswap
-}
+; bswap(X)
+;
 
-define i32 @var_bswap_i32(i32 %a) {
-; ALL-LABEL: 'var_bswap_i32'
-; ALL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call i32 @llvm.bswap.i32(i32 %a)
-; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 %bswap
+define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
+; SSE2-LABEL: 'cost_bswap_i64'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i64'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i64'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-  %bswap = call i32 @llvm.bswap.i32(i32 %a)
-  ret i32 %bswap
+; AVX1-LABEL: 'cost_bswap_i64'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i64'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i64'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i64'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i64'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+  %I64    = call i64 @llvm.bswap.i64(i64 %a64)
+  %V2I64  = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+  %V4I64  = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+  %V8I64  = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+  ret void
 }
 
-define i64 @var_bswap_i64(i64 %a) {
-; X64-LABEL: 'var_bswap_i64'
-; X64-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bswap = call i64 @llvm.bswap.i64(i64 %a)
-; X64-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %bswap
+define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
+; SSE2-LABEL: 'cost_bswap_i32'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i32'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i32'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i32'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-; X86-LABEL: 'var_bswap_i64'
-; X86-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call i64 @llvm.bswap.i64(i64 %a)
-; X86-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i64 %bswap
+; AVX2-LABEL: 'cost_bswap_i32'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-  %bswap = call i64 @llvm.bswap.i64(i64 %a)
-  ret i64 %bswap
+; AVX512F-LABEL: 'cost_bswap_i32'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512DQ-LABEL: 'cost_bswap_i32'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512BW-LABEL: 'cost_bswap_i32'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+  %I32   = call i32 @llvm.bswap.i32(i32 %a32)
+  %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+  %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+  %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+  ret void
 }
 
-define i128 @var_bswap_i128(i128 %a) {
-; X64-LABEL: 'var_bswap_i128'
-; X64-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bswap = call i128 @llvm.bswap.i128(i128 %a)
-; X64-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i128 %bswap
+define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
+; SSE2-LABEL: 'cost_bswap_i16'
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 7 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_bswap_i16'
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_bswap_i16'
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX1-LABEL: 'cost_bswap_i16'
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX2-LABEL: 'cost_bswap_i16'
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512F-LABEL: 'cost_bswap_i16'
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-; X86-LABEL: 'var_bswap_i128'
-; X86-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %bswap = call i128 @llvm.bswap.i128(i128 %a)
-; X86-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i128 %bswap
+; AVX512DQ-LABEL: 'cost_bswap_i16'
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512DQ-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
-  %bswap = call i128 @llvm.bswap.i128(i128 %a)
-  ret i128 %bswap
+; AVX512BW-LABEL: 'cost_bswap_i16'
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+  %I16    = call i16 @llvm.bswap.i16(i16 %a16)
+  %V8I16  = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+  %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+  %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+  ret void
 }
+
+declare i64 @llvm.bswap.i64(i64)
+declare i32 @llvm.bswap.i32(i32)
+declare i16 @llvm.bswap.i16(i16)
+
+declare <2 x i64>  @llvm.bswap.v2i64(<2 x i64>)
+declare <4 x i32>  @llvm.bswap.v4i32(<4 x i32>)
+declare <8 x i16>  @llvm.bswap.v8i16(<8 x i16>)
+
+declare <4 x i64>  @llvm.bswap.v4i64(<4 x i64>)
+declare <8 x i32>  @llvm.bswap.v8i32(<8 x i32>)
+declare <16 x i16> @llvm.bswap.v16i16(<16 x i16>)
+
+declare <8 x i64>  @llvm.bswap.v8i64(<8 x i64>)
+declare <16 x i32> @llvm.bswap.v16i32(<16 x i32>)
+declare <32 x i16> @llvm.bswap.v32i16(<32 x i16>)