[llvm] c692008 - [CostModel][X86] Add abs intrinsics cost model tests
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Tue Jul 28 11:56:46 PDT 2020
Author: Simon Pilgrim
Date: 2020-07-28T19:56:10+01:00
New Revision: c6920081a8b97a8a911803d2bac50fa1db7a8123
URL: https://github.com/llvm/llvm-project/commit/c6920081a8b97a8a911803d2bac50fa1db7a8123
DIFF: https://github.com/llvm/llvm-project/commit/c6920081a8b97a8a911803d2bac50fa1db7a8123.diff
LOG: [CostModel][X86] Add abs intrinsics cost model tests
abs costs currently falls back in scalar generic intrinsic calls
Added:
llvm/test/Analysis/CostModel/X86/abs.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/Analysis/CostModel/X86/abs.ll b/llvm/test/Analysis/CostModel/X86/abs.ll
new file mode 100644
index 000000000000..ee61f5c07a39
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/abs.ll
@@ -0,0 +1,354 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+sse2 | FileCheck %s -check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+sse2 | FileCheck %s -check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+ssse3 | FileCheck %s -check-prefixes=CHECK,SSE,SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+sse4.2 | FileCheck %s -check-prefixes=CHECK,SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx | FileCheck %s -check-prefixes=CHECK,AVX,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx2 | FileCheck %s -check-prefixes=CHECK,AVX,AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f | FileCheck %s -check-prefixes=CHECK,AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512vl,+avx512dq | FileCheck %s -check-prefixes=CHECK,AVX512,AVX512DQ
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512vl,+avx512bw | FileCheck %s -check-prefixes=CHECK,AVX512,AVX512BW
+
+;
+; abs(X, not_posion)
+;
+
+define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
+; SSE2-LABEL: 'cost_abs_i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_abs_i64'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_abs_i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 0)
+ %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 0)
+ %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 0)
+ %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 0)
+ ret void
+}
+
+define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
+; SSE2-LABEL: 'cost_abs_i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_abs_i32'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_abs_i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i32'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 63 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 0)
+ %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 0)
+ %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 0)
+ %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 0)
+ ret void
+}
+
+define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
+; SSE-LABEL: 'cost_abs_i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i16'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 57 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 114 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i16'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 57 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 123 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 0)
+ %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 0)
+ %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 0)
+ %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 0)
+ ret void
+}
+
+define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
+; SSE2-LABEL: 'cost_abs_i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_abs_i8'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_abs_i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 192 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 113 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 226 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 113 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 243 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 0)
+ %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 0)
+ %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 0)
+ %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 0)
+ ret void
+}
+
+;
+; abs(X, posion)
+;
+
+define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
+; SSE2-LABEL: 'cost_abs_i64_poison'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_abs_i64_poison'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_abs_i64_poison'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i64_poison'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i64_poison'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 -1)
+ %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 -1)
+ %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 -1)
+ %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 -1)
+ ret void
+}
+
+define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
+; SSE2-LABEL: 'cost_abs_i32_poison'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_abs_i32_poison'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_abs_i32_poison'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i32_poison'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i32_poison'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 63 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 -1)
+ %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 -1)
+ %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 -1)
+ %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 -1)
+ ret void
+}
+
+define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
+; SSE-LABEL: 'cost_abs_i16_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i16_poison'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 57 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 114 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i16_poison'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 57 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 123 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 -1)
+ %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 -1)
+ %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 -1)
+ %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 -1)
+ ret void
+}
+
+define void @cost_abs_i8_poison(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
+; SSE2-LABEL: 'cost_abs_i8_poison'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSSE3-LABEL: 'cost_abs_i8_poison'
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 312 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; SSE42-LABEL: 'cost_abs_i8_poison'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 192 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX-LABEL: 'cost_abs_i8_poison'
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 113 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 226 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; AVX512-LABEL: 'cost_abs_i8_poison'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 113 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 243 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 -1)
+ %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 -1)
+ %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 -1)
+ %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 -1)
+ ret void
+}
+
+declare i64 @llvm.abs.i64(i64, i1)
+declare i32 @llvm.abs.i32(i32, i1)
+declare i16 @llvm.abs.i16(i16, i1)
+declare i8 @llvm.abs.i8(i8, i1)
+
+declare <2 x i64> @llvm.abs.v2i64(<2 x i64>, i1)
+declare <4 x i32> @llvm.abs.v4i32(<4 x i32>, i1)
+declare <8 x i16> @llvm.abs.v8i16(<8 x i16>, i1)
+declare <16 x i8> @llvm.abs.v16i8(<16 x i8>, i1)
+
+declare <4 x i64> @llvm.abs.v4i64(<4 x i64>, i1)
+declare <8 x i32> @llvm.abs.v8i32(<8 x i32>, i1)
+declare <16 x i16> @llvm.abs.v16i16(<16 x i16>, i1)
+declare <32 x i8> @llvm.abs.v32i8(<32 x i8>, i1)
+
+declare <8 x i64> @llvm.abs.v8i64(<8 x i64>, i1)
+declare <16 x i32> @llvm.abs.v16i32(<16 x i32>, i1)
+declare <32 x i16> @llvm.abs.v32i16(<32 x i16>, i1)
+declare <64 x i8> @llvm.abs.v64i8(<64 x i8>, i1)
More information about the llvm-commits
mailing list