[llvm] 816b0a9 - [CostModel] add cl option to check size and latency costs; NFC
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Sun Sep 27 06:55:59 PDT 2020
Author: Sanjay Patel
Date: 2020-09-27T09:52:56-04:00
New Revision: 816b0a9c9f46cbfce253f49dad7cbf93867307ac
URL: https://github.com/llvm/llvm-project/commit/816b0a9c9f46cbfce253f49dad7cbf93867307ac
DIFF: https://github.com/llvm/llvm-project/commit/816b0a9c9f46cbfce253f49dad7cbf93867307ac.diff
LOG: [CostModel] add cl option to check size and latency costs; NFC
This is a setting used by SimplifyCFG, LoopUnroll, and InlineCost,
but there is apparently no direct test coverage for any of those
cost model values.
Added:
llvm/test/Analysis/CostModel/X86/fmaxnum-size-latency.ll
llvm/test/Analysis/CostModel/X86/fminnum-size-latency.ll
Modified:
llvm/lib/Analysis/CostModel.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Analysis/CostModel.cpp b/llvm/lib/Analysis/CostModel.cpp
index 953da964c435..0fcd69fe7dc4 100644
--- a/llvm/lib/Analysis/CostModel.cpp
+++ b/llvm/lib/Analysis/CostModel.cpp
@@ -35,7 +35,10 @@ static cl::opt<TargetTransformInfo::TargetCostKind> CostKind(
clEnumValN(TargetTransformInfo::TCK_Latency,
"latency", "Instruction latency"),
clEnumValN(TargetTransformInfo::TCK_CodeSize,
- "code-size", "Code size")));
+ "code-size", "Code size"),
+ clEnumValN(TargetTransformInfo::TCK_SizeAndLatency,
+ "size-latency", "Code size and latency")));
+
#define CM_NAME "cost-model"
#define DEBUG_TYPE CM_NAME
diff --git a/llvm/test/Analysis/CostModel/X86/fmaxnum-size-latency.ll b/llvm/test/Analysis/CostModel/X86/fmaxnum-size-latency.ll
new file mode 100644
index 000000000000..9e57bfc35614
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/fmaxnum-size-latency.ll
@@ -0,0 +1,49 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -cost-model -analyze -cost-kind=size-latency -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE2
+; RUN: opt < %s -cost-model -analyze -cost-kind=size-latency -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
+
+define i32 @f32(i32 %arg) {
+; CHECK-LABEL: 'f32'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call float @llvm.maxnum.f32(float undef, float undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+ %F32 = call float @llvm.maxnum.f32(float undef, float undef)
+ %V2F32 = call <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef)
+ %V4F32 = call <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef)
+ %V8F32 = call <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef)
+ %V16F32 = call <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef)
+ ret i32 undef
+}
+
+define i32 @f64(i32 %arg) {
+; CHECK-LABEL: 'f64'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call double @llvm.maxnum.f64(double undef, double undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+ %f64 = call double @llvm.maxnum.f64(double undef, double undef)
+ %V2f64 = call <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef)
+ %V4f64 = call <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef)
+ %V8f64 = call <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef)
+ %V16f64 = call <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef)
+ ret i32 undef
+}
+
+declare float @llvm.maxnum.f32(float, float)
+declare <2 x float> @llvm.maxnum.v2f32(<2 x float>, <2 x float>)
+declare <4 x float> @llvm.maxnum.v4f32(<4 x float>, <4 x float>)
+declare <8 x float> @llvm.maxnum.v8f32(<8 x float>, <8 x float>)
+declare <16 x float> @llvm.maxnum.v16f32(<16 x float>, <16 x float>)
+
+declare double @llvm.maxnum.f64(double, double)
+declare <2 x double> @llvm.maxnum.v2f64(<2 x double>, <2 x double>)
+declare <4 x double> @llvm.maxnum.v4f64(<4 x double>, <4 x double>)
+declare <8 x double> @llvm.maxnum.v8f64(<8 x double>, <8 x double>)
+declare <16 x double> @llvm.maxnum.v16f64(<16 x double>, <16 x double>)
diff --git a/llvm/test/Analysis/CostModel/X86/fminnum-size-latency.ll b/llvm/test/Analysis/CostModel/X86/fminnum-size-latency.ll
new file mode 100644
index 000000000000..5db3f3c20a2b
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/fminnum-size-latency.ll
@@ -0,0 +1,49 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -cost-model -analyze -cost-kind=size-latency -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE2
+; RUN: opt < %s -cost-model -analyze -cost-kind=size-latency -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
+
+define i32 @f32(i32 %arg) {
+; CHECK-LABEL: 'f32'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call float @llvm.minnum.f32(float undef, float undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+ %F32 = call float @llvm.minnum.f32(float undef, float undef)
+ %V2F32 = call <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef)
+ %V4F32 = call <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef)
+ %V8F32 = call <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef)
+ %V16F32 = call <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef)
+ ret i32 undef
+}
+
+define i32 @f64(i32 %arg) {
+; CHECK-LABEL: 'f64'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call double @llvm.minnum.f64(double undef, double undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+ %f64 = call double @llvm.minnum.f64(double undef, double undef)
+ %V2f64 = call <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef)
+ %V4f64 = call <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef)
+ %V8f64 = call <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef)
+ %V16f64 = call <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef)
+ ret i32 undef
+}
+
+declare float @llvm.minnum.f32(float, float)
+declare <2 x float> @llvm.minnum.v2f32(<2 x float>, <2 x float>)
+declare <4 x float> @llvm.minnum.v4f32(<4 x float>, <4 x float>)
+declare <8 x float> @llvm.minnum.v8f32(<8 x float>, <8 x float>)
+declare <16 x float> @llvm.minnum.v16f32(<16 x float>, <16 x float>)
+
+declare double @llvm.minnum.f64(double, double)
+declare <2 x double> @llvm.minnum.v2f64(<2 x double>, <2 x double>)
+declare <4 x double> @llvm.minnum.v4f64(<4 x double>, <4 x double>)
+declare <8 x double> @llvm.minnum.v8f64(<8 x double>, <8 x double>)
+declare <16 x double> @llvm.minnum.v16f64(<16 x double>, <16 x double>)
More information about the llvm-commits
mailing list