Skip to content

[AArch64] Add getVectorInstrCost Codesize costs handling. #130946

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Mar 27, 2025
Merged

[AArch64] Add getVectorInstrCost Codesize costs handling. #130946

merged 1 commit into from
Mar 27, 2025

Conversation

davemgreen
Copy link
Collaborator

We have a lot of missing Codesize costs for vector operations. This patch starts things off by adding codesize costs for getVectorInstrCost, returning a single cost instead of the VectorInsertExtractBaseCost (which is typically 2). Insert of a load are given a cost of 0 as they use ld1, otherwise the cost is 1.

@llvmbot
Copy link
Member

llvmbot commented Mar 12, 2025 •
edited
Loading

@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-analysis

Author: David Green (davemgreen)

Changes

We have a lot of missing Codesize costs for vector operations. This patch starts things off by adding codesize costs for getVectorInstrCost, returning a single cost instead of the VectorInsertExtractBaseCost (which is typically 2). Insert of a load are given a cost of 0 as they use ld1, otherwise the cost is 1.

Patch is 57.29 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/130946.diff

8 Files Affected:

  • (modified) llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp (+14-9)
  • (modified) llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h (+2-2)
  • (modified) llvm/test/Analysis/CostModel/AArch64/arith-fp.ll (+6-6)
  • (modified) llvm/test/Analysis/CostModel/AArch64/insert-extract.ll (+30-30)
  • (modified) llvm/test/Analysis/CostModel/AArch64/min-max.ll (+16-16)
  • (modified) llvm/test/Analysis/CostModel/AArch64/reduce-fadd.ll (+20-20)
  • (modified) llvm/test/Analysis/CostModel/AArch64/shuffle-load.ll (+7-7)
  • (modified) llvm/test/Analysis/CostModel/AArch64/sve-intrinsics.ll (+28-28)
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index 7cec8a17dfaaa..70f78cbcd6119 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -3302,8 +3302,8 @@ InstructionCost AArch64TTIImpl::getCFInstrCost(unsigned Opcode,
 }
 
 InstructionCost AArch64TTIImpl::getVectorInstrCostHelper(
-    unsigned Opcode, Type *Val, unsigned Index, bool HasRealUse,
-    const Instruction *I, Value *Scalar,
+    unsigned Opcode, Type *Val, TTI::TargetCostKind CostKind, unsigned Index,
+    bool HasRealUse, const Instruction *I, Value *Scalar,
     ArrayRef<std::tuple<Value *, User *, int>> ScalarUserAndIdx) {
   assert(Val->isVectorTy() && "This must be a vector type");
 
@@ -3336,12 +3336,16 @@ InstructionCost AArch64TTIImpl::getVectorInstrCostHelper(
     // and its second operand is a load, then we will generate a LD1, which
     // are expensive instructions.
     if (I && dyn_cast<LoadInst>(I->getOperand(1)))
-      return ST->getVectorInsertExtractBaseCost() + 1;
+      return CostKind == TTI::TCK_CodeSize
+                 ? 0
+                 : ST->getVectorInsertExtractBaseCost() + 1;
 
     // i1 inserts and extract will include an extra cset or cmp of the vector
     // value. Increase the cost by 1 to account.
     if (Val->getScalarSizeInBits() == 1)
-      return ST->getVectorInsertExtractBaseCost() + 1;
+      return CostKind == TTI::TCK_CodeSize
+                 ? 2
+                 : ST->getVectorInsertExtractBaseCost() + 1;
 
     // FIXME:
     // If the extract-element and insert-element instructions could be
@@ -3465,7 +3469,8 @@ InstructionCost AArch64TTIImpl::getVectorInstrCostHelper(
     return 0;
 
   // All other insert/extracts cost this much.
-  return ST->getVectorInsertExtractBaseCost();
+  return CostKind == TTI::TCK_CodeSize ? 1
+                                       : ST->getVectorInsertExtractBaseCost();
 }
 
 InstructionCost AArch64TTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
@@ -3474,22 +3479,22 @@ InstructionCost AArch64TTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
                                                    Value *Op1) {
   bool HasRealUse =
       Opcode == Instruction::InsertElement && Op0 && !isa<UndefValue>(Op0);
-  return getVectorInstrCostHelper(Opcode, Val, Index, HasRealUse);
+  return getVectorInstrCostHelper(Opcode, Val, CostKind, Index, HasRealUse);
 }
 
 InstructionCost AArch64TTIImpl::getVectorInstrCost(
     unsigned Opcode, Type *Val, TTI::TargetCostKind CostKind, unsigned Index,
     Value *Scalar,
     ArrayRef<std::tuple<Value *, User *, int>> ScalarUserAndIdx) {
-  return getVectorInstrCostHelper(Opcode, Val, Index, false, nullptr, Scalar,
-                                  ScalarUserAndIdx);
+  return getVectorInstrCostHelper(Opcode, Val, CostKind, Index, false, nullptr,
+                                  Scalar, ScalarUserAndIdx);
 }
 
 InstructionCost AArch64TTIImpl::getVectorInstrCost(const Instruction &I,
                                                    Type *Val,
                                                    TTI::TargetCostKind CostKind,
                                                    unsigned Index) {
-  return getVectorInstrCostHelper(I.getOpcode(), Val, Index,
+  return getVectorInstrCostHelper(I.getOpcode(), Val, CostKind, Index,
                                   true /* HasRealUse */, &I);
 }
 
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h
index 8a3fd11705640..a5db213cf1f84 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.h
@@ -73,8 +73,8 @@ class AArch64TTIImpl : public BasicTTIImplBase<AArch64TTIImpl> {
   /// of the extract(nullptr if user is not known before vectorization) and
   /// 'Idx' being the extract lane.
   InstructionCost getVectorInstrCostHelper(
-      unsigned Opcode, Type *Val, unsigned Index, bool HasRealUse,
-      const Instruction *I = nullptr, Value *Scalar = nullptr,
+      unsigned Opcode, Type *Val, TTI::TargetCostKind CostKind, unsigned Index,
+      bool HasRealUse, const Instruction *I = nullptr, Value *Scalar = nullptr,
       ArrayRef<std::tuple<Value *, User *, int>> ScalarUserAndIdx = {});
 
 public:
diff --git a/llvm/test/Analysis/CostModel/AArch64/arith-fp.ll b/llvm/test/Analysis/CostModel/AArch64/arith-fp.ll
index 1efd41426c4ef..de1b39db1539c 100644
--- a/llvm/test/Analysis/CostModel/AArch64/arith-fp.ll
+++ b/llvm/test/Analysis/CostModel/AArch64/arith-fp.ll
@@ -536,9 +536,9 @@ define void @fsqrt() {
 define void @fsqrt_fp16() {
 ; CHECK-BASE-LABEL: 'fsqrt_fp16'
 ; CHECK-BASE-NEXT:  Cost Model: Found costs of 1 for: %F16 = call half @llvm.sqrt.f16(half undef)
-; CHECK-BASE-NEXT:  Cost Model: Found costs of 10 for: %V4F16 = call <4 x half> @llvm.sqrt.v4f16(<4 x half> undef)
-; CHECK-BASE-NEXT:  Cost Model: Found costs of 22 for: %V8F16 = call <8 x half> @llvm.sqrt.v8f16(<8 x half> undef)
-; CHECK-BASE-NEXT:  Cost Model: Found costs of 44 for: %V16F16 = call <16 x half> @llvm.sqrt.v16f16(<16 x half> undef)
+; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:10 CodeSize:7 Lat:10 SizeLat:10 for: %V4F16 = call <4 x half> @llvm.sqrt.v4f16(<4 x half> undef)
+; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:22 CodeSize:15 Lat:22 SizeLat:22 for: %V8F16 = call <8 x half> @llvm.sqrt.v8f16(<8 x half> undef)
+; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:44 CodeSize:30 Lat:44 SizeLat:44 for: %V16F16 = call <16 x half> @llvm.sqrt.v16f16(<16 x half> undef)
 ; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; CHECK-FP16-LABEL: 'fsqrt_fp16'
@@ -679,9 +679,9 @@ define void @fma() {
 define void @fma_fp16() {
 ; CHECK-BASE-LABEL: 'fma_fp16'
 ; CHECK-BASE-NEXT:  Cost Model: Found costs of 1 for: %F16 = call half @llvm.fma.f16(half undef, half undef, half undef)
-; CHECK-BASE-NEXT:  Cost Model: Found costs of 10 for: %V4F16 = call <4 x half> @llvm.fma.v4f16(<4 x half> undef, <4 x half> undef, <4 x half> undef)
-; CHECK-BASE-NEXT:  Cost Model: Found costs of 22 for: %V8F16 = call <8 x half> @llvm.fma.v8f16(<8 x half> undef, <8 x half> undef, <8 x half> undef)
-; CHECK-BASE-NEXT:  Cost Model: Found costs of 44 for: %V16F16 = call <16 x half> @llvm.fma.v16f16(<16 x half> undef, <16 x half> undef, <16 x half> undef)
+; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:10 CodeSize:7 Lat:10 SizeLat:10 for: %V4F16 = call <4 x half> @llvm.fma.v4f16(<4 x half> undef, <4 x half> undef, <4 x half> undef)
+; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:22 CodeSize:15 Lat:22 SizeLat:22 for: %V8F16 = call <8 x half> @llvm.fma.v8f16(<8 x half> undef, <8 x half> undef, <8 x half> undef)
+; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:44 CodeSize:30 Lat:44 SizeLat:44 for: %V16F16 = call <16 x half> @llvm.fma.v16f16(<16 x half> undef, <16 x half> undef, <16 x half> undef)
 ; CHECK-BASE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; CHECK-FP16-LABEL: 'fma_fp16'
diff --git a/llvm/test/Analysis/CostModel/AArch64/insert-extract.ll b/llvm/test/Analysis/CostModel/AArch64/insert-extract.ll
index 3baeb075f1813..cef6cf1a081e6 100644
--- a/llvm/test/Analysis/CostModel/AArch64/insert-extract.ll
+++ b/llvm/test/Analysis/CostModel/AArch64/insert-extract.ll
@@ -11,38 +11,38 @@ target triple = "aarch64--linux-gnu"
 
 define void @vectorInstrCost() {
 ; CHECK-LABEL: 'vectorInstrCost'
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %ta0 = extractelement <8 x i1> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %ta1 = extractelement <8 x i1> undef, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t1 = extractelement <8 x i8> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t2 = extractelement <8 x i8> undef, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t3 = extractelement <4 x i16> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t4 = extractelement <4 x i16> undef, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t5 = extractelement <2 x i32> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t6 = extractelement <2 x i32> undef, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t7 = extractelement <2 x i64> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t8 = extractelement <2 x i64> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:2 Lat:3 SizeLat:3 for: %ta0 = extractelement <8 x i1> undef, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:2 Lat:3 SizeLat:3 for: %ta1 = extractelement <8 x i1> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t1 = extractelement <8 x i8> undef, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t2 = extractelement <8 x i8> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t3 = extractelement <4 x i16> undef, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t4 = extractelement <4 x i16> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t5 = extractelement <2 x i32> undef, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t6 = extractelement <2 x i32> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t7 = extractelement <2 x i64> undef, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t8 = extractelement <2 x i64> undef, i32 1
 ; CHECK-NEXT:  Cost Model: Found costs of 0 for: %t9 = extractelement <4 x half> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t10 = extractelement <4 x half> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t10 = extractelement <4 x half> undef, i32 1
 ; CHECK-NEXT:  Cost Model: Found costs of 0 for: %t11 = extractelement <2 x float> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t12 = extractelement <2 x float> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t12 = extractelement <2 x float> undef, i32 1
 ; CHECK-NEXT:  Cost Model: Found costs of 0 for: %t13 = extractelement <2 x double> undef, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t14 = extractelement <2 x double> undef, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %t31 = insertelement <8 x i1> undef, i1 false, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %t41 = insertelement <8 x i1> undef, i1 true, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t30 = insertelement <8 x i8> undef, i8 0, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t40 = insertelement <8 x i8> undef, i8 1, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t50 = insertelement <4 x i16> undef, i16 2, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t60 = insertelement <4 x i16> undef, i16 3, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t70 = insertelement <2 x i32> undef, i32 4, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t80 = insertelement <2 x i32> undef, i32 5, i32 1
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t90 = insertelement <2 x i64> undef, i64 6, i32 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t100 = insertelement <2 x i64> undef, i64 7, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t14 = extractelement <2 x double> undef, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:2 Lat:3 SizeLat:3 for: %t31 = insertelement <8 x i1> undef, i1 false, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:2 Lat:3 SizeLat:3 for: %t41 = insertelement <8 x i1> undef, i1 true, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t30 = insertelement <8 x i8> undef, i8 0, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t40 = insertelement <8 x i8> undef, i8 1, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t50 = insertelement <4 x i16> undef, i16 2, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t60 = insertelement <4 x i16> undef, i16 3, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t70 = insertelement <2 x i32> undef, i32 4, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t80 = insertelement <2 x i32> undef, i32 5, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t90 = insertelement <2 x i64> undef, i64 6, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t100 = insertelement <2 x i64> undef, i64 7, i32 1
 ; CHECK-NEXT:  Cost Model: Found costs of 0 for: %t110 = insertelement <4 x half> zeroinitializer, half 0xH0000, i64 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t120 = insertelement <4 x half> zeroinitializer, half 0xH0000, i64 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t120 = insertelement <4 x half> zeroinitializer, half 0xH0000, i64 1
 ; CHECK-NEXT:  Cost Model: Found costs of 0 for: %t130 = insertelement <2 x float> zeroinitializer, float 0.000000e+00, i64 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t140 = insertelement <2 x float> zeroinitializer, float 0.000000e+00, i64 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t140 = insertelement <2 x float> zeroinitializer, float 0.000000e+00, i64 1
 ; CHECK-NEXT:  Cost Model: Found costs of 0 for: %t150 = insertelement <2 x double> zeroinitializer, double 0.000000e+00, i64 0
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %t160 = insertelement <2 x double> zeroinitializer, double 0.000000e+00, i64 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %t160 = insertelement <2 x double> zeroinitializer, double 0.000000e+00, i64 1
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %ta0 = extractelement <8 x i1> undef, i32 0
@@ -86,7 +86,7 @@ define void @vectorInstrCost() {
 define <8 x i8> @LD1_B(<8 x i8> %vec, ptr noundef %i) {
 ; CHECK-LABEL: 'LD1_B'
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %v1 = load i8, ptr %i, align 1
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %v2 = insertelement <8 x i8> %vec, i8 %v1, i32 1
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:0 Lat:3 SizeLat:3 for: %v2 = insertelement <8 x i8> %vec, i8 %v1, i32 1
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <8 x i8> %v2
 ;
 entry:
@@ -98,7 +98,7 @@ entry:
 define <4 x i16> @LD1_H(<4 x i16> %vec, ptr noundef %i) {
 ; CHECK-LABEL: 'LD1_H'
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %v1 = load i16, ptr %i, align 2
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %v2 = insertelement <4 x i16> %vec, i16 %v1, i32 2
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:0 Lat:3 SizeLat:3 for: %v2 = insertelement <4 x i16> %vec, i16 %v1, i32 2
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i16> %v2
 ;
 entry:
@@ -110,7 +110,7 @@ entry:
 define <4 x i32> @LD1_W(<4 x i32> %vec, ptr noundef %i) {
 ; CHECK-LABEL: 'LD1_W'
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %v1 = load i32, ptr %i, align 4
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %v2 = insertelement <4 x i32> %vec, i32 %v1, i32 3
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:0 Lat:3 SizeLat:3 for: %v2 = insertelement <4 x i32> %vec, i32 %v1, i32 3
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <4 x i32> %v2
 ;
 entry:
@@ -122,7 +122,7 @@ entry:
 define <2 x i64> @LD1_X(<2 x i64> %vec, ptr noundef %i) {
 ; CHECK-LABEL: 'LD1_X'
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %v1 = load i64, ptr %i, align 8
-; CHECK-NEXT:  Cost Model: Found costs of 3 for: %v2 = insertelement <2 x i64> %vec, i64 %v1, i32 0
+; CHECK-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:0 Lat:3 SizeLat:3 for: %v2 = insertelement <2 x i64> %vec, i64 %v1, i32 0
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret <2 x i64> %v2
 ;
 entry:
diff --git a/llvm/test/Analysis/CostModel/AArch64/min-max.ll b/llvm/test/Analysis/CostModel/AArch64/min-max.ll
index ed4b9dd2ba571..b824f5309adc1 100644
--- a/llvm/test/Analysis/CostModel/AArch64/min-max.ll
+++ b/llvm/test/Analysis/CostModel/AArch64/min-max.ll
@@ -195,10 +195,10 @@ define void @smax() {
 define void @minnum16() {
 ; CHECK-NOF16-LABEL: 'minnum16'
 ; CHECK-NOF16-NEXT:  Cost Model: Found costs of 1 for: %f16 = call half @llvm.minnum.f16(half undef, half undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 4 for: %V2f16 = call <2 x half> @llvm.minnum.v2f16(<2 x half> undef, <2 x half> undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 10 for: %V4f16 = call <4 x half> @llvm.minnum.v4f16(<4 x half> undef, <4 x half> undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 22 for: %V8f16 = call <8 x half> @llvm.minnum.v8f16(<8 x half> undef, <8 x half> undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 44 for: %V16f16 = call <16 x half> @llvm.minnum.v16f16(<16 x half> undef, <16 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:3 Lat:4 SizeLat:4 for: %V2f16 = call <2 x half> @llvm.minnum.v2f16(<2 x half> undef, <2 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:10 CodeSize:7 Lat:10 SizeLat:10 for: %V4f16 = call <4 x half> @llvm.minnum.v4f16(<4 x half> undef, <4 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:22 CodeSize:15 Lat:22 SizeLat:22 for: %V8f16 = call <8 x half> @llvm.minnum.v8f16(<8 x half> undef, <8 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:44 CodeSize:30 Lat:44 SizeLat:44 for: %V16f16 = call <16 x half> @llvm.minnum.v16f16(<16 x half> undef, <16 x half> undef)
 ; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; CHECK-F16-LABEL: 'minnum16'
@@ -220,10 +220,10 @@ define void @minnum16() {
 define void @maxnum16() {
 ; CHECK-NOF16-LABEL: 'maxnum16'
 ; CHECK-NOF16-NEXT:  Cost Model: Found costs of 1 for: %f16 = call half @llvm.maxnum.f16(half undef, half undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 4 for: %V2f16 = call <2 x half> @llvm.maxnum.v2f16(<2 x half> undef, <2 x half> undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 10 for: %V4f16 = call <4 x half> @llvm.maxnum.v4f16(<4 x half> undef, <4 x half> undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 22 for: %V8f16 = call <8 x half> @llvm.maxnum.v8f16(<8 x half> undef, <8 x half> undef)
-; CHECK-NOF16-NEXT:  Cost Model: Found costs of 44 for: %V16f16 = call <16 x half> @llvm.maxnum.v16f16(<16 x half> undef, <16 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:3 Lat:4 SizeLat:4 for: %V2f16 = call <2 x half> @llvm.maxnum.v2f16(<2 x half> undef, <2 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:10 CodeSize:7 Lat:10 SizeLat:10 for: %V4f16 = call <4 x half> @llvm.maxnum.v4f16(<4 x half> undef, <4 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:22 CodeSize:15 Lat:22 SizeLat:22 for: %V8f16 = call <8 x half> @llvm.maxnum.v8f16(<8 x half> undef, <8 x half> undef)
+; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:44 CodeSize:30 Lat:44 SizeLat:44 for: %V16f16 = call <16 x half> @llvm.maxnum.v16f16(<16 x half> undef, <16 x half> undef)
 ; CHECK-NOF16-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; CHECK-F16-LABEL: 'maxnum16'
@@ -288,10 +288,10 @@ define void @maxnum() {
 define void @minimum16() {
 ; CHECK-NOF16-LABEL: 'minimum16'
 ; CHECK-NOF16-NEXT:  Cost Model: Found costs of 1 for: %f16 = call half @llvm.minimum.f16(half undef, ...
[truncated]

nasherm
nasherm approved these changes Mar 13, 2025
View reviewed changes
Copy link
Contributor

@nasherm nasherm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@davemgreen
[AArch64] Add getVectorInstrCost Codesize costs handling.
c6ca913 
We have a lot of missing Codesize costs for vector operations. This patch
starts things off by adding codesize costs for getVectorInstrCost, returning a
single cost instead of the VectorInsertExtractBaseCost (which is typically 2).
Insert of a load are given a cost of 0 as they use ld1, otherwise the cost
is 1.
@davemgreen davemgreen force-pushed the gh-a64-vecinstrcodesizecost branch from 5e3db3c to c6ca913 Compare March 27, 2025 15:04
@davemgreen davemgreen merged commit c6406c8 into llvm:main Mar 27, 2025
11 checks passed
@davemgreen davemgreen deleted the gh-a64-vecinstrcodesizecost branch March 27, 2025 17:25
davemgreen added a commit that referenced this pull request Apr 11, 2025
@davemgreen
[AArch64] Use a lower Costsize cost in getScalarizationOverhead.
052225d 
This is a follow on to #130946 to use the same codesize cost override in
getScalarizationOverhead for vector instructions.
var-const pushed a commit to ldionne/llvm-project that referenced this pull request Apr 17, 2025
@davemgreen @var-const
[AArch64] Use a lower Costsize cost in getScalarizationOverhead.
580261a 
This is a follow on to llvm#130946 to use the same codesize cost override in
getScalarizationOverhead for vector instructions.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@nasherm nasherm nasherm approved these changes

@SamTebbs33 SamTebbs33 Awaiting requested review from SamTebbs33

@c-rhodes c-rhodes Awaiting requested review from c-rhodes

@sjoerdmeijer sjoerdmeijer Awaiting requested review from sjoerdmeijer

@sushgokh sushgokh Awaiting requested review from sushgokh

@john-brawn-arm john-brawn-arm Awaiting requested review from john-brawn-arm

Assignees
No one assigned
Labels
backend:AArch64 llvm:analysis
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@davemgreen @llvmbot @nasherm