Skip to content
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

[CostModel][X86] SK_InsertSubvector - if the legalized base/subvector types match then the insert is now just a copy #84048

Closed
wants to merge 1 commit into from
Closed

[CostModel][X86] SK_InsertSubvector - if the legalized base/subvector types match then the insert is now just a copy #84048

wants to merge 1 commit into from

Conversation

RKSimon
Copy link
Collaborator

@RKSimon RKSimon commented Mar 5, 2024

No description provided.

@llvmbot
Copy link
Collaborator

llvmbot commented Mar 5, 2024 •
edited

@llvm/pr-subscribers-llvm-analysis
@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

Patch is 1.31 MiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/84048.diff

25 Files Affected:

  • (modified) llvm/lib/Target/X86/X86TargetTransformInfo.cpp (+6)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-intrinsic-cost-inseltpoison.ll (+204-204)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-intrinsic-cost.ll (+204-204)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-load-i16.ll (+2-2)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-load-i32.ll (+3-3)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-load-i8.ll (+3-3)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-store-i16.ll (+2-2)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-store-i32.ll (+3-3)
  • (modified) llvm/test/Analysis/CostModel/X86/masked-store-i8.ll (+3-3)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-concat_subvector-codesize.ll (+36-36)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-concat_subvector-latency.ll (+36-36)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-concat_subvector-sizelatency.ll (+36-36)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-concat_subvector.ll (+36-36)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-insert_subvector-codesize.ll (+294-374)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-insert_subvector-latency.ll (+294-374)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-insert_subvector-sizelatency.ll (+294-374)
  • (modified) llvm/test/Analysis/CostModel/X86/shuffle-insert_subvector.ll (+294-374)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/alternate-fp-inseltpoison.ll (+8-22)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/alternate-fp.ll (+8-22)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/crash_clear_undefs.ll (+1-1)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/load-merge-inseltpoison.ll (+12-18)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/load-merge.ll (+12-18)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/load-partial-vector-shuffle.ll (+14-23)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/remark-partial-loads-vectorize.ll (+2-2)
  • (modified) llvm/test/Transforms/SLPVectorizer/X86/resched.ll (+16-16)
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 4cca291a245622..851679918fea3d 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -1550,6 +1550,12 @@ InstructionCost X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
   if (Kind == TTI::SK_InsertSubvector && LT.second.isVector()) {
     int NumElts = LT.second.getVectorNumElements();
     std::pair<InstructionCost, MVT> SubLT = getTypeLegalizationCost(SubTp);
+
+    // If we're inserting into the same legalized type then its not an
+    // insert anymore, its a copy.
+    if (LT.second == SubLT.second)
+      return TTI::TCC_Free;
+
     if (SubLT.second.isVector()) {
       int NumSubElts = SubLT.second.getVectorNumElements();
       if ((Index % NumSubElts) == 0 && (NumElts % NumSubElts) == 0)
diff --git a/llvm/test/Analysis/CostModel/X86/masked-intrinsic-cost-inseltpoison.ll b/llvm/test/Analysis/CostModel/X86/masked-intrinsic-cost-inseltpoison.ll
index 897344d622d0f8..7265cf1ab1a938 100644
--- a/llvm/test/Analysis/CostModel/X86/masked-intrinsic-cost-inseltpoison.ll
+++ b/llvm/test/Analysis/CostModel/X86/masked-intrinsic-cost-inseltpoison.ll
@@ -129,52 +129,52 @@ define i32 @masked_load() {
 ;
 ; AVX-LABEL: 'masked_load'
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8F64 = call <8 x double> @llvm.masked.load.v8f64.p0(ptr undef, i32 1, <8 x i1> undef, <8 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V7F64 = call <7 x double> @llvm.masked.load.v7f64.p0(ptr undef, i32 1, <7 x i1> undef, <7 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V6F64 = call <6 x double> @llvm.masked.load.v6f64.p0(ptr undef, i32 1, <6 x i1> undef, <6 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V5F64 = call <5 x double> @llvm.masked.load.v5f64.p0(ptr undef, i32 1, <5 x i1> undef, <5 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V7F64 = call <7 x double> @llvm.masked.load.v7f64.p0(ptr undef, i32 1, <7 x i1> undef, <7 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V6F64 = call <6 x double> @llvm.masked.load.v6f64.p0(ptr undef, i32 1, <6 x i1> undef, <6 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V5F64 = call <5 x double> @llvm.masked.load.v5f64.p0(ptr undef, i32 1, <5 x i1> undef, <5 x double> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F64 = call <4 x double> @llvm.masked.load.v4f64.p0(ptr undef, i32 1, <4 x i1> undef, <4 x double> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V3F64 = call <3 x double> @llvm.masked.load.v3f64.p0(ptr undef, i32 1, <3 x i1> undef, <3 x double> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3F64 = call <3 x double> @llvm.masked.load.v3f64.p0(ptr undef, i32 1, <3 x i1> undef, <3 x double> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F64 = call <2 x double> @llvm.masked.load.v2f64.p0(ptr undef, i32 1, <2 x i1> undef, <2 x double> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F64 = call <1 x double> @llvm.masked.load.v1f64.p0(ptr undef, i32 1, <1 x i1> undef, <1 x double> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = call <16 x float> @llvm.masked.load.v16f32.p0(ptr undef, i32 1, <16 x i1> undef, <16 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V15F32 = call <15 x float> @llvm.masked.load.v15f32.p0(ptr undef, i32 1, <15 x i1> undef, <15 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V14F32 = call <14 x float> @llvm.masked.load.v14f32.p0(ptr undef, i32 1, <14 x i1> undef, <14 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V13F32 = call <13 x float> @llvm.masked.load.v13f32.p0(ptr undef, i32 1, <13 x i1> undef, <13 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V12F32 = call <12 x float> @llvm.masked.load.v12f32.p0(ptr undef, i32 1, <12 x i1> undef, <12 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V11F32 = call <11 x float> @llvm.masked.load.v11f32.p0(ptr undef, i32 1, <11 x i1> undef, <11 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V10F32 = call <10 x float> @llvm.masked.load.v10f32.p0(ptr undef, i32 1, <10 x i1> undef, <10 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V9F32 = call <9 x float> @llvm.masked.load.v9f32.p0(ptr undef, i32 1, <9 x i1> undef, <9 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V15F32 = call <15 x float> @llvm.masked.load.v15f32.p0(ptr undef, i32 1, <15 x i1> undef, <15 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V14F32 = call <14 x float> @llvm.masked.load.v14f32.p0(ptr undef, i32 1, <14 x i1> undef, <14 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V13F32 = call <13 x float> @llvm.masked.load.v13f32.p0(ptr undef, i32 1, <13 x i1> undef, <13 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V12F32 = call <12 x float> @llvm.masked.load.v12f32.p0(ptr undef, i32 1, <12 x i1> undef, <12 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V11F32 = call <11 x float> @llvm.masked.load.v11f32.p0(ptr undef, i32 1, <11 x i1> undef, <11 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V10F32 = call <10 x float> @llvm.masked.load.v10f32.p0(ptr undef, i32 1, <10 x i1> undef, <10 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V9F32 = call <9 x float> @llvm.masked.load.v9f32.p0(ptr undef, i32 1, <9 x i1> undef, <9 x float> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call <8 x float> @llvm.masked.load.v8f32.p0(ptr undef, i32 1, <8 x i1> undef, <8 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V7F32 = call <7 x float> @llvm.masked.load.v7f32.p0(ptr undef, i32 1, <7 x i1> undef, <7 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V6F32 = call <6 x float> @llvm.masked.load.v6f32.p0(ptr undef, i32 1, <6 x i1> undef, <6 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V5F32 = call <5 x float> @llvm.masked.load.v5f32.p0(ptr undef, i32 1, <5 x i1> undef, <5 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7F32 = call <7 x float> @llvm.masked.load.v7f32.p0(ptr undef, i32 1, <7 x i1> undef, <7 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6F32 = call <6 x float> @llvm.masked.load.v6f32.p0(ptr undef, i32 1, <6 x i1> undef, <6 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V5F32 = call <5 x float> @llvm.masked.load.v5f32.p0(ptr undef, i32 1, <5 x i1> undef, <5 x float> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call <4 x float> @llvm.masked.load.v4f32.p0(ptr undef, i32 1, <4 x i1> undef, <4 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V3F32 = call <3 x float> @llvm.masked.load.v3f32.p0(ptr undef, i32 1, <3 x i1> undef, <3 x float> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call <2 x float> @llvm.masked.load.v2f32.p0(ptr undef, i32 1, <2 x i1> undef, <2 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3F32 = call <3 x float> @llvm.masked.load.v3f32.p0(ptr undef, i32 1, <3 x i1> undef, <3 x float> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call <2 x float> @llvm.masked.load.v2f32.p0(ptr undef, i32 1, <2 x i1> undef, <2 x float> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F32 = call <1 x float> @llvm.masked.load.v1f32.p0(ptr undef, i32 1, <1 x i1> undef, <1 x float> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.masked.load.v8i64.p0(ptr undef, i32 1, <8 x i1> undef, <8 x i64> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V7I64 = call <7 x i64> @llvm.masked.load.v7i64.p0(ptr undef, i32 1, <7 x i1> undef, <7 x i64> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V6I64 = call <6 x i64> @llvm.masked.load.v6i64.p0(ptr undef, i32 1, <6 x i1> undef, <6 x i64> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V5I64 = call <5 x i64> @llvm.masked.load.v5i64.p0(ptr undef, i32 1, <5 x i1> undef, <5 x i64> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V7I64 = call <7 x i64> @llvm.masked.load.v7i64.p0(ptr undef, i32 1, <7 x i1> undef, <7 x i64> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V6I64 = call <6 x i64> @llvm.masked.load.v6i64.p0(ptr undef, i32 1, <6 x i1> undef, <6 x i64> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V5I64 = call <5 x i64> @llvm.masked.load.v5i64.p0(ptr undef, i32 1, <5 x i1> undef, <5 x i64> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr undef, i32 1, <4 x i1> undef, <4 x i64> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V3I64 = call <3 x i64> @llvm.masked.load.v3i64.p0(ptr undef, i32 1, <3 x i1> undef, <3 x i64> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3I64 = call <3 x i64> @llvm.masked.load.v3i64.p0(ptr undef, i32 1, <3 x i1> undef, <3 x i64> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.masked.load.v2i64.p0(ptr undef, i32 1, <2 x i1> undef, <2 x i64> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1I64 = call <1 x i64> @llvm.masked.load.v1i64.p0(ptr undef, i32 1, <1 x i1> undef, <1 x i64> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V16I32 = call <16 x i32> @llvm.masked.load.v16i32.p0(ptr undef, i32 1, <16 x i1> undef, <16 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V15I32 = call <15 x i32> @llvm.masked.load.v15i32.p0(ptr undef, i32 1, <15 x i1> undef, <15 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V14I32 = call <14 x i32> @llvm.masked.load.v14i32.p0(ptr undef, i32 1, <14 x i1> undef, <14 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V13I32 = call <13 x i32> @llvm.masked.load.v13i32.p0(ptr undef, i32 1, <13 x i1> undef, <13 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V12I32 = call <12 x i32> @llvm.masked.load.v12i32.p0(ptr undef, i32 1, <12 x i1> undef, <12 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V11I32 = call <11 x i32> @llvm.masked.load.v11i32.p0(ptr undef, i32 1, <11 x i1> undef, <11 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V10I32 = call <10 x i32> @llvm.masked.load.v10i32.p0(ptr undef, i32 1, <10 x i1> undef, <10 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V9I32 = call <9 x i32> @llvm.masked.load.v9i32.p0(ptr undef, i32 1, <9 x i1> undef, <9 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V15I32 = call <15 x i32> @llvm.masked.load.v15i32.p0(ptr undef, i32 1, <15 x i1> undef, <15 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V14I32 = call <14 x i32> @llvm.masked.load.v14i32.p0(ptr undef, i32 1, <14 x i1> undef, <14 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V13I32 = call <13 x i32> @llvm.masked.load.v13i32.p0(ptr undef, i32 1, <13 x i1> undef, <13 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V12I32 = call <12 x i32> @llvm.masked.load.v12i32.p0(ptr undef, i32 1, <12 x i1> undef, <12 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V11I32 = call <11 x i32> @llvm.masked.load.v11i32.p0(ptr undef, i32 1, <11 x i1> undef, <11 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V10I32 = call <10 x i32> @llvm.masked.load.v10i32.p0(ptr undef, i32 1, <10 x i1> undef, <10 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %V9I32 = call <9 x i32> @llvm.masked.load.v9i32.p0(ptr undef, i32 1, <9 x i1> undef, <9 x i32> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <8 x i32> @llvm.masked.load.v8i32.p0(ptr undef, i32 1, <8 x i1> undef, <8 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V7I32 = call <7 x i32> @llvm.masked.load.v7i32.p0(ptr undef, i32 1, <7 x i1> undef, <7 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V6I32 = call <6 x i32> @llvm.masked.load.v6i32.p0(ptr undef, i32 1, <6 x i1> undef, <6 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V5I32 = call <5 x i32> @llvm.masked.load.v5i32.p0(ptr undef, i32 1, <5 x i1> undef, <5 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7I32 = call <7 x i32> @llvm.masked.load.v7i32.p0(ptr undef, i32 1, <7 x i1> undef, <7 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6I32 = call <6 x i32> @llvm.masked.load.v6i32.p0(ptr undef, i32 1, <6 x i1> undef, <6 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V5I32 = call <5 x i32> @llvm.masked.load.v5i32.p0(ptr undef, i32 1, <5 x i1> undef, <5 x i32> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr undef, i32 1, <4 x i1> undef, <4 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V3I32 = call <3 x i32> @llvm.masked.load.v3i32.p0(ptr undef, i32 1, <3 x i1> undef, <3 x i32> undef)
-; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <2 x i32> @llvm.masked.load.v2i32.p0(ptr undef, i32 1, <2 x i1> undef, <2 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3I32 = call <3 x i32> @llvm.masked.load.v3i32.p0(ptr undef, i32 1, <3 x i1> undef, <3 x i32> undef)
+; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <2 x i32> @llvm.masked.load.v2i32.p0(ptr undef, i32 1, <2 x i1> undef, <2 x i32> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1I32 = call <1 x i32> @llvm.masked.load.v1i32.p0(ptr undef, i32 1, <1 x i1> undef, <1 x i32> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 131 for instruction: %V32I16 = call <32 x i16> @llvm.masked.load.v32i16.p0(ptr undef, i32 1, <32 x i1> undef, <32 x i16> undef)
 ; AVX-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V16I16 = call <16 x i16> @llvm.masked.load.v16i16.p0(ptr undef, i32 1, <16 x i1> undef, <16 x i16> undef)
@@ -188,52 +188,52 @@ define i32 @masked_load() {
 ;
 ; KNL-LABEL: 'masked_load'
 ; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8F64 = call <8 x double> @llvm.masked.load.v8f64.p0(ptr undef, i32 1, <8 x i1> undef, <8 x double> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V7F64 = call <7 x double> @llvm.masked.load.v7f64.p0(ptr undef, i32 1, <7 x i1> undef, <7 x double> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V6F64 = call <6 x double> @llvm.masked.load.v6f64.p0(ptr undef, i32 1, <6 x i1> undef, <6 x double> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V5F64 = call <5 x double> @llvm.masked.load.v5f64.p0(ptr undef, i32 1, <5 x i1> undef, <5 x double> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V7F64 = call <7 x double> @llvm.masked.load.v7f64.p0(ptr undef, i32 1, <7 x i1> undef, <7 x double> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V6F64 = call <6 x double> @llvm.masked.load.v6f64.p0(ptr undef, i32 1, <6 x i1> undef, <6 x double> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V5F64 = call <5 x double> @llvm.masked.load.v5f64.p0(ptr undef, i32 1, <5 x i1> undef, <5 x double> undef)
 ; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4F64 = call <4 x double> @llvm.masked.load.v4f64.p0(ptr undef, i32 1, <4 x i1> undef, <4 x double> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V3F64 = call <3 x double> @llvm.masked.load.v3f64.p0(ptr undef, i32 1, <3 x i1> undef, <3 x double> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V3F64 = call <3 x double> @llvm.masked.load.v3f64.p0(ptr undef, i32 1, <3 x i1> undef, <3 x double> undef)
 ; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2F64 = call <2 x double> @llvm.masked.load.v2f64.p0(ptr undef, i32 1, <2 x i1> undef, <2 x double> undef)
 ; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V1F64 = call <1 x double> @llvm.masked.load.v1f64.p0(ptr undef, i32 1, <1 x i1> undef, <1 x double> undef)
 ; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call <16 x float> @llvm.masked.load.v16f32.p0(ptr undef, i32 1, <16 x i1> undef, <16 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V15F32 = call <15 x float> @llvm.masked.load.v15f32.p0(ptr undef, i32 1, <15 x i1> undef, <15 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V14F32 = call <14 x float> @llvm.masked.load.v14f32.p0(ptr undef, i32 1, <14 x i1> undef, <14 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V13F32 = call <13 x float> @llvm.masked.load.v13f32.p0(ptr undef, i32 1, <13 x i1> undef, <13 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V12F32 = call <12 x float> @llvm.masked.load.v12f32.p0(ptr undef, i32 1, <12 x i1> undef, <12 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V11F32 = call <11 x float> @llvm.masked.load.v11f32.p0(ptr undef, i32 1, <11 x i1> undef, <11 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V10F32 = call <10 x float> @llvm.masked.load.v10f32.p0(ptr undef, i32 1, <10 x i1> undef, <10 x float> undef)
-; KNL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V9F32 = call <9 x float> @llvm.masked.load.v9f32.p0(ptr undef, i32 1, <9 x i1> undef, <9 x float> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V15F32 = call <15 x float> @llvm.masked.load.v15f32.p0(ptr undef, i32 1, <15 x i1> undef, <15 x float> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V14F32 = call <14 x float> @llvm.masked.load.v14f32.p0(ptr undef, i32 1, <14 x i1> undef, <14 x float> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V13F32 = call <13 x float> @llvm.masked.load.v13f32.p0(ptr undef, i32 1, <13 x i1> undef, <13 x float> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V12F32 = call <12 x float> @llvm.masked.load.v12f32.p0(ptr undef, i32 1, <12 x i1> undef, <12 x float> undef)
+; KNL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V11F32 = call <11 x float> @llvm.masked.load.v11f32.p0(pt...
[truncated]

alexey-bataev
alexey-bataev reviewed Mar 5, 2024
View reviewed changes
llvm/lib/Target/X86/X86TargetTransformInfo.cpp Outdated
Comment on lines 1556 to 1559
if (LT.second == SubLT.second)
return TTI::TCC_Free;
Copy link
Member

Choose a reason for hiding this comment

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

This is probably not quite correct. shufflevector <4 x i32>, <4 xi32> , <0,1,2,4> is a subvector insert of 2 vectors with the same types, but it is not free.

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

But wouldn't the SubTp not match?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

Sorry, just realized what you meant - I'll take another look soon.

@RKSimon RKSimon marked this pull request as draft March 5, 2024 19:04
@RKSimon
[CostModel][X86] SK_InsertSubvector - if the legalized base/subvector…
1547c01 
… types match then the insert is now just a copy
@RKSimon RKSimon marked this pull request as ready for review March 7, 2024 16:22
@RKSimon RKSimon marked this pull request as draft March 26, 2024 15:02
@RKSimon
Copy link
Collaborator Author

RKSimon commented Jun 5, 2024

broken + defunct

@RKSimon RKSimon closed this Jun 5, 2024
@RKSimon RKSimon deleted the free-insertsubvector branch June 5, 2024 15:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@alexey-bataev alexey-bataev alexey-bataev left review comments

Assignees
No one assigned
Labels
backend:X86 llvm:analysis llvm:transforms
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@RKSimon @llvmbot @alexey-bataev