[llvm] [RISCV][TTI] Model broadcast loads as zero-stride loads (PR #198446)
via llvm-commits
llvm-commits at lists.llvm.org
Mon May 18 21:59:52 PDT 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
Author: Pengcheng Wang (wangpc-pp)
<details>
<summary>Changes</summary>
We have done this optimization in ISel and this PR just models it
in TTI.
---
Patch is 20.95 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/198446.diff
3 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+31)
- (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h (+3)
- (added) llvm/test/Analysis/CostModel/RISCV/shuffle-load.ll (+163)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index 3d2a70a826dd1..f271ba02b8ca3 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -873,6 +873,13 @@ RISCVTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, VectorType *DstTy,
LT.second, CostKind));
}
case TTI::SK_Broadcast: {
+ // Check for broadcast loads, which are synthesized by optimized zero-stride
+ // loads (this is checked in RISCVTTIImpl::isLegalBroadcastLoad).
+ bool IsLoad = !Args.empty() && isa<LoadInst>(Args[0]);
+ if (IsLoad && isLegalBroadcastLoad(SrcTy->getElementType(),
+ LT.second.getVectorElementCount()))
+ return 0;
+
bool HasScalar = (Args.size() > 0) && (Operator::getOpcode(Args[0]) ==
Instruction::InsertElement);
if (LT.second.getScalarSizeInBits() == 1) {
@@ -3408,6 +3415,30 @@ bool RISCVTTIImpl::isLegalMaskedCompressStore(Type *DataTy,
return true;
}
+bool RISCVTTIImpl::isLegalBroadcastLoad(Type *ElementTy,
+ ElementCount NumElements) const {
+ // Optimized zero-stride loads can be treated as broadcasts.
+ if (!(ST->hasVInstructions() && ST->hasOptimizedZeroStrideLoad()))
+ return false;
+
+ switch (ElementTy->getScalarSizeInBits()) {
+ case 8:
+ return ElementTy->isIntegerTy();
+ case 16:
+ return ElementTy->isIntegerTy() ||
+ (ElementTy->isHalfTy() && ST->hasVInstructionsF16Minimal()) ||
+ (ElementTy->isBFloatTy() && ST->hasVInstructionsBF16Minimal());
+ case 32:
+ return ElementTy->isIntegerTy() ||
+ (ElementTy->isFloatTy() && ST->hasVInstructionsF32());
+ case 64:
+ return (ElementTy->isIntegerTy() && ST->hasVInstructionsI64()) ||
+ (ElementTy->isFloatingPointTy() && ST->hasVInstructionsF64());
+ }
+
+ return false;
+}
+
/// See if \p I should be considered for address type promotion. We check if \p
/// I is a sext with right type and used in memory accesses. If it used in a
/// "complex" getelementptr, we allow it to be promoted without finding other
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
index 477c419228a6b..2b43b93daa6c7 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
@@ -364,6 +364,9 @@ class RISCVTTIImpl final : public BasicTTIImplBase<RISCVTTIImpl> {
bool isLegalMaskedCompressStore(Type *DataTy, Align Alignment) const override;
+ bool isLegalBroadcastLoad(Type *ElementTy,
+ ElementCount NumElements) const override;
+
/// \returns How the target needs this vector-predicated operation to be
/// transformed.
TargetTransformInfo::VPLegalization
diff --git a/llvm/test/Analysis/CostModel/RISCV/shuffle-load.ll b/llvm/test/Analysis/CostModel/RISCV/shuffle-load.ll
new file mode 100644
index 0000000000000..de2f76ce8793e
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/shuffle-load.ll
@@ -0,0 +1,163 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
+; RUN: opt < %s -mtriple=riscv64-linux-gnu -mattr=+v,+zvfhmin,+zvfbfmin \
+; RUN: -passes="print<cost-model>" -cost-kind=all 2>&1 -disable-output \
+; RUN: | FileCheck %s --check-prefix=NO-OPTIMIZED-ZERO-STRIDE-LOAD
+; RUN: opt < %s -mtriple=riscv64-linux-gnu -mattr=+v,+zvfhmin,+zvfbfmin,+optimized-zero-stride-load \
+; RUN: -passes="print<cost-model>" -cost-kind=all 2>&1 -disable-output \
+; RUN: | FileCheck %s --check-prefix=HAS-OPTIMIZED-ZERO-STRIDE-LOAD
+
+; These tests check the costs of broadcast instructions, through the
+; isLegalBroadcastLoad method.
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+
+; The tests use vector loads and splats, as opposed to scalar loads, inserts
+; and splats as that is how getShuffleCost currently recognizes them.
+define void @shuffle() {
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-LABEL: 'shuffle'
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i8 = load <2 x i8>, ptr undef, align 2
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2i8 = shufflevector <2 x i8> %lv2i8, <2 x i8> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4i8 = load <4 x i8>, ptr undef, align 4
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv4i8 = shufflevector <4 x i8> %lv4i8, <4 x i8> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv8i8 = load <8 x i8>, ptr undef, align 8
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv8i8 = shufflevector <8 x i8> %lv8i8, <8 x i8> undef, <8 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv16i8 = load <16 x i8>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv16i8 = shufflevector <16 x i8> %lv16i8, <16 x i8> undef, <16 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i16 = load <2 x i16>, ptr undef, align 4
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2i16 = shufflevector <2 x i16> %lv2i16, <2 x i16> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4i16 = load <4 x i16>, ptr undef, align 8
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv4i16 = shufflevector <4 x i16> %lv4i16, <4 x i16> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv8i16 = load <8 x i16>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv8i16 = shufflevector <8 x i16> %lv8i16, <8 x i16> undef, <8 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv16i16 = load <16 x i16>, ptr undef, align 32
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %sv16i16 = shufflevector <16 x i16> %lv16i16, <16 x i16> undef, <16 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i32 = load <2 x i32>, ptr undef, align 8
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2i32 = shufflevector <2 x i32> %lv2i32, <2 x i32> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4i32 = load <4 x i32>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv4i32 = shufflevector <4 x i32> %lv4i32, <4 x i32> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv8i32 = load <8 x i32>, ptr undef, align 32
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %sv8i32 = shufflevector <8 x i32> %lv8i32, <8 x i32> undef, <8 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i64 = load <2 x i64>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2i64 = shufflevector <2 x i64> %lv2i64, <2 x i64> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv4i64 = load <4 x i64>, ptr undef, align 32
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %sv4i64 = shufflevector <4 x i64> %lv4i64, <4 x i64> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2f16 = load <2 x half>, ptr undef, align 4
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2f16 = shufflevector <2 x half> %lv2f16, <2 x half> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4f16 = load <4 x half>, ptr undef, align 8
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv4f16 = shufflevector <4 x half> %lv4f16, <4 x half> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv8f16 = load <8 x half>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv8f16 = shufflevector <8 x half> %lv8f16, <8 x half> undef, <8 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv16f16 = load <16 x half>, ptr undef, align 32
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %sv16f16 = shufflevector <16 x half> %lv16f16, <16 x half> undef, <16 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2f32 = load <2 x float>, ptr undef, align 8
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2f32 = shufflevector <2 x float> %lv2f32, <2 x float> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4f32 = load <4 x float>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv4f32 = shufflevector <4 x float> %lv4f32, <4 x float> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv8f32 = load <8 x float>, ptr undef, align 32
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %sv8f32 = shufflevector <8 x float> %lv8f32, <8 x float> undef, <8 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2f64 = load <2 x double>, ptr undef, align 16
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 1 for: %sv2f64 = shufflevector <2 x double> %lv2f64, <2 x double> undef, <2 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv4f64 = load <4 x double>, ptr undef, align 32
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:2 SizeLat:2 for: %sv4f64 = shufflevector <4 x double> %lv4f64, <4 x double> undef, <4 x i32> zeroinitializer
+; NO-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
+;
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-LABEL: 'shuffle'
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i8 = load <2 x i8>, ptr undef, align 2
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2i8 = shufflevector <2 x i8> %lv2i8, <2 x i8> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4i8 = load <4 x i8>, ptr undef, align 4
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4i8 = shufflevector <4 x i8> %lv4i8, <4 x i8> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv8i8 = load <8 x i8>, ptr undef, align 8
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv8i8 = shufflevector <8 x i8> %lv8i8, <8 x i8> undef, <8 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv16i8 = load <16 x i8>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv16i8 = shufflevector <16 x i8> %lv16i8, <16 x i8> undef, <16 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i16 = load <2 x i16>, ptr undef, align 4
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2i16 = shufflevector <2 x i16> %lv2i16, <2 x i16> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4i16 = load <4 x i16>, ptr undef, align 8
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4i16 = shufflevector <4 x i16> %lv4i16, <4 x i16> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv8i16 = load <8 x i16>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv8i16 = shufflevector <8 x i16> %lv8i16, <8 x i16> undef, <8 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv16i16 = load <16 x i16>, ptr undef, align 32
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv16i16 = shufflevector <16 x i16> %lv16i16, <16 x i16> undef, <16 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i32 = load <2 x i32>, ptr undef, align 8
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2i32 = shufflevector <2 x i32> %lv2i32, <2 x i32> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4i32 = load <4 x i32>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4i32 = shufflevector <4 x i32> %lv4i32, <4 x i32> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv8i32 = load <8 x i32>, ptr undef, align 32
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv8i32 = shufflevector <8 x i32> %lv8i32, <8 x i32> undef, <8 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2i64 = load <2 x i64>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2i64 = shufflevector <2 x i64> %lv2i64, <2 x i64> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv4i64 = load <4 x i64>, ptr undef, align 32
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4i64 = shufflevector <4 x i64> %lv4i64, <4 x i64> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2f16 = load <2 x half>, ptr undef, align 4
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2f16 = shufflevector <2 x half> %lv2f16, <2 x half> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4f16 = load <4 x half>, ptr undef, align 8
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4f16 = shufflevector <4 x half> %lv4f16, <4 x half> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv8f16 = load <8 x half>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv8f16 = shufflevector <8 x half> %lv8f16, <8 x half> undef, <8 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv16f16 = load <16 x half>, ptr undef, align 32
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv16f16 = shufflevector <16 x half> %lv16f16, <16 x half> undef, <16 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2f32 = load <2 x float>, ptr undef, align 8
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2f32 = shufflevector <2 x float> %lv2f32, <2 x float> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv4f32 = load <4 x float>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4f32 = shufflevector <4 x float> %lv4f32, <4 x float> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv8f32 = load <8 x float>, ptr undef, align 32
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv8f32 = shufflevector <8 x float> %lv8f32, <8 x float> undef, <8 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:1 CodeSize:1 Lat:4 SizeLat:1 for: %lv2f64 = load <2 x double>, ptr undef, align 16
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv2f64 = shufflevector <2 x double> %lv2f64, <2 x double> undef, <2 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:2 CodeSize:1 Lat:4 SizeLat:2 for: %lv4f64 = load <4 x double>, ptr undef, align 32
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of 0 for: %sv4f64 = shufflevector <4 x double> %lv4f64, <4 x double> undef, <4 x i32> zeroinitializer
+; HAS-OPTIMIZED-ZERO-STRIDE-LOAD-NEXT: Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
+;
+ %lv2i8 = load <2 x i8>, ptr undef
+ %sv2i8 = shufflevector <2 x i8> %lv2i8, <2 x i8> undef, <2 x i32> zeroinitializer
+ %lv4i8 = load <4 x i8>, ptr undef
+ %sv4i8 = shufflevector <4 x i8> %lv4i8, <4 x i8> undef, <4 x i32> zeroinitializer
+ %lv8i8 = load <8 x i8>, ptr undef
+ %sv8i8 = shufflevector <8 x i8> %lv8i8, <8 x i8> undef, <8 x i32> zeroinitializer
+ %lv16i8 = load <16 x i8>, ptr undef
+ %sv16i8 = shufflevector <16 x i8> %lv16i8, <16 x i8> undef, <16 x i32> zeroinitializer
+
+ %lv2i16 = load <2 x i16>, ptr undef
+ %sv2i16 = shufflevector <2 x i16> %lv2i16, <2 x i16> undef, <2 x i32> zeroinitializer
+ %lv4i16 = load <4 x i16>, ptr undef
+ %sv4i16 = shufflevector <4 x i16> %lv4i16, <4 x i16> undef, <4 x i32> zeroinitializer
+ %lv8i16 = load <8 x i16>, ptr undef
+ %sv8i16 = shufflevector <8 x i16> %lv8i16, <8 x i16> undef, <8 x i32> zeroinitializer
+ %lv16i16 = load <16 x i16>, ptr undef
+ %sv16i16 = shufflevector <16 x i16> %lv16i16, <16 x i16> undef, <16 x i32> zeroinitializer
+
+ %lv2i32 = load <2 x i32>, ptr undef
+ %sv2i32 = shufflevector <2 x i32> %lv2i32, <2 x i32> undef, <2 x i32> zeroinitializer
+ %lv4i32 = load <4 x i32>, ptr undef
+ %sv4i32 = shufflevector <4 x i32> %lv4i32, <4 x i32> undef, <4 x i32> zeroinitializer
+ %lv8i32 = load <8 x i32>, ptr undef
+ %sv8i32 = shufflevector <8 x i32> %lv8i32, <8 x i32> undef, <8 x i32> zeroinitializer
+
+ %lv2i64 = load <2 x i64>, ptr undef
+ %sv2i64 = shufflevector <2 x i64> %lv2i64, <2 x i64> undef, <2 x i32> zeroinitializer
+ %lv4i64 = load <4 x i64>, ptr undef
+ %sv4i64 = shufflevector <4 x i64> %lv4i64, <4 x i64> undef, <4 x i32> zeroinitializer
+
+ %lv2f16 = load <2 x half>, ptr undef
+ %sv2f16 = shufflevector <2 x half> %lv2f16, <2 x half> undef, <2 x i32> zeroinitializer
+ %lv4f16 = load <4 x half>, ptr undef
+ %sv4f16 = shufflevector <4 x half> %lv4f16, <4 x half> undef, <4 x i32> zeroinitial...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/198446
More information about the llvm-commits
mailing list