[llvm] [RISCV][CostModel] Estimate cost of Extract/InsertElement with non-constant index when vector instructions are not available (PR #67334)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 25 07:21:59 PDT 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
<details>
<summary>Changes</summary>
This patch fixes the compilation time issue of matrix-types-spec test from test-suite.
Reproduction of the problem:
```
clang++ -DNDEBUG --target=riscv64-linux-gnu --sysroot=<sysroot path> --gcc-toolchain=<gcc path> -O2 -fenable-matrix <test-suite-path>/SingleSource/UnitTests/matrix-types-spec.cpp
```
On my machine, compilation takes 50.44s. In comparison, the same test with RVV (-march=rv64gcv) compiles in 3.06s, and for x86-64 target it takes 1.71s. It turns out that the main issue is unrolling of loop in multiplySpec function, that has extractelements with non-constant index:
```
for.body9.i: ; preds = %for.body9.i, %for.cond6.preheader.i
%indvars.iv.i92 = phi i64 [ 0, %for.cond6.preheader.i ], [ %indvars.iv.next.i93, %for.body9.i ]
%Elt.033.i = phi double [ 0.000000e+00, %for.cond6.preheader.i ], [ %80, %for.body9.i ]
%77 = mul nuw nsw i64 %indvars.iv.i92, 25
%78 = add nuw nsw i64 %77, %indvars.iv39.i91
%matrixext.i = extractelement <475 x double> %62, i64 %78
%79 = add nuw nsw i64 %indvars.iv.i92, %74
%matrixext13.i = extractelement <209 x double> %73, i64 %79
%80 = tail call double @<!-- -->llvm.fmuladd.f64(double %matrixext.i, double %matrixext13.i, double %Elt.033.i)
%indvars.iv.next.i93 = add nuw nsw i64 %indvars.iv.i92, 1
%exitcond.not.i94 = icmp eq i64 %indvars.iv.next.i93, 19
br i1 %exitcond.not.i94, label %for.cond.cleanup8.i, label %for.body9.i, !llvm.loop !21
```
When RVV is supported, extractelement/insertelement with non-constant index can be lowered quite efficiently with vslidedown/vslideup; otherwise it's lowered via stack memory operations, i.e. for extractelement each vector element is stored on stack and then the needed element is loaded back; for insertelement is stores all vector elements, rewrites the required element value and then loads vector back. Currently the cost of such expensive operation is estimated as zero, so loop unroll processes the loop more aggresively. The proper estimation of cost (in a way like in X86 target) prohibits unrolling of this loop and fixes compilation time (2.77s on my machine).
---
Patch is 21.08 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/67334.diff
3 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+18)
- (added) llvm/test/Analysis/CostModel/RISCV/extractelement.ll (+114)
- (added) llvm/test/Analysis/CostModel/RISCV/insertelement.ll (+114)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index 4a34362a792f150..2df4ea77033f1c6 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1456,6 +1456,24 @@ InstructionCost RISCVTTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
Opcode != Instruction::InsertElement)
return BaseT::getVectorInstrCost(Opcode, Val, CostKind, Index, Op0, Op1);
+ // Extract/InsertElement with non-constant index is very costly without
+ // V instructions; estimate cost of loads/stores sequence via the stack:
+ // ExtractElement cost: store vector to stack, load scalar;
+ // InsertElement cost: store vector to stack, store scalar, load vector.
+ if (Index == -1U && !ST->hasVInstructions()) {
+ auto *VecTy = cast<VectorType>(Val);
+ auto *ElemTy = VecTy->getElementType();
+ auto NumElems = VecTy->getElementCount().getKnownMinValue();
+ auto Align = DL.getPrefTypeAlign(ElemTy);
+ InstructionCost LoadCost =
+ getMemoryOpCost(Instruction::Load, ElemTy, Align, 0, CostKind);
+ InstructionCost StoreCost =
+ getMemoryOpCost(Instruction::Store, ElemTy, Align, 0, CostKind);
+ return Opcode == Instruction::ExtractElement
+ ? StoreCost * NumElems + LoadCost
+ : (StoreCost + LoadCost) * NumElems + StoreCost;
+ }
+
// Legalize the type.
std::pair<InstructionCost, MVT> LT = getTypeLegalizationCost(Val);
diff --git a/llvm/test/Analysis/CostModel/RISCV/extractelement.ll b/llvm/test/Analysis/CostModel/RISCV/extractelement.ll
new file mode 100644
index 000000000000000..8f9275f717f1176
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/extractelement.ll
@@ -0,0 +1,114 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 3
+; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=riscv32 -mattr=+f,+d,+zfh < %s | FileCheck %s --check-prefixes=RV32
+; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=riscv64 -mattr=+f,+d,+zfh < %s | FileCheck %s --check-prefixes=RV64
+
+define void @extractelement_int(i32 %x) {
+; RV32-LABEL: 'extractelement_int'
+; RV32-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8 = extractelement <2 x i8> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i8 = extractelement <4 x i8> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i8 = extractelement <8 x i8> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i8 = extractelement <16 x i8> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i16 = extractelement <2 x i16> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i16 = extractelement <4 x i16> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i16 = extractelement <8 x i16> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i16 = extractelement <16 x i16> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i32 = extractelement <2 x i32> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i32 = extractelement <4 x i32> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i32 = extractelement <8 x i32> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i32 = extractelement <16 x i32> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %v2i64 = extractelement <2 x i64> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %v4i64 = extractelement <4 x i64> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %v8i64 = extractelement <8 x i64> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %v16i64 = extractelement <16 x i64> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; RV64-LABEL: 'extractelement_int'
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i8 = extractelement <2 x i8> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i8 = extractelement <4 x i8> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i8 = extractelement <8 x i8> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i8 = extractelement <16 x i8> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i16 = extractelement <2 x i16> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i16 = extractelement <4 x i16> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i16 = extractelement <8 x i16> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i16 = extractelement <16 x i16> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i32 = extractelement <2 x i32> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i32 = extractelement <4 x i32> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i32 = extractelement <8 x i32> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i32 = extractelement <16 x i32> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2i64 = extractelement <2 x i64> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4i64 = extractelement <4 x i64> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8i64 = extractelement <8 x i64> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16i64 = extractelement <16 x i64> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %v2i8 = extractelement <2 x i8> undef, i32 %x
+ %v4i8 = extractelement <4 x i8> undef, i32 %x
+ %v8i8 = extractelement <8 x i8> undef, i32 %x
+ %v16i8 = extractelement <16 x i8> undef, i32 %x
+
+ %v2i16 = extractelement <2 x i16> undef, i32 %x
+ %v4i16 = extractelement <4 x i16> undef, i32 %x
+ %v8i16 = extractelement <8 x i16> undef, i32 %x
+ %v16i16 = extractelement <16 x i16> undef, i32 %x
+
+ %v2i32 = extractelement <2 x i32> undef, i32 %x
+ %v4i32 = extractelement <4 x i32> undef, i32 %x
+ %v8i32 = extractelement <8 x i32> undef, i32 %x
+ %v16i32 = extractelement <16 x i32> undef, i32 %x
+
+ %v2i64 = extractelement <2 x i64> undef, i32 %x
+ %v4i64 = extractelement <4 x i64> undef, i32 %x
+ %v8i64 = extractelement <8 x i64> undef, i32 %x
+ %v16i64 = extractelement <16 x i64> undef, i32 %x
+
+ ret void
+}
+
+define void @extractelement_fp(i32 %x) {
+; RV32-LABEL: 'extractelement_fp'
+; RV32-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2f16 = extractelement <2 x half> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4f16 = extractelement <4 x half> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8f16 = extractelement <8 x half> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16f16 = extractelement <16 x half> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2f32 = extractelement <2 x float> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4f32 = extractelement <4 x float> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8f32 = extractelement <8 x float> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16f32 = extractelement <16 x float> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2f64 = extractelement <2 x double> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4f64 = extractelement <4 x double> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8f64 = extractelement <8 x double> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16f64 = extractelement <16 x double> undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; RV64-LABEL: 'extractelement_fp'
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2f16 = extractelement <2 x half> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4f16 = extractelement <4 x half> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8f16 = extractelement <8 x half> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16f16 = extractelement <16 x half> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2f32 = extractelement <2 x float> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4f32 = extractelement <4 x float> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8f32 = extractelement <8 x float> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16f32 = extractelement <16 x float> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v2f64 = extractelement <2 x double> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v4f64 = extractelement <4 x double> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v8f64 = extractelement <8 x double> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v16f64 = extractelement <16 x double> undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %v2f16 = extractelement <2 x half> undef, i32 %x
+ %v4f16 = extractelement <4 x half> undef, i32 %x
+ %v8f16 = extractelement <8 x half> undef, i32 %x
+ %v16f16 = extractelement <16 x half> undef, i32 %x
+
+ %v2f32 = extractelement <2 x float> undef, i32 %x
+ %v4f32 = extractelement <4 x float> undef, i32 %x
+ %v8f32 = extractelement <8 x float> undef, i32 %x
+ %v16f32 = extractelement <16 x float> undef, i32 %x
+
+ %v2f64 = extractelement <2 x double> undef, i32 %x
+ %v4f64 = extractelement <4 x double> undef, i32 %x
+ %v8f64 = extractelement <8 x double> undef, i32 %x
+ %v16f64 = extractelement <16 x double> undef, i32 %x
+
+ ret void
+}
diff --git a/llvm/test/Analysis/CostModel/RISCV/insertelement.ll b/llvm/test/Analysis/CostModel/RISCV/insertelement.ll
new file mode 100644
index 000000000000000..4cdeaf26e19e4d1
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/RISCV/insertelement.ll
@@ -0,0 +1,114 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 3
+; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=riscv32 -mattr=+f,+d,+zfh < %s | FileCheck %s --check-prefixes=RV32
+; RUN: opt -passes="print<cost-model>" 2>&1 -disable-output -mtriple=riscv64 -mattr=+f,+d,+zfh < %s | FileCheck %s --check-prefixes=RV64
+
+define void @insertelement_int(i32 %x) {
+; RV32-LABEL: 'insertelement_int'
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8 = insertelement <2 x i8> undef, i8 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i8 = insertelement <4 x i8> undef, i8 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i8 = insertelement <8 x i8> undef, i8 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i8 = insertelement <16 x i8> undef, i8 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i16 = insertelement <2 x i16> undef, i16 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i16 = insertelement <4 x i16> undef, i16 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i16 = insertelement <8 x i16> undef, i16 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i16 = insertelement <16 x i16> undef, i16 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i32 = insertelement <2 x i32> undef, i32 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i32 = insertelement <4 x i32> undef, i32 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i32 = insertelement <8 x i32> undef, i32 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i32 = insertelement <16 x i32> undef, i32 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %v2i64 = insertelement <2 x i64> undef, i64 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %v4i64 = insertelement <4 x i64> undef, i64 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %v8i64 = insertelement <8 x i64> undef, i64 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %v16i64 = insertelement <16 x i64> undef, i64 undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; RV64-LABEL: 'insertelement_int'
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i8 = insertelement <2 x i8> undef, i8 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i8 = insertelement <4 x i8> undef, i8 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i8 = insertelement <8 x i8> undef, i8 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i8 = insertelement <16 x i8> undef, i8 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i16 = insertelement <2 x i16> undef, i16 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i16 = insertelement <4 x i16> undef, i16 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i16 = insertelement <8 x i16> undef, i16 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i16 = insertelement <16 x i16> undef, i16 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i32 = insertelement <2 x i32> undef, i32 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i32 = insertelement <4 x i32> undef, i32 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i32 = insertelement <8 x i32> undef, i32 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i32 = insertelement <16 x i32> undef, i32 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2i64 = insertelement <2 x i64> undef, i64 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4i64 = insertelement <4 x i64> undef, i64 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8i64 = insertelement <8 x i64> undef, i64 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16i64 = insertelement <16 x i64> undef, i64 undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ %v2i8 = insertelement <2 x i8> undef, i8 undef, i32 %x
+ %v4i8 = insertelement <4 x i8> undef, i8 undef, i32 %x
+ %v8i8 = insertelement <8 x i8> undef, i8 undef, i32 %x
+ %v16i8 = insertelement <16 x i8> undef, i8 undef, i32 %x
+
+ %v2i16 = insertelement <2 x i16> undef, i16 undef, i32 %x
+ %v4i16 = insertelement <4 x i16> undef, i16 undef, i32 %x
+ %v8i16 = insertelement <8 x i16> undef, i16 undef, i32 %x
+ %v16i16 = insertelement <16 x i16> undef, i16 undef, i32 %x
+
+ %v2i32 = insertelement <2 x i32> undef, i32 undef, i32 %x
+ %v4i32 = insertelement <4 x i32> undef, i32 undef, i32 %x
+ %v8i32 = insertelement <8 x i32> undef, i32 undef, i32 %x
+ %v16i32 = insertelement <16 x i32> undef, i32 undef, i32 %x
+
+ %v2i64 = insertelement <2 x i64> undef, i64 undef, i32 %x
+ %v4i64 = insertelement <4 x i64> undef, i64 undef, i32 %x
+ %v8i64 = insertelement <8 x i64> undef, i64 undef, i32 %x
+ %v16i64 = insertelement <16 x i64> undef, i64 undef, i32 %x
+
+ ret void
+}
+
+define void @insertelement_fp(i32 %x) {
+; RV32-LABEL: 'insertelement_fp'
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2f16 = insertelement <2 x half> undef, half undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4f16 = insertelement <4 x half> undef, half undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8f16 = insertelement <8 x half> undef, half undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16f16 = insertelement <16 x half> undef, half undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2f32 = insertelement <2 x float> undef, float undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4f32 = insertelement <4 x float> undef, float undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8f32 = insertelement <8 x float> undef, float undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16f32 = insertelement <16 x float> undef, float undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2f64 = insertelement <2 x double> undef, double undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4f64 = insertelement <4 x double> undef, double undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8f64 = insertelement <8 x double> undef, double undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16f64 = insertelement <16 x double> undef, double undef, i32 %x
+; RV32-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+; RV64-LABEL: 'insertelement_fp'
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2f16 = insertelement <2 x half> undef, half undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4f16 = insertelement <4 x half> undef, half undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8f16 = insertelement <8 x half> undef, half undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16f16 = insertelement <16 x half> undef, half undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2f32 = insertelement <2 x float> undef, float undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4f32 = insertelement <4 x float> undef, float undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %v8f32 = insertelement <8 x float> undef, float undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %v16f32 = insertelement <16 x float> undef, float undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %v2f64 = insertelement <2 x double> undef, double undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %v4f64 = insertelement <4 x double> undef, double undef, i32 %x
+; RV64-NEXT: Cost Model: Found an estimated cost of 17 for inst...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/67334
More information about the llvm-commits
mailing list