Skip to content

[RISCV][CostModel] Estimate cost of Extract/InsertElement with non-constant index when vector instructions are not available #67334

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

Merged
merged 2 commits into from
Sep 27, 2023
Merged

[RISCV][CostModel] Estimate cost of Extract/InsertElement with non-constant index when vector instructions are not available #67334

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
bf83e01
[RISCV][CostModel] Add pre-commit test
skachkov-sc Sep 26, 2023
27394ae
[RISCV][CostModel] Calculate cost of Extract/InsertElement with non-c…
skachkov-sc Sep 25, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 20 additions & 2 deletions llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1463,8 +1463,26 @@ InstructionCost RISCVTTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val,
std::pair<InstructionCost, MVT> LT = getTypeLegalizationCost(Val);

// This type is legalized to a scalar type.
if (!LT.second.isVector())
return 0;
if (!LT.second.isVector()) {
auto *FixedVecTy = cast<FixedVectorType>(Val);
// If Index is a known constant, cost is zero.
if (Index != -1U)
return 0;
// Extract/InsertElement with non-constant index is very costly when
// scalarized; 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.
Type *ElemTy = FixedVecTy->getElementType();
auto NumElems = FixedVecTy->getNumElements();
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;
}

// For unsupported scalable vector.
if (LT.second.isScalableVector() && !LT.first.isValid())
Expand Down
135 changes: 135 additions & 0 deletions llvm/test/Analysis/CostModel/RISCV/extractelement.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,135 @@
; 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: Invalid cost for instruction: %nxv16i8 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i16 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i32 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i64 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i8 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i16 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i32 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16i64 = extractelement <vscale x 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
%nxv16i8 = extractelement <vscale x 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
%nxv16i16 = extractelement <vscale x 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
%nxv16i32 = extractelement <vscale x 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
%nxv16i64 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16f16 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16f32 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16f64 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16f16 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16f32 = extractelement <vscale x 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: Invalid cost for instruction: %nxv16f64 = extractelement <vscale x 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
%nxv16f16 = extractelement <vscale x 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
%nxv16f32 = extractelement <vscale x 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
%nxv16f64 = extractelement <vscale x 16 x double> undef, i32 %x

ret void
}
Loading