Skip to content

[RISCV][TTI] Cost non-power-of-two size changing casts #101047

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
Aug 13, 2024
Merged

[RISCV][TTI] Cost non-power-of-two size changing casts #101047

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
6b6f3c3
[RISCV][TTI] Cost non-power-of-two size changing casts
preames Jul 29, 2024
facafef
Address review comment
preames Jul 30, 2024
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
23 changes: 17 additions & 6 deletions llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1030,17 +1030,28 @@ InstructionCost RISCVTTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
if (!IsVectorType)
return BaseT::getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I);

bool IsTypeLegal = isTypeLegal(Src) && isTypeLegal(Dst) &&
(Src->getScalarSizeInBits() <= ST->getELen()) &&
(Dst->getScalarSizeInBits() <= ST->getELen());

// FIXME: Need to compute legalizing cost for illegal types.
if (!IsTypeLegal)
// FIXME: Need to compute legalizing cost for illegal types. The current
// code handles only legal types and those which can be trivially
// promoted to legal.
if (!ST->hasVInstructions() || Src->getScalarSizeInBits() > ST->getELen() ||
Dst->getScalarSizeInBits() > ST->getELen())
return BaseT::getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I);

std::pair<InstructionCost, MVT> SrcLT = getTypeLegalizationCost(Src);
std::pair<InstructionCost, MVT> DstLT = getTypeLegalizationCost(Dst);

// Our actual lowering for the case where a wider legal type is available
// uses promotion to the wider type. This is reflected in the result of
// getTypeLegalizationCost, but BasicTTI assumes the widened cases are
// scalarized if the legalized Src and Dst are not equal sized.
const DataLayout &DL = this->getDataLayout();
if (!SrcLT.second.isVector() || !DstLT.second.isVector() ||
!TypeSize::isKnownLE(DL.getTypeSizeInBits(Src),
SrcLT.second.getSizeInBits()) ||
!TypeSize::isKnownLE(DL.getTypeSizeInBits(Dst),
DstLT.second.getSizeInBits()))
return BaseT::getCastInstrCost(Opcode, Dst, Src, CCH, CostKind, I);

Comment on lines +1047 to +1054
Copy link
Contributor

Choose a reason for hiding this comment

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

Before the patch, the split cost is done by BasicTTIImplBase::getCastInstrCost.
Could we add a check SrcLT.first > 1 || DstLT.first > 1 here to keep the split cost since isTypeLegal() is removed?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

The size checks should catch this case. The only way it wouldn't would be if we split to larger types; that would be quite weird.

int ISD = TLI->InstructionOpcodeToISD(Opcode);
assert(ISD && "Invalid opcode");

Expand Down
30 changes: 15 additions & 15 deletions llvm/test/Analysis/CostModel/RISCV/cast.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4317,24 +4317,24 @@ define void @uitofp() {

define void @oddvec_sizes() {
; CHECK-LABEL: 'oddvec_sizes'
; CHECK-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %1 = sext <3 x i8> undef to <3 x i16>
; CHECK-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %2 = sext <7 x i8> undef to <7 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %3 = sext <15 x i8> undef to <15 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %4 = zext <3 x i8> undef to <3 x i16>
; CHECK-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %5 = zext <7 x i8> undef to <7 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %6 = zext <15 x i8> undef to <15 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %7 = trunc <3 x i32> undef to <3 x i8>
; CHECK-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %8 = trunc <7 x i32> undef to <7 x i8>
; CHECK-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %9 = trunc <15 x i32> undef to <15 x i8>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %1 = sext <3 x i8> undef to <3 x i16>
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %2 = sext <7 x i8> undef to <7 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %3 = sext <15 x i8> undef to <15 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %4 = zext <3 x i8> undef to <3 x i16>
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %5 = zext <7 x i8> undef to <7 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %6 = zext <15 x i8> undef to <15 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %7 = trunc <3 x i32> undef to <3 x i8>
; CHECK-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %8 = trunc <7 x i32> undef to <7 x i8>
; CHECK-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %9 = trunc <15 x i32> undef to <15 x i8>
Copy link
Collaborator Author

@preames preames Jul 29, 2024 •
edited
Loading

Choose a reason for hiding this comment

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

This exposes a bug in the base costing for truncs - we're using the destination LMUL where we should be using the source LMUL (since it's wider). Will fix that separately. (Edit, see #101051)

For the fp<->int cases a few lines below, that's the known lack of LMUL in the cost model. See #87506

; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %10 = bitcast <3 x i32> undef to <3 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %11 = bitcast <7 x i32> undef to <7 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %12 = bitcast <15 x i32> undef to <15 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %13 = sitofp <3 x i32> undef to <3 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %14 = sitofp <7 x i32> undef to <7 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %15 = sitofp <15 x i32> undef to <15 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %16 = uitofp <3 x i32> undef to <3 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %17 = uitofp <7 x i32> undef to <7 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %18 = uitofp <15 x i32> undef to <15 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %13 = sitofp <3 x i32> undef to <3 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %14 = sitofp <7 x i32> undef to <7 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %15 = sitofp <15 x i32> undef to <15 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %16 = uitofp <3 x i32> undef to <3 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %17 = uitofp <7 x i32> undef to <7 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %18 = uitofp <15 x i32> undef to <15 x float>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %19 = fptosi <3 x float> undef to <3 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %20 = fptosi <7 x float> undef to <7 x i32>
; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %21 = fptosi <15 x float> undef to <15 x i32>
Expand Down
Loading