Skip to content

[AMDGPU] Simplify VOP3PWMMA_Profile. NFC. #79377

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
rampitec merged 1 commit into from
Jan 24, 2024
Merged

[AMDGPU] Simplify VOP3PWMMA_Profile. NFC. #79377

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
39 changes: 7 additions & 32 deletions llvm/lib/Target/AMDGPU/VOP3PInstructions.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -998,39 +998,14 @@ class VOP3PWMMA_Profile<list<ValueType> ArgTy, bit _IsSWMMAC, int _IndexType,
bit NegLoAny = !or(NegLo01, NegLo2);
bit NegHiAny = !or(NegHi01, NegHi2);

let DstRC = !cond(!eq(ArgTy[0], v8f32): VDst_256,
!eq(ArgTy[0], v8i32): VDst_256,
!eq(ArgTy[0], v8f16): VDst_128,
!eq(ArgTy[0], v8i16): VDst_128,
!eq(ArgTy[0], v4f32): VDst_128,
!eq(ArgTy[0], v4i32): VDst_128,
!eq(ArgTy[0], v4f16): VDst_64,
!eq(ArgTy[0], v4i16): VDst_64);
let Src0RC64 = !cond(!eq(ArgTy[1], v8f16): VRegSrc_128,
!eq(ArgTy[1], v4f16): VRegSrc_64,
!eq(ArgTy[1], v4i16): VRegSrc_64,
!eq(ArgTy[1], v8i16): VRegSrc_128,
!eq(ArgTy[1], v4i32): VRegSrc_128,
!eq(ArgTy[1], v2i32): VRegSrc_64,
!eq(ArgTy[1], i32) : VRegSrc_32);
let Src1RC64 = !cond(!eq(ArgTy[2], v16f16): VRegSrc_256,
!eq(ArgTy[2], v16i16): VRegSrc_256,
!eq(ArgTy[2], v8f16): VRegSrc_128,
!eq(ArgTy[2], v8i16): VRegSrc_128,
!eq(ArgTy[2], v4i32): VRegSrc_128,
!eq(ArgTy[1], v4i16): VRegSrc_64,
!eq(ArgTy[1], v4f16): VRegSrc_64,
!eq(ArgTy[2], v2i32): VRegSrc_64,
!eq(ArgTy[2], i32) : VRegSrc_32);
let DstRC = !cast<RegisterOperand>("VDst_"#ArgTy[0].Size);
let Src0RC64 = !cast<RegisterOperand>("VRegSrc_"#ArgTy[1].Size);
let Src1RC64 = !cast<RegisterOperand>("VRegSrc_"#ArgTy[2].Size);
let Src2RC64 = !if(IsSWMMAC, DstRC,
!cond(!eq(ArgTy[3], v8f32): VISrc_256_f32,
!eq(ArgTy[3], v8i32): VISrc_256_b32,
!eq(ArgTy[3], v8f16): VISrc_128_f16,
!eq(ArgTy[3], v8i16): VISrc_128_f32, // bf16
!eq(ArgTy[3], v4f16): VISrc_64_f16,
!eq(ArgTy[3], v4i16): VISrc_64_b32,
!eq(ArgTy[3], v4i32): VISrc_128_b32,
!eq(ArgTy[3], v4f32): VISrc_128_f32));
!cast<RegisterOperand>("VISrc_"#ArgTy[3].Size#
!cond(IsC_F32: "_f32",
IsC_F16: "_f16",
1: "_b32")));

// For f16 and bf16 matrices A and B, each element can be modified by
// fneg(neg_lo,neg_hi = 1). For iu4 and iu8 matrices A and B neg_lo is
Expand Down