[all-commits] [llvm/llvm-project] df10cc: [AMDGPU] Rework VOPD constraints for gfx12+ with d...
Joe Nash via All-commits
all-commits at lists.llvm.org
Wed May 6 12:26:21 PDT 2026
Branch: refs/heads/main
Home: https://github.com/llvm/llvm-project
Commit: df10cc9b0566443dab78ba546785f370f9c1b78d
https://github.com/llvm/llvm-project/commit/df10cc9b0566443dab78ba546785f370f9c1b78d
Author: Joe Nash <joseph.nash at amd.com>
Date: 2026-05-06 (Wed, 06 May 2026)
Changed paths:
M llvm/lib/Target/AMDGPU/GCNCreateVOPD.cpp
M llvm/lib/Target/AMDGPU/GCNVOPDUtils.cpp
M llvm/lib/Target/AMDGPU/GCNVOPDUtils.h
M llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
M llvm/lib/Target/AMDGPU/SIInstrInfo.h
M llvm/test/CodeGen/AMDGPU/GlobalISel/fshl.ll
M llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i16.ll
M llvm/test/CodeGen/AMDGPU/GlobalISel/insertelement.i8.ll
M llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.load.1d.d16.ll
M llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.image.load.1d.ll
M llvm/test/CodeGen/AMDGPU/GlobalISel/load-uniform-in-vgpr.ll
M llvm/test/CodeGen/AMDGPU/GlobalISel/shl.ll
M llvm/test/CodeGen/AMDGPU/addrspacecast-gas.ll
M llvm/test/CodeGen/AMDGPU/amdgcn.bitcast.1024bit.ll
M llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll
M llvm/test/CodeGen/AMDGPU/atomics-system-scope.ll
M llvm/test/CodeGen/AMDGPU/bf16.ll
M llvm/test/CodeGen/AMDGPU/expand-waitcnt-profiling.ll
M llvm/test/CodeGen/AMDGPU/fcopysign.f16.ll
M llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll
M llvm/test/CodeGen/AMDGPU/global_atomics.ll
M llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.buffer.load.format.ll
M llvm/test/CodeGen/AMDGPU/mad_64_32.ll
M llvm/test/CodeGen/AMDGPU/match-perm-extract-vector-elt-bug.ll
M llvm/test/CodeGen/AMDGPU/narrow_math_for_and.ll
M llvm/test/CodeGen/AMDGPU/vopd-combine-gfx1250.mir
M llvm/test/CodeGen/AMDGPU/vopd-combine.mir
A llvm/test/CodeGen/AMDGPU/vopd-gfx12-f64-relaxed.mir
Log Message:
-----------
[AMDGPU] Rework VOPD constraints for gfx12+ with data deps (#191264)
Follow-up to #178772. Relax the constraint that blocks VOPD formation
when SecondMI writes to registers that FirstMI reads from, except if the
resulting VOPD would take multiple cycles to issue. That can happen if
the same source VGPR is used in the same position in the other Op
(AllowSameVGPR), or if one of the following opcodes is used for OpX:
- v_fma_f64
- v_add_f64
- v_mul_f64
- v_max_num_f64
- v_min_num_f64
De-duplicate the check for which instructions can be paired in the
scheduling and formation passes, and use the same check logic in both
passes (previously scheduling was looser).
---------
Co-authored-by: Claude Opus 4.6 <noreply at anthropic.com>
To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications
More information about the All-commits
mailing list