This is an archive of the discontinued LLVM Phabricator instance.

Differential D59536

[AMDGPU] Add convergent attribute to WWM.
ClosedPublic

Authored by sheredom on Mar 19 2019, 1:44 AM.

Details

Reviewers
arsenm
nhaehnle
cwabbott
Commits
rG47c2bd2b349f: [AMDGPU] Add convergent attribute to WWM.
rL356470: [AMDGPU] Add convergent attribute to WWM.
Summary

Add the convergent attribute to the WWM intrinsic to stop it ever being sunk out of cfg.

Diff Detail

Event Timeline

sheredom created this revision.Mar 19 2019, 1:44 AM
Herald added a project: Restricted Project. · View Herald TranscriptMar 19 2019, 1:44 AM
Herald added subscribers: llvm-commits, t-tye, tpr and 5 others. · View Herald Transcript
arsenm accepted this revision.Mar 19 2019, 7:14 AM

LGTM, though a test wouldn't hurt

This revision is now accepted and ready to land.Mar 19 2019, 7:14 AM
Closed by commit rL356470: [AMDGPU] Add convergent attribute to WWM. (authored by sheredom). · Explain WhyMar 19 2019, 9:31 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
include/
llvm/
IR/
2 lines

Diff 191259

include/llvm/IR/IntrinsicsAMDGPU.td

Show First 20 Lines • Show All 1,355 Lines • ▼ Show 20 Lines
def int_amdgcn_kill : Intrinsic<[], [llvm_i1_ty], []>; def int_amdgcn_kill : Intrinsic<[], [llvm_i1_ty], []>;
// Copies the active channels of the source value to the destination value, // Copies the active channels of the source value to the destination value,
// with the guarantee that the source value is computed as if the entire // with the guarantee that the source value is computed as if the entire
// program were executed in Whole Wavefront Mode, i.e. with all channels // program were executed in Whole Wavefront Mode, i.e. with all channels
// enabled, with a few exceptions: - Phi nodes with require WWM return an // enabled, with a few exceptions: - Phi nodes with require WWM return an
// undefined value. // undefined value.
def int_amdgcn_wwm : Intrinsic<[llvm_any_ty], def int_amdgcn_wwm : Intrinsic<[llvm_any_ty],
[LLVMMatchType<0>], [IntrNoMem, IntrSpeculatable] [LLVMMatchType<0>], [IntrNoMem, IntrSpeculatable, IntrConvergent]
>; >;
// Given a value, copies it while setting all the inactive lanes to a given // Given a value, copies it while setting all the inactive lanes to a given
// value. Note that OpenGL helper lanes are considered active, so if the // value. Note that OpenGL helper lanes are considered active, so if the
// program ever uses WQM, then the instruction and the first source will be // program ever uses WQM, then the instruction and the first source will be
// computed in WQM. // computed in WQM.
def int_amdgcn_set_inactive : def int_amdgcn_set_inactive :
Intrinsic<[llvm_anyint_ty], Intrinsic<[llvm_anyint_ty],
▲ Show 20 Lines • Show All 203 Lines • Show Last 20 Lines