Skip to content

[NFC][AMDGPU] IGLP: Fixes for unsigned int handling #135090

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 1 commit into from
Nov 18, 2025
Merged

[NFC][AMDGPU] IGLP: Fixes for unsigned int handling #135090

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
6 changes: 3 additions & 3 deletions llvm/lib/Target/AMDGPU/AMDGPUIGroupLP.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2183,7 +2183,7 @@ bool MFMASmallGemmSingleWaveOpt::applyIGLPStrategy(
SG->initSchedGroup(SyncedInstrs[SG->getSyncID()]);

// Interleave MFMA with DS_READ prefetch
for (unsigned I = 0; I < DSRCount - 4; ++I) {
for (unsigned I = 4; I < DSRCount; ++I) {
SG = &SyncedSchedGroups[PipelineSyncID].emplace_back(
SchedGroupMask::DS_READ, 1, PipelineSyncID, DAG, TII);
SG->initSchedGroup(SyncedInstrs[SG->getSyncID()]);
Expand All @@ -2196,7 +2196,7 @@ bool MFMASmallGemmSingleWaveOpt::applyIGLPStrategy(
// Phase 2a: Loop carried dependency with V_PERM
// Schedule VPerm & DS_WRITE as closely as possible to the VMEM_READ they
// depend on. Interleave MFMA to keep XDL unit busy throughout.
for (unsigned I = 0; I < DSWWithPermCount - DSWWithSharedVMEMCount; ++I) {
for (unsigned I = DSWWithSharedVMEMCount; I < DSWWithPermCount; ++I) {
SG = &SyncedSchedGroups[PipelineSyncID].emplace_back(
SchedGroupMask::VALU, 4, PipelineSyncID, DAG, TII);
SG->addRule(std::make_shared<IsPermForDSW>(TII, SG->getSGID(), true));
Expand Down Expand Up @@ -2233,7 +2233,7 @@ bool MFMASmallGemmSingleWaveOpt::applyIGLPStrategy(
// Phase 2b: Loop carried dependency without V_PERM
// Schedule DS_WRITE as closely as possible to the VMEM_READ they depend on.
// Interleave MFMA to keep XDL unit busy throughout.
for (unsigned I = 0; I < DSWCount - DSWWithPermCount; I++) {
for (unsigned I = DSWWithPermCount; I < DSWCount; I++) {
SG = &SyncedSchedGroups[PipelineSyncID].emplace_back(
SchedGroupMask::DS_WRITE, 1, PipelineSyncID, DAG, TII);
SG->initSchedGroup(SyncedInstrs[SG->getSyncID()]);
Expand Down
Loading