llvm-project
[AMDGPU]: Rewrite mbcnt_lo/mbcnt_hi to work item ID where applicable
#160496
Merged
Merged
[AMDGPU]: Rewrite mbcnt_lo/mbcnt_hi to work item ID where applicable #160496
arsenm
merged
14 commits into
llvm:main
from
TejaX-Alaghari:mbcnt-to-workitem
jayfoad
commented
on 2025-09-24
jayfoad
commented
on 2025-09-24
krzysz00
commented
on 2025-09-29
jmmartinez
approved these changes
on 2025-10-29
[AMDGPU] Move mbcnt optimization from InstCombine to AMDGPUCodeGenPre…
fbe9999b
[AMDGPU] Enhance mbcnt.lo optimization to support bitmask cases
c07893f7
Fix mbcnt.hi wave32 optimization to respect reqd_work_group_size meta…
e75a2b9c
Address review feedback: Add periods to all comments
0f7355ef
Address review comments from @arsenm
1468263a
Refactored test cases and implementation comments
0e69f69d
Simplified the implementation per review suggestions -
f1eda342
Addressed feedback on -
0759c5b6
Remove redundant isPower2_32 check and combine 2 match functions in v…
e908b3e7
Removed 4*Wave case from the tests
7708c9bd
Simplified negative cases test file as the arch specific cases are mo…
9863672b
Address feedback about -
1e5ccac9
Fixed format issue in AMDGPUCodeGenPrepare.cpp
db1b8934
Refactor the code to remove redundant logic
4938371d
arsenm
approved these changes
on 2025-12-18
Reviewers
Assignees
No one assigned
Labels
backend:AMDGPU
llvm:instcombine
llvm:transforms
Milestone
No milestone
Login to write a write a comment.
Login via GitHub