[PATCH] D115559: AMDGPU: Propagate amdgpu-waves-per-eu with attributor

[Matt Arsenault via Phabricator via llvm-commits]

arsenm created this revision.
arsenm added reviewers: rampitec, jdoerfert, kzhuravl, t-tye, yaxunl.
Herald added subscribers: foad, okura, kuter, uenoku, kerbowa, hiraditya, tpr, dstuttard, nhaehnle, jvesely.
Herald added a reviewer: uenoku.
arsenm requested review of this revision.
Herald added a subscriber: wdng.
Herald added a reviewer: sstefan1.
Herald added a reviewer: baziotis.
Herald added a project: LLVM.

This will do a value range merging down the callgraph, unlike the
current pass which can only propagate values to undecorated functions
from a kernel.

      

This one is a bit weird due to the interaction with the implied range
from amdgpu-flat-workgroup-size. At the default group range of 1,1024,
the minimum implied bounds is 4 so this ends up introducing the
attribute on undecorated functions. We could probably simplify this by
ignoring it and propagating the raw values. The subtarget interaction
and the interaction with amdgpu-flat-workgroup-size only really clamp
invalid values (plus the lower bound doesn't seem to do anything as
far as I can tell anyway).


https://reviews.llvm.org/D115559

Files:
  llvm/lib/Target/AMDGPU/AMDGPUAttributor.cpp
  llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
  llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
  llvm/lib/Target/AMDGPU/GCNSubtarget.h
  llvm/test/CodeGen/AMDGPU/addrspacecast-constantexpr.ll
  llvm/test/CodeGen/AMDGPU/annotate-existing-abi-attributes.ll
  llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa-call.ll
  llvm/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll
  llvm/test/CodeGen/AMDGPU/direct-indirect-call.ll
  llvm/test/CodeGen/AMDGPU/duplicate-attribute-indirect.ll
  llvm/test/CodeGen/AMDGPU/propagate-flat-work-group-size.ll
  llvm/test/CodeGen/AMDGPU/propagate-waves-per-eu.ll
  llvm/test/CodeGen/AMDGPU/simple-indirect-call.ll
  llvm/test/CodeGen/AMDGPU/uniform-work-group-attribute-missing.ll
  llvm/test/CodeGen/AMDGPU/uniform-work-group-multistep.ll
  llvm/test/CodeGen/AMDGPU/uniform-work-group-nested-function-calls.ll
  llvm/test/CodeGen/AMDGPU/uniform-work-group-prevent-attribute-propagation.ll
  llvm/test/CodeGen/AMDGPU/uniform-work-group-recursion-test.ll
  llvm/test/CodeGen/AMDGPU/uniform-work-group-test.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D115559.393613.patch
Type: text/x-patch
Size: 75580 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211210/a6379441/attachment.bin>