[PATCH] D83674: [AMDGPU] Calculate minimum allowed occupancy based on threads per lane

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec requested changes to this revision.
rampitec added inline comments.
This revision now requires changes to proceed.


================
Comment at: llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp:200
+  // - 16 waves per SIMD for GFX10 wave32
+  unsigned MinOccupancy = ST.getTotalNumVGPRs() / 64;
+  return std::min(Occupancy, MinOccupancy);
----------------
This is not expressable in a number of VGPRs. The limiter is about memory and not about register budget at all. Meanwhile it looks like 8 waves might be more reasonable for wave32, but wave64 shall still use the same value across the targets, because at the end of the day it boils down to a number of concurrent memory operations.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D83674/new/

https://reviews.llvm.org/D83674