[PATCH] D156301: [WIP] Support FP global atomics in AMDGPUAtomicOptimizer.

[Pravin Jagtap via Phabricator via llvm-commits]

pravinjagtap created this revision.
pravinjagtap added a reviewer: arsenm.
Herald added subscribers: foad, kerbowa, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl.
Herald added a project: All.
pravinjagtap requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

Completed:

1. Uniform & Divergent input values when intermediate scan result are not needed later in the kernel computations
2. Uniform input value when intermediate scan results are needed later in the kernel computations. (V_WRITELANE_B32 865 is not required in this case)

TODO:

1. Divergent value input when intermediate scan results are needed later in the kernel computations. (This case needs V_WRITELANE_B32 for float values)
2. Float64 support in all cases.
3. Fix lit tests


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D156301

Files:
  llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
  llvm/test/CodeGen/AMDGPU/global_atomics_iterative_scan_fp.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D156301.544260.patch
Type: text/x-patch
Size: 9737 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230726/2bd7fb58/attachment.bin>