[PATCH] D120499: [NVPTX] Fix nvvm.match.sync*.i64 intrinsics return type (i64 -> i32)

[Kristina Bessonova via Phabricator via cfe-commits]

krisb created this revision.
krisb added a reviewer: tra.
Herald added subscribers: asavonic, hiraditya.
krisb requested review of this revision.
Herald added subscribers: llvm-commits, cfe-commits, jdoerfert.
Herald added projects: clang, LLVM.

NVVM IR specification defines them with i32 return type [0]:

  The following intrinsics synchronize a subset of threads in a warp and then
  broadcast and compare a value across threads in the subset.
  
  declare i32 @llvm.nvvm.match.any.sync.i64(i32 %membermask, i64 %value)
  declare {i32, i1} @llvm.nvvm.match.all.sync.i64(i32 %membermask, i64 %value)
  ...
  The i32 return value is a 32-bit mask where bit position in mask corresponds
  to thread’s laneid.

as well as PTX ISA spec [1]:

  9.7.12.8. Parallel Synchronization and Communication Instructions: match.sync
  ...
  Syntax
  match.any.sync.type  d, a, membermask;
  match.all.sync.type  d[|p], a, membermask;
  
  Description
  ...
  Destination d is a 32-bit mask where bit position in mask corresponds
  to thread’s laneid.

So it doesn't make sense to define them with any other return type.

Additionally, ptxas doesn't accept intructions, produced by NVPTX
backend. Here is the ptxas output for `llvm/test/CodeGen/NVPTX/match.ll`:

  ptxas match.ptx, line 44; error   : Arguments mismatch for instruction 'match'
  ptxas match.ptx, line 45; error   : Arguments mismatch for instruction 'match'
  ptxas match.ptx, line 46; error   : Arguments mismatch for instruction 'match'
  ptxas match.ptx, line 47; error   : Arguments mismatch for instruction 'match'
  ptxas match.ptx, line 98; error   : Arguments mismatch for instruction 'match'

After this patch, it compiles with no issues.

[0] https://docs.nvidia.com/cuda/nvvm-ir-spec/index.html#unique_341827171
[1] https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#parallel-synchronization-and-communication-instructions-match-sync


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D120499

Files:
  clang/include/clang/Basic/BuiltinsNVPTX.def
  clang/lib/Headers/__clang_cuda_intrinsics.h
  clang/test/CodeGen/builtins-nvptx-ptx60.cu
  llvm/include/llvm/IR/IntrinsicsNVVM.td
  llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
  llvm/test/CodeGen/NVPTX/match.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D120499.411167.patch
Type: text/x-patch
Size: 13092 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20220224/7380b089/attachment-0001.bin>