[PATCH] D104847: [Clang][NVPTX] Add NVPTX intrinsics and builtins for CUDA PTX 6.5 and 7.0 WMMA and MMA instructions

[Artem Belevich via Phabricator via llvm-commits]

tra added inline comments.


================
Comment at: clang/include/clang/Basic/BuiltinsNVPTX.def:727
 TARGET_BUILTIN(__bmma_m8n8k128_ld_c, "vi*iC*UiIi", "", AND(SM_75,PTX63))
 TARGET_BUILTIN(__bmma_m8n8k128_mma_xor_popc_b1, "vi*iC*iC*iC*Ii", "", AND(SM_75,PTX63))
 TARGET_BUILTIN(__bmma_m8n8k128_st_c_i32, "vi*iC*UiIi", "", AND(SM_75,PTX63))
----------------
Bummer. mma.h in CUDA-11.3 still does not compile for Ampere.

We appear to be missing the new `__bmma_m8n8k128_mma_and_popc_b1` builtin for the `.and` variant of 1-bit `mma` introduced in PTX 7.1 and not included in this patch.

https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#warp-level-matrix-instructions-wmma-mma

Do you, by any chance, have upcoming patch for PTX7.1, too. :-) 


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104847/new/

https://reviews.llvm.org/D104847