[Mlir-commits] [llvm] [mlir] [NVPTX] Added more MMA intrinsics for F8F6F4 and FP64 types. (PR #156040)

[Kirill Vedernikov]

================
@@ -90,6 +90,21 @@ def __init__(self, geom, frag, ptx_elt_type, is_mma_sparse=False):
             "m16n8k32:b:s8": 2,
----------------
kvederni wrote:

Of course, I haven't tested the new tests only, I run all the tests using `cmake --build $REPO/build --target check-llvm -- -j 64` with ptxas from cuda-13.0 and cuda-12.8. The tests passed. WMMA tests are generated from `llvm-project/llvm/test/CodeGen/NVPTX/wmma-ptxX-smY.py`. Should I include all generated `build/test/CodeGen/NVPTX/Output/wmma-ptxX-smY.py.tmp-ptxX-sm_Y.ll` files up to 30K lines each to this review?

https://github.com/llvm/llvm-project/pull/156040