[Mlir-commits] [mlir] [MLIR][NVVM] Unify and move to a single tcgen05_mma_kind attr for all tcgen05.mma Ops (PR #184433)

[Rajat Bajpai]

================
@@ -1,39 +1,40 @@
 // RUN: mlir-translate --mlir-to-llvmir %s | FileCheck %s
 
+
 // CHECK-LABEL: @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1
 llvm.func @nvvm_tcgen05_mma_mxf8f6f4_block_scale_cta_1(%d_tmem : !llvm.ptr<6>, %a_desc: i64, %adesc: i64, %b_desc: i64, %idesc: i32, %enable_input_d: i1, %scale_a: !llvm.ptr<6>, %scale_b : !llvm.ptr<6>) {
 
   // CHECK: call void @llvm.nvvm.tcgen05.mma.shared.mxf8f6f4.block_scale(ptr addrspace(6) {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}}, i32 {{%[0-9]+}}, i1 {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, ptr addrspace(6) {{%[0-9]+}}, i32 1, i32 0)
----------------
rajatbajpai wrote:

Kind is modelled as a name not flag value so in this case having no update in llvm call intrinsic makes sense.

https://github.com/llvm/llvm-project/pull/184433