[Mlir-commits] [mlir] [mlir][amdgpu] Lower amdgpu.make_dma_base (PR #169817)

Krzysztof Drewniak llvmlistbot at llvm.org
Tue Dec 2 08:29:06 PST 2025


================
@@ -1230,14 +1230,13 @@ def AMDGPU_ScaledMFMAOp :
 def AMDGPU_MakeDmaBaseOp :
     AMDGPU_Op<"make_dma_base", [Pure, AttrSizedOperandSegments]>,
     Arguments<(ins
-                   Arg<AnyMemRef, "buffer to read from">:$src,
-                   Variadic<Index>:$src_indices,
-                   Arg<AnyMemRef, "buffer to write to">:$dst,
-                   Variadic<Index>:$dst_indices)>,
+                   Arg<AnyMemRef>:$lds,
+                   Variadic<Index>:$lds_indices,
+                   Arg<AnyMemRef>:$global,
----------------
krzysz00 wrote:

What I meant is that we should have `global` first in the argument list, since we're generally going to want to do `MakeDmaBaseOp::create(global, indices, lds, indices, ...)`

https://github.com/llvm/llvm-project/pull/169817


More information about the Mlir-commits mailing list