[Mlir-commits] [mlir] [mlir][llvm][x86vector] One-to-one intrinsic op interface (PR #140055)

[Adam Siemieniuk]

================
@@ -14,57 +14,18 @@
 #define X86VECTOR_INTERFACES
 
 include "mlir/IR/Interfaces.td"
+include "mlir/Dialect/LLVMIR/LLVMInterfaces.td"
 
 //===----------------------------------------------------------------------===//
-// One-to-One Intrinsic Interface
+// Intrinsic Interface
 //===----------------------------------------------------------------------===//
 
-def OneToOneIntrinsicOpInterface : OpInterface<"OneToOneIntrinsicOp"> {
+def IntrinsicOpInterface
+    : OpInterface<"IntrinsicOp", [OneToOneIntrinsicOpInterface]> {
----------------
adam-smnk wrote:

For now yes, I intend to specifically target only x86 and later only amx ops in their respective legalization passes.
To be fair, so far the two conversions (after initial testing) are identical and we could probably get away with a single generic converter. But I want to keep these separate for now, see how that holds up, and maybe refactor later.

https://github.com/llvm/llvm-project/pull/140055