[Mlir-commits] [mlir] [mlir][spirv] Add gpu printf op lowering to spirv.CL.printf op (PR #78510)

[Jakub Kuderski]

================
@@ -875,7 +875,7 @@ def SPIRV_CLPrintfOp : SPIRV_CLOp<"printf", 184, []> {
     #### Example:
 
     ```mlir
-    %0 = spirv.CL.printf %0 %1 %2 : (!spirv.ptr<i8, UniformConstant>, (i32, i32)) -> i32
+    %0 = spirv.CL.printf %0 : !spirv.ptr<i8, UniformConstant>(%1, %2 : i32, i32) -> i32
----------------
kuhar wrote:

This syntax seems unusual to me in the context of the SPIR-V dialect. Why would we prefer it like this?

If we were to update it, I think something like this would be the most canonical form:
```mlir
%x = spirv.CL.printf %0, %1, %2 : !spirv.ptr<i8, UniformConstant>, i32, i32 -> i32
```
BTW, `%0` is redefined in this example 

https://github.com/llvm/llvm-project/pull/78510