[Mlir-commits] [mlir] [MLIR][Conversion] Add new option "use-opencl" for "convert-gpu-to-spirv" conversion pass (PR #66445)

[Lei Zhang]

================
@@ -0,0 +1,81 @@
+// RUN: mlir-opt -convert-gpu-to-spirv='use-64bit-index=true use-opencl=true' %s -o - | FileCheck %s
+
+module attributes {
+  gpu.container_module,
+  spirv.target_env = #spirv.target_env<
+    #spirv.vce<v1.0, [Kernel, Int64, Addresses], []>, api=OpenCL, #spirv.resource_limits<>>
+} {
+  func.func @load_store(%arg0: memref<12x4xf32, #spirv.storage_class<CrossWorkgroup>>, %arg1: memref<12x4xf32, #spirv.storage_class<CrossWorkgroup>>, %arg2: memref<12x4xf32, #spirv.storage_class<CrossWorkgroup>>) {
----------------
antiagainst wrote:

Shouldn't this use GPU memory space attributes or numeric numbers as input so we can check the conversion is really going down the OpenCL mapping?

https://github.com/llvm/llvm-project/pull/66445