[Mlir-commits] [mlir] [MLIR][NVVM][NVGPU] Combine prefetch and prefetch.tensormap (PR #153134)

[Guray Ozen]

================
@@ -45,3 +45,17 @@ llvm.func @prefetch_L1_uniform(%gen_ptr: !llvm.ptr) {
   nvvm.prefetch level = L1 uniform, %gen_ptr : !llvm.ptr
   llvm.return
 }
+
+llvm.func @prefetch_tensormap(%gen_ptr: !llvm.ptr, %const_ptr: !llvm.ptr<4>) {
+  // CHECK-LABEL: define void @prefetch_tensormap(ptr %0, ptr addrspace(4) %1) {
+  // CHECK-NEXT: call void @llvm.nvvm.prefetch.tensormap.p0(ptr %0)
+  // CHECK-NEXT: call void @llvm.nvvm.prefetch.tensormap.p4(ptr addrspace(4) %1)
+  // CHECK-NEXT: %3 = addrspacecast ptr %0 to ptr addrspace(101)
+  // CHECK-NEXT: call void @llvm.nvvm.prefetch.tensormap.p101(ptr addrspace(101) %3)
+  // CHECK-NEXT: ret void
+  // CHECK-NEXT: }
+  nvvm.prefetch tensormap, %gen_ptr : !llvm.ptr
----------------
grypp wrote:

ah okay you disallow that. I see

https://github.com/llvm/llvm-project/pull/153134