Switch lds and global

Author: amd-eochoalo

mlir/include/mlir/Dialect/AMDGPU/IR/AMDGPU.td

@@ -1229,11 +1229,10 @@ def AMDGPU_ScaledMFMAOp :
 
 def AMDGPU_MakeDmaBaseOp :
     AMDGPU_Op<"make_dma_base", [Pure, AttrSizedOperandSegments]>,
-    Arguments<(ins
+    Arguments<(ins Arg<AnyMemRef>:$global,
+                   Variadic<Index>:$global_indices,
                    Arg<AnyMemRef>:$lds,
-                   Variadic<Index>:$lds_indices,
-                   Arg<AnyMemRef>:$global,
-                   Variadic<Index>:$global_indices)>,
+                   Variadic<Index>:$lds_indices)>,
     Results<(outs AMDGPU_TDMBaseType: $base)> {
 
   // TODO:
@@ -1250,7 +1249,7 @@ def AMDGPU_MakeDmaBaseOp :
     For example:
 
     ```mlir
-      %base = amdgpu.make_dma_base %lds[%idx0, %idx1], %global[%idx2, %idx3] : memref<64x64xi32, #gpu.address_space<workgroup>>, memref<64x64xi32> -> !amdgpu.tdm_base<i32>
+      %base = amdgpu.make_dma_base %global[%idx0, %idx1], %lds[%idx2, %idx3] : memref<64x64xi32>, memref<64x64xi32, #gpu.address_space<workgroup>> -> !amdgpu.tdm_base<i32>
       %descriptor = amdgpu.make_dma_descriptor %base globalSize [2, 2] globalStride [2, 1] sharedSize [2, 2] : !amdgpu.tdm_base<i32> -> !amdgpu.tdm_descriptor
       amdgpu.tensor_load_to_lds %descriptor : !amdgpu.tdm_descriptor
     ```
@@ -1279,7 +1278,7 @@ def AMDGPU_MakeDmaBaseOp :
   }];
 
   let assemblyFormat = [{
-    $lds `[` $lds_indices `]` `,` $global `[` $global_indices `]` attr-dict `:` type($lds) `,` type($global) `->` type(results)
+    $global `[` $global_indices `]` `,` $lds `[` $lds_indices `]` attr-dict `:` type($global) `,` type($lds) `->` type(results)
   }];
 
   let hasVerifier = 1;

mlir/test/Conversion/AMDGPUToROCDL/gfx1250.mlir

@@ -206,7 +206,7 @@ func.func @make_dma_base(%idx: index, %mem: memref<8xi32, #gpu_global_addrspace>
   // CHECK: %[[V4I32_0_3:.+]] = llvm.insertelement %[[MEM_INT_LOW]], %[[V4I32_0_2]][%[[C2]] : i32]
   // CHECK: %[[V4I32_0_4:.+]] = llvm.insertelement %[[MEM_INT_HIGH_TYPE]], %[[V4I32_0_3]][%[[C3]] : i32]
 
-  %0 = amdgpu.make_dma_base %smem[%idx], %mem[%idx] : memref<8xi32, #gpu_lds_addrspace>, memref<8xi32, #gpu_global_addrspace> -> !amdgpu.tdm_base<i32>
+  %0 = amdgpu.make_dma_base %mem[%idx], %smem[%idx] : memref<8xi32, #gpu_global_addrspace>, memref<8xi32, #gpu_lds_addrspace> -> !amdgpu.tdm_base<i32>
 
   func.return %0 : !amdgpu.tdm_base<i32>
 }

mlir/test/Dialect/AMDGPU/ops.mlir

@@ -689,8 +689,8 @@ func.func @memory_counter_wait() {
 // CHECK-LABEL: func @make_dma_base
 // CHECK-SAME: (%[[IDX:.+]]: index, %[[MEM:.+]]: memref<8xi32>, %[[SMEM:.+]]: memref<8xi32, #gpu.address_space<workgroup>>)
 func.func @make_dma_base(%idx: index, %mem: memref<8xi32>, %smem: memref<8xi32, #gpu.address_space<workgroup>>) {
-  // CHECK: amdgpu.make_dma_base %[[SMEM]][%[[IDX]]], %[[MEM]][%[[IDX]]] : memref<8xi32, #gpu.address_space<workgroup>>, memref<8xi32> -> !amdgpu.tdm_base<i32>
-  amdgpu.make_dma_base %smem[%idx], %mem[%idx] : memref<8xi32, #gpu.address_space<workgroup>>, memref<8xi32> -> !amdgpu.tdm_base<i32>
+  // CHECK: amdgpu.make_dma_base %[[MEM]][%[[IDX]]], %[[SMEM]][%[[IDX]]] : memref<8xi32>, memref<8xi32, #gpu.address_space<workgroup>> -> !amdgpu.tdm_base<i32>
+  amdgpu.make_dma_base %mem[%idx], %smem[%idx] : memref<8xi32>, memref<8xi32, #gpu.address_space<workgroup>> -> !amdgpu.tdm_base<i32>
   func.return
 }