[mlir][VectorToGPU] Support transposed+broadcasted 2D MMA load

[antiagainst]

Summary

This is loading from 2-D memref, in addition to D139655 where we
load from 1-D memref cases.

mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp

if (useNvGpu &&
     return false;
 
   return true;
 }
 
 // Return true if the given map represents a transposed matrix load,
 // i.e. (d0, d1, ...) -> (dn-1, dn-2).
 static bool isTransposeMatrixLoadMap(OpBuilder &b, AffineMap permutationMap) {
+  MLIRContext *ctx = b.getContext();
   auto nDim = permutationMap.getNumDims();
+  AffineExpr zero = b.getAffineConstantExpr(0);
   if (nDim < 2) {
     // Support transposed+broadcasted cases: affine_map<(d0) -> (d0, 0)>.
     AffineExpr dim0 = b.getAffineDimExpr(0);
-    AffineExpr zero = b.getAffineConstantExpr(0);
-    return permutationMap == AffineMap::get(1, 0, {dim0, zero}, b.getContext());
+    return permutationMap == AffineMap::get(1, 0, {dim0, zero}, ctx);
   }
 
   AffineExpr innerDim = b.getAffineDimExpr(nDim - 1);
   AffineExpr outerDim = b.getAffineDimExpr(nDim - 2);
-  return permutationMap ==
-         AffineMap::get(nDim, 0, {innerDim, outerDim}, b.getContext());
+  // Support both transposed and transposed+broadcasted cases.
+  return permutationMap == AffineMap::get(nDim, 0, {innerDim, outerDim}, ctx) ||
+         permutationMap == AffineMap::get(nDim, 0, {innerDim, zero}, ctx);
 }
 
 // Return the stide for the dimension 0 of |type| if it is a memref and has a
 // constant stride.
 static std::optional<int64_t>
 getMemrefConstantHorizontalStride(ShapedType type) {
   auto memrefType = type.dyn_cast<MemRefType>();
   if (!memrefType)

mlir/test/Conversion/VectorToGPU/vector-to-mma-ops.mlir

func.func @matmul_transposed(%arg0: memref<16x16xf16>, %arg1: memref<16x16xf16>, %arg2: memref<16x16xf16>) {
   %A = vector.transfer_read %arg0[%c0, %c0], %cst {in_bounds = [true, true]} : memref<16x16xf16>, vector<16x16xf16>
   %B = vector.transfer_read %arg1[%c0, %c0], %cst {permutation_map = #map5, in_bounds = [true, true]} : memref<16x16xf16>, vector<16x16xf16>
   %C = vector.transfer_read %arg2[%c0, %c0], %cst {in_bounds = [true, true]} : memref<16x16xf16>, vector<16x16xf16>
   %D = vector.contract {indexing_maps = [#map1, #map2, #map3], iterator_types = ["parallel", "parallel", "reduction"], kind = #vector.kind<add>} %A, %B, %C : vector<16x16xf16>, vector<16x16xf16> into vector<16x16xf16>
   vector.transfer_write %D, %arg2[%c0, %c0] {in_bounds = [true, true]} : vector<16x16xf16>, memref<16x16xf16>
   return
 }
 
-// CHECK-LABEL: func @matmul_transposed_broadcasted
+// CHECK-LABEL: func @matmul_transposed_broadcasted_1d
 //   CHECK-DAG:   %[[A:.+]] = gpu.subgroup_mma_load_matrix %{{.*}}[%{{.*}}] {leadDimension = 0 : index, transpose} : memref<16xf16> -> !gpu.mma_matrix<16x16xf16, "AOp">
 //   CHECK-DAG:   %[[B:.+]] = gpu.subgroup_mma_load_matrix %{{.*}}[%{{.*}}] {leadDimension = 0 : index} : memref<16xf16> -> !gpu.mma_matrix<16x16xf16, "BOp">
 //   CHECK-DAG:   %[[C:.+]] = gpu.subgroup_mma_load_matrix %{{.*}}[%{{.*}}, %{{.*}}] {leadDimension = 16 : index} : memref<16x16xf16> -> !gpu.mma_matrix<16x16xf16, "COp">
 //       CHECK:   %[[D:.+]] = gpu.subgroup_mma_compute %[[A]], %[[B]], %[[C]] : !gpu.mma_matrix<16x16xf16, "AOp">, !gpu.mma_matrix<16x16xf16, "BOp"> -> !gpu.mma_matrix<16x16xf16, "COp">
 //       CHECK:   gpu.subgroup_mma_store_matrix %[[D]], %{{.*}}[%{{.*}}, %{{.*}}] {leadDimension = 16 : index} : !gpu.mma_matrix<16x16xf16, "COp">, memref<16x16xf16>
-func.func @matmul_transposed_broadcasted(%arg0: memref<16xf16>, %arg1: memref<16xf16>, %arg2: memref<16x16xf16>) {
+func.func @matmul_transposed_broadcasted_1d(%arg0: memref<16xf16>, %arg1: memref<16xf16>, %arg2: memref<16x16xf16>) {
   %cst_0 = arith.constant dense<0.000000e+00> : vector<16x16xf16>
   %c0 = arith.constant 0 : index
   %cst = arith.constant 0.000000e+00 : f16
   %A = vector.transfer_read %arg0[%c0], %cst {in_bounds = [true, true], permutation_map = affine_map<(d0) -> (d0, 0)>} : memref<16xf16>, vector<16x16xf16>
   %B = vector.transfer_read %arg1[%c0], %cst {in_bounds = [true, true], permutation_map = affine_map<(d0) -> (d0, 0)>} : memref<16xf16>, vector<16x16xf16>
   %C = vector.transfer_read %arg2[%c0, %c0], %cst {in_bounds = [true, true]} : memref<16x16xf16>, vector<16x16xf16>
   %D = vector.contract {indexing_maps = [#map1, #map2, #map3], iterator_types = ["parallel", "parallel", "reduction"], kind = #vector.kind<add>} %A, %B, %C : vector<16x16xf16>, vector<16x16xf16> into vector<16x16xf16>
   vector.transfer_write %D, %arg2[%c0, %c0] {in_bounds = [true, true]} : vector<16x16xf16>, memref<16x16xf16>
   return
 }
+
+// CHECK-LABEL: func @matmul_transposed_broadcasted_2d
+//   CHECK-DAG:   %[[A:.+]] = gpu.subgroup_mma_load_matrix %{{.*}}[%{{.*}}] {leadDimension = 0 : index, transpose} : memref<32x32xf16> -> !gpu.mma_matrix<16x16xf16, "AOp">
+//   CHECK-DAG:   %[[B:.+]] = gpu.subgroup_mma_load_matrix %{{.*}}[%{{.*}}] {leadDimension = 0 : index} : memref<32x32xf16> -> !gpu.mma_matrix<16x16xf16, "BOp">
+//   CHECK-DAG:   %[[C:.+]] = gpu.subgroup_mma_load_matrix %{{.*}}[%{{.*}}, %{{.*}}] {leadDimension = 16 : index} : memref<16x16xf16> -> !gpu.mma_matrix<16x16xf16, "COp">
+//       CHECK:   %[[D:.+]] = gpu.subgroup_mma_compute %[[A]], %[[B]], %[[C]] : !gpu.mma_matrix<16x16xf16, "AOp">, !gpu.mma_matrix<16x16xf16, "BOp"> -> !gpu.mma_matrix<16x16xf16, "COp">
+//       CHECK:   gpu.subgroup_mma_store_matrix %[[D]], %{{.*}}[%{{.*}}, %{{.*}}] {leadDimension = 16 : index} : !gpu.mma_matrix<16x16xf16, "COp">, memref<16x16xf16>
+func.func @matmul_transposed_broadcasted_2d(%arg0: memref<32x32xf16>, %arg1: memref<32x32xf16>, %arg2: memref<16x16xf16>) {
+  %cst_0 = arith.constant dense<0.000000e+00> : vector<16x16xf16>
+  %c0 = arith.constant 0 : index
+  %cst = arith.constant 0.000000e+00 : f16
+  %A = vector.transfer_read %arg0[%c0, %c0], %cst {in_bounds = [true, true], permutation_map = affine_map<(d0, d1) -> (d1, 0)>} : memref<32x32xf16>, vector<16x16xf16>
+  %B = vector.transfer_read %arg1[%c0, %c0], %cst {in_bounds = [true, true], permutation_map = affine_map<(d0, d1) -> (d1, 0)>} : memref<32x32xf16>, vector<16x16xf16>
+  %C = vector.transfer_read %arg2[%c0, %c0], %cst {in_bounds = [true, true]} : memref<16x16xf16>, vector<16x16xf16>
+  %D = vector.contract {indexing_maps = [#map1, #map2, #map3], iterator_types = ["parallel", "parallel", "reduction"], kind = #vector.kind<add>} %A, %B, %C : vector<16x16xf16>, vector<16x16xf16> into vector<16x16xf16>
+  vector.transfer_write %D, %arg2[%c0, %c0] {in_bounds = [true, true]} : vector<16x16xf16>, memref<16x16xf16>
+  return
+}