[mlir] [MLIR][Linalg] Introduce transpose/broadcast semantic to linalg.batch… (PR #130944)

[Andrzej Warzyński]

================
@@ -1024,6 +1024,36 @@ func.func @batch_matmul(%arg0: tensor<2x3x5xf32>, %arg1: tensor<2x5x7xf32>, %arg
 
 // -----
 
+// CHECK: #[[$ATTR_0:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d1, d3)>
+// CHECK: #[[$ATTR_1:.+]] = affine_map<(d0, d1, d2, d3) -> (d0, d3, d2)>
+// CHECK: #[[$ATTR_2:.+]] = affine_map<(d0, d1, d2, d3) -> (d1, d2)>
+
+// CHECK-LABEL:   func.func @batch_reduce_matmul(
+// CHECK-SAME:                                   %[[VAL_0:[0-9]+|[a-zA-Z$._-][a-zA-Z0-9$._-]*]]: tensor<2x3x5xf32>,
+// CHECK-SAME:                                   %[[VAL_1:[0-9]+|[a-zA-Z$._-][a-zA-Z0-9$._-]*]]: tensor<2x5x7xf32>,
+// CHECK-SAME:                                   %[[VAL_2:[0-9]+|[a-zA-Z$._-][a-zA-Z0-9$._-]*]]: tensor<3x7xf32>) -> tensor<3x7xf32> {
+// CHECK:           %[[VAL_3:.*]] = linalg.generic {indexing_maps = [#[[$ATTR_0]], #[[$ATTR_1]], #[[$ATTR_2]]], iterator_types = ["reduction", "parallel", "parallel", "reduction"]} ins(%[[VAL_0]], %[[VAL_1]] : tensor<2x3x5xf32>, tensor<2x5x7xf32>) outs(%[[VAL_2]] : tensor<3x7xf32>) {
+// CHECK:           ^bb0(%[[VAL_4:.*]]: f32, %[[VAL_5:.*]]: f32, %[[VAL_6:.*]]: f32):
+// CHECK:             %[[VAL_7:.*]] = arith.mulf %[[VAL_4]], %[[VAL_5]] : f32
+// CHECK:             %[[VAL_8:.*]] = arith.addf %[[VAL_6]], %[[VAL_7]] : f32
+// CHECK:             linalg.yield %[[VAL_8]] : f32
+// CHECK:           } -> tensor<3x7xf32>
+// CHECK:           return %[[VAL_3]] : tensor<3x7xf32>
+// CHECK:         }
----------------
banach-space wrote:

Please only check what's necessary: https://mlir.llvm.org/getting_started/TestingGuide/#filecheck-best-practices

https://github.com/llvm/llvm-project/pull/130944