[Mlir-commits] [mlir] [MLIR][Linalg] Introduce linalg.contract (PR #123618)

[Andrzej Warzyński]

================
@@ -996,9 +994,198 @@ func.func @matmul_transpose_a_b_explicit(%arg0: memref<5x3xf32>, %arg1: memref<7
                       ]
                       ins(%arg0, %arg1 : memref<5x3xf32>, memref<7x5xf32>)
                       outs(%arg2: memref<3x7xf32>)
-                      
   return
 }
 
 // -----
 
+// CHECK: #[[$ATTR_0:.+]] = affine_map<(d0, d1, d2) -> (d0, d2)>
+// CHECK: #[[$ATTR_1:.+]] = affine_map<(d0, d1, d2) -> (d2, d1)>
+// CHECK: #[[$ATTR_2:.+]] = affine_map<(d0, d1, d2) -> (d0, d1)>
+
+// CHECK-LABEL:   func.func @contract_matmul(
+// CHECK-SAME:                                             %[[VAL_0:.*]]: memref<3x5xf32>,
+// CHECK-SAME:                                             %[[VAL_1:.*]]: memref<5x7xf32>,
+// CHECK-SAME:                                             %[[VAL_2:.*]]: memref<3x7xf32>) {
+
+// CHECK:           linalg.generic {indexing_maps = [#[[$ATTR_0]], #[[$ATTR_1]], #[[$ATTR_2]]], iterator_types = ["parallel", "parallel", "reduction"]}
+// CHECK-NEXT:     ^{{.+}}(
+// CHECK-NEXT:      arith.mulf
+// CHECK-NEXT:      arith.addf
+// CHECK-NEXT:      linalg.yield
+
+func.func @contract_matmul(%arg0: memref<3x5xf32>, %arg1: memref<5x7xf32>, %arg2: memref<3x7xf32>) {
----------------
banach-space wrote:

Would you object to using more descriptive variable names (MLIR + LIT). For example, `%A` instead of `%arg0`. It's a small change, but does reduce the cognitive load quite substantially.

https://github.com/llvm/llvm-project/pull/123618