[Mlir-commits] [mlir] [mlir][x86] Adds extra pattern for AMX lowering (PR #213228)

[Adam Siemieniuk]

================
@@ -483,6 +483,62 @@ func.func @brgemm_bf16_loop(%arg0: memref<16x64x64x2xbf16>, %arg1: memref<16x64x
 // CHECK-NOT: scf.for {{.*}} vector<16x16xf32>, vector<16x16xf32>, vector<16x16xf32>, vector<16x16xf32>
 // CHECK-NOT: vector.contract
 
+module attributes {transform.with_named_sequence} {
+  transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+    %func = transform.structured.match ops{["func.func"]} in %arg1 : (!transform.any_op) -> !transform.any_op
+    transform.apply_patterns to %func {
+      transform.apply_patterns.x86.vector_contract_to_amx_dot_product
+    } : !transform.any_op
+    transform.yield
+  }
+}
+
+// -----
+
+#map = affine_map<(d0, d1, d2, d3) -> (d0, d1, d3)>
+#map1 = affine_map<(d0, d1, d2, d3) -> (d0, d3, d2)>
+#map2 = affine_map<(d0, d1, d2, d3) -> (d1, d2)>
+
+func.func @M_N_K_Can(%arg0: memref<16x32x32xbf16>, %arg1: memref<16x32x32xbf16>, %arg2: memref<32x32xf32>) {
----------------
adam-smnk wrote:

Just one more nit: all lower case for consistency

https://github.com/llvm/llvm-project/pull/213228