[Mlir-commits] [mlir] [mlir][linalg] Add mixed precision folding pattern in vectorize_children_and_apply_patterns TD Op (PR #148684)

[Adam Siemieniuk]

================
@@ -1715,65 +1828,79 @@ module attributes {transform.with_named_sequence} {
 
 // -----
 
-// Input identical as the test in vectorization.mlir. Output is different -
-// vector sizes are inferred (rather than user-specified) and hence _no_
-// masking was used.
+// Mixed precision vectorization tests.
 
-func.func @test_vectorize_pack(%arg0: tensor<32x8x16xf32>, %arg1: tensor<4x1x32x16x2xf32>) -> tensor<4x1x32x16x2xf32> {
-  %pack = linalg.pack %arg0 outer_dims_perm = [1, 2, 0] inner_dims_pos = [2, 1] inner_tiles = [16, 2] into %arg1 : tensor<32x8x16xf32> -> tensor<4x1x32x16x2xf32>
-  return %pack : tensor<4x1x32x16x2xf32>
+// CHECK-LABEL: func @float_mixed_precision_generic_as_contract
+// CHECK-COUNT-3: vector.transfer_read
+// CHECK-NOT:     arith.extf
+// CHECK:         vector.contract {{.*}} : vector<8x32x16xbf16>, vector<8x32x16xbf16> into vector<8x32xf32>
+// CHECK:         vector.transfer_write
+func.func @float_mixed_precision_generic_as_contract(%A: memref<8x16xbf16>, %B: memref<16x32xbf16>,
----------------
adam-smnk wrote:

nit: wouldn't it be `...mixed_precision_matmul_as_generic` following above convention?

https://github.com/llvm/llvm-project/pull/148684