[Mlir-commits] [mlir] [mlir][linalg] Consolidate tests for scalable vectorization (PR #141469)
Han-Chung Wang
llvmlistbot at llvm.org
Tue May 27 01:22:21 PDT 2025
================
@@ -509,6 +670,183 @@ module attributes {transform.with_named_sequence} {
// -----
+func.func @vectorize_static_shape_with_mask_scalable(%arg0: tensor<8x30xf32>,
+ %arg1: tensor<8x30xf32>,
+ %arg2: tensor<8x30xf32>) -> tensor<8x30xf32> {
+ %0 = linalg.generic { indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,
+ affine_map<(d0, d1) -> (d0, d1)>,
+ affine_map<(d0, d1) -> (d0, d1)>],
+ iterator_types = ["parallel", "parallel"] }
+ ins(%arg0, %arg1 : tensor<8x30xf32>, tensor<8x30xf32>)
+ outs(%arg2 : tensor<8x30xf32>) {
+ ^bb(%in0: f32, %in1: f32, %out: f32) :
+ %0 = arith.addf %in0, %in1 : f32
+ linalg.yield %0 : f32
+ } -> tensor<8x30xf32>
+ return %0 : tensor<8x30xf32>
+}
+
+// CHECK-LABEL: func.func @vectorize_static_shape_with_mask_scalable(
+// CHECK-SAME: %[[VAL_0:.*]]: tensor<8x30xf32>, %[[VAL_1:.*]]: tensor<8x30xf32>, %[[VAL_2:.*]]: tensor<8x30xf32>) -> tensor<8x30xf32> {
+// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 0 : index
+// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index
+// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 30 : index
+// CHECK: %[[VAL_7:.*]] = vector.create_mask %[[VAL_5]], %[[VAL_6]] : vector<8x[32]xi1>
+// CHECK: %[[VAL_8:.*]] = vector.mask %[[VAL_7]] { vector.transfer_read %[[VAL_0]][%[[VAL_3]], %[[VAL_3]]], %[[VAL_4]] {in_bounds = [true, true]} : tensor<8x30xf32>, vector<8x[32]xf32> } : vector<8x[32]xi1> -> vector<8x[32]xf32>
+// CHECK: %[[VAL_9:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[VAL_10:.*]] = vector.mask %[[VAL_7]] { vector.transfer_read %[[VAL_1]][%[[VAL_3]], %[[VAL_3]]], %[[VAL_9]] {in_bounds = [true, true]} : tensor<8x30xf32>, vector<8x[32]xf32> } : vector<8x[32]xi1> -> vector<8x[32]xf32>
+// CHECK: %[[VAL_11:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[VAL_12:.*]] = vector.mask %[[VAL_7]] { vector.transfer_read %[[VAL_2]][%[[VAL_3]], %[[VAL_3]]], %[[VAL_11]] {in_bounds = [true, true]} : tensor<8x30xf32>, vector<8x[32]xf32> } : vector<8x[32]xi1> -> vector<8x[32]xf32>
+// CHECK: %[[VAL_13:.*]] = arith.addf %[[VAL_8]], %[[VAL_10]] : vector<8x[32]xf32>
+// CHECK: %[[VAL_14:.*]] = arith.constant 0 : index
+// CHECK: %[[VAL_15:.*]] = vector.mask %[[VAL_7]] { vector.transfer_write %[[VAL_13]], %[[VAL_2]][%[[VAL_14]], %[[VAL_14]]] {in_bounds = [true, true]} : vector<8x[32]xf32>, tensor<8x30xf32> } : vector<8x[32]xi1> -> tensor<8x30xf32>
+
+module attributes {transform.with_named_sequence} {
+ transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+ %0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!transform.any_op) -> !transform.any_op
+ transform.structured.vectorize %0 vector_sizes [8, [32]] : !transform.any_op
+ transform.yield
+ }
+}
+
+// -----
+
+///----------------------------------------------------------------------------------------
+/// Tests for linalg.matvec
+///----------------------------------------------------------------------------------------
+
+// Scalable _reduction_ dimension.
+
+func.func @vectorize_dynamic_matvec_trailing_reduction_dim(%arg0: tensor<?x?xf32>,
+ %arg1: tensor<?xf32>,
+ %arg2: tensor<?xf32>) {
+ linalg.matvec ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?xf32>)
+ outs(%arg2 : tensor<?xf32>) -> tensor<?xf32>
+ return
+}
+
+// CHECK-LABEL: func.func @vectorize_dynamic_matvec_trailing_reduction_dim(
+// CHECK-SAME: %[[ARG_0:.*]]: tensor<?x?xf32>, %[[ARG_1:.*]]: tensor<?xf32>, %[[ARG_2:.*]]: tensor<?xf32>) {
+// CHECK: %[[C0_idx:.*]] = arith.constant 0 : index
+// CHECK: %[[DIM_A0_0:.*]] = tensor.dim %[[ARG_0]], %[[C0_idx]] : tensor<?x?xf32>
+// CHECK: %[[C1_idx:.*]] = arith.constant 1 : index
+// CHECK: %[[DIM_A0_1:.*]] = tensor.dim %[[ARG_0]], %[[C1_idx]] : tensor<?x?xf32>
+// CHECK: %[[C0_idx:.*]] = arith.constant 0 : index
+// CHECK: %[[C0_f32:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[MASK_2d:.*]] = vector.create_mask %[[DIM_A0_0]], %[[DIM_A0_1]] : vector<4x[4]xi1>
+// CHECK: %[[VEC_RD_0:.*]] = vector.mask %[[MASK_2d]] { vector.transfer_read %[[ARG_0]][%[[C0_idx]], %[[C0_idx]]], %[[C0_f32]] {in_bounds = [true, true]} : tensor<?x?xf32>, vector<4x[4]xf32> } : vector<4x[4]xi1> -> vector<4x[4]xf32>
+// CHECK: %[[C0_f32:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[MASK_d1:.*]] = vector.create_mask %[[DIM_A0_1]] : vector<[4]xi1>
+// CHECK: %[[VEC_RD_1:.*]] = vector.mask %[[MASK_d1]] { vector.transfer_read %[[ARG_1]][%[[C0_idx]]], %[[C0_f32]] {in_bounds = [true, true], permutation_map = #map} : tensor<?xf32>, vector<4x[4]xf32> } : vector<[4]xi1> -> vector<4x[4]xf32>
+// CHECK: %[[C0_f32:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[MASK_d2:.*]] = vector.create_mask %[[DIM_A0_0]] : vector<4xi1>
+// CHECK: %[[VEC_RD_2:.*]] = vector.mask %[[MASK_d2]] { vector.transfer_read %[[ARG_2]][%[[C0_idx]]], %[[C0_f32]] {in_bounds = [true]} : tensor<?xf32>, vector<4xf32> } : vector<4xi1> -> vector<4xf32>
+// CHECK: %[[MUL:.*]] = arith.mulf %[[VEC_RD_0:.*]], %[[VEC_RD_1:.*]] : vector<4x[4]xf32>
+// CHECK: %[[REDUCE:.*]] = vector.mask %[[MASK_2d]] { vector.multi_reduction <add>, %[[MUL]], %[[VEC_RD_2]] [1] : vector<4x[4]xf32> to vector<4xf32> } : vector<4x[4]xi1> -> vector<4xf32>
+// CHECK: %[[C0_idx:.*]] = arith.constant 0 : index
+// CHECK: %{{.*}} = vector.mask %[[MASK_d2]] { vector.transfer_write %[[REDUCE]], %[[ARG_2]][%[[C0_idx]]] {in_bounds = [true]} : vector<4xf32>, tensor<?xf32> } : vector<4xi1> -> tensor<?xf32>
+
+module attributes {transform.with_named_sequence} {
+ transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+ %0 = transform.structured.match ops{["linalg.matvec"]} in %arg1 : (!transform.any_op) -> !transform.any_op
+ transform.structured.vectorize %0 vector_sizes [4, [4]] : !transform.any_op
+ transform.yield
+ }
+}
+
+// -----
+
+// Scalable _parallel_ dimension.
+
+func.func @vectorize_dynamic_matvec_trailing_reduction_dim(%arg0: tensor<?x?xf32>,
+ %arg1: tensor<?xf32>,
+ %arg2:
+ tensor<?xf32>) ->
+ tensor<?xf32>{
+ %0 = linalg.matvec ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?xf32>)
+ outs(%arg2 : tensor<?xf32>) -> tensor<?xf32>
+ return %0 : tensor<?xf32>
+}
+
+// CHECK-LABEL: func.func @vectorize_dynamic_matvec_trailing_reduction_dim(
+// CHECK-SAME: %[[ARG_0:.*]]: tensor<?x?xf32>, %[[ARG_1:.*]]: tensor<?xf32>, %[[ARG_2:.*]]: tensor<?xf32>) -> tensor<?xf32> {
+// CHECK: %[[C0_idx:.*]] = arith.constant 0 : index
+// CHECK: %[[DIM_A0_0:.*]] = tensor.dim %[[ARG_0]], %[[C0_idx]] : tensor<?x?xf32>
+// CHECK: %[[C1_idx:.*]] = arith.constant 1 : index
+// CHECK: %[[DIM_A0_1:.*]] = tensor.dim %[[ARG_0]], %[[C1_idx]] : tensor<?x?xf32>
+// CHECK: %[[C0_idx:.*]] = arith.constant 0 : index
+// CHECK: %[[C0_f32:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[MASK_2d:.*]] = vector.create_mask %[[DIM_A0_0]], %[[DIM_A0_1]] : vector<[4]x4xi1>
+// CHECK: %[[VEC_RD_0:.*]] = vector.mask %[[MASK_2d]] { vector.transfer_read %[[ARG_0]][%[[C0_idx]], %[[C0_idx]]], %[[C0_f32]] {in_bounds = [true, true]} : tensor<?x?xf32>, vector<[4]x4xf32> } : vector<[4]x4xi1> -> vector<[4]x4xf32>
+// CHECK: %[[C0_f32:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[MASK_d1:.*]] = vector.create_mask %[[DIM_A0_1]] : vector<4xi1>
+// CHECK: %[[VEC_RD_1:.*]] = vector.mask %[[MASK_d1]] { vector.transfer_read %[[ARG_1]][%[[C0_idx]]], %[[C0_f32]] {in_bounds = [true, true], permutation_map = #map} : tensor<?xf32>, vector<[4]x4xf32> } : vector<4xi1> -> vector<[4]x4xf32>
+// CHECK: %[[C0_f32:.*]] = arith.constant 0.000000e+00 : f32
+// CHECK: %[[MASK_d2:.*]] = vector.create_mask %[[DIM_A0_0]] : vector<[4]xi1>
+// CHECK: %[[VEC_RD_2:.*]] = vector.mask %[[MASK_d2]] { vector.transfer_read %[[ARG_2]][%[[C0_idx]]], %[[C0_f32]] {in_bounds = [true]} : tensor<?xf32>, vector<[4]xf32> } : vector<[4]xi1> -> vector<[4]xf32>
+// CHECK: %[[MUL:.*]] = arith.mulf %[[VEC_RD_0:.*]], %[[VEC_RD_1:.*]] : vector<[4]x4xf32>
+// CHECK: %[[REDUCE:.*]] = vector.mask %[[MASK_2d]] { vector.multi_reduction <add>, %[[MUL]], %[[VEC_RD_2]] [1] : vector<[4]x4xf32> to vector<[4]xf32> } : vector<[4]x4xi1> -> vector<[4]xf32>
+// CHECK: %[[C0_idx:.*]] = arith.constant 0 : index
+// CHECK: %{{.*}} = vector.mask %[[MASK_d2]] { vector.transfer_write %[[REDUCE]], %[[ARG_2]][%[[C0_idx]]] {in_bounds = [true]} : vector<[4]xf32>, tensor<?xf32> } : vector<[4]xi1> -> tensor<?xf32>
+
+
----------------
hanhanW wrote:
nit: remove one blank line
https://github.com/llvm/llvm-project/pull/141469
More information about the Mlir-commits
mailing list