Skip to content

[Mlir] Skip default schedule for linalg.fill #63

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
ElectrikSpace wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[Mlir] Skip default schedule for linalg.fill #63

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions src/xtc/backends/mlir/MlirCompilerPasses.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -219,6 +219,9 @@ def _generate_scheduling(self) -> OpResult:
for schedule in self._nodes_schedules:
if schedule.node_ident in unscheduled_handles:
continue
# Skip linalg.fill
if schedule.node_name[-1] == "0": # identify with naming convention
continue
self._create_sdist_meshes(schedule)
handle = structured_match(
results_=transform.AnyOpType.get(),
Expand Down
238 changes: 86 additions & 152 deletions tests/filecheck/backends/padding/test_gen_pad_dict_conv2d_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

211 changes: 80 additions & 131 deletions tests/filecheck/backends/padding/test_gen_pad_int_matmul_unpad_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

238 changes: 86 additions & 152 deletions tests/filecheck/backends/padding/test_gen_pad_tuple_conv2d_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

211 changes: 80 additions & 131 deletions tests/filecheck/backends/padding/test_gen_pad_tuple_matmul_unpad_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

215 changes: 82 additions & 133 deletions tests/filecheck/backends/padding/test_pad2d_dict_matmul_unpad_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

232 changes: 83 additions & 149 deletions tests/filecheck/backends/padding/test_pad_constant_conv2d_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

232 changes: 83 additions & 149 deletions tests/filecheck/backends/padding/test_pad_conv2d_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

215 changes: 82 additions & 133 deletions tests/filecheck/backends/padding/test_pad_matmul_unpad_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

215 changes: 82 additions & 133 deletions tests/filecheck/backends/padding/test_pad_tuple_matmul_unpad_mlir.py
View file
Open in desktop

Large diffs are not rendered by default.

227 changes: 84 additions & 143 deletions tests/filecheck/backends/tensor_dialect/test_conv2d_mini_mlir_tensor.py
View file
Open in desktop

Large diffs are not rendered by default.

275 changes: 114 additions & 161 deletions tests/filecheck/backends/tensor_dialect/test_conv2d_r181_mlir_tensor.py
View file
Open in desktop

Large diffs are not rendered by default.

123 changes: 38 additions & 85 deletions tests/filecheck/backends/tensor_dialect/test_conv2d_relu_tensor.py
View file
Open in desktop

Large diffs are not rendered by default.

401 changes: 187 additions & 214 deletions tests/filecheck/backends/tensor_dialect/test_matmul_mlir_tensor.py
View file
Open in desktop

Large diffs are not rendered by default.

97 changes: 46 additions & 51 deletions tests/filecheck/backends/tensor_dialect/test_matmul_mlir_tensor_layout.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,56 +28,51 @@
executor = module.get_executor(validate=True)
res = executor.execute()
print(f"CODE: {res}")
# CHECK: // -----// IR Dump Before transform //----- //
# CHECK-NEXT: #map = affine_map<(d0, d1, d2) -> (d2, d0)>
# CHECK-NEXT: #map1 = affine_map<(d0, d1, d2) -> (d2, d1)>
# CHECK-NEXT: #map2 = affine_map<(d0, d1, d2) -> (d0, d1)>
# CHECK-NEXT: module attributes {transform.with_named_sequence} {
# CHECK-NEXT: func.func @matmul_layout(%arg0: tensor<512x4xf32> {llvm.noalias}, %arg1: tensor<512x32xf32> {llvm.noalias}, %arg2: memref<4x32xf32> {llvm.noalias}) {
# CHECK-NEXT: %0 = tensor.empty() : tensor<4x32xf32>
# CHECK-NEXT: %cst = arith.constant 0.000000e+00 : f32
# CHECK-NEXT: %1 = linalg.fill {__xtc_id_C_0_} ins(%cst : f32) outs(%0 : tensor<4x32xf32>) -> tensor<4x32xf32>
# CHECK-NEXT: %2 = linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = ["parallel", "parallel", "reduction"]} ins(%arg0, %arg1 : tensor<512x4xf32>, tensor<512x32xf32>) outs(%1 : tensor<4x32xf32>) attrs = {__xtc_id_C_} {
# CHECK-NEXT: ^bb0(%in: f32, %in_0: f32, %out: f32):
# CHECK-NEXT: %3 = arith.mulf %in, %in_0 : f32
# CHECK-NEXT: %4 = arith.addf %out, %3 : f32
# CHECK-NEXT: linalg.yield %4 : f32
# CHECK-NEXT: } -> tensor<4x32xf32>
# CHECK-NEXT: bufferization.materialize_in_destination %2 in restrict writable %arg2 : (tensor<4x32xf32>, memref<4x32xf32>) -> ()
# CHECK-NEXT: return
# CHECK-NEXT: }
# CHECK-NEXT: transform.named_sequence @_vecto(%arg0: !transform.any_op {transform.consumed}) {
# CHECK-NEXT: transform.structured.vectorize %arg0 : !transform.any_op
# CHECK-NEXT: transform.yield
# CHECK-NEXT: }
# CHECK-NEXT: transform.named_sequence @_post_bufferize(%arg0: !transform.any_op {transform.readonly}) {
# CHECK-NEXT: transform.yield
# CHECK-NEXT: }
# CHECK-NEXT: transform.named_sequence @__transform_main(%arg0: !transform.any_op {transform.readonly}) {
# CHECK-NEXT: %0 = transform.structured.match attributes {__xtc_id_C_0_} in %arg0 : (!transform.any_op) -> !transform.any_op
# CHECK-NEXT: %tiled_linalg_op, %loops = transform.structured.tile_using_for %0 tile_sizes [1, 0] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops "./i" : !transform.any_op
# CHECK-NEXT: %tiled_linalg_op_0, %loops_1 = transform.structured.tile_using_for %tiled_linalg_op tile_sizes [0, 1] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops_1 "./j" : !transform.any_op
# CHECK-NEXT: %1 = transform.structured.match attributes {__xtc_id_C_} in %arg0 : (!transform.any_op) -> !transform.any_op
# CHECK-NEXT: %tiled_linalg_op_2, %loops_3 = transform.structured.tile_using_for %1 tile_sizes [1, 0, 0] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops_3 "./i" : !transform.any_op
# CHECK-NEXT: %tiled_linalg_op_4, %loops_5 = transform.structured.tile_using_for %tiled_linalg_op_2 tile_sizes [0, 1, 0] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops_5 "./j" : !transform.any_op
# CHECK-NEXT: %tiled_linalg_op_6, %loops_7 = transform.structured.tile_using_for %tiled_linalg_op_4 tile_sizes [0, 0, 1] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops_7 "./k" : !transform.any_op
# CHECK-NEXT: transform.yield
# CHECK-NEXT: }
# CHECK-NEXT: }
# CHECK: // -----// IR Dump Before transform //----- //
# CHECK-NEXT: #map = affine_map<(d0, d1, d2) -> (d2, d0)>
# CHECK-NEXT: #map1 = affine_map<(d0, d1, d2) -> (d2, d1)>
# CHECK-NEXT: #map2 = affine_map<(d0, d1, d2) -> (d0, d1)>
# CHECK-NEXT: module attributes {transform.with_named_sequence} {
# CHECK-NEXT: func.func @matmul_layout(%arg0: tensor<512x4xf32> {llvm.noalias}, %arg1: tensor<512x32xf32> {llvm.noalias}, %arg2: memref<4x32xf32> {llvm.noalias}) {
# CHECK-NEXT: %0 = tensor.empty() : tensor<4x32xf32>
# CHECK-NEXT: %cst = arith.constant 0.000000e+00 : f32
# CHECK-NEXT: %1 = linalg.fill {__xtc_id_C_0_} ins(%cst : f32) outs(%0 : tensor<4x32xf32>) -> tensor<4x32xf32>
# CHECK-NEXT: %2 = linalg.generic {indexing_maps = [#map, #map1, #map2], iterator_types = ["parallel", "parallel", "reduction"]} ins(%arg0, %arg1 : tensor<512x4xf32>, tensor<512x32xf32>) outs(%1 : tensor<4x32xf32>) attrs = {__xtc_id_C_} {
# CHECK-NEXT: ^bb0(%in: f32, %in_0: f32, %out: f32):
# CHECK-NEXT: %3 = arith.mulf %in, %in_0 : f32
# CHECK-NEXT: %4 = arith.addf %out, %3 : f32
# CHECK-NEXT: linalg.yield %4 : f32
# CHECK-NEXT: } -> tensor<4x32xf32>
# CHECK-NEXT: bufferization.materialize_in_destination %2 in restrict writable %arg2 : (tensor<4x32xf32>, memref<4x32xf32>) -> ()
# CHECK-NEXT: return
# CHECK-NEXT: }
# CHECK-NEXT: transform.named_sequence @_vecto(%arg0: !transform.any_op {transform.consumed}) {
# CHECK-NEXT: transform.structured.vectorize %arg0 : !transform.any_op
# CHECK-NEXT: transform.yield
# CHECK-NEXT: }
# CHECK-NEXT: transform.named_sequence @_post_bufferize(%arg0: !transform.any_op {transform.readonly}) {
# CHECK-NEXT: transform.yield
# CHECK-NEXT: }
# CHECK-NEXT: transform.named_sequence @__transform_main(%arg0: !transform.any_op {transform.readonly}) {
# CHECK-NEXT: %0 = transform.structured.match attributes {__xtc_id_C_} in %arg0 : (!transform.any_op) -> !transform.any_op
# CHECK-NEXT: %tiled_linalg_op, %loops = transform.structured.tile_using_for %0 tile_sizes [1, 0, 0] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops "./i" : !transform.any_op
# CHECK-NEXT: %tiled_linalg_op_0, %loops_1 = transform.structured.tile_using_for %tiled_linalg_op tile_sizes [0, 1, 0] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops_1 "./j" : !transform.any_op
# CHECK-NEXT: %tiled_linalg_op_2, %loops_3 = transform.structured.tile_using_for %tiled_linalg_op_0 tile_sizes [0, 0, 1] : (!transform.any_op) -> (!transform.any_op, !transform.any_op)
# CHECK-NEXT: transform.annotate %loops_3 "./k" : !transform.any_op
# CHECK-NEXT: transform.yield
# CHECK-NEXT: }
# CHECK-NEXT: }
# CHECK-NEXT:
# CHECK-NEXT: graph:
# CHECK-NEXT: name: matmul_layout
# CHECK-NEXT: inputs:
# CHECK-NEXT: - %0 : 4x512xfloat32, <(0,1)->(1,0)>
# CHECK-NEXT: - %1 : 512x32xfloat32
# CHECK-NEXT: outputs:
# CHECK-NEXT: - %2 : 4x32xfloat32
# CHECK-NEXT: nodes:
# CHECK-NEXT: - %2: matmul(%0, %1) {name = 'C'} : [4x512xfloat32, <(0,1)->(1,0)>, 512x32xfloat32] -> [4x32xfloat32]
# CHECK-NEXT: graph:
# CHECK-NEXT: name: matmul_layout
# CHECK-NEXT: inputs:
# CHECK-NEXT: - %0 : 4x512xfloat32, <(0,1)->(1,0)>
# CHECK-NEXT: - %1 : 512x32xfloat32
# CHECK-NEXT: outputs:
# CHECK-NEXT: - %2 : 4x32xfloat32
# CHECK-NEXT: nodes:
# CHECK-NEXT: - %2: matmul(%0, %1) {name = 'C'} : [4x512xfloat32, <(0,1)->(1,0)>, 512x32xfloat32] -> [4x32xfloat32]
# CHECK-NEXT:
# CHECK-NEXT: CODE: 0
# CHECK-NEXT: CODE: 0
Loading
Loading