Skip to content

Commit 84170e4

Browse files
swolchokswolchok
committed
rebase, fix lint in #9432
[ghstack-poisoned]
2 parents c0ad8ac + a4df2da commit 84170e4

File tree

33 files changed

+749
-241
lines changed

33 files changed

+749
-241
lines changed

‎.lintrunner.toml‎

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -392,15 +392,12 @@ exclude_patterns = [
392392
# backends
393393
"backends/vulkan/quantizer/**",
394394
"backends/vulkan/test/**",
395-
"backends/qualcomm/quantizer/**",
396-
"examples/qualcomm/**",
397395
"backends/xnnpack/quantizer/**",
398396
"backends/xnnpack/test/**",
399397
"exir/tests/test_passes.py",
400398
"extension/llm/export/builder.py",
401399
"extension/llm/export/quantizer_lib.py",
402400
"exir/tests/test_memory_planning.py",
403-
"backends/transforms/duplicate_dynamic_quant_chain.py",
404401
"exir/backend/test/demos/test_xnnpack_qnnpack.py",
405402
]
406403

‎CMakeLists.txt‎

Lines changed: 5 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -143,6 +143,11 @@ else()
143143
set(CMAKE_CXX_FLAGS_RELEASE "-O2 ${CMAKE_CXX_FLAGS_RELEASE}")
144144
endif()
145145

146+
if(EXECUTORCH_BUILD_TESTS)
147+
set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
148+
include(CTest)
149+
endif()
150+
146151
add_subdirectory(third-party)
147152

148153
if(EXECUTORCH_BUILD_EXTENSION_TRAINING)
@@ -254,11 +259,6 @@ if(EXECUTORCH_BUILD_PTHREADPOOL)
254259
)
255260
endif()
256261

257-
if(EXECUTORCH_BUILD_TESTS)
258-
set(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
259-
include(CTest)
260-
endif()
261-
262262
# TODO(dbort): Fix these warnings and remove this flag.
263263
set(_common_compile_options -Wno-deprecated-declarations -fPIC)
264264

@@ -472,11 +472,6 @@ install(
472472
)
473473
install(FILES tools/cmake/executorch-config.cmake DESTINATION lib/cmake/ExecuTorch)
474474

475-
# Add googletest if any test targets should be built
476-
if(BUILD_TESTING)
477-
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/third-party/googletest)
478-
endif()
479-
480475
if(EXECUTORCH_BUILD_ARM_BAREMETAL)
481476
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/backends/arm)
482477
endif()

‎backends/cadence/aot/tests/test_simplify_ops_passes.py‎

Lines changed: 30 additions & 80 deletions
Original file line numberDiff line numberDiff line change
@@ -12,7 +12,6 @@
1212

1313
import executorch.backends.cadence.aot.ops_registrations # noqa
1414
import torch
15-
from executorch.backends.cadence.aot.compiler import export_to_edge
1615
from executorch.backends.cadence.aot.graph_builder import single_op_builder
1716
from executorch.backends.cadence.aot.pass_utils import count_node
1817
from executorch.backends.cadence.aot.simplify_ops import (
@@ -40,82 +39,47 @@ def test_simplify_slice_scatter_op(
4039
end: Optional[int] = None,
4140
step: int = 1,
4241
):
43-
class SliceScatter(torch.nn.Module):
44-
def __init__(
45-
self, dim: int, start: Optional[int], end: Optional[int], step: int
46-
):
47-
super().__init__()
48-
self.dim = dim
49-
self.start = start
50-
self.end = end
51-
self.step = step
52-
53-
def forward(self, x: torch.Tensor, y: torch.Tensor):
54-
return torch.slice_scatter(
55-
x, y, self.dim, self.start, self.end, self.step
56-
)
57-
58-
model = SliceScatter(dim, start, end, step)
59-
x = torch.randn(in_shape)
60-
y = torch.randn(src_shape)
61-
graph_module = export_to_edge(model, (x, y)).exported_program().graph_module
62-
63-
p = SimplifySliceOpPass()
64-
65-
graph_after_passes = cast(PassResult, p(graph_module)).graph_module
66-
67-
self.assertEqual(
68-
count_node(graph_after_passes, exir_ops.edge.aten.slice_scatter.default), 0
42+
x = torch.randn(*in_shape)
43+
y = torch.randn(*src_shape)
44+
gm = single_op_builder(
45+
placeholders=(x, y),
46+
op=exir_ops.edge.aten.slice_scatter.default,
47+
args=(x, y, dim, start, end, step),
6948
)
49+
p = SimplifySliceOpPass()
50+
gm = cast(PassResult, p(gm)).graph_module
51+
self.assertEqual(count_node(gm, exir_ops.edge.aten.slice_scatter.default), 0)
7052

7153
@parameterized.expand(
7254
[
73-
[(3, 16, 5), (3, 0, 5), 1, 15, 3, 3],
55+
[(3, 16, 5), 1, 15, 3, 3],
7456
]
7557
)
7658
@torch.no_grad()
7759
def test_simplify_slice_op(
7860
self,
7961
in_shape: Tuple[int],
80-
src_shape: Tuple[int],
8162
dim: int,
8263
start: Optional[int] = None,
8364
end: Optional[int] = None,
8465
step: int = 1,
8566
):
86-
class SliceCopy(torch.nn.Module):
87-
def __init__(
88-
self, dim: int, start: Optional[int], end: Optional[int], step: int
89-
):
90-
super().__init__()
91-
self.dim = dim
92-
self.start = start
93-
self.end = end
94-
self.step = step
95-
96-
def forward(self, x: torch.Tensor) -> torch.Tensor:
97-
return torch.slice_copy(
98-
x, dim=self.dim, start=self.start, end=self.end, step=self.step
99-
)
100-
101-
# Create a model with single slice copy op.
102-
model = SliceCopy(dim, start, end, step)
103-
x = torch.randn(in_shape)
104-
graph_module = export_to_edge(model, (x,)).exported_program().graph_module
105-
self.assertEqual(
106-
count_node(graph_module, exir_ops.edge.aten.slice_copy.Tensor), 1
67+
x = torch.randn(*in_shape)
68+
gm = single_op_builder(
69+
placeholders=(x,),
70+
op=exir_ops.edge.aten.slice_copy.Tensor,
71+
args=(
72+
x,
73+
dim,
74+
start,
75+
end,
76+
step,
77+
),
10778
)
108-
10979
p = SimplifySliceOpPass()
110-
111-
graph_after_passes = cast(PassResult, p(graph_module)).graph_module
112-
113-
self.assertEqual(
114-
count_node(graph_after_passes, exir_ops.edge.aten.slice_copy.Tensor), 0
115-
)
116-
self.assertEqual(
117-
count_node(graph_after_passes, exir_ops.edge.aten.full.default), 1
118-
)
80+
gm = cast(PassResult, p(gm)).graph_module
81+
self.assertEqual(count_node(gm, exir_ops.edge.aten.slice_copy.Tensor), 0)
82+
self.assertEqual(count_node(gm, exir_ops.edge.aten.full.default), 1)
11983

12084
def test_simplify_slice_op_args(self) -> None:
12185
x = torch.rand(4, 5)
@@ -125,24 +89,10 @@ def test_simplify_slice_op_args(self) -> None:
12589
args=(x, 1),
12690
kwargs={"end": 3},
12791
)
128-
self.assertEqual(
129-
[
130-
(n.args[1:], n.kwargs)
131-
for n in gm.graph.find_nodes(
132-
op="call_function", target=exir_ops.edge.aten.slice_copy.Tensor
133-
)
134-
],
135-
[((1,), {"end": 3})],
136-
)
137-
92+
original_slice_copy = list(gm.graph.nodes)[1]
93+
self.assertEqual(original_slice_copy.args[1:], (1,))
94+
self.assertEqual(original_slice_copy.kwargs, {"end": 3})
13895
gm = BindOptionalArgsPass().call(gm).graph_module
139-
140-
self.assertEqual(
141-
[
142-
(n.args[1:], n.kwargs)
143-
for n in gm.graph.find_nodes(
144-
op="call_function", target=exir_ops.edge.aten.slice_copy.Tensor
145-
)
146-
],
147-
[((1, None, 3, 1), {})],
148-
)
96+
modified_slice_copy = list(gm.graph.nodes)[1]
97+
self.assertEqual(modified_slice_copy.args[1:], (1, None, 3, 1))
98+
self.assertEqual(modified_slice_copy.kwargs, {})

‎backends/qualcomm/_passes/qnn_pass_manager.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -135,8 +135,8 @@ def get_to_edge_transform_passes(
135135
from executorch.backends.qualcomm.builders import node_visitor
136136
from executorch.exir.dialects._ops import ops as exir_ops
137137

138-
node_visitor.q_ops.add(exir_ops.edge.pt2e_quant.quantize_affine.default)
139-
node_visitor.dq_ops.add(exir_ops.edge.pt2e_quant.dequantize_affine.default)
138+
node_visitor.q_ops.add(exir_ops.edge.torchao.quantize_affine.default)
139+
node_visitor.dq_ops.add(exir_ops.edge.torchao.dequantize_affine.default)
140140

141141
passes_job = (
142142
passes_job if passes_job is not None else get_capture_program_passes()

‎backends/qualcomm/builders/node_visitor.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -265,8 +265,8 @@ def get_quant_encoding_conf(
265265
)
266266
# TODO: refactor this when target could be correctly detected
267267
per_block_encoding = {
268-
exir_ops.edge.pt2e_quant.quantize_affine.default,
269-
exir_ops.edge.pt2e_quant.dequantize_affine.default,
268+
exir_ops.edge.torchao.quantize_affine.default,
269+
exir_ops.edge.torchao.dequantize_affine.default,
270270
}
271271
if quant_attrs[QCOM_ENCODING] in per_block_encoding:
272272
return self.make_qnn_per_block_config(node, quant_attrs)

‎backends/qualcomm/partition/utils.py‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -57,7 +57,7 @@ def get_skip_decomp_table() -> List[torch._ops.OperatorBase]:
5757
torch.ops.aten.upsample_bicubic2d.vec,
5858
# This request is ignored because it is in a blocklist. Refer to exir/program/_program.py
5959
torch.ops.aten.unbind.int,
60-
torch.ops.pt2e_quant.quantize_affine.default,
61-
torch.ops.pt2e_quant.dequantize_affine.default,
60+
torch.ops.torchao.quantize_affine.default,
61+
torch.ops.torchao.dequantize_affine.default,
6262
]
6363
return do_not_decompose

0 commit comments

Comments
 (0)