Skip to content

Commit 4e03115

Browse files
Erik-LundellErik-Lundell
authored
Merge branch 'main' into experimental/clone_support
2 parents 62ea5c1 + a93f59e commit 4e03115

File tree

103 files changed

+3271
-701
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

103 files changed

+3271
-701
lines changed

‎.github/workflows/cuda.yml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,7 @@ jobs:
2424
strategy:
2525
fail-fast: false
2626
matrix:
27-
cuda-version: ["12.6", "12.8", "13.0"]
27+
cuda-version: ["12.6", "12.8", "12.9", "13.0"]
2828

2929
name: test-executorch-cuda-build-${{ matrix.cuda-version }}
3030
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main

‎.github/workflows/pull.yml‎

Lines changed: 11 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -862,15 +862,24 @@ jobs:
862862
# Install Node.js and Emscripten
863863
source .ci/scripts/setup-emscripten.sh
864864
865+
export PNPM_VERSION=10.24.0
866+
867+
curl -fsSL https://get.pnpm.io/install.sh | env PNPM_VERSION=$PNPM_VERSION SHELL="$(which bash)" sh -
868+
869+
export PNPM_HOME="$HOME/.local/share/pnpm"
870+
export PATH="$PNPM_HOME:$PATH"
871+
872+
pnpm --version
873+
865874
# Test selective build
866875
bash scripts/build_wasm_tests.sh ${{ matrix.enable-etdump }}
867876
868877
# Install Jest
869878
cd cmake-out-wasm/extension/wasm/test
870-
npm install --save-dev jest
879+
pnpm add -D jest@30.2.0 --ignore-scripts
871880
872881
# Run unit test
873-
npm test
882+
pnpm test
874883
875884
unittest-nxp-neutron:
876885
uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main

‎backends/aoti/aoti_backend.py‎

Lines changed: 9 additions & 24 deletions
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@
99
import typing
1010
from abc import ABC, abstractmethod
1111
from enum import Enum
12-
from typing import Any, Dict, List, Optional, Set
12+
from typing import Any, Dict, List, Set
1313

1414
import torch
1515
from executorch.backends.aoti.passes.replace_view_copy_with_view import (
@@ -91,39 +91,24 @@ def collect_unsupported_fallback_kernels(cls, missing_fallback_kernels: Set[str]
9191
)
9292

9393
def generate_c_shim_extern_kernel_call_and_collect_unsupported_kernels(
94-
self,
95-
kernel: str,
96-
args: list[str],
97-
device: str,
98-
*,
99-
debug_args: Optional[list[str]] = None,
100-
debug_handle: Optional[int] = None,
101-
):
94+
self, kernel: str, *args: Any, **kwargs: Any
95+
) -> None:
10296
if kernel not in supported_kernels:
10397
missing_fallback_kernels.add(kernel)
10498

105-
original_generate_c_shim_extern_kernel_call(
106-
self,
107-
kernel,
108-
args,
109-
device,
110-
debug_args=debug_args,
111-
debug_handle=debug_handle,
99+
return original_generate_c_shim_extern_kernel_call(
100+
self, kernel, *args, **kwargs
112101
)
113102

114103
def generate_fallback_kernel_with_runtime_lookup_aot_and_collect_unsupported_kernels(
115-
self,
116-
op_overload,
117-
raw_args,
118-
output_args,
119-
raw_outputs,
120-
):
104+
self, op_overload: Any, *args: Any, **kwargs: Any
105+
) -> None:
121106
kernel_name = getattr(op_overload, "_name", str(op_overload))
122107
if kernel_name not in supported_kernels:
123108
missing_fallback_kernels.add(kernel_name)
124109

125-
original_generate_fallback_kernel_with_runtime_lookup_aot(
126-
self, op_overload, raw_args, output_args, raw_outputs
110+
return original_generate_fallback_kernel_with_runtime_lookup_aot(
111+
self, op_overload, *args, **kwargs
127112
)
128113

129114
CppWrapperCpu.generate_c_shim_extern_kernel_call = (

‎backends/cadence/aot/ops_registrations.py‎

Lines changed: 70 additions & 54 deletions
Original file line numberDiff line numberDiff line change
@@ -1030,9 +1030,9 @@ def quantized_conv2d_nhwc_meta(
10301030
get_conv1d_output_size(
10311031
in_size,
10321032
out_channels,
1033-
stride[1],
1034-
padding[1],
1035-
dilation[1],
1033+
stride[-1],
1034+
padding[-1],
1035+
dilation[-1],
10361036
kernel_size[0],
10371037
True,
10381038
)
@@ -1074,9 +1074,9 @@ def quantized_conv2d_nchw_meta(
10741074
get_conv1d_output_size(
10751075
in_size,
10761076
out_channels,
1077-
stride[1],
1078-
padding[1],
1079-
dilation[1],
1077+
stride[-1],
1078+
padding[-1],
1079+
dilation[-1],
10801080
kernel_size[0],
10811081
False,
10821082
)
@@ -1118,9 +1118,9 @@ def quantized_conv2d_nchw_per_tensor_meta(
11181118
get_conv1d_output_size(
11191119
in_size,
11201120
out_channels,
1121-
stride[1],
1122-
padding[1],
1123-
dilation[1],
1121+
stride[-1],
1122+
padding[-1],
1123+
dilation[-1],
11241124
kernel_size[0],
11251125
False,
11261126
)
@@ -1162,9 +1162,9 @@ def quantized_conv2d_nhwc_per_tensor_meta(
11621162
get_conv1d_output_size(
11631163
in_size,
11641164
out_channels,
1165-
stride[1],
1166-
padding[1],
1167-
dilation[1],
1165+
stride[-1],
1166+
padding[-1],
1167+
dilation[-1],
11681168
kernel_size[0],
11691169
True,
11701170
)
@@ -1211,9 +1211,9 @@ def quantized_conv2d_nchw_asym8sxsym8s_asym8s_per_tensor_meta(
12111211
get_conv1d_output_size(
12121212
in_size,
12131213
out_channels,
1214-
stride[1],
1215-
padding[1],
1216-
dilation[1],
1214+
stride[-1],
1215+
padding[-1],
1216+
dilation[-1],
12171217
kernel_size[0],
12181218
False,
12191219
)
@@ -1260,9 +1260,9 @@ def quantized_conv2d_nchw_asym8uxsym8u_asym8u_per_tensor_meta(
12601260
get_conv1d_output_size(
12611261
in_size,
12621262
out_channels,
1263-
stride[1],
1264-
padding[1],
1265-
dilation[1],
1263+
stride[-1],
1264+
padding[-1],
1265+
dilation[-1],
12661266
kernel_size[0],
12671267
False,
12681268
)
@@ -1309,9 +1309,9 @@ def quantized_conv2d_nhwc_asym8sxsym8s_asym8s_per_tensor_meta(
13091309
get_conv1d_output_size(
13101310
in_size,
13111311
out_channels,
1312-
stride[1],
1313-
padding[1],
1314-
dilation[1],
1312+
stride[-1],
1313+
padding[-1],
1314+
dilation[-1],
13151315
kernel_size[0],
13161316
True,
13171317
)
@@ -1358,9 +1358,9 @@ def quantized_conv2d_nhwc_asym8uxsym8u_asym8u_per_tensor_meta(
13581358
get_conv1d_output_size(
13591359
in_size,
13601360
out_channels,
1361-
stride[1],
1362-
padding[1],
1363-
dilation[1],
1361+
stride[-1],
1362+
padding[-1],
1363+
dilation[-1],
13641364
kernel_size[0],
13651365
True,
13661366
)
@@ -1407,9 +1407,9 @@ def quantized_conv2d_nchw_dilated_asym8sxsym8s_asym8s_per_tensor_meta(
14071407
get_conv1d_output_size(
14081408
in_size,
14091409
out_channels,
1410-
stride[1],
1411-
padding[1],
1412-
dilation[1],
1410+
stride[-1],
1411+
padding[-1],
1412+
dilation[-1],
14131413
kernel_size[0],
14141414
False,
14151415
)
@@ -1456,9 +1456,9 @@ def quantized_conv2d_nchw_dilated_asym8uxsym8u_asym8u_per_tensor_meta(
14561456
get_conv1d_output_size(
14571457
in_size,
14581458
out_channels,
1459-
stride[1],
1460-
padding[1],
1461-
dilation[1],
1459+
stride[-1],
1460+
padding[-1],
1461+
dilation[-1],
14621462
kernel_size[0],
14631463
False,
14641464
)
@@ -1505,9 +1505,9 @@ def quantized_conv2d_nhwc_dilated_asym8sxsym8s_asym8s_per_tensor_meta(
15051505
get_conv1d_output_size(
15061506
in_size,
15071507
out_channels,
1508-
stride[1],
1509-
padding[1],
1510-
dilation[1],
1508+
stride[-1],
1509+
padding[-1],
1510+
dilation[-1],
15111511
kernel_size[0],
15121512
True,
15131513
)
@@ -1554,9 +1554,9 @@ def quantized_conv2d_nhwc_dilated_asym8uxsym8u_asym8u_per_tensor_meta(
15541554
get_conv1d_output_size(
15551555
in_size,
15561556
out_channels,
1557-
stride[1],
1558-
padding[1],
1559-
dilation[1],
1557+
stride[-1],
1558+
padding[-1],
1559+
dilation[-1],
15601560
kernel_size[0],
15611561
True,
15621562
)
@@ -1605,9 +1605,9 @@ def quantized_conv2d_nchw_depthwise_asym8sxsym8s_asym8s_per_tensor_meta(
16051605
get_conv1d_output_size(
16061606
in_size,
16071607
out_channels,
1608-
stride[1],
1609-
padding[1],
1610-
dilation[1],
1608+
stride[-1],
1609+
padding[-1],
1610+
dilation[-1],
16111611
kernel_size[0],
16121612
False,
16131613
)
@@ -1656,9 +1656,9 @@ def quantized_conv2d_nchw_depthwise_asym8uxsym8u_asym8u_per_tensor_meta(
16561656
get_conv1d_output_size(
16571657
in_size,
16581658
out_channels,
1659-
stride[1],
1660-
padding[1],
1661-
dilation[1],
1659+
stride[-1],
1660+
padding[-1],
1661+
dilation[-1],
16621662
kernel_size[0],
16631663
False,
16641664
)
@@ -1707,9 +1707,9 @@ def quantized_conv2d_nhwc_depthwise_asym8sxsym8s_asym8s_per_tensor_meta(
17071707
get_conv1d_output_size(
17081708
in_size,
17091709
out_channels,
1710-
stride[1],
1711-
padding[1],
1712-
dilation[1],
1710+
stride[-1],
1711+
padding[-1],
1712+
dilation[-1],
17131713
kernel_size[0],
17141714
True,
17151715
)
@@ -1758,9 +1758,9 @@ def quantized_conv2d_nhwc_depthwise_asym8uxsym8u_asym8u_per_tensor_meta(
17581758
get_conv1d_output_size(
17591759
in_size,
17601760
out_channels,
1761-
stride[1],
1762-
padding[1],
1763-
dilation[1],
1761+
stride[-1],
1762+
padding[-1],
1763+
dilation[-1],
17641764
kernel_size[0],
17651765
True,
17661766
)
@@ -2178,15 +2178,31 @@ def conv1d_meta(
21782178
dilation: Tuple[int],
21792179
groups: int,
21802180
) -> torch.Tensor:
2181+
# Validate tensor dimensions
2182+
assert len(input.shape) == 3, f"Conv1d expects 3D input, got {len(input.shape)}D"
2183+
assert len(weight.shape) == 3, f"Conv1d expects 3D weight, got {len(weight.shape)}D"
2184+
2185+
# Extract dimensions
2186+
batch_size, in_channels, length = input.shape
2187+
out_channels, weight_in_channels, kernel_size = weight.shape
2188+
2189+
# Validate groups parameter and channel consistency
2190+
assert groups > 0, f"groups must be positive, got {groups}"
21812191
assert (
2182-
len(weight.shape) == 3
2183-
), f"Conv1d expects a 3D weight, got {len(weight.shape)}D"
2184-
out_channels, _, kernel_size = weight.shape
2185-
in_size = input.shape
2186-
assert len(in_size) == 3, f"conv1d expects 3D input, got {len(in_size)}D"
2192+
in_channels % groups == 0
2193+
), f"in_channels ({in_channels}) must be divisible by groups ({groups})"
2194+
assert (
2195+
out_channels % groups == 0
2196+
), f"out_channels ({out_channels}) must be divisible by groups ({groups})"
2197+
2198+
# Validate weight channels match input channels divided by groups
2199+
expected_weight_in_channels = in_channels // groups
2200+
assert (
2201+
weight_in_channels == expected_weight_in_channels
2202+
), f"Expected weight to have {expected_weight_in_channels} input channels (in_channels/groups), but got {weight_in_channels}"
21872203

21882204
output_size = get_conv1d_output_size(
2189-
in_size,
2205+
input.shape,
21902206
out_channels,
21912207
stride[0],
21922208
padding[0],

‎backends/cadence/aot/ref_implementations.py‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -788,9 +788,9 @@ def quantized_conv_per_tensor(
788788
(input_tensor - in_zero_point).float(),
789789
(weight - weight_zero_point).float(),
790790
(bias * bias_scale).float(),
791-
stride[1],
792-
padding[1],
793-
dilation[1],
791+
stride[-1],
792+
padding[-1],
793+
dilation[-1],
794794
groups,
795795
)
796796

0 commit comments

Comments
 (0)