Skip to content

Revert "Arm backend: Fix sdpa in generate_op_support.py" #23409

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

Closed
zingo wants to merge 1 commit into from
Closed

Revert "Arm backend: Fix sdpa in generate_op_support.py" #23409

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
28 changes: 0 additions & 28 deletions backends/arm/scripts/docgen/generate_op_support.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -356,7 +356,6 @@ def _activate_backend(backend: str | BackendConfig) -> BackendConfig:
"torch.ops.aten.native_layer_norm.default",
"torch.ops.aten.embedding.default",
"torch.ops.aten.pow.Tensor_Tensor",
"torch.ops.aten.scaled_dot_product_attention.default",
}


Expand Down Expand Up @@ -405,18 +404,6 @@ def _activate_backend(backend: str | BackendConfig) -> BackendConfig:
): {
"INT": {"torch.ops.aten.embedding.default"},
},
(
"backends/arm/test/ops/test_sdpa.py",
"test_sdpa_vgf_no_quant",
): {
"FP": {"torch.ops.aten.scaled_dot_product_attention.default"},
},
(
"backends/arm/test/ops/test_sdpa.py",
"test_sdpa_vgf_quant",
): {
"INT": {"torch.ops.aten.scaled_dot_product_attention.default"},
},
}

# Existing U55 runtime tests below intentionally suppress direct ATen/Edge
Expand Down Expand Up @@ -459,12 +446,6 @@ def _activate_backend(backend: str | BackendConfig) -> BackendConfig:
): {
"INT": {"torch.ops.aten.pow.Tensor_Tensor"},
},
(
"backends/arm/test/ops/test_sdpa.py",
"test_sdpa_u55_INT",
): {
"INT": {"torch.ops.aten.scaled_dot_product_attention.default"},
},
}

# Existing U85 runtime tests below intentionally suppress direct ATen/Edge
Expand Down Expand Up @@ -497,12 +478,6 @@ def _activate_backend(backend: str | BackendConfig) -> BackendConfig:
): {
"INT": {"torch.ops.aten.pow.Tensor_Tensor"},
},
(
"backends/arm/test/ops/test_sdpa.py",
"test_sdpa_u85_INT",
): {
"INT": {"torch.ops.aten.scaled_dot_product_attention.default"},
},
}


Expand Down Expand Up @@ -635,9 +610,6 @@ def _active_explicit_backend_coverage() -> dict[tuple[str, str], dict[str, set[s
"torch.nn.Embedding",
"torch.nn.functional.embedding",
),
"torch.ops.aten.scaled_dot_product_attention.default": (
"torch.nn.functional.scaled_dot_product_attention",
),
# Activations.
"torch.ops.aten.relu.default": ("torch.relu", "torch.nn.ReLU"),
"torch.ops.aten.relu_.default": ("torch.Tensor.relu_",),
Expand Down
33 changes: 0 additions & 33 deletions backends/arm/test/misc/test_docgen_op_support.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -215,39 +215,6 @@ def test_u85_explicit_coverage_attribution() -> None:
docgen._activate_backend(original)


SDPA_OP = "torch.ops.aten.scaled_dot_product_attention.default"


def test_sdpa_is_decomposed_with_public_api_alias() -> None:
assert SDPA_OP in docgen.DECOMPOSED_OPS
assert docgen._pytorch_api_aliases(SDPA_OP) == (
"torch.nn.functional.scaled_dot_product_attention",
)


@pytest.mark.parametrize(
("backend", "function", "profile"),
[
("vgf", "test_sdpa_vgf_no_quant", "FP"),
("vgf", "test_sdpa_vgf_quant", "INT"),
("u55", "test_sdpa_u55_INT", "INT"),
("u85", "test_sdpa_u85_INT", "INT"),
],
)
def test_sdpa_explicit_backend_coverage(
backend: str, function: str, profile: str
) -> None:
original = docgen.ACTIVE_BACKEND_KEY
try:
docgen._activate_backend(backend)
coverage = docgen._active_explicit_backend_coverage()
assert coverage[("backends/arm/test/ops/test_sdpa.py", function)][profile] == {
SDPA_OP
}
finally:
docgen._activate_backend(original)


def test_non_vgf_backend_does_not_collect_vgf_custom_partition_ops() -> None:
original = docgen.ACTIVE_BACKEND_KEY
try:
Expand Down
3 changes: 1 addition & 2 deletions docs/source/backends/arm-ethos-u/U55_op_support.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ This page lists Ethos-U55-supported PyTorch APIs and the dtype and quantization

`8x8` means 8-bit activations and 8-bit weights. `16x8` means 16-bit activations and 8-bit weights. `8x4` means 8-bit activations and 4-bit weights.

Total supported PyTorch APIs: **105**.
Total supported PyTorch APIs: **104**.

| PyTorch API | Support profile | DType | Quantization mode |
| --- | --- | --- | --- |
Expand Down Expand Up @@ -68,7 +68,6 @@ Total supported PyTorch APIs: **105**.
| `torch.nn.Conv2d` / `torch.nn.functional.conv2d` | INT | `INT8`, `INT16`, `INT4` | 8x8, 8x4, 16x8 |
| `torch.nn.ELU` / `torch.nn.functional.elu` | INT | `INT8` | 8x8 |
| `torch.nn.functional.pad` | INT | `INT8` | 8x8 |
| `torch.nn.functional.scaled_dot_product_attention` | INT | `INT8` | 8x8 |
| `torch.nn.GELU` / `torch.nn.functional.gelu` | INT | `INT8`, `INT16` | 8x8, 16x8 |
| `torch.nn.Hardsigmoid` / `torch.nn.functional.hardsigmoid` | INT | `INT8` | 8x8 |
| `torch.nn.Hardswish` / `torch.nn.functional.hardswish` | INT | `INT8` | 8x8 |
Expand Down
3 changes: 1 addition & 2 deletions docs/source/backends/arm-ethos-u/U85_op_support.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ This page lists Ethos-U85-supported PyTorch APIs and the dtype and quantization

`8x8` means 8-bit activations and 8-bit weights. `16x8` means 16-bit activations and 8-bit weights. `8x4` means 8-bit activations and 4-bit weights.

Total supported PyTorch APIs: **128**.
Total supported PyTorch APIs: **127**.

| PyTorch API | Support profile | DType | Quantization mode |
| --- | --- | --- | --- |
Expand Down Expand Up @@ -87,7 +87,6 @@ Total supported PyTorch APIs: **128**.
| `torch.nn.ELU` / `torch.nn.functional.elu` | INT | `INT8` | 8x8 |
| `torch.nn.Embedding` / `torch.nn.functional.embedding` | INT | `INT8` | 8x8 |
| `torch.nn.functional.pad` | INT | `INT8` | 8x8 |
| `torch.nn.functional.scaled_dot_product_attention` | INT | `INT8` | 8x8 |
| `torch.nn.GELU` / `torch.nn.functional.gelu` | INT | `INT8`, `INT16` | 8x8, 16x8 |
| `torch.nn.Hardsigmoid` / `torch.nn.functional.hardsigmoid` | INT | `INT8` | 8x8 |
| `torch.nn.Hardswish` / `torch.nn.functional.hardswish` | INT | `INT8` | 8x8 |
Expand Down
1 change: 0 additions & 1 deletion docs/source/backends/arm-vgf/VGF_op_support.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,7 +105,6 @@ Total supported PyTorch APIs: **158**.
| `torch.nn.ELU` / `torch.nn.functional.elu` | FP, INT | `FP32`, `INT8` | 8x8 |
| `torch.nn.Embedding` / `torch.nn.functional.embedding` | FP, INT | `FP32`, `INT8` | 8x8 |
| `torch.nn.functional.pad` | FP, INT | `FP32`, `FP16`, `BF16`, `INT8`, `INT16` | 8x8, 16x8 |
| `torch.nn.functional.scaled_dot_product_attention` | FP, INT | `FP32`, `INT8` | 8x8 |
| `torch.nn.GELU` / `torch.nn.functional.gelu` | FP, INT | `FP32`, `BF16`, `INT8` | 8x8 |
| `torch.nn.GroupNorm` / `torch.nn.functional.group_norm` | FP, INT | `FP32`, `INT8` | 8x8 |
| `torch.nn.Hardsigmoid` / `torch.nn.functional.hardsigmoid` | FP, INT | `FP32`, `INT8` | 8x8 |
Expand Down
Loading