torchvulkan 0.2.1__tar.gz → 0.2.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/.github/workflows/release.yaml +13 -10
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/.github/workflows/test_torchvulkan.yml +6 -7
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/CMakeLists.txt +3 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/PKG-INFO +5 -4
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/README.md +2 -2
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/pyproject.toml +4 -4
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/backend.cpp +1 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/decompositions.cpp +6 -1
- torchvulkan-0.2.2/src/api/fallback.cpp +13 -0
- torchvulkan-0.2.2/src/api/fallback.h +11 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/binary.cpp +3 -2
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/bitwise.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/compare.cpp +4 -3
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/fill.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/grid_sample.cpp +1 -2
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/indexing.cpp +6 -5
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/linalg.cpp +5 -11
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/matmul.cpp +5 -4
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/nllloss.cpp +3 -2
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/reduce.cpp +3 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/scan.cpp +3 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/scatter.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/search.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/sort.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/special.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/unary.cpp +3 -2
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/registration.cpp +0 -14
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/tensor/transfer.cpp +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/core/dtype.slang +3 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/eig.slang +1 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/grid_sample.slang +120 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/householder.slang +2 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/allocator.cpp +50 -17
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/allocator.h +0 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/cache.cpp +34 -11
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/cache.h +11 -5
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/device_context.cpp +36 -2
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/device_context.h +5 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/shader.cpp +0 -7
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/shader.h +0 -1
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/vulkan_context.cpp +1 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/vulkan_context.h +2 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/tests/test_pytorch_ops.py +12 -7
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/torchvulkan/__init__.py +15 -5
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/torchvulkan/_version.py +3 -3
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/tox.ini +1 -3
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/.github/actions/setup-toolchain/action.yml +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/.github/scripts/fetch-moltenvk-macos.sh +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/.github/scripts/install-toolchain-linux.sh +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/.gitignore +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/LICENSE +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/benchmark/benchmark.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/examples/conv.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/examples/llm.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/examples/mlp.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/extern/vma/vk_mem_alloc.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/extern/volk/volk.c +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/extern/volk/volk.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/licenses/MoltenVK-LICENSE +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/media/benchmark-cuda.png +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/requirements.txt +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/scripts/build_shaders.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/scripts/config.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/scripts/pack_shaders.py +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/guard.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/hook.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/complex.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/conv.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/ctc.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/distributions.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/dropout.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/fft.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/helpers.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/internal.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/nn.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/optim.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/pool.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/random.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/sampling.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/shape.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/ops/upsample.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/tensor/allocation.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/tensor/cast.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/api/tensor/views.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/CMakeLists.txt +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/core/cast.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/core/indexing.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/core/philox.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/core/special_functions.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/arg_reduce.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/binary.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/cast.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/cholesky.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/compare.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/conv.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/copy.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/ctc.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/distribution.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/dropout.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/embedding_backward.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/fft.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/fill.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/index_gather.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/integer_binary.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/jacobi.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/ldl.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/lu.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/matmul_coop.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/matmul_simd.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/multinomial.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/nllloss.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/pool.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/random.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/reduce.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/reduce_subgroup.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/resample.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/resample_u8.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/scan.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/scan_arg.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/scatter.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/searchsorted.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/sort.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/special.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/triangular_solve.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/unary.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/unary_ops.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/shaders/ops/where.slang +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/builders.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/dispatch.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/dispatch.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/generator.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/generator.h +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/memory.cpp +0 -0
- {torchvulkan-0.2.1 → torchvulkan-0.2.2}/src/vulkan/memory.h +0 -0
|
@@ -21,7 +21,7 @@ jobs:
|
|
|
21
21
|
|
|
22
22
|
- uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5
|
|
23
23
|
with:
|
|
24
|
-
python-version: "3.
|
|
24
|
+
python-version: "3.15"
|
|
25
25
|
|
|
26
26
|
- name: Build sdist
|
|
27
27
|
run: |
|
|
@@ -34,18 +34,20 @@ jobs:
|
|
|
34
34
|
path: dist/*.tar.gz
|
|
35
35
|
|
|
36
36
|
build_wheels:
|
|
37
|
-
name: Wheels (${{ matrix.name }})
|
|
37
|
+
name: Wheels (${{ matrix.name }} / ${{ matrix.python }})
|
|
38
38
|
runs-on: ${{ matrix.os }}
|
|
39
39
|
strategy:
|
|
40
40
|
fail-fast: false
|
|
41
41
|
matrix:
|
|
42
|
+
os: [ubuntu-latest, macos-15, windows-latest]
|
|
43
|
+
python: [cp311, cp312, cp313, cp314, cp315]
|
|
42
44
|
include:
|
|
43
|
-
-
|
|
44
|
-
|
|
45
|
-
-
|
|
46
|
-
|
|
47
|
-
-
|
|
48
|
-
|
|
45
|
+
- os: ubuntu-latest
|
|
46
|
+
name: Linux x86_64
|
|
47
|
+
- os: macos-15
|
|
48
|
+
name: macOS Apple Silicon
|
|
49
|
+
- os: windows-latest
|
|
50
|
+
name: Windows x86_64
|
|
49
51
|
steps:
|
|
50
52
|
- uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
|
|
51
53
|
with:
|
|
@@ -67,13 +69,14 @@ jobs:
|
|
|
67
69
|
uses: ./.github/actions/setup-toolchain
|
|
68
70
|
|
|
69
71
|
- name: Build wheels
|
|
70
|
-
uses: pypa/cibuildwheel@
|
|
72
|
+
uses: pypa/cibuildwheel@f9b966cfadf4109556cadc5bb39ddb72d0ce664e # v4.3.5
|
|
71
73
|
env:
|
|
72
74
|
CIBW_BUILD_VERBOSITY: "1"
|
|
75
|
+
CIBW_BUILD: ${{ matrix.python }}-*
|
|
73
76
|
|
|
74
77
|
- uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
|
|
75
78
|
with:
|
|
76
|
-
name: dist-${{ matrix.os }}
|
|
79
|
+
name: dist-${{ matrix.os }}-${{ matrix.python }}
|
|
77
80
|
path: wheelhouse/*.whl
|
|
78
81
|
|
|
79
82
|
pypi-publish:
|
|
@@ -8,7 +8,7 @@ on:
|
|
|
8
8
|
|
|
9
9
|
jobs:
|
|
10
10
|
build-and-test:
|
|
11
|
-
name: ${{ matrix.os }} / Py${{ matrix.python-version }}
|
|
11
|
+
name: ${{ matrix.os }} / Py${{ matrix.python-version }}
|
|
12
12
|
runs-on: ${{ matrix.os }}
|
|
13
13
|
env:
|
|
14
14
|
SKBUILD_CMAKE_GENERATOR: Ninja
|
|
@@ -16,8 +16,7 @@ jobs:
|
|
|
16
16
|
fail-fast: false
|
|
17
17
|
matrix:
|
|
18
18
|
os: [ubuntu-latest, macos-15, windows-latest]
|
|
19
|
-
python-version: ["3.
|
|
20
|
-
torch-factor: ["torch210", "torch213"]
|
|
19
|
+
python-version: ["3.11", "3.13", "3.15"]
|
|
21
20
|
|
|
22
21
|
steps:
|
|
23
22
|
- name: Checkout Code
|
|
@@ -50,7 +49,7 @@ jobs:
|
|
|
50
49
|
DNNL_MAX_CPU_ISA: AVX2 # legacy name for the same oneDNN ISA cap (older bundled oneDNN)
|
|
51
50
|
LP_NATIVE_VECTOR_WIDTH: "256" # cap lavapipe's LLVM JIT to AVX2; AVX-512 codegen faults (0xc000001d) on GHA runners
|
|
52
51
|
run: |
|
|
53
|
-
tox -e py
|
|
52
|
+
tox -e py
|
|
54
53
|
|
|
55
54
|
validation:
|
|
56
55
|
name: validation layers (lavapipe, debug)
|
|
@@ -83,7 +82,7 @@ jobs:
|
|
|
83
82
|
run: |
|
|
84
83
|
python -m pip install --upgrade pip
|
|
85
84
|
pip install tox
|
|
86
|
-
tox -e py
|
|
85
|
+
tox -e py -- -k "test_correctness_add or test_correctness_mm or test_correctness_sum or test_correctness_sort or test_correctness_index or test_correctness_cumsum or test_gradients_nn_functional_linear"
|
|
87
86
|
|
|
88
87
|
sanitizers:
|
|
89
88
|
name: ASan + UBSan (lavapipe)
|
|
@@ -108,7 +107,7 @@ jobs:
|
|
|
108
107
|
SKBUILD_CMAKE_ARGS: "-DCMAKE_CXX_FLAGS=-fsanitize=address,undefined -fno-omit-frame-pointer;-DCMAKE_MODULE_LINKER_FLAGS=-fsanitize=address,undefined"
|
|
109
108
|
run: |
|
|
110
109
|
python -m pip install --upgrade pip
|
|
111
|
-
pip install "
|
|
110
|
+
pip install $(python -c 'import tomllib; print(" ".join(tomllib.load(open("pyproject.toml", "rb"))["build-system"]["requires"]))') pytest expecttest numpy ninja
|
|
112
111
|
pip install . --no-build-isolation -v
|
|
113
112
|
|
|
114
113
|
- name: Run tests with the sanitizer runtimes preloaded
|
|
@@ -123,7 +122,7 @@ jobs:
|
|
|
123
122
|
run: |
|
|
124
123
|
cd "$RUNNER_TEMP"
|
|
125
124
|
export LD_PRELOAD="$(gcc -print-file-name=libasan.so):$(gcc -print-file-name=libubsan.so)"
|
|
126
|
-
python -m pytest "$GITHUB_WORKSPACE/tests/test_pytorch_ops.py" -q -k "
|
|
125
|
+
python -m pytest "$GITHUB_WORKSPACE/tests/test_pytorch_ops.py" -q -k "test_correctness_add or test_correctness_mm or test_correctness_sort or test_correctness_index or test_correctness_sum"
|
|
127
126
|
|
|
128
127
|
testing-success:
|
|
129
128
|
name: testing success
|
|
@@ -30,6 +30,9 @@ target_compile_features(torchvulkan_lib PRIVATE cxx_std_20)
|
|
|
30
30
|
|
|
31
31
|
install(TARGETS torchvulkan_lib DESTINATION torchvulkan)
|
|
32
32
|
|
|
33
|
+
file(WRITE "${TORCHVULKAN_GENERATED_DIR}/_build_info.py" "torch_version = \"${Torch_VERSION}\"\n")
|
|
34
|
+
install(FILES "${TORCHVULKAN_GENERATED_DIR}/_build_info.py" DESTINATION torchvulkan)
|
|
35
|
+
|
|
33
36
|
if(APPLE)
|
|
34
37
|
find_library(TORCHVULKAN_MOLTENVK NAMES libMoltenVK.dylib)
|
|
35
38
|
if(TORCHVULKAN_MOLTENVK)
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: torchvulkan
|
|
3
|
-
Version: 0.2.
|
|
3
|
+
Version: 0.2.2
|
|
4
4
|
Summary: Cross-Platform Vulkan Backend for PyTorch
|
|
5
5
|
Author: Zain Sharief
|
|
6
6
|
License-Expression: MIT
|
|
@@ -9,7 +9,8 @@ License-File: licenses/MoltenVK-LICENSE
|
|
|
9
9
|
Project-URL: Homepage, https://github.com/zainsharief/torchvulkan
|
|
10
10
|
Project-URL: Repository, https://github.com/zainsharief/torchvulkan
|
|
11
11
|
Project-URL: Issues, https://github.com/zainsharief/torchvulkan/issues
|
|
12
|
-
Requires-
|
|
12
|
+
Requires-Python: >=3.11
|
|
13
|
+
Requires-Dist: torch<2.16,>=2.15.0
|
|
13
14
|
Provides-Extra: test
|
|
14
15
|
Requires-Dist: pytest; extra == "test"
|
|
15
16
|
Requires-Dist: tox; extra == "test"
|
|
@@ -23,7 +24,7 @@ Description-Content-Type: text/markdown
|
|
|
23
24
|
<p align="center">
|
|
24
25
|
<a href="https://github.com/zainsharief/torchvulkan/actions/workflows/test_torchvulkan.yml"><img alt="CI" src="https://github.com/zainsharief/torchvulkan/actions/workflows/test_torchvulkan.yml/badge.svg"></a>
|
|
25
26
|
<a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-blue.svg"></a>
|
|
26
|
-
<img alt="Python 3.
|
|
27
|
+
<img alt="Python 3.11+" src="https://img.shields.io/badge/python-3.11%2B-blue">
|
|
27
28
|
<img alt="Status: experimental" src="https://img.shields.io/badge/status-experimental-orange">
|
|
28
29
|
</p>
|
|
29
30
|
|
|
@@ -195,6 +196,7 @@ TORCHVULKAN_STRICT=1 python your_script.py
|
|
|
195
196
|
| Variable | Effect |
|
|
196
197
|
| --- | --- |
|
|
197
198
|
| `TORCHVULKAN_STRICT=1` | Raise an error instead of falling back to the CPU. |
|
|
199
|
+
| `TORCHVULKAN_STRICT_ALLOCATION=1` | Raise an out-of-memory error instead of allocating past the VRAM budget. By default torchvulkan allocates anyway (with a one-time warning), and the driver may put the memory in system RAM, which is much slower. |
|
|
198
200
|
| `TORCHVULKAN_SYSTEM_VULKAN=1` | macOS: use the system Vulkan loader instead of the bundled MoltenVK. |
|
|
199
201
|
| `TORCHVULKAN_VALIDATION=1` | Enable the Khronos validation layer, if it's installed. `=sync` also enables synchronization validation. On macOS this needs `TORCHVULKAN_SYSTEM_VULKAN=1`. |
|
|
200
202
|
| `TORCHVULKAN_PROFILE=1` | Print each compute pipeline's driver statistics (such as register use) when it's compiled, on drivers that support `VK_KHR_pipeline_executable_properties`. |
|
|
@@ -208,7 +210,6 @@ TORCHVULKAN_STRICT=1 python your_script.py
|
|
|
208
210
|
- `float64` runs on the GPU only on devices with `shaderFloat64`; elsewhere it falls back to the CPU. Transcendental functions such as `exp`, `sin` and `tanh` are computed in `float32` precision for `float64` tensors.
|
|
209
211
|
- `bfloat16` runs on the GPU only on devices with `VK_KHR_shader_bfloat16`; elsewhere it falls back to the CPU.
|
|
210
212
|
- Complex tensors can live on the device and FFTs run on the GPU, but most complex arithmetic falls back to the CPU.
|
|
211
|
-
- `torch>=2.10.0` only. The extension is compiled against 2.13.0 but has testing support for >=2.10.0.
|
|
212
213
|
- APIs and coverage may change before `1.0.0`.
|
|
213
214
|
|
|
214
215
|
## How it works
|
|
@@ -6,7 +6,7 @@
|
|
|
6
6
|
<p align="center">
|
|
7
7
|
<a href="https://github.com/zainsharief/torchvulkan/actions/workflows/test_torchvulkan.yml"><img alt="CI" src="https://github.com/zainsharief/torchvulkan/actions/workflows/test_torchvulkan.yml/badge.svg"></a>
|
|
8
8
|
<a href="LICENSE"><img alt="License: MIT" src="https://img.shields.io/badge/License-MIT-blue.svg"></a>
|
|
9
|
-
<img alt="Python 3.
|
|
9
|
+
<img alt="Python 3.11+" src="https://img.shields.io/badge/python-3.11%2B-blue">
|
|
10
10
|
<img alt="Status: experimental" src="https://img.shields.io/badge/status-experimental-orange">
|
|
11
11
|
</p>
|
|
12
12
|
|
|
@@ -178,6 +178,7 @@ TORCHVULKAN_STRICT=1 python your_script.py
|
|
|
178
178
|
| Variable | Effect |
|
|
179
179
|
| --- | --- |
|
|
180
180
|
| `TORCHVULKAN_STRICT=1` | Raise an error instead of falling back to the CPU. |
|
|
181
|
+
| `TORCHVULKAN_STRICT_ALLOCATION=1` | Raise an out-of-memory error instead of allocating past the VRAM budget. By default torchvulkan allocates anyway (with a one-time warning), and the driver may put the memory in system RAM, which is much slower. |
|
|
181
182
|
| `TORCHVULKAN_SYSTEM_VULKAN=1` | macOS: use the system Vulkan loader instead of the bundled MoltenVK. |
|
|
182
183
|
| `TORCHVULKAN_VALIDATION=1` | Enable the Khronos validation layer, if it's installed. `=sync` also enables synchronization validation. On macOS this needs `TORCHVULKAN_SYSTEM_VULKAN=1`. |
|
|
183
184
|
| `TORCHVULKAN_PROFILE=1` | Print each compute pipeline's driver statistics (such as register use) when it's compiled, on drivers that support `VK_KHR_pipeline_executable_properties`. |
|
|
@@ -191,7 +192,6 @@ TORCHVULKAN_STRICT=1 python your_script.py
|
|
|
191
192
|
- `float64` runs on the GPU only on devices with `shaderFloat64`; elsewhere it falls back to the CPU. Transcendental functions such as `exp`, `sin` and `tanh` are computed in `float32` precision for `float64` tensors.
|
|
192
193
|
- `bfloat16` runs on the GPU only on devices with `VK_KHR_shader_bfloat16`; elsewhere it falls back to the CPU.
|
|
193
194
|
- Complex tensors can live on the device and FFTs run on the GPU, but most complex arithmetic falls back to the CPU.
|
|
194
|
-
- `torch>=2.10.0` only. The extension is compiled against 2.13.0 but has testing support for >=2.10.0.
|
|
195
195
|
- APIs and coverage may change before `1.0.0`.
|
|
196
196
|
|
|
197
197
|
## How it works
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
[build-system]
|
|
2
|
-
requires = ["torch==2.
|
|
2
|
+
requires = ["torch==2.15.0", "scikit-build-core", "setuptools_scm"]
|
|
3
3
|
build-backend = "scikit_build_core.build"
|
|
4
4
|
|
|
5
5
|
[project]
|
|
@@ -7,7 +7,8 @@ name = "torchvulkan"
|
|
|
7
7
|
description = "Cross-Platform Vulkan Backend for PyTorch"
|
|
8
8
|
readme = "README.md"
|
|
9
9
|
dynamic = ["version"]
|
|
10
|
-
|
|
10
|
+
requires-python = ">=3.11"
|
|
11
|
+
dependencies = ["torch>=2.15.0,<2.16"]
|
|
11
12
|
license = "MIT"
|
|
12
13
|
license-files = ["LICENSE", "licenses/MoltenVK-LICENSE"]
|
|
13
14
|
authors = [{ name = "Zain Sharief"}]
|
|
@@ -33,10 +34,9 @@ write_to = "torchvulkan/_version.py"
|
|
|
33
34
|
test = ["pytest", "tox"]
|
|
34
35
|
|
|
35
36
|
[tool.cibuildwheel]
|
|
36
|
-
build = "
|
|
37
|
+
build = "cp311-* cp312-* cp313-* cp314-* cp315-*"
|
|
37
38
|
skip = "*-musllinux_* *_i686 *-win32"
|
|
38
39
|
test-command = 'python -c "import torchvulkan; print(torchvulkan.__version__)"'
|
|
39
|
-
test-requires = "torch>=2.10.0"
|
|
40
40
|
|
|
41
41
|
[tool.cibuildwheel.linux]
|
|
42
42
|
manylinux-x86_64-image = "manylinux_2_34"
|
|
@@ -49,7 +49,7 @@ PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
|
|
|
49
49
|
});
|
|
50
50
|
|
|
51
51
|
m.def("empty_cache", []() {
|
|
52
|
-
for (DeviceContext* device : VulkanContext::Instance().devices) device->
|
|
52
|
+
for (DeviceContext* device : VulkanContext::Instance().devices) device->releaseCachedBuffers(SIZE_MAX, MemoryUsage::DEVICE_ONLY);
|
|
53
53
|
}, py::call_guard<py::gil_scoped_release>());
|
|
54
54
|
|
|
55
55
|
m.def("_save_pipeline_caches", []() {
|
|
@@ -338,7 +338,12 @@ at::Tensor mish_vulkan(const at::Tensor& self)
|
|
|
338
338
|
at::Tensor logit_vulkan(const at::Tensor& self, c10::optional<double> eps)
|
|
339
339
|
{
|
|
340
340
|
at::Tensor x = promote_to_float(self);
|
|
341
|
-
|
|
341
|
+
// lo is checked first, matching the CPU when eps > 0.5 makes lo > hi
|
|
342
|
+
if (eps.has_value()) {
|
|
343
|
+
at::Tensor lo = at::full({}, *eps, x.options());
|
|
344
|
+
at::Tensor hi = at::full({}, 1.0 - *eps, x.options());
|
|
345
|
+
x = at::where(at::lt(x, lo), lo, at::where(at::gt(x, hi), hi, x));
|
|
346
|
+
}
|
|
342
347
|
return at::log(at::div(x, at::rsub(x, 1)));
|
|
343
348
|
}
|
|
344
349
|
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
#include <torch/extension.h>
|
|
2
|
+
#include <ATen/native/CPUFallback.h>
|
|
3
|
+
#include "api/fallback.h"
|
|
4
|
+
|
|
5
|
+
void vulkan_cpu_fallback(const c10::OperatorHandle& op, torch::jit::Stack* stack)
|
|
6
|
+
{
|
|
7
|
+
TORCHVULKAN_FALLBACK("Silent fallback detected for operation: ", op.schema().operator_name(), ".");
|
|
8
|
+
at::native::cpu_fallback(op, stack);
|
|
9
|
+
}
|
|
10
|
+
|
|
11
|
+
TORCH_LIBRARY_IMPL(_, PrivateUse1, m) {
|
|
12
|
+
m.fallback(torch::CppFunction::makeFromBoxedFunction<&vulkan_cpu_fallback>());
|
|
13
|
+
}
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
#pragma once
|
|
2
|
+
#include <c10/util/Exception.h>
|
|
3
|
+
#include "vulkan/vulkan_context.h"
|
|
4
|
+
|
|
5
|
+
// every CPU fallback, for unimplemented ops and for registered ops with an unsupported dtype or shape;
|
|
6
|
+
// an error under TORCHVULKAN_STRICT=1
|
|
7
|
+
#define TORCHVULKAN_FALLBACK(...) \
|
|
8
|
+
do { \
|
|
9
|
+
if (VulkanContext::Instance().isStrict()) TORCH_CHECK(false, "torchvulkan [NOT IMPLEMENTED]: ", __VA_ARGS__, " Set TORCHVULKAN_STRICT=0 to disable strict mode."); \
|
|
10
|
+
TORCH_WARN_ONCE("torchvulkan [WARNING]: ", __VA_ARGS__, " Set TORCHVULKAN_STRICT=1 to enable strict mode."); \
|
|
11
|
+
} while (0)
|
|
@@ -9,6 +9,7 @@
|
|
|
9
9
|
#include "vulkan/allocator.h"
|
|
10
10
|
#include "api/ops/helpers.h"
|
|
11
11
|
#include "api/ops/internal.h"
|
|
12
|
+
#include "api/fallback.h"
|
|
12
13
|
|
|
13
14
|
namespace {
|
|
14
15
|
|
|
@@ -120,7 +121,7 @@ at::Tensor binary_op_vulkan(
|
|
|
120
121
|
std::optional<at::Tensor> out = dispatch_binary_shader(self, at::Tensor(), other, alpha, operation, at::result_type(self, other));
|
|
121
122
|
if (out.has_value()) return *out;
|
|
122
123
|
|
|
123
|
-
|
|
124
|
+
TORCHVULKAN_FALLBACK("A binary op on ", self.scalar_type(), " can't run on the Vulkan device. Falling back to CPU.");
|
|
124
125
|
return fallback(self.cpu(), other).to(self.device());
|
|
125
126
|
}
|
|
126
127
|
|
|
@@ -145,7 +146,7 @@ at::Tensor binary_op_vulkan(
|
|
|
145
146
|
std::optional<at::Tensor> out = dispatch_binary_shader(lhs, rhs, at::Scalar(0), alpha, operation, at::result_type(self, other));
|
|
146
147
|
if (out.has_value()) return *out;
|
|
147
148
|
|
|
148
|
-
|
|
149
|
+
TORCHVULKAN_FALLBACK("A binary op on ", at::result_type(self, other), " can't run on the Vulkan device. Falling back to CPU.");
|
|
149
150
|
return fallback(lhs.cpu(), rhs.cpu()).to(lhs.device());
|
|
150
151
|
}
|
|
151
152
|
|
|
@@ -3,6 +3,7 @@
|
|
|
3
3
|
#include "vulkan/memory.h"
|
|
4
4
|
#include "vulkan/vulkan_context.h"
|
|
5
5
|
#include "api/ops/helpers.h"
|
|
6
|
+
#include "api/fallback.h"
|
|
6
7
|
|
|
7
8
|
namespace {
|
|
8
9
|
|
|
@@ -116,7 +117,7 @@ at::Tensor& integer_binary_out(
|
|
|
116
117
|
? dispatch_integer_binary_shader(self, at::Tensor(), other.item(), result, operation)
|
|
117
118
|
: dispatch_integer_binary_shader(self, other, at::Scalar(0), result, operation);
|
|
118
119
|
if (!ran) {
|
|
119
|
-
|
|
120
|
+
TORCHVULKAN_FALLBACK(name, " on ", dtype, " can't run on the Vulkan device. Falling back to CPU.");
|
|
120
121
|
result.copy_(cpu_integer_binary(self.cpu(), other.cpu(), operation));
|
|
121
122
|
}
|
|
122
123
|
if (!result.is_same(out)) fill_out(out, result);
|
|
@@ -5,6 +5,7 @@
|
|
|
5
5
|
#include "vulkan/vulkan_context.h"
|
|
6
6
|
#include "vulkan/allocator.h"
|
|
7
7
|
#include "api/ops/helpers.h"
|
|
8
|
+
#include "api/fallback.h"
|
|
8
9
|
|
|
9
10
|
namespace {
|
|
10
11
|
|
|
@@ -132,7 +133,7 @@ at::Tensor where_vulkan(
|
|
|
132
133
|
DeviceContext* context = get_device_context(out);
|
|
133
134
|
int32_t out_dims = static_cast<int32_t>(iter.ndim());
|
|
134
135
|
if (!is_dtype_supported(context, promoted) || out_dims > MAX_DIMS || !fits_32bit_indexing(iter)) {
|
|
135
|
-
|
|
136
|
+
TORCHVULKAN_FALLBACK("This where can't run on the Vulkan device. Falling back to CPU.");
|
|
136
137
|
return at::where(c.cpu(), a.cpu(), b.cpu()).to(device);
|
|
137
138
|
}
|
|
138
139
|
|
|
@@ -172,7 +173,7 @@ at::Tensor& compare_scalar_out_vulkan(const at::Tensor& self, const at::Scalar&
|
|
|
172
173
|
if (!out.sizes().equals(self.sizes())) out.resize_(self.sizes());
|
|
173
174
|
c10::ScalarType promoted = at::result_type(self, other);
|
|
174
175
|
if (!dispatch_compare_shader(self, nullptr, other, out, op, promoted)) {
|
|
175
|
-
|
|
176
|
+
TORCHVULKAN_FALLBACK("This comparison can't run on the Vulkan device. Falling back to CPU.");
|
|
176
177
|
out.copy_(cpu_compare_scalar(self.cpu(), other, op));
|
|
177
178
|
}
|
|
178
179
|
return out;
|
|
@@ -195,7 +196,7 @@ at::Tensor& compare_out_vulkan(const at::Tensor& self, const at::Tensor& other,
|
|
|
195
196
|
at::Tensor lhs = self.to(out.device());
|
|
196
197
|
at::Tensor rhs = other.to(out.device());
|
|
197
198
|
if (!dispatch_compare_shader(lhs, &rhs, at::Scalar(0), out, op, promoted)) {
|
|
198
|
-
|
|
199
|
+
TORCHVULKAN_FALLBACK("This comparison can't run on the Vulkan device. Falling back to CPU.");
|
|
199
200
|
out.copy_(cpu_compare(self.cpu(), other.cpu(), op));
|
|
200
201
|
}
|
|
201
202
|
return out;
|
|
@@ -3,6 +3,7 @@
|
|
|
3
3
|
#include "vulkan/vulkan_context.h"
|
|
4
4
|
#include "vulkan/allocator.h"
|
|
5
5
|
#include "api/ops/helpers.h"
|
|
6
|
+
#include "api/fallback.h"
|
|
6
7
|
|
|
7
8
|
namespace {
|
|
8
9
|
|
|
@@ -17,7 +18,7 @@ at::Tensor& fill_scalar_vulkan(
|
|
|
17
18
|
|
|
18
19
|
int32_t out_dims = static_cast<int32_t>(iter.ndim());
|
|
19
20
|
if (!is_dtype_supported(self, iter.dtype()) || out_dims > MAX_DIMS || !fits_32bit_indexing(iter)) {
|
|
20
|
-
|
|
21
|
+
TORCHVULKAN_FALLBACK("Can't fill a ", iter.dtype(), " tensor of this shape on the Vulkan device. Falling back to CPU.");
|
|
21
22
|
self.copy_(at::empty(self.sizes(), self.options().device(at::kCPU)).fill_(value));
|
|
22
23
|
return self;
|
|
23
24
|
}
|
|
@@ -56,7 +56,7 @@ int64_t outputs_of(const GridSampleProblem& p) { return p.out[0] * p.out[1] * p.
|
|
|
56
56
|
int64_t tap_count(int64_t interpolation, int64_t dims)
|
|
57
57
|
{
|
|
58
58
|
if (interpolation == 1) return 1;
|
|
59
|
-
if (interpolation == 2) return 16;
|
|
59
|
+
if (interpolation == 2) return dims == 3 ? 64 : 16;
|
|
60
60
|
return dims == 3 ? 8 : 4;
|
|
61
61
|
}
|
|
62
62
|
|
|
@@ -138,7 +138,6 @@ std::vector<int64_t> output_shape(const GridSampleProblem& p)
|
|
|
138
138
|
|
|
139
139
|
at::Tensor grid_sampler_forward(const at::Tensor& input, const at::Tensor& grid, int64_t dims, int64_t interpolation, int64_t padding, bool align_corners)
|
|
140
140
|
{
|
|
141
|
-
TORCH_CHECK(dims == 2 || interpolation != 2, "grid_sampler(): bicubic interpolation only supports 4D input");
|
|
142
141
|
GridSampleProblem p = make_problem(input, grid, dims);
|
|
143
142
|
at::Tensor output = at::empty(output_shape(p), p.input.options());
|
|
144
143
|
if (p.channels > 0) dispatch_grid_sample_shader(p, output, at::Tensor(), at::Tensor(), at::Tensor(), at::Tensor(), GridSampleMode::FORWARD, interpolation, padding, align_corners);
|
|
@@ -7,6 +7,7 @@
|
|
|
7
7
|
#include "vulkan/memory.h"
|
|
8
8
|
#include "vulkan/vulkan_context.h"
|
|
9
9
|
#include "api/ops/helpers.h"
|
|
10
|
+
#include "api/fallback.h"
|
|
10
11
|
|
|
11
12
|
namespace {
|
|
12
13
|
|
|
@@ -120,7 +121,7 @@ at::Tensor& gather_out_vulkan(const at::Tensor& self, int64_t dim, const at::Ten
|
|
|
120
121
|
if (src.size(dim) == 0 || !is_dtype_copyable(self, self.scalar_type()) ||
|
|
121
122
|
!dispatch_index_gather_shader(result, restride_for_dim(src, dim, idx.sizes()), {idx}, {src.size(dim)}, {src.stride(dim)}, false))
|
|
122
123
|
{
|
|
123
|
-
|
|
124
|
+
TORCHVULKAN_FALLBACK("This gather can't run on the Vulkan device. Falling back to CPU.");
|
|
124
125
|
out.copy_(at::gather(self.cpu(), dim, index.cpu()));
|
|
125
126
|
}
|
|
126
127
|
return out;
|
|
@@ -132,7 +133,7 @@ at::Tensor index_select_vulkan(const at::Tensor& self, int64_t dim, const at::Te
|
|
|
132
133
|
TORCH_CHECK_INDEX(index.scalar_type() == at::kLong || index.scalar_type() == at::kInt, "index_select(): Expected dtype int32 or int64 for index");
|
|
133
134
|
|
|
134
135
|
if (self.dim() == 0) {
|
|
135
|
-
|
|
136
|
+
TORCHVULKAN_FALLBACK("index_select on a 0-dim tensor runs on the CPU.");
|
|
136
137
|
return at::index_select(self.cpu(), dim, index.cpu()).to(self.device());
|
|
137
138
|
}
|
|
138
139
|
|
|
@@ -150,7 +151,7 @@ at::Tensor index_select_vulkan(const at::Tensor& self, int64_t dim, const at::Te
|
|
|
150
151
|
if (self.size(dim) == 0 || !is_dtype_copyable(self, self.scalar_type()) ||
|
|
151
152
|
!dispatch_index_gather_shader(out, restride_for_dim(self, dim, out_sizes), {idx}, {self.size(dim)}, {self.stride(dim)}, false))
|
|
152
153
|
{
|
|
153
|
-
|
|
154
|
+
TORCHVULKAN_FALLBACK("This index_select can't run on the Vulkan device. Falling back to CPU.");
|
|
154
155
|
return at::index_select(self.cpu(), dim, index.cpu()).to(self.device());
|
|
155
156
|
}
|
|
156
157
|
return out;
|
|
@@ -229,7 +230,7 @@ at::Tensor& index_out_vulkan(const at::Tensor& self, const c10::List<std::option
|
|
|
229
230
|
if (!is_dtype_copyable(self, self.scalar_type()) ||
|
|
230
231
|
!dispatch_index_gather_shader(out, plan.src, plan.indices, plan.index_sizes, plan.index_strides, /*wrap_negative=*/true))
|
|
231
232
|
{
|
|
232
|
-
|
|
233
|
+
TORCHVULKAN_FALLBACK("This index can't run on the Vulkan device. Falling back to CPU.");
|
|
233
234
|
c10::List<std::optional<at::Tensor>> cpu_indices;
|
|
234
235
|
for (const std::optional<at::Tensor>& index : indices) {
|
|
235
236
|
cpu_indices.push_back(index.has_value() && index->defined() ? std::optional<at::Tensor>(index->cpu()) : std::nullopt);
|
|
@@ -276,7 +277,7 @@ at::Tensor embedding_dense_backward_vulkan(
|
|
|
276
277
|
if (!at::isFloatingType(grad_output.scalar_type()) || !is_dtype_supported(device, grad_output.scalar_type()) ||
|
|
277
278
|
!fits_32bit_indexing({flat_grad, grad_weight}))
|
|
278
279
|
{
|
|
279
|
-
|
|
280
|
+
TORCHVULKAN_FALLBACK("This embedding backward can't run on the Vulkan device. Falling back to CPU.");
|
|
280
281
|
return at::embedding_dense_backward(grad_output.cpu(), indices.cpu(), num_weights_, padding_idx_, scale_grad_by_freq).to(grad_output.device());
|
|
281
282
|
}
|
|
282
283
|
|
|
@@ -1,5 +1,4 @@
|
|
|
1
1
|
#include <torch/extension.h>
|
|
2
|
-
#include <torch/version.h>
|
|
3
2
|
#include <ATen/ExpandUtils.h>
|
|
4
3
|
#include <ATen/WrapDimUtils.h>
|
|
5
4
|
#include <algorithm>
|
|
@@ -12,6 +11,7 @@
|
|
|
12
11
|
#include "vulkan/memory.h"
|
|
13
12
|
#include "vulkan/vulkan_context.h"
|
|
14
13
|
#include "api/ops/helpers.h"
|
|
14
|
+
#include "api/fallback.h"
|
|
15
15
|
|
|
16
16
|
namespace {
|
|
17
17
|
|
|
@@ -318,7 +318,7 @@ at::Tensor solve_triangular_left(const at::Tensor& A, const at::Tensor& B, bool
|
|
|
318
318
|
auto [a, b] = broadcast_batches(A, B);
|
|
319
319
|
if (b.numel() == 0 || a.size(-1) == 0) return b.clone();
|
|
320
320
|
if (a.is_complex() || b.is_complex()) {
|
|
321
|
-
|
|
321
|
+
TORCHVULKAN_FALLBACK("A complex triangular solve can't run on the Vulkan device. Falling back to CPU.");
|
|
322
322
|
at::ScalarType type = at::promote_types(a.scalar_type(), b.scalar_type());
|
|
323
323
|
return at::linalg_solve_triangular(a.cpu().to(type), b.cpu().to(type), upper, /*left=*/true, unit).to(B.device());
|
|
324
324
|
}
|
|
@@ -354,7 +354,7 @@ std::tuple<at::Tensor, at::Tensor, at::Tensor> lu_factor_impl(const at::Tensor&
|
|
|
354
354
|
{
|
|
355
355
|
TORCH_CHECK(A.dim() >= 2, name, ": Expected tensor with 2 or more dimensions. Got size: ", A.sizes(), " instead");
|
|
356
356
|
if (A.is_complex()) {
|
|
357
|
-
|
|
357
|
+
TORCHVULKAN_FALLBACK("A complex LU factorization can't run on the Vulkan device. Falling back to CPU.");
|
|
358
358
|
auto [lu, pivots, info] = at::linalg_lu_factor_ex(A.cpu());
|
|
359
359
|
return std::make_tuple(lu.to(A.device()), pivots.to(A.device()), info.to(A.device()));
|
|
360
360
|
}
|
|
@@ -492,14 +492,8 @@ std::tuple<at::Tensor&, at::Tensor&, at::Tensor&, at::Tensor&> linalg_solve_ex_o
|
|
|
492
492
|
check_square(A, "torch.linalg.solve");
|
|
493
493
|
bool vector_case = is_vector_rhs(A, B);
|
|
494
494
|
at::Tensor b = vector_case ? B.unsqueeze(left ? -1 : -2) : B;
|
|
495
|
-
|
|
496
|
-
|
|
497
|
-
bool use_A_T = A.is_contiguous() && !A.is_complex();
|
|
498
|
-
#else
|
|
499
|
-
bool use_A_T = false;
|
|
500
|
-
#endif
|
|
501
|
-
auto [lu, piv, inf] = lu_factor_impl(use_A_T ? A.mT() : A, "torch.linalg.solve");
|
|
502
|
-
at::Tensor x = lu_solve_impl(lu, piv, b, left, /*adjoint=*/use_A_T);
|
|
495
|
+
auto [lu, piv, inf] = lu_factor_impl(A, "torch.linalg.solve");
|
|
496
|
+
at::Tensor x = lu_solve_impl(lu, piv, b, left, /*adjoint=*/false);
|
|
503
497
|
if (vector_case) x = x.squeeze(left ? -1 : -2);
|
|
504
498
|
fill_out(result, x);
|
|
505
499
|
fill_out(LU, lu);
|
|
@@ -6,6 +6,7 @@
|
|
|
6
6
|
#include "vulkan/allocator.h"
|
|
7
7
|
#include "api/ops/helpers.h"
|
|
8
8
|
#include "api/ops/internal.h"
|
|
9
|
+
#include "api/fallback.h"
|
|
9
10
|
|
|
10
11
|
namespace {
|
|
11
12
|
|
|
@@ -171,7 +172,7 @@ at::Tensor dispatch_matmul_simd_shader(
|
|
|
171
172
|
DeviceContext* device = get_device_context(self_);
|
|
172
173
|
c10::ScalarType result_dtype = at::result_type(shape.self, shape.other);
|
|
173
174
|
if (!is_dtype_supported(device, result_dtype) || WG * WG > device->properties.limits.maxComputeWorkGroupInvocations) {
|
|
174
|
-
|
|
175
|
+
TORCHVULKAN_FALLBACK("A ", result_dtype, " matmul can't run on the Vulkan device. Falling back to CPU.");
|
|
175
176
|
return cpu_fallback();
|
|
176
177
|
}
|
|
177
178
|
|
|
@@ -190,7 +191,7 @@ at::Tensor dispatch_matmul_simd_shader(
|
|
|
190
191
|
uint32_t tile_m = c10::elementSize(result_dtype) < 8 ? 128 : 64;
|
|
191
192
|
if (tile_m == 128 && shared_bytes(128) > max_shared_memory) tile_m = 64;
|
|
192
193
|
if (shared_bytes(tile_m) > max_shared_memory) {
|
|
193
|
-
|
|
194
|
+
TORCHVULKAN_FALLBACK("a ", result_dtype, " matmul tile needs ", shared_bytes(tile_m),
|
|
194
195
|
" bytes of shared memory but the device provides ", max_shared_memory, ". Falling back to CPU.");
|
|
195
196
|
return cpu_fallback();
|
|
196
197
|
}
|
|
@@ -204,7 +205,7 @@ at::Tensor dispatch_matmul_simd_shader(
|
|
|
204
205
|
return finish_matmul_output(apply_matmul_activation(result, activation), shape);
|
|
205
206
|
}
|
|
206
207
|
if (!fits_matmul_indexing(ops, out)) {
|
|
207
|
-
|
|
208
|
+
TORCHVULKAN_FALLBACK("A matmul is too large for 32-bit indexing. Falling back to CPU.");
|
|
208
209
|
return cpu_fallback();
|
|
209
210
|
}
|
|
210
211
|
|
|
@@ -299,7 +300,7 @@ at::Tensor dispatch_matmul_coop_shader(
|
|
|
299
300
|
DeviceContext* device = get_device_context(self_);
|
|
300
301
|
c10::ScalarType promoted_type = at::result_type(self_, other_);
|
|
301
302
|
if (!is_dtype_supported(device, promoted_type)) {
|
|
302
|
-
|
|
303
|
+
TORCHVULKAN_FALLBACK("Vulkan device does not support ", promoted_type, ". Falling back to CPU.");
|
|
303
304
|
return cpu_fallback();
|
|
304
305
|
}
|
|
305
306
|
|
|
@@ -6,6 +6,7 @@
|
|
|
6
6
|
#include "vulkan/allocator.h"
|
|
7
7
|
#include "api/ops/helpers.h"
|
|
8
8
|
#include "api/ops/internal.h"
|
|
9
|
+
#include "api/fallback.h"
|
|
9
10
|
|
|
10
11
|
namespace {
|
|
11
12
|
|
|
@@ -84,7 +85,7 @@ std::tuple<at::Tensor, at::Tensor> nll_loss_forward_vulkan(
|
|
|
84
85
|
TORCH_CHECK(self.dim() == 1 || self.dim() == 2, "torchvulkan [ERROR]: nll_loss only supports 1D (unbatched) or 2D input.");
|
|
85
86
|
|
|
86
87
|
if (!is_dtype_supported(self, self.scalar_type()) || self.numel() > std::numeric_limits<int32_t>::max()) {
|
|
87
|
-
|
|
88
|
+
TORCHVULKAN_FALLBACK("Vulkan device does not support ", self.scalar_type(), ". Falling back to CPU.");
|
|
88
89
|
c10::optional<at::Tensor> cpu_weight = weight.has_value() && weight->defined() ? c10::optional<at::Tensor>(weight->cpu()) : c10::nullopt;
|
|
89
90
|
auto [cpu_output, cpu_total_weight] = at::nll_loss_forward(self.cpu(), target.cpu(), cpu_weight, reduction, ignore_index);
|
|
90
91
|
return {cpu_output.to(self.device()), cpu_total_weight.to(self.device())};
|
|
@@ -124,7 +125,7 @@ at::Tensor nll_loss_backward_vulkan(
|
|
|
124
125
|
TORCH_CHECK(self.dim() == 1 || self.dim() == 2, "torchvulkan [ERROR]: nll_loss only supports 1D (unbatched) or 2D input.");
|
|
125
126
|
|
|
126
127
|
if (!is_dtype_supported(self, self.scalar_type()) || self.numel() > std::numeric_limits<int32_t>::max()) {
|
|
127
|
-
|
|
128
|
+
TORCHVULKAN_FALLBACK("Vulkan device does not support ", self.scalar_type(), ". Falling back to CPU.");
|
|
128
129
|
c10::optional<at::Tensor> cpu_weight = weight.has_value() && weight->defined() ? c10::optional<at::Tensor>(weight->cpu()) : c10::nullopt;
|
|
129
130
|
at::Tensor cpu_grad_input = at::nll_loss_backward(
|
|
130
131
|
grad_output.cpu(), self.cpu(), target.cpu(), cpu_weight, reduction, ignore_index, total_weight.cpu());
|
|
@@ -9,6 +9,7 @@
|
|
|
9
9
|
#include "vulkan/allocator.h"
|
|
10
10
|
#include "api/ops/helpers.h"
|
|
11
11
|
#include "api/ops/internal.h"
|
|
12
|
+
#include "api/fallback.h"
|
|
12
13
|
|
|
13
14
|
namespace {
|
|
14
15
|
|
|
@@ -148,7 +149,7 @@ at::Tensor reduce_dims_vulkan(
|
|
|
148
149
|
bool has_subgroup)
|
|
149
150
|
{
|
|
150
151
|
if (!is_dtype_supported(self, self.scalar_type()) || self.dim() > MAX_DIMS || !fits_32bit_indexing({self})) {
|
|
151
|
-
|
|
152
|
+
TORCHVULKAN_FALLBACK("A ", self.scalar_type(), " reduction of this shape can't run on the Vulkan device. Falling back to CPU.");
|
|
152
153
|
return cpu_reduce_fallback(self.cpu(), dims, keepdim, operation).to(self.device());
|
|
153
154
|
}
|
|
154
155
|
|
|
@@ -297,6 +298,7 @@ at::Tensor argreduce(
|
|
|
297
298
|
}
|
|
298
299
|
|
|
299
300
|
if (!is_dtype_supported(x, x.scalar_type()) || x.dim() == 0 || x.dim() > MAX_DIMS || x.numel() == 0 || !fits_32bit_indexing({x})) {
|
|
301
|
+
TORCHVULKAN_FALLBACK("This ", op == ArgOp::ARGMAX ? "argmax" : "argmin", " on ", x.scalar_type(), " can't run on the Vulkan device. Falling back to CPU.");
|
|
300
302
|
at::Tensor cpu = op == ArgOp::ARGMAX ? at::argmax(self.cpu(), dim, keepdim) : at::argmin(self.cpu(), dim, keepdim);
|
|
301
303
|
return cpu.to(self.device());
|
|
302
304
|
}
|
|
@@ -6,6 +6,7 @@
|
|
|
6
6
|
#include "vulkan/vulkan_context.h"
|
|
7
7
|
#include "vulkan/allocator.h"
|
|
8
8
|
#include "api/ops/helpers.h"
|
|
9
|
+
#include "api/fallback.h"
|
|
9
10
|
|
|
10
11
|
namespace {
|
|
11
12
|
|
|
@@ -25,6 +26,7 @@ at::Tensor cumscan(
|
|
|
25
26
|
|
|
26
27
|
int64_t d = c10::maybe_wrap_dim(dim, x.dim());
|
|
27
28
|
if (!is_dtype_supported(x, compute) || x.dim() > MAX_DIMS || !fits_32bit_indexing({x})) {
|
|
29
|
+
TORCHVULKAN_FALLBACK("A ", compute, " cumulative op of this shape can't run on the Vulkan device. Falling back to CPU.");
|
|
28
30
|
at::Tensor cpu = op == ScanOp::SUM ? at::cumsum(x.cpu(), d) : op == ScanOp::PROD ? at::cumprod(x.cpu(), d) : at::logcumsumexp(x.cpu(), d);
|
|
29
31
|
return cpu.to(out_dtype).to(self.device());
|
|
30
32
|
}
|
|
@@ -124,6 +126,7 @@ void cumscanarg(
|
|
|
124
126
|
}
|
|
125
127
|
int64_t d = c10::maybe_wrap_dim(dim, self.dim());
|
|
126
128
|
if (!is_dtype_supported(self, self.scalar_type()) || self.dim() > MAX_DIMS || !fits_32bit_indexing({self})) {
|
|
129
|
+
TORCHVULKAN_FALLBACK("A ", self.scalar_type(), op == CumArgOp::MAX ? " cummax" : " cummin", " of this shape can't run on the Vulkan device. Falling back to CPU.");
|
|
127
130
|
auto r = op == CumArgOp::MAX ? at::cummax(self.cpu(), dim) : at::cummin(self.cpu(), dim);
|
|
128
131
|
values.copy_(std::get<0>(r));
|
|
129
132
|
indices.copy_(std::get<1>(r));
|
|
@@ -7,6 +7,7 @@
|
|
|
7
7
|
#include "vulkan/memory.h"
|
|
8
8
|
#include "vulkan/vulkan_context.h"
|
|
9
9
|
#include "api/ops/helpers.h"
|
|
10
|
+
#include "api/fallback.h"
|
|
10
11
|
|
|
11
12
|
namespace {
|
|
12
13
|
|
|
@@ -181,7 +182,7 @@ at::Tensor scatter_along_dim(const at::Tensor& self, int64_t dim, const at::Tens
|
|
|
181
182
|
}
|
|
182
183
|
|
|
183
184
|
if (!can_scatter(s, idx.numel())) {
|
|
184
|
-
|
|
185
|
+
TORCHVULKAN_FALLBACK("This scatter can't run on the Vulkan device. Falling back to CPU.");
|
|
185
186
|
at::Tensor cpu = s.cpu().clone();
|
|
186
187
|
if (mode == ScatterMode::LAST) cpu.scatter_(dim, idx.cpu(), values.cpu());
|
|
187
188
|
else cpu.scatter_reduce_(dim, idx.cpu(), values.cpu(), mode == ScatterMode::SUM ? "sum" : mode == ScatterMode::PROD ? "prod" : mode == ScatterMode::MEAN ? "mean" : mode == ScatterMode::AMAX ? "amax" : "amin", include_self);
|