nvmath-python 1.0.0__cp314-cp314t-win_amd64.whl
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.
- nvmath/__init__.pxd +0 -0
- nvmath/__init__.py +45 -0
- nvmath/_internal/__init__.py +0 -0
- nvmath/_internal/attribute_ifc_factory.py +330 -0
- nvmath/_internal/layout.py +70 -0
- nvmath/_internal/templates.py +130 -0
- nvmath/_internal/threadsafe.py +106 -0
- nvmath/_internal/utils.py +43 -0
- nvmath/_internal/workspace.py +490 -0
- nvmath/_utils.py +147 -0
- nvmath/bindings/__init__.py +60 -0
- nvmath/bindings/_internal/__init__.pxd +0 -0
- nvmath/bindings/_internal/__init__.py +0 -0
- nvmath/bindings/_internal/common_types.pxd +31 -0
- nvmath/bindings/_internal/cublas.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cublas.pxd +530 -0
- nvmath/bindings/_internal/cublasLt.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cublasLt.pxd +59 -0
- nvmath/bindings/_internal/cublasMp.pxd +52 -0
- nvmath/bindings/_internal/cudss.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cudss.pxd +54 -0
- nvmath/bindings/_internal/cufft.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cufft.pxd +70 -0
- nvmath/bindings/_internal/cufftMp.pxd +77 -0
- nvmath/bindings/_internal/curand.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/curand.pxd +42 -0
- nvmath/bindings/_internal/cusolver.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cusolver.pxd +15 -0
- nvmath/bindings/_internal/cusolverDn.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cusolverDn.pxd +406 -0
- nvmath/bindings/_internal/cusolverMp.pxd +71 -0
- nvmath/bindings/_internal/cusolverSp.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cusolverSp.pxd +75 -0
- nvmath/bindings/_internal/cusparse.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cusparse.pxd +471 -0
- nvmath/bindings/_internal/cusparseLt.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cusparseLt.pxd +48 -0
- nvmath/bindings/_internal/cutensor.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/cutensor.pxd +58 -0
- nvmath/bindings/_internal/mathdx.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/mathdx.pxd +116 -0
- nvmath/bindings/_internal/nvshmem.pxd +29 -0
- nvmath/bindings/_internal/utils.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/_internal/utils.pxd +174 -0
- nvmath/bindings/_internal/utils.pyi +10 -0
- nvmath/bindings/cublas.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cublas.pxd +558 -0
- nvmath/bindings/cublas.pyi +812 -0
- nvmath/bindings/cublasLt.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cublasLt.pxd +109 -0
- nvmath/bindings/cublasLt.pyi +1461 -0
- nvmath/bindings/cublasMp.pxd +85 -0
- nvmath/bindings/cublasMp.pyi +267 -0
- nvmath/bindings/cudss.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cudss.pxd +98 -0
- nvmath/bindings/cudss.pyi +443 -0
- nvmath/bindings/cufft.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cufft.pxd +118 -0
- nvmath/bindings/cufft.pyi +301 -0
- nvmath/bindings/cufftMp.pxd +124 -0
- nvmath/bindings/cufftMp.pyi +326 -0
- nvmath/bindings/curand.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/curand.pxd +71 -0
- nvmath/bindings/curand.pyi +189 -0
- nvmath/bindings/cusolver.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cusolver.pxd +62 -0
- nvmath/bindings/cusolver.pyi +320 -0
- nvmath/bindings/cusolverDn.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cusolverDn.pxd +430 -0
- nvmath/bindings/cusolverDn.pyi +422 -0
- nvmath/bindings/cusolverMp.pxd +98 -0
- nvmath/bindings/cusolverMp.pyi +114 -0
- nvmath/bindings/cusolverSp.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cusolverSp.pxd +95 -0
- nvmath/bindings/cusolverSp.pyi +70 -0
- nvmath/bindings/cusparse.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cusparse.pxd +546 -0
- nvmath/bindings/cusparse.pyi +1017 -0
- nvmath/bindings/cusparseLt.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cusparseLt.pxd +99 -0
- nvmath/bindings/cusparseLt.pyi +252 -0
- nvmath/bindings/cutensor.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cutensor.pxd +98 -0
- nvmath/bindings/cutensor.pyi +324 -0
- nvmath/bindings/cycublas.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycublas.pxd +664 -0
- nvmath/bindings/cycublasLt.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycublasLt.pxd +1045 -0
- nvmath/bindings/cycublasMp.pxd +171 -0
- nvmath/bindings/cycudss.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycudss.pxd +277 -0
- nvmath/bindings/cycufft.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycufft.pxd +333 -0
- nvmath/bindings/cycufftMp.pxd +342 -0
- nvmath/bindings/cycurand.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycurand.pxd +141 -0
- nvmath/bindings/cycusolver.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycusolver.pxd +137 -0
- nvmath/bindings/cycusolverDn.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycusolverDn.pxd +443 -0
- nvmath/bindings/cycusolverMp.pxd +107 -0
- nvmath/bindings/cycusolverSp.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycusolverSp.pxd +93 -0
- nvmath/bindings/cycusparse.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycusparse.pxd +679 -0
- nvmath/bindings/cycusparseLt.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycusparseLt.pxd +135 -0
- nvmath/bindings/cycutensor.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cycutensor.pxd +189 -0
- nvmath/bindings/cymathdx.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/cymathdx.pxd +552 -0
- nvmath/bindings/cynvshmem.pxd +118 -0
- nvmath/bindings/mathdx.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/mathdx.pxd +182 -0
- nvmath/bindings/mathdx.pyi +1562 -0
- nvmath/bindings/nvpl/__init__.pxd +0 -0
- nvmath/bindings/nvpl/__init__.py +13 -0
- nvmath/bindings/nvpl/_internal/__init__.pxd +0 -0
- nvmath/bindings/nvpl/_internal/__init__.py +0 -0
- nvmath/bindings/nvpl/_internal/blas.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/nvpl/_internal/blas.pxd +237 -0
- nvmath/bindings/nvpl/_internal/fft.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/nvpl/_internal/fft.pxd +36 -0
- nvmath/bindings/nvpl/blas.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/nvpl/blas.pxd +131 -0
- nvmath/bindings/nvpl/blas.pyi +168 -0
- nvmath/bindings/nvpl/cyblas.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/nvpl/cyblas.pxd +280 -0
- nvmath/bindings/nvpl/cyfft.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/nvpl/cyfft.pxd +93 -0
- nvmath/bindings/nvpl/fft.cp314t-win_amd64.pyd +0 -0
- nvmath/bindings/nvpl/fft.pxd +100 -0
- nvmath/bindings/nvpl/fft.pyi +168 -0
- nvmath/bindings/nvshmem.pxd +54 -0
- nvmath/bindings/nvshmem.pyi +191 -0
- nvmath/device/__init__.py +38 -0
- nvmath/device/_deprecated.py +33 -0
- nvmath/device/common.py +315 -0
- nvmath/device/common_backend.py +131 -0
- nvmath/device/common_cuda.py +201 -0
- nvmath/device/common_numba.py +300 -0
- nvmath/device/common_numba_cuda_mlir.py +202 -0
- nvmath/device/common_opaque_tensor.py +201 -0
- nvmath/device/cublasdx.py +1606 -0
- nvmath/device/cublasdx_backend.py +860 -0
- nvmath/device/cublasdx_numba.py +1534 -0
- nvmath/device/cublasdx_numba_cuda_mlir.py +208 -0
- nvmath/device/cufftdx.py +373 -0
- nvmath/device/cufftdx_backend.py +220 -0
- nvmath/device/cufftdx_numba.py +140 -0
- nvmath/device/cufftdx_numba_cuda_mlir.py +79 -0
- nvmath/device/curand_kernel.py +9147 -0
- nvmath/device/cusolverdx.py +2708 -0
- nvmath/device/cusolverdx_backend.py +440 -0
- nvmath/device/cusolverdx_numba.py +567 -0
- nvmath/device/cusolverdx_numba_cuda_mlir.py +604 -0
- nvmath/device/cusolverdx_overload_backend.py +1029 -0
- nvmath/device/llvm_array.py +29 -0
- nvmath/device/random.py +441 -0
- nvmath/device/random_helpers.py +23 -0
- nvmath/device/random_states.py +187 -0
- nvmath/device/types.py +138 -0
- nvmath/device/vector_types_numba.py +259 -0
- nvmath/distributed/__init__.py +200 -0
- nvmath/distributed/_internal/__init__.py +0 -0
- nvmath/distributed/_internal/nccl.py +86 -0
- nvmath/distributed/_internal/nvshmem.py +307 -0
- nvmath/distributed/_internal/symmetric_memory.py +35 -0
- nvmath/distributed/_internal/tensor_ifc.py +70 -0
- nvmath/distributed/_internal/tensor_ifc_cupy.py +68 -0
- nvmath/distributed/_internal/tensor_ifc_host_device.py +172 -0
- nvmath/distributed/_internal/tensor_ifc_numpy.py +46 -0
- nvmath/distributed/_internal/tensor_ifc_torch.py +162 -0
- nvmath/distributed/_internal/tensor_wrapper.py +81 -0
- nvmath/distributed/_utils.py +167 -0
- nvmath/distributed/distribution/__init__.py +30 -0
- nvmath/distributed/distribution/_configuration.py +39 -0
- nvmath/distributed/distribution/distributions.py +1024 -0
- nvmath/distributed/distribution/redistribute.py +1284 -0
- nvmath/distributed/fft/__init__.py +7 -0
- nvmath/distributed/fft/_configuration.py +82 -0
- nvmath/distributed/fft/fft.py +2742 -0
- nvmath/distributed/linalg/__init__.py +22 -0
- nvmath/distributed/linalg/_internal/__init__.py +3 -0
- nvmath/distributed/linalg/_internal/epilog_protocol.py +586 -0
- nvmath/distributed/linalg/_internal/matmul_desc_ifc.py +28 -0
- nvmath/distributed/linalg/advanced/__init__.py +8 -0
- nvmath/distributed/linalg/advanced/_configuration.py +171 -0
- nvmath/distributed/linalg/advanced/matmulmod.py +3573 -0
- nvmath/distributed/linalg/generic/__init__.py +8 -0
- nvmath/distributed/linalg/generic/_caching.py +66 -0
- nvmath/distributed/linalg/generic/_configuration.py +61 -0
- nvmath/distributed/linalg/generic/_factorization.py +172 -0
- nvmath/distributed/linalg/generic/_initialization.py +966 -0
- nvmath/distributed/linalg/generic/_problem_spec.py +511 -0
- nvmath/distributed/linalg/generic/solvermod.py +1368 -0
- nvmath/distributed/process_group.py +408 -0
- nvmath/fft/__init__.py +7 -0
- nvmath/fft/_configuration.py +189 -0
- nvmath/fft/_exec_utils.py +82 -0
- nvmath/fft/_helpers.py +237 -0
- nvmath/fft/fft.py +3122 -0
- nvmath/internal/__init__.pxd +3 -0
- nvmath/internal/__init__.py +10 -0
- nvmath/internal/_bindings.cp314t-win_amd64.pyd +0 -0
- nvmath/internal/_bindings.pxd +18 -0
- nvmath/internal/_device_utils.py +45 -0
- nvmath/internal/_layout/__init__.pxd +3 -0
- nvmath/internal/_layout/__init__.py +7 -0
- nvmath/internal/_layout/_layout.cp314t-win_amd64.pyd +0 -0
- nvmath/internal/_layout/_layout.pxd +1303 -0
- nvmath/internal/_layout/_layout.pyi +1145 -0
- nvmath/internal/enum_utils.py +142 -0
- nvmath/internal/formatters.py +87 -0
- nvmath/internal/mem_limit.py +51 -0
- nvmath/internal/memory.cp314t-win_amd64.pyd +0 -0
- nvmath/internal/memory.pxd +13 -0
- nvmath/internal/memory.pyi +50 -0
- nvmath/internal/ndbuffer/__init__.pxd +3 -0
- nvmath/internal/ndbuffer/__init__.py +9 -0
- nvmath/internal/ndbuffer/_copy_kernel.cp314t-win_amd64.pyd +0 -0
- nvmath/internal/ndbuffer/_copy_kernel.pxd +10 -0
- nvmath/internal/ndbuffer/_jit.cp314t-win_amd64.pyd +0 -0
- nvmath/internal/ndbuffer/_jit.pxd +7 -0
- nvmath/internal/ndbuffer/_ndbuffer.cp314t-win_amd64.pyd +0 -0
- nvmath/internal/ndbuffer/_ndbuffer.pxd +40 -0
- nvmath/internal/ndbuffer/_ndbuffer.pyi +463 -0
- nvmath/internal/ndbuffer/copy_kernel/args.h +34 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/array_view.h +52 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/elementwise.h +68 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/grid_indexer.h +69 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/transposed.h +242 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/type_utils.h +39 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/utils.h +132 -0
- nvmath/internal/ndbuffer/copy_kernel/copy_kernel_impl/vec.h +159 -0
- nvmath/internal/ndbuffer/copy_kernel/elementwise.h +53 -0
- nvmath/internal/ndbuffer/copy_kernel/transposed.h +58 -0
- nvmath/internal/package_ifc.py +168 -0
- nvmath/internal/package_ifc_cuda.py +57 -0
- nvmath/internal/package_ifc_cupy.py +67 -0
- nvmath/internal/package_ifc_torch.py +69 -0
- nvmath/internal/package_wrapper.py +14 -0
- nvmath/internal/tensor_ifc.py +179 -0
- nvmath/internal/tensor_ifc_cupy.py +234 -0
- nvmath/internal/tensor_ifc_ndbuffer.py +147 -0
- nvmath/internal/tensor_ifc_numpy.py +184 -0
- nvmath/internal/tensor_ifc_torch.py +178 -0
- nvmath/internal/tensor_wrapper.py +160 -0
- nvmath/internal/typemaps.py +113 -0
- nvmath/internal/utils.py +805 -0
- nvmath/linalg/__init__.py +56 -0
- nvmath/linalg/_internal/__init__.py +3 -0
- nvmath/linalg/_internal/algo_cap_ifc.py +82 -0
- nvmath/linalg/_internal/algo_config_ifc.py +43 -0
- nvmath/linalg/_internal/batch.py +234 -0
- nvmath/linalg/_internal/enum_to_tuples.py +64 -0
- nvmath/linalg/_internal/epilog_protocol.py +766 -0
- nvmath/linalg/_internal/layout.py +624 -0
- nvmath/linalg/_internal/matmul_desc_ifc.py +28 -0
- nvmath/linalg/_internal/matmul_pref_ifc.py +27 -0
- nvmath/linalg/_internal/matrix_layout_ifc.py +26 -0
- nvmath/linalg/_internal/solver_utils.py +432 -0
- nvmath/linalg/_internal/typemaps.py +144 -0
- nvmath/linalg/_internal/utils.py +157 -0
- nvmath/linalg/advanced/__init__.py +8 -0
- nvmath/linalg/advanced/_algorithmmod.py +170 -0
- nvmath/linalg/advanced/_configuration.py +351 -0
- nvmath/linalg/advanced/helpers/__init__.py +5 -0
- nvmath/linalg/advanced/helpers/matmul.py +1316 -0
- nvmath/linalg/advanced/matmulmod.py +3734 -0
- nvmath/linalg/generic/__init__.py +53 -0
- nvmath/linalg/generic/_configuration/__init__.py +39 -0
- nvmath/linalg/generic/_configuration/layout.py +263 -0
- nvmath/linalg/generic/_configuration/match.py +734 -0
- nvmath/linalg/generic/_configuration/qualifiers.py +493 -0
- nvmath/linalg/generic/_configuration/solver_configuration.py +59 -0
- nvmath/linalg/generic/_configuration/wrap.py +217 -0
- nvmath/linalg/generic/_dtype.py +15 -0
- nvmath/linalg/generic/matmulmod.py +2094 -0
- nvmath/linalg/generic/solvermod.py +1301 -0
- nvmath/memory.py +279 -0
- nvmath/sparse/__init__.py +38 -0
- nvmath/sparse/_internal/__init__.py +21 -0
- nvmath/sparse/_internal/common_utils.py +147 -0
- nvmath/sparse/_internal/cudss_config_ifc.py +702 -0
- nvmath/sparse/_internal/cudss_data_ifc.py +399 -0
- nvmath/sparse/_internal/cudss_utils.py +506 -0
- nvmath/sparse/_internal/cusparse_utils.py +382 -0
- nvmath/sparse/_internal/sparse_bsc_ifc.py +303 -0
- nvmath/sparse/_internal/sparse_bsr_ifc.py +305 -0
- nvmath/sparse/_internal/sparse_coo_ifc.py +256 -0
- nvmath/sparse/_internal/sparse_csc_ifc.py +268 -0
- nvmath/sparse/_internal/sparse_csr_ifc.py +288 -0
- nvmath/sparse/_internal/sparse_dia_ifc.py +242 -0
- nvmath/sparse/_internal/sparse_format_helpers.py +601 -0
- nvmath/sparse/_internal/sparse_tensor_ifc.py +133 -0
- nvmath/sparse/_internal/sparse_ust_ifc.py +141 -0
- nvmath/sparse/_internal/utils.py +56 -0
- nvmath/sparse/advanced/__init__.py +7 -0
- nvmath/sparse/advanced/_configuration.py +227 -0
- nvmath/sparse/advanced/direct_solver.py +2069 -0
- nvmath/sparse/generic/__init__.py +7 -0
- nvmath/sparse/generic/_configuration.py +129 -0
- nvmath/sparse/generic/_helpers.py +137 -0
- nvmath/sparse/generic/_thunks.py +21 -0
- nvmath/sparse/generic/matmulmod.py +2353 -0
- nvmath/sparse/ust/__init__.py +7 -0
- nvmath/sparse/ust/_converters.py +422 -0
- nvmath/sparse/ust/_cpp.py +28 -0
- nvmath/sparse/ust/_drawer.py +565 -0
- nvmath/sparse/ust/_emitter.py +1033 -0
- nvmath/sparse/ust/_jit.py +188 -0
- nvmath/sparse/ust/_kernel.py +282 -0
- nvmath/sparse/ust/_utils.py +149 -0
- nvmath/sparse/ust/interfaces/__init__.py +0 -0
- nvmath/sparse/ust/interfaces/torch_interface.py +476 -0
- nvmath/sparse/ust/tensor.py +1016 -0
- nvmath/sparse/ust/tensor_format.py +957 -0
- nvmath/tensor/__init__.py +6 -0
- nvmath/tensor/_configuration.py +120 -0
- nvmath/tensor/_internal/__init__.py +3 -0
- nvmath/tensor/_internal/cutensor_config_ifc.py +279 -0
- nvmath/tensor/_internal/cutensor_utils.py +230 -0
- nvmath/tensor/_internal/data.py +43 -0
- nvmath/tensor/_internal/einsum_parser.py +444 -0
- nvmath/tensor/_internal/typemaps.py +96 -0
- nvmath/tensor/contract.py +1900 -0
- nvmath_python-1.0.0.dist-info/METADATA +134 -0
- nvmath_python-1.0.0.dist-info/RECORD +332 -0
- nvmath_python-1.0.0.dist-info/WHEEL +5 -0
- nvmath_python-1.0.0.dist-info/licenses/LICENSE +177 -0
- nvmath_python-1.0.0.dist-info/top_level.txt +2 -0
|
@@ -0,0 +1,53 @@
|
|
|
1
|
+
# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
|
|
2
|
+
#
|
|
3
|
+
# SPDX-License-Identifier: Apache-2.0
|
|
4
|
+
|
|
5
|
+
from ._configuration import (
|
|
6
|
+
DiagonalMatrixQualifier,
|
|
7
|
+
GeneralMatrixQualifier,
|
|
8
|
+
HermitianMatrixQualifier,
|
|
9
|
+
MatmulOptions,
|
|
10
|
+
MatrixQualifier,
|
|
11
|
+
SymmetricMatrixQualifier,
|
|
12
|
+
TriangularMatrixQualifier,
|
|
13
|
+
matrix_qualifiers_dtype,
|
|
14
|
+
)
|
|
15
|
+
from .matmulmod import (
|
|
16
|
+
DiagType,
|
|
17
|
+
ExecutionCPU,
|
|
18
|
+
ExecutionCUDA,
|
|
19
|
+
FillMode,
|
|
20
|
+
InvalidMatmulState,
|
|
21
|
+
Matmul,
|
|
22
|
+
SideMode,
|
|
23
|
+
matmul,
|
|
24
|
+
)
|
|
25
|
+
from .solvermod import (
|
|
26
|
+
DirectSolver,
|
|
27
|
+
DirectSolverOptions,
|
|
28
|
+
InvalidDirectSolverState,
|
|
29
|
+
direct_solver,
|
|
30
|
+
)
|
|
31
|
+
|
|
32
|
+
__all__ = (
|
|
33
|
+
"DiagonalMatrixQualifier",
|
|
34
|
+
"ExecutionCPU",
|
|
35
|
+
"ExecutionCUDA",
|
|
36
|
+
"GeneralMatrixQualifier",
|
|
37
|
+
"HermitianMatrixQualifier",
|
|
38
|
+
"InvalidMatmulState",
|
|
39
|
+
"matmul",
|
|
40
|
+
"Matmul",
|
|
41
|
+
"MatmulOptions",
|
|
42
|
+
"MatrixQualifier",
|
|
43
|
+
"matrix_qualifiers_dtype",
|
|
44
|
+
"SymmetricMatrixQualifier",
|
|
45
|
+
"TriangularMatrixQualifier",
|
|
46
|
+
"SideMode",
|
|
47
|
+
"FillMode",
|
|
48
|
+
"DiagType",
|
|
49
|
+
"DirectSolverOptions",
|
|
50
|
+
"DirectSolver",
|
|
51
|
+
"direct_solver",
|
|
52
|
+
"InvalidDirectSolverState",
|
|
53
|
+
)
|
|
@@ -0,0 +1,39 @@
|
|
|
1
|
+
# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
|
|
2
|
+
#
|
|
3
|
+
# SPDX-License-Identifier: Apache-2.0
|
|
4
|
+
|
|
5
|
+
from .layout import (
|
|
6
|
+
CACHED_LAYOUT_CHECKERS,
|
|
7
|
+
mm_layout_checker_getter,
|
|
8
|
+
)
|
|
9
|
+
from .match import (
|
|
10
|
+
select_blas_group_mm_function,
|
|
11
|
+
select_blas_mm_function,
|
|
12
|
+
)
|
|
13
|
+
from .qualifiers import (
|
|
14
|
+
DiagonalMatrixQualifier,
|
|
15
|
+
GeneralMatrixQualifier,
|
|
16
|
+
HermitianMatrixQualifier,
|
|
17
|
+
MatmulOptions,
|
|
18
|
+
MatrixQualifier,
|
|
19
|
+
SymmetricMatrixQualifier,
|
|
20
|
+
TriangularMatrixQualifier,
|
|
21
|
+
matrix_qualifiers_dtype,
|
|
22
|
+
vector_to_square,
|
|
23
|
+
)
|
|
24
|
+
|
|
25
|
+
__all__ = [
|
|
26
|
+
"DiagonalMatrixQualifier",
|
|
27
|
+
"GeneralMatrixQualifier",
|
|
28
|
+
"HermitianMatrixQualifier",
|
|
29
|
+
"MatmulOptions",
|
|
30
|
+
"MatrixQualifier",
|
|
31
|
+
"mm_layout_checker_getter",
|
|
32
|
+
"matrix_qualifiers_dtype",
|
|
33
|
+
"SymmetricMatrixQualifier",
|
|
34
|
+
"TriangularMatrixQualifier",
|
|
35
|
+
"vector_to_square",
|
|
36
|
+
"select_blas_mm_function",
|
|
37
|
+
"select_blas_group_mm_function",
|
|
38
|
+
"CACHED_LAYOUT_CHECKERS",
|
|
39
|
+
]
|
|
@@ -0,0 +1,263 @@
|
|
|
1
|
+
# Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
|
|
2
|
+
#
|
|
3
|
+
# SPDX-License-Identifier: Apache-2.0
|
|
4
|
+
|
|
5
|
+
from nvmath.bindings import cublasLt as cublaslt
|
|
6
|
+
from nvmath.bindings.cublasLt import Order
|
|
7
|
+
from nvmath.linalg._internal.layout import BLASMatrixTraits, BLASMMTraitsView, MMLayoutChecker, MMLayoutCheckerLookupTable
|
|
8
|
+
|
|
9
|
+
from .qualifiers import MatrixQualifier
|
|
10
|
+
|
|
11
|
+
|
|
12
|
+
def mm_layout_checker_getter(qualifiers: MatrixQualifier) -> MMLayoutChecker:
|
|
13
|
+
"""Return a function which checks if a layout is supported for a BLAS matrix
|
|
14
|
+
multiplication."""
|
|
15
|
+
match (qualifiers[0]["abbreviation"], qualifiers[1]["abbreviation"], qualifiers[2]["abbreviation"]):
|
|
16
|
+
case ("ge", "ge", "ge"):
|
|
17
|
+
return is_supported_gemm_layout
|
|
18
|
+
case ("sy" | "he" | "dg", "ge", "ge"):
|
|
19
|
+
return is_supported_symm_layout_left
|
|
20
|
+
case ("ge", "sy" | "he" | "dg", "ge"):
|
|
21
|
+
return is_supported_symm_layout_right
|
|
22
|
+
case ("tr", "ge", "ge"):
|
|
23
|
+
return is_supported_trmm_layout_left
|
|
24
|
+
case ("ge", "tr", "ge"):
|
|
25
|
+
return is_supported_trmm_layout_right
|
|
26
|
+
case _:
|
|
27
|
+
raise NotImplementedError(
|
|
28
|
+
f"Layout supported checker for {qualifiers[0]['abbreviation']}, {qualifiers[1]['abbreviation']}, "
|
|
29
|
+
f"{qualifiers[2]['abbreviation']} is not implemented."
|
|
30
|
+
)
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
def is_supported_gemm_layout(traits: BLASMMTraitsView) -> bool:
|
|
34
|
+
"""Return True if the layout is supported for a BLAS matrix multiplication.
|
|
35
|
+
|
|
36
|
+
Rules:
|
|
37
|
+
- All operands must be COL order.
|
|
38
|
+
- C must be non-conjugate non-transpose.
|
|
39
|
+
- The general matrix, B, must not be conjugate non-transpose.
|
|
40
|
+
- The general matrix, A, must not be conjugate non-transpose.
|
|
41
|
+
"""
|
|
42
|
+
a_layout = traits.a_layout_traits
|
|
43
|
+
b_layout = traits.b_layout_traits
|
|
44
|
+
c_layout = traits.c_layout_traits
|
|
45
|
+
match (a_layout, b_layout, c_layout):
|
|
46
|
+
case (BLASMatrixTraits(is_conjugate=True, is_transpose=False), _, _):
|
|
47
|
+
# If A is conjugate non-transpose that is unsupported.
|
|
48
|
+
return False
|
|
49
|
+
case (_, BLASMatrixTraits(is_conjugate=True, is_transpose=False), _):
|
|
50
|
+
# If B is conjugate non-transpose that is unsupported.
|
|
51
|
+
return False
|
|
52
|
+
case (
|
|
53
|
+
BLASMatrixTraits(order=cublaslt.Order.COL),
|
|
54
|
+
BLASMatrixTraits(order=cublaslt.Order.COL),
|
|
55
|
+
BLASMatrixTraits(is_conjugate=False, is_transpose=False, order=cublaslt.Order.COL),
|
|
56
|
+
):
|
|
57
|
+
return True
|
|
58
|
+
case _:
|
|
59
|
+
return False
|
|
60
|
+
|
|
61
|
+
|
|
62
|
+
GEMM_LAYOUT_LOOKUP_TABLE: MMLayoutCheckerLookupTable = {
|
|
63
|
+
(Order.COL, Order.COL, Order.COL, False, False, False): [],
|
|
64
|
+
(Order.COL, Order.COL, Order.ROW, False, False, False): ["transpose_and_reorder_C", "swap_AB_and_transpose_ABC"],
|
|
65
|
+
(Order.COL, Order.COL, Order.ROW, False, True, False): ["transpose_and_reorder_C", "swap_AB_and_transpose_ABC"],
|
|
66
|
+
(Order.COL, Order.COL, Order.ROW, True, False, False): ["transpose_and_reorder_C", "swap_AB_and_transpose_ABC"],
|
|
67
|
+
(Order.COL, Order.COL, Order.ROW, True, True, False): ["transpose_and_reorder_C", "swap_AB_and_transpose_ABC"],
|
|
68
|
+
(Order.COL, Order.ROW, Order.COL, False, False, False): ["transpose_and_reorder_B"],
|
|
69
|
+
(Order.COL, Order.ROW, Order.COL, False, True, False): ["transpose_and_reorder_B"],
|
|
70
|
+
(Order.COL, Order.ROW, Order.ROW, False, False, False): [
|
|
71
|
+
"transpose_and_reorder_B",
|
|
72
|
+
"transpose_and_reorder_C",
|
|
73
|
+
"swap_AB_and_transpose_ABC",
|
|
74
|
+
],
|
|
75
|
+
(Order.COL, Order.ROW, Order.ROW, True, False, False): [
|
|
76
|
+
"transpose_and_reorder_B",
|
|
77
|
+
"transpose_and_reorder_C",
|
|
78
|
+
"swap_AB_and_transpose_ABC",
|
|
79
|
+
],
|
|
80
|
+
(Order.ROW, Order.COL, Order.COL, False, False, False): ["transpose_and_reorder_A"],
|
|
81
|
+
(Order.ROW, Order.COL, Order.COL, True, False, False): ["transpose_and_reorder_A"],
|
|
82
|
+
(Order.ROW, Order.COL, Order.ROW, False, False, False): [
|
|
83
|
+
"transpose_and_reorder_A",
|
|
84
|
+
"transpose_and_reorder_C",
|
|
85
|
+
"swap_AB_and_transpose_ABC",
|
|
86
|
+
],
|
|
87
|
+
(Order.ROW, Order.COL, Order.ROW, False, True, False): [
|
|
88
|
+
"transpose_and_reorder_A",
|
|
89
|
+
"transpose_and_reorder_C",
|
|
90
|
+
"swap_AB_and_transpose_ABC",
|
|
91
|
+
],
|
|
92
|
+
(Order.ROW, Order.ROW, Order.COL, False, False, False): ["transpose_and_reorder_A", "transpose_and_reorder_B"],
|
|
93
|
+
(Order.ROW, Order.ROW, Order.COL, False, True, False): ["transpose_and_reorder_A", "transpose_and_reorder_B"],
|
|
94
|
+
(Order.ROW, Order.ROW, Order.COL, True, False, False): ["transpose_and_reorder_A", "transpose_and_reorder_B"],
|
|
95
|
+
(Order.ROW, Order.ROW, Order.COL, True, True, False): ["transpose_and_reorder_A", "transpose_and_reorder_B"],
|
|
96
|
+
(Order.ROW, Order.ROW, Order.ROW, False, False, False): [
|
|
97
|
+
"transpose_and_reorder_A",
|
|
98
|
+
"transpose_and_reorder_B",
|
|
99
|
+
"transpose_and_reorder_C",
|
|
100
|
+
"swap_AB_and_transpose_ABC",
|
|
101
|
+
],
|
|
102
|
+
}
|
|
103
|
+
|
|
104
|
+
|
|
105
|
+
def is_supported_symm_layout(traits: BLASMMTraitsView, left_side: bool) -> bool:
|
|
106
|
+
"""Return True if the layout is supported for a BLAS symmetric matrix multiplication.
|
|
107
|
+
|
|
108
|
+
Rules:
|
|
109
|
+
- All operands must be COL order.
|
|
110
|
+
- The general matrix, C, must be non-conjugate and non-transpose.
|
|
111
|
+
- The general matrix, B, must be non-conjugate and non-transpose.
|
|
112
|
+
- The symmetric matrix, A, must be non-conjugate; transpose is allowed.
|
|
113
|
+
"""
|
|
114
|
+
if left_side:
|
|
115
|
+
t_layout = traits.a_layout_traits
|
|
116
|
+
g_layout = traits.b_layout_traits
|
|
117
|
+
else:
|
|
118
|
+
g_layout = traits.a_layout_traits
|
|
119
|
+
t_layout = traits.b_layout_traits
|
|
120
|
+
c_layout = traits.c_layout_traits
|
|
121
|
+
match (t_layout, g_layout, c_layout):
|
|
122
|
+
case (
|
|
123
|
+
BLASMatrixTraits(is_conjugate=False, order=cublaslt.Order.COL),
|
|
124
|
+
BLASMatrixTraits(is_conjugate=False, is_transpose=False, order=cublaslt.Order.COL),
|
|
125
|
+
BLASMatrixTraits(is_conjugate=False, is_transpose=False, order=cublaslt.Order.COL),
|
|
126
|
+
):
|
|
127
|
+
return True
|
|
128
|
+
case _:
|
|
129
|
+
return False
|
|
130
|
+
|
|
131
|
+
|
|
132
|
+
def is_supported_symm_layout_left(traits: BLASMMTraitsView) -> bool:
|
|
133
|
+
return is_supported_symm_layout(traits, left_side=True)
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
SYMM_LAYOUT_LOOKUP_TABLE_LEFT: MMLayoutCheckerLookupTable = {
|
|
137
|
+
(Order.COL, Order.COL, Order.COL, False, False, False): [],
|
|
138
|
+
(Order.ROW, Order.COL, Order.COL, False, False, False): ["transpose_and_reorder_A"],
|
|
139
|
+
(Order.ROW, Order.COL, Order.ROW, False, False, False): [
|
|
140
|
+
"transpose_and_reorder_A",
|
|
141
|
+
"transpose_and_reorder_C",
|
|
142
|
+
"swap_AB_and_transpose_ABC",
|
|
143
|
+
],
|
|
144
|
+
(Order.ROW, Order.ROW, Order.ROW, False, False, False): [
|
|
145
|
+
"transpose_and_reorder_A",
|
|
146
|
+
"transpose_and_reorder_B",
|
|
147
|
+
"transpose_and_reorder_C",
|
|
148
|
+
"swap_AB_and_transpose_ABC",
|
|
149
|
+
],
|
|
150
|
+
}
|
|
151
|
+
|
|
152
|
+
|
|
153
|
+
def is_supported_symm_layout_right(traits: BLASMMTraitsView) -> bool:
|
|
154
|
+
return is_supported_symm_layout(traits, left_side=False)
|
|
155
|
+
|
|
156
|
+
|
|
157
|
+
SYMM_LAYOUT_LOOKUP_TABLE_RIGHT: MMLayoutCheckerLookupTable = {
|
|
158
|
+
(Order.COL, Order.COL, Order.COL, False, False, False): [],
|
|
159
|
+
(Order.COL, Order.ROW, Order.COL, False, False, False): ["transpose_and_reorder_B"],
|
|
160
|
+
(Order.COL, Order.ROW, Order.ROW, False, False, False): [
|
|
161
|
+
"transpose_and_reorder_B",
|
|
162
|
+
"transpose_and_reorder_C",
|
|
163
|
+
"swap_AB_and_transpose_ABC",
|
|
164
|
+
],
|
|
165
|
+
(Order.ROW, Order.ROW, Order.ROW, False, False, False): [
|
|
166
|
+
"transpose_and_reorder_A",
|
|
167
|
+
"transpose_and_reorder_B",
|
|
168
|
+
"transpose_and_reorder_C",
|
|
169
|
+
"swap_AB_and_transpose_ABC",
|
|
170
|
+
],
|
|
171
|
+
}
|
|
172
|
+
|
|
173
|
+
|
|
174
|
+
def is_supported_trmm_layout(traits: BLASMMTraitsView, left_side: bool) -> bool:
|
|
175
|
+
"""Return True if the layout is supported for a BLAS triangular matrix multiplication.
|
|
176
|
+
|
|
177
|
+
Rules:
|
|
178
|
+
- All operands must be COL order.
|
|
179
|
+
- The general matrix, C, must be non-conjugate and non-transpose.
|
|
180
|
+
- The general matrix, B, must be non-conjugate and non-transpose.
|
|
181
|
+
- The triangular matrix, A, must not be conjugate non-transpose.
|
|
182
|
+
"""
|
|
183
|
+
if left_side:
|
|
184
|
+
t_layout = traits.a_layout_traits
|
|
185
|
+
g_layout = traits.b_layout_traits
|
|
186
|
+
else:
|
|
187
|
+
g_layout = traits.a_layout_traits
|
|
188
|
+
t_layout = traits.b_layout_traits
|
|
189
|
+
c_layout = traits.c_layout_traits
|
|
190
|
+
match (t_layout, g_layout, c_layout):
|
|
191
|
+
case (BLASMatrixTraits(is_conjugate=True, is_transpose=False), _, _):
|
|
192
|
+
# If A is conjugate non-transpose that is unsupported.
|
|
193
|
+
return False
|
|
194
|
+
case (
|
|
195
|
+
BLASMatrixTraits(order=cublaslt.Order.COL),
|
|
196
|
+
BLASMatrixTraits(is_conjugate=False, is_transpose=False, order=cublaslt.Order.COL),
|
|
197
|
+
BLASMatrixTraits(is_conjugate=False, is_transpose=False, order=cublaslt.Order.COL),
|
|
198
|
+
):
|
|
199
|
+
return True
|
|
200
|
+
case _:
|
|
201
|
+
return False
|
|
202
|
+
|
|
203
|
+
|
|
204
|
+
def is_supported_trmm_layout_left(traits: BLASMMTraitsView) -> bool:
|
|
205
|
+
return is_supported_trmm_layout(traits, left_side=True)
|
|
206
|
+
|
|
207
|
+
|
|
208
|
+
TRMM_LAYOUT_LOOKUP_TABLE_LEFT: MMLayoutCheckerLookupTable = {
|
|
209
|
+
(Order.COL, Order.COL, Order.COL, False, False, False): [],
|
|
210
|
+
(Order.ROW, Order.COL, Order.COL, False, False, False): ["transpose_and_reorder_A"],
|
|
211
|
+
(Order.ROW, Order.COL, Order.COL, True, False, False): ["transpose_and_reorder_A"],
|
|
212
|
+
(Order.ROW, Order.COL, Order.ROW, False, False, False): [
|
|
213
|
+
"transpose_and_reorder_A",
|
|
214
|
+
"transpose_and_reorder_C",
|
|
215
|
+
"swap_AB_and_transpose_ABC",
|
|
216
|
+
],
|
|
217
|
+
(Order.ROW, Order.COL, Order.ROW, False, True, False): [
|
|
218
|
+
"transpose_and_reorder_A",
|
|
219
|
+
"transpose_and_reorder_C",
|
|
220
|
+
"swap_AB_and_transpose_ABC",
|
|
221
|
+
],
|
|
222
|
+
(Order.ROW, Order.ROW, Order.ROW, False, False, False): [
|
|
223
|
+
"transpose_and_reorder_A",
|
|
224
|
+
"transpose_and_reorder_B",
|
|
225
|
+
"transpose_and_reorder_C",
|
|
226
|
+
"swap_AB_and_transpose_ABC",
|
|
227
|
+
],
|
|
228
|
+
}
|
|
229
|
+
|
|
230
|
+
|
|
231
|
+
def is_supported_trmm_layout_right(traits: BLASMMTraitsView) -> bool:
|
|
232
|
+
return is_supported_trmm_layout(traits, left_side=False)
|
|
233
|
+
|
|
234
|
+
|
|
235
|
+
TRMM_LAYOUT_LOOKUP_TABLE_RIGHT: MMLayoutCheckerLookupTable = {
|
|
236
|
+
(Order.COL, Order.COL, Order.COL, False, False, False): [],
|
|
237
|
+
(Order.COL, Order.ROW, Order.COL, False, False, False): ["transpose_and_reorder_B"],
|
|
238
|
+
(Order.COL, Order.ROW, Order.COL, False, True, False): ["transpose_and_reorder_B"],
|
|
239
|
+
(Order.COL, Order.ROW, Order.ROW, False, False, False): [
|
|
240
|
+
"transpose_and_reorder_B",
|
|
241
|
+
"transpose_and_reorder_C",
|
|
242
|
+
"swap_AB_and_transpose_ABC",
|
|
243
|
+
],
|
|
244
|
+
(Order.COL, Order.ROW, Order.ROW, True, False, False): [
|
|
245
|
+
"transpose_and_reorder_B",
|
|
246
|
+
"transpose_and_reorder_C",
|
|
247
|
+
"swap_AB_and_transpose_ABC",
|
|
248
|
+
],
|
|
249
|
+
(Order.ROW, Order.ROW, Order.ROW, False, False, False): [
|
|
250
|
+
"transpose_and_reorder_A",
|
|
251
|
+
"transpose_and_reorder_B",
|
|
252
|
+
"transpose_and_reorder_C",
|
|
253
|
+
"swap_AB_and_transpose_ABC",
|
|
254
|
+
],
|
|
255
|
+
}
|
|
256
|
+
|
|
257
|
+
CACHED_LAYOUT_CHECKERS: dict[MMLayoutChecker, MMLayoutCheckerLookupTable] = {
|
|
258
|
+
is_supported_gemm_layout: GEMM_LAYOUT_LOOKUP_TABLE,
|
|
259
|
+
is_supported_symm_layout_left: SYMM_LAYOUT_LOOKUP_TABLE_LEFT,
|
|
260
|
+
is_supported_symm_layout_right: SYMM_LAYOUT_LOOKUP_TABLE_RIGHT,
|
|
261
|
+
is_supported_trmm_layout_left: TRMM_LAYOUT_LOOKUP_TABLE_LEFT,
|
|
262
|
+
is_supported_trmm_layout_right: TRMM_LAYOUT_LOOKUP_TABLE_RIGHT,
|
|
263
|
+
}
|