Home
last modified time | relevance | path

Searched +full:opt +full:- +full:einsum (Results 1 – 25 of 34) sorted by relevance

12

/external/pytorch/test/torch_np/numpy_tests/core/
Dtest_einsum.py29 # Setup for optimize einsum
41 (TypeError, IndexError, ValueError), np.einsum, optimize=do_opt
43 assert_raises((IndexError, ValueError), np.einsum, "", optimize=do_opt)
46 assert_raises((AttributeError, TypeError), np.einsum, 0, 0, optimize=do_opt)
49 assert_raises(TypeError, np.einsum, "", 0, out="test", optimize=do_opt)
54 np.einsum,
62 assert_raises(ValueError, np.einsum, "", 0, casting="blah", optimize=do_opt)
66 TypeError, np.einsum, "", 0, dtype="bad_data_type", optimize=do_opt
70 assert_raises(TypeError, np.einsum, "", 0, bad_arg=0, optimize=do_opt)
74 (RuntimeError, TypeError), np.einsum, *(None,) * 63, optimize=do_opt
[all …]
/external/pytorch/torch/backends/opt_einsum/
D__init__.py1 # mypy: allow-untyped-defs
18 def is_available() -> bool:
23 def get_opt_einsum() -> Any:
28 def _set_enabled(_enabled: bool) -> None:
32 "the benefits of calculating an optimal path for einsum. torch.einsum will "
34 "calculation, please install opt-einsum."
40 def _get_enabled() -> bool:
44 def _set_strategy(_strategy: str) -> None:
48 "torch.einsum will bypass path calculation and simply contract from left to right. "
54 "torch.einsum will bypass path calculation and simply contract from left to right. "
[all …]
/external/tensorflow/tensorflow/compiler/mlir/tensorflow/tests/
Dbatchmatmul_to_einsum.mlir1 // RUN: tf-opt %s -tf-batch-matmul-to-tf-einsum | FileCheck %s
3 func.func @test_batch_matmul_to_einsum(%arg0: tensor<1x2x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<…
4 // CHECK-LABEL: test_batch_matmul_to_einsum
5 …// CHECK: "tf.Einsum"(%arg0, %arg1) {equation = "...mk,...kn->...mn"} : (tensor<1x2x3xf32>, tensor…
6 …, %arg1) {adj_x = false, adj_y = false} : (tensor<1x2x3xf32>, tensor<3x4xf32>) -> tensor<1x2x4xf32>
10 …tmul_broadcast_to_einsum(%arg0: tensor<2x2x4xf32>, %arg1: tensor<2x4x2xf32>) -> tensor<2x2x2xf32> {
11 // CHECK-LABEL: test_batch_matmul_broadcast_to_einsum
12 …// CHECK: "tf.Einsum"(%arg0, %arg1) {equation = "...mk,...kn->...mn"} : (tensor<2x2x4xf32>, tensor…
13 …%arg1) {adj_x = false, adj_y = false} : (tensor<2x2x4xf32>, tensor<2x4x2xf32>) -> tensor<2x2x2xf32>
17 …c_shape_both_arg_to_einsum(%arg0: tensor<1x2x?xf32>, %arg1: tensor<?x4xf32>) -> tensor<1x2x4xf32> {
[all …]
Deinsum.mlir1 // RUN: tf-opt -split-input-file -verify-diagnostics -tf-einsum %s | FileCheck %s
3 func.func @einsum_basic(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x5x6xf32>) -> tensor<3x4x6xf32> {
4 …%0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ijk,ikm->ijm"}: (tensor<3x4x5x…
6 // CHECK-LABEL: einsum_basic
7 …%arg1) {adj_x = false, adj_y = false} : (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>
10 func.func @einsum_matmul(%arg0: tensor<7x9xf32>, %arg1: tensor<9x5xf32>) -> tensor<7x5xf32> {
11 …%0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ae,ed->ad"}: (tensor<7x9xf32>,…
13 // CHECK-LABEL: einsum_matmul
14 …arg0, %arg1) {adj_x = false, adj_y = false} : (tensor<7x9xf32>, tensor<9x5xf32>) -> tensor<7x5xf32>
18 func.func @einsum_broadcast(%arg0: tensor<3x4x5xf32>, %arg1: tensor<5x6xf32>) -> tensor<3x4x6xf32> {
[all …]
/external/tensorflow/tensorflow/compiler/xla/mlir_hlo/tests/Dialect/mhlo/
Dhlo-legalize-einsum-to-dot-general.mlir1 // RUN: mlir-hlo-opt -mhlo-legalize-einsum-to-dot-general %s -o - | FileCheck %s
3 func.func @einsum_diag(%arg0: tensor<6x6xf32>) -> tensor<6xf32> {
5 …%1 = "mhlo.einsum"(%0, %arg0) {einsum_config = ",ii->i"} : (tensor<f32>, tensor<6x6xf32>) -> tenso…
8 // CHECK-LABEL: func @einsum_diag
9 // NOTE: 2-operand diagonal not supported as a dot_general lowering.
12 // CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]+]]
14 // CHECK: "mhlo.einsum"
16 …_high_rank_vector_mul(%arg0: tensor<8x2x6xf32>, %arg1: tensor<8x5x3x6xf32>) -> tensor<8x5x3x2xf32>…
17 …%0 = "mhlo.einsum"(%arg0, %arg1) {einsum_config = "bxy,bijy->bijx"} : (tensor<8x2x6xf32>, tensor<8…
20 // CHECK-LABEL: func @einsum_batched_matrix_high_rank_vector_mul
[all …]
Dhlo-legalize-to-linalg.mlir1 // RUN: mlir-hlo-opt %s -hlo-legalize-to-linalg -split-input-file --canonicalize | FILECHECK_OPTS="…
3 // CHECK: #map = affine_map<(d0, d1) -> (d0, d1)>
4 // CHECK-LABEL: func @float_add
6 %rhs: tensor<2x2xf32>) -> tensor<2x2xf32> {
8 // CHECK-SAME: {someattr}
9 // CHECK: ^{{[a-z0-9_]*}}
10 // CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]*]]: f32
11 // CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]*]]: f32
12 // CHECK: %[[RESULT:[a-zA-Z0-9_]*]] = arith.addf %[[ARG0]], %[[ARG1]]
15 : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
[all …]
/external/tensorflow/tensorflow/compiler/mlir/hlo/tests/
Dhlo-legalize-einsum-to-dot-general.mlir1 // RUN: mlir-hlo-opt -mhlo-legalize-einsum-to-dot-general %s -o - | FileCheck %s
3 func @einsum_diag(%arg0: tensor<6x6xf32>) -> tensor<6xf32> {
5 …%1 = "mhlo.einsum"(%0, %arg0) {einsum_config = ",ii->i"} : (tensor<f32>, tensor<6x6xf32>) -> tenso…
8 // CHECK-LABEL: func @einsum_diag
9 // CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]+]]
12 // CHECK-SAME: dot_dimension_numbers = {
13 // CHECK-SAME: lhs_batching_dimensions = dense<> : tensor<0xi64>,
14 // CHECK-SAME: lhs_contracting_dimensions = dense<> : tensor<0xi64>,
15 // CHECK-SAME: rhs_batching_dimensions = dense<> : tensor<0xi64>,
16 // CHECK-SAME: rhs_contracting_dimensions = dense<> : tensor<0xi64>}
[all …]
/external/pytorch/.github/requirements/
Dpip-requirements-macOS.txt8 # Use numba-0.49.1 or older on Intel Macs, but 0.56.0 on M1 machines, as older numba is not availab…
11 opt-einsum>=3.3
13 nvidia-ml-py==11.525.84
17 pytest-xdist==3.3.1
18 pytest-rerunfailures==10.3
19 pytest-flakefinder==1.1.0
23 unittest-xml-reporting<=3.2.0,>=2.0.0
26 pytest-cpp==2.3.0
28 z3-solver==4.12.2.0
/external/pytorch/aten/src/ATen/native/
DLinear.cpp43 static int value = -1; in parseLinearFlatten3d()
44 if (value == -1) { in parseLinearFlatten3d()
59 // can't use -1 in reshape because it errors when a dimension is 0 in _flatten_nd_linear()
61 for (int64_t i = 0, ndim = input_sizes.size(); i < ndim - 1; ++i) { in _flatten_nd_linear()
64 auto inp_reshape = input.reshape_symint({flattened_dim, input_sizes.at(input_sizes.size() -1)}); in _flatten_nd_linear()
66 auto new_size = input_sizes.slice(0, input_sizes.size() - 1); in _flatten_nd_linear()
94 if (input_dim == 2 && bias->defined()) { in linear()
98 if (bias->defined() && !input.is_xla()) { in linear()
103 …ous() && input.layout() == c10::kStrided && weight.layout() == c10::kStrided && bias->dim() == 1) { in linear()
112 if (bias->defined()) { in linear()
[all …]
/external/pytorch/.ci/docker/
Drequirements-ci.txt29 #Description: dill extends pickle with serializing and de-serializing for most built-ins
70 #mkl #this breaks linux-bionic-rocm4.5-py3.7
77 #mkl-devel
87 #MonkeyType # breaks pytorch-xla-linux-bionic-py3.7-clang8
114 #Description: Just-In-Time Compiler for Numerical Functions
120 #Description: Provides N-dimensional arrays and linear algebra
137 opt-einsum==3.3
138 #Description: Python library to optimize tensor contraction order, used in einsum
175 pytest-xdist==3.3.1
180 pytest-flakefinder==1.1.0
[all …]
/external/pytorch/.github/workflows/
D_docs.yml6 build-environment:
9 description: Top-level label for what's being built/tested.
10 docker-image:
19 run-doxygen:
24 sync-tag:
30 job with the same `sync-tag` is identical.
31 s3-bucket:
35 default: "gha-artifacts"
36 aws-role-to-assume:
41 upload-aws-role-to-assume:
[all …]
D_win-test.yml1 name: win-test
6 build-environment:
9 description: Top-level label for what's being built/tested.
10 cuda-version:
14 test-matrix:
18 sync-tag:
24 job with the same `sync-tag` is identical.
25 timeout-minutes:
38 if: github.repository_owner == 'pytorch' && toJSON(fromJSON(inputs.test-matrix).include) != '[]'
40 matrix: ${{ fromJSON(inputs.test-matrix) }}
[all …]
D_linux-test.yml1 name: linux-test
6 build-environment:
9 description: Top-level label for what's being built/tested.
10 test-matrix:
14 docker-image:
18 sync-tag:
24 job with the same `sync-tag` is identical.
25 timeout-minutes:
31 use-gha:
36 dashboard-tag:
[all …]
/external/pytorch/.github/actions/linux-test/
Daction.yml1 name: linux-test
4 build-environment:
7 description: Top-level label for what's being built/tested.
8 test-matrix:
12 docker-image:
16 sync-tag:
22 job with the same `sync-tag` is identical.
23 use-gha:
28 dashboard-tag:
32 s3-bucket:
[all …]
/external/pytorch/
Dsetup.py5 # build with -O0 and -g (debug symbols)
8 # build with optimizations and -g (debug symbols)
32 # build CUDA kernels with -lineinfo --source-in-ptx. Note that
67 # MKL-DNN threading mode: TBB or OMP (default)
70 # Prefer to link with MKL statically - Unix only
90 # disables use of system-wide nccl (we will use our submoduled
113 # specify the version of PyTorch, rather than the hard-coded version
127 # specify a namespace for ONNX built here rather than the hard-coded
147 # /usr/local/cuda-x.y
178 # ATen parallel backend to use for intra- and inter-op parallelism
[all …]
/external/pytorch/aten/src/ATen/
Dautocast_mode.h25 // deprecated CUDA-specific autocast APIs
207 // If nextArg is floating-point, compare its scalar_type with our
258 // Template to catch non-Tensor args (no-op that returns current best guess)
275 // promoted to float32. Non-Tensor arguments are ignored.
340 // Template to catch non-Tensor args.
389 // Policies correspond to op categories that need code-divergent handling.
411 // and redispatch to the type-aware overload.
420 https://stackoverflow.com/questions/46533698/how-to-deduce-argument-list-from-function-pointer)
503 // If ineligible, calls F with unaltered args. Does not set opt dtype,
504 // because setting opt dtype explicitly may interfere with internal
[all …]
/external/pytorch/c10/core/
DDispatchKey.h22 // customizable per-backend, then we also look at the lower BackendComponent
30 // built-in keys)
32 // If you add a new (non-privateuse) backend here,
53 // WARNING! If we add a new per-backend functionality key that has higher
79 // - dense kernels (e.g. DispatchKey::CPU)
80 // - sparse kernels (e.g. DispatchKey::SparseCPU)
81 // - quantized kernels (e.g. DispatchKey::QuantizedCPU)
82 // - autograd kernels (e.g. DispatchKey::AutogradCPU)
86 // per-backend functionality]
116 // (1) non-customizable backends (e.g. FPGA)
[all …]
/external/tensorflow/tensorflow/compiler/mlir/quantization/tensorflow/tests/
Dprepare_lifting.mlir7 // http://www.apache.org/licenses/LICENSE-2.0
15 // RUN: tf-quant-opt %s -quant-prepare-lifting | FileCheck %s
17 func.func @decompose_batch_norm(%arg0: tensor<*xf32>) -> (tensor<*xf32>) {
18 %cst = "tf.Const"() {value = dense<1.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
19 %cst_0 = "tf.Const"() {value = dense<0.500000e+00> : tensor<2xf32>} : () -> tensor<2xf32>
20 …-5 : f32, exponential_avg_factor = 1.000000e+00 : f32, is_training = false} : (tensor<*xf32>, tens…
24 // CHECK-DAG: %[[CONST:.*]] = "tf.Const"() {value = dense<2.49743462E-5> : tensor<2xf32>} : () -> t…
25 // CHECK-DAG: %[[CONST_0:.*]] = "tf.Const"() {value = dense<0.999950051> : tensor<2xf32>} : () -> t…
26 // CHECK: %[[mul:.*]] = "tf.Mul"(%arg0, %[[CONST_0]]) : (tensor<*xf32>, tensor<2xf32>) -> tensor<*x…
27 // CHECK: %[[add:.*]] = "tf.Add"(%[[mul]], %[[CONST]]) : (tensor<*xf32>, tensor<2xf32>) -> tensor<*…
[all …]
/external/tensorflow/tensorflow/compiler/mlir/tensorflow/ir/
Dtf_ops_a_m.cc7 http://www.apache.org/licenses/LICENSE-2.0
42 #include "mlir/Dialect/Func/IR/FuncOps.h" // from @llvm-project
43 #include "mlir/Dialect/Traits.h" // from @llvm-project
44 #include "mlir/IR/Attributes.h" // from @llvm-project
45 #include "mlir/IR/Builders.h" // from @llvm-project
46 #include "mlir/IR/BuiltinAttributes.h" // from @llvm-project
47 #include "mlir/IR/BuiltinOps.h" // from @llvm-project
48 #include "mlir/IR/BuiltinTypes.h" // from @llvm-project
49 #include "mlir/IR/Diagnostics.h" // from @llvm-project
50 #include "mlir/IR/DialectImplementation.h" // from @llvm-project
[all …]
/external/tensorflow/tensorflow/python/util/
Ddispatch.py7 # http://www.apache.org/licenses/LICENSE-2.0
15 """Type-based dispatch for TensorFlow's Python APIs.
27 * The "type-based dispatch" system checks the types of the parameters passed
34 dispatch system in favor of the type-based dispatch system, once all users have
47 ### Type-based Dispatch
49 The main interface for the type-based dispatch system is the `dispatch_for_api`
58 visible ops by default. APIs/ops that are implemented in Python can opt in to
64 import typing # pylint: disable=unused-import (used in doctests)
103 def handle(self, args, kwargs): # pylint: disable=unused-argument
200 # pylint: disable=g-doc-return-or-yield
[all …]
/external/tensorflow/tensorflow/compiler/mlir/xla/tests/translate/
Dexport.mlir1 // RUN: tf-mlir-translate -split-input-file -mlir-hlo-to-hlo-text %s | FileCheck %s
2 // RUN: tf-mlir-translate -split-input-file -mlir-hlo-to-hlo-text --via-builder=true %s | FileCheck…
5 func.func @main(%arg0: tensor<2xi1>) -> tensor<2xi1> {
6 %0 = "mhlo.add"(%arg0, %arg0) : (tensor<2xi1>, tensor<2xi1>) -> tensor<2xi1>
14 // -----
17 func.func @main(%arg0: !mhlo.token, %arg1: !mhlo.token) -> !mhlo.token {
18 %0 = "mhlo.after_all"(%arg0, %arg1) : (!mhlo.token, !mhlo.token) -> !mhlo.token
25 // CHECK: ROOT %[[RESULT:.*]] = token[] after-all(token[] %[[ARG0]], token[] %[[ARG1]])
27 // -----
30 func.func @main(%arg0: tensor<10xf32>) -> tensor<5xf32> {
[all …]
/external/tensorflow/tensorflow/compiler/mlir/xla/transforms/
Dlegalize_tf.cc7 http://www.apache.org/licenses/LICENSE-2.0
32 #include "mlir/Dialect/Arithmetic/IR/Arithmetic.h" // from @llvm-project
33 #include "mlir/Dialect/Func/IR/FuncOps.h" // from @llvm-project
34 #include "mlir/Dialect/Shape/IR/Shape.h" // from @llvm-project
35 #include "mlir/Dialect/Tensor/IR/Tensor.h" // from @llvm-project
36 #include "mlir/Dialect/Traits.h" // from @llvm-project
37 #include "mlir/IR/Attributes.h" // from @llvm-project
38 #include "mlir/IR/BuiltinOps.h" // from @llvm-project
39 #include "mlir/IR/BuiltinTypes.h" // from @llvm-project
40 #include "mlir/IR/Diagnostics.h" // from @llvm-project
[all …]
/external/tensorflow/
DRELEASE.md57 * tf.einsum is supported with multiple unknown shapes.
78 …xperimental/SharpnessAwareMinimization). This class implements the sharpness-aware minimization te…
82 …* Added support for cross-trainer data caching in tf.data service. This saves computation resour…
91 …mptionCheckpointHandler) to handle worker preemption/maintenance and cluster-wise consistent error…
95 … to 100 times depending on the size of k. When running on CPU and GPU, a non-optimized XLA kernel …
117 …inistically (and `tf.errors.UnimplementedError` is no longer thrown) when op-determinism has been …
141 …-Review-Doctor, DEKHTIARJonathan, Deven Desai, Djacon, Duncan Riach, fedotoff, fo40225, Frederic B…
163 …ers should switch to [TensorFlow Decision Forests](https://github.com/tensorflow/decision-forests).
166 …eels now specifically conform to [manylinux2014](https://peps.python.org/pep-0599/), an upgrade fr…
167 … Forum thread](https://discuss.tensorflow.org/t/tensorflow-linux-wheels-are-being-upgraded-to-many…
[all …]
/external/tensorflow/tensorflow/compiler/mlir/xla/tests/
Dlegalize-tf.mlir1 // RUN: xla-opt "-xla-legalize-tf=allow-partial-conversion legalize-chlo=false" -split-input-file %…
2 …-opt "-xla-legalize-tf=allow-partial-conversion legalize-chlo=true" -split-input-file -verify-diag…
10 //===----------------------------------------------------------------------===//
12 //===----------------------------------------------------------------------===//
14 // -----
19 // CHECK-LABEL: fusedBatchNormV2_noTraining
20 …r<8xf32>, %arg2: tensor<8xf32>, %arg3: tensor<8xf32>, %arg4: tensor<8xf32>) -> (tensor<8x8x8x8xf32…
21 …ilon = 1.000000e-03 : f32, feature_index = 3 : i64} : (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<…
22 …r<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>) -> (tensor<8x8x8x8xf32…
26 // -----
[all …]
Dlegalize-tf-no-tf2xla-fallback.mlir1 // RUN: xla-opt "-xla-legalize-tf-no-fallback=allow-partial-conversion" -split-input-file %s | FILE…
3 //===----------------------------------------------------------------------===//
5 //===----------------------------------------------------------------------===//
7 // -----
12 // CHECK-LABEL: fusedBatchNormV2_noTraining
13 …r<8xf32>, %arg2: tensor<8xf32>, %arg3: tensor<8xf32>, %arg4: tensor<8xf32>) -> (tensor<8x8x8x8xf32…
14 …ilon = 1.000000e-03 : f32, feature_index = 3 : i64} : (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<…
15 …r<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>) -> (tensor<8x8x8x8xf32…
19 // -----
21 // CHECK-LABEL: fusedBatchNormV2_training
[all …]

12