warp-lang 0.9.0__py3-none-win_amd64.whl → 0.11.0__py3-none-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.
Potentially problematic release.
This version of warp-lang might be problematic. Click here for more details.
- warp/__init__.py +15 -7
- warp/__init__.pyi +1 -0
- warp/bin/warp-clang.dll +0 -0
- warp/bin/warp.dll +0 -0
- warp/build.py +22 -443
- warp/build_dll.py +384 -0
- warp/builtins.py +998 -488
- warp/codegen.py +1307 -739
- warp/config.py +5 -3
- warp/constants.py +6 -0
- warp/context.py +1291 -548
- warp/dlpack.py +31 -31
- warp/fabric.py +326 -0
- warp/fem/__init__.py +27 -0
- warp/fem/cache.py +389 -0
- warp/fem/dirichlet.py +181 -0
- warp/fem/domain.py +263 -0
- warp/fem/field/__init__.py +101 -0
- warp/fem/field/field.py +149 -0
- warp/fem/field/nodal_field.py +299 -0
- warp/fem/field/restriction.py +21 -0
- warp/fem/field/test.py +181 -0
- warp/fem/field/trial.py +183 -0
- warp/fem/geometry/__init__.py +19 -0
- warp/fem/geometry/closest_point.py +70 -0
- warp/fem/geometry/deformed_geometry.py +271 -0
- warp/fem/geometry/element.py +744 -0
- warp/fem/geometry/geometry.py +186 -0
- warp/fem/geometry/grid_2d.py +373 -0
- warp/fem/geometry/grid_3d.py +435 -0
- warp/fem/geometry/hexmesh.py +953 -0
- warp/fem/geometry/partition.py +376 -0
- warp/fem/geometry/quadmesh_2d.py +532 -0
- warp/fem/geometry/tetmesh.py +840 -0
- warp/fem/geometry/trimesh_2d.py +577 -0
- warp/fem/integrate.py +1616 -0
- warp/fem/operator.py +191 -0
- warp/fem/polynomial.py +213 -0
- warp/fem/quadrature/__init__.py +2 -0
- warp/fem/quadrature/pic_quadrature.py +245 -0
- warp/fem/quadrature/quadrature.py +294 -0
- warp/fem/space/__init__.py +292 -0
- warp/fem/space/basis_space.py +489 -0
- warp/fem/space/collocated_function_space.py +105 -0
- warp/fem/space/dof_mapper.py +236 -0
- warp/fem/space/function_space.py +145 -0
- warp/fem/space/grid_2d_function_space.py +267 -0
- warp/fem/space/grid_3d_function_space.py +306 -0
- warp/fem/space/hexmesh_function_space.py +352 -0
- warp/fem/space/partition.py +350 -0
- warp/fem/space/quadmesh_2d_function_space.py +369 -0
- warp/fem/space/restriction.py +160 -0
- warp/fem/space/shape/__init__.py +15 -0
- warp/fem/space/shape/cube_shape_function.py +738 -0
- warp/fem/space/shape/shape_function.py +103 -0
- warp/fem/space/shape/square_shape_function.py +611 -0
- warp/fem/space/shape/tet_shape_function.py +567 -0
- warp/fem/space/shape/triangle_shape_function.py +429 -0
- warp/fem/space/tetmesh_function_space.py +292 -0
- warp/fem/space/topology.py +295 -0
- warp/fem/space/trimesh_2d_function_space.py +221 -0
- warp/fem/types.py +77 -0
- warp/fem/utils.py +495 -0
- warp/native/array.h +164 -55
- warp/native/builtin.h +150 -174
- warp/native/bvh.cpp +75 -328
- warp/native/bvh.cu +406 -23
- warp/native/bvh.h +37 -45
- warp/native/clang/clang.cpp +136 -24
- warp/native/crt.cpp +1 -76
- warp/native/crt.h +111 -104
- warp/native/cuda_crt.h +1049 -0
- warp/native/cuda_util.cpp +15 -3
- warp/native/cuda_util.h +3 -1
- warp/native/cutlass/tools/library/scripts/conv2d_operation.py +463 -0
- warp/native/cutlass/tools/library/scripts/conv3d_operation.py +321 -0
- warp/native/cutlass/tools/library/scripts/gemm_operation.py +988 -0
- warp/native/cutlass/tools/library/scripts/generator.py +4625 -0
- warp/native/cutlass/tools/library/scripts/library.py +799 -0
- warp/native/cutlass/tools/library/scripts/manifest.py +402 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/docs/source/conf.py +96 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/profile/conv/conv2d_f16_sm80.py +106 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/profile/gemm/gemm_f32_sm80.py +91 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/setup.py +80 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/__init__.py +48 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/arguments.py +118 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/c_types.py +241 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/compiler.py +432 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/conv2d_operation.py +631 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/epilogue.py +1026 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/frontend.py +104 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/gemm_operation.py +1276 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/library.py +744 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/memory_manager.py +74 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/operation.py +110 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/parser.py +619 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/reduction_operation.py +398 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/tensor_ref.py +70 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/__init__.py +4 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/conv2d_testbed.py +646 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/gemm_grouped_testbed.py +235 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/gemm_testbed.py +557 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/profiler.py +70 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/type_hint.py +39 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/utils/__init__.py +1 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/utils/device.py +76 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/utils/reference_model.py +255 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/__init__.py +0 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py +201 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +177 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py +98 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py +95 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_few_channels_f16nhwc_f16nhwc_f16nhwc_tensor_op_f32_sm80.py +163 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_fixed_channels_f16nhwc_f16nhwc_f16nhwc_tensor_op_f32_sm80.py +187 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py +309 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +54 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py +96 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py +107 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_strided_dgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +253 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py +97 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +242 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py +96 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py +107 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/run_all_tests.py +10 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/frontend/test_frontend.py +146 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/__init__.py +0 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_bf16_sm80.py +96 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_f16_sm80.py +447 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_f32_sm80.py +146 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_f64_sm80.py +102 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_grouped_sm80.py +203 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_s8_sm80.py +229 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/run_all_tests.py +9 -0
- warp/native/cutlass/tools/library/scripts/pycutlass/test/unit/test_sm80.py +453 -0
- warp/native/cutlass/tools/library/scripts/rank_2k_operation.py +398 -0
- warp/native/cutlass/tools/library/scripts/rank_k_operation.py +387 -0
- warp/native/cutlass/tools/library/scripts/rt.py +796 -0
- warp/native/cutlass/tools/library/scripts/symm_operation.py +400 -0
- warp/native/cutlass/tools/library/scripts/trmm_operation.py +407 -0
- warp/native/cutlass_gemm.cu +5 -3
- warp/native/exports.h +1240 -949
- warp/native/fabric.h +228 -0
- warp/native/hashgrid.cpp +4 -4
- warp/native/hashgrid.h +22 -2
- warp/native/initializer_array.h +2 -2
- warp/native/intersect.h +22 -7
- warp/native/intersect_adj.h +8 -8
- warp/native/intersect_tri.h +13 -16
- warp/native/marching.cu +157 -161
- warp/native/mat.h +119 -19
- warp/native/matnn.h +2 -2
- warp/native/mesh.cpp +108 -83
- warp/native/mesh.cu +243 -6
- warp/native/mesh.h +1547 -458
- warp/native/nanovdb/NanoVDB.h +1 -1
- warp/native/noise.h +272 -329
- warp/native/quat.h +51 -8
- warp/native/rand.h +45 -35
- warp/native/range.h +6 -2
- warp/native/reduce.cpp +157 -0
- warp/native/reduce.cu +348 -0
- warp/native/runlength_encode.cpp +62 -0
- warp/native/runlength_encode.cu +46 -0
- warp/native/scan.cu +11 -13
- warp/native/scan.h +1 -0
- warp/native/solid_angle.h +442 -0
- warp/native/sort.cpp +13 -0
- warp/native/sort.cu +9 -1
- warp/native/sparse.cpp +338 -0
- warp/native/sparse.cu +545 -0
- warp/native/spatial.h +2 -2
- warp/native/temp_buffer.h +30 -0
- warp/native/vec.h +126 -24
- warp/native/volume.h +120 -0
- warp/native/warp.cpp +658 -53
- warp/native/warp.cu +660 -68
- warp/native/warp.h +112 -12
- warp/optim/__init__.py +1 -0
- warp/optim/linear.py +922 -0
- warp/optim/sgd.py +92 -0
- warp/render/render_opengl.py +392 -152
- warp/render/render_usd.py +11 -11
- warp/sim/__init__.py +2 -2
- warp/sim/articulation.py +385 -185
- warp/sim/collide.py +21 -8
- warp/sim/import_mjcf.py +297 -106
- warp/sim/import_urdf.py +389 -210
- warp/sim/import_usd.py +198 -97
- warp/sim/inertia.py +17 -18
- warp/sim/integrator_euler.py +14 -8
- warp/sim/integrator_xpbd.py +161 -19
- warp/sim/model.py +795 -291
- warp/sim/optimizer.py +2 -6
- warp/sim/render.py +65 -3
- warp/sim/utils.py +3 -0
- warp/sparse.py +1227 -0
- warp/stubs.py +665 -223
- warp/tape.py +66 -15
- warp/tests/__main__.py +3 -6
- warp/tests/assets/curlnoise_golden.npy +0 -0
- warp/tests/assets/pnoise_golden.npy +0 -0
- warp/tests/assets/torus.usda +105 -105
- warp/tests/{test_class_kernel.py → aux_test_class_kernel.py} +9 -1
- warp/tests/aux_test_conditional_unequal_types_kernels.py +21 -0
- warp/tests/{test_dependent.py → aux_test_dependent.py} +2 -2
- warp/tests/{test_reference.py → aux_test_reference.py} +1 -1
- warp/tests/aux_test_unresolved_func.py +14 -0
- warp/tests/aux_test_unresolved_symbol.py +14 -0
- warp/tests/disabled_kinematics.py +239 -0
- warp/tests/run_coverage_serial.py +31 -0
- warp/tests/test_adam.py +103 -106
- warp/tests/test_arithmetic.py +128 -74
- warp/tests/test_array.py +1497 -211
- warp/tests/test_array_reduce.py +150 -0
- warp/tests/test_atomic.py +64 -28
- warp/tests/test_bool.py +99 -0
- warp/tests/test_builtins_resolution.py +1292 -0
- warp/tests/test_bvh.py +75 -43
- warp/tests/test_closest_point_edge_edge.py +54 -57
- warp/tests/test_codegen.py +233 -128
- warp/tests/test_compile_consts.py +28 -20
- warp/tests/test_conditional.py +108 -24
- warp/tests/test_copy.py +10 -12
- warp/tests/test_ctypes.py +112 -88
- warp/tests/test_dense.py +21 -14
- warp/tests/test_devices.py +98 -0
- warp/tests/test_dlpack.py +136 -108
- warp/tests/test_examples.py +277 -0
- warp/tests/test_fabricarray.py +955 -0
- warp/tests/test_fast_math.py +15 -11
- warp/tests/test_fem.py +1271 -0
- warp/tests/test_fp16.py +53 -19
- warp/tests/test_func.py +187 -74
- warp/tests/test_generics.py +194 -49
- warp/tests/test_grad.py +180 -116
- warp/tests/test_grad_customs.py +176 -0
- warp/tests/test_hash_grid.py +52 -37
- warp/tests/test_import.py +10 -23
- warp/tests/test_indexedarray.py +577 -24
- warp/tests/test_intersect.py +18 -9
- warp/tests/test_large.py +141 -0
- warp/tests/test_launch.py +251 -15
- warp/tests/test_lerp.py +64 -65
- warp/tests/test_linear_solvers.py +154 -0
- warp/tests/test_lvalue.py +493 -0
- warp/tests/test_marching_cubes.py +12 -13
- warp/tests/test_mat.py +508 -2778
- warp/tests/test_mat_lite.py +115 -0
- warp/tests/test_mat_scalar_ops.py +2889 -0
- warp/tests/test_math.py +103 -9
- warp/tests/test_matmul.py +305 -69
- warp/tests/test_matmul_lite.py +410 -0
- warp/tests/test_mesh.py +71 -14
- warp/tests/test_mesh_query_aabb.py +41 -25
- warp/tests/test_mesh_query_point.py +325 -34
- warp/tests/test_mesh_query_ray.py +39 -22
- warp/tests/test_mlp.py +30 -22
- warp/tests/test_model.py +92 -89
- warp/tests/test_modules_lite.py +39 -0
- warp/tests/test_multigpu.py +88 -114
- warp/tests/test_noise.py +12 -11
- warp/tests/test_operators.py +16 -20
- warp/tests/test_options.py +11 -11
- warp/tests/test_pinned.py +17 -18
- warp/tests/test_print.py +32 -11
- warp/tests/test_quat.py +275 -129
- warp/tests/test_rand.py +18 -16
- warp/tests/test_reload.py +38 -34
- warp/tests/test_rounding.py +50 -43
- warp/tests/test_runlength_encode.py +190 -0
- warp/tests/test_smoothstep.py +9 -11
- warp/tests/test_snippet.py +143 -0
- warp/tests/test_sparse.py +460 -0
- warp/tests/test_spatial.py +276 -243
- warp/tests/test_streams.py +110 -85
- warp/tests/test_struct.py +331 -85
- warp/tests/test_tape.py +39 -21
- warp/tests/test_torch.py +118 -89
- warp/tests/test_transient_module.py +12 -13
- warp/tests/test_types.py +614 -0
- warp/tests/test_utils.py +494 -0
- warp/tests/test_vec.py +354 -1987
- warp/tests/test_vec_lite.py +73 -0
- warp/tests/test_vec_scalar_ops.py +2099 -0
- warp/tests/test_volume.py +457 -293
- warp/tests/test_volume_write.py +124 -134
- warp/tests/unittest_serial.py +35 -0
- warp/tests/unittest_suites.py +341 -0
- warp/tests/unittest_utils.py +568 -0
- warp/tests/unused_test_misc.py +71 -0
- warp/tests/{test_debug.py → walkthough_debug.py} +3 -17
- warp/thirdparty/appdirs.py +36 -45
- warp/thirdparty/unittest_parallel.py +549 -0
- warp/torch.py +72 -30
- warp/types.py +1744 -713
- warp/utils.py +360 -350
- warp_lang-0.11.0.dist-info/LICENSE.md +36 -0
- warp_lang-0.11.0.dist-info/METADATA +238 -0
- warp_lang-0.11.0.dist-info/RECORD +332 -0
- {warp_lang-0.9.0.dist-info → warp_lang-0.11.0.dist-info}/WHEEL +1 -1
- warp/bin/warp-clang.exp +0 -0
- warp/bin/warp-clang.lib +0 -0
- warp/bin/warp.exp +0 -0
- warp/bin/warp.lib +0 -0
- warp/tests/test_all.py +0 -215
- warp/tests/test_array_scan.py +0 -60
- warp/tests/test_base.py +0 -208
- warp/tests/test_unresolved_func.py +0 -7
- warp/tests/test_unresolved_symbol.py +0 -7
- warp_lang-0.9.0.dist-info/METADATA +0 -20
- warp_lang-0.9.0.dist-info/RECORD +0 -177
- /warp/tests/{test_compile_consts_dummy.py → aux_test_compile_consts_dummy.py} +0 -0
- /warp/tests/{test_reference_reference.py → aux_test_reference_reference.py} +0 -0
- /warp/tests/{test_square.py → aux_test_square.py} +0 -0
- {warp_lang-0.9.0.dist-info → warp_lang-0.11.0.dist-info}/top_level.txt +0 -0
warp/tape.py
CHANGED
|
@@ -5,11 +5,39 @@
|
|
|
5
5
|
# distribution of this software and related documentation without an express
|
|
6
6
|
# license agreement from NVIDIA CORPORATION is strictly prohibited.
|
|
7
7
|
|
|
8
|
-
import numpy as np
|
|
9
8
|
import warp as wp
|
|
10
9
|
|
|
11
10
|
|
|
12
11
|
class Tape:
|
|
12
|
+
"""
|
|
13
|
+
Record kernel launches within a Tape scope to enable automatic differentiation.
|
|
14
|
+
Gradients can be computed after the operations have been recorded on the tape via
|
|
15
|
+
``tape.backward()``.
|
|
16
|
+
|
|
17
|
+
Example
|
|
18
|
+
-------
|
|
19
|
+
|
|
20
|
+
.. code-block:: python
|
|
21
|
+
|
|
22
|
+
tape = wp.Tape()
|
|
23
|
+
|
|
24
|
+
# forward pass
|
|
25
|
+
with tape:
|
|
26
|
+
wp.launch(kernel=compute1, inputs=[a, b], device="cuda")
|
|
27
|
+
wp.launch(kernel=compute2, inputs=[c, d], device="cuda")
|
|
28
|
+
wp.launch(kernel=loss, inputs=[d, l], device="cuda")
|
|
29
|
+
|
|
30
|
+
# reverse pass
|
|
31
|
+
tape.backward(l)
|
|
32
|
+
|
|
33
|
+
Gradients can be accessed via the ``tape.gradients`` dictionary, e.g.:
|
|
34
|
+
|
|
35
|
+
.. code-block:: python
|
|
36
|
+
|
|
37
|
+
print(tape.gradients[a])
|
|
38
|
+
|
|
39
|
+
"""
|
|
40
|
+
|
|
13
41
|
def __init__(self):
|
|
14
42
|
self.gradients = {}
|
|
15
43
|
self.const_gradients = set()
|
|
@@ -37,13 +65,24 @@ class Tape:
|
|
|
37
65
|
# adj_tensor = tape.gradients[tensor]
|
|
38
66
|
#
|
|
39
67
|
def backward(self, loss: wp.array = None, grads: dict = None):
|
|
68
|
+
"""
|
|
69
|
+
Evaluate the backward pass of the recorded operations on the tape.
|
|
70
|
+
A single-element array ``loss`` or a dictionary of arrays ``grads``
|
|
71
|
+
can be provided to assign the incoming gradients for the reverse-mode
|
|
72
|
+
automatic differentiation pass.
|
|
73
|
+
|
|
74
|
+
Args:
|
|
75
|
+
loss (wp.array): A single-element array that holds the loss function value whose gradient is to be computed
|
|
76
|
+
grads (dict): A dictionary of arrays that map from Warp arrays to their incoming gradients
|
|
77
|
+
|
|
78
|
+
"""
|
|
40
79
|
# if scalar loss is specified then initialize
|
|
41
80
|
# a 'seed' array for it, with gradient of one
|
|
42
81
|
if loss:
|
|
43
82
|
if loss.size > 1 or wp.types.type_length(loss.dtype) > 1:
|
|
44
83
|
raise RuntimeError("Can only return gradients for scalar loss functions.")
|
|
45
84
|
|
|
46
|
-
if loss.requires_grad
|
|
85
|
+
if not loss.requires_grad:
|
|
47
86
|
raise RuntimeError(
|
|
48
87
|
"Scalar loss arrays should have requires_grad=True set before calling Tape.backward()"
|
|
49
88
|
)
|
|
@@ -67,9 +106,10 @@ class Tape:
|
|
|
67
106
|
else:
|
|
68
107
|
kernel = launch[0]
|
|
69
108
|
dim = launch[1]
|
|
70
|
-
|
|
71
|
-
|
|
72
|
-
|
|
109
|
+
max_blocks = launch[2]
|
|
110
|
+
inputs = launch[3]
|
|
111
|
+
outputs = launch[4]
|
|
112
|
+
device = launch[5]
|
|
73
113
|
|
|
74
114
|
adj_inputs = []
|
|
75
115
|
adj_outputs = []
|
|
@@ -91,16 +131,21 @@ class Tape:
|
|
|
91
131
|
adj_outputs=adj_outputs,
|
|
92
132
|
device=device,
|
|
93
133
|
adjoint=True,
|
|
134
|
+
max_blocks=max_blocks,
|
|
94
135
|
)
|
|
95
136
|
|
|
96
137
|
# record a kernel launch on the tape
|
|
97
|
-
def record_launch(self, kernel, dim, inputs, outputs, device):
|
|
98
|
-
self.launches.append([kernel, dim, inputs, outputs, device])
|
|
138
|
+
def record_launch(self, kernel, dim, max_blocks, inputs, outputs, device):
|
|
139
|
+
self.launches.append([kernel, dim, max_blocks, inputs, outputs, device])
|
|
99
140
|
|
|
100
|
-
# records a custom function for the backward pass, can be any
|
|
101
|
-
# Callable python object. Callee should also pass arrays that
|
|
102
|
-
# take part in the function for gradient tracking.
|
|
103
141
|
def record_func(self, backward, arrays):
|
|
142
|
+
"""
|
|
143
|
+
Records a custom function to be executed only in the backward pass.
|
|
144
|
+
|
|
145
|
+
Args:
|
|
146
|
+
backward (Callable): A callable Python object (can be any function) that will be executed in the backward pass.
|
|
147
|
+
arrays (list): A list of arrays that are used by the function for gradient tracking.
|
|
148
|
+
"""
|
|
104
149
|
self.launches.append(backward)
|
|
105
150
|
|
|
106
151
|
for a in arrays:
|
|
@@ -125,11 +170,11 @@ class Tape:
|
|
|
125
170
|
return a.grad
|
|
126
171
|
|
|
127
172
|
elif isinstance(a, wp.codegen.StructInstance):
|
|
128
|
-
adj = a.
|
|
129
|
-
for name, _ in a.
|
|
173
|
+
adj = a._cls()
|
|
174
|
+
for name, _ in a._cls.ctype._fields_:
|
|
130
175
|
if name.startswith("_"):
|
|
131
176
|
continue
|
|
132
|
-
if isinstance(a.
|
|
177
|
+
if isinstance(a._cls.vars[name].type, wp.array):
|
|
133
178
|
arr = getattr(a, name)
|
|
134
179
|
if arr.grad:
|
|
135
180
|
grad = self.gradients[arr] = arr.grad
|
|
@@ -145,15 +190,21 @@ class Tape:
|
|
|
145
190
|
return None
|
|
146
191
|
|
|
147
192
|
def reset(self):
|
|
193
|
+
"""
|
|
194
|
+
Clear all operations recorded on the tape and zero out all gradients.
|
|
195
|
+
"""
|
|
148
196
|
self.launches = []
|
|
149
197
|
self.zero()
|
|
150
198
|
|
|
151
199
|
def zero(self):
|
|
200
|
+
"""
|
|
201
|
+
Zero out all gradients recorded on the tape.
|
|
202
|
+
"""
|
|
152
203
|
for a, g in self.gradients.items():
|
|
153
204
|
if a not in self.const_gradients:
|
|
154
205
|
if isinstance(a, wp.codegen.StructInstance):
|
|
155
|
-
for name in g.
|
|
156
|
-
if isinstance(g.
|
|
206
|
+
for name in g._cls.vars:
|
|
207
|
+
if isinstance(g._cls.vars[name].type, wp.array) and g._cls.vars[name].requires_grad:
|
|
157
208
|
getattr(g, name).zero_()
|
|
158
209
|
else:
|
|
159
210
|
g.zero_()
|
warp/tests/__main__.py
CHANGED
|
Binary file
|
|
Binary file
|