warp-lang 0.10.1__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.

Files changed (300) hide show
  1. warp/__init__.py +10 -4
  2. warp/__init__.pyi +1 -0
  3. warp/bin/warp-clang.dll +0 -0
  4. warp/bin/warp.dll +0 -0
  5. warp/build.py +5 -3
  6. warp/build_dll.py +29 -9
  7. warp/builtins.py +868 -507
  8. warp/codegen.py +1074 -638
  9. warp/config.py +3 -3
  10. warp/constants.py +6 -0
  11. warp/context.py +715 -222
  12. warp/fabric.py +326 -0
  13. warp/fem/__init__.py +27 -0
  14. warp/fem/cache.py +389 -0
  15. warp/fem/dirichlet.py +181 -0
  16. warp/fem/domain.py +263 -0
  17. warp/fem/field/__init__.py +101 -0
  18. warp/fem/field/field.py +149 -0
  19. warp/fem/field/nodal_field.py +299 -0
  20. warp/fem/field/restriction.py +21 -0
  21. warp/fem/field/test.py +181 -0
  22. warp/fem/field/trial.py +183 -0
  23. warp/fem/geometry/__init__.py +19 -0
  24. warp/fem/geometry/closest_point.py +70 -0
  25. warp/fem/geometry/deformed_geometry.py +271 -0
  26. warp/fem/geometry/element.py +744 -0
  27. warp/fem/geometry/geometry.py +186 -0
  28. warp/fem/geometry/grid_2d.py +373 -0
  29. warp/fem/geometry/grid_3d.py +435 -0
  30. warp/fem/geometry/hexmesh.py +953 -0
  31. warp/fem/geometry/partition.py +376 -0
  32. warp/fem/geometry/quadmesh_2d.py +532 -0
  33. warp/fem/geometry/tetmesh.py +840 -0
  34. warp/fem/geometry/trimesh_2d.py +577 -0
  35. warp/fem/integrate.py +1616 -0
  36. warp/fem/operator.py +191 -0
  37. warp/fem/polynomial.py +213 -0
  38. warp/fem/quadrature/__init__.py +2 -0
  39. warp/fem/quadrature/pic_quadrature.py +245 -0
  40. warp/fem/quadrature/quadrature.py +294 -0
  41. warp/fem/space/__init__.py +292 -0
  42. warp/fem/space/basis_space.py +489 -0
  43. warp/fem/space/collocated_function_space.py +105 -0
  44. warp/fem/space/dof_mapper.py +236 -0
  45. warp/fem/space/function_space.py +145 -0
  46. warp/fem/space/grid_2d_function_space.py +267 -0
  47. warp/fem/space/grid_3d_function_space.py +306 -0
  48. warp/fem/space/hexmesh_function_space.py +352 -0
  49. warp/fem/space/partition.py +350 -0
  50. warp/fem/space/quadmesh_2d_function_space.py +369 -0
  51. warp/fem/space/restriction.py +160 -0
  52. warp/fem/space/shape/__init__.py +15 -0
  53. warp/fem/space/shape/cube_shape_function.py +738 -0
  54. warp/fem/space/shape/shape_function.py +103 -0
  55. warp/fem/space/shape/square_shape_function.py +611 -0
  56. warp/fem/space/shape/tet_shape_function.py +567 -0
  57. warp/fem/space/shape/triangle_shape_function.py +429 -0
  58. warp/fem/space/tetmesh_function_space.py +292 -0
  59. warp/fem/space/topology.py +295 -0
  60. warp/fem/space/trimesh_2d_function_space.py +221 -0
  61. warp/fem/types.py +77 -0
  62. warp/fem/utils.py +495 -0
  63. warp/native/array.h +147 -44
  64. warp/native/builtin.h +122 -149
  65. warp/native/bvh.cpp +73 -325
  66. warp/native/bvh.cu +406 -23
  67. warp/native/bvh.h +34 -43
  68. warp/native/clang/clang.cpp +13 -8
  69. warp/native/crt.h +2 -0
  70. warp/native/cuda_crt.h +5 -0
  71. warp/native/cuda_util.cpp +15 -3
  72. warp/native/cuda_util.h +3 -1
  73. warp/native/cutlass/tools/library/scripts/conv2d_operation.py +463 -0
  74. warp/native/cutlass/tools/library/scripts/conv3d_operation.py +321 -0
  75. warp/native/cutlass/tools/library/scripts/gemm_operation.py +988 -0
  76. warp/native/cutlass/tools/library/scripts/generator.py +4625 -0
  77. warp/native/cutlass/tools/library/scripts/library.py +799 -0
  78. warp/native/cutlass/tools/library/scripts/manifest.py +402 -0
  79. warp/native/cutlass/tools/library/scripts/pycutlass/docs/source/conf.py +96 -0
  80. warp/native/cutlass/tools/library/scripts/pycutlass/profile/conv/conv2d_f16_sm80.py +106 -0
  81. warp/native/cutlass/tools/library/scripts/pycutlass/profile/gemm/gemm_f32_sm80.py +91 -0
  82. warp/native/cutlass/tools/library/scripts/pycutlass/setup.py +80 -0
  83. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/__init__.py +48 -0
  84. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/arguments.py +118 -0
  85. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/c_types.py +241 -0
  86. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/compiler.py +432 -0
  87. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/conv2d_operation.py +631 -0
  88. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/epilogue.py +1026 -0
  89. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/frontend.py +104 -0
  90. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/gemm_operation.py +1276 -0
  91. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/library.py +744 -0
  92. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/memory_manager.py +74 -0
  93. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/operation.py +110 -0
  94. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/parser.py +619 -0
  95. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/reduction_operation.py +398 -0
  96. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/tensor_ref.py +70 -0
  97. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/__init__.py +4 -0
  98. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/conv2d_testbed.py +646 -0
  99. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/gemm_grouped_testbed.py +235 -0
  100. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/gemm_testbed.py +557 -0
  101. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/test/profiler.py +70 -0
  102. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/type_hint.py +39 -0
  103. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/utils/__init__.py +1 -0
  104. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/utils/device.py +76 -0
  105. warp/native/cutlass/tools/library/scripts/pycutlass/src/pycutlass/utils/reference_model.py +255 -0
  106. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/__init__.py +0 -0
  107. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py +201 -0
  108. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +177 -0
  109. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py +98 -0
  110. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_dgrad_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py +95 -0
  111. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_few_channels_f16nhwc_f16nhwc_f16nhwc_tensor_op_f32_sm80.py +163 -0
  112. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_fixed_channels_f16nhwc_f16nhwc_f16nhwc_tensor_op_f32_sm80.py +187 -0
  113. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py +309 -0
  114. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +54 -0
  115. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py +96 -0
  116. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_fprop_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py +107 -0
  117. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_strided_dgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +253 -0
  118. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f16nhwc_f16nhwc_f16nhwc_tensor_op_f16_sm80.py +97 -0
  119. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f16nhwc_f16nhwc_f32nhwc_tensor_op_f32_sm80.py +242 -0
  120. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_f32nhwc_f32nhwc_f32nhwc_simt_f32_sm80.py +96 -0
  121. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/conv2d_wgrad_implicit_gemm_tf32nhwc_tf32nhwc_f32nhwc_tensor_op_f32_sm80.py +107 -0
  122. warp/native/cutlass/tools/library/scripts/pycutlass/test/conv/run_all_tests.py +10 -0
  123. warp/native/cutlass/tools/library/scripts/pycutlass/test/frontend/test_frontend.py +146 -0
  124. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/__init__.py +0 -0
  125. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_bf16_sm80.py +96 -0
  126. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_f16_sm80.py +447 -0
  127. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_f32_sm80.py +146 -0
  128. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_f64_sm80.py +102 -0
  129. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_grouped_sm80.py +203 -0
  130. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/gemm_s8_sm80.py +229 -0
  131. warp/native/cutlass/tools/library/scripts/pycutlass/test/gemm/run_all_tests.py +9 -0
  132. warp/native/cutlass/tools/library/scripts/pycutlass/test/unit/test_sm80.py +453 -0
  133. warp/native/cutlass/tools/library/scripts/rank_2k_operation.py +398 -0
  134. warp/native/cutlass/tools/library/scripts/rank_k_operation.py +387 -0
  135. warp/native/cutlass/tools/library/scripts/rt.py +796 -0
  136. warp/native/cutlass/tools/library/scripts/symm_operation.py +400 -0
  137. warp/native/cutlass/tools/library/scripts/trmm_operation.py +407 -0
  138. warp/native/cutlass_gemm.cu +5 -3
  139. warp/native/exports.h +1240 -952
  140. warp/native/fabric.h +228 -0
  141. warp/native/hashgrid.cpp +4 -4
  142. warp/native/hashgrid.h +22 -2
  143. warp/native/intersect.h +22 -7
  144. warp/native/intersect_adj.h +8 -8
  145. warp/native/intersect_tri.h +1 -1
  146. warp/native/marching.cu +157 -161
  147. warp/native/mat.h +80 -19
  148. warp/native/matnn.h +2 -2
  149. warp/native/mesh.cpp +33 -108
  150. warp/native/mesh.cu +114 -23
  151. warp/native/mesh.h +446 -46
  152. warp/native/noise.h +272 -329
  153. warp/native/quat.h +51 -8
  154. warp/native/rand.h +45 -35
  155. warp/native/range.h +6 -2
  156. warp/native/reduce.cpp +1 -1
  157. warp/native/reduce.cu +10 -12
  158. warp/native/runlength_encode.cu +6 -10
  159. warp/native/scan.cu +8 -11
  160. warp/native/sparse.cpp +4 -4
  161. warp/native/sparse.cu +164 -154
  162. warp/native/spatial.h +2 -2
  163. warp/native/temp_buffer.h +14 -30
  164. warp/native/vec.h +107 -23
  165. warp/native/volume.h +120 -0
  166. warp/native/warp.cpp +560 -30
  167. warp/native/warp.cu +431 -44
  168. warp/native/warp.h +13 -4
  169. warp/optim/__init__.py +1 -0
  170. warp/optim/linear.py +922 -0
  171. warp/optim/sgd.py +92 -0
  172. warp/render/render_opengl.py +335 -119
  173. warp/render/render_usd.py +11 -11
  174. warp/sim/__init__.py +2 -2
  175. warp/sim/articulation.py +385 -185
  176. warp/sim/collide.py +8 -0
  177. warp/sim/import_mjcf.py +297 -106
  178. warp/sim/import_urdf.py +389 -210
  179. warp/sim/import_usd.py +198 -97
  180. warp/sim/inertia.py +17 -18
  181. warp/sim/integrator_euler.py +14 -8
  182. warp/sim/integrator_xpbd.py +158 -16
  183. warp/sim/model.py +795 -291
  184. warp/sim/render.py +3 -3
  185. warp/sim/utils.py +3 -0
  186. warp/sparse.py +640 -150
  187. warp/stubs.py +606 -267
  188. warp/tape.py +61 -10
  189. warp/tests/__main__.py +3 -6
  190. warp/tests/assets/curlnoise_golden.npy +0 -0
  191. warp/tests/assets/pnoise_golden.npy +0 -0
  192. warp/tests/{test_class_kernel.py → aux_test_class_kernel.py} +9 -1
  193. warp/tests/aux_test_conditional_unequal_types_kernels.py +21 -0
  194. warp/tests/{test_dependent.py → aux_test_dependent.py} +2 -2
  195. warp/tests/{test_reference.py → aux_test_reference.py} +1 -1
  196. warp/tests/aux_test_unresolved_func.py +14 -0
  197. warp/tests/aux_test_unresolved_symbol.py +14 -0
  198. warp/tests/disabled_kinematics.py +239 -0
  199. warp/tests/run_coverage_serial.py +31 -0
  200. warp/tests/test_adam.py +103 -106
  201. warp/tests/test_arithmetic.py +128 -74
  202. warp/tests/test_array.py +212 -97
  203. warp/tests/test_array_reduce.py +57 -23
  204. warp/tests/test_atomic.py +64 -28
  205. warp/tests/test_bool.py +99 -0
  206. warp/tests/test_builtins_resolution.py +1292 -0
  207. warp/tests/test_bvh.py +42 -18
  208. warp/tests/test_closest_point_edge_edge.py +54 -57
  209. warp/tests/test_codegen.py +208 -130
  210. warp/tests/test_compile_consts.py +28 -20
  211. warp/tests/test_conditional.py +108 -24
  212. warp/tests/test_copy.py +10 -12
  213. warp/tests/test_ctypes.py +112 -88
  214. warp/tests/test_dense.py +21 -14
  215. warp/tests/test_devices.py +98 -0
  216. warp/tests/test_dlpack.py +75 -75
  217. warp/tests/test_examples.py +277 -0
  218. warp/tests/test_fabricarray.py +955 -0
  219. warp/tests/test_fast_math.py +15 -11
  220. warp/tests/test_fem.py +1271 -0
  221. warp/tests/test_fp16.py +53 -19
  222. warp/tests/test_func.py +187 -86
  223. warp/tests/test_generics.py +194 -49
  224. warp/tests/test_grad.py +178 -109
  225. warp/tests/test_grad_customs.py +176 -0
  226. warp/tests/test_hash_grid.py +52 -37
  227. warp/tests/test_import.py +10 -23
  228. warp/tests/test_indexedarray.py +32 -31
  229. warp/tests/test_intersect.py +18 -9
  230. warp/tests/test_large.py +141 -0
  231. warp/tests/test_launch.py +14 -41
  232. warp/tests/test_lerp.py +64 -65
  233. warp/tests/test_linear_solvers.py +154 -0
  234. warp/tests/test_lvalue.py +493 -0
  235. warp/tests/test_marching_cubes.py +12 -13
  236. warp/tests/test_mat.py +517 -2898
  237. warp/tests/test_mat_lite.py +115 -0
  238. warp/tests/test_mat_scalar_ops.py +2889 -0
  239. warp/tests/test_math.py +103 -9
  240. warp/tests/test_matmul.py +305 -69
  241. warp/tests/test_matmul_lite.py +410 -0
  242. warp/tests/test_mesh.py +71 -14
  243. warp/tests/test_mesh_query_aabb.py +41 -25
  244. warp/tests/test_mesh_query_point.py +140 -22
  245. warp/tests/test_mesh_query_ray.py +39 -22
  246. warp/tests/test_mlp.py +30 -22
  247. warp/tests/test_model.py +92 -89
  248. warp/tests/test_modules_lite.py +39 -0
  249. warp/tests/test_multigpu.py +88 -114
  250. warp/tests/test_noise.py +12 -11
  251. warp/tests/test_operators.py +16 -20
  252. warp/tests/test_options.py +11 -11
  253. warp/tests/test_pinned.py +17 -18
  254. warp/tests/test_print.py +32 -11
  255. warp/tests/test_quat.py +275 -129
  256. warp/tests/test_rand.py +18 -16
  257. warp/tests/test_reload.py +38 -34
  258. warp/tests/test_rounding.py +50 -43
  259. warp/tests/test_runlength_encode.py +168 -20
  260. warp/tests/test_smoothstep.py +9 -11
  261. warp/tests/test_snippet.py +143 -0
  262. warp/tests/test_sparse.py +261 -63
  263. warp/tests/test_spatial.py +276 -243
  264. warp/tests/test_streams.py +110 -85
  265. warp/tests/test_struct.py +268 -63
  266. warp/tests/test_tape.py +39 -21
  267. warp/tests/test_torch.py +118 -89
  268. warp/tests/test_transient_module.py +12 -13
  269. warp/tests/test_types.py +614 -0
  270. warp/tests/test_utils.py +494 -0
  271. warp/tests/test_vec.py +354 -2050
  272. warp/tests/test_vec_lite.py +73 -0
  273. warp/tests/test_vec_scalar_ops.py +2099 -0
  274. warp/tests/test_volume.py +457 -293
  275. warp/tests/test_volume_write.py +124 -134
  276. warp/tests/unittest_serial.py +35 -0
  277. warp/tests/unittest_suites.py +341 -0
  278. warp/tests/unittest_utils.py +568 -0
  279. warp/tests/unused_test_misc.py +71 -0
  280. warp/tests/{test_debug.py → walkthough_debug.py} +3 -17
  281. warp/thirdparty/appdirs.py +36 -45
  282. warp/thirdparty/unittest_parallel.py +549 -0
  283. warp/torch.py +9 -6
  284. warp/types.py +1089 -366
  285. warp/utils.py +93 -387
  286. warp_lang-0.11.0.dist-info/METADATA +238 -0
  287. warp_lang-0.11.0.dist-info/RECORD +332 -0
  288. {warp_lang-0.10.1.dist-info → warp_lang-0.11.0.dist-info}/WHEEL +1 -1
  289. warp/tests/test_all.py +0 -219
  290. warp/tests/test_array_scan.py +0 -60
  291. warp/tests/test_base.py +0 -208
  292. warp/tests/test_unresolved_func.py +0 -7
  293. warp/tests/test_unresolved_symbol.py +0 -7
  294. warp_lang-0.10.1.dist-info/METADATA +0 -21
  295. warp_lang-0.10.1.dist-info/RECORD +0 -188
  296. /warp/tests/{test_compile_consts_dummy.py → aux_test_compile_consts_dummy.py} +0 -0
  297. /warp/tests/{test_reference_reference.py → aux_test_reference_reference.py} +0 -0
  298. /warp/tests/{test_square.py → aux_test_square.py} +0 -0
  299. {warp_lang-0.10.1.dist-info → warp_lang-0.11.0.dist-info}/LICENSE.md +0 -0
  300. {warp_lang-0.10.1.dist-info → warp_lang-0.11.0.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,955 @@
1
+ # Copyright (c) 2023 NVIDIA CORPORATION. All rights reserved.
2
+ # NVIDIA CORPORATION and its licensors retain all intellectual property
3
+ # and proprietary rights in and to this software, related documentation
4
+ # and any modifications thereto. Any use, reproduction, disclosure or
5
+ # distribution of this software and related documentation without an express
6
+ # license agreement from NVIDIA CORPORATION is strictly prohibited.
7
+
8
+ import math
9
+ import unittest
10
+ from typing import Any
11
+
12
+ import numpy as np
13
+
14
+ import warp as wp
15
+ from warp.tests.unittest_utils import *
16
+
17
+ wp.init()
18
+
19
+
20
+ # types to test fabric arrays
21
+ _fabric_types = [
22
+ *wp.types.scalar_types,
23
+ *[wp.types.vector(2, T) for T in wp.types.scalar_types],
24
+ *[wp.types.vector(3, T) for T in wp.types.scalar_types],
25
+ *[wp.types.vector(4, T) for T in wp.types.scalar_types],
26
+ *[wp.types.matrix((2, 2), T) for T in wp.types.scalar_types],
27
+ *[wp.types.matrix((3, 3), T) for T in wp.types.scalar_types],
28
+ *[wp.types.matrix((4, 4), T) for T in wp.types.scalar_types],
29
+ *[wp.types.quaternion(T) for T in wp.types.float_types],
30
+ ]
31
+
32
+
33
+ def _warp_type_to_fabric(dtype, is_array=False):
34
+ scalar_map = {
35
+ wp.bool: "b",
36
+ wp.int8: "i1",
37
+ wp.int16: "i2",
38
+ wp.int32: "i4",
39
+ wp.int64: "i8",
40
+ wp.uint8: "u1",
41
+ wp.uint16: "u2",
42
+ wp.uint32: "u4",
43
+ wp.uint64: "u8",
44
+ wp.float16: "f2",
45
+ wp.float32: "f4",
46
+ wp.float64: "f8",
47
+ }
48
+
49
+ if hasattr(dtype, "_wp_scalar_type_"):
50
+ type_str = scalar_map[dtype._wp_scalar_type_]
51
+ if len(dtype._shape_) == 1:
52
+ role = "vector"
53
+ else:
54
+ role = "matrix"
55
+ else:
56
+ type_str = scalar_map[dtype]
57
+ role = ""
58
+
59
+ if is_array:
60
+ array_depth = 1
61
+ else:
62
+ array_depth = 0
63
+
64
+ return (True, type_str, dtype._length_, array_depth, role)
65
+
66
+
67
+ # returns a fabric array interface constructed from a regular array
68
+ def _create_fabric_array_interface(data: wp.array, attrib: str, bucket_sizes: list = None, copy=False):
69
+ assert isinstance(data, wp.array)
70
+ assert data.ndim == 1
71
+
72
+ assert isinstance(attrib, str)
73
+
74
+ if copy:
75
+ data = wp.clone(data)
76
+
77
+ if bucket_sizes is not None:
78
+ assert hasattr(bucket_sizes, "__len__")
79
+
80
+ # verify total size
81
+ total_size = 0
82
+ for bucket_size in bucket_sizes:
83
+ total_size += bucket_size
84
+
85
+ if total_size != data.size:
86
+ raise RuntimeError("Bucket sizes don't add up to the size of data array")
87
+
88
+ elif data.size > 0:
89
+ rng = np.random.default_rng(123)
90
+
91
+ # generate random bucket sizes
92
+ bucket_min = 1
93
+ bucket_max = math.ceil(0.5 * data.size)
94
+ total_size = data.size
95
+ size_remaining = total_size
96
+
97
+ bucket_sizes = []
98
+ while size_remaining >= bucket_max:
99
+ bucket_size = rng.integers(bucket_min, high=bucket_max, dtype=int)
100
+ bucket_sizes.append(bucket_size)
101
+ size_remaining -= bucket_size
102
+
103
+ if size_remaining > 0:
104
+ bucket_sizes.append(size_remaining)
105
+
106
+ else:
107
+ # empty data array
108
+ bucket_sizes = []
109
+
110
+ dtype_size = wp.types.type_size_in_bytes(data.dtype)
111
+ p = int(data.ptr) if data.ptr else 0
112
+ pointers = []
113
+ counts = []
114
+ for bucket_size in bucket_sizes:
115
+ pointers.append(p)
116
+ counts.append(bucket_size)
117
+ p += bucket_size * dtype_size
118
+
119
+ attrib_info = {}
120
+
121
+ attrib_info["type"] = _warp_type_to_fabric(data.dtype)
122
+ attrib_info["access"] = 2 # ReadWrite
123
+ attrib_info["pointers"] = pointers
124
+ attrib_info["counts"] = counts
125
+
126
+ iface = {}
127
+ iface["version"] = 1
128
+ iface["device"] = str(data.device)
129
+ iface["attribs"] = {attrib: attrib_info}
130
+ iface["_ref"] = data # backref to keep the array alive
131
+
132
+ return iface
133
+
134
+
135
+ # returns a fabric array array interface constructed from a list of regular arrays
136
+ def _create_fabric_array_array_interface(data: list, attrib: str, bucket_sizes: list = None):
137
+ # data should be a list of arrays
138
+ assert isinstance(data, list)
139
+
140
+ num_arrays = len(data)
141
+ assert num_arrays > 0
142
+
143
+ device = data[0].device
144
+ dtype = data[0].dtype
145
+
146
+ assert isinstance(attrib, str)
147
+
148
+ if bucket_sizes is not None:
149
+ assert hasattr(bucket_sizes, "__len__")
150
+
151
+ # verify total size
152
+ total_size = 0
153
+ for bucket_size in bucket_sizes:
154
+ total_size += bucket_size
155
+
156
+ if total_size != num_arrays:
157
+ raise RuntimeError("Bucket sizes don't add up to the number of given arrays")
158
+
159
+ else:
160
+ rng = np.random.default_rng(123)
161
+
162
+ # generate random bucket sizes
163
+ bucket_min = 1
164
+ bucket_max = math.ceil(0.5 * num_arrays)
165
+ total_size = num_arrays
166
+ size_remaining = total_size
167
+
168
+ bucket_sizes = []
169
+ while size_remaining >= bucket_max:
170
+ bucket_size = rng.integers(bucket_min, high=bucket_max, dtype=int)
171
+ bucket_sizes.append(bucket_size)
172
+ size_remaining -= bucket_size
173
+
174
+ if size_remaining > 0:
175
+ bucket_sizes.append(size_remaining)
176
+
177
+ # initialize array of pointers to arrays and their lengths
178
+ _array_pointers = []
179
+ _array_lengths = []
180
+ for i in range(num_arrays):
181
+ _array_pointers.append(data[i].ptr)
182
+ _array_lengths.append(data[i].size)
183
+
184
+ array_pointers = wp.array(_array_pointers, dtype=wp.uint64, device=device)
185
+ pointer_size = wp.types.type_size_in_bytes(array_pointers.dtype)
186
+
187
+ lengths = wp.array(_array_lengths, dtype=wp.uint64, device=device)
188
+ length_size = wp.types.type_size_in_bytes(lengths.dtype)
189
+
190
+ p_pointers = int(array_pointers.ptr)
191
+ p_lengths = int(lengths.ptr)
192
+ pointers = []
193
+ counts = []
194
+ array_lengths = []
195
+ for bucket_size in bucket_sizes:
196
+ pointers.append(p_pointers)
197
+ counts.append(bucket_size)
198
+ array_lengths.append(p_lengths)
199
+ p_pointers += bucket_size * pointer_size
200
+ p_lengths += bucket_size * length_size
201
+
202
+ attrib_info = {}
203
+
204
+ attrib_info["type"] = _warp_type_to_fabric(dtype, is_array=True)
205
+ attrib_info["access"] = 2 # ReadWrite
206
+ attrib_info["pointers"] = pointers
207
+ attrib_info["counts"] = counts
208
+ attrib_info["array_lengths"] = array_lengths
209
+
210
+ iface = {}
211
+ iface["version"] = 1
212
+ iface["device"] = str(device)
213
+ iface["attribs"] = {attrib: attrib_info}
214
+ iface["_ref"] = data # backref to keep the data arrays alive
215
+ iface["_ref_pointers"] = array_pointers # backref to keep the array pointers alive
216
+ iface["_ref_lengths"] = lengths # backref to keep the lengths array alive
217
+
218
+ return iface
219
+
220
+
221
+ @wp.kernel
222
+ def fa_kernel(a: wp.fabricarray(dtype=float), expected: wp.array(dtype=float)):
223
+ i = wp.tid()
224
+
225
+ wp.expect_eq(a[i], expected[i])
226
+
227
+ a[i] = 2.0 * a[i]
228
+
229
+ wp.atomic_add(a, i, 1.0)
230
+
231
+ wp.expect_eq(a[i], 2.0 * expected[i] + 1.0)
232
+
233
+
234
+ @wp.kernel
235
+ def fa_kernel_indexed(a: wp.indexedfabricarray(dtype=float), expected: wp.indexedarray(dtype=float)):
236
+ i = wp.tid()
237
+
238
+ wp.expect_eq(a[i], expected[i])
239
+
240
+ a[i] = 2.0 * a[i]
241
+
242
+ wp.atomic_add(a, i, 1.0)
243
+
244
+ wp.expect_eq(a[i], 2.0 * expected[i] + 1.0)
245
+
246
+
247
+ def test_fabricarray_kernel(test, device):
248
+ data = wp.array(data=np.arange(100, dtype=np.float32), device=device)
249
+ iface = _create_fabric_array_interface(data, "foo", copy=True)
250
+ fa = wp.fabricarray(data=iface, attrib="foo")
251
+
252
+ test.assertEqual(fa.dtype, data.dtype)
253
+ test.assertEqual(fa.ndim, 1)
254
+ test.assertEqual(fa.shape, data.shape)
255
+ test.assertEqual(fa.size, data.size)
256
+
257
+ wp.launch(fa_kernel, dim=fa.size, inputs=[fa, data], device=device)
258
+
259
+ # reset data
260
+ wp.copy(fa, data)
261
+
262
+ # test indexed
263
+ indices = wp.array(data=np.arange(1, data.size, 2, dtype=np.int32), device=device)
264
+ ifa = fa[indices]
265
+ idata = data[indices]
266
+
267
+ test.assertEqual(ifa.dtype, idata.dtype)
268
+ test.assertEqual(ifa.ndim, 1)
269
+ test.assertEqual(ifa.shape, idata.shape)
270
+ test.assertEqual(ifa.size, idata.size)
271
+
272
+ wp.launch(fa_kernel_indexed, dim=ifa.size, inputs=[ifa, idata], device=device)
273
+
274
+ wp.synchronize_device(device)
275
+
276
+
277
+ @wp.kernel
278
+ def fa_generic_dtype_kernel(a: wp.fabricarray(dtype=Any), b: wp.fabricarray(dtype=Any)):
279
+ i = wp.tid()
280
+ b[i] = a[i] + a[i]
281
+
282
+
283
+ @wp.kernel
284
+ def fa_generic_dtype_kernel_indexed(a: wp.indexedfabricarray(dtype=Any), b: wp.indexedfabricarray(dtype=Any)):
285
+ i = wp.tid()
286
+ b[i] = a[i] + a[i]
287
+
288
+
289
+ def test_fabricarray_generic_dtype(test, device):
290
+ for T in _fabric_types:
291
+ if hasattr(T, "_wp_scalar_type_"):
292
+ nptype = wp.types.warp_type_to_np_dtype[T._wp_scalar_type_]
293
+ else:
294
+ nptype = wp.types.warp_type_to_np_dtype[T]
295
+
296
+ data = wp.array(data=np.arange(10, dtype=nptype), device=device)
297
+ data_iface = _create_fabric_array_interface(data, "foo", copy=True)
298
+ fa = wp.fabricarray(data=data_iface, attrib="foo")
299
+
300
+ result = wp.zeros_like(data)
301
+ result_iface = _create_fabric_array_interface(result, "foo", copy=True)
302
+ fb = wp.fabricarray(data=result_iface, attrib="foo")
303
+
304
+ test.assertEqual(fa.dtype, fb.dtype)
305
+ test.assertEqual(fa.ndim, fb.ndim)
306
+ test.assertEqual(fa.shape, fb.shape)
307
+ test.assertEqual(fa.size, fb.size)
308
+
309
+ wp.launch(fa_generic_dtype_kernel, dim=fa.size, inputs=[fa, fb], device=device)
310
+
311
+ assert_np_equal(fb.numpy(), 2 * fa.numpy())
312
+
313
+ # reset data
314
+ wp.copy(fa, data)
315
+ wp.copy(fb, result)
316
+
317
+ # test indexed
318
+ indices = wp.array(data=np.arange(1, data.size, 2, dtype=np.int32), device=device)
319
+ ifa = fa[indices]
320
+ ifb = fb[indices]
321
+
322
+ test.assertEqual(ifa.dtype, ifb.dtype)
323
+ test.assertEqual(ifa.ndim, ifb.ndim)
324
+ test.assertEqual(ifa.shape, ifb.shape)
325
+ test.assertEqual(ifa.size, ifb.size)
326
+
327
+ wp.launch(fa_generic_dtype_kernel_indexed, dim=ifa.size, inputs=[ifa, ifb], device=device)
328
+
329
+ assert_np_equal(ifb.numpy(), 2 * ifa.numpy())
330
+
331
+
332
+ @wp.kernel
333
+ def fa_generic_array_kernel(a: Any, b: Any):
334
+ i = wp.tid()
335
+ b[i] = a[i] + a[i]
336
+
337
+
338
+ def test_fabricarray_generic_array(test, device):
339
+ for T in _fabric_types:
340
+ if hasattr(T, "_wp_scalar_type_"):
341
+ nptype = wp.types.warp_type_to_np_dtype[T._wp_scalar_type_]
342
+ else:
343
+ nptype = wp.types.warp_type_to_np_dtype[T]
344
+
345
+ data = wp.array(data=np.arange(100, dtype=nptype), device=device)
346
+ data_iface = _create_fabric_array_interface(data, "foo", copy=True)
347
+ fa = wp.fabricarray(data=data_iface, attrib="foo")
348
+
349
+ result = wp.zeros_like(data)
350
+ result_iface = _create_fabric_array_interface(result, "foo", copy=True)
351
+ fb = wp.fabricarray(data=result_iface, attrib="foo")
352
+
353
+ test.assertEqual(fa.dtype, fb.dtype)
354
+ test.assertEqual(fa.ndim, fb.ndim)
355
+ test.assertEqual(fa.shape, fb.shape)
356
+ test.assertEqual(fa.size, fb.size)
357
+
358
+ wp.launch(fa_generic_array_kernel, dim=fa.size, inputs=[fa, fb], device=device)
359
+
360
+ assert_np_equal(fb.numpy(), 2 * fa.numpy())
361
+
362
+ # reset data
363
+ wp.copy(fa, data)
364
+ wp.copy(fb, result)
365
+
366
+ # test indexed
367
+ indices = wp.array(data=np.arange(1, data.size, 2, dtype=np.int32), device=device)
368
+ ifa = fa[indices]
369
+ ifb = fb[indices]
370
+
371
+ test.assertEqual(ifa.dtype, ifb.dtype)
372
+ test.assertEqual(ifa.ndim, ifb.ndim)
373
+ test.assertEqual(ifa.shape, ifb.shape)
374
+ test.assertEqual(ifa.size, ifb.size)
375
+
376
+ wp.launch(fa_generic_array_kernel, dim=ifa.size, inputs=[ifa, ifb], device=device)
377
+
378
+ assert_np_equal(ifb.numpy(), 2 * ifa.numpy())
379
+
380
+
381
+ def test_fabricarray_empty(test, device):
382
+ # Test whether common operations work with empty (zero-sized) indexed arrays
383
+ # without throwing exceptions.
384
+
385
+ def test_empty_ops(nrows, ncols, wptype, nptype):
386
+ # scalar, vector, or matrix
387
+ if ncols > 0:
388
+ if nrows > 0:
389
+ wptype = wp.types.matrix((nrows, ncols), wptype)
390
+ else:
391
+ wptype = wp.types.vector(ncols, wptype)
392
+ dtype_shape = wptype._shape_
393
+ else:
394
+ dtype_shape = ()
395
+
396
+ fill_value = wptype(42)
397
+
398
+ # create an empty data array
399
+ data = wp.empty(0, dtype=wptype, device=device)
400
+ iface = _create_fabric_array_interface(data, "foo", copy=True)
401
+ fa = wp.fabricarray(data=iface, attrib="foo")
402
+
403
+ test.assertEqual(fa.size, 0)
404
+ test.assertEqual(fa.shape, (0,))
405
+
406
+ # all of these methods should succeed with zero-sized arrays
407
+ fa.zero_()
408
+ fa.fill_(fill_value)
409
+ fb = fa.contiguous()
410
+
411
+ fb = wp.empty_like(fa)
412
+ fb = wp.zeros_like(fa)
413
+ fb = wp.full_like(fa, fill_value)
414
+ fb = wp.clone(fa)
415
+
416
+ wp.copy(fa, fb)
417
+ fa.assign(fb)
418
+
419
+ na = fa.numpy()
420
+ test.assertEqual(na.size, 0)
421
+ test.assertEqual(na.shape, (0, *dtype_shape))
422
+ test.assertEqual(na.dtype, nptype)
423
+
424
+ test.assertEqual(fa.list(), [])
425
+
426
+ # test indexed
427
+
428
+ # create a zero-sized array of indices
429
+ indices = wp.empty(0, dtype=int, device=device)
430
+
431
+ ifa = fa[indices]
432
+
433
+ test.assertEqual(ifa.size, 0)
434
+ test.assertEqual(ifa.shape, (0,))
435
+
436
+ # all of these methods should succeed with zero-sized arrays
437
+ ifa.zero_()
438
+ ifa.fill_(fill_value)
439
+ ifb = ifa.contiguous()
440
+
441
+ ifb = wp.empty_like(ifa)
442
+ ifb = wp.zeros_like(ifa)
443
+ ifb = wp.full_like(ifa, fill_value)
444
+ ifb = wp.clone(ifa)
445
+
446
+ wp.copy(ifa, ifb)
447
+ ifa.assign(ifb)
448
+
449
+ na = ifa.numpy()
450
+ test.assertEqual(na.size, 0)
451
+ test.assertEqual(na.shape, (0, *dtype_shape))
452
+ test.assertEqual(na.dtype, nptype)
453
+
454
+ test.assertEqual(ifa.list(), [])
455
+
456
+ # test with scalars, vectors, and matrices
457
+ for nptype, wptype in wp.types.np_dtype_to_warp_type.items():
458
+ # scalars
459
+ test_empty_ops(0, 0, wptype, nptype)
460
+
461
+ for ncols in [2, 3, 4, 5]:
462
+ # vectors
463
+ test_empty_ops(0, ncols, wptype, nptype)
464
+ # square matrices (the Fabric interface only supports square matrices right now)
465
+ test_empty_ops(ncols, ncols, wptype, nptype)
466
+
467
+
468
+ def test_fabricarray_fill_scalar(test, device):
469
+ for nptype, wptype in wp.types.np_dtype_to_warp_type.items():
470
+ # create a data array
471
+ data = wp.zeros(100, dtype=wptype, device=device)
472
+ iface = _create_fabric_array_interface(data, "foo", copy=True)
473
+ fa = wp.fabricarray(data=iface, attrib="foo")
474
+
475
+ assert_np_equal(fa.numpy(), np.zeros(fa.shape, dtype=nptype))
476
+
477
+ # fill with int value
478
+ fill_value = 42
479
+ fa.fill_(fill_value)
480
+ assert_np_equal(fa.numpy(), np.full(fa.shape, fill_value, dtype=nptype))
481
+
482
+ fa.zero_()
483
+ assert_np_equal(fa.numpy(), np.zeros(fa.shape, dtype=nptype))
484
+
485
+ if wptype in wp.types.float_types:
486
+ # fill with float value
487
+ fill_value = 13.37
488
+ fa.fill_(fill_value)
489
+ assert_np_equal(fa.numpy(), np.full(fa.shape, fill_value, dtype=nptype))
490
+
491
+ # fill with Warp scalar value
492
+ fill_value = wptype(17)
493
+ fa.fill_(fill_value)
494
+ assert_np_equal(fa.numpy(), np.full(fa.shape, fill_value.value, dtype=nptype))
495
+
496
+ # reset data
497
+ wp.copy(fa, data)
498
+
499
+ # test indexed
500
+ indices1 = wp.array(data=np.arange(1, data.size, 2, dtype=np.int32), device=device)
501
+ ifa = fa[indices1]
502
+
503
+ # ensure that the other indices remain unchanged
504
+ indices2 = wp.array(data=np.arange(0, data.size, 2, dtype=np.int32), device=device)
505
+ ifb = fa[indices2]
506
+
507
+ assert_np_equal(ifa.numpy(), np.zeros(ifa.shape, dtype=nptype))
508
+ assert_np_equal(ifb.numpy(), np.zeros(ifb.shape, dtype=nptype))
509
+
510
+ # fill with int value
511
+ fill_value = 42
512
+ ifa.fill_(fill_value)
513
+ assert_np_equal(ifa.numpy(), np.full(ifa.shape, fill_value, dtype=nptype))
514
+ assert_np_equal(ifb.numpy(), np.zeros(ifb.shape, dtype=nptype))
515
+
516
+ ifa.zero_()
517
+ assert_np_equal(ifa.numpy(), np.zeros(ifa.shape, dtype=nptype))
518
+ assert_np_equal(ifb.numpy(), np.zeros(ifb.shape, dtype=nptype))
519
+
520
+ if wptype in wp.types.float_types:
521
+ # fill with float value
522
+ fill_value = 13.37
523
+ ifa.fill_(fill_value)
524
+ assert_np_equal(ifa.numpy(), np.full(ifa.shape, fill_value, dtype=nptype))
525
+ assert_np_equal(ifb.numpy(), np.zeros(ifb.shape, dtype=nptype))
526
+
527
+ # fill with Warp scalar value
528
+ fill_value = wptype(17)
529
+ ifa.fill_(fill_value)
530
+ assert_np_equal(ifa.numpy(), np.full(ifa.shape, fill_value.value, dtype=nptype))
531
+ assert_np_equal(ifb.numpy(), np.zeros(ifb.shape, dtype=nptype))
532
+
533
+
534
+ def test_fabricarray_fill_vector(test, device):
535
+ # test filling a vector array with scalar or vector values (vec_type, list, or numpy array)
536
+
537
+ for nptype, wptype in wp.types.np_dtype_to_warp_type.items():
538
+ # vector types
539
+ vector_types = [
540
+ wp.types.vector(2, wptype),
541
+ wp.types.vector(3, wptype),
542
+ wp.types.vector(4, wptype),
543
+ wp.types.vector(5, wptype),
544
+ ]
545
+
546
+ for vec_type in vector_types:
547
+ vec_len = vec_type._length_
548
+
549
+ data = wp.zeros(100, dtype=vec_type, device=device)
550
+ iface = _create_fabric_array_interface(data, "foo", copy=True)
551
+ fa = wp.fabricarray(data=iface, attrib="foo")
552
+
553
+ assert_np_equal(fa.numpy(), np.zeros((*fa.shape, vec_len), dtype=nptype))
554
+
555
+ # fill with int scalar
556
+ fill_value = 42
557
+ fa.fill_(fill_value)
558
+ assert_np_equal(fa.numpy(), np.full((*fa.shape, vec_len), fill_value, dtype=nptype))
559
+
560
+ # test zeroing
561
+ fa.zero_()
562
+ assert_np_equal(fa.numpy(), np.zeros((*fa.shape, vec_len), dtype=nptype))
563
+
564
+ # vector values can be passed as a list, numpy array, or Warp vector instance
565
+ fill_list = [17, 42, 99, 101, 127][:vec_len]
566
+ fill_arr = np.array(fill_list, dtype=nptype)
567
+ fill_vec = vec_type(fill_list)
568
+
569
+ expected = np.tile(fill_arr, fa.size).reshape((*fa.shape, vec_len))
570
+
571
+ # fill with list of vector length
572
+ fa.fill_(fill_list)
573
+ assert_np_equal(fa.numpy(), expected)
574
+
575
+ # clear
576
+ fa.zero_()
577
+
578
+ # fill with numpy array of vector length
579
+ fa.fill_(fill_arr)
580
+ assert_np_equal(fa.numpy(), expected)
581
+
582
+ # clear
583
+ fa.zero_()
584
+
585
+ # fill with vec instance
586
+ fa.fill_(fill_vec)
587
+ assert_np_equal(fa.numpy(), expected)
588
+
589
+ if wptype in wp.types.float_types:
590
+ # fill with float scalar
591
+ fill_value = 13.37
592
+ fa.fill_(fill_value)
593
+ assert_np_equal(fa.numpy(), np.full((*fa.shape, vec_len), fill_value, dtype=nptype))
594
+
595
+ # fill with float list of vector length
596
+ fill_list = [-2.5, -1.25, 1.25, 2.5, 5.0][:vec_len]
597
+
598
+ fa.fill_(fill_list)
599
+
600
+ expected = np.tile(np.array(fill_list, dtype=nptype), fa.size).reshape((*fa.shape, vec_len))
601
+
602
+ assert_np_equal(fa.numpy(), expected)
603
+
604
+ # reset data
605
+ wp.copy(fa, data)
606
+
607
+ # test indexed
608
+ indices1 = wp.array(data=np.arange(1, data.size, 2, dtype=np.int32), device=device)
609
+ ifa = fa[indices1]
610
+
611
+ # ensure that the other indices remain unchanged
612
+ indices2 = wp.array(data=np.arange(0, data.size, 2, dtype=np.int32), device=device)
613
+ ifb = fa[indices2]
614
+
615
+ assert_np_equal(ifa.numpy(), np.zeros((*ifa.shape, vec_len), dtype=nptype))
616
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
617
+
618
+ # fill with int scalar
619
+ fill_value = 42
620
+ ifa.fill_(fill_value)
621
+ assert_np_equal(ifa.numpy(), np.full((*ifa.shape, vec_len), fill_value, dtype=nptype))
622
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
623
+
624
+ # test zeroing
625
+ ifa.zero_()
626
+ assert_np_equal(ifa.numpy(), np.zeros((*ifa.shape, vec_len), dtype=nptype))
627
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
628
+
629
+ # vector values can be passed as a list, numpy array, or Warp vector instance
630
+ fill_list = [17, 42, 99, 101, 127][:vec_len]
631
+ fill_arr = np.array(fill_list, dtype=nptype)
632
+ fill_vec = vec_type(fill_list)
633
+
634
+ expected = np.tile(fill_arr, ifa.size).reshape((*ifa.shape, vec_len))
635
+
636
+ # fill with list of vector length
637
+ ifa.fill_(fill_list)
638
+ assert_np_equal(ifa.numpy(), expected)
639
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
640
+
641
+ # clear
642
+ ifa.zero_()
643
+
644
+ # fill with numpy array of vector length
645
+ ifa.fill_(fill_arr)
646
+ assert_np_equal(ifa.numpy(), expected)
647
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
648
+
649
+ # clear
650
+ ifa.zero_()
651
+
652
+ # fill with vec instance
653
+ ifa.fill_(fill_vec)
654
+ assert_np_equal(ifa.numpy(), expected)
655
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
656
+
657
+ if wptype in wp.types.float_types:
658
+ # fill with float scalar
659
+ fill_value = 13.37
660
+ ifa.fill_(fill_value)
661
+ assert_np_equal(ifa.numpy(), np.full((*ifa.shape, vec_len), fill_value, dtype=nptype))
662
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
663
+
664
+ # fill with float list of vector length
665
+ fill_list = [-2.5, -1.25, 1.25, 2.5, 5.0][:vec_len]
666
+
667
+ ifa.fill_(fill_list)
668
+
669
+ expected = np.tile(np.array(fill_list, dtype=nptype), ifa.size).reshape((*ifa.shape, vec_len))
670
+
671
+ assert_np_equal(ifa.numpy(), expected)
672
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, vec_len), dtype=nptype))
673
+
674
+
675
+ def test_fabricarray_fill_matrix(test, device):
676
+ # test filling a matrix array with scalar or matrix values (mat_type, nested list, or 2d numpy array)
677
+
678
+ for nptype, wptype in wp.types.np_dtype_to_warp_type.items():
679
+ # matrix types
680
+ matrix_types = [
681
+ # square matrices only
682
+ wp.types.matrix((2, 2), wptype),
683
+ wp.types.matrix((3, 3), wptype),
684
+ wp.types.matrix((4, 4), wptype),
685
+ wp.types.matrix((5, 5), wptype),
686
+ ]
687
+
688
+ for mat_type in matrix_types:
689
+ mat_len = mat_type._length_
690
+ mat_shape = mat_type._shape_
691
+
692
+ data = wp.zeros(100, dtype=mat_type, device=device)
693
+ iface = _create_fabric_array_interface(data, "foo", copy=True)
694
+ fa = wp.fabricarray(data=iface, attrib="foo")
695
+
696
+ assert_np_equal(fa.numpy(), np.zeros((*fa.shape, *mat_shape), dtype=nptype))
697
+
698
+ # fill with scalar
699
+ fill_value = 42
700
+ fa.fill_(fill_value)
701
+ assert_np_equal(fa.numpy(), np.full((*fa.shape, *mat_shape), fill_value, dtype=nptype))
702
+
703
+ # test zeroing
704
+ fa.zero_()
705
+ assert_np_equal(fa.numpy(), np.zeros((*fa.shape, *mat_shape), dtype=nptype))
706
+
707
+ # matrix values can be passed as a 1d numpy array, 2d numpy array, flat list, nested list, or Warp matrix instance
708
+ if wptype != wp.bool:
709
+ fill_arr1 = np.arange(mat_len, dtype=nptype)
710
+ else:
711
+ fill_arr1 = np.ones(mat_len, dtype=nptype)
712
+
713
+ fill_arr2 = fill_arr1.reshape(mat_shape)
714
+ fill_list1 = list(fill_arr1)
715
+ fill_list2 = [list(row) for row in fill_arr2]
716
+ fill_mat = mat_type(fill_arr1)
717
+
718
+ expected = np.tile(fill_arr1, fa.size).reshape((*fa.shape, *mat_shape))
719
+
720
+ # fill with 1d numpy array
721
+ fa.fill_(fill_arr1)
722
+ assert_np_equal(fa.numpy(), expected)
723
+
724
+ # clear
725
+ fa.zero_()
726
+
727
+ # fill with 2d numpy array
728
+ fa.fill_(fill_arr2)
729
+ assert_np_equal(fa.numpy(), expected)
730
+
731
+ # clear
732
+ fa.zero_()
733
+
734
+ # fill with flat list
735
+ fa.fill_(fill_list1)
736
+ assert_np_equal(fa.numpy(), expected)
737
+
738
+ # clear
739
+ fa.zero_()
740
+
741
+ # fill with nested list
742
+ fa.fill_(fill_list2)
743
+ assert_np_equal(fa.numpy(), expected)
744
+
745
+ # clear
746
+ fa.zero_()
747
+
748
+ # fill with mat instance
749
+ fa.fill_(fill_mat)
750
+ assert_np_equal(fa.numpy(), expected)
751
+
752
+ # reset data
753
+ wp.copy(fa, data)
754
+
755
+ # test indexed
756
+ indices1 = wp.array(data=np.arange(1, data.size, 2, dtype=np.int32), device=device)
757
+ ifa = fa[indices1]
758
+
759
+ # ensure that the other indices remain unchanged
760
+ indices2 = wp.array(data=np.arange(0, data.size, 2, dtype=np.int32), device=device)
761
+ ifb = fa[indices2]
762
+
763
+ assert_np_equal(ifa.numpy(), np.zeros((*ifa.shape, *mat_shape), dtype=nptype))
764
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
765
+
766
+ # fill with scalar
767
+ fill_value = 42
768
+ ifa.fill_(fill_value)
769
+ assert_np_equal(ifa.numpy(), np.full((*ifa.shape, *mat_shape), fill_value, dtype=nptype))
770
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
771
+
772
+ # test zeroing
773
+ ifa.zero_()
774
+ assert_np_equal(ifa.numpy(), np.zeros((*ifa.shape, *mat_shape), dtype=nptype))
775
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
776
+
777
+ # matrix values can be passed as a 1d numpy array, 2d numpy array, flat list, nested list, or Warp matrix instance
778
+ if wptype != wp.bool:
779
+ fill_arr1 = np.arange(mat_len, dtype=nptype)
780
+ else:
781
+ fill_arr1 = np.ones(mat_len, dtype=nptype)
782
+ fill_arr2 = fill_arr1.reshape(mat_shape)
783
+ fill_list1 = list(fill_arr1)
784
+ fill_list2 = [list(row) for row in fill_arr2]
785
+ fill_mat = mat_type(fill_arr1)
786
+
787
+ expected = np.tile(fill_arr1, ifa.size).reshape((*ifa.shape, *mat_shape))
788
+
789
+ # fill with 1d numpy array
790
+ ifa.fill_(fill_arr1)
791
+ assert_np_equal(ifa.numpy(), expected)
792
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
793
+
794
+ # clear
795
+ ifa.zero_()
796
+
797
+ # fill with 2d numpy array
798
+ ifa.fill_(fill_arr2)
799
+ assert_np_equal(ifa.numpy(), expected)
800
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
801
+
802
+ # clear
803
+ ifa.zero_()
804
+
805
+ # fill with flat list
806
+ ifa.fill_(fill_list1)
807
+ assert_np_equal(ifa.numpy(), expected)
808
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
809
+
810
+ # clear
811
+ ifa.zero_()
812
+
813
+ # fill with nested list
814
+ ifa.fill_(fill_list2)
815
+ assert_np_equal(ifa.numpy(), expected)
816
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
817
+
818
+ # clear
819
+ ifa.zero_()
820
+
821
+ # fill with mat instance
822
+ ifa.fill_(fill_mat)
823
+ assert_np_equal(ifa.numpy(), expected)
824
+ assert_np_equal(ifb.numpy(), np.zeros((*ifb.shape, *mat_shape), dtype=nptype))
825
+
826
+
827
+ @wp.kernel
828
+ def fa_generic_sums_kernel(a: wp.fabricarrayarray(dtype=Any), sums: wp.array(dtype=Any)):
829
+ i = wp.tid()
830
+
831
+ # get sub-array using wp::view()
832
+ row = a[i]
833
+
834
+ # get sub-array length
835
+ count = row.shape[0]
836
+
837
+ # compute sub-array sum
838
+ for j in range(count):
839
+ sums[i] = sums[i] + row[j]
840
+
841
+
842
+ @wp.kernel
843
+ def fa_generic_sums_kernel_indexed(a: wp.indexedfabricarrayarray(dtype=Any), sums: wp.array(dtype=Any)):
844
+ i = wp.tid()
845
+
846
+ # get sub-array using wp::view()
847
+ row = a[i]
848
+
849
+ # get sub-array length
850
+ count = row.shape[0]
851
+
852
+ # compute sub-array sum
853
+ for j in range(count):
854
+ sums[i] = sums[i] + row[j]
855
+
856
+
857
+ def test_fabricarrayarray(test, device):
858
+ for T in _fabric_types:
859
+ if hasattr(T, "_wp_scalar_type_"):
860
+ nptype = wp.types.warp_type_to_np_dtype[T._wp_scalar_type_]
861
+ else:
862
+ nptype = wp.types.warp_type_to_np_dtype[T]
863
+
864
+ n = 100
865
+
866
+ min_length = 1
867
+ max_length = 10
868
+ arrays = []
869
+ expected_sums = []
870
+ expected_sums_indexed = []
871
+
872
+ # generate data arrays
873
+ length = min_length
874
+ for i in range(n):
875
+ if length > max_length:
876
+ length = min_length
877
+
878
+ na = np.arange(1, length + 1, dtype=nptype)
879
+
880
+ arrays.append(wp.array(data=na, device=device))
881
+ expected_sums.append(na.sum())
882
+
883
+ # every second index
884
+ if i % 2 == 0:
885
+ expected_sums_indexed.append(na.sum())
886
+
887
+ length += 1
888
+
889
+ data_iface = _create_fabric_array_array_interface(arrays, "foo")
890
+ fa = wp.fabricarrayarray(data=data_iface, attrib="foo")
891
+
892
+ sums = wp.zeros_like(fa)
893
+
894
+ test.assertEqual(fa.dtype, sums.dtype)
895
+ test.assertEqual(fa.ndim, 2)
896
+ test.assertEqual(sums.ndim, 1)
897
+ test.assertEqual(fa.shape, sums.shape)
898
+ test.assertEqual(fa.size, sums.size)
899
+
900
+ wp.launch(fa_generic_sums_kernel, dim=fa.size, inputs=[fa, sums], device=device)
901
+
902
+ assert_np_equal(sums.numpy(), np.array(expected_sums, dtype=nptype))
903
+
904
+ # test indexed
905
+ indices = wp.array(data=np.arange(0, n, 2, dtype=np.int32), device=device)
906
+ ifa = fa[indices]
907
+
908
+ sums = wp.zeros_like(ifa)
909
+
910
+ test.assertEqual(ifa.dtype, sums.dtype)
911
+ test.assertEqual(ifa.ndim, 2)
912
+ test.assertEqual(sums.ndim, 1)
913
+ test.assertEqual(ifa.shape, sums.shape)
914
+ test.assertEqual(ifa.size, sums.size)
915
+
916
+ wp.launch(fa_generic_sums_kernel_indexed, dim=ifa.size, inputs=[ifa, sums], device=device)
917
+
918
+ assert_np_equal(sums.numpy(), np.array(expected_sums_indexed, dtype=nptype))
919
+
920
+
921
+ # explicit kernel overloads
922
+ for T in _fabric_types:
923
+ wp.overload(fa_generic_dtype_kernel, [wp.fabricarray(dtype=T), wp.fabricarray(dtype=T)])
924
+ wp.overload(fa_generic_dtype_kernel_indexed, [wp.indexedfabricarray(dtype=T), wp.indexedfabricarray(dtype=T)])
925
+
926
+ wp.overload(fa_generic_array_kernel, [wp.fabricarray(dtype=T), wp.fabricarray(dtype=T)])
927
+ wp.overload(fa_generic_array_kernel, [wp.indexedfabricarray(dtype=T), wp.indexedfabricarray(dtype=T)])
928
+
929
+ wp.overload(fa_generic_sums_kernel, [wp.fabricarrayarray(dtype=T), wp.array(dtype=T)])
930
+ wp.overload(fa_generic_sums_kernel_indexed, [wp.indexedfabricarrayarray(dtype=T), wp.array(dtype=T)])
931
+
932
+
933
+ devices = get_test_devices()
934
+
935
+
936
+ class TestFabricArray(unittest.TestCase):
937
+ pass
938
+
939
+
940
+ # fabric arrays
941
+ add_function_test(TestFabricArray, "test_fabricarray_kernel", test_fabricarray_kernel, devices=devices)
942
+ add_function_test(TestFabricArray, "test_fabricarray_empty", test_fabricarray_empty, devices=devices)
943
+ add_function_test(TestFabricArray, "test_fabricarray_generic_dtype", test_fabricarray_generic_dtype, devices=devices)
944
+ add_function_test(TestFabricArray, "test_fabricarray_generic_array", test_fabricarray_generic_array, devices=devices)
945
+ add_function_test(TestFabricArray, "test_fabricarray_fill_scalar", test_fabricarray_fill_scalar, devices=devices)
946
+ add_function_test(TestFabricArray, "test_fabricarray_fill_vector", test_fabricarray_fill_vector, devices=devices)
947
+ add_function_test(TestFabricArray, "test_fabricarray_fill_matrix", test_fabricarray_fill_matrix, devices=devices)
948
+
949
+ # fabric arrays of arrays
950
+ add_function_test(TestFabricArray, "test_fabricarrayarray", test_fabricarrayarray, devices=devices)
951
+
952
+
953
+ if __name__ == "__main__":
954
+ wp.build.clear_kernel_cache()
955
+ unittest.main(verbosity=2)