executorch 1.4.0__cp314-cp314-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.
Files changed (4120) hide show
  1. executorch/backends/__init__.py +0 -0
  2. executorch/backends/aoti/aoti_backend.py +360 -0
  3. executorch/backends/aoti/aoti_partitioner.py +141 -0
  4. executorch/backends/aoti/passes/replace_view_copy_with_view.py +118 -0
  5. executorch/backends/apple/coreml/compiler/__init__.py +9 -0
  6. executorch/backends/apple/coreml/compiler/coreml_preprocess.py +955 -0
  7. executorch/backends/apple/coreml/compiler/enumerated_shape_utils.py +244 -0
  8. executorch/backends/apple/coreml/compiler/torch_ops.py +240 -0
  9. executorch/backends/apple/coreml/executorchcoreml.pyi +10 -0
  10. executorch/backends/apple/coreml/logging.py +24 -0
  11. executorch/backends/apple/coreml/partition/__init__.py +9 -0
  12. executorch/backends/apple/coreml/partition/coreml_partitioner.py +382 -0
  13. executorch/backends/apple/coreml/quantizer/__init__.py +9 -0
  14. executorch/backends/apple/coreml/quantizer/_annotation_config.py +130 -0
  15. executorch/backends/apple/coreml/quantizer/_coreml_quantizer.py +628 -0
  16. executorch/backends/apple/coreml/quantizer/_coreml_quantizer_utils.py +800 -0
  17. executorch/backends/apple/coreml/quantizer/coreml_quantizer.py +7 -0
  18. executorch/backends/apple/coreml/recipes/__init__.py +17 -0
  19. executorch/backends/apple/coreml/recipes/coreml_recipe_provider.py +413 -0
  20. executorch/backends/apple/coreml/recipes/coreml_recipe_types.py +53 -0
  21. executorch/backends/apple/coreml/runtime/test/export_stateful_model.py +77 -0
  22. executorch/backends/apple/coreml/test/test_coreml_multifunction.py +334 -0
  23. executorch/backends/apple/coreml/test/test_coreml_partitioner.py +536 -0
  24. executorch/backends/apple/coreml/test/test_coreml_quantizer.py +172 -0
  25. executorch/backends/apple/coreml/test/test_coreml_recipes.py +552 -0
  26. executorch/backends/apple/coreml/test/test_coreml_utils.py +19 -0
  27. executorch/backends/apple/coreml/test/test_enumerated_shapes.py +112 -0
  28. executorch/backends/apple/coreml/test/test_torch_ops.py +360 -0
  29. executorch/backends/apple/coreml/test/tester.py +121 -0
  30. executorch/backends/apple/metal/metal_backend.py +118 -0
  31. executorch/backends/apple/metal/metal_partitioner.py +25 -0
  32. executorch/backends/apple/metal/ops/__init__.py +5 -0
  33. executorch/backends/apple/metal/ops/gated_delta_rule.py +92 -0
  34. executorch/backends/apple/metal/ops/gather_qmv.py +117 -0
  35. executorch/backends/apple/metal/passes/__init__.py +11 -0
  36. executorch/backends/apple/metal/passes/decompose_linear_pass.py +115 -0
  37. executorch/backends/apple/metal/tests/__init__.py +5 -0
  38. executorch/backends/apple/metal/tests/test_metal_backend.py +79 -0
  39. executorch/backends/apple/metal/tests/test_metal_partitioner.py +172 -0
  40. executorch/backends/apple/metal/tests/test_modules.py +1404 -0
  41. executorch/backends/apple/mps/__init__.py +19 -0
  42. executorch/backends/apple/mps/mps_preprocess.py +329 -0
  43. executorch/backends/apple/mps/operators/__init__.py +77 -0
  44. executorch/backends/apple/mps/operators/activation_ops.py +99 -0
  45. executorch/backends/apple/mps/operators/binary_ops.py +156 -0
  46. executorch/backends/apple/mps/operators/clamp_ops.py +60 -0
  47. executorch/backends/apple/mps/operators/constant_ops.py +128 -0
  48. executorch/backends/apple/mps/operators/convolution_ops.py +70 -0
  49. executorch/backends/apple/mps/operators/indexing_ops.py +225 -0
  50. executorch/backends/apple/mps/operators/linear_algebra_ops.py +52 -0
  51. executorch/backends/apple/mps/operators/node_visitor.py +473 -0
  52. executorch/backends/apple/mps/operators/normalization_ops.py +97 -0
  53. executorch/backends/apple/mps/operators/op_clone.py +54 -0
  54. executorch/backends/apple/mps/operators/op_getitem.py +29 -0
  55. executorch/backends/apple/mps/operators/op_quant_dequant.py +142 -0
  56. executorch/backends/apple/mps/operators/op_skip_ops.py +24 -0
  57. executorch/backends/apple/mps/operators/pad_ops.py +37 -0
  58. executorch/backends/apple/mps/operators/pooling_ops.py +142 -0
  59. executorch/backends/apple/mps/operators/range_ops.py +53 -0
  60. executorch/backends/apple/mps/operators/reduce_ops.py +39 -0
  61. executorch/backends/apple/mps/operators/shape_ops.py +263 -0
  62. executorch/backends/apple/mps/operators/unary_ops.py +130 -0
  63. executorch/backends/apple/mps/partition/__init__.py +20 -0
  64. executorch/backends/apple/mps/partition/mps_partitioner.py +159 -0
  65. executorch/backends/apple/mps/serialization/mps_graph_schema.py +817 -0
  66. executorch/backends/apple/mps/serialization/mps_graph_serialize.py +34 -0
  67. executorch/backends/apple/mps/serialization/schema.fbs +523 -0
  68. executorch/backends/apple/mps/test/test_mps.py +2507 -0
  69. executorch/backends/apple/mps/test/test_mps_binary_ops.py +296 -0
  70. executorch/backends/apple/mps/test/test_mps_deprecation.py +125 -0
  71. executorch/backends/apple/mps/test/test_mps_indexing_ops.py +264 -0
  72. executorch/backends/apple/mps/test/test_mps_linear.py +523 -0
  73. executorch/backends/apple/mps/test/test_mps_models.py +416 -0
  74. executorch/backends/apple/mps/test/test_mps_unary_ops.py +26 -0
  75. executorch/backends/apple/mps/test/test_mps_utils.py +326 -0
  76. executorch/backends/apple/mps/utils/mps_utils.py +82 -0
  77. executorch/backends/apple/mps/utils/quant_utils.py +41 -0
  78. executorch/backends/arm/__init__.py +55 -0
  79. executorch/backends/arm/_passes/__init__.py +213 -0
  80. executorch/backends/arm/_passes/_debug_passes.py +59 -0
  81. executorch/backends/arm/_passes/accumulate_index_put_pass.py +65 -0
  82. executorch/backends/arm/_passes/arm_pass.py +294 -0
  83. executorch/backends/arm/_passes/arm_pass_manager.py +779 -0
  84. executorch/backends/arm/_passes/arm_pass_utils.py +427 -0
  85. executorch/backends/arm/_passes/aten_to_tosa_activation_functions.py +148 -0
  86. executorch/backends/arm/_passes/aten_to_tosa_data_layout.py +98 -0
  87. executorch/backends/arm/_passes/aten_to_tosa_tensor_operators.py +118 -0
  88. executorch/backends/arm/_passes/broadcast_args_pass.py +90 -0
  89. executorch/backends/arm/_passes/canonicalize_gather_pass.py +203 -0
  90. executorch/backends/arm/_passes/canonicalize_view_copy_permute_pass.py +372 -0
  91. executorch/backends/arm/_passes/cast_int64_pass.py +99 -0
  92. executorch/backends/arm/_passes/cast_to_int32_pass.py +69 -0
  93. executorch/backends/arm/_passes/constant_folding_pass.py +81 -0
  94. executorch/backends/arm/_passes/control_flow_const_inline.py +78 -0
  95. executorch/backends/arm/_passes/conv1d_unsqueeze_pass.py +113 -0
  96. executorch/backends/arm/_passes/convert_elu_params.py +88 -0
  97. executorch/backends/arm/_passes/convert_expand_copy_to_repeat.py +80 -0
  98. executorch/backends/arm/_passes/convert_full_like_to_full_pass.py +49 -0
  99. executorch/backends/arm/_passes/convert_int64_const_ops_to_int32.py +81 -0
  100. executorch/backends/arm/_passes/convert_int64_output_ops_to_int32.py +199 -0
  101. executorch/backends/arm/_passes/convert_minmax_pass.py +166 -0
  102. executorch/backends/arm/_passes/convert_permute_singleton_to_view_pass.py +65 -0
  103. executorch/backends/arm/_passes/convert_split_to_slice.py +134 -0
  104. executorch/backends/arm/_passes/convert_squeezes_to_view.py +40 -0
  105. executorch/backends/arm/_passes/convert_to_clamp_pass.py +47 -0
  106. executorch/backends/arm/_passes/decompose_acosh_pass.py +75 -0
  107. executorch/backends/arm/_passes/decompose_adaptive_avg_pool2d_pass.py +462 -0
  108. executorch/backends/arm/_passes/decompose_adaptive_max_pool2d_pass.py +203 -0
  109. executorch/backends/arm/_passes/decompose_add_sub_alpha_pass.py +102 -0
  110. executorch/backends/arm/_passes/decompose_addmm_pass.py +75 -0
  111. executorch/backends/arm/_passes/decompose_any_pass.py +122 -0
  112. executorch/backends/arm/_passes/decompose_as_strided_copy_pass.py +116 -0
  113. executorch/backends/arm/_passes/decompose_asin_and_acos_pass.py +223 -0
  114. executorch/backends/arm/_passes/decompose_asinh_pass.py +72 -0
  115. executorch/backends/arm/_passes/decompose_atan_pass.py +139 -0
  116. executorch/backends/arm/_passes/decompose_atanh_pass.py +84 -0
  117. executorch/backends/arm/_passes/decompose_avg_pool2d_pass.py +178 -0
  118. executorch/backends/arm/_passes/decompose_batch_norm_no_stats.py +235 -0
  119. executorch/backends/arm/_passes/decompose_cosh_pass.py +69 -0
  120. executorch/backends/arm/_passes/decompose_cosine_similarity_pass.py +98 -0
  121. executorch/backends/arm/_passes/decompose_cumsum_pass.py +167 -0
  122. executorch/backends/arm/_passes/decompose_div_pass.py +59 -0
  123. executorch/backends/arm/_passes/decompose_div_tensor_mode.py +283 -0
  124. executorch/backends/arm/_passes/decompose_dynamic_adaptive_avg_pool2d_pass.py +57 -0
  125. executorch/backends/arm/_passes/decompose_dynamic_full_pass.py +56 -0
  126. executorch/backends/arm/_passes/decompose_einsum_pass.py +66 -0
  127. executorch/backends/arm/_passes/decompose_elu_pass.py +187 -0
  128. executorch/backends/arm/_passes/decompose_embedding_pass.py +132 -0
  129. executorch/backends/arm/_passes/decompose_erfinv_pass.py +374 -0
  130. executorch/backends/arm/_passes/decompose_expm1_pass.py +160 -0
  131. executorch/backends/arm/_passes/decompose_flip_pass.py +35 -0
  132. executorch/backends/arm/_passes/decompose_floor_divide_pass.py +76 -0
  133. executorch/backends/arm/_passes/decompose_gelu_pass.py +144 -0
  134. executorch/backends/arm/_passes/decompose_glu_pass.py +84 -0
  135. executorch/backends/arm/_passes/decompose_grouped_conv_pass.py +319 -0
  136. executorch/backends/arm/_passes/decompose_groupnorm_pass.py +223 -0
  137. executorch/backends/arm/_passes/decompose_gru_pass.py +337 -0
  138. executorch/backends/arm/_passes/decompose_index_copy_pass.py +33 -0
  139. executorch/backends/arm/_passes/decompose_index_select_to_gather_pass.py +182 -0
  140. executorch/backends/arm/_passes/decompose_index_tensor_to_gather_pass.py +400 -0
  141. executorch/backends/arm/_passes/decompose_int_pow_pass.py +87 -0
  142. executorch/backends/arm/_passes/decompose_layernorm_pass.py +203 -0
  143. executorch/backends/arm/_passes/decompose_leaky_relu_pass.py +80 -0
  144. executorch/backends/arm/_passes/decompose_linalg_vector_norm_pass.py +93 -0
  145. executorch/backends/arm/_passes/decompose_linear_pass.py +117 -0
  146. executorch/backends/arm/_passes/decompose_log1p_pass.py +96 -0
  147. executorch/backends/arm/_passes/decompose_logit_pass.py +119 -0
  148. executorch/backends/arm/_passes/decompose_lstm_pass.py +378 -0
  149. executorch/backends/arm/_passes/decompose_masked_fill_pass.py +60 -0
  150. executorch/backends/arm/_passes/decompose_matmul.py +47 -0
  151. executorch/backends/arm/_passes/decompose_maxpool2d_with_dilation_pass.py +242 -0
  152. executorch/backends/arm/_passes/decompose_meandim_pass.py +263 -0
  153. executorch/backends/arm/_passes/decompose_ne_pass.py +75 -0
  154. executorch/backends/arm/_passes/decompose_permute_for_u55_pass.py +370 -0
  155. executorch/backends/arm/_passes/decompose_quant_nodes.py +161 -0
  156. executorch/backends/arm/_passes/decompose_remainder_pass.py +74 -0
  157. executorch/backends/arm/_passes/decompose_rnn_pass.py +300 -0
  158. executorch/backends/arm/_passes/decompose_round_pass.py +85 -0
  159. executorch/backends/arm/_passes/decompose_sdpa_pass.py +41 -0
  160. executorch/backends/arm/_passes/decompose_select.py +80 -0
  161. executorch/backends/arm/_passes/decompose_select_scatter_pass.py +146 -0
  162. executorch/backends/arm/_passes/decompose_sign_pass.py +83 -0
  163. executorch/backends/arm/_passes/decompose_sinh_pass.py +74 -0
  164. executorch/backends/arm/_passes/decompose_slice_scatter_pass.py +201 -0
  165. executorch/backends/arm/_passes/decompose_softmax_pass.py +119 -0
  166. executorch/backends/arm/_passes/decompose_sqrt_pass.py +46 -0
  167. executorch/backends/arm/_passes/decompose_strided_slice_copy_pass.py +147 -0
  168. executorch/backends/arm/_passes/decompose_sum_pass.py +94 -0
  169. executorch/backends/arm/_passes/decompose_tan_pass.py +44 -0
  170. executorch/backends/arm/_passes/decompose_tosa_unsupported_clamp_pass.py +99 -0
  171. executorch/backends/arm/_passes/decompose_tril_pass.py +112 -0
  172. executorch/backends/arm/_passes/decompose_unfold_to_gather_pass.py +372 -0
  173. executorch/backends/arm/_passes/decompose_unsupported_bilinear_resize_pass.py +159 -0
  174. executorch/backends/arm/_passes/decompose_var_pass.py +107 -0
  175. executorch/backends/arm/_passes/decompose_where_scalar_other_pass.py +62 -0
  176. executorch/backends/arm/_passes/decorate_fp32_to_int32_casting_pass.py +90 -0
  177. executorch/backends/arm/_passes/deduplicate_get_attr_pass.py +96 -0
  178. executorch/backends/arm/_passes/dim_maps.py +959 -0
  179. executorch/backends/arm/_passes/ensure_unique_output_nodes_pass.py +82 -0
  180. executorch/backends/arm/_passes/exir_to_tosa_pass.py +136 -0
  181. executorch/backends/arm/_passes/fold_qdq_with_annotated_qparams_pass.py +442 -0
  182. executorch/backends/arm/_passes/fold_scalar_mul_into_conv_pass.py +335 -0
  183. executorch/backends/arm/_passes/fuse_batch_norm2d_pass.py +246 -0
  184. executorch/backends/arm/_passes/fuse_consecutive_concat_shapes.py +64 -0
  185. executorch/backends/arm/_passes/fuse_consecutive_rescales_pass.py +175 -0
  186. executorch/backends/arm/_passes/fuse_consecutive_slices_pass.py +112 -0
  187. executorch/backends/arm/_passes/fuse_constant_ops_pass.py +362 -0
  188. executorch/backends/arm/_passes/fuse_duplicate_users_pass.py +170 -0
  189. executorch/backends/arm/_passes/fuse_equal_placeholders_pass.py +125 -0
  190. executorch/backends/arm/_passes/fuse_identical_input_transforms_pass.py +401 -0
  191. executorch/backends/arm/_passes/fuse_quantized_activation_pass.py +82 -0
  192. executorch/backends/arm/_passes/fuse_view_copy_transform_pass.py +14 -0
  193. executorch/backends/arm/_passes/get_decomposition_pass.py +149 -0
  194. executorch/backends/arm/_passes/insert_const_shapes.py +78 -0
  195. executorch/backends/arm/_passes/insert_data_layout_casts_pass.py +134 -0
  196. executorch/backends/arm/_passes/insert_dynamic_padding.py +101 -0
  197. executorch/backends/arm/_passes/insert_int32_casts_after_int64_placeholders.py +129 -0
  198. executorch/backends/arm/_passes/insert_rescales_pass.py +667 -0
  199. executorch/backends/arm/_passes/insert_table_ops.py +345 -0
  200. executorch/backends/arm/_passes/match_arg_dtype_pass.py +105 -0
  201. executorch/backends/arm/_passes/match_arg_ranks_pass.py +116 -0
  202. executorch/backends/arm/_passes/mm_to_bmm_pass.py +114 -0
  203. executorch/backends/arm/_passes/normalize_delegate_io_layout_pass.py +138 -0
  204. executorch/backends/arm/_passes/normalize_index_put_bool_index_tensor_pass.py +102 -0
  205. executorch/backends/arm/_passes/normalize_index_put_none_indices_pass.py +144 -0
  206. executorch/backends/arm/_passes/normalize_while_initial_args_pass.py +115 -0
  207. executorch/backends/arm/_passes/promote_bool_operands_pass.py +84 -0
  208. executorch/backends/arm/_passes/propagate_view_copy_permute_pass.py +751 -0
  209. executorch/backends/arm/_passes/quant_args.py +131 -0
  210. executorch/backends/arm/_passes/remove_getitem_pass.py +14 -0
  211. executorch/backends/arm/_passes/remove_graph_asserts_pass.py +18 -0
  212. executorch/backends/arm/_passes/remove_noop_pass.py +43 -0
  213. executorch/backends/arm/_passes/remove_permutes_around_elementwise_tosa_ops.py +42 -0
  214. executorch/backends/arm/_passes/replace_inf_and_limit_values_pass.py +81 -0
  215. executorch/backends/arm/_passes/replace_scalar_with_tensor_pass.py +139 -0
  216. executorch/backends/arm/_passes/rewrite_adaptive_avg_pool2d.py +170 -0
  217. executorch/backends/arm/_passes/rewrite_avg_pool2d_pass.py +108 -0
  218. executorch/backends/arm/_passes/rewrite_bool_bitwise_to_logical_pass.py +50 -0
  219. executorch/backends/arm/_passes/rewrite_bool_to_fp32_cast_via_int8_pass.py +77 -0
  220. executorch/backends/arm/_passes/rewrite_conv_pass.py +699 -0
  221. executorch/backends/arm/_passes/rewrite_high_rank_singleton_permute_pass.py +114 -0
  222. executorch/backends/arm/_passes/rewrite_index_put_pass.py +200 -0
  223. executorch/backends/arm/_passes/rewrite_inplace_arithmetic_pass.py +41 -0
  224. executorch/backends/arm/_passes/rewrite_le_lt_to_ge_gt_pass.py +44 -0
  225. executorch/backends/arm/_passes/rewrite_matmul.py +125 -0
  226. executorch/backends/arm/_passes/rewrite_max_pool2d_pass.py +189 -0
  227. executorch/backends/arm/_passes/rewrite_mxfp_conv2d.py +407 -0
  228. executorch/backends/arm/_passes/rewrite_mxfp_linear.py +390 -0
  229. executorch/backends/arm/_passes/rewrite_pad.py +177 -0
  230. executorch/backends/arm/_passes/rewrite_slice.py +75 -0
  231. executorch/backends/arm/_passes/rewrite_upsample.py +267 -0
  232. executorch/backends/arm/_passes/scalars_to_attribute_pass.py +144 -0
  233. executorch/backends/arm/_passes/size_adjust_input_pass.py +287 -0
  234. executorch/backends/arm/_passes/symbolic_to_tosa_shape_pass.py +26 -0
  235. executorch/backends/arm/_passes/symbolic_value_range.py +189 -0
  236. executorch/backends/arm/_passes/unsqueeze_before_repeat_pass.py +66 -0
  237. executorch/backends/arm/_passes/unsqueeze_scalar_placeholders_pass.py +101 -0
  238. executorch/backends/arm/ao_ext/__init__.py +12 -0
  239. executorch/backends/arm/ao_ext/mxfp.py +137 -0
  240. executorch/backends/arm/ao_ext/mxfp_tosa_lib.py +12 -0
  241. executorch/backends/arm/ao_ext/mxfp_transform.py +27 -0
  242. executorch/backends/arm/ao_ext/ops/__init__.py +12 -0
  243. executorch/backends/arm/ao_ext/ops/mxfp_conv2d_op.py +325 -0
  244. executorch/backends/arm/ao_ext/ops/mxfp_linear_op.py +236 -0
  245. executorch/backends/arm/arm_vela.py +144 -0
  246. executorch/backends/arm/cmsis_pack/scripts/generate_components.py +462 -0
  247. executorch/backends/arm/cmsis_pack/scripts/generate_register_all_kernels.py +1215 -0
  248. executorch/backends/arm/cmsis_pack/test/validate_pack.py +85 -0
  249. executorch/backends/arm/common/__init__.py +4 -0
  250. executorch/backends/arm/common/annotation_meta.py +39 -0
  251. executorch/backends/arm/common/arm_compile_spec.py +385 -0
  252. executorch/backends/arm/common/as_strided_utils.py +72 -0
  253. executorch/backends/arm/common/debug.py +79 -0
  254. executorch/backends/arm/common/pipeline_config.py +109 -0
  255. executorch/backends/arm/common/type.py +28 -0
  256. executorch/backends/arm/constants.py +52 -0
  257. executorch/backends/arm/debug/__init__.py +4 -0
  258. executorch/backends/arm/debug/schema.py +145 -0
  259. executorch/backends/arm/ethosu/__init__.py +11 -0
  260. executorch/backends/arm/ethosu/backend.py +115 -0
  261. executorch/backends/arm/ethosu/compile_spec.py +154 -0
  262. executorch/backends/arm/ethosu/partitioner.py +41 -0
  263. executorch/backends/arm/operator_support/__init__.py +30 -0
  264. executorch/backends/arm/operator_support/as_strided_copy_support.py +113 -0
  265. executorch/backends/arm/operator_support/bool_bitwise_support.py +39 -0
  266. executorch/backends/arm/operator_support/clone_dim_order_support.py +77 -0
  267. executorch/backends/arm/operator_support/control_flow_support.py +194 -0
  268. executorch/backends/arm/operator_support/convolution_support.py +321 -0
  269. executorch/backends/arm/operator_support/embedding_support.py +56 -0
  270. executorch/backends/arm/operator_support/ethos_u55_support.py +306 -0
  271. executorch/backends/arm/operator_support/gather_support.py +166 -0
  272. executorch/backends/arm/operator_support/index_put_support.py +86 -0
  273. executorch/backends/arm/operator_support/index_select_support.py +128 -0
  274. executorch/backends/arm/operator_support/index_tensor_support.py +193 -0
  275. executorch/backends/arm/operator_support/minmax_support.py +59 -0
  276. executorch/backends/arm/operator_support/pool_2d_support.py +266 -0
  277. executorch/backends/arm/operator_support/reduce_sum_support.py +71 -0
  278. executorch/backends/arm/operator_support/right_shift_support.py +47 -0
  279. executorch/backends/arm/operator_support/slice_copy_support.py +119 -0
  280. executorch/backends/arm/operator_support/sym_size_int_support.py +25 -0
  281. executorch/backends/arm/operator_support/to_dim_order_copy_support.py +119 -0
  282. executorch/backends/arm/operator_support/tosa_profile_supported_op_lists.py +285 -0
  283. executorch/backends/arm/operator_support/tosa_supported_operators.py +1154 -0
  284. executorch/backends/arm/operator_support/unfold_copy_support.py +146 -0
  285. executorch/backends/arm/operator_support/upsample_support.py +91 -0
  286. executorch/backends/arm/operator_support/where_support.py +76 -0
  287. executorch/backends/arm/operators/__init__.py +79 -0
  288. executorch/backends/arm/operators/node_visitor.py +261 -0
  289. executorch/backends/arm/operators/op_amax.py +80 -0
  290. executorch/backends/arm/operators/op_amin.py +79 -0
  291. executorch/backends/arm/operators/op_any.py +59 -0
  292. executorch/backends/arm/operators/op_cond_if.py +67 -0
  293. executorch/backends/arm/operators/op_flip.py +77 -0
  294. executorch/backends/arm/operators/op_sum.py +58 -0
  295. executorch/backends/arm/operators/op_to_dim_order_copy.py +47 -0
  296. executorch/backends/arm/operators/op_tosa_abs.py +26 -0
  297. executorch/backends/arm/operators/op_tosa_add.py +26 -0
  298. executorch/backends/arm/operators/op_tosa_argmax.py +63 -0
  299. executorch/backends/arm/operators/op_tosa_avg_pool2d.py +73 -0
  300. executorch/backends/arm/operators/op_tosa_avg_pool2d_adaptive.py +71 -0
  301. executorch/backends/arm/operators/op_tosa_binary_ops.py +93 -0
  302. executorch/backends/arm/operators/op_tosa_bitwise_not.py +30 -0
  303. executorch/backends/arm/operators/op_tosa_cast_to_block_scaled.py +141 -0
  304. executorch/backends/arm/operators/op_tosa_ceil.py +30 -0
  305. executorch/backends/arm/operators/op_tosa_clamp.py +71 -0
  306. executorch/backends/arm/operators/op_tosa_clz.py +30 -0
  307. executorch/backends/arm/operators/op_tosa_concat.py +73 -0
  308. executorch/backends/arm/operators/op_tosa_conv2d.py +140 -0
  309. executorch/backends/arm/operators/op_tosa_conv2d_block_scaled.py +114 -0
  310. executorch/backends/arm/operators/op_tosa_conv3d.py +23 -0
  311. executorch/backends/arm/operators/op_tosa_cos.py +30 -0
  312. executorch/backends/arm/operators/op_tosa_custom.py +186 -0
  313. executorch/backends/arm/operators/op_tosa_depthwise_conv2d.py +25 -0
  314. executorch/backends/arm/operators/op_tosa_eq.py +29 -0
  315. executorch/backends/arm/operators/op_tosa_erf.py +30 -0
  316. executorch/backends/arm/operators/op_tosa_exp.py +30 -0
  317. executorch/backends/arm/operators/op_tosa_fft.py +76 -0
  318. executorch/backends/arm/operators/op_tosa_floor.py +30 -0
  319. executorch/backends/arm/operators/op_tosa_gather.py +90 -0
  320. executorch/backends/arm/operators/op_tosa_ge.py +31 -0
  321. executorch/backends/arm/operators/op_tosa_gt.py +31 -0
  322. executorch/backends/arm/operators/op_tosa_identity.py +38 -0
  323. executorch/backends/arm/operators/op_tosa_log.py +30 -0
  324. executorch/backends/arm/operators/op_tosa_logical_not.py +26 -0
  325. executorch/backends/arm/operators/op_tosa_matmul.py +98 -0
  326. executorch/backends/arm/operators/op_tosa_matmul_t_block_scaled.py +100 -0
  327. executorch/backends/arm/operators/op_tosa_max_pool2d.py +71 -0
  328. executorch/backends/arm/operators/op_tosa_max_pool2d_adaptive.py +60 -0
  329. executorch/backends/arm/operators/op_tosa_maximum.py +27 -0
  330. executorch/backends/arm/operators/op_tosa_minimum.py +27 -0
  331. executorch/backends/arm/operators/op_tosa_mul.py +60 -0
  332. executorch/backends/arm/operators/op_tosa_neg.py +81 -0
  333. executorch/backends/arm/operators/op_tosa_pad.py +78 -0
  334. executorch/backends/arm/operators/op_tosa_pow.py +30 -0
  335. executorch/backends/arm/operators/op_tosa_reciprocal.py +30 -0
  336. executorch/backends/arm/operators/op_tosa_rescale.py +265 -0
  337. executorch/backends/arm/operators/op_tosa_reshape.py +72 -0
  338. executorch/backends/arm/operators/op_tosa_resize.py +57 -0
  339. executorch/backends/arm/operators/op_tosa_reverse.py +64 -0
  340. executorch/backends/arm/operators/op_tosa_rshift_tensor.py +57 -0
  341. executorch/backends/arm/operators/op_tosa_rsqrt.py +30 -0
  342. executorch/backends/arm/operators/op_tosa_scatter.py +77 -0
  343. executorch/backends/arm/operators/op_tosa_shapes.py +257 -0
  344. executorch/backends/arm/operators/op_tosa_sigmoid.py +30 -0
  345. executorch/backends/arm/operators/op_tosa_sin.py +30 -0
  346. executorch/backends/arm/operators/op_tosa_slice.py +69 -0
  347. executorch/backends/arm/operators/op_tosa_sub.py +26 -0
  348. executorch/backends/arm/operators/op_tosa_table.py +68 -0
  349. executorch/backends/arm/operators/op_tosa_tanh.py +30 -0
  350. executorch/backends/arm/operators/op_tosa_tile.py +68 -0
  351. executorch/backends/arm/operators/op_tosa_transpose.py +71 -0
  352. executorch/backends/arm/operators/op_tosa_transpose_conv2d.py +132 -0
  353. executorch/backends/arm/operators/op_where.py +69 -0
  354. executorch/backends/arm/operators/op_while.py +105 -0
  355. executorch/backends/arm/operators/operator_validation_utils.py +209 -0
  356. executorch/backends/arm/operators/ops_quant_utils.py +36 -0
  357. executorch/backends/arm/operators/simple_node_visitor.py +70 -0
  358. executorch/backends/arm/process_node.py +326 -0
  359. executorch/backends/arm/quantizer/__init__.py +37 -0
  360. executorch/backends/arm/quantizer/arm_quantizer.py +1453 -0
  361. executorch/backends/arm/quantizer/arm_quantizer_utils.py +768 -0
  362. executorch/backends/arm/quantizer/quantization_annotator.py +1084 -0
  363. executorch/backends/arm/quantizer/quantization_config.py +424 -0
  364. executorch/backends/arm/quantizer/quantizer_support.py +246 -0
  365. executorch/backends/arm/scripts/aot_arm_compiler.py +1194 -0
  366. executorch/backends/arm/scripts/docgen/docgen.py +271 -0
  367. executorch/backends/arm/scripts/etdump_to_chrome_trace.py +109 -0
  368. executorch/backends/arm/scripts/evaluate_model.py +395 -0
  369. executorch/backends/arm/scripts/generate_grid_sampler_spirv.py +81 -0
  370. executorch/backends/arm/scripts/get_ethosu_scratch_from_pte.py +77 -0
  371. executorch/backends/arm/scripts/public_api_manifest/__init__.py +4 -0
  372. executorch/backends/arm/scripts/public_api_manifest/generate_public_api_manifest.py +155 -0
  373. executorch/backends/arm/scripts/public_api_manifest/validate_public_api_manifest.py +358 -0
  374. executorch/backends/arm/test/_custom_vgf_test_utils.py +999 -0
  375. executorch/backends/arm/test/common.py +357 -0
  376. executorch/backends/arm/test/conftest.py +149 -0
  377. executorch/backends/arm/test/deprecation/test_arm_compile_spec_deprecation.py +46 -0
  378. executorch/backends/arm/test/misc/test_bn_relu_folding_qat.py +124 -0
  379. executorch/backends/arm/test/misc/test_call_operator_submodule.py +72 -0
  380. executorch/backends/arm/test/misc/test_compile_spec.py +125 -0
  381. executorch/backends/arm/test/misc/test_const_shape.py +58 -0
  382. executorch/backends/arm/test/misc/test_conv_relu_residual_add.py +108 -0
  383. executorch/backends/arm/test/misc/test_convolution_operator_support.py +218 -0
  384. executorch/backends/arm/test/misc/test_count_tosa_ops.py +41 -0
  385. executorch/backends/arm/test/misc/test_custom_partition.py +183 -0
  386. executorch/backends/arm/test/misc/test_custom_shader_payload.py +371 -0
  387. executorch/backends/arm/test/misc/test_custom_shader_payloads.py +178 -0
  388. executorch/backends/arm/test/misc/test_debug_feats.py +416 -0
  389. executorch/backends/arm/test/misc/test_debug_hook.py +186 -0
  390. executorch/backends/arm/test/misc/test_dim_order.py +122 -0
  391. executorch/backends/arm/test/misc/test_dw_convs_with_shared_weights.py +46 -0
  392. executorch/backends/arm/test/misc/test_evaluate_model.py +72 -0
  393. executorch/backends/arm/test/misc/test_extract_io_params_tosa.py +115 -0
  394. executorch/backends/arm/test/misc/test_high_rank_permute_view_invariants.py +186 -0
  395. executorch/backends/arm/test/misc/test_int64.py +112 -0
  396. executorch/backends/arm/test/misc/test_lifted_tensor.py +146 -0
  397. executorch/backends/arm/test/misc/test_mixed_fp_bf16_partition.py +123 -0
  398. executorch/backends/arm/test/misc/test_mixed_type_lowering.py +83 -0
  399. executorch/backends/arm/test/misc/test_model_evaluator.py +54 -0
  400. executorch/backends/arm/test/misc/test_multiple_delegates.py +71 -0
  401. executorch/backends/arm/test/misc/test_multiple_outputs.py +70 -0
  402. executorch/backends/arm/test/misc/test_mxfp_conv2d_ao.py +318 -0
  403. executorch/backends/arm/test/misc/test_mxfp_linear_ao.py +207 -0
  404. executorch/backends/arm/test/misc/test_non_persistent_buffers.py +45 -0
  405. executorch/backends/arm/test/misc/test_partition_cycle_detection.py +92 -0
  406. executorch/backends/arm/test/misc/test_partition_decomposed_quantized_ops.py +188 -0
  407. executorch/backends/arm/test/misc/test_pass_pipeline_config.py +210 -0
  408. executorch/backends/arm/test/misc/test_pass_required_order.py +97 -0
  409. executorch/backends/arm/test/misc/test_post_quant_device_switch.py +232 -0
  410. executorch/backends/arm/test/misc/test_process_node.py +172 -0
  411. executorch/backends/arm/test/misc/test_public_api_manifest.py +117 -0
  412. executorch/backends/arm/test/misc/test_qat_training_loop.py +104 -0
  413. executorch/backends/arm/test/misc/test_quant_custom_meta.py +110 -0
  414. executorch/backends/arm/test/misc/test_rescale_range.py +206 -0
  415. executorch/backends/arm/test/misc/test_reshape_permute_5d_regression.py +189 -0
  416. executorch/backends/arm/test/misc/test_runner_utils.py +231 -0
  417. executorch/backends/arm/test/misc/test_save_exported_model.py +64 -0
  418. executorch/backends/arm/test/misc/test_shared_qspecs.py +689 -0
  419. executorch/backends/arm/test/misc/test_tosa_data_layout_visitors.py +61 -0
  420. executorch/backends/arm/test/misc/test_tosa_dialect_activation.py +195 -0
  421. executorch/backends/arm/test/misc/test_tosa_dialect_argmax.py +43 -0
  422. executorch/backends/arm/test/misc/test_tosa_dialect_fft.py +121 -0
  423. executorch/backends/arm/test/misc/test_tosa_dialect_max_pool2d_adaptive.py +128 -0
  424. executorch/backends/arm/test/misc/test_tosa_dialect_scatter.py +38 -0
  425. executorch/backends/arm/test/misc/test_tosa_dialect_unary_ops.py +394 -0
  426. executorch/backends/arm/test/misc/test_tosa_operator_support.py +89 -0
  427. executorch/backends/arm/test/misc/test_tosa_shape_node_visitors.py +583 -0
  428. executorch/backends/arm/test/misc/test_tosa_shape_support.py +170 -0
  429. executorch/backends/arm/test/misc/test_tosa_spec.py +227 -0
  430. executorch/backends/arm/test/misc/test_transpose_counts.py +580 -0
  431. executorch/backends/arm/test/misc/test_validate_public_api_manifest.py +262 -0
  432. executorch/backends/arm/test/misc/test_vgf_backend.py +332 -0
  433. executorch/backends/arm/test/misc/test_vgf_check_env.py +442 -0
  434. executorch/backends/arm/test/misc/test_vgf_smoke.py +61 -0
  435. executorch/backends/arm/test/misc/tosa_dialect/test_data_layout_ops.py +180 -0
  436. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_avg_pool2d_adaptive.py +324 -0
  437. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_conv2d.py +220 -0
  438. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_dialect_binary_ops.py +439 -0
  439. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_dialect_cast_to_block_scaled.py +106 -0
  440. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_dialect_conversion_ops.py +35 -0
  441. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_dialect_mxfp_conv2d.py +91 -0
  442. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_dialect_mxfp_linear.py +90 -0
  443. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_dw_conv2d.py +227 -0
  444. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_gather.py +44 -0
  445. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_identity.py +34 -0
  446. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_reduction_ops.py +134 -0
  447. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_rescale.py +60 -0
  448. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_resize.py +154 -0
  449. executorch/backends/arm/test/misc/tosa_dialect/test_tosa_shape_ops.py +815 -0
  450. executorch/backends/arm/test/models/Gemma3n/gemma3n_test_config.py +37 -0
  451. executorch/backends/arm/test/models/Gemma3n/test_gemma3nModel.py +811 -0
  452. executorch/backends/arm/test/models/Qwen3_VL/qwen3_vl_test_config.py +64 -0
  453. executorch/backends/arm/test/models/Qwen3_VL/test_qwen3_vl_layers.py +628 -0
  454. executorch/backends/arm/test/models/Qwen3_VL/test_qwen3_vl_model.py +407 -0
  455. executorch/backends/arm/test/models/phi3/phi3_module_test_configs.py +24 -0
  456. executorch/backends/arm/test/models/phi3/test_phi3_layers.py +277 -0
  457. executorch/backends/arm/test/models/stable_diffusion/stable_diffusion_module_test_configs.py +113 -0
  458. executorch/backends/arm/test/models/stable_diffusion/test_CLIPTextModelWithProjection.py +166 -0
  459. executorch/backends/arm/test/models/stable_diffusion/test_SD3Transformer2DModel.py +188 -0
  460. executorch/backends/arm/test/models/stable_diffusion/test_T5EncoderModel.py +165 -0
  461. executorch/backends/arm/test/models/stable_diffusion/test_vae_AutoencoderKL.py +121 -0
  462. executorch/backends/arm/test/models/test_T5ForConditionalGeneration_arm.py +170 -0
  463. executorch/backends/arm/test/models/test_conformer.py +144 -0
  464. executorch/backends/arm/test/models/test_deep_autoencoder.py +123 -0
  465. executorch/backends/arm/test/models/test_deit_tiny_arm.py +122 -0
  466. executorch/backends/arm/test/models/test_dl3_arm.py +120 -0
  467. executorch/backends/arm/test/models/test_ds_cnn.py +97 -0
  468. executorch/backends/arm/test/models/test_inception_v3_arm.py +139 -0
  469. executorch/backends/arm/test/models/test_llama.py +349 -0
  470. executorch/backends/arm/test/models/test_lstm_arm.py +198 -0
  471. executorch/backends/arm/test/models/test_mobilenet_v1_025.py +93 -0
  472. executorch/backends/arm/test/models/test_mobilenet_v2_arm.py +240 -0
  473. executorch/backends/arm/test/models/test_mobilenet_v3_arm.py +138 -0
  474. executorch/backends/arm/test/models/test_nn_functional.py +130 -0
  475. executorch/backends/arm/test/models/test_nn_modules.py +174 -0
  476. executorch/backends/arm/test/models/test_nss.py +148 -0
  477. executorch/backends/arm/test/models/test_residual_conv_block.py +143 -0
  478. executorch/backends/arm/test/models/test_resnet18.py +133 -0
  479. executorch/backends/arm/test/models/test_resnet8.py +91 -0
  480. executorch/backends/arm/test/models/test_swin2sr_arm.py +132 -0
  481. executorch/backends/arm/test/models/test_torch_functions.py +155 -0
  482. executorch/backends/arm/test/models/test_w2l_arm.py +144 -0
  483. executorch/backends/arm/test/modules/test_gru.py +178 -0
  484. executorch/backends/arm/test/modules/test_lstm.py +199 -0
  485. executorch/backends/arm/test/modules/test_rnn.py +242 -0
  486. executorch/backends/arm/test/modules/test_static_cache.py +228 -0
  487. executorch/backends/arm/test/ops/mxfp/__init__.py +4 -0
  488. executorch/backends/arm/test/ops/mxfp/common.py +122 -0
  489. executorch/backends/arm/test/ops/mxfp/test_mxfp_linear.py +437 -0
  490. executorch/backends/arm/test/ops/test_abs.py +117 -0
  491. executorch/backends/arm/test/ops/test_acos.py +121 -0
  492. executorch/backends/arm/test/ops/test_acosh.py +137 -0
  493. executorch/backends/arm/test/ops/test_adaptive_avg_pool2d.py +294 -0
  494. executorch/backends/arm/test/ops/test_add.py +550 -0
  495. executorch/backends/arm/test/ops/test_addmm.py +275 -0
  496. executorch/backends/arm/test/ops/test_alias_copy.py +127 -0
  497. executorch/backends/arm/test/ops/test_amax.py +319 -0
  498. executorch/backends/arm/test/ops/test_amin.py +326 -0
  499. executorch/backends/arm/test/ops/test_any.py +231 -0
  500. executorch/backends/arm/test/ops/test_arange.py +258 -0
  501. executorch/backends/arm/test/ops/test_argmax.py +133 -0
  502. executorch/backends/arm/test/ops/test_as_strided_copy.py +138 -0
  503. executorch/backends/arm/test/ops/test_asin.py +105 -0
  504. executorch/backends/arm/test/ops/test_asinh.py +104 -0
  505. executorch/backends/arm/test/ops/test_at.py +117 -0
  506. executorch/backends/arm/test/ops/test_atan.py +111 -0
  507. executorch/backends/arm/test/ops/test_atanh.py +116 -0
  508. executorch/backends/arm/test/ops/test_avg_pool2d.py +382 -0
  509. executorch/backends/arm/test/ops/test_batch_norm.py +397 -0
  510. executorch/backends/arm/test/ops/test_bitwise.py +769 -0
  511. executorch/backends/arm/test/ops/test_bitwise_not.py +135 -0
  512. executorch/backends/arm/test/ops/test_bmm.py +240 -0
  513. executorch/backends/arm/test/ops/test_cat.py +297 -0
  514. executorch/backends/arm/test/ops/test_ceil.py +152 -0
  515. executorch/backends/arm/test/ops/test_clamp.py +584 -0
  516. executorch/backends/arm/test/ops/test_clone.py +150 -0
  517. executorch/backends/arm/test/ops/test_cond.py +313 -0
  518. executorch/backends/arm/test/ops/test_constant_pad_nd.py +390 -0
  519. executorch/backends/arm/test/ops/test_conv1d.py +433 -0
  520. executorch/backends/arm/test/ops/test_conv2d.py +872 -0
  521. executorch/backends/arm/test/ops/test_conv3d.py +908 -0
  522. executorch/backends/arm/test/ops/test_conv_combos.py +667 -0
  523. executorch/backends/arm/test/ops/test_conv_constant_pad_nd.py +161 -0
  524. executorch/backends/arm/test/ops/test_copy.py +171 -0
  525. executorch/backends/arm/test/ops/test_cos.py +136 -0
  526. executorch/backends/arm/test/ops/test_cosh.py +111 -0
  527. executorch/backends/arm/test/ops/test_cumsum.py +120 -0
  528. executorch/backends/arm/test/ops/test_custom_shader_lowering.py +287 -0
  529. executorch/backends/arm/test/ops/test_depthwise_conv.py +589 -0
  530. executorch/backends/arm/test/ops/test_detach_copy.py +56 -0
  531. executorch/backends/arm/test/ops/test_div.py +209 -0
  532. executorch/backends/arm/test/ops/test_div_tensor_mode.py +235 -0
  533. executorch/backends/arm/test/ops/test_elu.py +241 -0
  534. executorch/backends/arm/test/ops/test_embedding.py +173 -0
  535. executorch/backends/arm/test/ops/test_eq.py +371 -0
  536. executorch/backends/arm/test/ops/test_erf.py +121 -0
  537. executorch/backends/arm/test/ops/test_erfinv.py +121 -0
  538. executorch/backends/arm/test/ops/test_exp.py +172 -0
  539. executorch/backends/arm/test/ops/test_expand.py +153 -0
  540. executorch/backends/arm/test/ops/test_expm1.py +113 -0
  541. executorch/backends/arm/test/ops/test_eye.py +156 -0
  542. executorch/backends/arm/test/ops/test_fft.py +62 -0
  543. executorch/backends/arm/test/ops/test_fill_scalar.py +108 -0
  544. executorch/backends/arm/test/ops/test_flip.py +226 -0
  545. executorch/backends/arm/test/ops/test_floor.py +148 -0
  546. executorch/backends/arm/test/ops/test_floor_div.py +162 -0
  547. executorch/backends/arm/test/ops/test_full.py +257 -0
  548. executorch/backends/arm/test/ops/test_gather.py +259 -0
  549. executorch/backends/arm/test/ops/test_ge.py +379 -0
  550. executorch/backends/arm/test/ops/test_gelu.py +215 -0
  551. executorch/backends/arm/test/ops/test_glu.py +134 -0
  552. executorch/backends/arm/test/ops/test_grid_sampler.py +62 -0
  553. executorch/backends/arm/test/ops/test_group_norm.py +175 -0
  554. executorch/backends/arm/test/ops/test_gt.py +383 -0
  555. executorch/backends/arm/test/ops/test_hardsigmoid.py +114 -0
  556. executorch/backends/arm/test/ops/test_hardswish.py +104 -0
  557. executorch/backends/arm/test/ops/test_hardtanh.py +113 -0
  558. executorch/backends/arm/test/ops/test_index_copy.py +216 -0
  559. executorch/backends/arm/test/ops/test_index_put.py +519 -0
  560. executorch/backends/arm/test/ops/test_index_select.py +231 -0
  561. executorch/backends/arm/test/ops/test_index_tensor.py +624 -0
  562. executorch/backends/arm/test/ops/test_layer_norm.py +213 -0
  563. executorch/backends/arm/test/ops/test_le.py +355 -0
  564. executorch/backends/arm/test/ops/test_leaky_relu.py +121 -0
  565. executorch/backends/arm/test/ops/test_linalg_vector_norm.py +161 -0
  566. executorch/backends/arm/test/ops/test_linear.py +409 -0
  567. executorch/backends/arm/test/ops/test_log.py +126 -0
  568. executorch/backends/arm/test/ops/test_log10.py +87 -0
  569. executorch/backends/arm/test/ops/test_log1p.py +78 -0
  570. executorch/backends/arm/test/ops/test_logical.py +429 -0
  571. executorch/backends/arm/test/ops/test_logit.py +122 -0
  572. executorch/backends/arm/test/ops/test_logsoftmax.py +124 -0
  573. executorch/backends/arm/test/ops/test_lshift.py +220 -0
  574. executorch/backends/arm/test/ops/test_lt.py +356 -0
  575. executorch/backends/arm/test/ops/test_masked_fill.py +171 -0
  576. executorch/backends/arm/test/ops/test_matmul.py +557 -0
  577. executorch/backends/arm/test/ops/test_max_pool.py +449 -0
  578. executorch/backends/arm/test/ops/test_max_pool1d.py +164 -0
  579. executorch/backends/arm/test/ops/test_maximum.py +195 -0
  580. executorch/backends/arm/test/ops/test_mean_dim.py +487 -0
  581. executorch/backends/arm/test/ops/test_minimum.py +195 -0
  582. executorch/backends/arm/test/ops/test_mm.py +93 -0
  583. executorch/backends/arm/test/ops/test_mul.py +416 -0
  584. executorch/backends/arm/test/ops/test_multihead_attention.py +131 -0
  585. executorch/backends/arm/test/ops/test_mxfp_conv2d.py +445 -0
  586. executorch/backends/arm/test/ops/test_ne.py +253 -0
  587. executorch/backends/arm/test/ops/test_neg.py +166 -0
  588. executorch/backends/arm/test/ops/test_ones.py +152 -0
  589. executorch/backends/arm/test/ops/test_permute.py +263 -0
  590. executorch/backends/arm/test/ops/test_pixel_shuffling.py +235 -0
  591. executorch/backends/arm/test/ops/test_pow.py +240 -0
  592. executorch/backends/arm/test/ops/test_reciprocal.py +176 -0
  593. executorch/backends/arm/test/ops/test_relu.py +166 -0
  594. executorch/backends/arm/test/ops/test_remainder.py +208 -0
  595. executorch/backends/arm/test/ops/test_repeat.py +271 -0
  596. executorch/backends/arm/test/ops/test_round.py +114 -0
  597. executorch/backends/arm/test/ops/test_rshift.py +226 -0
  598. executorch/backends/arm/test/ops/test_rsqrt.py +212 -0
  599. executorch/backends/arm/test/ops/test_rsub.py +121 -0
  600. executorch/backends/arm/test/ops/test_scalar_tensor.py +135 -0
  601. executorch/backends/arm/test/ops/test_scalars.py +396 -0
  602. executorch/backends/arm/test/ops/test_sdpa.py +154 -0
  603. executorch/backends/arm/test/ops/test_select.py +204 -0
  604. executorch/backends/arm/test/ops/test_select_scatter.py +188 -0
  605. executorch/backends/arm/test/ops/test_sigmoid.py +358 -0
  606. executorch/backends/arm/test/ops/test_sigmoid_32bit.py +219 -0
  607. executorch/backends/arm/test/ops/test_sign.py +111 -0
  608. executorch/backends/arm/test/ops/test_silu.py +193 -0
  609. executorch/backends/arm/test/ops/test_sin.py +130 -0
  610. executorch/backends/arm/test/ops/test_sinh.py +103 -0
  611. executorch/backends/arm/test/ops/test_slice.py +410 -0
  612. executorch/backends/arm/test/ops/test_slice_scatter.py +342 -0
  613. executorch/backends/arm/test/ops/test_softmax.py +239 -0
  614. executorch/backends/arm/test/ops/test_split.py +347 -0
  615. executorch/backends/arm/test/ops/test_sqrt.py +112 -0
  616. executorch/backends/arm/test/ops/test_squeeze.py +318 -0
  617. executorch/backends/arm/test/ops/test_stack.py +150 -0
  618. executorch/backends/arm/test/ops/test_sub.py +547 -0
  619. executorch/backends/arm/test/ops/test_sum.py +303 -0
  620. executorch/backends/arm/test/ops/test_t_copy.py +115 -0
  621. executorch/backends/arm/test/ops/test_tan.py +146 -0
  622. executorch/backends/arm/test/ops/test_tanh.py +198 -0
  623. executorch/backends/arm/test/ops/test_to_copy.py +655 -0
  624. executorch/backends/arm/test/ops/test_transpose_conv2d.py +670 -0
  625. executorch/backends/arm/test/ops/test_transpose_copy.py +114 -0
  626. executorch/backends/arm/test/ops/test_tril.py +177 -0
  627. executorch/backends/arm/test/ops/test_unary_combos.py +143 -0
  628. executorch/backends/arm/test/ops/test_unbind.py +82 -0
  629. executorch/backends/arm/test/ops/test_unflatten.py +110 -0
  630. executorch/backends/arm/test/ops/test_unfold_copy.py +262 -0
  631. executorch/backends/arm/test/ops/test_unsqueeze.py +107 -0
  632. executorch/backends/arm/test/ops/test_upsample_bilinear2d.py +717 -0
  633. executorch/backends/arm/test/ops/test_upsample_nearest2d.py +656 -0
  634. executorch/backends/arm/test/ops/test_var.py +441 -0
  635. executorch/backends/arm/test/ops/test_view.py +261 -0
  636. executorch/backends/arm/test/ops/test_where.py +319 -0
  637. executorch/backends/arm/test/ops/test_while.py +311 -0
  638. executorch/backends/arm/test/ops/test_zeros.py +158 -0
  639. executorch/backends/arm/test/passes/test_arm_op_targeted_pass.py +150 -0
  640. executorch/backends/arm/test/passes/test_arm_pass_manager_insertions.py +288 -0
  641. executorch/backends/arm/test/passes/test_broadcast_args_pass.py +57 -0
  642. executorch/backends/arm/test/passes/test_canonicalize_view_copy_permute_pass.py +573 -0
  643. executorch/backends/arm/test/passes/test_cast_int64_pass.py +81 -0
  644. executorch/backends/arm/test/passes/test_convert_expand_copy_to_repeat.py +49 -0
  645. executorch/backends/arm/test/passes/test_convert_int64_const_ops_to_int32.py +527 -0
  646. executorch/backends/arm/test/passes/test_convert_int64_output_ops_to_int32.py +173 -0
  647. executorch/backends/arm/test/passes/test_convert_permute_singleton_to_view_pass.py +100 -0
  648. executorch/backends/arm/test/passes/test_convert_split_to_slice.py +71 -0
  649. executorch/backends/arm/test/passes/test_convert_to_clamp.py +92 -0
  650. executorch/backends/arm/test/passes/test_custom_op_rewrite.py +257 -0
  651. executorch/backends/arm/test/passes/test_decompose_adaptive_avg_pool2d_pass.py +191 -0
  652. executorch/backends/arm/test/passes/test_decompose_adaptive_max_pool2d_pass.py +80 -0
  653. executorch/backends/arm/test/passes/test_decompose_avg_pool2d_pass.py +80 -0
  654. executorch/backends/arm/test/passes/test_decompose_cosine_similarity_pass.py +58 -0
  655. executorch/backends/arm/test/passes/test_decompose_div_pass.py +66 -0
  656. executorch/backends/arm/test/passes/test_decompose_dynamic_adaptive_avg_pool2d_pass.py +93 -0
  657. executorch/backends/arm/test/passes/test_decompose_dynamic_full_pass.py +176 -0
  658. executorch/backends/arm/test/passes/test_decompose_einsum_pass.py +73 -0
  659. executorch/backends/arm/test/passes/test_decompose_int_pow_pass.py +107 -0
  660. executorch/backends/arm/test/passes/test_decompose_layernorm_pass.py +67 -0
  661. executorch/backends/arm/test/passes/test_decompose_linalg_vector_norm_pass.py +109 -0
  662. executorch/backends/arm/test/passes/test_decompose_linear_pass.py +54 -0
  663. executorch/backends/arm/test/passes/test_decompose_meandim_pass.py +133 -0
  664. executorch/backends/arm/test/passes/test_decompose_quant_nodes.py +44 -0
  665. executorch/backends/arm/test/passes/test_decompose_select_pass.py +73 -0
  666. executorch/backends/arm/test/passes/test_decompose_softmax_pass.py +132 -0
  667. executorch/backends/arm/test/passes/test_decompose_tosa_unsupported_clamp_pass.py +73 -0
  668. executorch/backends/arm/test/passes/test_decompose_unsupported_bilinear_resize_pass.py +249 -0
  669. executorch/backends/arm/test/passes/test_decompose_var_pass.py +85 -0
  670. executorch/backends/arm/test/passes/test_decorate_fp32_to_int32_casting_pass.py +87 -0
  671. executorch/backends/arm/test/passes/test_deduplicate_get_attr_pass.py +37 -0
  672. executorch/backends/arm/test/passes/test_dim_maps.py +708 -0
  673. executorch/backends/arm/test/passes/test_ensure_unique_output_nodes_pass.py +64 -0
  674. executorch/backends/arm/test/passes/test_fold_qdq_pass.py +51 -0
  675. executorch/backends/arm/test/passes/test_fold_scalar_mul_into_conv_pass.py +196 -0
  676. executorch/backends/arm/test/passes/test_fuse_batchnorm_pass.py +192 -0
  677. executorch/backends/arm/test/passes/test_fuse_consecutive_concat_shapes.py +110 -0
  678. executorch/backends/arm/test/passes/test_fuse_consecutive_slices_pass.py +110 -0
  679. executorch/backends/arm/test/passes/test_fuse_constant_ops_pass.py +504 -0
  680. executorch/backends/arm/test/passes/test_fuse_duplicate_users_pass.py +169 -0
  681. executorch/backends/arm/test/passes/test_fuse_equal_placeholders_ops_pass.py +170 -0
  682. executorch/backends/arm/test/passes/test_fuse_identical_input_transforms_pass.py +860 -0
  683. executorch/backends/arm/test/passes/test_fuse_quantized_activation_pass.py +48 -0
  684. executorch/backends/arm/test/passes/test_fuse_view_copy.py +82 -0
  685. executorch/backends/arm/test/passes/test_insert_data_layout_casts_pass.py +140 -0
  686. executorch/backends/arm/test/passes/test_insert_dynamic_padding_pass.py +124 -0
  687. executorch/backends/arm/test/passes/test_insert_int32_casts_after_int64_placeholders_pass.py +109 -0
  688. executorch/backends/arm/test/passes/test_insert_rescale_i32_pass.py +107 -0
  689. executorch/backends/arm/test/passes/test_insert_table_ops_pass.py +100 -0
  690. executorch/backends/arm/test/passes/test_int32_cast_embedding_pass.py +46 -0
  691. executorch/backends/arm/test/passes/test_ioquantization_pass.py +617 -0
  692. executorch/backends/arm/test/passes/test_permute_view_swap.py +230 -0
  693. executorch/backends/arm/test/passes/test_promote_bool_operands_pass.py +101 -0
  694. executorch/backends/arm/test/passes/test_propagate_permutes_views_pass.py +1515 -0
  695. executorch/backends/arm/test/passes/test_rank_match_after_propagation.py +51 -0
  696. executorch/backends/arm/test/passes/test_remove_clone_pass.py +43 -0
  697. executorch/backends/arm/test/passes/test_remove_permutes_around_elementwise_tosa_ops.py +196 -0
  698. executorch/backends/arm/test/passes/test_replace_inf_values_pass.py +115 -0
  699. executorch/backends/arm/test/passes/test_rescale_optimization.py +244 -0
  700. executorch/backends/arm/test/passes/test_rescale_pass.py +95 -0
  701. executorch/backends/arm/test/passes/test_rewrite_adaptive_avg_pool2d_pass.py +328 -0
  702. executorch/backends/arm/test/passes/test_rewrite_avg_pool2d_pass.py +122 -0
  703. executorch/backends/arm/test/passes/test_rewrite_conv_pass.py +476 -0
  704. executorch/backends/arm/test/passes/test_rewrite_grid_sampler_to_tosa_custom_pass.py +455 -0
  705. executorch/backends/arm/test/passes/test_rewrite_le_lt_to_ge_gt_pass.py +51 -0
  706. executorch/backends/arm/test/passes/test_rewrite_max_pool2d_pass.py +198 -0
  707. executorch/backends/arm/test/passes/test_rewrite_mxfp_conv2d_pass.py +195 -0
  708. executorch/backends/arm/test/passes/test_rewrite_mxfp_linear_pass.py +177 -0
  709. executorch/backends/arm/test/passes/test_rewrite_upsample_pass.py +44 -0
  710. executorch/backends/arm/test/passes/test_size_adjust_input_pass.py +459 -0
  711. executorch/backends/arm/test/passes/test_symbolic_to_tosa_shape_pass.py +101 -0
  712. executorch/backends/arm/test/passes/test_symbolic_value_range.py +83 -0
  713. executorch/backends/arm/test/passes/test_unsqueeze_before_repeat_pass.py +59 -0
  714. executorch/backends/arm/test/passes/test_view_permute_swap.py +227 -0
  715. executorch/backends/arm/test/public_api_bc/run_public_api_bc_scenarios.py +104 -0
  716. executorch/backends/arm/test/public_api_bc/test_ethosu_flow.py +129 -0
  717. executorch/backends/arm/test/public_api_bc/test_vgf_fp_flow.py +75 -0
  718. executorch/backends/arm/test/public_api_bc/test_vgf_int_flow.py +113 -0
  719. executorch/backends/arm/test/quantizer/test_conv_relu_fusing.py +118 -0
  720. executorch/backends/arm/test/quantizer/test_generic_annotater.py +251 -0
  721. executorch/backends/arm/test/quantizer/test_partial_quantization.py +226 -0
  722. executorch/backends/arm/test/quantizer/test_preserve_kwargs.py +55 -0
  723. executorch/backends/arm/test/quantizer/test_selective_quantization.py +349 -0
  724. executorch/backends/arm/test/quantizer/test_set_module_name.py +158 -0
  725. executorch/backends/arm/test/quantizer/test_tosa_quantizer_validate.py +55 -0
  726. executorch/backends/arm/test/quantizer/test_uint8_io_quantization.py +177 -0
  727. executorch/backends/arm/test/runner_utils.py +1170 -0
  728. executorch/backends/arm/test/runtime/_vgf_runtime_test_utils.py +350 -0
  729. executorch/backends/arm/test/runtime/test_vgf_aliasing_runtime.py +133 -0
  730. executorch/backends/arm/test/runtime/test_vgf_combinations_runtime.py +465 -0
  731. executorch/backends/arm/test/runtime/test_vgf_multi_segment_runtime.py +153 -0
  732. executorch/backends/arm/test/runtime/test_vgf_sampler_image_runtime.py +110 -0
  733. executorch/backends/arm/test/runtime/test_vgf_tensor_buffer_runtime.py +165 -0
  734. executorch/backends/arm/test/test_memory_allocator_log.py +170 -0
  735. executorch/backends/arm/test/test_model.py +368 -0
  736. executorch/backends/arm/test/tester/__init__.py +6 -0
  737. executorch/backends/arm/test/tester/analyze_output_utils.py +473 -0
  738. executorch/backends/arm/test/tester/arm_tester.py +1408 -0
  739. executorch/backends/arm/test/tester/quantize.py +63 -0
  740. executorch/backends/arm/test/tester/serialize.py +96 -0
  741. executorch/backends/arm/test/tester/test_pipeline.py +1369 -0
  742. executorch/backends/arm/tosa/__init__.py +16 -0
  743. executorch/backends/arm/tosa/backend.py +491 -0
  744. executorch/backends/arm/tosa/cast_support.py +175 -0
  745. executorch/backends/arm/tosa/compile_spec.py +44 -0
  746. executorch/backends/arm/tosa/dialect/__init__.py +33 -0
  747. executorch/backends/arm/tosa/dialect/lib.py +83 -0
  748. executorch/backends/arm/tosa/dialect/ops/_common.py +46 -0
  749. executorch/backends/arm/tosa/dialect/ops/activation.py +138 -0
  750. executorch/backends/arm/tosa/dialect/ops/argmax.py +78 -0
  751. executorch/backends/arm/tosa/dialect/ops/avg_pool2d.py +173 -0
  752. executorch/backends/arm/tosa/dialect/ops/avg_pool2d_adaptive.py +82 -0
  753. executorch/backends/arm/tosa/dialect/ops/binary_elementwise.py +336 -0
  754. executorch/backends/arm/tosa/dialect/ops/conv2d.py +139 -0
  755. executorch/backends/arm/tosa/dialect/ops/conv2d_block_scaled.py +213 -0
  756. executorch/backends/arm/tosa/dialect/ops/conv3d.py +69 -0
  757. executorch/backends/arm/tosa/dialect/ops/conversion_ops.py +213 -0
  758. executorch/backends/arm/tosa/dialect/ops/custom.py +159 -0
  759. executorch/backends/arm/tosa/dialect/ops/data_layout_ops.py +274 -0
  760. executorch/backends/arm/tosa/dialect/ops/depthwise_conv2d.py +49 -0
  761. executorch/backends/arm/tosa/dialect/ops/fft.py +144 -0
  762. executorch/backends/arm/tosa/dialect/ops/gather.py +101 -0
  763. executorch/backends/arm/tosa/dialect/ops/identity.py +16 -0
  764. executorch/backends/arm/tosa/dialect/ops/matmul.py +76 -0
  765. executorch/backends/arm/tosa/dialect/ops/matmul_t_block_scaled.py +162 -0
  766. executorch/backends/arm/tosa/dialect/ops/max_pool2d.py +128 -0
  767. executorch/backends/arm/tosa/dialect/ops/max_pool2d_adaptive.py +70 -0
  768. executorch/backends/arm/tosa/dialect/ops/reduction_ops.py +186 -0
  769. executorch/backends/arm/tosa/dialect/ops/resize.py +116 -0
  770. executorch/backends/arm/tosa/dialect/ops/scatter.py +23 -0
  771. executorch/backends/arm/tosa/dialect/ops/shape_ops.py +506 -0
  772. executorch/backends/arm/tosa/dialect/ops/table.py +51 -0
  773. executorch/backends/arm/tosa/dialect/ops/transpose_conv2d.py +58 -0
  774. executorch/backends/arm/tosa/dialect/ops/unary_elementwise.py +224 -0
  775. executorch/backends/arm/tosa/dialect/ops_registration.py +92 -0
  776. executorch/backends/arm/tosa/dialect/real_impl.py +174 -0
  777. executorch/backends/arm/tosa/dialect/shape.py +21 -0
  778. executorch/backends/arm/tosa/mapping.py +335 -0
  779. executorch/backends/arm/tosa/partitioner.py +772 -0
  780. executorch/backends/arm/tosa/resize_utils.py +341 -0
  781. executorch/backends/arm/tosa/schemas/__init__.py +4 -0
  782. executorch/backends/arm/tosa/schemas/tosa_1.0.fbs +559 -0
  783. executorch/backends/arm/tosa/schemas/tosa_1.1.fbs +701 -0
  784. executorch/backends/arm/tosa/specification.py +514 -0
  785. executorch/backends/arm/tosa/utils.py +173 -0
  786. executorch/backends/arm/util/_factory.py +59 -0
  787. executorch/backends/arm/util/arm_model_evaluator.py +254 -0
  788. executorch/backends/arm/vgf/__init__.py +11 -0
  789. executorch/backends/arm/vgf/_passes/__init__.py +11 -0
  790. executorch/backends/arm/vgf/_passes/insert_grid_sampler_grid_dequant_pass.py +151 -0
  791. executorch/backends/arm/vgf/_passes/rewrite_grid_sampler_to_tosa_custom.py +458 -0
  792. executorch/backends/arm/vgf/backend.py +393 -0
  793. executorch/backends/arm/vgf/check_env.py +678 -0
  794. executorch/backends/arm/vgf/compile_spec.py +108 -0
  795. executorch/backends/arm/vgf/model_converter.py +229 -0
  796. executorch/backends/arm/vgf/partitioner.py +44 -0
  797. executorch/backends/arm/vgf/shaders/__init__.py +4 -0
  798. executorch/backends/arm/vgf/shaders/grid_sampler.glsl +41 -0
  799. executorch/backends/arm/vgf/shaders/grid_sampler.py +340 -0
  800. executorch/backends/arm/vgf/shaders/grid_sampler.spirv.b64 +1 -0
  801. executorch/backends/arm/vgf/shaders/grid_sampler_sampler.glsl +35 -0
  802. executorch/backends/arm/vgf/shaders/grid_sampler_sampler.spirv.b64 +1 -0
  803. executorch/backends/arm/vgf/shaders/grid_sampler_sampler_align_corners.glsl +40 -0
  804. executorch/backends/arm/vgf/shaders/grid_sampler_sampler_align_corners.spirv.b64 +1 -0
  805. executorch/backends/arm/vgf/shaders/grid_sampler_sampler_int8.glsl +44 -0
  806. executorch/backends/arm/vgf/shaders/grid_sampler_sampler_int8.spirv.b64 +1 -0
  807. executorch/backends/arm/vgf/shaders/grid_sampler_sampler_int8_align_corners.glsl +49 -0
  808. executorch/backends/arm/vgf/shaders/grid_sampler_sampler_int8_align_corners.spirv.b64 +1 -0
  809. executorch/backends/cadence/aot/__init__.py +5 -0
  810. executorch/backends/cadence/aot/compiler.py +514 -0
  811. executorch/backends/cadence/aot/compiler_funcs.py +739 -0
  812. executorch/backends/cadence/aot/compiler_utils.py +271 -0
  813. executorch/backends/cadence/aot/decompose_ops.py +120 -0
  814. executorch/backends/cadence/aot/export_example.py +123 -0
  815. executorch/backends/cadence/aot/functions.yaml +505 -0
  816. executorch/backends/cadence/aot/functions_fusion_g3.yaml +191 -0
  817. executorch/backends/cadence/aot/functions_hifi.yaml +590 -0
  818. executorch/backends/cadence/aot/functions_vision.yaml +275 -0
  819. executorch/backends/cadence/aot/fuse_ops.py +1111 -0
  820. executorch/backends/cadence/aot/graph_builder.py +13 -0
  821. executorch/backends/cadence/aot/memory_constraints.py +766 -0
  822. executorch/backends/cadence/aot/memory_planning.py +472 -0
  823. executorch/backends/cadence/aot/memory_planning_algo.py +341 -0
  824. executorch/backends/cadence/aot/ops_registrations.py +3349 -0
  825. executorch/backends/cadence/aot/pass_utils.py +231 -0
  826. executorch/backends/cadence/aot/passes.py +140 -0
  827. executorch/backends/cadence/aot/program_builder.py +9 -0
  828. executorch/backends/cadence/aot/quantizer/fusion_pass.py +843 -0
  829. executorch/backends/cadence/aot/quantizer/passes/fuse_ops.py +391 -0
  830. executorch/backends/cadence/aot/quantizer/pattern_utils.py +207 -0
  831. executorch/backends/cadence/aot/quantizer/patterns.py +1570 -0
  832. executorch/backends/cadence/aot/quantizer/quantizer.py +490 -0
  833. executorch/backends/cadence/aot/quantizer/utils.py +286 -0
  834. executorch/backends/cadence/aot/ref_implementations.py +2832 -0
  835. executorch/backends/cadence/aot/remove_ops.py +986 -0
  836. executorch/backends/cadence/aot/reorder_ops.py +1279 -0
  837. executorch/backends/cadence/aot/replace_ops.py +2646 -0
  838. executorch/backends/cadence/aot/simplify_ops.py +221 -0
  839. executorch/backends/cadence/aot/tests/test_decompose_ops_passes.py +80 -0
  840. executorch/backends/cadence/aot/tests/test_fusion_ops_passes.py +2030 -0
  841. executorch/backends/cadence/aot/tests/test_graph_builder.py +101 -0
  842. executorch/backends/cadence/aot/tests/test_idma_ops.py +31 -0
  843. executorch/backends/cadence/aot/tests/test_memory_passes.py +1367 -0
  844. executorch/backends/cadence/aot/tests/test_pass_filter.py +170 -0
  845. executorch/backends/cadence/aot/tests/test_pass_utils.py +85 -0
  846. executorch/backends/cadence/aot/tests/test_program_builder.py +222 -0
  847. executorch/backends/cadence/aot/tests/test_quantizer_ops.py +975 -0
  848. executorch/backends/cadence/aot/tests/test_ref_implementations.py +3633 -0
  849. executorch/backends/cadence/aot/tests/test_remove_bn_tracking_pass.py +209 -0
  850. executorch/backends/cadence/aot/tests/test_remove_ops_passes.py +1390 -0
  851. executorch/backends/cadence/aot/tests/test_reorder_ops_passes.py +1554 -0
  852. executorch/backends/cadence/aot/tests/test_replace_ops_passes.py +3660 -0
  853. executorch/backends/cadence/aot/tests/test_simplify_ops_passes.py +160 -0
  854. executorch/backends/cadence/aot/tests/test_to_out_var_pass.py +98 -0
  855. executorch/backends/cadence/aot/tests/test_type_dispatch_passes.py +424 -0
  856. executorch/backends/cadence/aot/to_out_var_pass.py +24 -0
  857. executorch/backends/cadence/aot/type_dispatch.py +184 -0
  858. executorch/backends/cadence/aot/typing_stubs.py +22 -0
  859. executorch/backends/cadence/aot/utils.py +372 -0
  860. executorch/backends/cadence/runtime/__init__.py +13 -0
  861. executorch/backends/cadence/runtime/etdump.py +173 -0
  862. executorch/backends/cadence/runtime/executor.py +208 -0
  863. executorch/backends/cadence/runtime/runtime.py +167 -0
  864. executorch/backends/cadence/runtime/utils.py +100 -0
  865. executorch/backends/cadence/utils/FACTO/calibrator/__init__.py +0 -0
  866. executorch/backends/cadence/utils/FACTO/calibrator/runner.py +209 -0
  867. executorch/backends/cadence/utils/FACTO/examples/config_example.py +96 -0
  868. executorch/backends/cadence/utils/FACTO/examples/example.py +59 -0
  869. executorch/backends/cadence/utils/FACTO/examples/minimal_example.py +21 -0
  870. executorch/backends/cadence/utils/FACTO/examples/minimal_modelgen_example.py +23 -0
  871. executorch/backends/cadence/utils/FACTO/examples/random_seed.py +35 -0
  872. executorch/backends/cadence/utils/FACTO/facto/__init__.py +0 -0
  873. executorch/backends/cadence/utils/FACTO/facto/inputgen/__init__.py +0 -0
  874. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/__init__.py +0 -0
  875. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/engine.py +129 -0
  876. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/gen.py +245 -0
  877. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/__init__.py +0 -0
  878. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/engine.py +303 -0
  879. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/gen.py +320 -0
  880. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/type.py +148 -0
  881. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/__init__.py +0 -0
  882. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/engine.py +58 -0
  883. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/model.py +133 -0
  884. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/solve.py +125 -0
  885. executorch/backends/cadence/utils/FACTO/facto/inputgen/specs/__init__.py +0 -0
  886. executorch/backends/cadence/utils/FACTO/facto/inputgen/specs/model.py +149 -0
  887. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/__init__.py +0 -0
  888. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/config.py +44 -0
  889. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/random_manager.py +36 -0
  890. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/__init__.py +0 -0
  891. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/constants.py +9 -0
  892. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/gen.py +283 -0
  893. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/solve.py +154 -0
  894. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/space.py +450 -0
  895. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/type.py +105 -0
  896. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/utils.py +30 -0
  897. executorch/backends/cadence/utils/FACTO/facto/modelgen/__init__.py +0 -0
  898. executorch/backends/cadence/utils/FACTO/facto/modelgen/gen.py +251 -0
  899. executorch/backends/cadence/utils/FACTO/facto/specdb/__init__.py +0 -0
  900. executorch/backends/cadence/utils/FACTO/facto/specdb/db.py +5973 -0
  901. executorch/backends/cadence/utils/FACTO/facto/specdb/default.py +36 -0
  902. executorch/backends/cadence/utils/FACTO/facto/specdb/function.py +585 -0
  903. executorch/backends/cadence/utils/FACTO/facto/utils/__init__.py +0 -0
  904. executorch/backends/cadence/utils/FACTO/facto/utils/dtypes.py +24 -0
  905. executorch/backends/cadence/utils/FACTO/facto/utils/ops.py +48 -0
  906. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argtuple_engine.py +53 -0
  907. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argtuple_generator.py +71 -0
  908. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_attributes.py +78 -0
  909. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_generator.py +98 -0
  910. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_types.py +28 -0
  911. executorch/backends/cadence/utils/FACTO/test/inputgen/test_attribute_engine.py +87 -0
  912. executorch/backends/cadence/utils/FACTO/test/inputgen/test_attribute_solver.py +98 -0
  913. executorch/backends/cadence/utils/FACTO/test/inputgen/test_config.py +269 -0
  914. executorch/backends/cadence/utils/FACTO/test/inputgen/test_constraints.py +27 -0
  915. executorch/backends/cadence/utils/FACTO/test/inputgen/test_meta_arg_engine.py +85 -0
  916. executorch/backends/cadence/utils/FACTO/test/inputgen/test_specs.py +48 -0
  917. executorch/backends/cadence/utils/FACTO/test/inputgen/test_structural_engine.py +32 -0
  918. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_generation.py +80 -0
  919. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_solving.py +193 -0
  920. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_space.py +231 -0
  921. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_types.py +71 -0
  922. executorch/backends/cadence/utils/FACTO/test/specdb/__init__.py +1 -0
  923. executorch/backends/cadence/utils/FACTO/test/specdb/base_model_test.py +78 -0
  924. executorch/backends/cadence/utils/FACTO/test/specdb/base_test.py +82 -0
  925. executorch/backends/cadence/utils/FACTO/test/specdb/test_executorch_portable.py +340 -0
  926. executorch/backends/cadence/utils/FACTO/test/specdb/test_model_cpu.py +70 -0
  927. executorch/backends/cadence/utils/FACTO/test/specdb/test_specdb_cpu.py +67 -0
  928. executorch/backends/cadence/utils/FACTO/test/specdb/test_specdb_mps.py +111 -0
  929. executorch/backends/cadence/utils/facto_util.py +612 -0
  930. executorch/backends/cadence/utils/gen_header.py +49 -0
  931. executorch/backends/cadence/utils/post_compilation.py +173 -0
  932. executorch/backends/cortex_m/library/__init__.py +4 -0
  933. executorch/backends/cortex_m/library/_cmsis_nn/cmsis_nn.cp314-win_amd64.pyd +0 -0
  934. executorch/backends/cortex_m/library/cmsis_nn.py +281 -0
  935. executorch/backends/cortex_m/ops/operators.py +1438 -0
  936. executorch/backends/cortex_m/ops/operators.yaml +108 -0
  937. executorch/backends/cortex_m/passes/__init__.py +15 -0
  938. executorch/backends/cortex_m/passes/activation_fusion_pass.py +200 -0
  939. executorch/backends/cortex_m/passes/aten_to_cortex_m_pass.py +1156 -0
  940. executorch/backends/cortex_m/passes/clamp_hardswish_pass.py +37 -0
  941. executorch/backends/cortex_m/passes/cortex_m_pass.py +35 -0
  942. executorch/backends/cortex_m/passes/cortex_m_pass_manager.py +122 -0
  943. executorch/backends/cortex_m/passes/decompose_hardswish_pass.py +128 -0
  944. executorch/backends/cortex_m/passes/decompose_mean_pass.py +45 -0
  945. executorch/backends/cortex_m/passes/matmul_to_bmm_pass.py +81 -0
  946. executorch/backends/cortex_m/passes/passes_utils.py +350 -0
  947. executorch/backends/cortex_m/passes/quantized_clamp_activation_pass.py +129 -0
  948. executorch/backends/cortex_m/passes/replace_quant_nodes_pass.py +76 -0
  949. executorch/backends/cortex_m/passes/scratch_buffer_sizes.py +290 -0
  950. executorch/backends/cortex_m/quantizer/__init__.py +19 -0
  951. executorch/backends/cortex_m/quantizer/node_finders.py +153 -0
  952. executorch/backends/cortex_m/quantizer/pattern_checkers.py +348 -0
  953. executorch/backends/cortex_m/quantizer/pattern_matcher.py +214 -0
  954. executorch/backends/cortex_m/quantizer/quantization_configs.py +205 -0
  955. executorch/backends/cortex_m/quantizer/quantizer.py +89 -0
  956. executorch/backends/cortex_m/quantizer/quantizer_support.py +172 -0
  957. executorch/backends/cortex_m/quantizer_reporter.py +443 -0
  958. executorch/backends/cortex_m/target_config.py +117 -0
  959. executorch/backends/cortex_m/test/conftest.py +97 -0
  960. executorch/backends/cortex_m/test/misc/test_cmsis_pybind.py +132 -0
  961. executorch/backends/cortex_m/test/misc/test_node_finders.py +186 -0
  962. executorch/backends/cortex_m/test/misc/test_pattern_matcher.py +254 -0
  963. executorch/backends/cortex_m/test/misc/test_portable_int8.py +784 -0
  964. executorch/backends/cortex_m/test/misc/test_quantization.py +411 -0
  965. executorch/backends/cortex_m/test/misc/test_quantizer_reporter.py +245 -0
  966. executorch/backends/cortex_m/test/misc/test_target_config.py +117 -0
  967. executorch/backends/cortex_m/test/models/__init__.py +4 -0
  968. executorch/backends/cortex_m/test/models/test_deep_autoencoder.py +44 -0
  969. executorch/backends/cortex_m/test/models/test_ds_cnn.py +57 -0
  970. executorch/backends/cortex_m/test/models/test_mobilenet_v1_025.py +54 -0
  971. executorch/backends/cortex_m/test/models/test_mobilenet_v2.py +91 -0
  972. executorch/backends/cortex_m/test/models/test_mobilenet_v3.py +102 -0
  973. executorch/backends/cortex_m/test/models/test_nn_functional.py +119 -0
  974. executorch/backends/cortex_m/test/models/test_nn_modules.py +197 -0
  975. executorch/backends/cortex_m/test/models/test_resnet8.py +55 -0
  976. executorch/backends/cortex_m/test/models/test_silero_vad.py +104 -0
  977. executorch/backends/cortex_m/test/models/test_torch_functions.py +154 -0
  978. executorch/backends/cortex_m/test/models/test_wav2letter.py +34 -0
  979. executorch/backends/cortex_m/test/models/test_yolo11.py +45 -0
  980. executorch/backends/cortex_m/test/ops/__init__.py +4 -0
  981. executorch/backends/cortex_m/test/ops/test_activation.py +716 -0
  982. executorch/backends/cortex_m/test/ops/test_activation_quant.py +156 -0
  983. executorch/backends/cortex_m/test/ops/test_add.py +255 -0
  984. executorch/backends/cortex_m/test/ops/test_avg_pool2d.py +145 -0
  985. executorch/backends/cortex_m/test/ops/test_batch_matmul.py +192 -0
  986. executorch/backends/cortex_m/test/ops/test_conv.py +345 -0
  987. executorch/backends/cortex_m/test/ops/test_conv_transpose.py +331 -0
  988. executorch/backends/cortex_m/test/ops/test_linear.py +295 -0
  989. executorch/backends/cortex_m/test/ops/test_lstm.py +102 -0
  990. executorch/backends/cortex_m/test/ops/test_max_pool2d.py +146 -0
  991. executorch/backends/cortex_m/test/ops/test_maximum.py +87 -0
  992. executorch/backends/cortex_m/test/ops/test_minimum.py +108 -0
  993. executorch/backends/cortex_m/test/ops/test_mul.py +160 -0
  994. executorch/backends/cortex_m/test/ops/test_pad.py +100 -0
  995. executorch/backends/cortex_m/test/ops/test_softmax.py +88 -0
  996. executorch/backends/cortex_m/test/ops/test_transpose.py +106 -0
  997. executorch/backends/cortex_m/test/test_helpers_passes_utils.py +103 -0
  998. executorch/backends/cortex_m/test/test_replace_quant_nodes.py +134 -0
  999. executorch/backends/cortex_m/test/tester.py +181 -0
  1000. executorch/backends/cuda/__init__.py +5 -0
  1001. executorch/backends/cuda/benchmarks/benchmark_moe.py +423 -0
  1002. executorch/backends/cuda/benchmarks/benchmark_sdpa.py +308 -0
  1003. executorch/backends/cuda/coalesced_int4_tensor.py +243 -0
  1004. executorch/backends/cuda/cuda_backend.py +684 -0
  1005. executorch/backends/cuda/cuda_partitioner.py +53 -0
  1006. executorch/backends/cuda/dp4a_planar_int5_tensor.py +323 -0
  1007. executorch/backends/cuda/dp4a_planar_int6_tensor.py +307 -0
  1008. executorch/backends/cuda/passes/__init__.py +0 -0
  1009. executorch/backends/cuda/passes/move_cond_predicate_to_cpu.py +90 -0
  1010. executorch/backends/cuda/passes/replace_int64_floordiv.py +152 -0
  1011. executorch/backends/cuda/passes/tests/__init__.py +0 -0
  1012. executorch/backends/cuda/passes/tests/test_move_cond_predicate_to_cpu.py +553 -0
  1013. executorch/backends/cuda/passes/tests/test_replace_int64_floordiv.py +216 -0
  1014. executorch/backends/cuda/quantize_op_dispatch/__init__.py +31 -0
  1015. executorch/backends/cuda/quantize_op_dispatch/_library.py +17 -0
  1016. executorch/backends/cuda/quantize_op_dispatch/int4_dispatch.py +181 -0
  1017. executorch/backends/cuda/quantize_op_dispatch/int5_dispatch.py +168 -0
  1018. executorch/backends/cuda/quantize_op_dispatch/int6_dispatch.py +128 -0
  1019. executorch/backends/cuda/quantize_op_dispatch/int8_dispatch.py +132 -0
  1020. executorch/backends/cuda/runtime/shims/tests/gen_plain_mm_test_vectors.py +460 -0
  1021. executorch/backends/cuda/tests/__init__.py +5 -0
  1022. executorch/backends/cuda/tests/test_chunk_gated_delta_rule.py +328 -0
  1023. executorch/backends/cuda/tests/test_cuda_export.py +509 -0
  1024. executorch/backends/cuda/tests/test_cuda_partitioner.py +458 -0
  1025. executorch/backends/cuda/tests/test_fused_deltanet_decode.py +357 -0
  1026. executorch/backends/cuda/tests/test_fused_moe.py +812 -0
  1027. executorch/backends/cuda/tests/test_int4_dispatch.py +365 -0
  1028. executorch/backends/cuda/tests/test_int4_matmul.py +287 -0
  1029. executorch/backends/cuda/tests/test_int5_dispatch.py +323 -0
  1030. executorch/backends/cuda/tests/test_int6_dispatch.py +321 -0
  1031. executorch/backends/cuda/tests/test_output_stride_rearrange.py +177 -0
  1032. executorch/backends/cuda/tests/test_sdpa_splitk_replacement.py +179 -0
  1033. executorch/backends/cuda/tests/test_sort_shim.py +128 -0
  1034. executorch/backends/cuda/tests/test_topk.py +308 -0
  1035. executorch/backends/cuda/tests/test_tq4_sdpa.py +1232 -0
  1036. executorch/backends/cuda/tests/test_triton_sdpa.py +522 -0
  1037. executorch/backends/cuda/tests/test_triton_sdpa_nan.py +253 -0
  1038. executorch/backends/cuda/tests/test_triton_sdpa_splitk.py +310 -0
  1039. executorch/backends/cuda/triton/__init__.py +17 -0
  1040. executorch/backends/cuda/triton/kernels/__init__.py +58 -0
  1041. executorch/backends/cuda/triton/kernels/chunk_gated_delta_rule.py +269 -0
  1042. executorch/backends/cuda/triton/kernels/fused_deltanet_decode.py +358 -0
  1043. executorch/backends/cuda/triton/kernels/fused_moe.py +1604 -0
  1044. executorch/backends/cuda/triton/kernels/int4_matmul.py +529 -0
  1045. executorch/backends/cuda/triton/kernels/sdpa.py +1551 -0
  1046. executorch/backends/cuda/triton/kernels/topk.py +220 -0
  1047. executorch/backends/cuda/triton/kernels/tq4_sdpa.py +1216 -0
  1048. executorch/backends/cuda/triton/replacement_pass.py +185 -0
  1049. executorch/backends/example/example_backend.py +41 -0
  1050. executorch/backends/example/example_backend_delegate_passes/merge_to_dim_pass.py +83 -0
  1051. executorch/backends/example/example_backend_delegate_passes/permute_memory_formats_pass.py +143 -0
  1052. executorch/backends/example/example_operators/adaptive_avg_pool2d.py +47 -0
  1053. executorch/backends/example/example_operators/add.py +44 -0
  1054. executorch/backends/example/example_operators/conv2d.py +52 -0
  1055. executorch/backends/example/example_operators/conv_relu.py +49 -0
  1056. executorch/backends/example/example_operators/dropout.py +44 -0
  1057. executorch/backends/example/example_operators/flatten.py +40 -0
  1058. executorch/backends/example/example_operators/linear.py +62 -0
  1059. executorch/backends/example/example_operators/op_base.py +15 -0
  1060. executorch/backends/example/example_operators/ops.py +32 -0
  1061. executorch/backends/example/example_operators/utils.py +34 -0
  1062. executorch/backends/example/example_partitioner.py +88 -0
  1063. executorch/backends/example/example_quantizer.py +108 -0
  1064. executorch/backends/example/test_example_delegate.py +112 -0
  1065. executorch/backends/mediatek/__init__.py +5 -0
  1066. executorch/backends/mediatek/_passes/__init__.py +0 -0
  1067. executorch/backends/mediatek/_passes/decompose_scaled_dot_product_attention.py +79 -0
  1068. executorch/backends/mediatek/partitioner.py +107 -0
  1069. executorch/backends/mediatek/preprocess.py +214 -0
  1070. executorch/backends/mediatek/quantizer/__init__.py +4 -0
  1071. executorch/backends/mediatek/quantizer/annotator.py +263 -0
  1072. executorch/backends/mediatek/quantizer/qconfig.py +174 -0
  1073. executorch/backends/mediatek/quantizer/quantizer.py +45 -0
  1074. executorch/backends/mlx/__init__.py +17 -0
  1075. executorch/backends/mlx/_logging.py +40 -0
  1076. executorch/backends/mlx/builder/__init__.py +16 -0
  1077. executorch/backends/mlx/builder/op_helpers.py +810 -0
  1078. executorch/backends/mlx/builder/op_registry.py +151 -0
  1079. executorch/backends/mlx/builder/pattern_matcher.py +64 -0
  1080. executorch/backends/mlx/builder/program_builder.py +1135 -0
  1081. executorch/backends/mlx/builder/slot_manager.py +252 -0
  1082. executorch/backends/mlx/custom_kernel_ops/__init__.py +7 -0
  1083. executorch/backends/mlx/custom_kernel_ops/gated_delta_rule.py +608 -0
  1084. executorch/backends/mlx/custom_kernel_ops/gguf/__init__.py +18 -0
  1085. executorch/backends/mlx/custom_kernel_ops/gguf/patterns.py +177 -0
  1086. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/__init__.py +32 -0
  1087. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/common.py +151 -0
  1088. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/embedding.py +147 -0
  1089. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/embedding_mlx_native.py +60 -0
  1090. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/linear.py +500 -0
  1091. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/linear_mlx_native.py +88 -0
  1092. executorch/backends/mlx/custom_kernel_ops/gguf/q4k/repack_mlx.py +60 -0
  1093. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/__init__.py +37 -0
  1094. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/common.py +167 -0
  1095. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/embedding.py +144 -0
  1096. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/embedding_mlx_native.py +60 -0
  1097. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/linear.py +534 -0
  1098. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/linear_mlx_native.py +88 -0
  1099. executorch/backends/mlx/custom_kernel_ops/gguf/q5k/repack_mlx.py +60 -0
  1100. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/__init__.py +40 -0
  1101. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/common.py +142 -0
  1102. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/embedding.py +151 -0
  1103. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/embedding_mlx_native.py +64 -0
  1104. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/linear.py +511 -0
  1105. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/linear_mlx_native.py +89 -0
  1106. executorch/backends/mlx/custom_kernel_ops/gguf/q6k/repack_mlx.py +87 -0
  1107. executorch/backends/mlx/custom_kernel_ops/gguf/test/__init__.py +5 -0
  1108. executorch/backends/mlx/custom_kernel_ops/gguf/test/test_embedding.py +270 -0
  1109. executorch/backends/mlx/custom_kernel_ops/gguf/test/test_linear.py +647 -0
  1110. executorch/backends/mlx/custom_kernel_ops/test/__init__.py +5 -0
  1111. executorch/backends/mlx/custom_kernel_ops/test/test_gated_delta_rule.py +987 -0
  1112. executorch/backends/mlx/custom_kernel_ops/test/test_tq4_compress.py +183 -0
  1113. executorch/backends/mlx/custom_kernel_ops/test/test_tq_dequant.py +166 -0
  1114. executorch/backends/mlx/custom_kernel_ops/test/test_tq_norm.py +150 -0
  1115. executorch/backends/mlx/custom_kernel_ops/tq4_compress.py +189 -0
  1116. executorch/backends/mlx/custom_kernel_ops/tq_dequant.py +216 -0
  1117. executorch/backends/mlx/custom_kernel_ops/tq_norm.py +170 -0
  1118. executorch/backends/mlx/custom_ops.py +467 -0
  1119. executorch/backends/mlx/examples/__init__.py +6 -0
  1120. executorch/backends/mlx/examples/llm/__init__.py +6 -0
  1121. executorch/backends/mlx/examples/llm/export_llm_hf.py +457 -0
  1122. executorch/backends/mlx/examples/llm/run_llm_hf.py +249 -0
  1123. executorch/backends/mlx/examples/voxtral/__init__.py +5 -0
  1124. executorch/backends/mlx/examples/voxtral/export_voxtral_hf.py +201 -0
  1125. executorch/backends/mlx/examples/whisper/__init__.py +7 -0
  1126. executorch/backends/mlx/examples/whisper/args.py +151 -0
  1127. executorch/backends/mlx/examples/whisper/export_whisper.py +617 -0
  1128. executorch/backends/mlx/examples/whisper/run_whisper.py +276 -0
  1129. executorch/backends/mlx/llm/__init__.py +6 -0
  1130. executorch/backends/mlx/llm/cache.py +556 -0
  1131. executorch/backends/mlx/llm/et_attention.py +252 -0
  1132. executorch/backends/mlx/llm/hf_attention.py +258 -0
  1133. executorch/backends/mlx/llm/quantization.py +57 -0
  1134. executorch/backends/mlx/llm/sampling.py +44 -0
  1135. executorch/backends/mlx/llm/source_transformation.py +318 -0
  1136. executorch/backends/mlx/llm/switch.py +330 -0
  1137. executorch/backends/mlx/llm/turboquant_cache.py +244 -0
  1138. executorch/backends/mlx/ops.py +4808 -0
  1139. executorch/backends/mlx/partitioner.py +298 -0
  1140. executorch/backends/mlx/passes.py +570 -0
  1141. executorch/backends/mlx/pattern_utils.py +360 -0
  1142. executorch/backends/mlx/patterns.py +1339 -0
  1143. executorch/backends/mlx/preprocess.py +168 -0
  1144. executorch/backends/mlx/pte_inspector.py +902 -0
  1145. executorch/backends/mlx/serialization/__init__.py +32 -0
  1146. executorch/backends/mlx/serialization/generate.py +1566 -0
  1147. executorch/backends/mlx/serialization/mlx_graph_serialize.py +417 -0
  1148. executorch/backends/mlx/serialization/schema.fbs +1273 -0
  1149. executorch/backends/mlx/test/__init__.py +5 -0
  1150. executorch/backends/mlx/test/export_multi_thread_test_model.py +124 -0
  1151. executorch/backends/mlx/test/run_all_tests.py +496 -0
  1152. executorch/backends/mlx/test/test_ops.py +7929 -0
  1153. executorch/backends/mlx/test/test_partitioner.py +45 -0
  1154. executorch/backends/mlx/test/test_passes.py +828 -0
  1155. executorch/backends/mlx/test/test_pattern_utils.py +592 -0
  1156. executorch/backends/mlx/test/test_quant_regroup.py +70 -0
  1157. executorch/backends/mlx/test/test_sample.py +295 -0
  1158. executorch/backends/mlx/test/test_serialization_dedup.py +84 -0
  1159. executorch/backends/mlx/test/test_slot_recycling.py +198 -0
  1160. executorch/backends/mlx/test/test_utils.py +1162 -0
  1161. executorch/backends/mlx/test/tester.py +78 -0
  1162. executorch/backends/mlx/third-party/mlx/benchmarks/numpy/single_ops.py +39 -0
  1163. executorch/backends/mlx/third-party/mlx/benchmarks/numpy/time_utils.py +20 -0
  1164. executorch/backends/mlx/third-party/mlx/benchmarks/python/batch_matmul_bench.py +62 -0
  1165. executorch/backends/mlx/third-party/mlx/benchmarks/python/blas/bench_gemm.py +191 -0
  1166. executorch/backends/mlx/third-party/mlx/benchmarks/python/blas/bench_gemv.py +220 -0
  1167. executorch/backends/mlx/third-party/mlx/benchmarks/python/block_masked_mm_bench.py +193 -0
  1168. executorch/backends/mlx/third-party/mlx/benchmarks/python/comparative/bench_mlx.py +519 -0
  1169. executorch/backends/mlx/third-party/mlx/benchmarks/python/comparative/bench_torch.py +482 -0
  1170. executorch/backends/mlx/third-party/mlx/benchmarks/python/comparative/compare.py +284 -0
  1171. executorch/backends/mlx/third-party/mlx/benchmarks/python/compile_bench.py +107 -0
  1172. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv1d_bench.py +123 -0
  1173. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv2d_bench_cpu.py +127 -0
  1174. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv2d_train_bench_cpu.py +143 -0
  1175. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv2d_transpose_bench_cpu.py +129 -0
  1176. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv3d_bench.py +152 -0
  1177. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv3d_bench_cpu.py +110 -0
  1178. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv3d_train_bench_cpu.py +143 -0
  1179. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv3d_transpose_bench_cpu.py +116 -0
  1180. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv_bench.py +135 -0
  1181. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv_transpose_bench.py +135 -0
  1182. executorch/backends/mlx/third-party/mlx/benchmarks/python/conv_unaligned_bench.py +107 -0
  1183. executorch/backends/mlx/third-party/mlx/benchmarks/python/distributed_bench.py +66 -0
  1184. executorch/backends/mlx/third-party/mlx/benchmarks/python/einsum_bench.py +84 -0
  1185. executorch/backends/mlx/third-party/mlx/benchmarks/python/fft_bench.py +118 -0
  1186. executorch/backends/mlx/third-party/mlx/benchmarks/python/gather_bench.py +52 -0
  1187. executorch/backends/mlx/third-party/mlx/benchmarks/python/gather_mm_bench.py +74 -0
  1188. executorch/backends/mlx/third-party/mlx/benchmarks/python/gather_qmm_bench.py +84 -0
  1189. executorch/backends/mlx/third-party/mlx/benchmarks/python/hadamard_bench.py +70 -0
  1190. executorch/backends/mlx/third-party/mlx/benchmarks/python/large_gemm_bench.py +119 -0
  1191. executorch/backends/mlx/third-party/mlx/benchmarks/python/layer_norm_bench.py +82 -0
  1192. executorch/backends/mlx/third-party/mlx/benchmarks/python/masked_scatter.py +236 -0
  1193. executorch/backends/mlx/third-party/mlx/benchmarks/python/rms_norm_bench.py +63 -0
  1194. executorch/backends/mlx/third-party/mlx/benchmarks/python/rope_bench.py +35 -0
  1195. executorch/backends/mlx/third-party/mlx/benchmarks/python/scatter_bench.py +96 -0
  1196. executorch/backends/mlx/third-party/mlx/benchmarks/python/sdpa_bench.py +229 -0
  1197. executorch/backends/mlx/third-party/mlx/benchmarks/python/sdpa_vector_bench.py +95 -0
  1198. executorch/backends/mlx/third-party/mlx/benchmarks/python/segmented_mm_bench.py +209 -0
  1199. executorch/backends/mlx/third-party/mlx/benchmarks/python/single_ops.py +132 -0
  1200. executorch/backends/mlx/third-party/mlx/benchmarks/python/slice_update_bench.py +109 -0
  1201. executorch/backends/mlx/third-party/mlx/benchmarks/python/synchronize_bench.py +55 -0
  1202. executorch/backends/mlx/third-party/mlx/benchmarks/python/time_utils.py +38 -0
  1203. executorch/backends/mlx/third-party/mlx/docs/src/conf.py +99 -0
  1204. executorch/backends/mlx/third-party/mlx/examples/export/eval_mlp.py +52 -0
  1205. executorch/backends/mlx/third-party/mlx/examples/export/train_mlp.py +76 -0
  1206. executorch/backends/mlx/third-party/mlx/examples/extensions/mlx_sample_extensions/__init__.py +5 -0
  1207. executorch/backends/mlx/third-party/mlx/examples/extensions/setup.py +18 -0
  1208. executorch/backends/mlx/third-party/mlx/examples/extensions/test.py +12 -0
  1209. executorch/backends/mlx/third-party/mlx/examples/python/linear_regression.py +46 -0
  1210. executorch/backends/mlx/third-party/mlx/examples/python/logistic_regression.py +49 -0
  1211. executorch/backends/mlx/third-party/mlx/examples/python/qqmm.py +117 -0
  1212. executorch/backends/mlx/third-party/mlx/python/mlx/__main__.py +27 -0
  1213. executorch/backends/mlx/third-party/mlx/python/mlx/_distributed_utils/common.py +135 -0
  1214. executorch/backends/mlx/third-party/mlx/python/mlx/_distributed_utils/config.py +639 -0
  1215. executorch/backends/mlx/third-party/mlx/python/mlx/_distributed_utils/launch.py +562 -0
  1216. executorch/backends/mlx/third-party/mlx/python/mlx/_reprlib_fix.py +16 -0
  1217. executorch/backends/mlx/third-party/mlx/python/mlx/extension.py +88 -0
  1218. executorch/backends/mlx/third-party/mlx/python/mlx/nn/__init__.py +9 -0
  1219. executorch/backends/mlx/third-party/mlx/python/mlx/nn/init.py +451 -0
  1220. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/__init__.py +105 -0
  1221. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/activations.py +667 -0
  1222. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/base.py +689 -0
  1223. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/containers.py +24 -0
  1224. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/convolution.py +235 -0
  1225. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/convolution_transpose.py +245 -0
  1226. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/distributed.py +619 -0
  1227. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/dropout.py +137 -0
  1228. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/embedding.py +53 -0
  1229. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/linear.py +180 -0
  1230. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/normalization.py +381 -0
  1231. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/pooling.py +398 -0
  1232. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/positional_encoding.py +174 -0
  1233. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/quantized.py +426 -0
  1234. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/recurrent.py +291 -0
  1235. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/transformer.py +354 -0
  1236. executorch/backends/mlx/third-party/mlx/python/mlx/nn/layers/upsample.py +277 -0
  1237. executorch/backends/mlx/third-party/mlx/python/mlx/nn/losses.py +614 -0
  1238. executorch/backends/mlx/third-party/mlx/python/mlx/nn/utils.py +321 -0
  1239. executorch/backends/mlx/third-party/mlx/python/mlx/optimizers/__init__.py +4 -0
  1240. executorch/backends/mlx/third-party/mlx/python/mlx/optimizers/optimizers.py +974 -0
  1241. executorch/backends/mlx/third-party/mlx/python/mlx/optimizers/schedulers.py +158 -0
  1242. executorch/backends/mlx/third-party/mlx/python/mlx/py.typed +1 -0
  1243. executorch/backends/mlx/third-party/mlx/python/mlx/utils.py +326 -0
  1244. executorch/backends/mlx/third-party/mlx/python/tests/__main__.py +5 -0
  1245. executorch/backends/mlx/third-party/mlx/python/tests/mlx_distributed_tests.py +324 -0
  1246. executorch/backends/mlx/third-party/mlx/python/tests/mlx_tests.py +93 -0
  1247. executorch/backends/mlx/third-party/mlx/python/tests/mpi_test_distributed.py +142 -0
  1248. executorch/backends/mlx/third-party/mlx/python/tests/nccl_test_distributed.py +346 -0
  1249. executorch/backends/mlx/third-party/mlx/python/tests/ring_test_distributed.py +130 -0
  1250. executorch/backends/mlx/third-party/mlx/python/tests/test_array.py +2743 -0
  1251. executorch/backends/mlx/third-party/mlx/python/tests/test_autograd.py +1421 -0
  1252. executorch/backends/mlx/third-party/mlx/python/tests/test_bf16.py +228 -0
  1253. executorch/backends/mlx/third-party/mlx/python/tests/test_blas.py +1432 -0
  1254. executorch/backends/mlx/third-party/mlx/python/tests/test_compile.py +1506 -0
  1255. executorch/backends/mlx/third-party/mlx/python/tests/test_constants.py +41 -0
  1256. executorch/backends/mlx/third-party/mlx/python/tests/test_conv.py +1214 -0
  1257. executorch/backends/mlx/third-party/mlx/python/tests/test_conv_transpose.py +810 -0
  1258. executorch/backends/mlx/third-party/mlx/python/tests/test_device.py +150 -0
  1259. executorch/backends/mlx/third-party/mlx/python/tests/test_double.py +306 -0
  1260. executorch/backends/mlx/third-party/mlx/python/tests/test_einsum.py +363 -0
  1261. executorch/backends/mlx/third-party/mlx/python/tests/test_eval.py +200 -0
  1262. executorch/backends/mlx/third-party/mlx/python/tests/test_export_import.py +709 -0
  1263. executorch/backends/mlx/third-party/mlx/python/tests/test_fast.py +1122 -0
  1264. executorch/backends/mlx/third-party/mlx/python/tests/test_fast_sdpa.py +687 -0
  1265. executorch/backends/mlx/third-party/mlx/python/tests/test_fft.py +413 -0
  1266. executorch/backends/mlx/third-party/mlx/python/tests/test_graph.py +37 -0
  1267. executorch/backends/mlx/third-party/mlx/python/tests/test_init.py +158 -0
  1268. executorch/backends/mlx/third-party/mlx/python/tests/test_linalg.py +896 -0
  1269. executorch/backends/mlx/third-party/mlx/python/tests/test_load.py +529 -0
  1270. executorch/backends/mlx/third-party/mlx/python/tests/test_losses.py +476 -0
  1271. executorch/backends/mlx/third-party/mlx/python/tests/test_memory.py +77 -0
  1272. executorch/backends/mlx/third-party/mlx/python/tests/test_nn.py +2198 -0
  1273. executorch/backends/mlx/third-party/mlx/python/tests/test_ops.py +3520 -0
  1274. executorch/backends/mlx/third-party/mlx/python/tests/test_optimizers.py +615 -0
  1275. executorch/backends/mlx/third-party/mlx/python/tests/test_quantized.py +1388 -0
  1276. executorch/backends/mlx/third-party/mlx/python/tests/test_random.py +392 -0
  1277. executorch/backends/mlx/third-party/mlx/python/tests/test_reduce.py +223 -0
  1278. executorch/backends/mlx/third-party/mlx/python/tests/test_threads.py +112 -0
  1279. executorch/backends/mlx/third-party/mlx/python/tests/test_tree.py +126 -0
  1280. executorch/backends/mlx/third-party/mlx/python/tests/test_upsample.py +100 -0
  1281. executorch/backends/mlx/third-party/mlx/python/tests/test_vmap.py +949 -0
  1282. executorch/backends/mlx/third-party/mlx/setup.py +336 -0
  1283. executorch/backends/nxp/__init__.py +0 -0
  1284. executorch/backends/nxp/aten_passes/convert_1d_conv_to_2d.py +489 -0
  1285. executorch/backends/nxp/aten_passes/convert_div_to_mul.py +177 -0
  1286. executorch/backends/nxp/aten_passes/decompose_split_to_slices_pass.py +196 -0
  1287. executorch/backends/nxp/aten_passes/fuse_batch_norm_with_conv_pass.py +146 -0
  1288. executorch/backends/nxp/aten_passes/fuse_batch_norm_with_linear_pass.py +188 -0
  1289. executorch/backends/nxp/aten_passes/fuse_linear_and_add_pass.py +204 -0
  1290. executorch/backends/nxp/aten_passes/move_activation_before_concat.py +102 -0
  1291. executorch/backends/nxp/aten_passes/neutron_aten_pass_manager.py +82 -0
  1292. executorch/backends/nxp/aten_passes/remove_nodes_with_known_outputs.py +196 -0
  1293. executorch/backends/nxp/aten_passes/simulated_linear_bn_fusion_passes/__init__.py +11 -0
  1294. executorch/backends/nxp/aten_passes/simulated_linear_bn_fusion_passes/add_simulated_linear_bn_fusion_qat_pass.py +395 -0
  1295. executorch/backends/nxp/aten_passes/simulated_linear_bn_fusion_passes/remove_simulated_linear_bn_fusion_qat_pass.py +245 -0
  1296. executorch/backends/nxp/aten_passes/split_group_convolution.py +300 -0
  1297. executorch/backends/nxp/aten_passes/split_gru_based_on_num_layers.py +307 -0
  1298. executorch/backends/nxp/backend/custom_delegation_options.py +17 -0
  1299. executorch/backends/nxp/backend/data_format.py +49 -0
  1300. executorch/backends/nxp/backend/edge_helper.py +515 -0
  1301. executorch/backends/nxp/backend/edge_program_converter.py +360 -0
  1302. executorch/backends/nxp/backend/graph_utils.py +129 -0
  1303. executorch/backends/nxp/backend/ir/conversion_config.py +54 -0
  1304. executorch/backends/nxp/backend/ir/conversion_context.py +43 -0
  1305. executorch/backends/nxp/backend/ir/converter/__init__.py +0 -0
  1306. executorch/backends/nxp/backend/ir/converter/builder/__init__.py +0 -0
  1307. executorch/backends/nxp/backend/ir/converter/builder/aten_model_builder_director.py +156 -0
  1308. executorch/backends/nxp/backend/ir/converter/builder/model_builder.py +1806 -0
  1309. executorch/backends/nxp/backend/ir/converter/builder/quantization_verification.py +387 -0
  1310. executorch/backends/nxp/backend/ir/converter/conversion/__init__.py +0 -0
  1311. executorch/backends/nxp/backend/ir/converter/conversion/aten_translator.py +71 -0
  1312. executorch/backends/nxp/backend/ir/converter/conversion/common.py +177 -0
  1313. executorch/backends/nxp/backend/ir/converter/conversion/translator.py +689 -0
  1314. executorch/backends/nxp/backend/ir/converter/node_converter.py +535 -0
  1315. executorch/backends/nxp/backend/ir/converter/node_converters/__init__.py +4 -0
  1316. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/__init__.py +162 -0
  1317. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/abs_converter.py +52 -0
  1318. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/adaptive_avg_pool_2d_converter.py +107 -0
  1319. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/add_tensor_converter.py +66 -0
  1320. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/addmm_converter.py +140 -0
  1321. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/amax_converter.py +81 -0
  1322. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/amin_converter.py +81 -0
  1323. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/avg_pool_2d_converter.py +137 -0
  1324. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/bmm_converter.py +101 -0
  1325. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/cat_converter.py +122 -0
  1326. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/clamp_converter.py +250 -0
  1327. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/clone_converter.py +118 -0
  1328. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/constant_pad_nd_converter.py +162 -0
  1329. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/convolution_converter.py +542 -0
  1330. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/exp_converter.py +59 -0
  1331. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/getitem_converter.py +56 -0
  1332. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/hardtanh_converter.py +37 -0
  1333. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/leaky_relu_converter.py +63 -0
  1334. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/log_converter.py +59 -0
  1335. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/max_pool2d_with_indices_converter.py +178 -0
  1336. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/maximum_converter.py +68 -0
  1337. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mean_dim_converter.py +107 -0
  1338. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/minimum_converter.py +68 -0
  1339. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mm_converter.py +94 -0
  1340. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mul_tensor_converter.py +63 -0
  1341. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/neg_converter.py +77 -0
  1342. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/permute_copy_converter.py +434 -0
  1343. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/prelu_converter.py +66 -0
  1344. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/qdq_dequantize_converter.py +127 -0
  1345. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/qdq_quantize_converter.py +46 -0
  1346. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/relu_converter.py +66 -0
  1347. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sigmoid_converter.py +62 -0
  1348. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/slice_tensor_converter.py +115 -0
  1349. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/softmax_converter.py +150 -0
  1350. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sub_tensor_converter.py +71 -0
  1351. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sum_dim_int_list_converter.py +83 -0
  1352. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/tanh_converter.py +60 -0
  1353. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/upsample_bilinear2d_converter.py +152 -0
  1354. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/upsample_nearest2d_converter.py +136 -0
  1355. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/view_copy_converter.py +99 -0
  1356. executorch/backends/nxp/backend/ir/converter/node_converters/shared/__init__.py +0 -0
  1357. executorch/backends/nxp/backend/ir/converter/node_converters/shared/conv_utils.py +123 -0
  1358. executorch/backends/nxp/backend/ir/converter/node_converters/shared/recurrent_utils.py +64 -0
  1359. executorch/backends/nxp/backend/ir/converter/node_converters/shared/reduce_utils.py +167 -0
  1360. executorch/backends/nxp/backend/ir/converter/node_converters/shared/reshape_transposition.py +233 -0
  1361. executorch/backends/nxp/backend/ir/converter/quantization_utils.py +161 -0
  1362. executorch/backends/nxp/backend/ir/converter/tensor_utils.py +55 -0
  1363. executorch/backends/nxp/backend/ir/lib/__init__.py +0 -0
  1364. executorch/backends/nxp/backend/ir/lib/tflite/ATan2Options.py +50 -0
  1365. executorch/backends/nxp/backend/ir/lib/tflite/AbsOptions.py +50 -0
  1366. executorch/backends/nxp/backend/ir/lib/tflite/ActivationFunctionType.py +12 -0
  1367. executorch/backends/nxp/backend/ir/lib/tflite/AddNOptions.py +50 -0
  1368. executorch/backends/nxp/backend/ir/lib/tflite/AddOptions.py +82 -0
  1369. executorch/backends/nxp/backend/ir/lib/tflite/ArgMaxOptions.py +65 -0
  1370. executorch/backends/nxp/backend/ir/lib/tflite/ArgMinOptions.py +65 -0
  1371. executorch/backends/nxp/backend/ir/lib/tflite/AssignVariableOptions.py +50 -0
  1372. executorch/backends/nxp/backend/ir/lib/tflite/BatchMatMulOptions.py +101 -0
  1373. executorch/backends/nxp/backend/ir/lib/tflite/BatchToSpaceNDOptions.py +50 -0
  1374. executorch/backends/nxp/backend/ir/lib/tflite/BidirectionalSequenceLSTMOptions.py +160 -0
  1375. executorch/backends/nxp/backend/ir/lib/tflite/BidirectionalSequenceRNNOptions.py +126 -0
  1376. executorch/backends/nxp/backend/ir/lib/tflite/BitcastOptions.py +50 -0
  1377. executorch/backends/nxp/backend/ir/lib/tflite/BitwiseXorOptions.py +50 -0
  1378. executorch/backends/nxp/backend/ir/lib/tflite/BroadcastToOptions.py +50 -0
  1379. executorch/backends/nxp/backend/ir/lib/tflite/BucketizeOptions.py +98 -0
  1380. executorch/backends/nxp/backend/ir/lib/tflite/Buffer.py +132 -0
  1381. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOperator.py +212 -0
  1382. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOptions.py +133 -0
  1383. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOptions2.py +27 -0
  1384. executorch/backends/nxp/backend/ir/lib/tflite/CallOnceOptions.py +65 -0
  1385. executorch/backends/nxp/backend/ir/lib/tflite/CallOptions.py +67 -0
  1386. executorch/backends/nxp/backend/ir/lib/tflite/CastOptions.py +80 -0
  1387. executorch/backends/nxp/backend/ir/lib/tflite/CombinerType.py +9 -0
  1388. executorch/backends/nxp/backend/ir/lib/tflite/ConcatEmbeddingsOptions.py +163 -0
  1389. executorch/backends/nxp/backend/ir/lib/tflite/ConcatenationOptions.py +80 -0
  1390. executorch/backends/nxp/backend/ir/lib/tflite/Conv2DOptions.py +155 -0
  1391. executorch/backends/nxp/backend/ir/lib/tflite/Conv3DOptions.py +170 -0
  1392. executorch/backends/nxp/backend/ir/lib/tflite/CosOptions.py +50 -0
  1393. executorch/backends/nxp/backend/ir/lib/tflite/CumsumOptions.py +84 -0
  1394. executorch/backends/nxp/backend/ir/lib/tflite/CustomOptionsFormat.py +7 -0
  1395. executorch/backends/nxp/backend/ir/lib/tflite/CustomQuantization.py +98 -0
  1396. executorch/backends/nxp/backend/ir/lib/tflite/DensifyOptions.py +50 -0
  1397. executorch/backends/nxp/backend/ir/lib/tflite/DepthToSpaceOptions.py +65 -0
  1398. executorch/backends/nxp/backend/ir/lib/tflite/DepthwiseConv2DOptions.py +157 -0
  1399. executorch/backends/nxp/backend/ir/lib/tflite/DequantizeOptions.py +50 -0
  1400. executorch/backends/nxp/backend/ir/lib/tflite/DilateOptions.py +50 -0
  1401. executorch/backends/nxp/backend/ir/lib/tflite/DimensionMetadata.py +152 -0
  1402. executorch/backends/nxp/backend/ir/lib/tflite/DimensionType.py +8 -0
  1403. executorch/backends/nxp/backend/ir/lib/tflite/DivOptions.py +65 -0
  1404. executorch/backends/nxp/backend/ir/lib/tflite/DynamicUpdateSliceOptions.py +52 -0
  1405. executorch/backends/nxp/backend/ir/lib/tflite/EmbeddingLookupSparseOptions.py +67 -0
  1406. executorch/backends/nxp/backend/ir/lib/tflite/EqualOptions.py +50 -0
  1407. executorch/backends/nxp/backend/ir/lib/tflite/ExpOptions.py +50 -0
  1408. executorch/backends/nxp/backend/ir/lib/tflite/ExpandDimsOptions.py +50 -0
  1409. executorch/backends/nxp/backend/ir/lib/tflite/FakeQuantOptions.py +116 -0
  1410. executorch/backends/nxp/backend/ir/lib/tflite/FillOptions.py +50 -0
  1411. executorch/backends/nxp/backend/ir/lib/tflite/FloorDivOptions.py +50 -0
  1412. executorch/backends/nxp/backend/ir/lib/tflite/FloorModOptions.py +50 -0
  1413. executorch/backends/nxp/backend/ir/lib/tflite/FullyConnectedOptions.py +129 -0
  1414. executorch/backends/nxp/backend/ir/lib/tflite/FullyConnectedOptionsWeightsFormat.py +8 -0
  1415. executorch/backends/nxp/backend/ir/lib/tflite/GatherNdOptions.py +50 -0
  1416. executorch/backends/nxp/backend/ir/lib/tflite/GatherOptions.py +80 -0
  1417. executorch/backends/nxp/backend/ir/lib/tflite/GeluOptions.py +67 -0
  1418. executorch/backends/nxp/backend/ir/lib/tflite/GreaterEqualOptions.py +50 -0
  1419. executorch/backends/nxp/backend/ir/lib/tflite/GreaterOptions.py +50 -0
  1420. executorch/backends/nxp/backend/ir/lib/tflite/HardSwishOptions.py +50 -0
  1421. executorch/backends/nxp/backend/ir/lib/tflite/HashtableFindOptions.py +50 -0
  1422. executorch/backends/nxp/backend/ir/lib/tflite/HashtableImportOptions.py +52 -0
  1423. executorch/backends/nxp/backend/ir/lib/tflite/HashtableOptions.py +95 -0
  1424. executorch/backends/nxp/backend/ir/lib/tflite/HashtableSizeOptions.py +50 -0
  1425. executorch/backends/nxp/backend/ir/lib/tflite/IfOptions.py +80 -0
  1426. executorch/backends/nxp/backend/ir/lib/tflite/Int32Vector.py +98 -0
  1427. executorch/backends/nxp/backend/ir/lib/tflite/L2NormOptions.py +65 -0
  1428. executorch/backends/nxp/backend/ir/lib/tflite/LSHProjectionOptions.py +65 -0
  1429. executorch/backends/nxp/backend/ir/lib/tflite/LSHProjectionType.py +9 -0
  1430. executorch/backends/nxp/backend/ir/lib/tflite/LSTMKernelType.py +8 -0
  1431. executorch/backends/nxp/backend/ir/lib/tflite/LSTMOptions.py +131 -0
  1432. executorch/backends/nxp/backend/ir/lib/tflite/LeakyReluOptions.py +67 -0
  1433. executorch/backends/nxp/backend/ir/lib/tflite/LessEqualOptions.py +50 -0
  1434. executorch/backends/nxp/backend/ir/lib/tflite/LessOptions.py +50 -0
  1435. executorch/backends/nxp/backend/ir/lib/tflite/LocalResponseNormalizationOptions.py +118 -0
  1436. executorch/backends/nxp/backend/ir/lib/tflite/LogSoftmaxOptions.py +50 -0
  1437. executorch/backends/nxp/backend/ir/lib/tflite/LogicalAndOptions.py +50 -0
  1438. executorch/backends/nxp/backend/ir/lib/tflite/LogicalNotOptions.py +50 -0
  1439. executorch/backends/nxp/backend/ir/lib/tflite/LogicalOrOptions.py +50 -0
  1440. executorch/backends/nxp/backend/ir/lib/tflite/MatrixDiagOptions.py +50 -0
  1441. executorch/backends/nxp/backend/ir/lib/tflite/MatrixSetDiagOptions.py +50 -0
  1442. executorch/backends/nxp/backend/ir/lib/tflite/MaximumMinimumOptions.py +50 -0
  1443. executorch/backends/nxp/backend/ir/lib/tflite/Metadata.py +84 -0
  1444. executorch/backends/nxp/backend/ir/lib/tflite/MirrorPadMode.py +8 -0
  1445. executorch/backends/nxp/backend/ir/lib/tflite/MirrorPadOptions.py +65 -0
  1446. executorch/backends/nxp/backend/ir/lib/tflite/Model.py +352 -0
  1447. executorch/backends/nxp/backend/ir/lib/tflite/MulOptions.py +65 -0
  1448. executorch/backends/nxp/backend/ir/lib/tflite/NegOptions.py +50 -0
  1449. executorch/backends/nxp/backend/ir/lib/tflite/NonMaxSuppressionV4Options.py +52 -0
  1450. executorch/backends/nxp/backend/ir/lib/tflite/NonMaxSuppressionV5Options.py +52 -0
  1451. executorch/backends/nxp/backend/ir/lib/tflite/NotEqualOptions.py +50 -0
  1452. executorch/backends/nxp/backend/ir/lib/tflite/OneHotOptions.py +65 -0
  1453. executorch/backends/nxp/backend/ir/lib/tflite/Operator.py +428 -0
  1454. executorch/backends/nxp/backend/ir/lib/tflite/OperatorCode.py +112 -0
  1455. executorch/backends/nxp/backend/ir/lib/tflite/PackOptions.py +80 -0
  1456. executorch/backends/nxp/backend/ir/lib/tflite/PadOptions.py +50 -0
  1457. executorch/backends/nxp/backend/ir/lib/tflite/PadV2Options.py +50 -0
  1458. executorch/backends/nxp/backend/ir/lib/tflite/Padding.py +8 -0
  1459. executorch/backends/nxp/backend/ir/lib/tflite/Pool2DOptions.py +140 -0
  1460. executorch/backends/nxp/backend/ir/lib/tflite/PowOptions.py +50 -0
  1461. executorch/backends/nxp/backend/ir/lib/tflite/QuantizationDetails.py +8 -0
  1462. executorch/backends/nxp/backend/ir/lib/tflite/QuantizationParameters.py +295 -0
  1463. executorch/backends/nxp/backend/ir/lib/tflite/QuantizeOptions.py +50 -0
  1464. executorch/backends/nxp/backend/ir/lib/tflite/RNNOptions.py +82 -0
  1465. executorch/backends/nxp/backend/ir/lib/tflite/RandomOptions.py +80 -0
  1466. executorch/backends/nxp/backend/ir/lib/tflite/RangeOptions.py +50 -0
  1467. executorch/backends/nxp/backend/ir/lib/tflite/RankOptions.py +50 -0
  1468. executorch/backends/nxp/backend/ir/lib/tflite/ReadVariableOptions.py +50 -0
  1469. executorch/backends/nxp/backend/ir/lib/tflite/ReduceWindowFunction.py +13 -0
  1470. executorch/backends/nxp/backend/ir/lib/tflite/ReduceWindowOptions.py +65 -0
  1471. executorch/backends/nxp/backend/ir/lib/tflite/ReducerOptions.py +67 -0
  1472. executorch/backends/nxp/backend/ir/lib/tflite/ReshapeOptions.py +98 -0
  1473. executorch/backends/nxp/backend/ir/lib/tflite/ResizeBilinearOptions.py +84 -0
  1474. executorch/backends/nxp/backend/ir/lib/tflite/ResizeNearestNeighborOptions.py +86 -0
  1475. executorch/backends/nxp/backend/ir/lib/tflite/ReverseSequenceOptions.py +82 -0
  1476. executorch/backends/nxp/backend/ir/lib/tflite/ReverseV2Options.py +50 -0
  1477. executorch/backends/nxp/backend/ir/lib/tflite/Rfft2dOptions.py +50 -0
  1478. executorch/backends/nxp/backend/ir/lib/tflite/RightShiftOptions.py +50 -0
  1479. executorch/backends/nxp/backend/ir/lib/tflite/RngAlgorithm.py +9 -0
  1480. executorch/backends/nxp/backend/ir/lib/tflite/SVDFOptions.py +97 -0
  1481. executorch/backends/nxp/backend/ir/lib/tflite/ScatterNdOptions.py +50 -0
  1482. executorch/backends/nxp/backend/ir/lib/tflite/SegmentSumOptions.py +50 -0
  1483. executorch/backends/nxp/backend/ir/lib/tflite/SelectOptions.py +50 -0
  1484. executorch/backends/nxp/backend/ir/lib/tflite/SelectV2Options.py +50 -0
  1485. executorch/backends/nxp/backend/ir/lib/tflite/SequenceRNNOptions.py +99 -0
  1486. executorch/backends/nxp/backend/ir/lib/tflite/ShapeOptions.py +65 -0
  1487. executorch/backends/nxp/backend/ir/lib/tflite/SignOptions.py +50 -0
  1488. executorch/backends/nxp/backend/ir/lib/tflite/SignatureDef.py +172 -0
  1489. executorch/backends/nxp/backend/ir/lib/tflite/SkipGramOptions.py +97 -0
  1490. executorch/backends/nxp/backend/ir/lib/tflite/SliceOptions.py +50 -0
  1491. executorch/backends/nxp/backend/ir/lib/tflite/SoftmaxOptions.py +67 -0
  1492. executorch/backends/nxp/backend/ir/lib/tflite/SpaceToBatchNDOptions.py +50 -0
  1493. executorch/backends/nxp/backend/ir/lib/tflite/SpaceToDepthOptions.py +65 -0
  1494. executorch/backends/nxp/backend/ir/lib/tflite/SparseIndexVector.py +10 -0
  1495. executorch/backends/nxp/backend/ir/lib/tflite/SparseToDenseOptions.py +67 -0
  1496. executorch/backends/nxp/backend/ir/lib/tflite/SparsityParameters.py +190 -0
  1497. executorch/backends/nxp/backend/ir/lib/tflite/SplitOptions.py +65 -0
  1498. executorch/backends/nxp/backend/ir/lib/tflite/SplitVOptions.py +65 -0
  1499. executorch/backends/nxp/backend/ir/lib/tflite/SquareOptions.py +50 -0
  1500. executorch/backends/nxp/backend/ir/lib/tflite/SquaredDifferenceOptions.py +52 -0
  1501. executorch/backends/nxp/backend/ir/lib/tflite/SqueezeOptions.py +98 -0
  1502. executorch/backends/nxp/backend/ir/lib/tflite/StablehloBroadcastInDimOptions.py +102 -0
  1503. executorch/backends/nxp/backend/ir/lib/tflite/StablehloCompareOptions.py +86 -0
  1504. executorch/backends/nxp/backend/ir/lib/tflite/StablehloComparisonDirection.py +12 -0
  1505. executorch/backends/nxp/backend/ir/lib/tflite/StablehloComparisonType.py +11 -0
  1506. executorch/backends/nxp/backend/ir/lib/tflite/StablehloConcatenateOptions.py +67 -0
  1507. executorch/backends/nxp/backend/ir/lib/tflite/StablehloConvolutionOptions.py +634 -0
  1508. executorch/backends/nxp/backend/ir/lib/tflite/StablehloCustomCallOptions.py +214 -0
  1509. executorch/backends/nxp/backend/ir/lib/tflite/StablehloDotGeneralOptions.py +304 -0
  1510. executorch/backends/nxp/backend/ir/lib/tflite/StablehloDynamicSliceOptions.py +100 -0
  1511. executorch/backends/nxp/backend/ir/lib/tflite/StablehloGatherOptions.py +276 -0
  1512. executorch/backends/nxp/backend/ir/lib/tflite/StablehloIotaOptions.py +65 -0
  1513. executorch/backends/nxp/backend/ir/lib/tflite/StablehloPadOptions.py +194 -0
  1514. executorch/backends/nxp/backend/ir/lib/tflite/StablehloPrecisionConfig.py +9 -0
  1515. executorch/backends/nxp/backend/ir/lib/tflite/StablehloReduceOptions.py +115 -0
  1516. executorch/backends/nxp/backend/ir/lib/tflite/StablehloReduceWindowOptions.py +307 -0
  1517. executorch/backends/nxp/backend/ir/lib/tflite/StablehloRngBitGeneratorOptions.py +67 -0
  1518. executorch/backends/nxp/backend/ir/lib/tflite/StablehloScatterOptions.py +268 -0
  1519. executorch/backends/nxp/backend/ir/lib/tflite/StablehloSliceOptions.py +194 -0
  1520. executorch/backends/nxp/backend/ir/lib/tflite/StablehloSortOptions.py +97 -0
  1521. executorch/backends/nxp/backend/ir/lib/tflite/StablehloTransposeOptions.py +100 -0
  1522. executorch/backends/nxp/backend/ir/lib/tflite/StablehloWhileOptions.py +80 -0
  1523. executorch/backends/nxp/backend/ir/lib/tflite/StridedSliceOptions.py +142 -0
  1524. executorch/backends/nxp/backend/ir/lib/tflite/SubGraph.py +251 -0
  1525. executorch/backends/nxp/backend/ir/lib/tflite/SubOptions.py +82 -0
  1526. executorch/backends/nxp/backend/ir/lib/tflite/Tensor.py +317 -0
  1527. executorch/backends/nxp/backend/ir/lib/tflite/TensorMap.py +84 -0
  1528. executorch/backends/nxp/backend/ir/lib/tflite/TensorType.py +24 -0
  1529. executorch/backends/nxp/backend/ir/lib/tflite/TileOptions.py +50 -0
  1530. executorch/backends/nxp/backend/ir/lib/tflite/TopKV2Options.py +50 -0
  1531. executorch/backends/nxp/backend/ir/lib/tflite/TransposeConvOptions.py +125 -0
  1532. executorch/backends/nxp/backend/ir/lib/tflite/TransposeOptions.py +50 -0
  1533. executorch/backends/nxp/backend/ir/lib/tflite/Uint16Vector.py +98 -0
  1534. executorch/backends/nxp/backend/ir/lib/tflite/Uint8Vector.py +98 -0
  1535. executorch/backends/nxp/backend/ir/lib/tflite/UnidirectionalSequenceLSTMOptions.py +164 -0
  1536. executorch/backends/nxp/backend/ir/lib/tflite/UniqueOptions.py +65 -0
  1537. executorch/backends/nxp/backend/ir/lib/tflite/UnpackOptions.py +80 -0
  1538. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentMaxOptions.py +52 -0
  1539. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentMinOptions.py +52 -0
  1540. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentProdOptions.py +52 -0
  1541. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentSumOptions.py +52 -0
  1542. executorch/backends/nxp/backend/ir/lib/tflite/VarHandleOptions.py +84 -0
  1543. executorch/backends/nxp/backend/ir/lib/tflite/VariantSubType.py +130 -0
  1544. executorch/backends/nxp/backend/ir/lib/tflite/WhereOptions.py +50 -0
  1545. executorch/backends/nxp/backend/ir/lib/tflite/WhileOptions.py +80 -0
  1546. executorch/backends/nxp/backend/ir/lib/tflite/ZerosLikeOptions.py +50 -0
  1547. executorch/backends/nxp/backend/ir/lib/tflite/__init__.py +0 -0
  1548. executorch/backends/nxp/backend/ir/logger.py +341 -0
  1549. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/__init__.py +0 -0
  1550. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/graph_utils.py +115 -0
  1551. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/operator_rules.py +102 -0
  1552. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/__init__.py +0 -0
  1553. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/base_optimization.py +41 -0
  1554. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/permute_fully_connected_weights_after_reshape.py +121 -0
  1555. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/prune_transpose_operators.py +177 -0
  1556. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizer.py +127 -0
  1557. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/pattern_matcher.py +925 -0
  1558. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/tensor_rules.py +710 -0
  1559. executorch/backends/nxp/backend/ir/tflite_generator/__init__.py +0 -0
  1560. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/__init__.py +0 -0
  1561. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/abs_options.py +25 -0
  1562. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/add_n_options.py +32 -0
  1563. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/add_options.py +46 -0
  1564. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/arg_max_options.py +28 -0
  1565. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/arg_min_options.py +28 -0
  1566. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/average_pool_2d_options.py +63 -0
  1567. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/batch_mat_mul_options.py +45 -0
  1568. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bidirectional_sequence_lstm_options.py +65 -0
  1569. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bidirectional_sequence_rnn_options.py +53 -0
  1570. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bitwise_xor_options.py +23 -0
  1571. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/broadcast_to_options.py +25 -0
  1572. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/cast_options.py +32 -0
  1573. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/concatenation_options.py +38 -0
  1574. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/conv_2d_options.py +59 -0
  1575. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/conv_3d_options.py +67 -0
  1576. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/cum_sum_options.py +30 -0
  1577. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/depth_to_space_options.py +29 -0
  1578. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/depthwise_conv_2d_options.py +69 -0
  1579. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/dequantize_options.py +24 -0
  1580. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/div_options.py +43 -0
  1581. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/equal_options.py +23 -0
  1582. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/exp_options.py +23 -0
  1583. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/floor_mod_options.py +23 -0
  1584. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/fully_connected_options.py +57 -0
  1585. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gather_nd_options.py +23 -0
  1586. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gather_options.py +30 -0
  1587. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gelu_options.py +27 -0
  1588. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/greater_equal_options.py +25 -0
  1589. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/greater_options.py +23 -0
  1590. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/hard_swish_options.py +23 -0
  1591. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/leaky_relu_options.py +32 -0
  1592. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/less_equal_options.py +23 -0
  1593. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/less_options.py +23 -0
  1594. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/log_softmax_options.py +30 -0
  1595. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_and_options.py +23 -0
  1596. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_not_options.py +23 -0
  1597. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_or_options.py +23 -0
  1598. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/lrn_options.py +45 -0
  1599. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/lstm_options.py +60 -0
  1600. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/max_pool_2d_options.py +63 -0
  1601. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/maximum_options.py +21 -0
  1602. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mean_options.py +28 -0
  1603. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/minimum_options.py +21 -0
  1604. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mirror_pad_options.py +60 -0
  1605. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mul_options.py +36 -0
  1606. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/multinomial_options.py +30 -0
  1607. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/neg_options.py +23 -0
  1608. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/not_equal_options.py +23 -0
  1609. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/one_hot_options.py +27 -0
  1610. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pad_options.py +27 -0
  1611. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pad_v2_options.py +27 -0
  1612. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pow_options.py +24 -0
  1613. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/quantize_options.py +30 -0
  1614. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/range_options.py +23 -0
  1615. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_max_options.py +28 -0
  1616. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_min_options.py +28 -0
  1617. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_prod_options.py +28 -0
  1618. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reshape_options.py +53 -0
  1619. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/resize_bilinear_options.py +34 -0
  1620. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/resize_nearest_neighbor_options.py +37 -0
  1621. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reverse_sequence_options.py +32 -0
  1622. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/scatter_nd_options.py +23 -0
  1623. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/select_v2_options.py +24 -0
  1624. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/shape_options.py +34 -0
  1625. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sign_options.py +22 -0
  1626. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/slice_options.py +29 -0
  1627. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/softmax_options.py +36 -0
  1628. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/space_to_depth_options.py +29 -0
  1629. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/split_options.py +27 -0
  1630. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/split_v_options.py +27 -0
  1631. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/square_options.py +23 -0
  1632. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/squared_difference_options.py +25 -0
  1633. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/squeeze_options.py +42 -0
  1634. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/strided_slice_options.py +35 -0
  1635. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sub_options.py +46 -0
  1636. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sum_options.py +28 -0
  1637. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/tile_options.py +23 -0
  1638. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/transpose_conv_options.py +50 -0
  1639. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/transpose_options.py +30 -0
  1640. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/unidirectional_sequence_lstm_options.py +67 -0
  1641. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/unidirectional_sequence_rnn_options.py +49 -0
  1642. executorch/backends/nxp/backend/ir/tflite_generator/custom_options/flex_transpose_options.py +78 -0
  1643. executorch/backends/nxp/backend/ir/tflite_generator/meta/__init__.py +0 -0
  1644. executorch/backends/nxp/backend/ir/tflite_generator/meta/meta.py +255 -0
  1645. executorch/backends/nxp/backend/ir/tflite_generator/meta/types.py +198 -0
  1646. executorch/backends/nxp/backend/ir/tflite_generator/tflite_model.py +819 -0
  1647. executorch/backends/nxp/backend/neutron_converter_manager.py +151 -0
  1648. executorch/backends/nxp/backend/neutron_map.py +459 -0
  1649. executorch/backends/nxp/backend/neutron_operator_support.py +57 -0
  1650. executorch/backends/nxp/backend/neutron_target_spec.py +141 -0
  1651. executorch/backends/nxp/backend/node_format_inference.py +396 -0
  1652. executorch/backends/nxp/edge_passes/convert_reshaping_nodes_to_view.py +102 -0
  1653. executorch/backends/nxp/edge_passes/fold_redundant_qdq_pass.py +36 -0
  1654. executorch/backends/nxp/edge_passes/move_auxiliary_operator_into_separate_qdq_cluster_pass.py +301 -0
  1655. executorch/backends/nxp/edge_passes/neutron_edge_pass.py +55 -0
  1656. executorch/backends/nxp/edge_passes/neutron_edge_pass_manager.py +37 -0
  1657. executorch/backends/nxp/edge_passes/remove_additional_quantize_dequantize_nodes_pass.py +111 -0
  1658. executorch/backends/nxp/edge_passes/remove_as_strided_copy_nodes.py +66 -0
  1659. executorch/backends/nxp/edge_passes/remove_io_quant_ops_pass.py +79 -0
  1660. executorch/backends/nxp/neutron_node_extraction.py +152 -0
  1661. executorch/backends/nxp/neutron_partitioner.py +555 -0
  1662. executorch/backends/nxp/nxp_backend.py +476 -0
  1663. executorch/backends/nxp/quantizer/neutron_quantizer.py +398 -0
  1664. executorch/backends/nxp/quantizer/patterns.py +1395 -0
  1665. executorch/backends/nxp/quantizer/utils.py +257 -0
  1666. executorch/backends/nxp/tests/__init__.py +0 -0
  1667. executorch/backends/nxp/tests/calibration_dataset.py +30 -0
  1668. executorch/backends/nxp/tests/config.py +56 -0
  1669. executorch/backends/nxp/tests/config_importer.py +17 -0
  1670. executorch/backends/nxp/tests/conftest.py +38 -0
  1671. executorch/backends/nxp/tests/dataset_creator.py +417 -0
  1672. executorch/backends/nxp/tests/executorch_pipeline.py +323 -0
  1673. executorch/backends/nxp/tests/executors.py +433 -0
  1674. executorch/backends/nxp/tests/generic_tests/__init__.py +0 -0
  1675. executorch/backends/nxp/tests/generic_tests/conftest.py +35 -0
  1676. executorch/backends/nxp/tests/generic_tests/test_aot_example.py +172 -0
  1677. executorch/backends/nxp/tests/generic_tests/test_batch_norm_fusion.py +221 -0
  1678. executorch/backends/nxp/tests/generic_tests/test_cifarnet.py +104 -0
  1679. executorch/backends/nxp/tests/generic_tests/test_context_sensitive_delegation.py +180 -0
  1680. executorch/backends/nxp/tests/generic_tests/test_convert_div_to_mul.py +236 -0
  1681. executorch/backends/nxp/tests/generic_tests/test_debug_results.py +155 -0
  1682. executorch/backends/nxp/tests/generic_tests/test_decompose_split_to_slices.py +287 -0
  1683. executorch/backends/nxp/tests/generic_tests/test_gru_splitting.py +267 -0
  1684. executorch/backends/nxp/tests/generic_tests/test_integration.py +54 -0
  1685. executorch/backends/nxp/tests/generic_tests/test_kernel_selection.py +80 -0
  1686. executorch/backends/nxp/tests/generic_tests/test_linear_and_add_fusion.py +653 -0
  1687. executorch/backends/nxp/tests/generic_tests/test_move_activation_before_concatenation.py +901 -0
  1688. executorch/backends/nxp/tests/generic_tests/test_neutron_backend.py +63 -0
  1689. executorch/backends/nxp/tests/generic_tests/test_neutron_backend_executor.py +176 -0
  1690. executorch/backends/nxp/tests/generic_tests/test_neutron_converter_manager.py +55 -0
  1691. executorch/backends/nxp/tests/generic_tests/test_node_format_inference.py +121 -0
  1692. executorch/backends/nxp/tests/generic_tests/test_operator_selector.py +24 -0
  1693. executorch/backends/nxp/tests/generic_tests/test_per_channel_conversion.py +187 -0
  1694. executorch/backends/nxp/tests/generic_tests/test_profiling.py +466 -0
  1695. executorch/backends/nxp/tests/generic_tests/test_qdq_clustering_conv.py +31 -0
  1696. executorch/backends/nxp/tests/generic_tests/test_quantized_input_data.py +115 -0
  1697. executorch/backends/nxp/tests/generic_tests/test_quantizer.py +719 -0
  1698. executorch/backends/nxp/tests/generic_tests/test_removing_dead_code.py +65 -0
  1699. executorch/backends/nxp/tests/generic_tests/test_removing_nodes_with_known_outputs.py +116 -0
  1700. executorch/backends/nxp/tests/generic_tests/test_split_group_convolution.py +307 -0
  1701. executorch/backends/nxp/tests/graph_verifier.py +264 -0
  1702. executorch/backends/nxp/tests/ir/__init__.py +0 -0
  1703. executorch/backends/nxp/tests/ir/converter/__init__.py +0 -0
  1704. executorch/backends/nxp/tests/ir/converter/node_converter/__init__.py +0 -0
  1705. executorch/backends/nxp/tests/ir/converter/node_converter/test_abs_converter.py +102 -0
  1706. executorch/backends/nxp/tests/ir/converter/node_converter/test_adaptive_avg_pool2d_converter.py +142 -0
  1707. executorch/backends/nxp/tests/ir/converter/node_converter/test_add_tensor_converter.py +259 -0
  1708. executorch/backends/nxp/tests/ir/converter/node_converter/test_addmm_converter.py +190 -0
  1709. executorch/backends/nxp/tests/ir/converter/node_converter/test_amax_converter.py +432 -0
  1710. executorch/backends/nxp/tests/ir/converter/node_converter/test_amin_converter.py +432 -0
  1711. executorch/backends/nxp/tests/ir/converter/node_converter/test_avg_pool2d_converter.py +127 -0
  1712. executorch/backends/nxp/tests/ir/converter/node_converter/test_bmm_converter.py +118 -0
  1713. executorch/backends/nxp/tests/ir/converter/node_converter/test_cat_converter.py +163 -0
  1714. executorch/backends/nxp/tests/ir/converter/node_converter/test_clamp_converter.py +224 -0
  1715. executorch/backends/nxp/tests/ir/converter/node_converter/test_clone_converter.py +347 -0
  1716. executorch/backends/nxp/tests/ir/converter/node_converter/test_constant_pad_nd_converter.py +129 -0
  1717. executorch/backends/nxp/tests/ir/converter/node_converter/test_conv_converter.py +1394 -0
  1718. executorch/backends/nxp/tests/ir/converter/node_converter/test_exp_converter.py +72 -0
  1719. executorch/backends/nxp/tests/ir/converter/node_converter/test_hardtanh_converter.py +274 -0
  1720. executorch/backends/nxp/tests/ir/converter/node_converter/test_leaky_relu_converter.py +92 -0
  1721. executorch/backends/nxp/tests/ir/converter/node_converter/test_log_converter.py +76 -0
  1722. executorch/backends/nxp/tests/ir/converter/node_converter/test_max_pool_2d_converter.py +174 -0
  1723. executorch/backends/nxp/tests/ir/converter/node_converter/test_maximum_converter.py +231 -0
  1724. executorch/backends/nxp/tests/ir/converter/node_converter/test_mean_dim_converter.py +410 -0
  1725. executorch/backends/nxp/tests/ir/converter/node_converter/test_minimum_converter.py +231 -0
  1726. executorch/backends/nxp/tests/ir/converter/node_converter/test_mm_converter.py +115 -0
  1727. executorch/backends/nxp/tests/ir/converter/node_converter/test_mul_tensor_converter.py +224 -0
  1728. executorch/backends/nxp/tests/ir/converter/node_converter/test_neg_converter.py +116 -0
  1729. executorch/backends/nxp/tests/ir/converter/node_converter/test_permute_copy_converter.py +231 -0
  1730. executorch/backends/nxp/tests/ir/converter/node_converter/test_prelu_converter.py +238 -0
  1731. executorch/backends/nxp/tests/ir/converter/node_converter/test_relu_converter.py +190 -0
  1732. executorch/backends/nxp/tests/ir/converter/node_converter/test_sigmoid_converter.py +76 -0
  1733. executorch/backends/nxp/tests/ir/converter/node_converter/test_slice_tensor_converter.py +395 -0
  1734. executorch/backends/nxp/tests/ir/converter/node_converter/test_softmax_converter.py +233 -0
  1735. executorch/backends/nxp/tests/ir/converter/node_converter/test_sub_tensor_converter.py +254 -0
  1736. executorch/backends/nxp/tests/ir/converter/node_converter/test_sum_dim_int_list_converter.py +436 -0
  1737. executorch/backends/nxp/tests/ir/converter/node_converter/test_tanh_converter.py +100 -0
  1738. executorch/backends/nxp/tests/ir/converter/node_converter/test_upsample_bilinear2d.py +272 -0
  1739. executorch/backends/nxp/tests/ir/converter/node_converter/test_upsample_nearest2d.py +162 -0
  1740. executorch/backends/nxp/tests/ir/converter/node_converter/test_view_copy_converter.py +420 -0
  1741. executorch/backends/nxp/tests/ir/edge_passes/__init__.py +0 -0
  1742. executorch/backends/nxp/tests/ir/edge_passes/test_convert_reshaping_nodes_to_view.py +126 -0
  1743. executorch/backends/nxp/tests/ir/edge_passes/test_edge_passes.py +402 -0
  1744. executorch/backends/nxp/tests/ir/edge_passes/test_linear_bn_fusing.py +316 -0
  1745. executorch/backends/nxp/tests/ir/edge_passes/test_remove_io_quant_ops_pass.py +128 -0
  1746. executorch/backends/nxp/tests/ir/test_model_builder.py +69 -0
  1747. executorch/backends/nxp/tests/model_output_comparator.py +391 -0
  1748. executorch/backends/nxp/tests/models.py +1083 -0
  1749. executorch/backends/nxp/tests/nsys_testing.py +828 -0
  1750. executorch/backends/nxp/tests/ops_aliases.py +66 -0
  1751. executorch/backends/nxp/tests/outputs_dir.py +10 -0
  1752. executorch/backends/nxp/tests/outputs_dir_importer.py +17 -0
  1753. executorch/backends/nxp/tests/test_convert_1d_conv_to_2d.py +528 -0
  1754. executorch/backends/nxp/tests/test_fold_redundant_qdq.py +86 -0
  1755. executorch/backends/nxp/tests/test_fuse_batch_norm_single_user.py +112 -0
  1756. executorch/backends/nxp/tests/use_qat.py +21 -0
  1757. executorch/backends/nxp/tests/utils.py +66 -0
  1758. executorch/backends/openvino/__init__.py +5 -0
  1759. executorch/backends/openvino/_passes/__init__.py +9 -0
  1760. executorch/backends/openvino/_passes/decompose_floor_divide_pass.py +106 -0
  1761. executorch/backends/openvino/partitioner.py +238 -0
  1762. executorch/backends/openvino/preprocess.py +62 -0
  1763. executorch/backends/openvino/quantizer/__init__.py +9 -0
  1764. executorch/backends/openvino/quantizer/llm_compression.py +133 -0
  1765. executorch/backends/openvino/quantizer/observers.py +206 -0
  1766. executorch/backends/openvino/quantizer/quantizer.py +675 -0
  1767. executorch/backends/openvino/test/tester/__init__.py +19 -0
  1768. executorch/backends/openvino/test/tester/tester.py +94 -0
  1769. executorch/backends/openvino/tests/models/test_classification.py +38 -0
  1770. executorch/backends/openvino/tests/ops/base_openvino_op_test.py +86 -0
  1771. executorch/backends/openvino/tests/ops/test_add.py +22 -0
  1772. executorch/backends/openvino/tests/ops/test_addmm.py +28 -0
  1773. executorch/backends/openvino/tests/ops/test_arange.py +23 -0
  1774. executorch/backends/openvino/tests/ops/test_batch_norm.py +61 -0
  1775. executorch/backends/openvino/tests/ops/test_convolution.py +296 -0
  1776. executorch/backends/openvino/tests/ops/test_mean.py +90 -0
  1777. executorch/backends/openvino/tests/ops/test_permute.py +33 -0
  1778. executorch/backends/openvino/tests/ops/test_pooling.py +91 -0
  1779. executorch/backends/openvino/tests/ops/test_unary_ops.py +37 -0
  1780. executorch/backends/openvino/tests/ops/test_view.py +35 -0
  1781. executorch/backends/openvino/tests/quantizer/synthetic_test_models.py +22 -0
  1782. executorch/backends/openvino/tests/quantizer/test_llm_compression.py +249 -0
  1783. executorch/backends/openvino/tests/test_runner.py +73 -0
  1784. executorch/backends/qualcomm/__init__.py +41 -0
  1785. executorch/backends/qualcomm/_passes/__init__.py +135 -0
  1786. executorch/backends/qualcomm/_passes/annotate_avg_pool1d.py +63 -0
  1787. executorch/backends/qualcomm/_passes/annotate_quant_attrs.py +145 -0
  1788. executorch/backends/qualcomm/_passes/annotate_stack.py +44 -0
  1789. executorch/backends/qualcomm/_passes/annotate_unbind.py +42 -0
  1790. executorch/backends/qualcomm/_passes/backends/__init__.py +5 -0
  1791. executorch/backends/qualcomm/_passes/backends/gpu/__init__.py +11 -0
  1792. executorch/backends/qualcomm/_passes/backends/gpu/qnn_gpu_pass_manager.py +49 -0
  1793. executorch/backends/qualcomm/_passes/backends/htp/__init__.py +11 -0
  1794. executorch/backends/qualcomm/_passes/backends/htp/qnn_htp_pass_manager.py +49 -0
  1795. executorch/backends/qualcomm/_passes/backends/lpai/__init__.py +11 -0
  1796. executorch/backends/qualcomm/_passes/backends/lpai/qnn_lpai_pass_manager.py +95 -0
  1797. executorch/backends/qualcomm/_passes/build_quant_io.py +53 -0
  1798. executorch/backends/qualcomm/_passes/canonicalize_conv.py +216 -0
  1799. executorch/backends/qualcomm/_passes/convert_bmm_to_matmul.py +83 -0
  1800. executorch/backends/qualcomm/_passes/convert_linear_to_conv2d.py +232 -0
  1801. executorch/backends/qualcomm/_passes/convert_mha_to_sha.py +626 -0
  1802. executorch/backends/qualcomm/_passes/convert_square_to_pow.py +38 -0
  1803. executorch/backends/qualcomm/_passes/decompose_acos.py +75 -0
  1804. executorch/backends/qualcomm/_passes/decompose_addmm.py +115 -0
  1805. executorch/backends/qualcomm/_passes/decompose_any.py +64 -0
  1806. executorch/backends/qualcomm/_passes/decompose_atan2.py +210 -0
  1807. executorch/backends/qualcomm/_passes/decompose_binary_alpha.py +61 -0
  1808. executorch/backends/qualcomm/_passes/decompose_cdist.py +74 -0
  1809. executorch/backends/qualcomm/_passes/decompose_col_im.py +139 -0
  1810. executorch/backends/qualcomm/_passes/decompose_diagonal.py +234 -0
  1811. executorch/backends/qualcomm/_passes/decompose_div_mode.py +91 -0
  1812. executorch/backends/qualcomm/_passes/decompose_einsum.py +51 -0
  1813. executorch/backends/qualcomm/_passes/decompose_expm1.py +46 -0
  1814. executorch/backends/qualcomm/_passes/decompose_fill.py +61 -0
  1815. executorch/backends/qualcomm/_passes/decompose_floor_divide.py +62 -0
  1816. executorch/backends/qualcomm/_passes/decompose_glu.py +55 -0
  1817. executorch/backends/qualcomm/_passes/decompose_hardsigmoid.py +59 -0
  1818. executorch/backends/qualcomm/_passes/decompose_hyperbolic_variants.py +202 -0
  1819. executorch/backends/qualcomm/_passes/decompose_linalg_vector_norm.py +81 -0
  1820. executorch/backends/qualcomm/_passes/decompose_log_variants.py +114 -0
  1821. executorch/backends/qualcomm/_passes/decompose_maxpool3d.py +133 -0
  1822. executorch/backends/qualcomm/_passes/decompose_minmaxdim.py +109 -0
  1823. executorch/backends/qualcomm/_passes/decompose_pad.py +57 -0
  1824. executorch/backends/qualcomm/_passes/decompose_reciprocal.py +39 -0
  1825. executorch/backends/qualcomm/_passes/decompose_remainder.py +118 -0
  1826. executorch/backends/qualcomm/_passes/decompose_roll.py +78 -0
  1827. executorch/backends/qualcomm/_passes/decompose_select_scatter.py +96 -0
  1828. executorch/backends/qualcomm/_passes/decompose_silu.py +45 -0
  1829. executorch/backends/qualcomm/_passes/decompose_tan.py +71 -0
  1830. executorch/backends/qualcomm/_passes/decompose_threshold.py +61 -0
  1831. executorch/backends/qualcomm/_passes/decompose_triu.py +71 -0
  1832. executorch/backends/qualcomm/_passes/decompose_trunc.py +93 -0
  1833. executorch/backends/qualcomm/_passes/decompose_var.py +177 -0
  1834. executorch/backends/qualcomm/_passes/decompose_wrap_with_autocast.py +79 -0
  1835. executorch/backends/qualcomm/_passes/expand_broadcast_tensor_shape.py +66 -0
  1836. executorch/backends/qualcomm/_passes/fixed_linear_keep_dim.py +94 -0
  1837. executorch/backends/qualcomm/_passes/fold_qdq.py +83 -0
  1838. executorch/backends/qualcomm/_passes/fuse_consecutive_cast.py +115 -0
  1839. executorch/backends/qualcomm/_passes/fuse_consecutive_transpose.py +108 -0
  1840. executorch/backends/qualcomm/_passes/i64_to_i32.py +184 -0
  1841. executorch/backends/qualcomm/_passes/insert_cast_for_fp_act_quantized_weight.py +141 -0
  1842. executorch/backends/qualcomm/_passes/insert_io_qdq.py +85 -0
  1843. executorch/backends/qualcomm/_passes/insert_requantize.py +111 -0
  1844. executorch/backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py +58 -0
  1845. executorch/backends/qualcomm/_passes/layout_transform.py +360 -0
  1846. executorch/backends/qualcomm/_passes/lift_constant_scalar_operands.py +196 -0
  1847. executorch/backends/qualcomm/_passes/lpai_partition_fallback_support.py +326 -0
  1848. executorch/backends/qualcomm/_passes/qnn_pass_manager.py +522 -0
  1849. executorch/backends/qualcomm/_passes/recompose_pad_maxpool2d.py +144 -0
  1850. executorch/backends/qualcomm/_passes/recompose_pixel_unshuffle.py +81 -0
  1851. executorch/backends/qualcomm/_passes/recompose_rms_norm.py +155 -0
  1852. executorch/backends/qualcomm/_passes/remove_0d_tensor.py +36 -0
  1853. executorch/backends/qualcomm/_passes/remove_redundancy.py +72 -0
  1854. executorch/backends/qualcomm/_passes/replace_arange_args.py +48 -0
  1855. executorch/backends/qualcomm/_passes/resolve_debug_handle.py +47 -0
  1856. executorch/backends/qualcomm/_passes/seq_mse.py +237 -0
  1857. executorch/backends/qualcomm/_passes/tag_quant_io.py +45 -0
  1858. executorch/backends/qualcomm/_passes/utils.py +371 -0
  1859. executorch/backends/qualcomm/builders/__init__.py +235 -0
  1860. executorch/backends/qualcomm/builders/node_visitor.py +572 -0
  1861. executorch/backends/qualcomm/builders/node_visitor_manager.py +81 -0
  1862. executorch/backends/qualcomm/builders/op_abs.py +57 -0
  1863. executorch/backends/qualcomm/builders/op_adaptive_avg_pool2d.py +126 -0
  1864. executorch/backends/qualcomm/builders/op_adaptive_max_pool2d.py +151 -0
  1865. executorch/backends/qualcomm/builders/op_add.py +62 -0
  1866. executorch/backends/qualcomm/builders/op_amax.py +90 -0
  1867. executorch/backends/qualcomm/builders/op_amin.py +90 -0
  1868. executorch/backends/qualcomm/builders/op_and.py +60 -0
  1869. executorch/backends/qualcomm/builders/op_arange.py +41 -0
  1870. executorch/backends/qualcomm/builders/op_argmax.py +78 -0
  1871. executorch/backends/qualcomm/builders/op_argmin.py +78 -0
  1872. executorch/backends/qualcomm/builders/op_asin.py +56 -0
  1873. executorch/backends/qualcomm/builders/op_atan.py +55 -0
  1874. executorch/backends/qualcomm/builders/op_avg_pool2d.py +151 -0
  1875. executorch/backends/qualcomm/builders/op_avg_pool3d.py +293 -0
  1876. executorch/backends/qualcomm/builders/op_batch_norm.py +155 -0
  1877. executorch/backends/qualcomm/builders/op_binary.py +84 -0
  1878. executorch/backends/qualcomm/builders/op_bmm.py +59 -0
  1879. executorch/backends/qualcomm/builders/op_cat.py +89 -0
  1880. executorch/backends/qualcomm/builders/op_ceil.py +56 -0
  1881. executorch/backends/qualcomm/builders/op_channel_shuffle.py +73 -0
  1882. executorch/backends/qualcomm/builders/op_clamp.py +80 -0
  1883. executorch/backends/qualcomm/builders/op_conv.py +263 -0
  1884. executorch/backends/qualcomm/builders/op_copy.py +112 -0
  1885. executorch/backends/qualcomm/builders/op_cos.py +56 -0
  1886. executorch/backends/qualcomm/builders/op_cum_sum.py +87 -0
  1887. executorch/backends/qualcomm/builders/op_custom_op.py +107 -0
  1888. executorch/backends/qualcomm/builders/op_depth_to_space.py +78 -0
  1889. executorch/backends/qualcomm/builders/op_dequantize.py +72 -0
  1890. executorch/backends/qualcomm/builders/op_div.py +62 -0
  1891. executorch/backends/qualcomm/builders/op_elu.py +68 -0
  1892. executorch/backends/qualcomm/builders/op_embedding.py +252 -0
  1893. executorch/backends/qualcomm/builders/op_eq.py +62 -0
  1894. executorch/backends/qualcomm/builders/op_exp.py +60 -0
  1895. executorch/backends/qualcomm/builders/op_expand.py +84 -0
  1896. executorch/backends/qualcomm/builders/op_flip.py +80 -0
  1897. executorch/backends/qualcomm/builders/op_floor.py +56 -0
  1898. executorch/backends/qualcomm/builders/op_full.py +42 -0
  1899. executorch/backends/qualcomm/builders/op_full_like.py +41 -0
  1900. executorch/backends/qualcomm/builders/op_gather.py +75 -0
  1901. executorch/backends/qualcomm/builders/op_ge.py +62 -0
  1902. executorch/backends/qualcomm/builders/op_gelu.py +57 -0
  1903. executorch/backends/qualcomm/builders/op_grid_sampler_2d.py +162 -0
  1904. executorch/backends/qualcomm/builders/op_group_norm.py +96 -0
  1905. executorch/backends/qualcomm/builders/op_gt.py +62 -0
  1906. executorch/backends/qualcomm/builders/op_hardsigmoid.py +78 -0
  1907. executorch/backends/qualcomm/builders/op_hardswish.py +57 -0
  1908. executorch/backends/qualcomm/builders/op_hardtanh.py +79 -0
  1909. executorch/backends/qualcomm/builders/op_index.py +85 -0
  1910. executorch/backends/qualcomm/builders/op_index_put.py +481 -0
  1911. executorch/backends/qualcomm/builders/op_index_select.py +80 -0
  1912. executorch/backends/qualcomm/builders/op_instance_norm.py +112 -0
  1913. executorch/backends/qualcomm/builders/op_is_inf.py +71 -0
  1914. executorch/backends/qualcomm/builders/op_is_nan.py +66 -0
  1915. executorch/backends/qualcomm/builders/op_layer_norm.py +139 -0
  1916. executorch/backends/qualcomm/builders/op_le.py +62 -0
  1917. executorch/backends/qualcomm/builders/op_linear.py +98 -0
  1918. executorch/backends/qualcomm/builders/op_log.py +58 -0
  1919. executorch/backends/qualcomm/builders/op_log_softmax.py +81 -0
  1920. executorch/backends/qualcomm/builders/op_logical_not.py +56 -0
  1921. executorch/backends/qualcomm/builders/op_lt.py +62 -0
  1922. executorch/backends/qualcomm/builders/op_matmul.py +59 -0
  1923. executorch/backends/qualcomm/builders/op_max.py +62 -0
  1924. executorch/backends/qualcomm/builders/op_max_dim.py +89 -0
  1925. executorch/backends/qualcomm/builders/op_max_pool2d.py +145 -0
  1926. executorch/backends/qualcomm/builders/op_mean_dim.py +100 -0
  1927. executorch/backends/qualcomm/builders/op_min.py +62 -0
  1928. executorch/backends/qualcomm/builders/op_min_dim.py +89 -0
  1929. executorch/backends/qualcomm/builders/op_mul.py +62 -0
  1930. executorch/backends/qualcomm/builders/op_ne.py +62 -0
  1931. executorch/backends/qualcomm/builders/op_neg.py +54 -0
  1932. executorch/backends/qualcomm/builders/op_or.py +60 -0
  1933. executorch/backends/qualcomm/builders/op_pad.py +107 -0
  1934. executorch/backends/qualcomm/builders/op_pow.py +75 -0
  1935. executorch/backends/qualcomm/builders/op_prelu.py +69 -0
  1936. executorch/backends/qualcomm/builders/op_quantize.py +73 -0
  1937. executorch/backends/qualcomm/builders/op_rand.py +79 -0
  1938. executorch/backends/qualcomm/builders/op_randn.py +79 -0
  1939. executorch/backends/qualcomm/builders/op_relu.py +58 -0
  1940. executorch/backends/qualcomm/builders/op_repeat.py +68 -0
  1941. executorch/backends/qualcomm/builders/op_reshape.py +55 -0
  1942. executorch/backends/qualcomm/builders/op_resize.py +85 -0
  1943. executorch/backends/qualcomm/builders/op_rms_norm.py +149 -0
  1944. executorch/backends/qualcomm/builders/op_round.py +58 -0
  1945. executorch/backends/qualcomm/builders/op_rsqrt.py +58 -0
  1946. executorch/backends/qualcomm/builders/op_scalar_tensor.py +51 -0
  1947. executorch/backends/qualcomm/builders/op_scatter_elements.py +103 -0
  1948. executorch/backends/qualcomm/builders/op_select_copy.py +89 -0
  1949. executorch/backends/qualcomm/builders/op_sigmoid.py +58 -0
  1950. executorch/backends/qualcomm/builders/op_sign.py +56 -0
  1951. executorch/backends/qualcomm/builders/op_sin.py +56 -0
  1952. executorch/backends/qualcomm/builders/op_skip_ops.py +55 -0
  1953. executorch/backends/qualcomm/builders/op_slice_copy.py +95 -0
  1954. executorch/backends/qualcomm/builders/op_slice_scatter.py +122 -0
  1955. executorch/backends/qualcomm/builders/op_softmax.py +79 -0
  1956. executorch/backends/qualcomm/builders/op_space_to_depth.py +78 -0
  1957. executorch/backends/qualcomm/builders/op_split_with_sizes.py +99 -0
  1958. executorch/backends/qualcomm/builders/op_sqrt.py +60 -0
  1959. executorch/backends/qualcomm/builders/op_squeeze.py +57 -0
  1960. executorch/backends/qualcomm/builders/op_stack.py +71 -0
  1961. executorch/backends/qualcomm/builders/op_sub.py +62 -0
  1962. executorch/backends/qualcomm/builders/op_sum_int_list.py +84 -0
  1963. executorch/backends/qualcomm/builders/op_tanh.py +57 -0
  1964. executorch/backends/qualcomm/builders/op_to.py +149 -0
  1965. executorch/backends/qualcomm/builders/op_topk.py +112 -0
  1966. executorch/backends/qualcomm/builders/op_transpose.py +76 -0
  1967. executorch/backends/qualcomm/builders/op_unary.py +66 -0
  1968. executorch/backends/qualcomm/builders/op_unbind.py +73 -0
  1969. executorch/backends/qualcomm/builders/op_unsqueeze.py +57 -0
  1970. executorch/backends/qualcomm/builders/op_upsample_bilinear2d.py +68 -0
  1971. executorch/backends/qualcomm/builders/op_upsample_nearest2d.py +68 -0
  1972. executorch/backends/qualcomm/builders/op_where.py +82 -0
  1973. executorch/backends/qualcomm/builders/op_xor.py +60 -0
  1974. executorch/backends/qualcomm/builders/qnn_constants.py +704 -0
  1975. executorch/backends/qualcomm/builders/utils.py +162 -0
  1976. executorch/backends/qualcomm/custom_op/annotator.py +137 -0
  1977. executorch/backends/qualcomm/custom_op/interface.py +112 -0
  1978. executorch/backends/qualcomm/debugger/format_outputs.py +235 -0
  1979. executorch/backends/qualcomm/debugger/qcom_numerical_comparator_base.py +194 -0
  1980. executorch/backends/qualcomm/debugger/qcom_numerical_comparator_sample.py +75 -0
  1981. executorch/backends/qualcomm/debugger/qnn_intermediate_debugger.py +101 -0
  1982. executorch/backends/qualcomm/debugger/utils.py +435 -0
  1983. executorch/backends/qualcomm/export_utils.py +969 -0
  1984. executorch/backends/qualcomm/genai_pipeline/__init__.py +49 -0
  1985. executorch/backends/qualcomm/genai_pipeline/configs/__init__.py +41 -0
  1986. executorch/backends/qualcomm/genai_pipeline/configs/compilation_input_config.py +38 -0
  1987. executorch/backends/qualcomm/genai_pipeline/configs/compilation_output_config.py +29 -0
  1988. executorch/backends/qualcomm/genai_pipeline/configs/inference_input_config.py +36 -0
  1989. executorch/backends/qualcomm/genai_pipeline/configs/inference_output_config.py +30 -0
  1990. executorch/backends/qualcomm/genai_pipeline/configs/model_preparation_input_config.py +23 -0
  1991. executorch/backends/qualcomm/genai_pipeline/configs/model_preparation_output_config.py +33 -0
  1992. executorch/backends/qualcomm/genai_pipeline/configs/quantization_input_config.py +38 -0
  1993. executorch/backends/qualcomm/genai_pipeline/configs/quantization_output_config.py +19 -0
  1994. executorch/backends/qualcomm/genai_pipeline/engine_proxy.py +77 -0
  1995. executorch/backends/qualcomm/genai_pipeline/exceptions.py +52 -0
  1996. executorch/backends/qualcomm/genai_pipeline/pipeline_context.py +116 -0
  1997. executorch/backends/qualcomm/genai_pipeline/pipeline_types.py +29 -0
  1998. executorch/backends/qualcomm/genai_pipeline/tests/configs/test_compilation_input_config.py +28 -0
  1999. executorch/backends/qualcomm/genai_pipeline/tests/configs/test_inference_input_config.py +29 -0
  2000. executorch/backends/qualcomm/genai_pipeline/tests/configs/test_model_preparation_input_config.py +28 -0
  2001. executorch/backends/qualcomm/genai_pipeline/tests/configs/test_quantization_input_config.py +29 -0
  2002. executorch/backends/qualcomm/genai_pipeline/tests/test_engine_proxy.py +80 -0
  2003. executorch/backends/qualcomm/genai_pipeline/tests/test_exceptions.py +49 -0
  2004. executorch/backends/qualcomm/genai_pipeline/tests/test_pipeline_context.py +157 -0
  2005. executorch/backends/qualcomm/genai_pipeline/tests/test_utils.py +27 -0
  2006. executorch/backends/qualcomm/partition/common_defs.py +39 -0
  2007. executorch/backends/qualcomm/partition/qnn_partitioner.py +241 -0
  2008. executorch/backends/qualcomm/partition/utils.py +79 -0
  2009. executorch/backends/qualcomm/qnn_preprocess.py +247 -0
  2010. executorch/backends/qualcomm/quantizer/annotators/__init__.py +0 -0
  2011. executorch/backends/qualcomm/quantizer/annotators/htp_rules.py +1879 -0
  2012. executorch/backends/qualcomm/quantizer/annotators/lpai_rules.py +1134 -0
  2013. executorch/backends/qualcomm/quantizer/backend_opinfo_adapter.py +138 -0
  2014. executorch/backends/qualcomm/quantizer/custom_annotation.py +466 -0
  2015. executorch/backends/qualcomm/quantizer/observers/concat_observer.py +76 -0
  2016. executorch/backends/qualcomm/quantizer/observers/per_block_param_observer.py +146 -0
  2017. executorch/backends/qualcomm/quantizer/observers/per_channel_param_observer.py +157 -0
  2018. executorch/backends/qualcomm/quantizer/qconfig.py +1031 -0
  2019. executorch/backends/qualcomm/quantizer/quant_recipe.py +500 -0
  2020. executorch/backends/qualcomm/quantizer/quantizer.py +766 -0
  2021. executorch/backends/qualcomm/quantizer/registry_loader.py +26 -0
  2022. executorch/backends/qualcomm/quantizer/rules.py +258 -0
  2023. executorch/backends/qualcomm/quantizer/validators.py +461 -0
  2024. executorch/backends/qualcomm/recipes/__init__.py +16 -0
  2025. executorch/backends/qualcomm/recipes/qnn_recipe_provider.py +188 -0
  2026. executorch/backends/qualcomm/recipes/qnn_recipe_types.py +28 -0
  2027. executorch/backends/qualcomm/scripts/download_qnn_sdk.py +863 -0
  2028. executorch/backends/qualcomm/serialization/qc_compiler_spec.fbs +369 -0
  2029. executorch/backends/qualcomm/serialization/qc_schema.py +305 -0
  2030. executorch/backends/qualcomm/serialization/qc_schema_serialize.py +59 -0
  2031. executorch/backends/qualcomm/tests/conftest.py +61 -0
  2032. executorch/backends/qualcomm/tests/models.py +2968 -0
  2033. executorch/backends/qualcomm/tests/rework/common/pass/test.py +5 -0
  2034. executorch/backends/qualcomm/tests/rework/common/utils/test.py +5 -0
  2035. executorch/backends/qualcomm/tests/rework/conftest.py +589 -0
  2036. executorch/backends/qualcomm/tests/rework/gpu/conftest.py +5 -0
  2037. executorch/backends/qualcomm/tests/rework/gpu/e2e/genai/test.py +5 -0
  2038. executorch/backends/qualcomm/tests/rework/gpu/e2e/oss/test.py +5 -0
  2039. executorch/backends/qualcomm/tests/rework/gpu/feature/test.py +5 -0
  2040. executorch/backends/qualcomm/tests/rework/gpu/op/test.py +5 -0
  2041. executorch/backends/qualcomm/tests/rework/htp/conftest.py +102 -0
  2042. executorch/backends/qualcomm/tests/rework/htp/e2e/genai/v68/test.py +5 -0
  2043. executorch/backends/qualcomm/tests/rework/htp/e2e/genai/v69/test.py +5 -0
  2044. executorch/backends/qualcomm/tests/rework/htp/e2e/genai/v73/test.py +5 -0
  2045. executorch/backends/qualcomm/tests/rework/htp/e2e/genai/v75/test.py +5 -0
  2046. executorch/backends/qualcomm/tests/rework/htp/e2e/genai/v79/test.py +5 -0
  2047. executorch/backends/qualcomm/tests/rework/htp/e2e/genai/v81/test.py +5 -0
  2048. executorch/backends/qualcomm/tests/rework/htp/e2e/oss/v68/test.py +5 -0
  2049. executorch/backends/qualcomm/tests/rework/htp/e2e/oss/v69/test.py +5 -0
  2050. executorch/backends/qualcomm/tests/rework/htp/e2e/oss/v73/test.py +5 -0
  2051. executorch/backends/qualcomm/tests/rework/htp/e2e/oss/v75/test.py +5 -0
  2052. executorch/backends/qualcomm/tests/rework/htp/e2e/oss/v79/test.py +5 -0
  2053. executorch/backends/qualcomm/tests/rework/htp/e2e/oss/v81/test.py +5 -0
  2054. executorch/backends/qualcomm/tests/rework/htp/feature/conftest.py +37 -0
  2055. executorch/backends/qualcomm/tests/rework/htp/feature/v68/test.py +112 -0
  2056. executorch/backends/qualcomm/tests/rework/htp/feature/v69/test.py +5 -0
  2057. executorch/backends/qualcomm/tests/rework/htp/feature/v73/test.py +5 -0
  2058. executorch/backends/qualcomm/tests/rework/htp/feature/v75/test.py +5 -0
  2059. executorch/backends/qualcomm/tests/rework/htp/feature/v79/test.py +5 -0
  2060. executorch/backends/qualcomm/tests/rework/htp/feature/v81/test.py +5 -0
  2061. executorch/backends/qualcomm/tests/rework/htp/op/v68/test.py +1357 -0
  2062. executorch/backends/qualcomm/tests/rework/htp/op/v69/test.py +5 -0
  2063. executorch/backends/qualcomm/tests/rework/htp/op/v73/test.py +5 -0
  2064. executorch/backends/qualcomm/tests/rework/htp/op/v75/test.py +5 -0
  2065. executorch/backends/qualcomm/tests/rework/htp/op/v79/test.py +5 -0
  2066. executorch/backends/qualcomm/tests/rework/htp/op/v81/test.py +5 -0
  2067. executorch/backends/qualcomm/tests/rework/lpai/conftest.py +5 -0
  2068. executorch/backends/qualcomm/tests/rework/lpai/e2e/oss/v6/test.py +5 -0
  2069. executorch/backends/qualcomm/tests/rework/lpai/feature/v6/test.py +5 -0
  2070. executorch/backends/qualcomm/tests/rework/lpai/op/v6/test.py +5 -0
  2071. executorch/backends/qualcomm/tests/rework/src/feature.py +692 -0
  2072. executorch/backends/qualcomm/tests/rework/src/op.py +5032 -0
  2073. executorch/backends/qualcomm/tests/rework/src/pattern.py +5 -0
  2074. executorch/backends/qualcomm/tests/test_passes.py +475 -0
  2075. executorch/backends/qualcomm/tests/test_qnn_delegate.py +11079 -0
  2076. executorch/backends/qualcomm/tests/test_qnn_delegate_x86.py +47 -0
  2077. executorch/backends/qualcomm/tests/tester.py +117 -0
  2078. executorch/backends/qualcomm/tests/utils.py +930 -0
  2079. executorch/backends/qualcomm/utils/check_qnn_version.py +74 -0
  2080. executorch/backends/qualcomm/utils/constants.py +64 -0
  2081. executorch/backends/qualcomm/utils/qnn_manager_lifecycle.py +94 -0
  2082. executorch/backends/qualcomm/utils/utils.py +1427 -0
  2083. executorch/backends/samsung/_passes/annotate_qparams.py +201 -0
  2084. executorch/backends/samsung/_passes/annotate_scalar_parameters.py +65 -0
  2085. executorch/backends/samsung/_passes/conv1d_to_conv2d.py +97 -0
  2086. executorch/backends/samsung/_passes/customized_constant_prop.py +40 -0
  2087. executorch/backends/samsung/_passes/fold_qdq.py +36 -0
  2088. executorch/backends/samsung/_passes/fuse_conv_act.py +77 -0
  2089. executorch/backends/samsung/_passes/insert_qdq.py +164 -0
  2090. executorch/backends/samsung/_passes/remove_useless_ops.py +87 -0
  2091. executorch/backends/samsung/_passes/replace_scalar_ops.py +46 -0
  2092. executorch/backends/samsung/_passes/utils.py +15 -0
  2093. executorch/backends/samsung/builders/__init__.py +97 -0
  2094. executorch/backends/samsung/builders/node_visitor.py +116 -0
  2095. executorch/backends/samsung/builders/op_add.py +41 -0
  2096. executorch/backends/samsung/builders/op_avg_pool2d.py +70 -0
  2097. executorch/backends/samsung/builders/op_batch_norm.py +56 -0
  2098. executorch/backends/samsung/builders/op_bmm.py +43 -0
  2099. executorch/backends/samsung/builders/op_cat.py +50 -0
  2100. executorch/backends/samsung/builders/op_clamp.py +47 -0
  2101. executorch/backends/samsung/builders/op_constant_pad_nd.py +56 -0
  2102. executorch/backends/samsung/builders/op_conv2d.py +89 -0
  2103. executorch/backends/samsung/builders/op_dequantize.py +19 -0
  2104. executorch/backends/samsung/builders/op_div.py +42 -0
  2105. executorch/backends/samsung/builders/op_embedding.py +41 -0
  2106. executorch/backends/samsung/builders/op_expand_copy.py +77 -0
  2107. executorch/backends/samsung/builders/op_gelu.py +40 -0
  2108. executorch/backends/samsung/builders/op_getitem.py +32 -0
  2109. executorch/backends/samsung/builders/op_hardsigmoid.py +35 -0
  2110. executorch/backends/samsung/builders/op_hardswish.py +35 -0
  2111. executorch/backends/samsung/builders/op_hardtanh.py +42 -0
  2112. executorch/backends/samsung/builders/op_layer_norm.py +53 -0
  2113. executorch/backends/samsung/builders/op_leaky_relu.py +58 -0
  2114. executorch/backends/samsung/builders/op_linear.py +51 -0
  2115. executorch/backends/samsung/builders/op_log_softmax.py +39 -0
  2116. executorch/backends/samsung/builders/op_max_pool2d.py +96 -0
  2117. executorch/backends/samsung/builders/op_maximum.py +37 -0
  2118. executorch/backends/samsung/builders/op_mean_dim.py +49 -0
  2119. executorch/backends/samsung/builders/op_minimum.py +39 -0
  2120. executorch/backends/samsung/builders/op_mul.py +43 -0
  2121. executorch/backends/samsung/builders/op_permute.py +37 -0
  2122. executorch/backends/samsung/builders/op_pixel_shuffle.py +38 -0
  2123. executorch/backends/samsung/builders/op_quantize.py +60 -0
  2124. executorch/backends/samsung/builders/op_relu.py +37 -0
  2125. executorch/backends/samsung/builders/op_reshape.py +34 -0
  2126. executorch/backends/samsung/builders/op_rsqrt.py +35 -0
  2127. executorch/backends/samsung/builders/op_select.py +52 -0
  2128. executorch/backends/samsung/builders/op_slice_copy.py +59 -0
  2129. executorch/backends/samsung/builders/op_softmax.py +39 -0
  2130. executorch/backends/samsung/builders/op_sqrt.py +35 -0
  2131. executorch/backends/samsung/builders/op_squeeze.py +37 -0
  2132. executorch/backends/samsung/builders/op_sub.py +39 -0
  2133. executorch/backends/samsung/builders/op_to_copy.py +44 -0
  2134. executorch/backends/samsung/builders/op_unsqueeze.py +35 -0
  2135. executorch/backends/samsung/builders/op_upsample_bilinear2d.py +53 -0
  2136. executorch/backends/samsung/builders/op_upsample_nearest2d.py +52 -0
  2137. executorch/backends/samsung/builders/utils.py +64 -0
  2138. executorch/backends/samsung/enn_preprocess.py +110 -0
  2139. executorch/backends/samsung/partition/enn_partitioner.py +134 -0
  2140. executorch/backends/samsung/quantizer/__init__.py +10 -0
  2141. executorch/backends/samsung/quantizer/annotator.py +871 -0
  2142. executorch/backends/samsung/quantizer/qconfig.py +174 -0
  2143. executorch/backends/samsung/quantizer/quantizer.py +65 -0
  2144. executorch/backends/samsung/serialization/compile_options.py +78 -0
  2145. executorch/backends/samsung/serialization/compile_options_def.fbs +23 -0
  2146. executorch/backends/samsung/serialization/enn_graph_schema.py +141 -0
  2147. executorch/backends/samsung/test/models/test_deeplab_v3.py +28 -0
  2148. executorch/backends/samsung/test/models/test_edsr.py +30 -0
  2149. executorch/backends/samsung/test/models/test_inception_v3.py +33 -0
  2150. executorch/backends/samsung/test/models/test_inception_v4.py +30 -0
  2151. executorch/backends/samsung/test/models/test_mobilenet_v2.py +28 -0
  2152. executorch/backends/samsung/test/models/test_mobilenet_v3.py +33 -0
  2153. executorch/backends/samsung/test/models/test_resnet18.py +30 -0
  2154. executorch/backends/samsung/test/models/test_resnet50.py +30 -0
  2155. executorch/backends/samsung/test/models/test_torchvision_vit.py +32 -0
  2156. executorch/backends/samsung/test/models/test_wav2letter.py +28 -0
  2157. executorch/backends/samsung/test/ops/test_add.py +63 -0
  2158. executorch/backends/samsung/test/ops/test_avg_pool2d.py +66 -0
  2159. executorch/backends/samsung/test/ops/test_batch_norm.py +52 -0
  2160. executorch/backends/samsung/test/ops/test_bmm.py +50 -0
  2161. executorch/backends/samsung/test/ops/test_cat.py +49 -0
  2162. executorch/backends/samsung/test/ops/test_clamp.py +50 -0
  2163. executorch/backends/samsung/test/ops/test_constant_pad_nd.py +46 -0
  2164. executorch/backends/samsung/test/ops/test_conv2d.py +95 -0
  2165. executorch/backends/samsung/test/ops/test_div.py +50 -0
  2166. executorch/backends/samsung/test/ops/test_embedding.py +46 -0
  2167. executorch/backends/samsung/test/ops/test_expand_copy.py +46 -0
  2168. executorch/backends/samsung/test/ops/test_gelu.py +65 -0
  2169. executorch/backends/samsung/test/ops/test_leaky_relu.py +47 -0
  2170. executorch/backends/samsung/test/ops/test_linear.py +48 -0
  2171. executorch/backends/samsung/test/ops/test_log_softmax.py +46 -0
  2172. executorch/backends/samsung/test/ops/test_max_pool2d.py +72 -0
  2173. executorch/backends/samsung/test/ops/test_mean_dim.py +51 -0
  2174. executorch/backends/samsung/test/ops/test_minimum.py +45 -0
  2175. executorch/backends/samsung/test/ops/test_mul.py +63 -0
  2176. executorch/backends/samsung/test/ops/test_permute.py +55 -0
  2177. executorch/backends/samsung/test/ops/test_pixel_shuffle.py +49 -0
  2178. executorch/backends/samsung/test/ops/test_relu.py +65 -0
  2179. executorch/backends/samsung/test/ops/test_reshape.py +46 -0
  2180. executorch/backends/samsung/test/ops/test_rsqrt.py +46 -0
  2181. executorch/backends/samsung/test/ops/test_select.py +52 -0
  2182. executorch/backends/samsung/test/ops/test_slice_copy.py +46 -0
  2183. executorch/backends/samsung/test/ops/test_softmax.py +46 -0
  2184. executorch/backends/samsung/test/ops/test_sqrt.py +46 -0
  2185. executorch/backends/samsung/test/ops/test_squeeze.py +47 -0
  2186. executorch/backends/samsung/test/ops/test_sub.py +63 -0
  2187. executorch/backends/samsung/test/ops/test_to_copy.py +49 -0
  2188. executorch/backends/samsung/test/ops/test_unsqueeze.py +47 -0
  2189. executorch/backends/samsung/test/ops/test_upsample_bilinear2d.py +54 -0
  2190. executorch/backends/samsung/test/ops/test_upsample_nearest2d.py +53 -0
  2191. executorch/backends/samsung/test/tester/__init__.py +11 -0
  2192. executorch/backends/samsung/test/tester/samsung_tester.py +125 -0
  2193. executorch/backends/samsung/test/utils/__init__.py +11 -0
  2194. executorch/backends/samsung/test/utils/runtime_executor.py +186 -0
  2195. executorch/backends/samsung/utils/constants.py +45 -0
  2196. executorch/backends/samsung/utils/export_utils.py +93 -0
  2197. executorch/backends/samsung/utils/utils.py +125 -0
  2198. executorch/backends/test/facto/__init__.py +0 -0
  2199. executorch/backends/test/facto/facto_specs.py +59 -0
  2200. executorch/backends/test/facto/test_facto.py +301 -0
  2201. executorch/backends/test/graph_builder.py +158 -0
  2202. executorch/backends/test/harness/__init__.py +3 -0
  2203. executorch/backends/test/harness/error_statistics.py +99 -0
  2204. executorch/backends/test/harness/stages/__init__.py +23 -0
  2205. executorch/backends/test/harness/stages/export.py +32 -0
  2206. executorch/backends/test/harness/stages/partition.py +26 -0
  2207. executorch/backends/test/harness/stages/quantize.py +130 -0
  2208. executorch/backends/test/harness/stages/run_passes.py +66 -0
  2209. executorch/backends/test/harness/stages/serialize.py +59 -0
  2210. executorch/backends/test/harness/stages/stage.py +91 -0
  2211. executorch/backends/test/harness/stages/to_edge.py +27 -0
  2212. executorch/backends/test/harness/stages/to_edge_transform_and_lower.py +50 -0
  2213. executorch/backends/test/harness/stages/to_executorch.py +54 -0
  2214. executorch/backends/test/harness/tester.py +543 -0
  2215. executorch/backends/test/harness/tests/test_error_statistics.py +67 -0
  2216. executorch/backends/test/program_builder.py +138 -0
  2217. executorch/backends/test/suite/__init__.py +74 -0
  2218. executorch/backends/test/suite/conftest.py +221 -0
  2219. executorch/backends/test/suite/context.py +34 -0
  2220. executorch/backends/test/suite/discovery.py +100 -0
  2221. executorch/backends/test/suite/flow.py +202 -0
  2222. executorch/backends/test/suite/flows/__init__.py +7 -0
  2223. executorch/backends/test/suite/flows/arm.py +98 -0
  2224. executorch/backends/test/suite/flows/coreml.py +31 -0
  2225. executorch/backends/test/suite/flows/mlx.py +14 -0
  2226. executorch/backends/test/suite/flows/openvino.py +60 -0
  2227. executorch/backends/test/suite/flows/portable.py +19 -0
  2228. executorch/backends/test/suite/flows/qualcomm.py +61 -0
  2229. executorch/backends/test/suite/flows/vulkan.py +44 -0
  2230. executorch/backends/test/suite/flows/webgpu.py +36 -0
  2231. executorch/backends/test/suite/flows/xnnpack.py +79 -0
  2232. executorch/backends/test/suite/generate_markdown_summary.py +249 -0
  2233. executorch/backends/test/suite/generate_markdown_summary_json.py +229 -0
  2234. executorch/backends/test/suite/models/__init__.py +7 -0
  2235. executorch/backends/test/suite/models/test_torchaudio.py +110 -0
  2236. executorch/backends/test/suite/models/test_torchvision.py +194 -0
  2237. executorch/backends/test/suite/operators/__init__.py +108 -0
  2238. executorch/backends/test/suite/operators/test_abs.py +60 -0
  2239. executorch/backends/test/suite/operators/test_adaptive_avgpool1d.py +102 -0
  2240. executorch/backends/test/suite/operators/test_adaptive_avgpool2d.py +112 -0
  2241. executorch/backends/test/suite/operators/test_adaptive_avgpool3d.py +112 -0
  2242. executorch/backends/test/suite/operators/test_adaptive_maxpool1d.py +125 -0
  2243. executorch/backends/test/suite/operators/test_adaptive_maxpool2d.py +135 -0
  2244. executorch/backends/test/suite/operators/test_adaptive_maxpool3d.py +135 -0
  2245. executorch/backends/test/suite/operators/test_add.py +76 -0
  2246. executorch/backends/test/suite/operators/test_amax.py +257 -0
  2247. executorch/backends/test/suite/operators/test_amin.py +259 -0
  2248. executorch/backends/test/suite/operators/test_argmax.py +201 -0
  2249. executorch/backends/test/suite/operators/test_argmin.py +201 -0
  2250. executorch/backends/test/suite/operators/test_avgpool1d.py +155 -0
  2251. executorch/backends/test/suite/operators/test_avgpool2d.py +168 -0
  2252. executorch/backends/test/suite/operators/test_avgpool3d.py +163 -0
  2253. executorch/backends/test/suite/operators/test_cat.py +176 -0
  2254. executorch/backends/test/suite/operators/test_ceil.py +60 -0
  2255. executorch/backends/test/suite/operators/test_clamp.py +86 -0
  2256. executorch/backends/test/suite/operators/test_conv1d.py +144 -0
  2257. executorch/backends/test/suite/operators/test_conv2d.py +167 -0
  2258. executorch/backends/test/suite/operators/test_conv3d.py +162 -0
  2259. executorch/backends/test/suite/operators/test_convtranspose1d.py +144 -0
  2260. executorch/backends/test/suite/operators/test_convtranspose2d.py +171 -0
  2261. executorch/backends/test/suite/operators/test_convtranspose3d.py +166 -0
  2262. executorch/backends/test/suite/operators/test_div.py +110 -0
  2263. executorch/backends/test/suite/operators/test_elu.py +49 -0
  2264. executorch/backends/test/suite/operators/test_embedding.py +89 -0
  2265. executorch/backends/test/suite/operators/test_embedding_bag.py +118 -0
  2266. executorch/backends/test/suite/operators/test_exp.py +65 -0
  2267. executorch/backends/test/suite/operators/test_expand.py +122 -0
  2268. executorch/backends/test/suite/operators/test_floor.py +57 -0
  2269. executorch/backends/test/suite/operators/test_floor_divide.py +216 -0
  2270. executorch/backends/test/suite/operators/test_gelu.py +52 -0
  2271. executorch/backends/test/suite/operators/test_glu.py +52 -0
  2272. executorch/backends/test/suite/operators/test_hardsigmoid.py +50 -0
  2273. executorch/backends/test/suite/operators/test_hardswish.py +50 -0
  2274. executorch/backends/test/suite/operators/test_hardtanh.py +57 -0
  2275. executorch/backends/test/suite/operators/test_index_put.py +455 -0
  2276. executorch/backends/test/suite/operators/test_index_select.py +128 -0
  2277. executorch/backends/test/suite/operators/test_leaky_relu.py +56 -0
  2278. executorch/backends/test/suite/operators/test_linear.py +124 -0
  2279. executorch/backends/test/suite/operators/test_log.py +80 -0
  2280. executorch/backends/test/suite/operators/test_log10.py +60 -0
  2281. executorch/backends/test/suite/operators/test_log1p.py +60 -0
  2282. executorch/backends/test/suite/operators/test_log2.py +60 -0
  2283. executorch/backends/test/suite/operators/test_logsigmoid.py +40 -0
  2284. executorch/backends/test/suite/operators/test_lstm.py +247 -0
  2285. executorch/backends/test/suite/operators/test_masked_fill.py +101 -0
  2286. executorch/backends/test/suite/operators/test_maxpool1d.py +185 -0
  2287. executorch/backends/test/suite/operators/test_maxpool2d.py +191 -0
  2288. executorch/backends/test/suite/operators/test_maxpool3d.py +189 -0
  2289. executorch/backends/test/suite/operators/test_mean.py +305 -0
  2290. executorch/backends/test/suite/operators/test_median.py +188 -0
  2291. executorch/backends/test/suite/operators/test_mul.py +66 -0
  2292. executorch/backends/test/suite/operators/test_neg.py +70 -0
  2293. executorch/backends/test/suite/operators/test_permute.py +109 -0
  2294. executorch/backends/test/suite/operators/test_pow.py +149 -0
  2295. executorch/backends/test/suite/operators/test_prelu.py +55 -0
  2296. executorch/backends/test/suite/operators/test_relu.py +45 -0
  2297. executorch/backends/test/suite/operators/test_reshape.py +82 -0
  2298. executorch/backends/test/suite/operators/test_round.py +129 -0
  2299. executorch/backends/test/suite/operators/test_rsqrt.py +76 -0
  2300. executorch/backends/test/suite/operators/test_select.py +89 -0
  2301. executorch/backends/test/suite/operators/test_sigmoid.py +40 -0
  2302. executorch/backends/test/suite/operators/test_silu.py +50 -0
  2303. executorch/backends/test/suite/operators/test_slice.py +146 -0
  2304. executorch/backends/test/suite/operators/test_split.py +119 -0
  2305. executorch/backends/test/suite/operators/test_sqrt.py +71 -0
  2306. executorch/backends/test/suite/operators/test_square.py +67 -0
  2307. executorch/backends/test/suite/operators/test_squeeze.py +86 -0
  2308. executorch/backends/test/suite/operators/test_stack.py +114 -0
  2309. executorch/backends/test/suite/operators/test_sub.py +84 -0
  2310. executorch/backends/test/suite/operators/test_tanh.py +40 -0
  2311. executorch/backends/test/suite/operators/test_threshold.py +72 -0
  2312. executorch/backends/test/suite/operators/test_transpose.py +143 -0
  2313. executorch/backends/test/suite/operators/test_trunc.py +73 -0
  2314. executorch/backends/test/suite/operators/test_unsqueeze.py +110 -0
  2315. executorch/backends/test/suite/operators/test_upsample_bilinear2d.py +273 -0
  2316. executorch/backends/test/suite/operators/test_upsample_nearest2d.py +137 -0
  2317. executorch/backends/test/suite/operators/test_view.py +82 -0
  2318. executorch/backends/test/suite/reporting.py +427 -0
  2319. executorch/backends/test/suite/runner.py +350 -0
  2320. executorch/backends/test/suite/tests/__init__.py +0 -0
  2321. executorch/backends/test/suite/tests/test_reporting.py +144 -0
  2322. executorch/backends/transforms/__init__.py +12 -0
  2323. executorch/backends/transforms/addmm_mm_to_linear.py +170 -0
  2324. executorch/backends/transforms/aten_to_dialect_pass.py +121 -0
  2325. executorch/backends/transforms/channels_last_ops.py +153 -0
  2326. executorch/backends/transforms/convert_dtype_pass.py +103 -0
  2327. executorch/backends/transforms/decompose_channels_last_pass.py +92 -0
  2328. executorch/backends/transforms/decompose_sdpa.py +208 -0
  2329. executorch/backends/transforms/duplicate_dynamic_quant_chain.py +195 -0
  2330. executorch/backends/transforms/fuse_batch_norm_with_conv.py +161 -0
  2331. executorch/backends/transforms/fuse_cascaded_transpose_or_permute_ops.py +181 -0
  2332. executorch/backends/transforms/fuse_cascaded_view_ops.py +43 -0
  2333. executorch/backends/transforms/fuse_conv_with_clamp.py +78 -0
  2334. executorch/backends/transforms/fuse_transpose_or_permute_op_pairs_pass.py +102 -0
  2335. executorch/backends/transforms/fuse_view_copy.py +139 -0
  2336. executorch/backends/transforms/mean_to_sum_div.py +41 -0
  2337. executorch/backends/transforms/permute_pass_utils.py +278 -0
  2338. executorch/backends/transforms/postpone_permute_below_squeeze_view.py +207 -0
  2339. executorch/backends/transforms/quantize_fused_convbn_bias_pass.py +363 -0
  2340. executorch/backends/transforms/rank_0_to_rank_1.py +18 -0
  2341. executorch/backends/transforms/remove_clone_ops.py +109 -0
  2342. executorch/backends/transforms/remove_getitem_op.py +109 -0
  2343. executorch/backends/transforms/remove_permutes_around_elementwise_ops.py +589 -0
  2344. executorch/backends/transforms/replace_nop_transpose_or_permute_with_view.py +90 -0
  2345. executorch/backends/transforms/replace_scalar_with_tensor.py +99 -0
  2346. executorch/backends/transforms/test/common.py +69 -0
  2347. executorch/backends/transforms/test/test_aten_to_dialect_pass.py +208 -0
  2348. executorch/backends/transforms/test/test_channels_last_ops.py +226 -0
  2349. executorch/backends/transforms/test/test_create_delete_constant_placeholder.py +122 -0
  2350. executorch/backends/transforms/test/test_create_mutable_buffer.py +503 -0
  2351. executorch/backends/transforms/test/test_decompose_channels_last_pass.py +210 -0
  2352. executorch/backends/transforms/test/test_decompose_sdpa.py +96 -0
  2353. executorch/backends/transforms/test/test_duplicate_dynamic_quant_chain.py +184 -0
  2354. executorch/backends/transforms/test/test_permute_optimization_passes.py +1099 -0
  2355. executorch/backends/transforms/test/test_quantize_fused_convbn_bias_pass.py +174 -0
  2356. executorch/backends/transforms/test/test_rank_0_to_rank_1.py +32 -0
  2357. executorch/backends/transforms/test/test_remove_clone_ops.py +197 -0
  2358. executorch/backends/transforms/test/test_to_contiguous_channels_last_pass.py +700 -0
  2359. executorch/backends/transforms/utils.py +386 -0
  2360. executorch/backends/transforms/view_copy_to_squeeze_unsqueeze.py +135 -0
  2361. executorch/backends/vulkan/__init__.py +14 -0
  2362. executorch/backends/vulkan/_passes/__init__.py +41 -0
  2363. executorch/backends/vulkan/_passes/fold_qdq.py +40 -0
  2364. executorch/backends/vulkan/_passes/fuse_patterns.py +34 -0
  2365. executorch/backends/vulkan/_passes/fuse_quantized_ops.py +236 -0
  2366. executorch/backends/vulkan/_passes/insert_dtype_promotion.py +102 -0
  2367. executorch/backends/vulkan/_passes/insert_prepack_nodes.py +105 -0
  2368. executorch/backends/vulkan/_passes/remove_asserts.py +52 -0
  2369. executorch/backends/vulkan/_passes/remove_redundant_ops.py +83 -0
  2370. executorch/backends/vulkan/_passes/squeeze_unsqueeze_inputs.py +93 -0
  2371. executorch/backends/vulkan/_passes/tag_memory_meta_pass.py +549 -0
  2372. executorch/backends/vulkan/custom_ops_lib.py +1321 -0
  2373. executorch/backends/vulkan/op_registry.py +1852 -0
  2374. executorch/backends/vulkan/partitioner/vulkan_partitioner.py +424 -0
  2375. executorch/backends/vulkan/patterns/__init__.py +151 -0
  2376. executorch/backends/vulkan/patterns/pattern_registry.py +116 -0
  2377. executorch/backends/vulkan/patterns/quantized_binary.py +161 -0
  2378. executorch/backends/vulkan/patterns/quantized_convolution.py +347 -0
  2379. executorch/backends/vulkan/patterns/quantized_embedding.py +342 -0
  2380. executorch/backends/vulkan/patterns/quantized_linear.py +645 -0
  2381. executorch/backends/vulkan/patterns/quantized_pixel_shuffle.py +180 -0
  2382. executorch/backends/vulkan/patterns/quantized_unary.py +121 -0
  2383. executorch/backends/vulkan/patterns/rms_norm.py +280 -0
  2384. executorch/backends/vulkan/patterns/rope.py +173 -0
  2385. executorch/backends/vulkan/patterns/rope_hf.py +188 -0
  2386. executorch/backends/vulkan/patterns/sdpa.py +155 -0
  2387. executorch/backends/vulkan/patterns/select_as_symint.py +104 -0
  2388. executorch/backends/vulkan/patterns/weight_packing_utils.py +51 -0
  2389. executorch/backends/vulkan/quantizer/vulkan_quantizer.py +183 -0
  2390. executorch/backends/vulkan/quantizer/vulkan_quantizer_utils.py +223 -0
  2391. executorch/backends/vulkan/runtime/gen_vulkan_spv.py +1673 -0
  2392. executorch/backends/vulkan/runtime/graph/ops/glsl/adamw_step.yaml +17 -0
  2393. executorch/backends/vulkan/runtime/graph/ops/glsl/apply_rotary_emb_interleaved.yaml +13 -0
  2394. executorch/backends/vulkan/runtime/graph/ops/glsl/arange_buffer.yaml +17 -0
  2395. executorch/backends/vulkan/runtime/graph/ops/glsl/arange_texture.yaml +16 -0
  2396. executorch/backends/vulkan/runtime/graph/ops/glsl/avg_pool2d.yaml +18 -0
  2397. executorch/backends/vulkan/runtime/graph/ops/glsl/batchnorm.yaml +11 -0
  2398. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_op_buffer.yaml +59 -0
  2399. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_op_texture.yaml +61 -0
  2400. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.yaml +26 -0
  2401. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_texture.yaml +26 -0
  2402. executorch/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.yaml +22 -0
  2403. executorch/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.yaml +20 -0
  2404. executorch/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.yaml +25 -0
  2405. executorch/backends/vulkan/runtime/graph/ops/glsl/choose_qparams_per_row.yaml +23 -0
  2406. executorch/backends/vulkan/runtime/graph/ops/glsl/clone.yaml +13 -0
  2407. executorch/backends/vulkan/runtime/graph/ops/glsl/col2im.yaml +19 -0
  2408. executorch/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.yaml +15 -0
  2409. executorch/backends/vulkan/runtime/graph/ops/glsl/concat_texture.yaml +15 -0
  2410. executorch/backends/vulkan/runtime/graph/ops/glsl/conv1d.yaml +19 -0
  2411. executorch/backends/vulkan/runtime/graph/ops/glsl/conv1d_dw.yaml +22 -0
  2412. executorch/backends/vulkan/runtime/graph/ops/glsl/conv1d_pw.yaml +31 -0
  2413. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d.yaml +19 -0
  2414. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.yaml +19 -0
  2415. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.yaml +34 -0
  2416. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_prepack_weights.yaml +19 -0
  2417. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.yaml +20 -0
  2418. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_sned_output_tile.yaml +25 -0
  2419. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_gemm.yaml +26 -0
  2420. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_im2col.yaml +22 -0
  2421. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_prepack_weights.yaml +19 -0
  2422. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.yaml +21 -0
  2423. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_q8ta_q8csw_q8to_tiled.yaml +20 -0
  2424. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_s1p0.yaml +19 -0
  2425. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_tiled.yaml +20 -0
  2426. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8csw_linear_tiled.yaml +22 -0
  2427. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_linear_tiled.yaml +22 -0
  2428. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to.yaml +21 -0
  2429. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to_linear_tiled.yaml +21 -0
  2430. executorch/backends/vulkan/runtime/graph/ops/glsl/conv_transpose2d.yaml +19 -0
  2431. executorch/backends/vulkan/runtime/graph/ops/glsl/conv_transpose2d_prepack_weights.yaml +20 -0
  2432. executorch/backends/vulkan/runtime/graph/ops/glsl/coopmat_mm.yaml +43 -0
  2433. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.yaml +16 -0
  2434. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_legacy.yaml +12 -0
  2435. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_q4gsw.yaml +38 -0
  2436. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_texture.yaml +15 -0
  2437. executorch/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.yaml +11 -0
  2438. executorch/backends/vulkan/runtime/graph/ops/glsl/expand_texture.yaml +11 -0
  2439. executorch/backends/vulkan/runtime/graph/ops/glsl/flip.yaml +14 -0
  2440. executorch/backends/vulkan/runtime/graph/ops/glsl/full_buffer.yaml +18 -0
  2441. executorch/backends/vulkan/runtime/graph/ops/glsl/full_texture.yaml +18 -0
  2442. executorch/backends/vulkan/runtime/graph/ops/glsl/fused_ce.yaml +15 -0
  2443. executorch/backends/vulkan/runtime/graph/ops/glsl/fused_ce_sum.yaml +15 -0
  2444. executorch/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.yaml +18 -0
  2445. executorch/backends/vulkan/runtime/graph/ops/glsl/gather_texture.yaml +17 -0
  2446. executorch/backends/vulkan/runtime/graph/ops/glsl/grid_priors.yaml +12 -0
  2447. executorch/backends/vulkan/runtime/graph/ops/glsl/grid_sampler_2d.yaml +16 -0
  2448. executorch/backends/vulkan/runtime/graph/ops/glsl/group_norm_reduce_texture.yaml +15 -0
  2449. executorch/backends/vulkan/runtime/graph/ops/glsl/group_norm_texture.yaml +15 -0
  2450. executorch/backends/vulkan/runtime/graph/ops/glsl/im2col.yaml +18 -0
  2451. executorch/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.yaml +15 -0
  2452. executorch/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.yaml +35 -0
  2453. executorch/backends/vulkan/runtime/graph/ops/glsl/index_select.yaml +13 -0
  2454. executorch/backends/vulkan/runtime/graph/ops/glsl/index_select_channel.yaml +13 -0
  2455. executorch/backends/vulkan/runtime/graph/ops/glsl/index_tensor_buffer.yaml +16 -0
  2456. executorch/backends/vulkan/runtime/graph/ops/glsl/index_tensor_texture.yaml +16 -0
  2457. executorch/backends/vulkan/runtime/graph/ops/glsl/int8x4_buffer_to_nchw.yaml +11 -0
  2458. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_dW.yaml +17 -0
  2459. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_dq8ca_q4gsw_tiled.yaml +33 -0
  2460. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_dq8ca_qw_coopmat.yaml +40 -0
  2461. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.yaml +72 -0
  2462. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_tiled.yaml +28 -0
  2463. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q8csw_tiled.yaml +28 -0
  2464. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q8ta_q8csw_tiled.yaml +24 -0
  2465. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.yaml +23 -0
  2466. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.yaml +38 -0
  2467. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.yaml +54 -0
  2468. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qw_coopmat.yaml +40 -0
  2469. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_scalar.yaml +39 -0
  2470. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_vec.yaml +41 -0
  2471. executorch/backends/vulkan/runtime/graph/ops/glsl/matmul_scalar.yaml +35 -0
  2472. executorch/backends/vulkan/runtime/graph/ops/glsl/matmul_vec.yaml +36 -0
  2473. executorch/backends/vulkan/runtime/graph/ops/glsl/max_pool2d.yaml +17 -0
  2474. executorch/backends/vulkan/runtime/graph/ops/glsl/native_layer_norm_buffer.yaml +16 -0
  2475. executorch/backends/vulkan/runtime/graph/ops/glsl/native_layer_norm_texture.yaml +15 -0
  2476. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_bitw8_image_nobitw8buffer.yaml +22 -0
  2477. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.yaml +25 -0
  2478. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_image.yaml +28 -0
  2479. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_int8x4_buffer.yaml +11 -0
  2480. executorch/backends/vulkan/runtime/graph/ops/glsl/no_op.yaml +24 -0
  2481. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_conv2d_gemm_weight.yaml +20 -0
  2482. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_fp_linear_weight.yaml +22 -0
  2483. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_int4_linear_weight_transposed_interleaved.yaml +19 -0
  2484. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q4_linear_weight.yaml +14 -0
  2485. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q4_linear_weight__w_4x8.yaml +22 -0
  2486. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_conv2d_dw_weights.yaml +15 -0
  2487. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_conv2d_weights.yaml +15 -0
  2488. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_linear_weight.yaml +14 -0
  2489. executorch/backends/vulkan/runtime/graph/ops/glsl/pad_buffer.yaml +11 -0
  2490. executorch/backends/vulkan/runtime/graph/ops/glsl/pad_texture.yaml +11 -0
  2491. executorch/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.yaml +11 -0
  2492. executorch/backends/vulkan/runtime/graph/ops/glsl/permute_texture.yaml +11 -0
  2493. executorch/backends/vulkan/runtime/graph/ops/glsl/pixel_shuffle_buffer.yaml +16 -0
  2494. executorch/backends/vulkan/runtime/graph/ops/glsl/pixel_shuffle_texture.yaml +16 -0
  2495. executorch/backends/vulkan/runtime/graph/ops/glsl/q4gsw_backward.yaml +17 -0
  2496. executorch/backends/vulkan/runtime/graph/ops/glsl/q4gsw_linear_gemm__tin__w_4x8.yaml +34 -0
  2497. executorch/backends/vulkan/runtime/graph/ops/glsl/q4gsw_linear_gemm__w_4x8.yaml +29 -0
  2498. executorch/backends/vulkan/runtime/graph/ops/glsl/q4gsw_linear_gemv_coop__w_4x8.yaml +41 -0
  2499. executorch/backends/vulkan/runtime/graph/ops/glsl/q4gsw_requant.yaml +17 -0
  2500. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_binary.yaml +12 -0
  2501. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_clone.yaml +11 -0
  2502. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_conv2d.yaml +17 -0
  2503. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_conv2d_dw.yaml +14 -0
  2504. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_conv2d_pw.yaml +17 -0
  2505. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_conv2d_transposed.yaml +17 -0
  2506. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_dequantize.yaml +18 -0
  2507. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_im2col.yaml +11 -0
  2508. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_linear.yaml +21 -0
  2509. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_linear_gemv.yaml +21 -0
  2510. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_pixel_shuffle.yaml +11 -0
  2511. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_quantize.yaml +18 -0
  2512. executorch/backends/vulkan/runtime/graph/ops/glsl/q8ta_unary.yaml +12 -0
  2513. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4h4w.yaml +25 -0
  2514. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4h4w_with_group_sums.yaml +34 -0
  2515. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4w4c.yaml +23 -0
  2516. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_im2col.yaml +18 -0
  2517. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce.yaml +29 -0
  2518. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce2d.yaml +29 -0
  2519. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce_per_row_buffer.yaml +42 -0
  2520. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_buffer.yaml +13 -0
  2521. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_channel.yaml +12 -0
  2522. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_interleave.yaml +12 -0
  2523. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_texture.yaml +12 -0
  2524. executorch/backends/vulkan/runtime/graph/ops/glsl/rms_norm_buffer.yaml +16 -0
  2525. executorch/backends/vulkan/runtime/graph/ops/glsl/rms_norm_texture.yaml +15 -0
  2526. executorch/backends/vulkan/runtime/graph/ops/glsl/rotary_embedding.yaml +13 -0
  2527. executorch/backends/vulkan/runtime/graph/ops/glsl/rotary_embedding_hf.yaml +13 -0
  2528. executorch/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.yaml +27 -0
  2529. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_attn_weights_softmax.yaml +35 -0
  2530. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_attn_weights_coop.yaml +25 -0
  2531. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_attn_weights_tiled.yaml +60 -0
  2532. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_out_coop.yaml +25 -0
  2533. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_out_tiled.yaml +39 -0
  2534. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_kv_cache_update.yaml +23 -0
  2535. executorch/backends/vulkan/runtime/graph/ops/glsl/set_zero.yaml +8 -0
  2536. executorch/backends/vulkan/runtime/graph/ops/glsl/softmax.yaml +21 -0
  2537. executorch/backends/vulkan/runtime/graph/ops/glsl/softmax_buffer.yaml +21 -0
  2538. executorch/backends/vulkan/runtime/graph/ops/glsl/split_buffer.yaml +18 -0
  2539. executorch/backends/vulkan/runtime/graph/ops/glsl/split_texture.yaml +17 -0
  2540. executorch/backends/vulkan/runtime/graph/ops/glsl/tan.yaml +13 -0
  2541. executorch/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.yaml +22 -0
  2542. executorch/backends/vulkan/runtime/graph/ops/glsl/transfer_texture.yaml +22 -0
  2543. executorch/backends/vulkan/runtime/graph/ops/glsl/transpose_cast_contig_to_vectorized.yaml +26 -0
  2544. executorch/backends/vulkan/runtime/graph/ops/glsl/transpose_cast_contig_to_vectorized_4x4.yaml +25 -0
  2545. executorch/backends/vulkan/runtime/graph/ops/glsl/unary_op.yaml +51 -0
  2546. executorch/backends/vulkan/runtime/graph/ops/glsl/unpack_4w4c_and_dequantize.yaml +23 -0
  2547. executorch/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.yaml +13 -0
  2548. executorch/backends/vulkan/runtime/graph/ops/glsl/upsample_2d.yaml +19 -0
  2549. executorch/backends/vulkan/runtime/graph/ops/glsl/var_buffer.yaml +16 -0
  2550. executorch/backends/vulkan/runtime/graph/ops/glsl/var_texture3d.yaml +16 -0
  2551. executorch/backends/vulkan/runtime/graph/ops/glsl/view_buffer.yaml +20 -0
  2552. executorch/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.yaml +26 -0
  2553. executorch/backends/vulkan/runtime/graph/ops/glsl/view_convert_texture.yaml +26 -0
  2554. executorch/backends/vulkan/runtime/graph/ops/glsl/view_texture.yaml +13 -0
  2555. executorch/backends/vulkan/runtime/graph/ops/glsl/where.yaml +12 -0
  2556. executorch/backends/vulkan/serialization/schema.fbs +162 -0
  2557. executorch/backends/vulkan/serialization/vulkan_graph_builder.py +490 -0
  2558. executorch/backends/vulkan/serialization/vulkan_graph_schema.py +165 -0
  2559. executorch/backends/vulkan/serialization/vulkan_graph_serialize.py +274 -0
  2560. executorch/backends/vulkan/test/custom_ops/glsl/add.yaml +29 -0
  2561. executorch/backends/vulkan/test/custom_ops/glsl/q4gsw_linear_gemv__w_4x8.yaml +40 -0
  2562. executorch/backends/vulkan/test/glsl/all_shaders.yaml +68 -0
  2563. executorch/backends/vulkan/test/glsl/dynamic_dispatch_test.yaml +7 -0
  2564. executorch/backends/vulkan/test/glsl/reference_matmul_common.yaml +9 -0
  2565. executorch/backends/vulkan/test/op_tests/cases.py +2252 -0
  2566. executorch/backends/vulkan/test/op_tests/generate_op_benchmarks.py +88 -0
  2567. executorch/backends/vulkan/test/op_tests/generate_op_correctness_tests.py +90 -0
  2568. executorch/backends/vulkan/test/op_tests/utils/aten_types.py +31 -0
  2569. executorch/backends/vulkan/test/op_tests/utils/gen_benchmark_vk.py +363 -0
  2570. executorch/backends/vulkan/test/op_tests/utils/gen_computegraph.py +836 -0
  2571. executorch/backends/vulkan/test/op_tests/utils/gen_correctness_base.py +427 -0
  2572. executorch/backends/vulkan/test/op_tests/utils/gen_correctness_vk.py +177 -0
  2573. executorch/backends/vulkan/test/op_tests/utils/test_suite.py +53 -0
  2574. executorch/backends/vulkan/test/test_serialization.py +135 -0
  2575. executorch/backends/vulkan/test/test_vulkan_compile_options.py +48 -0
  2576. executorch/backends/vulkan/test/test_vulkan_delegate.py +2716 -0
  2577. executorch/backends/vulkan/test/test_vulkan_delegate_header.py +104 -0
  2578. executorch/backends/vulkan/test/test_vulkan_passes.py +781 -0
  2579. executorch/backends/vulkan/test/test_vulkan_tensor_repr.py +991 -0
  2580. executorch/backends/vulkan/test/tester.py +91 -0
  2581. executorch/backends/vulkan/test/utils.py +1111 -0
  2582. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/apiconventions.py +23 -0
  2583. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/base_generator.py +1425 -0
  2584. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/cgenerator.py +575 -0
  2585. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/generator.py +1478 -0
  2586. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/parse_dependency.py +474 -0
  2587. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/reg.py +2115 -0
  2588. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/spec_tools/conventions.py +561 -0
  2589. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/spec_tools/util.py +58 -0
  2590. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/stripAPI.py +42 -0
  2591. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/vkconventions.py +330 -0
  2592. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/vulkan_object.py +690 -0
  2593. executorch/backends/vulkan/third-party/VulkanMemoryAllocator/tools/GpuMemDumpVis/GpuMemDumpVis.py +333 -0
  2594. executorch/backends/vulkan/third-party/volk/generate.py +212 -0
  2595. executorch/backends/vulkan/tools/gpuinfo/glsl/buf_bandwidth.yaml +18 -0
  2596. executorch/backends/vulkan/tools/gpuinfo/glsl/buf_cacheline_size.yaml +12 -0
  2597. executorch/backends/vulkan/tools/gpuinfo/glsl/reg_count.yaml +16 -0
  2598. executorch/backends/vulkan/tools/gpuinfo/glsl/tex_bandwidth.yaml +15 -0
  2599. executorch/backends/vulkan/tools/gpuinfo/glsl/tex_cacheline_concurr.yaml +14 -0
  2600. executorch/backends/vulkan/tools/gpuinfo/glsl/warp_size.yaml +16 -0
  2601. executorch/backends/vulkan/utils.py +2016 -0
  2602. executorch/backends/vulkan/vulkan_preprocess.py +256 -0
  2603. executorch/backends/webgpu/__init__.py +5 -0
  2604. executorch/backends/webgpu/runtime/ops/binary_op/binary_op.yaml +11 -0
  2605. executorch/backends/webgpu/runtime/ops/boolean_op/boolean_op.yaml +20 -0
  2606. executorch/backends/webgpu/runtime/ops/quantized_linear/q4gsw_linear_gemm_steel.yaml +22 -0
  2607. executorch/backends/webgpu/runtime/ops/rms_norm/rms_norm.yaml +12 -0
  2608. executorch/backends/webgpu/runtime/ops/sdpa/sdpa_compute_attn_weights.yaml +11 -0
  2609. executorch/backends/webgpu/runtime/ops/sdpa/sdpa_compute_out.yaml +11 -0
  2610. executorch/backends/webgpu/runtime/ops/sdpa_fd_decode/sdpa_fd_split.yaml +11 -0
  2611. executorch/backends/webgpu/runtime/ops/softmax/softmax.yaml +8 -0
  2612. executorch/backends/webgpu/runtime/ops/update_cache/update_cache.yaml +11 -0
  2613. executorch/backends/webgpu/scripts/gen_wgsl_headers.py +584 -0
  2614. executorch/backends/webgpu/test/conftest.py +28 -0
  2615. executorch/backends/webgpu/test/op_tests/__init__.py +5 -0
  2616. executorch/backends/webgpu/test/op_tests/cases.py +288 -0
  2617. executorch/backends/webgpu/test/op_tests/generate_op_tests.py +198 -0
  2618. executorch/backends/webgpu/test/op_tests/test_generator.py +111 -0
  2619. executorch/backends/webgpu/test/op_tests/test_schema.py +84 -0
  2620. executorch/backends/webgpu/test/op_tests/test_suite.py +98 -0
  2621. executorch/backends/webgpu/test/ops/__init__.py +0 -0
  2622. executorch/backends/webgpu/test/ops/dynamic_shape/test_dynamic_shape_export.py +459 -0
  2623. executorch/backends/webgpu/test/ops/index/__init__.py +0 -0
  2624. executorch/backends/webgpu/test/ops/index/test_index.py +109 -0
  2625. executorch/backends/webgpu/test/ops/test_adamw.py +142 -0
  2626. executorch/backends/webgpu/test/ops/test_add.py +116 -0
  2627. executorch/backends/webgpu/test/ops/test_bmm.py +104 -0
  2628. executorch/backends/webgpu/test/ops/test_cat.py +82 -0
  2629. executorch/backends/webgpu/test/ops/test_compare.py +109 -0
  2630. executorch/backends/webgpu/test/ops/test_dim_order.py +95 -0
  2631. executorch/backends/webgpu/test/ops/test_dispatch_order.py +119 -0
  2632. executorch/backends/webgpu/test/ops/test_div.py +94 -0
  2633. executorch/backends/webgpu/test/ops/test_embedding.py +103 -0
  2634. executorch/backends/webgpu/test/ops/test_embedding_q4gsw.py +161 -0
  2635. executorch/backends/webgpu/test/ops/test_expand_copy.py +120 -0
  2636. executorch/backends/webgpu/test/ops/test_fill.py +91 -0
  2637. executorch/backends/webgpu/test/ops/test_fused_ce.py +130 -0
  2638. executorch/backends/webgpu/test/ops/test_gather.py +123 -0
  2639. executorch/backends/webgpu/test/ops/test_linear.py +116 -0
  2640. executorch/backends/webgpu/test/ops/test_linear_dW.py +88 -0
  2641. executorch/backends/webgpu/test/ops/test_log_softmax.py +97 -0
  2642. executorch/backends/webgpu/test/ops/test_logical_not.py +116 -0
  2643. executorch/backends/webgpu/test/ops/test_matmul.py +67 -0
  2644. executorch/backends/webgpu/test/ops/test_mul.py +69 -0
  2645. executorch/backends/webgpu/test/ops/test_permute.py +76 -0
  2646. executorch/backends/webgpu/test/ops/test_prepack.py +142 -0
  2647. executorch/backends/webgpu/test/ops/test_q4gsw_requant.py +123 -0
  2648. executorch/backends/webgpu/test/ops/test_quantized_linear.py +193 -0
  2649. executorch/backends/webgpu/test/ops/test_quantized_linear_backward.py +157 -0
  2650. executorch/backends/webgpu/test/ops/test_reduce.py +127 -0
  2651. executorch/backends/webgpu/test/ops/test_rms_norm.py +192 -0
  2652. executorch/backends/webgpu/test/ops/test_rope.py +129 -0
  2653. executorch/backends/webgpu/test/ops/test_sdpa.py +533 -0
  2654. executorch/backends/webgpu/test/ops/test_select.py +67 -0
  2655. executorch/backends/webgpu/test/ops/test_sigmoid.py +80 -0
  2656. executorch/backends/webgpu/test/ops/test_slice.py +66 -0
  2657. executorch/backends/webgpu/test/ops/test_softmax.py +116 -0
  2658. executorch/backends/webgpu/test/ops/test_squeeze.py +75 -0
  2659. executorch/backends/webgpu/test/ops/test_sub.py +114 -0
  2660. executorch/backends/webgpu/test/ops/test_unsqueeze.py +75 -0
  2661. executorch/backends/webgpu/test/ops/test_update_cache.py +196 -0
  2662. executorch/backends/webgpu/test/ops/test_view_copy.py +66 -0
  2663. executorch/backends/webgpu/test/ops/test_where.py +86 -0
  2664. executorch/backends/webgpu/test/test_wgsl_codegen.py +473 -0
  2665. executorch/backends/webgpu/test/tester.py +91 -0
  2666. executorch/backends/xnnpack/__init__.py +28 -0
  2667. executorch/backends/xnnpack/_passes/__init__.py +113 -0
  2668. executorch/backends/xnnpack/_passes/channels_last_tagged_reshape_pass.py +570 -0
  2669. executorch/backends/xnnpack/_passes/conv1d_unsqueeze_pass.py +203 -0
  2670. executorch/backends/xnnpack/_passes/convert_to_linear.py +198 -0
  2671. executorch/backends/xnnpack/_passes/convert_to_sdpa.py +114 -0
  2672. executorch/backends/xnnpack/_passes/convert_to_upsample_bilinear2d.py +65 -0
  2673. executorch/backends/xnnpack/_passes/decompose_batch_norm.py +296 -0
  2674. executorch/backends/xnnpack/_passes/decompose_cat.py +109 -0
  2675. executorch/backends/xnnpack/_passes/fuse_activation_pass.py +87 -0
  2676. executorch/backends/xnnpack/_passes/fuse_batch_norm.py +279 -0
  2677. executorch/backends/xnnpack/_passes/insert_pad_qdq.py +88 -0
  2678. executorch/backends/xnnpack/_passes/prelu_reshape_pass.py +75 -0
  2679. executorch/backends/xnnpack/_passes/propagate_custom_meta_pass.py +44 -0
  2680. executorch/backends/xnnpack/_passes/remove_noop_expand_copy_pass.py +59 -0
  2681. executorch/backends/xnnpack/_passes/remove_redundant_copy_pass.py +158 -0
  2682. executorch/backends/xnnpack/_passes/xnnpack_pass.py +22 -0
  2683. executorch/backends/xnnpack/operators/__init__.py +61 -0
  2684. executorch/backends/xnnpack/operators/node_visitor.py +758 -0
  2685. executorch/backends/xnnpack/operators/op_abs.py +52 -0
  2686. executorch/backends/xnnpack/operators/op_add.py +77 -0
  2687. executorch/backends/xnnpack/operators/op_addmm.py +70 -0
  2688. executorch/backends/xnnpack/operators/op_avg_pooling2d.py +75 -0
  2689. executorch/backends/xnnpack/operators/op_bmm.py +54 -0
  2690. executorch/backends/xnnpack/operators/op_cat.py +121 -0
  2691. executorch/backends/xnnpack/operators/op_ceiling.py +52 -0
  2692. executorch/backends/xnnpack/operators/op_clamp.py +64 -0
  2693. executorch/backends/xnnpack/operators/op_clone.py +71 -0
  2694. executorch/backends/xnnpack/operators/op_conv2d.py +183 -0
  2695. executorch/backends/xnnpack/operators/op_cos.py +52 -0
  2696. executorch/backends/xnnpack/operators/op_div.py +56 -0
  2697. executorch/backends/xnnpack/operators/op_dynamic_dequantize_ops.py +114 -0
  2698. executorch/backends/xnnpack/operators/op_dynamic_quantize_ops.py +179 -0
  2699. executorch/backends/xnnpack/operators/op_elu.py +56 -0
  2700. executorch/backends/xnnpack/operators/op_exp.py +52 -0
  2701. executorch/backends/xnnpack/operators/op_floor.py +48 -0
  2702. executorch/backends/xnnpack/operators/op_gelu.py +52 -0
  2703. executorch/backends/xnnpack/operators/op_hardswish.py +52 -0
  2704. executorch/backends/xnnpack/operators/op_hardtanh.py +67 -0
  2705. executorch/backends/xnnpack/operators/op_leaky_relu.py +65 -0
  2706. executorch/backends/xnnpack/operators/op_linear.py +110 -0
  2707. executorch/backends/xnnpack/operators/op_log.py +52 -0
  2708. executorch/backends/xnnpack/operators/op_matrix_multiplication.py +66 -0
  2709. executorch/backends/xnnpack/operators/op_max_dim.py +206 -0
  2710. executorch/backends/xnnpack/operators/op_max_pool2d.py +136 -0
  2711. executorch/backends/xnnpack/operators/op_maximum.py +55 -0
  2712. executorch/backends/xnnpack/operators/op_mean_dim.py +87 -0
  2713. executorch/backends/xnnpack/operators/op_minimum.py +54 -0
  2714. executorch/backends/xnnpack/operators/op_multiply.py +77 -0
  2715. executorch/backends/xnnpack/operators/op_negate.py +51 -0
  2716. executorch/backends/xnnpack/operators/op_permute.py +87 -0
  2717. executorch/backends/xnnpack/operators/op_prelu.py +56 -0
  2718. executorch/backends/xnnpack/operators/op_quant_dequant.py +196 -0
  2719. executorch/backends/xnnpack/operators/op_relu.py +61 -0
  2720. executorch/backends/xnnpack/operators/op_rsqrt.py +52 -0
  2721. executorch/backends/xnnpack/operators/op_sigmoid.py +48 -0
  2722. executorch/backends/xnnpack/operators/op_sin.py +52 -0
  2723. executorch/backends/xnnpack/operators/op_skip_ops.py +89 -0
  2724. executorch/backends/xnnpack/operators/op_slice_copy.py +100 -0
  2725. executorch/backends/xnnpack/operators/op_softmax.py +58 -0
  2726. executorch/backends/xnnpack/operators/op_square.py +52 -0
  2727. executorch/backends/xnnpack/operators/op_square_root.py +52 -0
  2728. executorch/backends/xnnpack/operators/op_squeeze.py +147 -0
  2729. executorch/backends/xnnpack/operators/op_static_constant_pad.py +145 -0
  2730. executorch/backends/xnnpack/operators/op_static_resize_bilinear_2d.py +61 -0
  2731. executorch/backends/xnnpack/operators/op_sub.py +77 -0
  2732. executorch/backends/xnnpack/operators/op_tanh.py +52 -0
  2733. executorch/backends/xnnpack/operators/op_to_copy.py +91 -0
  2734. executorch/backends/xnnpack/operators/op_view_copy.py +96 -0
  2735. executorch/backends/xnnpack/operators/quant_params.py +412 -0
  2736. executorch/backends/xnnpack/partition/config/__init__.py +134 -0
  2737. executorch/backends/xnnpack/partition/config/gemm_configs.py +599 -0
  2738. executorch/backends/xnnpack/partition/config/generic_node_configs.py +783 -0
  2739. executorch/backends/xnnpack/partition/config/node_configs.py +139 -0
  2740. executorch/backends/xnnpack/partition/config/quant_affine_configs.py +56 -0
  2741. executorch/backends/xnnpack/partition/config/xnnpack_config.py +252 -0
  2742. executorch/backends/xnnpack/partition/graphs/bilinear_2d.py +55 -0
  2743. executorch/backends/xnnpack/partition/graphs/sdpa.py +86 -0
  2744. executorch/backends/xnnpack/partition/xnnpack_partitioner.py +160 -0
  2745. executorch/backends/xnnpack/quantizer/xnnpack_quantizer.py +462 -0
  2746. executorch/backends/xnnpack/quantizer/xnnpack_quantizer_utils.py +1172 -0
  2747. executorch/backends/xnnpack/recipes/__init__.py +19 -0
  2748. executorch/backends/xnnpack/recipes/xnnpack_recipe_provider.py +200 -0
  2749. executorch/backends/xnnpack/recipes/xnnpack_recipe_types.py +37 -0
  2750. executorch/backends/xnnpack/serialization/runtime_schema.fbs +389 -0
  2751. executorch/backends/xnnpack/serialization/schema.fbs +381 -0
  2752. executorch/backends/xnnpack/serialization/xnnpack_graph_schema.py +542 -0
  2753. executorch/backends/xnnpack/serialization/xnnpack_graph_serialize.py +393 -0
  2754. executorch/backends/xnnpack/test/__init__.py +0 -0
  2755. executorch/backends/xnnpack/test/fragments/test_dim_order.py +40 -0
  2756. executorch/backends/xnnpack/test/models/deeplab_v3.py +76 -0
  2757. executorch/backends/xnnpack/test/models/edsr.py +87 -0
  2758. executorch/backends/xnnpack/test/models/emformer_rnnt.py +146 -0
  2759. executorch/backends/xnnpack/test/models/inception_v3.py +117 -0
  2760. executorch/backends/xnnpack/test/models/inception_v4.py +95 -0
  2761. executorch/backends/xnnpack/test/models/llama2_et_example.py +49 -0
  2762. executorch/backends/xnnpack/test/models/mobilebert.py +72 -0
  2763. executorch/backends/xnnpack/test/models/mobilenet_v2.py +103 -0
  2764. executorch/backends/xnnpack/test/models/mobilenet_v3.py +84 -0
  2765. executorch/backends/xnnpack/test/models/resnet.py +90 -0
  2766. executorch/backends/xnnpack/test/models/torchvision_vit.py +105 -0
  2767. executorch/backends/xnnpack/test/models/very_big_model.py +45 -0
  2768. executorch/backends/xnnpack/test/models/w2l.py +60 -0
  2769. executorch/backends/xnnpack/test/ops/test_abs.py +87 -0
  2770. executorch/backends/xnnpack/test/ops/test_add.py +266 -0
  2771. executorch/backends/xnnpack/test/ops/test_avgpool2d.py +92 -0
  2772. executorch/backends/xnnpack/test/ops/test_batch_norm.py +380 -0
  2773. executorch/backends/xnnpack/test/ops/test_bilinear2d.py +152 -0
  2774. executorch/backends/xnnpack/test/ops/test_bmm.py +49 -0
  2775. executorch/backends/xnnpack/test/ops/test_cat.py +297 -0
  2776. executorch/backends/xnnpack/test/ops/test_ceil.py +63 -0
  2777. executorch/backends/xnnpack/test/ops/test_check_quant_params.py +107 -0
  2778. executorch/backends/xnnpack/test/ops/test_clamp.py +79 -0
  2779. executorch/backends/xnnpack/test/ops/test_clone.py +165 -0
  2780. executorch/backends/xnnpack/test/ops/test_conv1d.py +243 -0
  2781. executorch/backends/xnnpack/test/ops/test_conv2d.py +1025 -0
  2782. executorch/backends/xnnpack/test/ops/test_cos.py +87 -0
  2783. executorch/backends/xnnpack/test/ops/test_div.py +78 -0
  2784. executorch/backends/xnnpack/test/ops/test_elu.py +98 -0
  2785. executorch/backends/xnnpack/test/ops/test_exp.py +70 -0
  2786. executorch/backends/xnnpack/test/ops/test_floor.py +58 -0
  2787. executorch/backends/xnnpack/test/ops/test_gelu.py +91 -0
  2788. executorch/backends/xnnpack/test/ops/test_hardswish.py +70 -0
  2789. executorch/backends/xnnpack/test/ops/test_hardtanh.py +83 -0
  2790. executorch/backends/xnnpack/test/ops/test_leaky_relu.py +131 -0
  2791. executorch/backends/xnnpack/test/ops/test_linear.py +1059 -0
  2792. executorch/backends/xnnpack/test/ops/test_log.py +45 -0
  2793. executorch/backends/xnnpack/test/ops/test_lstm.py +67 -0
  2794. executorch/backends/xnnpack/test/ops/test_max_dim.py +66 -0
  2795. executorch/backends/xnnpack/test/ops/test_maximum.py +66 -0
  2796. executorch/backends/xnnpack/test/ops/test_maxpool2d.py +221 -0
  2797. executorch/backends/xnnpack/test/ops/test_mean_dim.py +128 -0
  2798. executorch/backends/xnnpack/test/ops/test_minimum.py +49 -0
  2799. executorch/backends/xnnpack/test/ops/test_multiply.py +149 -0
  2800. executorch/backends/xnnpack/test/ops/test_negate.py +58 -0
  2801. executorch/backends/xnnpack/test/ops/test_permute.py +134 -0
  2802. executorch/backends/xnnpack/test/ops/test_pow.py +62 -0
  2803. executorch/backends/xnnpack/test/ops/test_prelu.py +50 -0
  2804. executorch/backends/xnnpack/test/ops/test_quantize_per_tensor.py +70 -0
  2805. executorch/backends/xnnpack/test/ops/test_relu.py +38 -0
  2806. executorch/backends/xnnpack/test/ops/test_rsqrt.py +45 -0
  2807. executorch/backends/xnnpack/test/ops/test_sigmoid.py +45 -0
  2808. executorch/backends/xnnpack/test/ops/test_sin.py +87 -0
  2809. executorch/backends/xnnpack/test/ops/test_slice_copy.py +170 -0
  2810. executorch/backends/xnnpack/test/ops/test_softmax.py +73 -0
  2811. executorch/backends/xnnpack/test/ops/test_sqrt.py +45 -0
  2812. executorch/backends/xnnpack/test/ops/test_square.py +48 -0
  2813. executorch/backends/xnnpack/test/ops/test_static_constant_pad.py +294 -0
  2814. executorch/backends/xnnpack/test/ops/test_sub.py +178 -0
  2815. executorch/backends/xnnpack/test/ops/test_tanh.py +43 -0
  2816. executorch/backends/xnnpack/test/ops/test_to_copy.py +113 -0
  2817. executorch/backends/xnnpack/test/passes/__init__.py +0 -0
  2818. executorch/backends/xnnpack/test/passes/test_activation_fusion.py +301 -0
  2819. executorch/backends/xnnpack/test/passes/test_batch_norm_fusion.py +133 -0
  2820. executorch/backends/xnnpack/test/passes/test_channels_last_tagged_reshape.py +667 -0
  2821. executorch/backends/xnnpack/test/passes/test_convert_to_linear.py +43 -0
  2822. executorch/backends/xnnpack/test/passes/test_decompose_batch_norm.py +196 -0
  2823. executorch/backends/xnnpack/test/passes/test_decompose_cat_pass.py +112 -0
  2824. executorch/backends/xnnpack/test/passes/test_insert_pad_qdq.py +73 -0
  2825. executorch/backends/xnnpack/test/passes/test_propagate_custom_meta_pass.py +185 -0
  2826. executorch/backends/xnnpack/test/passes/test_remove_get_item_pass.py +99 -0
  2827. executorch/backends/xnnpack/test/passes/test_remove_noop_expand_copy_pass.py +75 -0
  2828. executorch/backends/xnnpack/test/passes/test_remove_redundant_copy_pass.py +182 -0
  2829. executorch/backends/xnnpack/test/quantizer/test_pt2e_quantization.py +818 -0
  2830. executorch/backends/xnnpack/test/quantizer/test_representation.py +308 -0
  2831. executorch/backends/xnnpack/test/quantizer/test_xnnpack_quantizer.py +1217 -0
  2832. executorch/backends/xnnpack/test/recipes/test_xnnpack_recipes.py +320 -0
  2833. executorch/backends/xnnpack/test/serialization/test_serialization.py +50 -0
  2834. executorch/backends/xnnpack/test/serialization/test_xnnheader.py +104 -0
  2835. executorch/backends/xnnpack/test/test_xnnpack_partitioner.py +163 -0
  2836. executorch/backends/xnnpack/test/test_xnnpack_utils.py +533 -0
  2837. executorch/backends/xnnpack/test/test_xnnpack_utils_classes.py +44 -0
  2838. executorch/backends/xnnpack/test/tester/__init__.py +29 -0
  2839. executorch/backends/xnnpack/test/tester/tester.py +134 -0
  2840. executorch/backends/xnnpack/third-party/FP16/configure.py +65 -0
  2841. executorch/backends/xnnpack/third-party/FP16/confu.yaml +8 -0
  2842. executorch/backends/xnnpack/third-party/FP16/include/fp16/__init__.py +0 -0
  2843. executorch/backends/xnnpack/third-party/FP16/include/fp16/avx.py +50 -0
  2844. executorch/backends/xnnpack/third-party/FP16/include/fp16/avx2.py +53 -0
  2845. executorch/backends/xnnpack/third-party/FP16/test/peachpy/stubs.py +38 -0
  2846. executorch/backends/xnnpack/third-party/FXdiv/configure.py +30 -0
  2847. executorch/backends/xnnpack/third-party/FXdiv/confu.yaml +6 -0
  2848. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/aarch32_template.py +361 -0
  2849. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/aarch64_template.py +347 -0
  2850. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/arm_template.py +78 -0
  2851. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512bf16_template.py +81 -0
  2852. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512f_template.py +393 -0
  2853. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512vnni_template.py +703 -0
  2854. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/base_architecture.py +276 -0
  2855. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/fma3_template.py +280 -0
  2856. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate.py +71 -0
  2857. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_bf16_f32_gemm_microkernels.py +39 -0
  2858. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_f32_gemm_microkernels.py +104 -0
  2859. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_gemm_microkernels_main.py +29 -0
  2860. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qd8_f32_qc4w_gemm_microkernels.py +81 -0
  2861. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qd8_f32_qc8w_gemm_microkernels.py +105 -0
  2862. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qs8_qc4w_gemm_microkernels.py +42 -0
  2863. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qs8_qc8w_gemm_microkernels.py +43 -0
  2864. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neondot_template.py +474 -0
  2865. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neonfma_template.py +248 -0
  2866. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neonmlal_aarch32_template.py +394 -0
  2867. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/x64_template.py +525 -0
  2868. executorch/backends/xnnpack/third-party/XNNPACK/scripts/check_config.py +152 -0
  2869. executorch/backends/xnnpack/third-party/XNNPACK/scripts/check_files_changed.py +53 -0
  2870. executorch/backends/xnnpack/third-party/XNNPACK/scripts/generate-build-identifier.py +106 -0
  2871. executorch/backends/xnnpack/third-party/XNNPACK/scripts/run-gn-tests.py +198 -0
  2872. executorch/backends/xnnpack/third-party/XNNPACK/scripts/sort-filenames.py +134 -0
  2873. executorch/backends/xnnpack/third-party/XNNPACK/test/bf16-f32-gemm-minmax.yaml +121 -0
  2874. executorch/backends/xnnpack/third-party/XNNPACK/test/bf16-gemm-minmax.yaml +106 -0
  2875. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-conv-hwc2chw.yaml +8 -0
  2876. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-dwconv2d-chw.yaml +88 -0
  2877. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-f32acc-gemm-minmax.yaml +42 -0
  2878. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-f32acc-igemm-minmax.yaml +42 -0
  2879. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-gemm-minmax.yaml +233 -0
  2880. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-ibilinear-chw.yaml +9 -0
  2881. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-ibilinear.yaml +12 -0
  2882. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-igemm-minmax.yaml +179 -0
  2883. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-raddstoreexpminusmax.yaml +27 -0
  2884. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-argmaxpool.yaml +15 -0
  2885. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-conv-hwc2chw.yaml +32 -0
  2886. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-dwconv2d-chw.yaml +712 -0
  2887. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm-minmax.yaml +1584 -0
  2888. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm-relu.yaml +123 -0
  2889. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm.yaml +127 -0
  2890. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ibilinear-chw.yaml +22 -0
  2891. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ibilinear.yaml +23 -0
  2892. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm-minmax.yaml +1021 -0
  2893. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm-relu.yaml +123 -0
  2894. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm.yaml +123 -0
  2895. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ppmm-minmax.yaml +122 -0
  2896. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc4w-gemm-minmax.yaml +277 -0
  2897. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm-minmax.yaml +757 -0
  2898. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm-relu.yaml +67 -0
  2899. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm.yaml +115 -0
  2900. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-raddexpminusmax.yaml +38 -0
  2901. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-raddstoreexpminusmax.yaml +88 -0
  2902. executorch/backends/xnnpack/third-party/XNNPACK/test/pf16-f16-igemm-minmax.yaml +24 -0
  2903. executorch/backends/xnnpack/third-party/XNNPACK/test/pf16-gemm-minmax.yaml +34 -0
  2904. executorch/backends/xnnpack/third-party/XNNPACK/test/pf32-gemm-minmax.yaml +34 -0
  2905. executorch/backends/xnnpack/third-party/XNNPACK/test/pqs8-qc8w-gemm-minmax.yaml +27 -0
  2906. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qb4w-gemm-minmax.yaml +260 -0
  2907. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc2w-gemm-minmax.yaml +148 -0
  2908. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc4w-gemm-minmax.yaml +678 -0
  2909. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc8w-gemm-minmax.yaml +412 -0
  2910. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc8w-igemm-minmax.yaml +360 -0
  2911. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qb4w-gemm-minmax.yaml +537 -0
  2912. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc2w-gemm-minmax.yaml +230 -0
  2913. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc4w-gemm-minmax.yaml +2218 -0
  2914. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc8w-gemm-minmax.yaml +1600 -0
  2915. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc8w-igemm-minmax.yaml +1035 -0
  2916. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qb4w-gemm-minmax.yaml +54 -0
  2917. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qc4w-gemm-minmax.yaml +66 -0
  2918. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qc8w-gemm-minmax.yaml +54 -0
  2919. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc2w-gemm-minmax-fp32.yaml +204 -0
  2920. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc4w-gemm-minmax-fp32.yaml +784 -0
  2921. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc8w-gemm-minmax-fp32.yaml +1997 -0
  2922. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc8w-igemm-minmax-fp32.yaml +1775 -0
  2923. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-gemm-minmax-fp32.yaml +584 -0
  2924. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-gemm-minmax-rndnu.yaml +148 -0
  2925. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-igemm-minmax-fp32.yaml +581 -0
  2926. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-igemm-minmax-rndnu.yaml +110 -0
  2927. executorch/backends/xnnpack/third-party/XNNPACK/test/s8-ibilinear.yaml +22 -0
  2928. executorch/backends/xnnpack/third-party/XNNPACK/test/u8-ibilinear.yaml +22 -0
  2929. executorch/backends/xnnpack/third-party/XNNPACK/test/x8-lut.yaml +47 -0
  2930. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-argmaxpool-test.py +692 -0
  2931. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-conv-hwc-test.py +543 -0
  2932. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-conv-hwc2chw-test.py +591 -0
  2933. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-dwconv-test.py +343 -0
  2934. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-dwconv2d-chw-test.py +463 -0
  2935. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-gemm-test.py +1152 -0
  2936. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-ibilinear-chw-test.py +236 -0
  2937. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-ibilinear-test.py +237 -0
  2938. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-lut-test.py +164 -0
  2939. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-raddexpminusmax-test.py +159 -0
  2940. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-raddstoreexpminusmax-test.py +210 -0
  2941. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-reduce-discontiguous-test.py +495 -0
  2942. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-reduce-test.py +289 -0
  2943. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-vbinary-test.py +179 -0
  2944. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-vunary-test.py +280 -0
  2945. executorch/backends/xnnpack/third-party/XNNPACK/tools/parse-microkernel-bench.py +189 -0
  2946. executorch/backends/xnnpack/third-party/XNNPACK/tools/primes.py +1015 -0
  2947. executorch/backends/xnnpack/third-party/XNNPACK/tools/update-microkernels.py +567 -0
  2948. executorch/backends/xnnpack/third-party/XNNPACK/tools/xngen.py +159 -0
  2949. executorch/backends/xnnpack/third-party/XNNPACK/tools/xnncommon.py +287 -0
  2950. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/binary/generator.py +13 -0
  2951. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/binary/kernels.py +81 -0
  2952. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/arm.py +210 -0
  2953. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/arm_bf16_bf16_fp32.py +66 -0
  2954. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/arm_bf16_bf16_fp32_k2.py +64 -0
  2955. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/arm_fp32.py +49 -0
  2956. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/arm_fp64.py +45 -0
  2957. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/arm_int8_int8_int32.py +152 -0
  2958. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/dot_base.py +538 -0
  2959. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/generator.py +113 -0
  2960. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86.py +316 -0
  2961. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_bf16_bf16_fp32.py +144 -0
  2962. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_bf16_bf16_fp32_k1.py +92 -0
  2963. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_fp16_fp16_fp32.py +88 -0
  2964. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_fp32.py +105 -0
  2965. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_fp32_k2.py +150 -0
  2966. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_fp32_k4.py +105 -0
  2967. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_fp64.py +105 -0
  2968. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_int8_int8_int32.py +138 -0
  2969. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_int8_int8_int32_k1.py +77 -0
  2970. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_int8_int8_int32_k16.py +88 -0
  2971. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_uint8_int8_int32.py +47 -0
  2972. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/dot/generator/x86_uint8_int8_int32_k16.py +85 -0
  2973. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/elementwise/arm.py +284 -0
  2974. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/elementwise/common_rules.py +89 -0
  2975. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/elementwise/compiler.py +1804 -0
  2976. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/elementwise/compiler_test.py +241 -0
  2977. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/elementwise/generator.py +54 -0
  2978. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/elementwise/x86.py +980 -0
  2979. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/ternary/convert.py +36 -0
  2980. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/ternary/generator.py +14 -0
  2981. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/ternary/kernels.py +49 -0
  2982. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/convert.py +32 -0
  2983. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/exp.py +121 -0
  2984. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/generator.py +18 -0
  2985. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/kernels.py +55 -0
  2986. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/sigmoid.py +49 -0
  2987. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/sine_cosine.py +86 -0
  2988. executorch/backends/xnnpack/third-party/XNNPACK/ynnpack/kernels/unary/tanh.py +65 -0
  2989. executorch/backends/xnnpack/third-party/cpuinfo/configure.py +110 -0
  2990. executorch/backends/xnnpack/third-party/cpuinfo/confu.yaml +9 -0
  2991. executorch/backends/xnnpack/third-party/cpuinfo/deps/clog/configure.py +26 -0
  2992. executorch/backends/xnnpack/third-party/cpuinfo/deps/clog/confu.yaml +5 -0
  2993. executorch/backends/xnnpack/third-party/cpuinfo/scripts/android-device-dump.py +400 -0
  2994. executorch/backends/xnnpack/third-party/cpuinfo/scripts/arm-linux-filesystem-dump.py +107 -0
  2995. executorch/backends/xnnpack/third-party/cpuinfo/scripts/parse-x86-cpuid-dump.py +47 -0
  2996. executorch/backends/xnnpack/third-party/generate-cpuinfo-wrappers.py +93 -0
  2997. executorch/backends/xnnpack/third-party/pthreadpool/configure.py +38 -0
  2998. executorch/backends/xnnpack/third-party/pthreadpool/confu.yaml +8 -0
  2999. executorch/backends/xnnpack/utils/configs.py +41 -0
  3000. executorch/backends/xnnpack/utils/quant_utils.py +301 -0
  3001. executorch/backends/xnnpack/utils/utils.py +214 -0
  3002. executorch/backends/xnnpack/utils/xnnpack_constants.py +37 -0
  3003. executorch/backends/xnnpack/xnnpack_preprocess.py +194 -0
  3004. executorch/codegen/__init__.py +0 -0
  3005. executorch/codegen/api/__init__.py +0 -0
  3006. executorch/codegen/api/custom_ops.py +151 -0
  3007. executorch/codegen/api/et_cpp.py +367 -0
  3008. executorch/codegen/api/types/__init__.py +5 -0
  3009. executorch/codegen/api/types/signatures.py +77 -0
  3010. executorch/codegen/api/types/types.py +76 -0
  3011. executorch/codegen/api/unboxing.py +232 -0
  3012. executorch/codegen/gen.py +1049 -0
  3013. executorch/codegen/model.py +220 -0
  3014. executorch/codegen/parse.py +158 -0
  3015. executorch/codegen/test/test_executorch_custom_ops.py +153 -0
  3016. executorch/codegen/test/test_executorch_gen.py +698 -0
  3017. executorch/codegen/test/test_executorch_signatures.py +65 -0
  3018. executorch/codegen/test/test_executorch_types.py +121 -0
  3019. executorch/codegen/test/test_executorch_unboxing.py +165 -0
  3020. executorch/codegen/test/test_selective_build.py +51 -0
  3021. executorch/codegen/tools/__init__.py +0 -0
  3022. executorch/codegen/tools/combine_prim_ops_headers.py +164 -0
  3023. executorch/codegen/tools/gen_all_oplist.py +252 -0
  3024. executorch/codegen/tools/gen_max_kernel_num.py +174 -0
  3025. executorch/codegen/tools/gen_oplist.py +369 -0
  3026. executorch/codegen/tools/gen_ops_def.py +96 -0
  3027. executorch/codegen/tools/gen_selected_op_variants.py +133 -0
  3028. executorch/codegen/tools/gen_selected_prim_ops.py +96 -0
  3029. executorch/codegen/tools/merge_yaml.py +125 -0
  3030. executorch/codegen/tools/selective_build.cp314-win_amd64.pyd +0 -0
  3031. executorch/codegen/tools/selective_build.pyi +23 -0
  3032. executorch/codegen/tools/test/test_gen_all_oplist.py +122 -0
  3033. executorch/codegen/tools/test/test_gen_max_kernel_num.py +226 -0
  3034. executorch/codegen/tools/test/test_gen_oplist.py +263 -0
  3035. executorch/codegen/tools/test/test_gen_oplist_real_model.py +50 -0
  3036. executorch/codegen/tools/test/test_gen_selected_op_variants.py +204 -0
  3037. executorch/codegen/tools/test/test_tools_selective_build.py +129 -0
  3038. executorch/codegen/tools/yaml_util.py +15 -0
  3039. executorch/data/bin/__init__.py +44 -0
  3040. executorch/data/bin/flatc.exe +0 -0
  3041. executorch/data/lib/aoti_cuda_shims.lib +0 -0
  3042. executorch/devtools/__init__.py +19 -0
  3043. executorch/devtools/backend_debug/__init__.py +13 -0
  3044. executorch/devtools/backend_debug/delegation_info.py +183 -0
  3045. executorch/devtools/backend_debug/tests/test_delegation_info.py +79 -0
  3046. executorch/devtools/bundled_program/config.py +120 -0
  3047. executorch/devtools/bundled_program/core.py +395 -0
  3048. executorch/devtools/bundled_program/schema/__init__.py +9 -0
  3049. executorch/devtools/bundled_program/schema/bundled_program_schema.fbs +97 -0
  3050. executorch/devtools/bundled_program/schema/bundled_program_schema.py +99 -0
  3051. executorch/devtools/bundled_program/schema/scalar_type.fbs +43 -0
  3052. executorch/devtools/bundled_program/serialize/__init__.py +298 -0
  3053. executorch/devtools/bundled_program/serialize/bundled_program_schema.fbs +97 -0
  3054. executorch/devtools/bundled_program/serialize/generate_bundled_program.py +174 -0
  3055. executorch/devtools/bundled_program/serialize/generated/__init__.py +9 -0
  3056. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/Bool.py +92 -0
  3057. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/BundledMethodTestCase.py +194 -0
  3058. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/BundledMethodTestSuite.py +158 -0
  3059. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/BundledProgram.py +212 -0
  3060. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/Double.py +92 -0
  3061. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/Int.py +92 -0
  3062. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/Tensor.py +267 -0
  3063. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/Value.py +125 -0
  3064. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/ValueUnion.py +29 -0
  3065. executorch/devtools/bundled_program/serialize/generated/bundled_program_flatbuffer/__init__.py +23 -0
  3066. executorch/devtools/bundled_program/serialize/generated/executorch_flatbuffer/ScalarType.py +28 -0
  3067. executorch/devtools/bundled_program/serialize/generated/executorch_flatbuffer/__init__.py +7 -0
  3068. executorch/devtools/bundled_program/serialize/scalar_type.fbs +43 -0
  3069. executorch/devtools/bundled_program/serialize/test/test_serialize.py +37 -0
  3070. executorch/devtools/bundled_program/test/test_bundle_data.py +168 -0
  3071. executorch/devtools/bundled_program/test/test_config.py +131 -0
  3072. executorch/devtools/bundled_program/test/test_end2end.py +64 -0
  3073. executorch/devtools/bundled_program/util/test_util.py +289 -0
  3074. executorch/devtools/bundled_program/version.py +13 -0
  3075. executorch/devtools/debug_format/base_schema.py +61 -0
  3076. executorch/devtools/debug_format/et_schema.py +432 -0
  3077. executorch/devtools/etdump/etdump_schema_flatcc.fbs +181 -0
  3078. executorch/devtools/etdump/scalar_type.fbs +43 -0
  3079. executorch/devtools/etdump/schema_flatcc.py +154 -0
  3080. executorch/devtools/etdump/serialize.py +108 -0
  3081. executorch/devtools/etdump/tests/serialize_test.py +142 -0
  3082. executorch/devtools/etrecord/__init__.py +13 -0
  3083. executorch/devtools/etrecord/_etrecord.py +929 -0
  3084. executorch/devtools/etrecord/tests/etrecord_test.py +1936 -0
  3085. executorch/devtools/inspector/__init__.py +31 -0
  3086. executorch/devtools/inspector/_inspector.py +1712 -0
  3087. executorch/devtools/inspector/_inspector_utils.py +1252 -0
  3088. executorch/devtools/inspector/_intermediate_output_capturer.py +62 -0
  3089. executorch/devtools/inspector/inspector_cli.py +76 -0
  3090. executorch/devtools/inspector/numerical_comparator/__init__.py +35 -0
  3091. executorch/devtools/inspector/numerical_comparator/l1_numerical_comparator.py +35 -0
  3092. executorch/devtools/inspector/numerical_comparator/mse_numerical_comparator.py +37 -0
  3093. executorch/devtools/inspector/numerical_comparator/numerical_comparator_base.py +266 -0
  3094. executorch/devtools/inspector/numerical_comparator/snr_numerical_comparator.py +47 -0
  3095. executorch/devtools/inspector/tests/event_blocks_test.py +657 -0
  3096. executorch/devtools/inspector/tests/inspector_test.py +2279 -0
  3097. executorch/devtools/inspector/tests/inspector_test_utils.py +189 -0
  3098. executorch/devtools/inspector/tests/inspector_utils_test.py +877 -0
  3099. executorch/devtools/inspector/tests/intermediate_output_capturer_test.py +101 -0
  3100. executorch/devtools/inspector/tests/l1_comparator_test.py +49 -0
  3101. executorch/devtools/inspector/tests/mse_comparator_test.py +49 -0
  3102. executorch/devtools/inspector/tests/snr_comparator_test.py +52 -0
  3103. executorch/devtools/inspector/vgf_neural_statistics.py +124 -0
  3104. executorch/devtools/intermediate_output_tap/__init__.py +93 -0
  3105. executorch/devtools/intermediate_output_tap/_convenience.py +191 -0
  3106. executorch/devtools/intermediate_output_tap/_reducers.py +319 -0
  3107. executorch/devtools/intermediate_output_tap/_selectors.py +231 -0
  3108. executorch/devtools/intermediate_output_tap/_spec.py +55 -0
  3109. executorch/devtools/intermediate_output_tap/_strip_pass.py +92 -0
  3110. executorch/devtools/intermediate_output_tap/_tap_pass.py +278 -0
  3111. executorch/devtools/intermediate_output_tap/custom_ops_lib.py +24 -0
  3112. executorch/devtools/intermediate_output_tap/tests/test_coreml_e2e.py +284 -0
  3113. executorch/devtools/intermediate_output_tap/tests/test_reducers.py +131 -0
  3114. executorch/devtools/intermediate_output_tap/tests/test_selectors.py +198 -0
  3115. executorch/devtools/intermediate_output_tap/tests/test_strip_pass.py +130 -0
  3116. executorch/devtools/intermediate_output_tap/tests/test_tap_pass.py +239 -0
  3117. executorch/devtools/intermediate_output_tap/tests/test_xnnpack_e2e.py +482 -0
  3118. executorch/devtools/pte_tool/diff_pte.py +895 -0
  3119. executorch/devtools/pte_tool/tests/test_diff_pte.py +341 -0
  3120. executorch/devtools/scripts/generate_profiling_csv.py +62 -0
  3121. executorch/devtools/size_analysis_tool/size_analysis_tool.py +194 -0
  3122. executorch/devtools/size_analysis_tool/size_analysis_tool_test.py +107 -0
  3123. executorch/devtools/visualization/__init__.py +13 -0
  3124. executorch/devtools/visualization/visualization_utils.py +305 -0
  3125. executorch/devtools/visualization/visualization_utils_test.py +177 -0
  3126. executorch/examples/apple/coreml/llama/export.py +224 -0
  3127. executorch/examples/apple/coreml/llama/export_static_llm_coreml.py +717 -0
  3128. executorch/examples/apple/coreml/llama/llama_transformer.py +899 -0
  3129. executorch/examples/apple/coreml/llama/run.py +206 -0
  3130. executorch/examples/apple/coreml/llama/run_lookahead.py +284 -0
  3131. executorch/examples/apple/coreml/llama/run_static_llm.py +293 -0
  3132. executorch/examples/apple/coreml/llama/run_static_llm_multifunction.py +325 -0
  3133. executorch/examples/apple/coreml/llama/test.py +48 -0
  3134. executorch/examples/apple/coreml/llama/utils.py +334 -0
  3135. executorch/examples/apple/coreml/scripts/coreml_compute_plan.py +236 -0
  3136. executorch/examples/apple/coreml/scripts/debugger_cli.py +181 -0
  3137. executorch/examples/apple/coreml/scripts/export.py +280 -0
  3138. executorch/examples/apple/coreml/scripts/extract_coreml_models.py +143 -0
  3139. executorch/examples/apple/coreml/scripts/inspector_cli.py +64 -0
  3140. executorch/examples/apple/coreml/scripts/inspector_utils.py +425 -0
  3141. executorch/examples/apple/coreml/scripts/test_coreml_compute_plan.py +161 -0
  3142. executorch/examples/apple/mps/scripts/bench_utils.py +108 -0
  3143. executorch/examples/apple/mps/scripts/mps_example.py +239 -0
  3144. executorch/examples/cuda/scripts/__init__.py +0 -0
  3145. executorch/examples/cuda/scripts/export.py +107 -0
  3146. executorch/examples/llm_pte_finetuning/__init__.py +19 -0
  3147. executorch/examples/llm_pte_finetuning/llama3_config.yaml +96 -0
  3148. executorch/examples/llm_pte_finetuning/model_exporter.py +87 -0
  3149. executorch/examples/llm_pte_finetuning/model_loading_lib.py +143 -0
  3150. executorch/examples/llm_pte_finetuning/phi3_alpaca_code_config.yaml +44 -0
  3151. executorch/examples/llm_pte_finetuning/phi3_config.yaml +42 -0
  3152. executorch/examples/llm_pte_finetuning/qwen_05b_config.yaml +43 -0
  3153. executorch/examples/llm_pte_finetuning/runner.py +135 -0
  3154. executorch/examples/llm_pte_finetuning/training_lib.py +157 -0
  3155. executorch/examples/llm_server/__init__.py +5 -0
  3156. executorch/examples/llm_server/conformance/test_openai_contract.py +207 -0
  3157. executorch/examples/llm_server/python/__init__.py +7 -0
  3158. executorch/examples/llm_server/python/chat_template.py +338 -0
  3159. executorch/examples/llm_server/python/errors.py +105 -0
  3160. executorch/examples/llm_server/python/openai_transcript.py +337 -0
  3161. executorch/examples/llm_server/python/protocol.py +153 -0
  3162. executorch/examples/llm_server/python/server.py +244 -0
  3163. executorch/examples/llm_server/python/serving_chat.py +664 -0
  3164. executorch/examples/llm_server/python/session_runtime.py +250 -0
  3165. executorch/examples/llm_server/python/tests/conftest.py +143 -0
  3166. executorch/examples/llm_server/python/tests/test_contract.py +457 -0
  3167. executorch/examples/llm_server/python/tests/test_gemma_tool_parser.py +131 -0
  3168. executorch/examples/llm_server/python/tests/test_hermes_tool_parser.py +100 -0
  3169. executorch/examples/llm_server/python/tests/test_qwen_tool_parser.py +218 -0
  3170. executorch/examples/llm_server/python/tests/test_sampling_params.py +269 -0
  3171. executorch/examples/llm_server/python/tests/test_session_runtime.py +244 -0
  3172. executorch/examples/llm_server/python/tests/test_sessions.py +426 -0
  3173. executorch/examples/llm_server/python/tests/test_streaming_stops.py +274 -0
  3174. executorch/examples/llm_server/python/tests/test_template.py +382 -0
  3175. executorch/examples/llm_server/python/tests/test_tool_calls.py +221 -0
  3176. executorch/examples/llm_server/python/tests/test_warm_resume_scaffold.py +745 -0
  3177. executorch/examples/llm_server/python/tests/test_worker_client.py +261 -0
  3178. executorch/examples/llm_server/python/tool_parsers/__init__.py +29 -0
  3179. executorch/examples/llm_server/python/tool_parsers/gemma.py +258 -0
  3180. executorch/examples/llm_server/python/tool_parsers/hermes.py +119 -0
  3181. executorch/examples/llm_server/python/tool_parsers/qwen.py +193 -0
  3182. executorch/examples/llm_server/python/tool_parsers/types.py +33 -0
  3183. executorch/examples/llm_server/python/worker_client.py +275 -0
  3184. executorch/examples/models/__init__.py +120 -0
  3185. executorch/examples/models/checkpoint.py +142 -0
  3186. executorch/examples/models/codegen/__init__.py +16 -0
  3187. executorch/examples/models/codegen/convert_weight.py +93 -0
  3188. executorch/examples/models/deeplab_v3/__init__.py +9 -0
  3189. executorch/examples/models/deeplab_v3/model.py +50 -0
  3190. executorch/examples/models/deepseek-r1-distill-llama-8B/config/deepseek_xnnpack_q8da4w.yaml +16 -0
  3191. executorch/examples/models/deit_tiny/__init__.py +8 -0
  3192. executorch/examples/models/deit_tiny/model.py +36 -0
  3193. executorch/examples/models/dinov2/__init__.py +0 -0
  3194. executorch/examples/models/dinov2/export_dinov2.py +187 -0
  3195. executorch/examples/models/edsr/__init__.py +11 -0
  3196. executorch/examples/models/edsr/model.py +28 -0
  3197. executorch/examples/models/efficient_sam/__init__.py +11 -0
  3198. executorch/examples/models/efficient_sam/efficient_sam_core/build_efficient_sam.py +25 -0
  3199. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam.py +316 -0
  3200. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam_decoder.py +373 -0
  3201. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam_encoder.py +261 -0
  3202. executorch/examples/models/efficient_sam/efficient_sam_core/mlp.py +31 -0
  3203. executorch/examples/models/efficient_sam/efficient_sam_core/two_way_transformer.py +268 -0
  3204. executorch/examples/models/efficient_sam/model.py +34 -0
  3205. executorch/examples/models/emformer_rnnt/__init__.py +17 -0
  3206. executorch/examples/models/emformer_rnnt/model.py +132 -0
  3207. executorch/examples/models/gemma/__init__.py +16 -0
  3208. executorch/examples/models/gemma/convert_weights.py +104 -0
  3209. executorch/examples/models/gemma2/__init__.py +19 -0
  3210. executorch/examples/models/gemma2/convert_weights.py +113 -0
  3211. executorch/examples/models/gemma3/__init__.py +16 -0
  3212. executorch/examples/models/gemma3/convert_weights.py +110 -0
  3213. executorch/examples/models/gemma4/__init__.py +13 -0
  3214. executorch/examples/models/gemma4/export_gemma4.py +884 -0
  3215. executorch/examples/models/gemma4/image_utils.py +155 -0
  3216. executorch/examples/models/gemma4/quant_utils.py +93 -0
  3217. executorch/examples/models/gemma4/run_gemma4.py +365 -0
  3218. executorch/examples/models/gemma4/speech_transform.py +149 -0
  3219. executorch/examples/models/gemma4/tests/test_speech_transform.py +107 -0
  3220. executorch/examples/models/gemma4/text_decoder/__init__.py +17 -0
  3221. executorch/examples/models/gemma4/text_decoder/convert_weights.py +314 -0
  3222. executorch/examples/models/gemma4/text_decoder/gemma4_attention.py +629 -0
  3223. executorch/examples/models/gemma4/text_decoder/gemma4_config.py +299 -0
  3224. executorch/examples/models/gemma4/text_decoder/gemma4_cross_decoder.py +83 -0
  3225. executorch/examples/models/gemma4/text_decoder/gemma4_decoder_layer.py +185 -0
  3226. executorch/examples/models/gemma4/text_decoder/gemma4_model.py +207 -0
  3227. executorch/examples/models/gemma4/text_decoder/gemma4_self_decoder.py +229 -0
  3228. executorch/examples/models/gemma4/text_decoder/gemma4_transformer.py +171 -0
  3229. executorch/examples/models/gemma4_31b/__init__.py +5 -0
  3230. executorch/examples/models/gemma4_31b/cuda_packers.py +160 -0
  3231. executorch/examples/models/gemma4_31b/cuda_source_transformations.py +353 -0
  3232. executorch/examples/models/gemma4_31b/export.py +612 -0
  3233. executorch/examples/models/gemma4_31b/gguf_loader.py +60 -0
  3234. executorch/examples/models/gemma4_31b/inference.py +263 -0
  3235. executorch/examples/models/gemma4_31b/mlx_source_transformations.py +234 -0
  3236. executorch/examples/models/gemma4_31b/model.py +645 -0
  3237. executorch/examples/models/gemma4_31b/quant/quantize.py +340 -0
  3238. executorch/examples/models/gemma4_31b/quant/recipe.py +62 -0
  3239. executorch/examples/models/gemma4_31b/quant/tests/test_pack_cuda.py +492 -0
  3240. executorch/examples/models/gemma4_31b/quant/tests/test_quantize.py +230 -0
  3241. executorch/examples/models/gemma4_31b/quant/tests/test_recipe.py +167 -0
  3242. executorch/examples/models/gemma4_31b/quant/tests/test_safetensors_roundtrip.py +143 -0
  3243. executorch/examples/models/gemma4_31b/quantize_and_save.py +158 -0
  3244. executorch/examples/models/gemma4_31b/sampler.py +37 -0
  3245. executorch/examples/models/gemma4_31b/serve.py +214 -0
  3246. executorch/examples/models/gemma4_31b/test_ondevice_serving.py +62 -0
  3247. executorch/examples/models/gemma4_31b/test_serve.py +184 -0
  3248. executorch/examples/models/gemma4_31b/tests/test_cuda_packers.py +501 -0
  3249. executorch/examples/models/gemma4_31b/tests/test_cuda_pipeline.py +334 -0
  3250. executorch/examples/models/gemma4_31b/tests/test_mlx_pipeline.py +582 -0
  3251. executorch/examples/models/gemma4_31b/tests/test_pipeline.py +504 -0
  3252. executorch/examples/models/gemma4_31b/tests/test_recipe.py +61 -0
  3253. executorch/examples/models/glm/__init__.py +16 -0
  3254. executorch/examples/models/glm/convert_weights.py +79 -0
  3255. executorch/examples/models/granite/__init__.py +16 -0
  3256. executorch/examples/models/granite/convert_weights.py +106 -0
  3257. executorch/examples/models/granite_speech/__init__.py +16 -0
  3258. executorch/examples/models/granite_speech/convert_weights.py +111 -0
  3259. executorch/examples/models/inception_v3/__init__.py +11 -0
  3260. executorch/examples/models/inception_v3/model.py +28 -0
  3261. executorch/examples/models/inception_v4/__init__.py +11 -0
  3262. executorch/examples/models/inception_v4/model.py +26 -0
  3263. executorch/examples/models/internvl3/__init__.py +16 -0
  3264. executorch/examples/models/internvl3/convert_weights.py +82 -0
  3265. executorch/examples/models/lfm2/__init__.py +10 -0
  3266. executorch/examples/models/lfm2/config/lfm2_mlx_4w.yaml +15 -0
  3267. executorch/examples/models/lfm2/config/lfm2_xnnpack_fp32.yaml +12 -0
  3268. executorch/examples/models/lfm2/config/lfm2_xnnpack_q8da4w.yaml +15 -0
  3269. executorch/examples/models/lfm2/convert_weights.py +92 -0
  3270. executorch/examples/models/lfm2/short_conv.py +112 -0
  3271. executorch/examples/models/lfm2/test_lfm2_5_mlx.py +102 -0
  3272. executorch/examples/models/llama/__init__.py +20 -0
  3273. executorch/examples/models/llama/attention.py +882 -0
  3274. executorch/examples/models/llama/config/llama_attention_sink.yaml +30 -0
  3275. executorch/examples/models/llama/config/llama_bf16.yaml +7 -0
  3276. executorch/examples/models/llama/config/llama_q8da4w.yaml +11 -0
  3277. executorch/examples/models/llama/config/llama_xnnpack.yaml +17 -0
  3278. executorch/examples/models/llama/config/llama_xnnpack_qat.yaml +23 -0
  3279. executorch/examples/models/llama/config/llama_xnnpack_spinquant.yaml +22 -0
  3280. executorch/examples/models/llama/config/test_llm_config.py +106 -0
  3281. executorch/examples/models/llama/convert_weights.py +61 -0
  3282. executorch/examples/models/llama/eval_llama.py +38 -0
  3283. executorch/examples/models/llama/eval_llama_lib.py +441 -0
  3284. executorch/examples/models/llama/evaluate/__init__.py +11 -0
  3285. executorch/examples/models/llama/evaluate/eager_eval.py +89 -0
  3286. executorch/examples/models/llama/experimental/__init__.py +0 -0
  3287. executorch/examples/models/llama/experimental/generate.py +190 -0
  3288. executorch/examples/models/llama/experimental/load_gguf_q4_0.py +184 -0
  3289. executorch/examples/models/llama/experimental/subclass.py +413 -0
  3290. executorch/examples/models/llama/experimental/test_subclass.py +28 -0
  3291. executorch/examples/models/llama/export_llama.py +56 -0
  3292. executorch/examples/models/llama/export_llama_lib.py +1932 -0
  3293. executorch/examples/models/llama/fairseq2.py +63 -0
  3294. executorch/examples/models/llama/feed_forward.py +72 -0
  3295. executorch/examples/models/llama/hf_download.py +50 -0
  3296. executorch/examples/models/llama/install_requirement_helper.py +29 -0
  3297. executorch/examples/models/llama/llama_transformer.py +523 -0
  3298. executorch/examples/models/llama/lora.py +80 -0
  3299. executorch/examples/models/llama/model.py +396 -0
  3300. executorch/examples/models/llama/model_args.py +234 -0
  3301. executorch/examples/models/llama/norm.py +172 -0
  3302. executorch/examples/models/llama/rope.py +402 -0
  3303. executorch/examples/models/llama/runner/eager.py +142 -0
  3304. executorch/examples/models/llama/runner/generation.py +234 -0
  3305. executorch/examples/models/llama/runner/native.py +145 -0
  3306. executorch/examples/models/llama/source_transformation/__init__.py +0 -0
  3307. executorch/examples/models/llama/source_transformation/apply_spin_quant_r1_r2.py +202 -0
  3308. executorch/examples/models/llama/source_transformation/attention.py +223 -0
  3309. executorch/examples/models/llama/source_transformation/attention_sink.py +378 -0
  3310. executorch/examples/models/llama/source_transformation/custom_kv_cache.py +714 -0
  3311. executorch/examples/models/llama/source_transformation/lora.py +153 -0
  3312. executorch/examples/models/llama/source_transformation/pre_quantization.py +340 -0
  3313. executorch/examples/models/llama/source_transformation/prune_vocab.py +119 -0
  3314. executorch/examples/models/llama/source_transformation/quantize.py +904 -0
  3315. executorch/examples/models/llama/source_transformation/rms_norm.py +23 -0
  3316. executorch/examples/models/llama/source_transformation/rope.py +39 -0
  3317. executorch/examples/models/llama/source_transformation/sdpa.py +550 -0
  3318. executorch/examples/models/llama/source_transformation/spin_quant.py +88 -0
  3319. executorch/examples/models/llama/source_transformation/test_attention_sink.py +462 -0
  3320. executorch/examples/models/llama/source_transformation/test_quantized_kv_cache.py +126 -0
  3321. executorch/examples/models/llama/source_transformation/test_quantized_sdpa.py +167 -0
  3322. executorch/examples/models/llama/source_transformation/test_sdpa_with_quantized_kv_cache.py +97 -0
  3323. executorch/examples/models/llama/static_attention.py +1601 -0
  3324. executorch/examples/models/llama/tests/test_export_llama_lib.py +158 -0
  3325. executorch/examples/models/llama/tests/test_pre_quantization_transforms.py +188 -0
  3326. executorch/examples/models/llama/tests/test_qwen3_5_attention.py +128 -0
  3327. executorch/examples/models/llama/tests/test_replace_kv_cache.py +158 -0
  3328. executorch/examples/models/llama/tests/test_ring_attention.py +359 -0
  3329. executorch/examples/models/llama/tests/test_ring_kv_cache.py +492 -0
  3330. executorch/examples/models/llama/tests/test_simple_sdpa.py +66 -0
  3331. executorch/examples/models/llama/tests/test_static_attention.py +624 -0
  3332. executorch/examples/models/llama/tests/test_yoco_attention.py +708 -0
  3333. executorch/examples/models/llama/tokenizer/tiktoken.py +224 -0
  3334. executorch/examples/models/llama3_2_vision/__init__.py +14 -0
  3335. executorch/examples/models/llama3_2_vision/preprocess/__init__.py +0 -0
  3336. executorch/examples/models/llama3_2_vision/preprocess/export_preprocess.py +45 -0
  3337. executorch/examples/models/llama3_2_vision/preprocess/model.py +72 -0
  3338. executorch/examples/models/llama3_2_vision/preprocess/test_preprocess.py +338 -0
  3339. executorch/examples/models/llama3_2_vision/runner/eager.py +59 -0
  3340. executorch/examples/models/llama3_2_vision/runner/exported.py +94 -0
  3341. executorch/examples/models/llama3_2_vision/runner/generation.py +102 -0
  3342. executorch/examples/models/llama3_2_vision/runner/native.py +143 -0
  3343. executorch/examples/models/llama3_2_vision/text_decoder/model.py +199 -0
  3344. executorch/examples/models/llama3_2_vision/text_decoder/test/__init__.py +0 -0
  3345. executorch/examples/models/llama3_2_vision/text_decoder/test/test_text_decoder.py +88 -0
  3346. executorch/examples/models/llama3_2_vision/vision_encoder/__init__.py +12 -0
  3347. executorch/examples/models/llama3_2_vision/vision_encoder/model.py +96 -0
  3348. executorch/examples/models/llama3_2_vision/vision_encoder/test/__init__.py +0 -0
  3349. executorch/examples/models/llama3_2_vision/vision_encoder/test/test_vision_encoder.py +46 -0
  3350. executorch/examples/models/llava/__init__.py +11 -0
  3351. executorch/examples/models/llava/export_llava.py +365 -0
  3352. executorch/examples/models/llava/model.py +415 -0
  3353. executorch/examples/models/llava/test/test_llava.py +151 -0
  3354. executorch/examples/models/llava/test/test_pte.py +105 -0
  3355. executorch/examples/models/lstm/__init__.py +12 -0
  3356. executorch/examples/models/lstm/model.py +31 -0
  3357. executorch/examples/models/mlperf_tiny/__init__.py +20 -0
  3358. executorch/examples/models/mlperf_tiny/deep_autoencoder.py +40 -0
  3359. executorch/examples/models/mlperf_tiny/ds_cnn.py +86 -0
  3360. executorch/examples/models/mlperf_tiny/mobilenet_v1_025.py +94 -0
  3361. executorch/examples/models/mlperf_tiny/resnet8.py +93 -0
  3362. executorch/examples/models/mobilebert/__init__.py +11 -0
  3363. executorch/examples/models/mobilebert/model.py +31 -0
  3364. executorch/examples/models/mobilenet_v2/__init__.py +12 -0
  3365. executorch/examples/models/mobilenet_v2/model.py +41 -0
  3366. executorch/examples/models/mobilenet_v3/__init__.py +11 -0
  3367. executorch/examples/models/mobilenet_v3/model.py +30 -0
  3368. executorch/examples/models/model_base.py +47 -0
  3369. executorch/examples/models/model_factory.py +60 -0
  3370. executorch/examples/models/moshi/mimi/test_mimi.py +245 -0
  3371. executorch/examples/models/parakeet/export_parakeet_tdt.py +826 -0
  3372. executorch/examples/models/parakeet/quantize.py +8 -0
  3373. executorch/examples/models/phi-3-mini/__init__.py +11 -0
  3374. executorch/examples/models/phi-3-mini-lora/export_model.py +144 -0
  3375. executorch/examples/models/phi_4_mini/__init__.py +16 -0
  3376. executorch/examples/models/phi_4_mini/config/phi_4_mini_xnnpack.yaml +12 -0
  3377. executorch/examples/models/phi_4_mini/convert_weights.py +153 -0
  3378. executorch/examples/models/qwen2_5/__init__.py +16 -0
  3379. executorch/examples/models/qwen2_5/config/qwen2_5_coder_xnnpack_q8da4w.yaml +20 -0
  3380. executorch/examples/models/qwen2_5/config/qwen2_5_xnnpack_q8da4w.yaml +14 -0
  3381. executorch/examples/models/qwen2_5/convert_weights.py +105 -0
  3382. executorch/examples/models/qwen3/__init__.py +16 -0
  3383. executorch/examples/models/qwen3/config/qwen3_multimethod.yaml +34 -0
  3384. executorch/examples/models/qwen3/config/qwen3_xnnpack_lora.yaml +21 -0
  3385. executorch/examples/models/qwen3/config/qwen3_xnnpack_q8da4w.yaml +20 -0
  3386. executorch/examples/models/qwen3/convert_weights.py +142 -0
  3387. executorch/examples/models/qwen3_5/__init__.py +19 -0
  3388. executorch/examples/models/qwen3_5/config/qwen3_5_xnnpack_fp32.yaml +21 -0
  3389. executorch/examples/models/qwen3_5/convert_weights.py +165 -0
  3390. executorch/examples/models/qwen3_5/tests/__init__.py +2 -0
  3391. executorch/examples/models/qwen3_5/tests/test_convert_weights.py +108 -0
  3392. executorch/examples/models/qwen3_5_moe/__init__.py +0 -0
  3393. executorch/examples/models/qwen3_5_moe/export.py +1383 -0
  3394. executorch/examples/models/qwen3_5_moe/inference.py +196 -0
  3395. executorch/examples/models/qwen3_5_moe/metal_source_transformations.py +336 -0
  3396. executorch/examples/models/qwen3_5_moe/mlx_source_transformations.py +371 -0
  3397. executorch/examples/models/qwen3_5_moe/model.py +955 -0
  3398. executorch/examples/models/qwen3_5_moe/quantize_and_save.py +570 -0
  3399. executorch/examples/models/qwen3_5_moe/run.py +329 -0
  3400. executorch/examples/models/qwen3_5_moe/sampler.py +60 -0
  3401. executorch/examples/models/qwen3_5_moe/serve.py +182 -0
  3402. executorch/examples/models/qwen3_5_moe/test_chunked_prefill.py +124 -0
  3403. executorch/examples/models/qwen3_5_moe/test_quantize_roundtrip.py +323 -0
  3404. executorch/examples/models/qwen3_5_moe/test_sampler.py +142 -0
  3405. executorch/examples/models/qwen3_5_moe/test_serve.py +158 -0
  3406. executorch/examples/models/qwen3_5_moe/test_turboquant.py +211 -0
  3407. executorch/examples/models/resnet/__init__.py +12 -0
  3408. executorch/examples/models/resnet/model.py +48 -0
  3409. executorch/examples/models/silero_vad/export_silero_vad.py +232 -0
  3410. executorch/examples/models/smollm2/__init__.py +16 -0
  3411. executorch/examples/models/smollm2/convert_weights.py +84 -0
  3412. executorch/examples/models/smollm3/__init__.py +16 -0
  3413. executorch/examples/models/smollm3/convert_weights.py +112 -0
  3414. executorch/examples/models/smolvlm/__init__.py +16 -0
  3415. executorch/examples/models/smolvlm/convert_weights.py +77 -0
  3416. executorch/examples/models/sortformer/export_sortformer.py +409 -0
  3417. executorch/examples/models/sortformer/validate_pte.py +136 -0
  3418. executorch/examples/models/stable_diffusion/__init__.py +21 -0
  3419. executorch/examples/models/stable_diffusion/model.py +202 -0
  3420. executorch/examples/models/test/__init__.py +0 -0
  3421. executorch/examples/models/test/test_export.py +161 -0
  3422. executorch/examples/models/torchvision_vit/__init__.py +11 -0
  3423. executorch/examples/models/torchvision_vit/model.py +28 -0
  3424. executorch/examples/models/toy_model/__init__.py +25 -0
  3425. executorch/examples/models/toy_model/model.py +143 -0
  3426. executorch/examples/models/voxtral_realtime/__init__.py +5 -0
  3427. executorch/examples/models/voxtral_realtime/export_voxtral_rt.py +677 -0
  3428. executorch/examples/models/voxtral_realtime/model.py +1517 -0
  3429. executorch/examples/models/voxtral_realtime/tests/test_ring_kv_cache.py +42 -0
  3430. executorch/examples/models/voxtral_tts/__init__.py +5 -0
  3431. executorch/examples/models/voxtral_tts/export_voxtral_tts.py +720 -0
  3432. executorch/examples/models/voxtral_tts/model.py +1609 -0
  3433. executorch/examples/models/voxtral_tts/voice.py +91 -0
  3434. executorch/examples/models/wav2letter/__init__.py +11 -0
  3435. executorch/examples/models/wav2letter/model.py +29 -0
  3436. executorch/examples/models/whisper/__init__.py +8 -0
  3437. executorch/examples/models/whisper/model.py +43 -0
  3438. executorch/examples/models/yolo26/__init__.py +8 -0
  3439. executorch/examples/models/yolo26/export_and_validate.py +397 -0
  3440. executorch/examples/models/yolo26/model.py +30 -0
  3441. executorch/examples/nxp/experimental/cifar_net/cifar_net.py +301 -0
  3442. executorch/examples/qualcomm/__init__.py +5 -0
  3443. executorch/examples/qualcomm/custom_op/custom_ops_1.py +305 -0
  3444. executorch/examples/qualcomm/custom_op/custom_ops_2.py +315 -0
  3445. executorch/examples/qualcomm/oss_scripts/__init__.py +5 -0
  3446. executorch/examples/qualcomm/oss_scripts/albert.py +175 -0
  3447. executorch/examples/qualcomm/oss_scripts/bert.py +163 -0
  3448. executorch/examples/qualcomm/oss_scripts/conv_former.py +135 -0
  3449. executorch/examples/qualcomm/oss_scripts/convnext_small.py +139 -0
  3450. executorch/examples/qualcomm/oss_scripts/cvt.py +201 -0
  3451. executorch/examples/qualcomm/oss_scripts/deit.py +153 -0
  3452. executorch/examples/qualcomm/oss_scripts/depthanything_v2_small.py +214 -0
  3453. executorch/examples/qualcomm/oss_scripts/dino_v2.py +139 -0
  3454. executorch/examples/qualcomm/oss_scripts/distilbert.py +164 -0
  3455. executorch/examples/qualcomm/oss_scripts/dit.py +152 -0
  3456. executorch/examples/qualcomm/oss_scripts/efficientSAM/efficientSAM.py +329 -0
  3457. executorch/examples/qualcomm/oss_scripts/efficientSAM/source_transformation/__init__.py +17 -0
  3458. executorch/examples/qualcomm/oss_scripts/efficientSAM/source_transformation/mask_decoder.py +125 -0
  3459. executorch/examples/qualcomm/oss_scripts/efficientSAM/source_transformation/pos_emb.py +64 -0
  3460. executorch/examples/qualcomm/oss_scripts/efficientnet.py +144 -0
  3461. executorch/examples/qualcomm/oss_scripts/esrgan.py +172 -0
  3462. executorch/examples/qualcomm/oss_scripts/eurobert.py +173 -0
  3463. executorch/examples/qualcomm/oss_scripts/fastvit.py +218 -0
  3464. executorch/examples/qualcomm/oss_scripts/fbnet.py +137 -0
  3465. executorch/examples/qualcomm/oss_scripts/focalnet.py +138 -0
  3466. executorch/examples/qualcomm/oss_scripts/gMLP_image_classification.py +127 -0
  3467. executorch/examples/qualcomm/oss_scripts/hf_causal_lm.py +260 -0
  3468. executorch/examples/qualcomm/oss_scripts/llama/__init__.py +604 -0
  3469. executorch/examples/qualcomm/oss_scripts/llama/dataset/__init__.py +53 -0
  3470. executorch/examples/qualcomm/oss_scripts/llama/dataset/builders.py +348 -0
  3471. executorch/examples/qualcomm/oss_scripts/llama/dataset/collators.py +75 -0
  3472. executorch/examples/qualcomm/oss_scripts/llama/dataset/config.py +41 -0
  3473. executorch/examples/qualcomm/oss_scripts/llama/dataset/datasets.py +41 -0
  3474. executorch/examples/qualcomm/oss_scripts/llama/dataset/loaders.py +105 -0
  3475. executorch/examples/qualcomm/oss_scripts/llama/dataset/preprocessors.py +122 -0
  3476. executorch/examples/qualcomm/oss_scripts/llama/dataset/schema.py +34 -0
  3477. executorch/examples/qualcomm/oss_scripts/llama/decoder_constants.py +66 -0
  3478. executorch/examples/qualcomm/oss_scripts/llama/encoder/__init__.py +31 -0
  3479. executorch/examples/qualcomm/oss_scripts/llama/encoder/encoder_config.py +120 -0
  3480. executorch/examples/qualcomm/oss_scripts/llama/encoder/encoder_quant_recipe.py +118 -0
  3481. executorch/examples/qualcomm/oss_scripts/llama/evaluator/__init__.py +25 -0
  3482. executorch/examples/qualcomm/oss_scripts/llama/evaluator/device_evaluator.py +794 -0
  3483. executorch/examples/qualcomm/oss_scripts/llama/evaluator/lm_eval_adapter.py +99 -0
  3484. executorch/examples/qualcomm/oss_scripts/llama/inference/__init__.py +25 -0
  3485. executorch/examples/qualcomm/oss_scripts/llama/inference/decoder.py +148 -0
  3486. executorch/examples/qualcomm/oss_scripts/llama/inference/encoder.py +22 -0
  3487. executorch/examples/qualcomm/oss_scripts/llama/inference/model.py +59 -0
  3488. executorch/examples/qualcomm/oss_scripts/llama/llama.py +870 -0
  3489. executorch/examples/qualcomm/oss_scripts/llama/masking_utils.py +312 -0
  3490. executorch/examples/qualcomm/oss_scripts/llama/mix_precision_analyzer.py +694 -0
  3491. executorch/examples/qualcomm/oss_scripts/llama/model/__init__.py +16 -0
  3492. executorch/examples/qualcomm/oss_scripts/llama/model/apply_rope.py +52 -0
  3493. executorch/examples/qualcomm/oss_scripts/llama/model/audio_encoder.py +157 -0
  3494. executorch/examples/qualcomm/oss_scripts/llama/model/embedding.py +39 -0
  3495. executorch/examples/qualcomm/oss_scripts/llama/model/feed_forward.py +141 -0
  3496. executorch/examples/qualcomm/oss_scripts/llama/model/layernorm.py +50 -0
  3497. executorch/examples/qualcomm/oss_scripts/llama/model/static_llama.py +1088 -0
  3498. executorch/examples/qualcomm/oss_scripts/llama/model/vision_encoder.py +364 -0
  3499. executorch/examples/qualcomm/oss_scripts/llama/quantize/__init__.py +12 -0
  3500. executorch/examples/qualcomm/oss_scripts/llama/quantize/ptq.py +77 -0
  3501. executorch/examples/qualcomm/oss_scripts/llama/quantize/strategy.py +34 -0
  3502. executorch/examples/qualcomm/oss_scripts/llama/range_setting_pt2e.py +279 -0
  3503. executorch/examples/qualcomm/oss_scripts/llama/static_llm_quant_recipe.py +738 -0
  3504. executorch/examples/qualcomm/oss_scripts/llama/tokenizer.py +414 -0
  3505. executorch/examples/qualcomm/oss_scripts/llama/utils.py +70 -0
  3506. executorch/examples/qualcomm/oss_scripts/llama/wrappers/__init__.py +23 -0
  3507. executorch/examples/qualcomm/oss_scripts/llama/wrappers/attention_sink_wrappers.py +469 -0
  3508. executorch/examples/qualcomm/oss_scripts/llama/wrappers/base_component.py +197 -0
  3509. executorch/examples/qualcomm/oss_scripts/llama/wrappers/llm_wrappers.py +1344 -0
  3510. executorch/examples/qualcomm/oss_scripts/llm_utils/decoder_model_wrapper.py +165 -0
  3511. executorch/examples/qualcomm/oss_scripts/llm_utils/eval_decoder_model_qnn.py +299 -0
  3512. executorch/examples/qualcomm/oss_scripts/llm_utils/qnn_decoder_model_manager.py +341 -0
  3513. executorch/examples/qualcomm/oss_scripts/maxvit_t.py +247 -0
  3514. executorch/examples/qualcomm/oss_scripts/mobilevit_v1.py +166 -0
  3515. executorch/examples/qualcomm/oss_scripts/mobilevit_v2.py +171 -0
  3516. executorch/examples/qualcomm/oss_scripts/moshi/mimi.py +525 -0
  3517. executorch/examples/qualcomm/oss_scripts/moshi/model/static_conv.py +145 -0
  3518. executorch/examples/qualcomm/oss_scripts/moshi/model/static_convtr.py +175 -0
  3519. executorch/examples/qualcomm/oss_scripts/moshi/model/static_mimi.py +679 -0
  3520. executorch/examples/qualcomm/oss_scripts/moshi/model/static_seanet_decoder.py +313 -0
  3521. executorch/examples/qualcomm/oss_scripts/pvt.py +137 -0
  3522. executorch/examples/qualcomm/oss_scripts/regnet.py +144 -0
  3523. executorch/examples/qualcomm/oss_scripts/retinanet.py +319 -0
  3524. executorch/examples/qualcomm/oss_scripts/roberta.py +191 -0
  3525. executorch/examples/qualcomm/oss_scripts/squeezenet.py +129 -0
  3526. executorch/examples/qualcomm/oss_scripts/ssd300_vgg16.py +275 -0
  3527. executorch/examples/qualcomm/oss_scripts/swin_transformer.py +191 -0
  3528. executorch/examples/qualcomm/oss_scripts/swin_v2_t.py +189 -0
  3529. executorch/examples/qualcomm/oss_scripts/t5/t5.py +380 -0
  3530. executorch/examples/qualcomm/oss_scripts/t5/t5_model.py +550 -0
  3531. executorch/examples/qualcomm/oss_scripts/vit_b_16.py +140 -0
  3532. executorch/examples/qualcomm/oss_scripts/whisper/whisper.py +543 -0
  3533. executorch/examples/qualcomm/oss_scripts/whisper/whisper_model.py +114 -0
  3534. executorch/examples/qualcomm/qaihub_scripts/llama/llama2/qaihub_llama2_7b.py +241 -0
  3535. executorch/examples/qualcomm/qaihub_scripts/llama/llama3/qaihub_llama3_8b.py +248 -0
  3536. executorch/examples/qualcomm/qaihub_scripts/stable_diffusion/qaihub_stable_diffusion.py +403 -0
  3537. executorch/examples/qualcomm/qaihub_scripts/stable_diffusion/stable_diffusion_lib.py +22 -0
  3538. executorch/examples/qualcomm/qaihub_scripts/utils/export.py +507 -0
  3539. executorch/examples/qualcomm/qaihub_scripts/utils/utils.py +82 -0
  3540. executorch/examples/qualcomm/scripts/deeplab_v3.py +200 -0
  3541. executorch/examples/qualcomm/scripts/edsr.py +231 -0
  3542. executorch/examples/qualcomm/scripts/export_example.py +127 -0
  3543. executorch/examples/qualcomm/scripts/inception_v3.py +133 -0
  3544. executorch/examples/qualcomm/scripts/inception_v4.py +132 -0
  3545. executorch/examples/qualcomm/scripts/mobilebert_fine_tune.py +378 -0
  3546. executorch/examples/qualcomm/scripts/mobilenet_v2.py +133 -0
  3547. executorch/examples/qualcomm/scripts/mobilenet_v3.py +140 -0
  3548. executorch/examples/qualcomm/scripts/torchvision_vit.py +180 -0
  3549. executorch/examples/qualcomm/scripts/wav2letter.py +218 -0
  3550. executorch/examples/qualcomm/util_scripts/cli.py +681 -0
  3551. executorch/examples/qualcomm/util_scripts/gen_etrecord.py +104 -0
  3552. executorch/examples/qualcomm/util_scripts/qairt_visualizer_demo.py +93 -0
  3553. executorch/examples/qualcomm/util_scripts/qnn_intermediate_debugger_demo.py +217 -0
  3554. executorch/examples/qualcomm/utils.py +364 -0
  3555. executorch/examples/xnnpack/__init__.py +51 -0
  3556. executorch/examples/xnnpack/aot_compiler.py +148 -0
  3557. executorch/examples/xnnpack/quantization/example.py +205 -0
  3558. executorch/examples/xnnpack/quantization/utils.py +42 -0
  3559. executorch/exir/__init__.py +57 -0
  3560. executorch/exir/_program_utils.py +104 -0
  3561. executorch/exir/_serialize/__init__.py +20 -0
  3562. executorch/exir/_serialize/_cord.py +49 -0
  3563. executorch/exir/_serialize/_dataclass.py +145 -0
  3564. executorch/exir/_serialize/_flatbuffer.py +426 -0
  3565. executorch/exir/_serialize/_flatbuffer_program.py +359 -0
  3566. executorch/exir/_serialize/_named_data_store.py +246 -0
  3567. executorch/exir/_serialize/_program.py +827 -0
  3568. executorch/exir/_serialize/_serialize.py +123 -0
  3569. executorch/exir/_serialize/data_serializer.py +80 -0
  3570. executorch/exir/_serialize/generate_program.py +167 -0
  3571. executorch/exir/_serialize/generated/__init__.py +7 -0
  3572. executorch/exir/_serialize/generated/executorch_flatbuffer/AllocationDetails.py +130 -0
  3573. executorch/exir/_serialize/generated/executorch_flatbuffer/BackendDelegate.py +189 -0
  3574. executorch/exir/_serialize/generated/executorch_flatbuffer/BackendDelegateDataReference.py +113 -0
  3575. executorch/exir/_serialize/generated/executorch_flatbuffer/BackendDelegateInlineData.py +141 -0
  3576. executorch/exir/_serialize/generated/executorch_flatbuffer/Bool.py +96 -0
  3577. executorch/exir/_serialize/generated/executorch_flatbuffer/BoolList.py +141 -0
  3578. executorch/exir/_serialize/generated/executorch_flatbuffer/Buffer.py +141 -0
  3579. executorch/exir/_serialize/generated/executorch_flatbuffer/Chain.py +317 -0
  3580. executorch/exir/_serialize/generated/executorch_flatbuffer/CompileSpec.py +162 -0
  3581. executorch/exir/_serialize/generated/executorch_flatbuffer/ContainerMetadata.py +120 -0
  3582. executorch/exir/_serialize/generated/executorch_flatbuffer/DataLocation.py +7 -0
  3583. executorch/exir/_serialize/generated/executorch_flatbuffer/DataSegment.py +113 -0
  3584. executorch/exir/_serialize/generated/executorch_flatbuffer/DelegateCall.py +158 -0
  3585. executorch/exir/_serialize/generated/executorch_flatbuffer/DeviceType.py +7 -0
  3586. executorch/exir/_serialize/generated/executorch_flatbuffer/Double.py +96 -0
  3587. executorch/exir/_serialize/generated/executorch_flatbuffer/DoubleList.py +141 -0
  3588. executorch/exir/_serialize/generated/executorch_flatbuffer/EValue.py +137 -0
  3589. executorch/exir/_serialize/generated/executorch_flatbuffer/ExecutionPlan.py +595 -0
  3590. executorch/exir/_serialize/generated/executorch_flatbuffer/ExtraTensorInfo.py +168 -0
  3591. executorch/exir/_serialize/generated/executorch_flatbuffer/Frame.py +157 -0
  3592. executorch/exir/_serialize/generated/executorch_flatbuffer/FrameList.py +143 -0
  3593. executorch/exir/_serialize/generated/executorch_flatbuffer/FreeCall.py +96 -0
  3594. executorch/exir/_serialize/generated/executorch_flatbuffer/Instruction.py +131 -0
  3595. executorch/exir/_serialize/generated/executorch_flatbuffer/InstructionArguments.py +33 -0
  3596. executorch/exir/_serialize/generated/executorch_flatbuffer/Int.py +96 -0
  3597. executorch/exir/_serialize/generated/executorch_flatbuffer/IntList.py +141 -0
  3598. executorch/exir/_serialize/generated/executorch_flatbuffer/JumpFalseCall.py +113 -0
  3599. executorch/exir/_serialize/generated/executorch_flatbuffer/KernelCall.py +158 -0
  3600. executorch/exir/_serialize/generated/executorch_flatbuffer/KernelTypes.py +57 -0
  3601. executorch/exir/_serialize/generated/executorch_flatbuffer/MoveCall.py +113 -0
  3602. executorch/exir/_serialize/generated/executorch_flatbuffer/NamedData.py +117 -0
  3603. executorch/exir/_serialize/generated/executorch_flatbuffer/NonConstBufferDevice.py +130 -0
  3604. executorch/exir/_serialize/generated/executorch_flatbuffer/Null.py +80 -0
  3605. executorch/exir/_serialize/generated/executorch_flatbuffer/Operator.py +120 -0
  3606. executorch/exir/_serialize/generated/executorch_flatbuffer/OptionalTensorList.py +141 -0
  3607. executorch/exir/_serialize/generated/executorch_flatbuffer/Program.py +474 -0
  3608. executorch/exir/_serialize/generated/executorch_flatbuffer/ScalarType.py +28 -0
  3609. executorch/exir/_serialize/generated/executorch_flatbuffer/String.py +100 -0
  3610. executorch/exir/_serialize/generated/executorch_flatbuffer/SubsegmentOffsets.py +158 -0
  3611. executorch/exir/_serialize/generated/executorch_flatbuffer/Tensor.py +355 -0
  3612. executorch/exir/_serialize/generated/executorch_flatbuffer/TensorDataLocation.py +7 -0
  3613. executorch/exir/_serialize/generated/executorch_flatbuffer/TensorList.py +141 -0
  3614. executorch/exir/_serialize/generated/executorch_flatbuffer/TensorShapeDynamism.py +8 -0
  3615. executorch/exir/_serialize/generated/executorch_flatbuffer/__init__.py +91 -0
  3616. executorch/exir/_serialize/padding.py +39 -0
  3617. executorch/exir/_serialize/program.fbs +534 -0
  3618. executorch/exir/_serialize/scalar_type.fbs +43 -0
  3619. executorch/exir/_serialize/test/__init__.py +0 -0
  3620. executorch/exir/_serialize/test/test_cord.py +63 -0
  3621. executorch/exir/_serialize/test/test_flatbuffer.py +273 -0
  3622. executorch/exir/_serialize/test/test_flatbuffer_program.py +174 -0
  3623. executorch/exir/_serialize/test/test_named_data_store.py +249 -0
  3624. executorch/exir/_serialize/test/test_program.py +1369 -0
  3625. executorch/exir/_serialize/test/test_serialize.py +289 -0
  3626. executorch/exir/_warnings.py +44 -0
  3627. executorch/exir/backend/backend_api.py +794 -0
  3628. executorch/exir/backend/backend_details.py +150 -0
  3629. executorch/exir/backend/canonical_partitioners/all_node_partitioner.py +55 -0
  3630. executorch/exir/backend/canonical_partitioners/config_partitioner.py +262 -0
  3631. executorch/exir/backend/canonical_partitioners/duplicate_constant_node_pass.py +151 -0
  3632. executorch/exir/backend/canonical_partitioners/duplicate_dequant_node_pass.py +111 -0
  3633. executorch/exir/backend/canonical_partitioners/group_partitioner.py +382 -0
  3634. executorch/exir/backend/canonical_partitioners/pattern_op_partitioner.py +185 -0
  3635. executorch/exir/backend/compile_spec_schema.py +17 -0
  3636. executorch/exir/backend/operator_support.py +127 -0
  3637. executorch/exir/backend/partitioner.py +114 -0
  3638. executorch/exir/backend/test/backend_with_compiler_demo.py +142 -0
  3639. executorch/exir/backend/test/backend_with_delegate_mapping_demo.py +210 -0
  3640. executorch/exir/backend/test/backend_with_named_data_map.py +115 -0
  3641. executorch/exir/backend/test/backend_with_preprocess_all_demo.py +297 -0
  3642. executorch/exir/backend/test/demo_backend.py +35 -0
  3643. executorch/exir/backend/test/demos/rpc/executor_backend_partitioner.py +77 -0
  3644. executorch/exir/backend/test/demos/rpc/executor_backend_preprocess.py +39 -0
  3645. executorch/exir/backend/test/demos/rpc/test_rpc.py +189 -0
  3646. executorch/exir/backend/test/demos/test_delegate_aten_mode.py +72 -0
  3647. executorch/exir/backend/test/device_util.py +112 -0
  3648. executorch/exir/backend/test/hta_partitioner_demo.py +247 -0
  3649. executorch/exir/backend/test/op_partitioner_demo.py +266 -0
  3650. executorch/exir/backend/test/test_backend_with_named_data_map.py +85 -0
  3651. executorch/exir/backend/test/test_backends_lifted.py +1322 -0
  3652. executorch/exir/backend/test/test_backends_nested.py +258 -0
  3653. executorch/exir/backend/test/test_compatibility.py +118 -0
  3654. executorch/exir/backend/test/test_debug_handle_map.py +136 -0
  3655. executorch/exir/backend/test/test_delegate_map_builder.py +274 -0
  3656. executorch/exir/backend/test/test_graph_partition.py +248 -0
  3657. executorch/exir/backend/test/test_group_partitioner.py +1770 -0
  3658. executorch/exir/backend/test/test_lowered_backend_module.py +281 -0
  3659. executorch/exir/backend/test/test_partitioner.py +731 -0
  3660. executorch/exir/backend/test/test_passes.py +88 -0
  3661. executorch/exir/backend/test/test_to_backend_multi_method.py +595 -0
  3662. executorch/exir/backend/test/test_utils.py +392 -0
  3663. executorch/exir/backend/utils.py +724 -0
  3664. executorch/exir/capture/__init__.py +28 -0
  3665. executorch/exir/capture/_capture.py +381 -0
  3666. executorch/exir/capture/_config.py +140 -0
  3667. executorch/exir/capture/_unlift.py +159 -0
  3668. executorch/exir/common.py +140 -0
  3669. executorch/exir/control_flow.py +204 -0
  3670. executorch/exir/debug_handle_utils.py +29 -0
  3671. executorch/exir/delegate.py +171 -0
  3672. executorch/exir/delegate.pyi +21 -0
  3673. executorch/exir/dialects/__init__.py +0 -0
  3674. executorch/exir/dialects/_ops.py +166 -0
  3675. executorch/exir/dialects/backend/_ops.py +123 -0
  3676. executorch/exir/dialects/backend/test/test_backend_ops.py +88 -0
  3677. executorch/exir/dialects/edge/_ops.py +405 -0
  3678. executorch/exir/dialects/edge/arg/model.py +238 -0
  3679. executorch/exir/dialects/edge/arg/type.py +46 -0
  3680. executorch/exir/dialects/edge/dtype/runner.py +122 -0
  3681. executorch/exir/dialects/edge/dtype/supported.py +49 -0
  3682. executorch/exir/dialects/edge/dtype/utils.py +36 -0
  3683. executorch/exir/dialects/edge/edge.yaml +8388 -0
  3684. executorch/exir/dialects/edge/op/api.py +86 -0
  3685. executorch/exir/dialects/edge/op/sample_input.py +1283 -0
  3686. executorch/exir/dialects/edge/op/test/test_api.py +51 -0
  3687. executorch/exir/dialects/edge/spec/gen.py +523 -0
  3688. executorch/exir/dialects/edge/spec/utils.py +328 -0
  3689. executorch/exir/dialects/edge/test/test_edge_ops.py +443 -0
  3690. executorch/exir/dialects/edge/test/test_edge_yaml.py +225 -0
  3691. executorch/exir/dialects/test/test_exir_dialect_ops.py +22 -0
  3692. executorch/exir/dim_order_utils.py +85 -0
  3693. executorch/exir/dynamic_shape.py +12 -0
  3694. executorch/exir/emit/__init__.py +11 -0
  3695. executorch/exir/emit/_emit_program.py +216 -0
  3696. executorch/exir/emit/_emitter.py +2117 -0
  3697. executorch/exir/emit/test/test_emit.py +2812 -0
  3698. executorch/exir/error.py +55 -0
  3699. executorch/exir/graph.py +86 -0
  3700. executorch/exir/graph_module.py +153 -0
  3701. executorch/exir/lowered_backend_module.py +1073 -0
  3702. executorch/exir/memory.py +50 -0
  3703. executorch/exir/memory_planning.py +1610 -0
  3704. executorch/exir/operator/convert.py +371 -0
  3705. executorch/exir/operator/manip.py +89 -0
  3706. executorch/exir/operator/test/test_operator.py +172 -0
  3707. executorch/exir/operator/util.py +68 -0
  3708. executorch/exir/pass_base.py +979 -0
  3709. executorch/exir/pass_manager.py +241 -0
  3710. executorch/exir/passes/__init__.py +545 -0
  3711. executorch/exir/passes/_device_copy_ops_registry.py +58 -0
  3712. executorch/exir/passes/_quant_patterns_and_replacements.py +1564 -0
  3713. executorch/exir/passes/const_prop_pass.py +84 -0
  3714. executorch/exir/passes/constant_prop_pass.py +402 -0
  3715. executorch/exir/passes/convert_constant_dim_order_pass.py +86 -0
  3716. executorch/exir/passes/cse_pass.py +201 -0
  3717. executorch/exir/passes/debug_handle_generator_pass.py +81 -0
  3718. executorch/exir/passes/dim_order_ops_registry.py +90 -0
  3719. executorch/exir/passes/dynamic_shape_prop_pass.py +301 -0
  3720. executorch/exir/passes/executorch_prim_ops_registry.py +186 -0
  3721. executorch/exir/passes/external_constants_pass.py +113 -0
  3722. executorch/exir/passes/init_mutable_pass.py +32 -0
  3723. executorch/exir/passes/insert_write_back_for_buffers_pass.py +185 -0
  3724. executorch/exir/passes/legalize_portable_dim_order_pass.py +69 -0
  3725. executorch/exir/passes/memory_format_ops_pass.py +139 -0
  3726. executorch/exir/passes/memory_planning_pass.py +329 -0
  3727. executorch/exir/passes/normalize_transpose_pass.py +24 -0
  3728. executorch/exir/passes/normalize_view_copy_base_pass.py +62 -0
  3729. executorch/exir/passes/pass_registry.py +89 -0
  3730. executorch/exir/passes/propagate_device_config.py +56 -0
  3731. executorch/exir/passes/propagate_device_pass.py +458 -0
  3732. executorch/exir/passes/prune_empty_tensors_pass.py +68 -0
  3733. executorch/exir/passes/quant_fusion_pass.py +174 -0
  3734. executorch/exir/passes/quantize_io_pass.py +414 -0
  3735. executorch/exir/passes/reinplace.py +532 -0
  3736. executorch/exir/passes/remove_graph_asserts_pass.py +62 -0
  3737. executorch/exir/passes/remove_mixed_type_operators.py +95 -0
  3738. executorch/exir/passes/remove_noop_pass.py +129 -0
  3739. executorch/exir/passes/remove_unused_parameters_pass.py +55 -0
  3740. executorch/exir/passes/replace_aten_with_edge_pass.py +59 -0
  3741. executorch/exir/passes/replace_broken_ops_with_function_ops_pass.py +42 -0
  3742. executorch/exir/passes/replace_edge_with_backend_pass.py +56 -0
  3743. executorch/exir/passes/replace_sym_size_op_pass.py +35 -0
  3744. executorch/exir/passes/replace_view_copy_with_view_pass.py +327 -0
  3745. executorch/exir/passes/scalar_to_tensor_pass.py +44 -0
  3746. executorch/exir/passes/spec_prop_pass.py +132 -0
  3747. executorch/exir/passes/sym_shape_eval_pass.py +289 -0
  3748. executorch/exir/passes/sym_to_tensor_pass.py +64 -0
  3749. executorch/exir/passes/to_device_pass.py +45 -0
  3750. executorch/exir/passes/weights_to_outputs_pass.py +88 -0
  3751. executorch/exir/print_program.py +428 -0
  3752. executorch/exir/program/__init__.py +32 -0
  3753. executorch/exir/program/_fake_program.py +67 -0
  3754. executorch/exir/program/_program.py +1998 -0
  3755. executorch/exir/program/test/__init__.py +0 -0
  3756. executorch/exir/program/test/test_fake_program.py +78 -0
  3757. executorch/exir/program/test/test_program.py +952 -0
  3758. executorch/exir/scalar_type.py +38 -0
  3759. executorch/exir/schema.py +330 -0
  3760. executorch/exir/serde/__init__.py +0 -0
  3761. executorch/exir/serde/export_serialize.py +2875 -0
  3762. executorch/exir/serde/schema.py +409 -0
  3763. executorch/exir/serde/schema_check.py +293 -0
  3764. executorch/exir/serde/serialize.py +838 -0
  3765. executorch/exir/serde/union.py +77 -0
  3766. executorch/exir/sym_util.py +105 -0
  3767. executorch/exir/tensor.py +413 -0
  3768. executorch/exir/tensor_layout.py +39 -0
  3769. executorch/exir/tests/asr_joiner.py +82 -0
  3770. executorch/exir/tests/common.py +96 -0
  3771. executorch/exir/tests/control_flow_models.py +105 -0
  3772. executorch/exir/tests/dynamic_shape_models.py +51 -0
  3773. executorch/exir/tests/models.py +576 -0
  3774. executorch/exir/tests/test_arg_validator.py +72 -0
  3775. executorch/exir/tests/test_common.py +81 -0
  3776. executorch/exir/tests/test_delegate.py +308 -0
  3777. executorch/exir/tests/test_device_copy_ops.py +73 -0
  3778. executorch/exir/tests/test_dim_order_utils.py +49 -0
  3779. executorch/exir/tests/test_dynamic_shape_propagation.py +174 -0
  3780. executorch/exir/tests/test_error.py +19 -0
  3781. executorch/exir/tests/test_extract_io_quant_params.py +93 -0
  3782. executorch/exir/tests/test_joint_graph.py +124 -0
  3783. executorch/exir/tests/test_legalize_portable_dim_order_pass.py +228 -0
  3784. executorch/exir/tests/test_memory_format_ops_pass.py +443 -0
  3785. executorch/exir/tests/test_memory_format_ops_pass_aten.py +142 -0
  3786. executorch/exir/tests/test_memory_format_ops_pass_utils.py +266 -0
  3787. executorch/exir/tests/test_memory_planning.py +1767 -0
  3788. executorch/exir/tests/test_op_convert.py +88 -0
  3789. executorch/exir/tests/test_pass_infra.py +559 -0
  3790. executorch/exir/tests/test_passes.py +2898 -0
  3791. executorch/exir/tests/test_print_program.py +54 -0
  3792. executorch/exir/tests/test_propagate_device_pass.py +1275 -0
  3793. executorch/exir/tests/test_prune_empty_tensors_pass.py +112 -0
  3794. executorch/exir/tests/test_quant_fusion_pass.py +548 -0
  3795. executorch/exir/tests/test_quantization.py +120 -0
  3796. executorch/exir/tests/test_quantize_io_pass.py +156 -0
  3797. executorch/exir/tests/test_reinplace_pass.py +725 -0
  3798. executorch/exir/tests/test_remove_unused_parameters_pass.py +205 -0
  3799. executorch/exir/tests/test_remove_view_copy.py +236 -0
  3800. executorch/exir/tests/test_serde.py +352 -0
  3801. executorch/exir/tests/test_tensor.py +416 -0
  3802. executorch/exir/tests/test_verification.py +326 -0
  3803. executorch/exir/tests/test_warnings.py +169 -0
  3804. executorch/exir/tests/transformer.py +122 -0
  3805. executorch/exir/tracer.py +803 -0
  3806. executorch/exir/types.py +18 -0
  3807. executorch/exir/verification/arg_validator.py +140 -0
  3808. executorch/exir/verification/dev_html.py +68 -0
  3809. executorch/exir/verification/interpreter.py +425 -0
  3810. executorch/exir/verification/test/test_verifier.py +137 -0
  3811. executorch/exir/verification/verifier.py +346 -0
  3812. executorch/exir/version.py +9 -0
  3813. executorch/exir/wrap.py +32 -0
  3814. executorch/export/__init__.py +40 -0
  3815. executorch/export/export.py +696 -0
  3816. executorch/export/recipe.py +328 -0
  3817. executorch/export/recipe_provider.py +60 -0
  3818. executorch/export/recipe_registry.py +86 -0
  3819. executorch/export/stages.py +686 -0
  3820. executorch/export/target_recipes.py +185 -0
  3821. executorch/export/tests/test_executorch_export.py +34 -0
  3822. executorch/export/tests/test_export_recipe.py +131 -0
  3823. executorch/export/tests/test_export_session.py +973 -0
  3824. executorch/export/tests/test_export_stages.py +534 -0
  3825. executorch/export/tests/test_recipe_provider.py +98 -0
  3826. executorch/export/tests/test_recipe_registry.py +139 -0
  3827. executorch/export/tests/test_target_recipes.py +539 -0
  3828. executorch/export/types.py +22 -0
  3829. executorch/export/utils.py +51 -0
  3830. executorch/extension/apple/ExecuTorch/__tests__/resources/generate_coreml_test_models.py +81 -0
  3831. executorch/extension/audio/mel_spectrogram.py +332 -0
  3832. executorch/extension/export_util/__init__.py +13 -0
  3833. executorch/extension/export_util/utils.py +147 -0
  3834. executorch/extension/flat_tensor/__init__.py +0 -0
  3835. executorch/extension/flat_tensor/serialize/__init__.py +0 -0
  3836. executorch/extension/flat_tensor/serialize/flat_tensor.fbs +73 -0
  3837. executorch/extension/flat_tensor/serialize/flat_tensor_schema.py +34 -0
  3838. executorch/extension/flat_tensor/serialize/scalar_type.fbs +43 -0
  3839. executorch/extension/flat_tensor/serialize/serialize.py +446 -0
  3840. executorch/extension/flat_tensor/test/test_serialize.py +326 -0
  3841. executorch/extension/gguf_util/convert_main.py +53 -0
  3842. executorch/extension/gguf_util/converter.py +27 -0
  3843. executorch/extension/gguf_util/converters/llama_converter.py +123 -0
  3844. executorch/extension/gguf_util/load_gguf.py +103 -0
  3845. executorch/extension/image/benchmark/compare_benchmarks.py +122 -0
  3846. executorch/extension/llm/custom_ops/__init__.py +0 -0
  3847. executorch/extension/llm/custom_ops/custom_ops.py +581 -0
  3848. executorch/extension/llm/custom_ops/custom_ops_aot_lib.dll +0 -0
  3849. executorch/extension/llm/custom_ops/model_sharding.py +102 -0
  3850. executorch/extension/llm/custom_ops/op_fallback.py +29 -0
  3851. executorch/extension/llm/custom_ops/op_tile_crop_aot.py +38 -0
  3852. executorch/extension/llm/custom_ops/preprocess_custom_ops.py +69 -0
  3853. executorch/extension/llm/custom_ops/spinquant/special_hadamard_code_gen.py +279 -0
  3854. executorch/extension/llm/custom_ops/spinquant/third-party/FFHT/example.py +20 -0
  3855. executorch/extension/llm/custom_ops/spinquant/third-party/FFHT/gen.py +869 -0
  3856. executorch/extension/llm/custom_ops/test_preprocess_custom_ops.py +34 -0
  3857. executorch/extension/llm/custom_ops/test_quantized_sdpa.py +550 -0
  3858. executorch/extension/llm/custom_ops/test_sdpa_with_kv_cache.py +685 -0
  3859. executorch/extension/llm/custom_ops/test_update_cache.py +754 -0
  3860. executorch/extension/llm/custom_ops/test_update_cross_attn_cache.py +286 -0
  3861. executorch/extension/llm/export/__init__.py +11 -0
  3862. executorch/extension/llm/export/builder.py +592 -0
  3863. executorch/extension/llm/export/config/llm_config.py +912 -0
  3864. executorch/extension/llm/export/export_llm.py +106 -0
  3865. executorch/extension/llm/export/export_passes.py +202 -0
  3866. executorch/extension/llm/export/gguf.py +577 -0
  3867. executorch/extension/llm/export/int4.py +193 -0
  3868. executorch/extension/llm/export/load.py +315 -0
  3869. executorch/extension/llm/export/nvfp4.py +191 -0
  3870. executorch/extension/llm/export/partitioner_lib.py +278 -0
  3871. executorch/extension/llm/export/quant/__init__.py +20 -0
  3872. executorch/extension/llm/export/quant/convert.py +216 -0
  3873. executorch/extension/llm/export/quant/quantize.py +386 -0
  3874. executorch/extension/llm/export/quant/recipe.py +62 -0
  3875. executorch/extension/llm/export/quant/test/test_convert.py +362 -0
  3876. executorch/extension/llm/export/quant/test/test_quantize.py +310 -0
  3877. executorch/extension/llm/export/quant/test/test_recipe.py +118 -0
  3878. executorch/extension/llm/export/quant/test/test_safetensors_roundtrip.py +143 -0
  3879. executorch/extension/llm/export/quantize.py +286 -0
  3880. executorch/extension/llm/export/quantizer_lib.py +443 -0
  3881. executorch/extension/llm/export/test/__init__.py +0 -0
  3882. executorch/extension/llm/export/test/test_builder.py +118 -0
  3883. executorch/extension/llm/export/test/test_export_llm.py +155 -0
  3884. executorch/extension/llm/export/test/test_gguf.py +409 -0
  3885. executorch/extension/llm/export/test/test_int4.py +139 -0
  3886. executorch/extension/llm/export/test_export_passes.py +209 -0
  3887. executorch/extension/llm/modules/__init__.py +24 -0
  3888. executorch/extension/llm/modules/_position_embeddings.py +729 -0
  3889. executorch/extension/llm/modules/attention.py +413 -0
  3890. executorch/extension/llm/modules/kv_cache.py +151 -0
  3891. executorch/extension/llm/modules/test/__init__.py +0 -0
  3892. executorch/extension/llm/modules/test/test_attention.py +353 -0
  3893. executorch/extension/llm/modules/test/test_kv_cache.py +218 -0
  3894. executorch/extension/llm/modules/test/test_position_embeddings.py +230 -0
  3895. executorch/extension/llm/modules/test/test_turboquant_kv_cache.py +262 -0
  3896. executorch/extension/llm/modules/turboquant/__init__.py +11 -0
  3897. executorch/extension/llm/modules/turboquant/codebook.py +124 -0
  3898. executorch/extension/llm/modules/turboquant/kv_cache.py +170 -0
  3899. executorch/extension/llm/runner/__init__.py +238 -0
  3900. executorch/extension/llm/runner/_llm_runner.pyi +596 -0
  3901. executorch/extension/llm/runner/test/test_runner_pybindings.py +266 -0
  3902. executorch/extension/llm/tokenizers/pytorch_tokenizers/__init__.py +63 -0
  3903. executorch/extension/llm/tokenizers/pytorch_tokenizers/constants.py +31 -0
  3904. executorch/extension/llm/tokenizers/pytorch_tokenizers/hf_tokenizer.py +57 -0
  3905. executorch/extension/llm/tokenizers/pytorch_tokenizers/llama2c.py +111 -0
  3906. executorch/extension/llm/tokenizers/pytorch_tokenizers/tiktoken.py +197 -0
  3907. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/__init__.py +0 -0
  3908. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/llama2c/__init__.py +0 -0
  3909. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/llama2c/convert.py +49 -0
  3910. executorch/extension/llm/tokenizers/setup.py +200 -0
  3911. executorch/extension/llm/tokenizers/test/test_hf_tokenizer.py +96 -0
  3912. executorch/extension/llm/tokenizers/test/test_python_bindings.py +131 -0
  3913. executorch/extension/llm/tokenizers/test/test_tekken_python.py +453 -0
  3914. executorch/extension/llm/tokenizers/test/test_tiktoken.py +49 -0
  3915. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/abseil.podspec.gen.py +243 -0
  3916. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/copts/copts.py +194 -0
  3917. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/copts/generate_copts.py +109 -0
  3918. executorch/extension/llm/tokenizers/third-party/abseil-cpp/conanfile.py +51 -0
  3919. executorch/extension/llm/tokenizers/third-party/abseil-cpp/create_lts.py +139 -0
  3920. executorch/extension/llm/tokenizers/third-party/json/docs/mkdocs/scripts/check_structure.py +179 -0
  3921. executorch/extension/llm/tokenizers/third-party/json/tests/thirdparty/imapdl/filterbr.py +109 -0
  3922. executorch/extension/llm/tokenizers/third-party/json/tools/amalgamate/amalgamate.py +299 -0
  3923. executorch/extension/llm/tokenizers/third-party/json/tools/gdb_pretty_printer/nlohmann-json.py +32 -0
  3924. executorch/extension/llm/tokenizers/third-party/json/tools/generate_natvis/generate_natvis.py +41 -0
  3925. executorch/extension/llm/tokenizers/third-party/json/tools/serve_header/serve_header.py +410 -0
  3926. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateCommon.py +354 -0
  3927. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateTest.py +186 -0
  3928. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcd.py +999 -0
  3929. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcpHeader.py +98 -0
  3930. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcpTables.py +203 -0
  3931. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateCommon.py +31 -0
  3932. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateDates.py +59 -0
  3933. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateRelease.py +29 -0
  3934. executorch/extension/llm/tokenizers/third-party/re2/benchlog/benchplot.py +98 -0
  3935. executorch/extension/llm/tokenizers/third-party/re2/python/re2.py +583 -0
  3936. executorch/extension/llm/tokenizers/third-party/re2/python/re2_test.py +489 -0
  3937. executorch/extension/llm/tokenizers/third-party/re2/python/setup.py +158 -0
  3938. executorch/extension/llm/tokenizers/third-party/re2/python/toolchains/generate.py +98 -0
  3939. executorch/extension/llm/tokenizers/third-party/re2/re2/make_unicode_casefold.py +151 -0
  3940. executorch/extension/llm/tokenizers/third-party/re2/re2/make_unicode_groups.py +117 -0
  3941. executorch/extension/llm/tokenizers/third-party/re2/re2/unicode.py +284 -0
  3942. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/setup.py +225 -0
  3943. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/__init__.py +1224 -0
  3944. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/_version.py +1 -0
  3945. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/sentencepiece_model_pb2.py +44 -0
  3946. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/sentencepiece_pb2.py +30 -0
  3947. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/test/__init__.py +0 -0
  3948. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/test/sentencepiece_test.py +937 -0
  3949. executorch/extension/module/test/resources/gen_bundled_program.py +99 -0
  3950. executorch/extension/pybindings/_portable_lib.cp314-win_amd64.pyd +0 -0
  3951. executorch/extension/pybindings/data_loader.cp314-win_amd64.pyd +0 -0
  3952. executorch/extension/pybindings/data_loader.pyi +12 -0
  3953. executorch/extension/pybindings/portable_lib.py +108 -0
  3954. executorch/extension/pybindings/pybindings.pyi +393 -0
  3955. executorch/extension/pybindings/test/make_test.py +186 -0
  3956. executorch/extension/pybindings/test/test_backend_pybinding.py +27 -0
  3957. executorch/extension/pybindings/test/test_pybindings.py +735 -0
  3958. executorch/extension/pytree/__init__.py +54 -0
  3959. executorch/extension/pytree/test/test.py +331 -0
  3960. executorch/extension/training/__init__.py +22 -0
  3961. executorch/extension/training/examples/CIFAR/data_utils.py +389 -0
  3962. executorch/extension/training/examples/CIFAR/export.py +220 -0
  3963. executorch/extension/training/examples/CIFAR/main.py +188 -0
  3964. executorch/extension/training/examples/CIFAR/model.py +73 -0
  3965. executorch/extension/training/examples/CIFAR/train_utils.py +622 -0
  3966. executorch/extension/training/examples/XOR/export_model.py +77 -0
  3967. executorch/extension/training/examples/XOR/model.py +34 -0
  3968. executorch/extension/training/examples/XOR/test/test_export.py +19 -0
  3969. executorch/extension/training/pybindings/_training_lib.pyi +45 -0
  3970. executorch/extension/training/pybindings/_training_module.py +86 -0
  3971. executorch/extension/training/pybindings/test/test.py +67 -0
  3972. executorch/extension/wasm/test/test_model.py +34 -0
  3973. executorch/include/executorch/extension/kernel_util/make_boxed_from_unboxed_functor.h +260 -0
  3974. executorch/include/executorch/extension/kernel_util/meta_programming.h +119 -0
  3975. executorch/include/executorch/extension/kernel_util/type_list.h +145 -0
  3976. executorch/include/executorch/extension/tensor/tensor.h +14 -0
  3977. executorch/include/executorch/extension/tensor/tensor_accessor.h +218 -0
  3978. executorch/include/executorch/extension/tensor/tensor_ptr.h +561 -0
  3979. executorch/include/executorch/extension/tensor/tensor_ptr_maker.h +687 -0
  3980. executorch/include/executorch/extension/threadpool/cpuinfo_utils.h +24 -0
  3981. executorch/include/executorch/extension/threadpool/threadpool.h +130 -0
  3982. executorch/include/executorch/extension/threadpool/threadpool_guard.h +37 -0
  3983. executorch/include/executorch/runtime/backend/backend_execution_context.h +80 -0
  3984. executorch/include/executorch/runtime/backend/backend_init_context.h +149 -0
  3985. executorch/include/executorch/runtime/backend/backend_option_context.h +34 -0
  3986. executorch/include/executorch/runtime/backend/backend_options_map.h +229 -0
  3987. executorch/include/executorch/runtime/backend/interface.h +230 -0
  3988. executorch/include/executorch/runtime/backend/options.h +208 -0
  3989. executorch/include/executorch/runtime/core/array_ref.h +264 -0
  3990. executorch/include/executorch/runtime/core/data_loader.h +141 -0
  3991. executorch/include/executorch/runtime/core/defines.h +20 -0
  3992. executorch/include/executorch/runtime/core/device_allocator.h +198 -0
  3993. executorch/include/executorch/runtime/core/device_memory_buffer.h +129 -0
  3994. executorch/include/executorch/runtime/core/error.h +264 -0
  3995. executorch/include/executorch/runtime/core/evalue.h +867 -0
  3996. executorch/include/executorch/runtime/core/event_tracer.h +592 -0
  3997. executorch/include/executorch/runtime/core/event_tracer_hooks.h +340 -0
  3998. executorch/include/executorch/runtime/core/event_tracer_hooks_delegate.h +204 -0
  3999. executorch/include/executorch/runtime/core/exec_aten/exec_aten.h +208 -0
  4000. executorch/include/executorch/runtime/core/exec_aten/testing_util/tensor_factory.h +1133 -0
  4001. executorch/include/executorch/runtime/core/exec_aten/testing_util/tensor_util.h +378 -0
  4002. executorch/include/executorch/runtime/core/exec_aten/util/dim_order_util.h +290 -0
  4003. executorch/include/executorch/runtime/core/exec_aten/util/scalar_type_util.h +1430 -0
  4004. executorch/include/executorch/runtime/core/exec_aten/util/tensor_dimension_limit.h +21 -0
  4005. executorch/include/executorch/runtime/core/exec_aten/util/tensor_shape_to_c_string.h +68 -0
  4006. executorch/include/executorch/runtime/core/exec_aten/util/tensor_util.h +1428 -0
  4007. executorch/include/executorch/runtime/core/freeable_buffer.h +218 -0
  4008. executorch/include/executorch/runtime/core/function_ref.h +108 -0
  4009. executorch/include/executorch/runtime/core/hierarchical_allocator.h +167 -0
  4010. executorch/include/executorch/runtime/core/memory_allocator.h +249 -0
  4011. executorch/include/executorch/runtime/core/named_data_map.h +75 -0
  4012. executorch/include/executorch/runtime/core/portable_type/bfloat16.h +27 -0
  4013. executorch/include/executorch/runtime/core/portable_type/bfloat16_math.h +14 -0
  4014. executorch/include/executorch/runtime/core/portable_type/bits_types.h +83 -0
  4015. executorch/include/executorch/runtime/core/portable_type/c10/c10/macros/Export.h +1 -0
  4016. executorch/include/executorch/runtime/core/portable_type/c10/c10/macros/Macros.h +1 -0
  4017. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-inl.h +1 -0
  4018. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-math.h +299 -0
  4019. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16.h +1 -0
  4020. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/Half-inl.h +1 -0
  4021. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/Half.h +8 -0
  4022. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/TypeSafeSignMath.h +1 -0
  4023. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/bit_cast.h +1 -0
  4024. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex.h +78 -0
  4025. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex_math.h +441 -0
  4026. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex_utils.h +46 -0
  4027. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/floating_point_utils.h +1 -0
  4028. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/irange.h +123 -0
  4029. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h +905 -0
  4030. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/overflows.h +100 -0
  4031. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/safe_numerics.h +114 -0
  4032. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Export.h +145 -0
  4033. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h +741 -0
  4034. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/BFloat16.h +481 -0
  4035. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/Half.h +788 -0
  4036. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h +148 -0
  4037. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/bit_cast.h +50 -0
  4038. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/complex.h +616 -0
  4039. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/floating_point_utils.h +38 -0
  4040. executorch/include/executorch/runtime/core/portable_type/complex.h +21 -0
  4041. executorch/include/executorch/runtime/core/portable_type/device.h +85 -0
  4042. executorch/include/executorch/runtime/core/portable_type/half.h +27 -0
  4043. executorch/include/executorch/runtime/core/portable_type/optional.h +37 -0
  4044. executorch/include/executorch/runtime/core/portable_type/qint_types.h +83 -0
  4045. executorch/include/executorch/runtime/core/portable_type/scalar.h +123 -0
  4046. executorch/include/executorch/runtime/core/portable_type/scalar_type.h +154 -0
  4047. executorch/include/executorch/runtime/core/portable_type/string_view.h +31 -0
  4048. executorch/include/executorch/runtime/core/portable_type/tensor.h +190 -0
  4049. executorch/include/executorch/runtime/core/portable_type/tensor_impl.h +340 -0
  4050. executorch/include/executorch/runtime/core/portable_type/tensor_options.h +60 -0
  4051. executorch/include/executorch/runtime/core/result.h +333 -0
  4052. executorch/include/executorch/runtime/core/span.h +110 -0
  4053. executorch/include/executorch/runtime/core/tag.h +90 -0
  4054. executorch/include/executorch/runtime/core/tensor_layout.h +93 -0
  4055. executorch/include/executorch/runtime/core/tensor_shape_dynamism.h +39 -0
  4056. executorch/include/executorch/runtime/core/test/mock_cuda_allocator.h +146 -0
  4057. executorch/include/executorch/runtime/core/testing_util/error_matchers.h +303 -0
  4058. executorch/include/executorch/runtime/executor/memory_manager.h +149 -0
  4059. executorch/include/executorch/runtime/executor/merged_data_map.h +150 -0
  4060. executorch/include/executorch/runtime/executor/method.h +465 -0
  4061. executorch/include/executorch/runtime/executor/method_meta.h +316 -0
  4062. executorch/include/executorch/runtime/executor/platform_memory_allocator.h +124 -0
  4063. executorch/include/executorch/runtime/executor/program.h +329 -0
  4064. executorch/include/executorch/runtime/executor/program_validation.h +51 -0
  4065. executorch/include/executorch/runtime/executor/pte_data_map.h +151 -0
  4066. executorch/include/executorch/runtime/executor/tensor_parser.h +167 -0
  4067. executorch/include/executorch/runtime/executor/test/managed_memory_manager.h +68 -0
  4068. executorch/include/executorch/runtime/kernel/kernel_includes.h +62 -0
  4069. executorch/include/executorch/runtime/kernel/kernel_runtime_context.h +131 -0
  4070. executorch/include/executorch/runtime/kernel/operator_registry.h +308 -0
  4071. executorch/include/executorch/runtime/kernel/test/test_util.h +40 -0
  4072. executorch/include/executorch/runtime/kernel/thread_parallel_interface.h +107 -0
  4073. executorch/include/executorch/runtime/platform/abort.h +36 -0
  4074. executorch/include/executorch/runtime/platform/assert.h +124 -0
  4075. executorch/include/executorch/runtime/platform/clock.h +43 -0
  4076. executorch/include/executorch/runtime/platform/compat_unistd.h +75 -0
  4077. executorch/include/executorch/runtime/platform/compiler.h +225 -0
  4078. executorch/include/executorch/runtime/platform/log.h +216 -0
  4079. executorch/include/executorch/runtime/platform/platform.h +276 -0
  4080. executorch/include/executorch/runtime/platform/profiler.h +293 -0
  4081. executorch/include/executorch/runtime/platform/runtime.h +35 -0
  4082. executorch/include/executorch/runtime/platform/system.h +49 -0
  4083. executorch/include/executorch/runtime/platform/test/pal_spy.h +86 -0
  4084. executorch/include/executorch/runtime/platform/test/stub_platform.h +76 -0
  4085. executorch/include/executorch/runtime/platform/types.h +24 -0
  4086. executorch/kernels/portable/__init__.py +26 -0
  4087. executorch/kernels/portable/custom_ops.yaml +32 -0
  4088. executorch/kernels/portable/et_copy_ops.yaml +16 -0
  4089. executorch/kernels/portable/functions.yaml +1071 -0
  4090. executorch/kernels/portable/test/op_grid_sampler_2d_test.py +234 -0
  4091. executorch/kernels/portable/test/op_upsample_bilinear2d_aa_test.py +356 -0
  4092. executorch/kernels/portable/test/op_upsample_bilinear2d_test.py +87 -0
  4093. executorch/kernels/portable/test/op_upsample_nearest2d_test.py +71 -0
  4094. executorch/kernels/portable/test/supported_features_def.yaml +1 -0
  4095. executorch/kernels/portable/test/test_grid_sampler_2d_executorch.py +458 -0
  4096. executorch/kernels/quantized/__init__.py +21 -0
  4097. executorch/kernels/quantized/quantized.yaml +113 -0
  4098. executorch/kernels/quantized/quantized_ops_aot_lib.dll +0 -0
  4099. executorch/kernels/quantized/test/supported_features_def.yaml +1 -0
  4100. executorch/kernels/quantized/test/test_out_variants.py +100 -0
  4101. executorch/kernels/quantized/test/test_quant_dequant_per_token.py +149 -0
  4102. executorch/runtime/__init__.py +334 -0
  4103. executorch/runtime/kernel/test/functions.yaml +13 -0
  4104. executorch/runtime/test/test_runtime.py +106 -0
  4105. executorch/runtime/test/test_runtime_etdump_gen.py +163 -0
  4106. executorch/runtime/test/test_runtime_xnnpack.py +181 -0
  4107. executorch/schema/program.fbs +534 -0
  4108. executorch/schema/scalar_type.fbs +43 -0
  4109. executorch/schema/test/test_schema.py +52 -0
  4110. executorch/share/cmake/executorch-config.cmake +78 -0
  4111. executorch/util/activation_memory_profiler.py +177 -0
  4112. executorch/util/collect_env.py +748 -0
  4113. executorch/util/python_profiler.py +96 -0
  4114. executorch/version.py +4 -0
  4115. executorch-1.4.0.dist-info/METADATA +109 -0
  4116. executorch-1.4.0.dist-info/RECORD +4120 -0
  4117. executorch-1.4.0.dist-info/WHEEL +5 -0
  4118. executorch-1.4.0.dist-info/entry_points.txt +2 -0
  4119. executorch-1.4.0.dist-info/licenses/LICENSE +37 -0
  4120. executorch-1.4.0.dist-info/top_level.txt +9 -0
@@ -0,0 +1,2646 @@
1
+ # Copyright (c) Meta Platforms, Inc. and affiliates.
2
+ # All rights reserved.
3
+ # Copyright 2025 Arm Limited and/or its affiliates.
4
+ #
5
+ # This source code is licensed under the BSD-style license found in the
6
+ # LICENSE file in the root directory of this source tree.
7
+
8
+
9
+ # This file contains all the functions that replace one op with another in the
10
+ # graph.
11
+
12
+ # pyre-unsafe
13
+
14
+ import logging
15
+ import math
16
+ import operator
17
+ from operator import neg
18
+ from typing import cast, Dict, Optional, Sequence
19
+
20
+ import torch
21
+ import torch.fx
22
+ from executorch.backends.cadence.aot.compiler_utils import quantize_tensor_multiplier
23
+ from executorch.backends.cadence.aot.fuse_ops import FuseCascadedTransposeOrPermuteOps
24
+ from executorch.backends.cadence.aot.pass_utils import (
25
+ CadencePassAttribute,
26
+ get_arg,
27
+ register_cadence_pass,
28
+ RemoveOrReplacePassInterface,
29
+ )
30
+ from executorch.backends.cadence.aot.utils import is_depthwise_conv
31
+ from executorch.backends.transforms.replace_nop_transpose_or_permute_with_view import (
32
+ ReplaceNopTransposeOrPermuteWithViewPass as _SharedReplaceNopTransposeOrPermuteWithViewPass,
33
+ )
34
+ from executorch.backends.transforms.replace_scalar_with_tensor import (
35
+ ReplaceScalarWithTensorArgPass,
36
+ )
37
+ from executorch.exir.dialects._ops import ops as exir_ops
38
+ from executorch.exir.dialects.edge._ops import EdgeOpOverload
39
+ from executorch.exir.pass_base import ExportPass, PassResult
40
+
41
+ # A map to represent ops that:
42
+ # (a) are functionally equivalent; and
43
+ # (b) have identical arguments
44
+ # An op whose target is 'key' in this dict can be replaced by the functionally euivalent
45
+ # op whose target is 'value'. The replacement would just involve changing the op target.
46
+ functionally_equivalent_op_targets: Dict[EdgeOpOverload, EdgeOpOverload] = {
47
+ exir_ops.edge.aten.relu_.default: exir_ops.edge.aten.relu.default,
48
+ exir_ops.edge.aten.unsafe_split.Tensor: exir_ops.edge.aten.split_copy.Tensor,
49
+ }
50
+
51
+
52
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
53
+ class ReplaceLogicalNotBooleanWhereWithWherePass(RemoveOrReplacePassInterface):
54
+ """
55
+ A where op with a logical_not and a boolean tensor can be replaced
56
+ by a where op with flipped inputs and the initial boolean tensor.
57
+ """
58
+
59
+ @property
60
+ def targets(self) -> list[EdgeOpOverload]:
61
+ return [exir_ops.edge.aten.where.self]
62
+
63
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
64
+ # If the first arg is not a logical_not, bail.
65
+ if not isinstance(node.args[0], torch.fx.Node):
66
+ return False
67
+
68
+ logical_not_node = cast(torch.fx.Node, node.args[0])
69
+ if logical_not_node.target != exir_ops.edge.aten.logical_not.default:
70
+ return False
71
+
72
+ # Get the first arg node and its input
73
+ if not isinstance(logical_not_node.args[0], torch.fx.Node):
74
+ return False
75
+
76
+ logical_not_input_node = cast(torch.fx.Node, logical_not_node.args[0])
77
+
78
+ # If the logical_not input is not a boolean tensor, bail.
79
+ if logical_not_input_node.meta["val"].dtype != torch.bool:
80
+ return False
81
+
82
+ # Replace the where op with another one, flipping the inputs and using the boolean
83
+ # tensor from logical_not.
84
+ with node.graph.inserting_before(node):
85
+ new_node = node.graph.call_function(
86
+ exir_ops.edge.aten.where.self,
87
+ args=(logical_not_input_node, node.args[2], node.args[1]),
88
+ )
89
+ new_node.meta = node.meta
90
+ # Replace all the uses
91
+ node.replace_all_uses_with(new_node)
92
+ return True
93
+
94
+
95
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
96
+ class ReplaceSafeSoftmaxWithSoftmax(RemoveOrReplacePassInterface): # keep
97
+ """
98
+ Replace _safe_softmax with _softmax
99
+ """
100
+
101
+ @property
102
+ def targets(self) -> list[EdgeOpOverload]:
103
+ return [torch.ops.aten._safe_softmax.default]
104
+
105
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
106
+ # Add False for the half_to_float argument of softmax
107
+ softmax_args = tuple(list(node.args) + [False])
108
+
109
+ with node.graph.inserting_before(node):
110
+ new_node = node.graph.call_function(
111
+ torch.ops.aten._softmax.default,
112
+ args=softmax_args,
113
+ kwargs=node.kwargs,
114
+ )
115
+ new_node.meta = node.meta
116
+ node.replace_all_uses_with(new_node)
117
+ return True
118
+
119
+
120
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
121
+ class ReplacePT2QuantWithCadenceQuantPass(RemoveOrReplacePassInterface):
122
+ """
123
+ Replace the pt2 quantization ops with cadence quantization ops.
124
+ We do not link kernels to the PT2 quantization ops, so we need to
125
+ replace them with cadence ops at all optimization levels.
126
+ """
127
+
128
+ @property
129
+ def targets(self) -> list[EdgeOpOverload]:
130
+ return [
131
+ torch.ops.quantized_decomposed.quantize_per_tensor.default,
132
+ exir_ops.edge.quantized_decomposed.quantize_per_tensor.default,
133
+ ]
134
+
135
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
136
+ ns = exir_ops.edge if isinstance(node.target, EdgeOpOverload) else torch.ops
137
+ with node.graph.inserting_before(node):
138
+ new_node = node.graph.call_function(
139
+ ns.cadence.quantize_per_tensor.default,
140
+ args=node.args,
141
+ kwargs=node.kwargs,
142
+ )
143
+ new_node.meta = node.meta
144
+ node.replace_all_uses_with(new_node)
145
+ return True
146
+
147
+
148
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
149
+ class ReplacePT2DequantWithCadenceDequantPass(RemoveOrReplacePassInterface):
150
+ """
151
+ Replace the pt2 dequantization ops with cadence dequantization ops.
152
+ We do not link kernels to the PT2 quantization ops, so we need to
153
+ replace them with cadence ops at all optimization levels.
154
+ """
155
+
156
+ @property
157
+ def targets(self) -> list[EdgeOpOverload]:
158
+ return [
159
+ torch.ops.quantized_decomposed.dequantize_per_tensor.default,
160
+ exir_ops.edge.quantized_decomposed.dequantize_per_tensor.default,
161
+ ]
162
+
163
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
164
+ ns = exir_ops.edge if isinstance(node.target, EdgeOpOverload) else torch.ops
165
+ out_dtype = node.kwargs.get("out_dtype")
166
+ kwargs = {k: v for k, v in node.kwargs.items() if k != "out_dtype"}
167
+ with node.graph.inserting_before(node):
168
+ new_node = node.graph.call_function(
169
+ ns.cadence.dequantize_per_tensor.default,
170
+ args=node.args,
171
+ kwargs=kwargs,
172
+ )
173
+ new_node.meta = node.meta.copy()
174
+ if (
175
+ out_dtype is not None
176
+ and out_dtype != torch.float32
177
+ and "val" in new_node.meta
178
+ ):
179
+ new_node.meta["val"] = new_node.meta["val"].to(torch.float32)
180
+ if out_dtype is not None and out_dtype != torch.float32:
181
+ with node.graph.inserting_after(new_node):
182
+ cast_node = node.graph.call_function(
183
+ ns.aten.to.dtype,
184
+ args=(new_node, out_dtype),
185
+ )
186
+ cast_node.meta = node.meta.copy()
187
+ node.replace_all_uses_with(cast_node)
188
+ else:
189
+ node.replace_all_uses_with(new_node)
190
+ return True
191
+
192
+
193
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
194
+ class ReplaceSqueezeAndUnsqueezeWithViewPass(RemoveOrReplacePassInterface):
195
+ """
196
+ When the shape is static, replace squeeze_copy and unsqueeze_copy ops with
197
+ view_copy op
198
+ """
199
+
200
+ @property
201
+ def targets(self) -> list[EdgeOpOverload]:
202
+ return [
203
+ exir_ops.edge.aten.squeeze_copy.default,
204
+ exir_ops.edge.aten.squeeze_copy.dim,
205
+ exir_ops.edge.aten.squeeze_copy.dims,
206
+ exir_ops.edge.aten.unsqueeze_copy.default,
207
+ ]
208
+
209
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
210
+ # Get the output tensor shape
211
+ out_shape = node.meta["val"].shape
212
+
213
+ # Bail out if any dim is not an int (dynamic shape)
214
+ for dim in list(out_shape):
215
+ if not isinstance(dim, int):
216
+ return False
217
+
218
+ # Replace with view op with the new shape
219
+ with node.graph.inserting_before(node):
220
+ new_node = node.graph.call_function(
221
+ exir_ops.edge.aten.view_copy.default,
222
+ args=(node.args[0], list(out_shape)),
223
+ )
224
+ # Do not remove the metadata copy!
225
+ new_node.meta = node.meta
226
+ node.replace_all_uses_with(new_node)
227
+ return True
228
+
229
+
230
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
231
+ class ReplaceFunctionallyEquivalentOpTargets(RemoveOrReplacePassInterface):
232
+ """
233
+ Replace an op with a functionally equivalent op by just switching the op
234
+ target, but without incurring any change to the op args.
235
+ """
236
+
237
+ @property
238
+ def targets(self) -> list[EdgeOpOverload]:
239
+ return list(functionally_equivalent_op_targets.keys())
240
+
241
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
242
+ assert isinstance(node.target, EdgeOpOverload)
243
+ target_op = functionally_equivalent_op_targets[node.target]
244
+ with node.graph.inserting_before(node):
245
+ new_node = node.graph.call_function(
246
+ target_op,
247
+ args=node.args,
248
+ kwargs=node.kwargs,
249
+ )
250
+ new_node.meta = node.meta
251
+ node.replace_all_uses_with(new_node)
252
+
253
+ # RemoveOrReplacePassInterface calls eliminate_dead_code, but this doesn't
254
+ # remove impure nodes (nodes which have side effects). Not sure if that is
255
+ # generally safe, so instead of modifying the interface, just erasing
256
+ # these nodes for this pass.
257
+ node.graph.erase_node(node)
258
+ return True
259
+
260
+
261
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
262
+ class ReplaceSelectWithViewOpPass(RemoveOrReplacePassInterface):
263
+ """
264
+ If the size along the select dim is 1, then the select op can be replaced
265
+ by view op.
266
+ """
267
+
268
+ @property
269
+ def targets(self) -> list[EdgeOpOverload]:
270
+ return [exir_ops.edge.aten.select_copy.int]
271
+
272
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
273
+ # Get the input tensor and shapes
274
+ in_tensor_node = node.args[0]
275
+ assert isinstance(in_tensor_node, torch.fx.Node)
276
+ in_shape = in_tensor_node.meta["val"].shape
277
+ out_shape = node.meta["val"].shape
278
+
279
+ # Get the select dimension
280
+ select_dim = node.args[1]
281
+ assert isinstance(select_dim, int)
282
+ select_dim = select_dim if select_dim >= 0 else select_dim + len(in_shape)
283
+
284
+ if in_shape[select_dim] == 1:
285
+ # Replace with view op with the new shape
286
+ with node.graph.inserting_before(node):
287
+ new_node = node.graph.call_function(
288
+ exir_ops.edge.aten.view_copy.default,
289
+ args=(node.args[0], list(out_shape)),
290
+ )
291
+ # Important to copy metadata
292
+ new_node.meta = node.meta
293
+ node.replace_all_uses_with(new_node)
294
+ return True
295
+
296
+ return False
297
+
298
+
299
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
300
+ class ReplaceMMWithAddMMPass(RemoveOrReplacePassInterface):
301
+ """
302
+ This pass replaces mm with addmm by introducing a zero bias.
303
+ mm is not supported, so this is an opt_level=0 pass.
304
+ """
305
+
306
+ @property
307
+ def targets(self) -> list[EdgeOpOverload]:
308
+ return [exir_ops.edge.aten.mm.default]
309
+
310
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
311
+ # The mm op has two args: input, mat2
312
+ assert len(node.args) == 2
313
+ X, mat2 = node.args
314
+ assert isinstance(X, torch.fx.Node)
315
+ assert isinstance(mat2, torch.fx.Node)
316
+
317
+ # Create a zero bias tensor, and insert it as a graph buffer before the
318
+ # current node
319
+ mat2_tensor = mat2.meta["val"]
320
+ bias_size = mat2_tensor.size(1)
321
+
322
+ with node.graph.inserting_before(node):
323
+ zero_bias = node.graph.call_function(
324
+ exir_ops.edge.aten.full.default,
325
+ args=([bias_size], 0.0),
326
+ kwargs={"dtype": torch.float32},
327
+ )
328
+ zero_bias.meta = node.meta
329
+
330
+ # Replace mm with addmm
331
+ new_args = (zero_bias, X, mat2)
332
+ with node.graph.inserting_before(node):
333
+ new_node = node.graph.call_function(
334
+ exir_ops.edge.aten.addmm.default,
335
+ args=new_args,
336
+ )
337
+ new_node.meta = node.meta
338
+
339
+ node.replace_all_uses_with(new_node)
340
+ return True
341
+
342
+
343
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
344
+ class ReplaceAddMMWithLinearPass(RemoveOrReplacePassInterface):
345
+ """
346
+ This pass replaces addmm with linear op.
347
+
348
+ AddMM computes: beta*bias + alpha*mm(mat1, mat2)
349
+ Linear computes: mat1 @ weight.T + bias
350
+
351
+ """
352
+
353
+ @property
354
+ def targets(self) -> list[EdgeOpOverload]:
355
+ return [exir_ops.edge.aten.addmm.default]
356
+
357
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
358
+ # The addmm op has three concrete args: bias, mat1, mat2
359
+ assert len(node.args) >= 3
360
+ (bias, mat1, mat2) = node.args[0:3]
361
+
362
+ # The other two args are optional scale args
363
+ beta = float(node.kwargs.get("beta", 1.0))
364
+ alpha = float(node.kwargs.get("alpha", 1.0))
365
+
366
+ bias, mat1, mat2 = cast(
367
+ tuple[torch.fx.Node, torch.fx.Node, torch.fx.Node],
368
+ (bias, mat1, mat2),
369
+ )
370
+
371
+ graph = node.graph
372
+
373
+ fit_bias = beta == 1.0
374
+ fit_mat2 = False
375
+
376
+ # Handle transpose: if mat2 is a transpose op, extract the original tensor
377
+ transposed_mat2 = False
378
+ if (
379
+ mat2.op == "call_function"
380
+ and mat2.target == exir_ops.edge.aten.transpose_copy.int
381
+ ):
382
+ # mat2 is already transposed, so we use the input to the transpose
383
+ mat2 = cast(torch.fx.Node, mat2.args[0])
384
+ transposed_mat2 = True
385
+ fit_mat2 = alpha == 1.0
386
+
387
+ if not (fit_bias and fit_mat2):
388
+ return False
389
+
390
+ # Multiply bias by beta if needed
391
+ if beta != 1.0:
392
+ # Create a scaled bias using element-wise multiplication in the graph
393
+ with graph.inserting_before(node):
394
+ beta_scalar = graph.call_function(
395
+ exir_ops.edge.aten.full.default,
396
+ args=([1], beta),
397
+ kwargs={"dtype": torch.float32},
398
+ )
399
+ beta_scalar.meta = node.meta
400
+ bias = graph.call_function(
401
+ exir_ops.edge.aten.mul.Tensor,
402
+ args=(bias, beta_scalar),
403
+ )
404
+
405
+ # Metadata copy important
406
+ bias.meta = node.meta
407
+
408
+ # Multiply mat2 by alpha if needed
409
+ if alpha != 1.0:
410
+ with graph.inserting_before(node):
411
+ alpha_scalar = graph.call_function(
412
+ exir_ops.edge.aten.full.default,
413
+ args=([1], alpha),
414
+ kwargs={"dtype": torch.float32},
415
+ )
416
+ alpha_scalar.meta = node.meta
417
+ mat2 = graph.call_function(
418
+ exir_ops.edge.aten.mul.Tensor,
419
+ args=(mat2, alpha_scalar),
420
+ )
421
+
422
+ # Metadata copy important
423
+ mat2.meta = node.meta
424
+
425
+ # Transpose mat2 if it wasn't already transposed
426
+ if not transposed_mat2:
427
+ with graph.inserting_before(node):
428
+ mat2 = graph.call_function(
429
+ exir_ops.edge.aten.transpose_copy.int,
430
+ args=(mat2, -1, -2),
431
+ )
432
+
433
+ # Metadata copy important
434
+ mat2.meta = node.meta
435
+
436
+ # Construct the linear node: linear(input, weight, bias)
437
+ # linear computes: input @ weight.T + bias
438
+ linear_args = (mat1, mat2, bias)
439
+ with graph.inserting_before(node):
440
+ linear_node = graph.call_function(
441
+ exir_ops.edge.aten.linear.default,
442
+ args=linear_args,
443
+ )
444
+
445
+ # Metadata copy important
446
+ linear_node.meta = node.meta
447
+
448
+ # Replace all uses of the addmm op with linear op
449
+ node.replace_all_uses_with(linear_node)
450
+ return True
451
+
452
+
453
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
454
+ class ReplacePermuteWithTransposePass(RemoveOrReplacePassInterface):
455
+ """
456
+ Replace permute op with transpose if the permutation is only along
457
+ two dimensions.
458
+ """
459
+
460
+ @property
461
+ def targets(self) -> list[EdgeOpOverload]:
462
+ return [exir_ops.edge.aten.permute_copy.default]
463
+
464
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
465
+ # Get the old dim and new dim order
466
+ in_tensor = node.args[0]
467
+ assert isinstance(in_tensor, torch.fx.Node)
468
+ in_shape = in_tensor.meta["val"].shape
469
+ old_dims = tuple(range(len(in_shape)))
470
+ new_dims = cast(Sequence[int], node.args[1])
471
+
472
+ # Compute the number of positions in which the old and new order differ
473
+ diff = [od for od, nd in zip(old_dims, new_dims) if od != nd]
474
+
475
+ # If the difference is zero, replace with identity (just the input)
476
+ if len(diff) == 0:
477
+ node.replace_all_uses_with(in_tensor)
478
+ return True
479
+
480
+ # If the difference is in two dimensions, we can replace this permute op
481
+ # with transpose op.
482
+ if len(diff) == 2:
483
+ with node.graph.inserting_before(node):
484
+ new_node = node.graph.call_function(
485
+ exir_ops.edge.aten.transpose_copy.int,
486
+ args=(node.args[0], diff[0], diff[1]),
487
+ )
488
+ new_node.meta = node.meta
489
+ node.replace_all_uses_with(new_node)
490
+ return True
491
+
492
+ return False
493
+
494
+
495
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
496
+ class ReplaceConvolutionOptionalArgsWithConcreteArgsPass(RemoveOrReplacePassInterface):
497
+ """
498
+ Replace optional tensors with concrete tensors. Currently, we
499
+ replace the optional bias tensor with a zero tensor.
500
+ """
501
+
502
+ @property
503
+ def targets(self) -> list[EdgeOpOverload]:
504
+ return [
505
+ exir_ops.edge.cadence.conv1d.default,
506
+ exir_ops.edge.cadence.conv2d.default,
507
+ exir_ops.edge.cadence.conv3d.default,
508
+ exir_ops.edge.cadence.transposed_convolution.default,
509
+ ]
510
+
511
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
512
+ # Check if this is a transposed convolution
513
+ assert isinstance(node.target, EdgeOpOverload)
514
+ is_transposed = (
515
+ node.target == exir_ops.edge.cadence.transposed_convolution.default
516
+ )
517
+ num_expected_args = 9 if is_transposed else 7
518
+ assert len(node.args) == num_expected_args
519
+ # Check if the bias is concrete
520
+ if node.args[2] is not None:
521
+ return False
522
+
523
+ # The bias length is the number of out channels.
524
+ out_shape = node.meta["val"].shape
525
+ bias_size = out_shape[1]
526
+
527
+ # Create a zero bias tensor
528
+ with node.graph.inserting_before(node):
529
+ zero_bias = node.graph.call_function(
530
+ exir_ops.edge.aten.full.default,
531
+ args=([bias_size], 0.0),
532
+ kwargs={"dtype": torch.float32},
533
+ )
534
+ # Create proper metadata for the zero_bias node
535
+ zero_bias.meta = node.meta
536
+ new_args = list(node.args)
537
+ new_args[2] = zero_bias
538
+ new_args = tuple(new_args)
539
+
540
+ new_node = node.graph.call_function(
541
+ # pyre-ignore[6]: Target is a call func, but type is union call func and str
542
+ node.target,
543
+ args=new_args,
544
+ kwargs=node.kwargs,
545
+ )
546
+ new_node.meta = node.meta
547
+
548
+ node.replace_all_uses_with(new_node)
549
+ return True
550
+
551
+
552
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
553
+ class ReplaceRepeatWithCatPass(RemoveOrReplacePassInterface):
554
+ """
555
+ Replace repeat op as successive cat ops along different dimensions.
556
+ repeat is not supported, so this is an opt_level=0 pass.
557
+ """
558
+
559
+ @property
560
+ def targets(self) -> list[EdgeOpOverload]:
561
+ return [exir_ops.edge.aten.repeat.default]
562
+
563
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
564
+ # Extract the input tensor, and the repeats from the args
565
+ in_tensor = node.args[0]
566
+ assert isinstance(in_tensor, torch.fx.Node)
567
+ repeats = cast(Sequence[int], node.args[1])
568
+
569
+ # Glean the shapes of input tensor
570
+ in_shape = list(in_tensor.meta["val"].shape)
571
+
572
+ # If the size of repeats is more than the dimensionality of the tensor,
573
+ # the output of repeat will be a higher-dimensional tensor. We reshape
574
+ # the input so that it has the same dimensionality as the output tensor.
575
+ diff = len(repeats) - len(in_shape)
576
+ assert (
577
+ diff >= 0
578
+ ), "Repeat arg malformed: expected a repeat along each dimension of input tensor"
579
+
580
+ graph = node.graph
581
+ result_node = in_tensor
582
+
583
+ if diff > 0:
584
+ # Extend the input shape with 1's along the higher dimensions
585
+ in_shape = ([1] * diff) + in_shape
586
+ # Insert a view op that reshapes the input tensor to have same
587
+ # dimensionality as the output tensor.
588
+ with graph.inserting_before(node):
589
+ result_node = graph.call_function(
590
+ exir_ops.edge.aten.view_copy.default,
591
+ args=(in_tensor, in_shape),
592
+ )
593
+ result_node.meta = node.meta
594
+ assert len(repeats) == len(in_shape)
595
+
596
+ # Repeat op is nothing but successive cat ops along each dimension.
597
+ for dim, repeat in reversed(list(enumerate(repeats))):
598
+ # We do not need to do anything if repeat factor is 1
599
+ if repeat == 1:
600
+ continue
601
+ cat_arg = [result_node] * repeat
602
+ with graph.inserting_before(node):
603
+ result_node = graph.call_function(
604
+ exir_ops.edge.aten.cat.default, args=(cat_arg, dim)
605
+ )
606
+ result_node.meta = node.meta
607
+
608
+ node.replace_all_uses_with(result_node)
609
+ return True
610
+
611
+
612
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
613
+ class ReplacePadWithCatPass(RemoveOrReplacePassInterface):
614
+ """
615
+ Replace constant pad nd op that does padding on outer-most dimension
616
+ with Cat(left_padding_constant_tensor, X, right_padding_constant_tensor)
617
+ """
618
+
619
+ @property
620
+ def targets(self) -> list[EdgeOpOverload]:
621
+ return [exir_ops.edge.aten.constant_pad_nd.default]
622
+
623
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
624
+ assert len(node.args) >= 2
625
+ input_node, orig_padding = node.args[:2]
626
+ assert isinstance(input_node, torch.fx.Node)
627
+
628
+ # if there is no padding, this op will be treated in removal pass.
629
+ if not orig_padding:
630
+ return False
631
+
632
+ value = 0 if len(node.args) == 2 else node.args[2]
633
+
634
+ arg_shape = input_node.meta["val"].shape
635
+ dtype = input_node.meta["val"].dtype
636
+
637
+ # Convert orig_padding to a list for manipulation
638
+ # pyre-ignore[6]: Argument type
639
+ padding_list = list(orig_padding)
640
+ padding = padding_list + ([0] * (len(padding_list) % 2 != 0))
641
+ assert len(padding) >= 2
642
+ (left_padding_size, right_padding_size) = padding[-2:]
643
+ # Replace only if constant_pad_nd is along the innermost padding dimension.
644
+ if (
645
+ any(x != 0 for x in padding[0:-2])
646
+ or left_padding_size < 0
647
+ or right_padding_size < 0
648
+ ):
649
+ return False
650
+
651
+ cat_tensors = []
652
+ dim = len(arg_shape) - len(padding) // 2
653
+ graph = node.graph
654
+
655
+ # add left_padding
656
+ if left_padding_size > 0:
657
+ left_padding_shape = (
658
+ arg_shape[:dim] + (left_padding_size,) + arg_shape[dim + 1 :]
659
+ )
660
+ with graph.inserting_before(node):
661
+ left_padding_node = graph.call_function(
662
+ exir_ops.edge.aten.full.default,
663
+ args=(
664
+ left_padding_shape,
665
+ value,
666
+ ),
667
+ kwargs={"dtype": dtype},
668
+ )
669
+ left_padding_node.meta = node.meta
670
+ cat_tensors.append(left_padding_node)
671
+
672
+ # input_node
673
+ cat_tensors.append(input_node)
674
+
675
+ # right_padding
676
+ if right_padding_size > 0:
677
+ right_padding_shape = (
678
+ arg_shape[:dim] + (right_padding_size,) + arg_shape[dim + 1 :]
679
+ )
680
+ with graph.inserting_before(node):
681
+ right_padding_node = graph.call_function(
682
+ exir_ops.edge.aten.full.default,
683
+ args=(
684
+ right_padding_shape,
685
+ value,
686
+ ),
687
+ kwargs={"dtype": dtype},
688
+ )
689
+ right_padding_node.meta = node.meta
690
+ cat_tensors.append(right_padding_node)
691
+
692
+ assert len(cat_tensors) == 1 + (left_padding_size > 0) + (
693
+ right_padding_size > 0
694
+ )
695
+
696
+ new_args = (cat_tensors, dim)
697
+ with graph.inserting_before(node):
698
+ new_node = graph.call_function(
699
+ exir_ops.edge.aten.cat.default,
700
+ args=new_args,
701
+ )
702
+ new_node.meta = node.meta
703
+
704
+ node.replace_all_uses_with(new_node)
705
+ return True
706
+
707
+
708
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
709
+ class ReplaceConstantPadNdWithSlicePass(RemoveOrReplacePassInterface):
710
+ """
711
+ Replace constant pad nd op that does padding on outer-most dimension
712
+ with exir_ops slice(left_padding_constant_tensor, X, right_padding_constant_tensor)
713
+ """
714
+
715
+ @property
716
+ def targets(self) -> list[EdgeOpOverload]:
717
+ return [exir_ops.edge.aten.constant_pad_nd.default]
718
+
719
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
720
+ assert len(node.args) >= 2
721
+ input_node = node.args[0]
722
+ orig_padding = cast(Sequence[int], node.args[1])
723
+ assert isinstance(input_node, torch.fx.Node)
724
+
725
+ # if there is no padding, this op will be treated in removal pass.
726
+ if not orig_padding:
727
+ return False
728
+
729
+ padding = list(orig_padding) + ([0] * (len(orig_padding) % 2 != 0))
730
+ assert len(padding) >= 2
731
+
732
+ # pyre-ignore[6]
733
+ (start, diff) = map(neg, padding[-2:])
734
+ # Replace only if constant_pad_nd is along the innermost padding dimension.
735
+ if any(x != 0 for x in padding[0:-2]) or start < 0 or diff < 0:
736
+ return False
737
+
738
+ arg_shape = input_node.meta["val"].shape
739
+ dim = len(arg_shape) - len(padding) // 2
740
+ stop = arg_shape[dim] - diff
741
+ assert start <= stop
742
+
743
+ with node.graph.inserting_before(node):
744
+ new_node = node.graph.call_function(
745
+ exir_ops.edge.aten.slice.Tensor,
746
+ args=(input_node, dim, start, stop),
747
+ )
748
+ new_node.meta = node.meta
749
+ node.replace_all_uses_with(new_node)
750
+ return True
751
+
752
+
753
+ # Make that pass runnable standalone at opt level 0.
754
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
755
+ class ReplaceAtenConvolutionWithCadenceConvolutionPass(RemoveOrReplacePassInterface):
756
+ """
757
+ Replace aten convolution op with jarvis-specific convolution op, since the
758
+ aten version is not supported by jarvis.
759
+ Also remove convolution stride if the output size along the strided dimension
760
+ is 1. We can enable more transformations (e.g., conv -> linear replacement)
761
+ for unit-stride convolutions.
762
+ """
763
+
764
+ @property
765
+ def targets(self) -> list[EdgeOpOverload]:
766
+ return [exir_ops.edge.aten.convolution.default]
767
+
768
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
769
+ # There must be 9 total args.
770
+ if len(node.args) != 9:
771
+ return False
772
+
773
+ # Unpack the args
774
+ (
775
+ in_tensor,
776
+ weight,
777
+ bias,
778
+ stride,
779
+ padding,
780
+ dilation,
781
+ transposed,
782
+ output_padding,
783
+ groups,
784
+ ) = node.args
785
+
786
+ # Cast to appropriate types
787
+ stride = cast(Sequence[int], stride)
788
+ padding = cast(Sequence[int], padding)
789
+ dilation = cast(Sequence[int], dilation)
790
+ output_padding = cast(Sequence[int], output_padding)
791
+
792
+ # Currently we only handle conversion to conv1d, conv2d, and conv3d, therefore
793
+ # verify that the stride, padding, dilation, and output_padding have
794
+ # len <=3.
795
+ if not (
796
+ (len(stride) == len(padding) == len(dilation) == len(output_padding) == 1)
797
+ or (
798
+ len(stride) == len(padding) == len(dilation) == len(output_padding) == 2
799
+ )
800
+ or (
801
+ len(stride) == len(padding) == len(dilation) == len(output_padding) == 3
802
+ )
803
+ ):
804
+ return False
805
+
806
+ # Determine if this is 1D, 2D, or 3D convolution based on parameter lengths
807
+ if transposed:
808
+ target = exir_ops.edge.cadence.transposed_convolution.default
809
+ elif len(stride) == 1:
810
+ target = exir_ops.edge.cadence.conv1d.default
811
+ elif len(stride) == 2:
812
+ target = exir_ops.edge.cadence.conv2d.default
813
+ else: # len(stride) == 3
814
+ target = exir_ops.edge.cadence.conv3d.default
815
+
816
+ with node.graph.inserting_before(node):
817
+ if transposed:
818
+ # Flip the height and width dimensions of weight, since we apply a
819
+ # gather stencil. Also, the first two dimensions of weight must be
820
+ # transposed/interchanged.
821
+ assert isinstance(weight, torch.fx.Node)
822
+ transposed_weight = node.graph.call_function(
823
+ exir_ops.edge.aten.transpose_copy.int,
824
+ args=(weight, 0, 1),
825
+ )
826
+ transposed_weight.meta = weight.meta
827
+
828
+ # Get the dimension for flip based on weight shape
829
+ weight_dim = len(weight.meta["val"].shape)
830
+ flip_dims = [-1] if weight_dim == 3 else [-1, -2]
831
+
832
+ flipped_weight = node.graph.call_function(
833
+ exir_ops.edge.aten.flip.default,
834
+ args=(transposed_weight, flip_dims),
835
+ )
836
+ flipped_weight.meta = transposed_weight.meta
837
+
838
+ new_args = (
839
+ in_tensor,
840
+ flipped_weight,
841
+ bias,
842
+ stride,
843
+ padding,
844
+ dilation,
845
+ output_padding,
846
+ groups,
847
+ False,
848
+ )
849
+ else:
850
+ # Verify that output_padding is 0.
851
+ if not all(x == 0 for x in output_padding):
852
+ return False
853
+
854
+ # Keep the original stride to maintain correct output dimensions
855
+ new_stride = stride
856
+
857
+ new_args = (
858
+ in_tensor,
859
+ weight,
860
+ bias,
861
+ new_stride,
862
+ padding,
863
+ dilation,
864
+ groups,
865
+ )
866
+
867
+ new_node = node.graph.call_function(target, args=new_args)
868
+ new_node.meta = node.meta
869
+
870
+ node.replace_all_uses_with(new_node)
871
+ return True
872
+
873
+
874
+ @register_cadence_pass(CadencePassAttribute(opt_level=2))
875
+ class ReplaceTrivialConvWithLinear(RemoveOrReplacePassInterface):
876
+ """
877
+ In nn.Conv1d, the operand shapes are:
878
+ input - [batch, in_channels, in_length]
879
+ weight - [out_channels, in_channels, weight_length]
880
+ output - [batch, out_channels, out_length]
881
+ When in_length == weight_length, out_length = 1. In this scenario, we can
882
+ view the input as a tensor shaped [batch, K], and weight as a tensor
883
+ shaped [out_channels, K], and replace nn.Conv1d with nn.Linear. This
884
+ optimization can be extended to nn.Conv2d as well, where in_length is a 2d
885
+ image, and weight_length can be replaced with a 2d filter the same shape as
886
+ the image.
887
+ """
888
+
889
+ trivial_conv_op_to_linear_op: Dict[EdgeOpOverload, EdgeOpOverload] = {
890
+ exir_ops.edge.cadence.conv1d.default: exir_ops.edge.aten.linear.default,
891
+ exir_ops.edge.cadence.conv2d.default: exir_ops.edge.aten.linear.default,
892
+ exir_ops.edge.cadence.conv3d.default: exir_ops.edge.aten.linear.default,
893
+ exir_ops.edge.cadence.quantized_conv1d_ncl.per_tensor: exir_ops.edge.cadence.quantized_linear.per_tensor,
894
+ exir_ops.edge.cadence.quantized_conv1d_nlc.per_tensor: exir_ops.edge.cadence.quantized_linear.per_tensor,
895
+ exir_ops.edge.cadence.quantized_conv2d_nchw.per_tensor: exir_ops.edge.cadence.quantized_linear.per_tensor,
896
+ exir_ops.edge.cadence.quantized_conv2d_nhwc.per_tensor: exir_ops.edge.cadence.quantized_linear.per_tensor,
897
+ }
898
+
899
+ quantized_conv_ops: frozenset[EdgeOpOverload] = frozenset(
900
+ {
901
+ exir_ops.edge.cadence.quantized_conv1d_ncl.per_tensor,
902
+ exir_ops.edge.cadence.quantized_conv1d_nlc.per_tensor,
903
+ exir_ops.edge.cadence.quantized_conv2d_nchw.per_tensor,
904
+ exir_ops.edge.cadence.quantized_conv2d_nhwc.per_tensor,
905
+ }
906
+ )
907
+
908
+ @property
909
+ def targets(self) -> list[EdgeOpOverload]:
910
+ return list(self.trivial_conv_op_to_linear_op.keys())
911
+
912
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
913
+ # Parse the necessary args of the convolution node. Both convolution
914
+ # and quantized_conv have the same first 8 args. The quantized op has
915
+ # extra args holding at least the zero point and scale of input, weight, bias,
916
+ # and output tensor.
917
+ assert isinstance(node.target, EdgeOpOverload)
918
+ quantized_op = node.target in self.quantized_conv_ops
919
+ assert (len(node.args) == 7 and not quantized_op) or (
920
+ len(node.args) >= 12 and quantized_op
921
+ ), "Inconsistent args for convolution"
922
+ (in_tensor, weight, bias, _, padding, dilation, groups) = node.args[0:7]
923
+
924
+ assert isinstance(in_tensor, torch.fx.Node)
925
+ assert isinstance(weight, torch.fx.Node)
926
+
927
+ # Glean the shapes of input, weight, and output
928
+ in_shape = in_tensor.meta["val"].shape
929
+ weight_shape = weight.meta["val"].shape
930
+ out_shape = node.meta["val"].shape
931
+ assert None not in {in_shape, weight_shape, out_shape}
932
+
933
+ # pyre-ignore[6]: Argument type for iteration
934
+ padding_list = list(padding)
935
+ # pyre-ignore[6]: Argument type for iteration
936
+ dilation_list = list(dilation)
937
+
938
+ # Check the condition under which conv can be replaced by linear: (1) this
939
+ # should not be a depthwise convolution; (2) the padding and dilation
940
+ # should be standard; (3) The [channels, height, width] of input must match the
941
+ # [channel, kernel_height, kernel_width] of the weight. These conditions would
942
+ # ensure that output height and width are 1, and the convolution can be replaced
943
+ # by linear. Note: stride is not checked because when in_shape[1:] ==
944
+ # weight_shape[1:], the output spatial dimensions are always 1 regardless of
945
+ # stride (there is only one valid convolution position).
946
+ if (
947
+ groups != 1
948
+ or any(x != 0 for x in padding_list)
949
+ or any(x != 1 for x in dilation_list)
950
+ or (list(in_shape[1:]) != list(weight_shape[1:]))
951
+ ):
952
+ return False
953
+
954
+ # Reshape the weight to [out_channels, in_channels * X]
955
+ K = math.prod(weight_shape[1:])
956
+
957
+ graph = node.graph
958
+
959
+ # Weight is always a Node, so we need a view_copy operation
960
+ with graph.inserting_before(node):
961
+ linear_weight = graph.call_function(
962
+ exir_ops.edge.aten.view_copy.default,
963
+ args=(
964
+ weight,
965
+ [weight_shape[0], K],
966
+ ),
967
+ )
968
+ linear_weight.meta = node.meta
969
+
970
+ # Reshape the input from 3d to 2d tensor
971
+ with graph.inserting_before(node):
972
+ in_view = graph.call_function(
973
+ exir_ops.edge.aten.view_copy.default,
974
+ args=(
975
+ in_tensor,
976
+ [in_shape[0], K],
977
+ ),
978
+ )
979
+ in_view.meta = node.meta
980
+
981
+ # Create the linear node, which multiplies the 2d input and weight
982
+ # tensors, and adds the 1d bias to produce a 2d output.
983
+ if quantized_op:
984
+ (
985
+ in_zero_point,
986
+ weight_zero_point,
987
+ bias_scale,
988
+ out_scale,
989
+ out_zero_point,
990
+ ) = node.args[7:12]
991
+ # Always compute out_multiplier and out_shift from bias_scale / out_scale.
992
+ # The conv reference implementations ignore the out_multiplier and out_shift
993
+ # args and use out_scale directly, but quantized_linear uses the computed
994
+ # values. So we must always recompute them to ensure numerical consistency.
995
+ # pyre-ignore[58]: Division operands
996
+ requantize_scale = bias_scale / out_scale
997
+ (out_multiplier, out_shift) = quantize_tensor_multiplier(
998
+ torch.tensor([requantize_scale])
999
+ )
1000
+ linear_args = (
1001
+ in_view,
1002
+ linear_weight,
1003
+ bias,
1004
+ in_zero_point,
1005
+ weight_zero_point,
1006
+ int(out_multiplier.item()),
1007
+ int(out_shift.item()),
1008
+ out_zero_point,
1009
+ None,
1010
+ )
1011
+ else:
1012
+ linear_args = (in_view, linear_weight, bias)
1013
+ with graph.inserting_before(node):
1014
+ linear_res = graph.call_function(
1015
+ self.trivial_conv_op_to_linear_op[cast(EdgeOpOverload, node.target)],
1016
+ args=linear_args,
1017
+ )
1018
+ linear_res.meta = node.meta
1019
+
1020
+ # Reshape the output of linear from 2d to 3d tensor
1021
+ with graph.inserting_before(node):
1022
+ out_res = graph.call_function(
1023
+ exir_ops.edge.aten.view_copy.default,
1024
+ args=(linear_res, list(out_shape)),
1025
+ )
1026
+ out_res.meta = node.meta
1027
+
1028
+ node.replace_all_uses_with(out_res)
1029
+ return True
1030
+
1031
+
1032
+ def canonicalize_transposed_dim(dim: int, shape: Sequence[int]) -> int:
1033
+ """Canonicalize transpose ops so it gets easier to pattern-match and fuse transpose ops."""
1034
+ if dim < 0:
1035
+ # Keep transpose dimensions positive.
1036
+ dim += len(shape)
1037
+ return dim
1038
+
1039
+
1040
+ @register_cadence_pass(CadencePassAttribute(opt_level=3))
1041
+ class ReplaceConvWithChannelLastConvPass(RemoveOrReplacePassInterface):
1042
+ """
1043
+ Replace NCHW convolutions with NHWC (channel-last) convolutions by adding
1044
+ transpose operations before and after the convolution.
1045
+ """
1046
+
1047
+ @property
1048
+ def targets(self) -> list[EdgeOpOverload]:
1049
+ return [
1050
+ exir_ops.edge.cadence.conv1d.default,
1051
+ exir_ops.edge.cadence.conv2d.default,
1052
+ exir_ops.edge.cadence.conv3d.default,
1053
+ exir_ops.edge.cadence.quantized_conv1d_ncl.per_tensor,
1054
+ exir_ops.edge.cadence.quantized_depthwise_conv1d_ncl.per_tensor,
1055
+ exir_ops.edge.cadence.quantized_conv2d_nchw.per_tensor,
1056
+ ]
1057
+
1058
+ def _transpose_dims(
1059
+ self, graph: torch.fx.Graph, node: torch.fx.Node, dim0: int, dim1: int
1060
+ ) -> torch.fx.Node:
1061
+ """Helper function to transpose dims of a node."""
1062
+ shape = node.meta["val"].shape
1063
+ dim0, dim1 = (
1064
+ canonicalize_transposed_dim(dim0, shape),
1065
+ canonicalize_transposed_dim(dim1, shape),
1066
+ )
1067
+ dim0, dim1 = min(dim0, dim1), max(dim0, dim1)
1068
+ transpose_node = graph.call_function(
1069
+ exir_ops.edge.aten.transpose_copy.int, (node, dim0, dim1), {}
1070
+ )
1071
+ transpose_node.meta = node.meta
1072
+ return transpose_node
1073
+
1074
+ def _change_nchw_to_nhwc(
1075
+ self, graph: torch.fx.Graph, node: torch.fx.Node
1076
+ ) -> torch.fx.Node:
1077
+ """Convert NCHW format to NHWC format."""
1078
+ shape = node.meta["val"].shape
1079
+ indices = list(range(len(shape)))
1080
+ permute_indices = [indices[0]] + indices[2:] + [indices[1]]
1081
+ permute_node = graph.call_function(
1082
+ exir_ops.edge.aten.permute_copy.default, (node, permute_indices), {}
1083
+ )
1084
+ permute_node.meta = node.meta
1085
+ return permute_node
1086
+
1087
+ def _change_nhwc_to_nchw(
1088
+ self, graph: torch.fx.Graph, node: torch.fx.Node
1089
+ ) -> torch.fx.Node:
1090
+ """Convert NHWC format to NCHW format."""
1091
+ shape = node.meta["val"].shape
1092
+ indices = list(range(len(shape)))
1093
+ permute_indices = [indices[0], indices[-1]] + indices[1:-1]
1094
+ permute_node = graph.call_function(
1095
+ exir_ops.edge.aten.permute_copy.default, (node, permute_indices), {}
1096
+ )
1097
+ permute_node.meta = node.meta
1098
+ return permute_node
1099
+
1100
+ def _change_depthwise_weight_to_hwc(
1101
+ self, graph: torch.fx.Graph, node: torch.fx.Node, is_2d: bool = True
1102
+ ) -> torch.fx.Node:
1103
+ """Convert depthwise weight from OIHW [OC, 1, KH, KW] to HWC [KH, KW, OC].
1104
+
1105
+ NNLib depthwise convolution expects weights in [KH, KW, OC] format when
1106
+ inp_data_format=0 (NHWC), but the standard NCHW->NHWC permutation produces
1107
+ [OC, KH, KW, 1]. This function applies the correct permutation for depthwise
1108
+ convolution weights.
1109
+
1110
+ For the 1D case, the C++ NLC kernels expect 3D weights in [OC, K, IC/groups]
1111
+ format, so we use the standard NCHW->NHWC transpose (swap dims 1 and -1)
1112
+ to produce [OC, K, 1].
1113
+ """
1114
+ if not is_2d:
1115
+ # 1D case: use standard transpose [OC, 1, K] -> [OC, K, 1]
1116
+ # The C++ NLC kernels expect 3D weight in [OC, K, IC/groups] format.
1117
+ return self._transpose_dims(graph, node, 1, -1)
1118
+
1119
+ # 2D case: [OC, 1, KH, KW], target is [KH, KW, OC]
1120
+ # Permute [0, 1, 2, 3] -> [2, 3, 0, 1] gives [KH, KW, OC, 1]
1121
+ # Then squeeze the last dim (which is 1) to get [KH, KW, OC]
1122
+ permute_indices = [2, 3, 0, 1]
1123
+ permute_node = graph.call_function(
1124
+ exir_ops.edge.aten.permute_copy.default, (node, permute_indices), {}
1125
+ )
1126
+ permute_node.meta = node.meta
1127
+
1128
+ # Squeeze the last dimension (which has size 1)
1129
+ squeeze_node = graph.call_function(
1130
+ exir_ops.edge.aten.squeeze_copy.dim, (permute_node, -1), {}
1131
+ )
1132
+ squeeze_node.meta = node.meta
1133
+ return squeeze_node
1134
+
1135
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1136
+ assert isinstance(node.target, EdgeOpOverload)
1137
+ quantized_op = node.target in {
1138
+ exir_ops.edge.cadence.quantized_conv1d_ncl.per_tensor,
1139
+ exir_ops.edge.cadence.quantized_depthwise_conv1d_ncl.per_tensor,
1140
+ exir_ops.edge.cadence.quantized_conv2d_nchw.per_tensor,
1141
+ }
1142
+
1143
+ # Check if already in NHWC/NLC layout
1144
+ if not quantized_op and len(node.args) == 8 and node.args[-1] is True:
1145
+ return False
1146
+
1147
+ # Get input shape to determine if it's 1D or 2D
1148
+ input_node = get_arg(node, "input", torch.fx.Node)
1149
+ input_shape = input_node.meta["val"].shape
1150
+ is_2d = len(input_shape) == 4
1151
+
1152
+ # Determine the new op target
1153
+ if quantized_op:
1154
+ if is_2d:
1155
+ new_op = exir_ops.edge.cadence.quantized_conv2d_nhwc.per_tensor
1156
+ else:
1157
+ assert len(input_shape) == 3
1158
+ if (
1159
+ node.target
1160
+ == exir_ops.edge.cadence.quantized_depthwise_conv1d_ncl.per_tensor
1161
+ ):
1162
+ new_op = (
1163
+ exir_ops.edge.cadence.quantized_depthwise_conv1d_nlc.per_tensor
1164
+ )
1165
+ else:
1166
+ new_op = exir_ops.edge.cadence.quantized_conv1d_nlc.per_tensor
1167
+ else:
1168
+ new_op = node.target
1169
+
1170
+ graph = node.graph
1171
+
1172
+ # Get weight node
1173
+ weight_node = get_arg(node, "weight", torch.fx.Node)
1174
+
1175
+ # Check if this is a depthwise convolution (groups == input_channels)
1176
+ # and weight is 4D with shape [OC, 1, KH, KW]
1177
+ groups = get_arg(node, "groups", int)
1178
+ weight_shape = weight_node.meta["val"].shape
1179
+ input_channels = input_shape[1] # NCHW format, channels at index 1
1180
+ # NCHW: also verify weight IC dim == 1.
1181
+ depthwise = is_depthwise_conv(groups, input_channels) and weight_shape[1] == 1
1182
+
1183
+ # Insert transpose operations before the node
1184
+ with graph.inserting_before(node):
1185
+ # Convert input from NCHW to NHWC
1186
+ input_nhwc = self._change_nchw_to_nhwc(graph, input_node)
1187
+ # Convert weight from NCHW to the appropriate format
1188
+ if depthwise:
1189
+ # For depthwise: [OC, 1, KH, KW] -> [KH, KW, OC] for NNLib
1190
+ weight_nhwc = self._change_depthwise_weight_to_hwc(
1191
+ graph, weight_node, is_2d
1192
+ )
1193
+ else:
1194
+ # For regular conv: [OC, IC, KH, KW] -> [OC, KH, KW, IC]
1195
+ weight_nhwc = self._change_nchw_to_nhwc(graph, weight_node)
1196
+
1197
+ # Non-quantized ops need to set the last optional argument to True
1198
+ channel_last_arg = [] if quantized_op else [True]
1199
+
1200
+ # Create new args with transposed input/weights
1201
+ new_args = (
1202
+ (input_nhwc, weight_nhwc)
1203
+ + tuple(node.args[2:])
1204
+ + tuple(channel_last_arg)
1205
+ )
1206
+
1207
+ # Create the new conv operation
1208
+ new_conv = graph.call_function(new_op, new_args, node.kwargs)
1209
+ new_conv.meta = node.meta
1210
+
1211
+ # Convert output back from NHWC to NCHW
1212
+ nchw_output = self._change_nhwc_to_nchw(graph, new_conv)
1213
+
1214
+ # Replace all uses with the final output
1215
+ node.replace_all_uses_with(nchw_output)
1216
+ return True
1217
+
1218
+
1219
+ @register_cadence_pass(CadencePassAttribute(opt_level=3))
1220
+ class ReplaceMaxPool2dWithChannelLastMaxPool2dPass(RemoveOrReplacePassInterface):
1221
+ """
1222
+ Replace NCHW max pooling with NHWC (channel-last) max pooling by adding
1223
+ permute operations before and after the max pooling.
1224
+ """
1225
+
1226
+ @property
1227
+ def targets(self) -> list[EdgeOpOverload]:
1228
+ return [
1229
+ exir_ops.edge.cadence.quantized_max_pool2d_nchw.default,
1230
+ ]
1231
+
1232
+ def _change_nchw_to_nhwc(
1233
+ self, graph: torch.fx.Graph, node: torch.fx.Node
1234
+ ) -> torch.fx.Node:
1235
+ """Convert NCHW format to NHWC format."""
1236
+ permute_node = graph.call_function(
1237
+ exir_ops.edge.aten.permute_copy.default, (node, [0, 2, 3, 1]), {}
1238
+ )
1239
+ permute_node.meta = node.meta
1240
+ return permute_node
1241
+
1242
+ def _change_nhwc_to_nchw(
1243
+ self, graph: torch.fx.Graph, node: torch.fx.Node
1244
+ ) -> torch.fx.Node:
1245
+ """Convert NHWC format to NCHW format."""
1246
+ permute_node = graph.call_function(
1247
+ exir_ops.edge.aten.permute_copy.default, (node, [0, 3, 1, 2]), {}
1248
+ )
1249
+ permute_node.meta = node.meta
1250
+ return permute_node
1251
+
1252
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1253
+ graph = node.graph
1254
+
1255
+ # Get input node
1256
+ input_node = cast(torch.fx.Node, node.args[0])
1257
+
1258
+ with graph.inserting_before(node):
1259
+ # Convert input from NCHW to NHWC
1260
+ input_nhwc = self._change_nchw_to_nhwc(graph, input_node)
1261
+
1262
+ # Create the NHWC max pooling with the same args (kernel_size, stride, padding, dilation, ceil_mode)
1263
+ new_args = (input_nhwc,) + tuple(node.args[1:])
1264
+
1265
+ new_pool = graph.call_function(
1266
+ exir_ops.edge.cadence.quantized_max_pool2d_nhwc.default,
1267
+ new_args,
1268
+ node.kwargs,
1269
+ )
1270
+ new_pool.meta = node.meta
1271
+
1272
+ # Convert output back from NHWC to NCHW
1273
+ nchw_output = self._change_nhwc_to_nchw(graph, new_pool)
1274
+
1275
+ # Replace all uses with the final output
1276
+ node.replace_all_uses_with(nchw_output)
1277
+ return True
1278
+
1279
+
1280
+ @register_cadence_pass(CadencePassAttribute(opt_level=3))
1281
+ class MakeSliceAndCatDimOutermostPass(RemoveOrReplacePassInterface):
1282
+ """
1283
+ Make the slice/cat dimension the outermost dimension by adding transpose
1284
+ operations before and after the slice/cat operation.
1285
+ """
1286
+
1287
+ @property
1288
+ def targets(self) -> list[EdgeOpOverload]:
1289
+ return [
1290
+ exir_ops.edge.aten.cat.default,
1291
+ exir_ops.edge.aten.slice_copy.Tensor,
1292
+ ]
1293
+
1294
+ def _transpose_dims(
1295
+ self, graph: torch.fx.Graph, node: torch.fx.Node, dim0: int, dim1: int
1296
+ ) -> torch.fx.Node:
1297
+ """Helper function to transpose dims of a node."""
1298
+ shape = node.meta["val"].shape
1299
+ dim0, dim1 = (
1300
+ canonicalize_transposed_dim(dim0, shape),
1301
+ canonicalize_transposed_dim(dim1, shape),
1302
+ )
1303
+ dim0, dim1 = min(dim0, dim1), max(dim0, dim1)
1304
+ transpose_node = graph.call_function(
1305
+ exir_ops.edge.aten.transpose_copy.int, (node, dim0, dim1), {}
1306
+ )
1307
+ transpose_node.meta = node.meta
1308
+ return transpose_node
1309
+
1310
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1311
+ # Get the dimension argument
1312
+ dim = cast(int, node.args[1]) if len(node.args) > 1 else 0
1313
+ output_shape = node.meta["val"].shape
1314
+
1315
+ # Canonicalize dim to be positive
1316
+ if dim < 0:
1317
+ dim += len(output_shape)
1318
+
1319
+ # Not needed if dim is already outermost or all dims before it are 1
1320
+ if dim == 0 or math.prod(output_shape[:dim]) == 1:
1321
+ return False
1322
+
1323
+ graph = node.graph
1324
+
1325
+ with graph.inserting_before(node):
1326
+ if node.target == exir_ops.edge.aten.slice_copy.Tensor:
1327
+ # Transpose input -> slice with dim=0 -> transpose back
1328
+ input_node = cast(torch.fx.Node, node.args[0])
1329
+ transposed_input = self._transpose_dims(graph, input_node, dim, 0)
1330
+
1331
+ # Create slice operation with dim=0
1332
+ slice_args = (transposed_input, 0) + node.args[2:]
1333
+ sliced = graph.call_function(
1334
+ exir_ops.edge.aten.slice_copy.Tensor, slice_args, node.kwargs
1335
+ )
1336
+ sliced.meta = node.meta
1337
+
1338
+ # Transpose back
1339
+ result = self._transpose_dims(graph, sliced, 0, dim)
1340
+ else:
1341
+ # Cat operation: transpose all inputs -> cat with dim=0 -> transpose back
1342
+ cat_inputs = cast(list[torch.fx.Node], node.args[0])
1343
+ transposed_inputs = [
1344
+ self._transpose_dims(graph, t, dim, 0) for t in cat_inputs
1345
+ ]
1346
+
1347
+ # Create cat operation with dim=0
1348
+ catted = graph.call_function(
1349
+ exir_ops.edge.aten.cat.default, (transposed_inputs, 0), node.kwargs
1350
+ )
1351
+ catted.meta = node.meta
1352
+
1353
+ # Transpose back
1354
+ result = self._transpose_dims(graph, catted, 0, dim)
1355
+
1356
+ # Replace all uses with the final result
1357
+ node.replace_all_uses_with(result)
1358
+ return True
1359
+
1360
+
1361
+ @register_cadence_pass(CadencePassAttribute(opt_level=2, use_im2row_transform=True))
1362
+ class ReplaceConvWithIm2RowAndLinear(RemoveOrReplacePassInterface):
1363
+ """
1364
+ Replace convolution where groups=1 with im2row followed by a linear op.
1365
+ """
1366
+
1367
+ # A map from the convolution op to the linear op that it should
1368
+ # decompose to.
1369
+ conv_op_to_linear_op: Dict[EdgeOpOverload, EdgeOpOverload] = {
1370
+ exir_ops.edge.cadence.conv2d.default: exir_ops.edge.aten.linear.default,
1371
+ exir_ops.edge.cadence.conv3d.default: exir_ops.edge.aten.linear.default,
1372
+ exir_ops.edge.cadence.quantized_conv2d_nchw.per_tensor: exir_ops.edge.cadence.quantized_linear.per_tensor,
1373
+ exir_ops.edge.cadence.quantized_conv2d_nhwc.per_tensor: exir_ops.edge.cadence.quantized_linear.per_tensor,
1374
+ }
1375
+
1376
+ # Set of quantized conv ops
1377
+ quantized_conv_ops: frozenset[EdgeOpOverload] = frozenset(
1378
+ {
1379
+ exir_ops.edge.cadence.quantized_conv2d_nchw.per_tensor,
1380
+ exir_ops.edge.cadence.quantized_conv2d_nhwc.per_tensor,
1381
+ }
1382
+ )
1383
+
1384
+ # Set of channel-last conv ops (NHWC for 2D)
1385
+ channel_last_conv_ops: frozenset[EdgeOpOverload] = frozenset(
1386
+ {
1387
+ exir_ops.edge.cadence.quantized_conv2d_nhwc.per_tensor,
1388
+ }
1389
+ )
1390
+
1391
+ @property
1392
+ def targets(self) -> list[EdgeOpOverload]:
1393
+ return list(self.conv_op_to_linear_op.keys())
1394
+
1395
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1396
+ # Get the relevant args from convolution node.
1397
+ assert isinstance(node.target, EdgeOpOverload)
1398
+ quantized_op = node.target in self.quantized_conv_ops
1399
+ assert (len(node.args) == 7 and not quantized_op) or (
1400
+ len(node.args) >= 12 and quantized_op
1401
+ ), "Inconsistent args for convolution"
1402
+ (in_tensor, weight, bias, stride, padding, dilation, groups) = node.args[0:7]
1403
+
1404
+ assert isinstance(in_tensor, torch.fx.Node)
1405
+ assert isinstance(weight, torch.fx.Node)
1406
+
1407
+ # We do not replace depthwise convolution with gemm yet.
1408
+ if groups != 1:
1409
+ return False
1410
+
1411
+ weight_shape = weight.meta["val"].shape
1412
+
1413
+ # pyre-ignore[6]: Argument type for iteration
1414
+ stride_list = list(stride)
1415
+ # pyre-ignore[6]: Argument type for iteration
1416
+ padding_list = list(padding)
1417
+ # pyre-ignore[6]: Argument type for iteration
1418
+ dilation_list = list(dilation)
1419
+
1420
+ # If this is a pointwise convolution, im2col will start dominating the
1421
+ # runtime. So we call convolution op for this case.
1422
+ if (
1423
+ all(x == 1 for x in weight_shape[2:])
1424
+ and all(x == 1 for x in stride_list)
1425
+ and all(x == 0 for x in padding_list)
1426
+ and all(x == 1 for x in dilation_list)
1427
+ ):
1428
+ return False
1429
+
1430
+ # Get the shapes
1431
+ out_shape = node.meta["val"].shape
1432
+ assert None not in {weight_shape, out_shape}
1433
+
1434
+ # Determine if the convolution is NCHW or NHWC (for 2D) / NCL or NLC (for 1D).
1435
+ # Channel-last layouts are NHWC for 2D and NLC for 1D.
1436
+ channel_last = node.target in self.channel_last_conv_ops
1437
+ # The weight tensor is [out_channels, in_channels, X] for NCHW layout,
1438
+ # and [out_channels, X, in_channels] for NHWC layout. Here, X is the
1439
+ # kernel_width for conv1d, and X = kernel_height * kernel_width for
1440
+ # conv2d. We extract X as the kernel_size for im2row.
1441
+ kernel_size = list(weight_shape[1:-1] if channel_last else weight_shape[2:])
1442
+ # If the convolution op was quantized, we need the input tensor's
1443
+ # zero_point for im2row. Otherwise in_zero_point defaults to a zero
1444
+ # tensor.
1445
+ in_zero_point = node.args[7] if quantized_op else 0
1446
+
1447
+ # im2row expects every kernel parameter to be 2d. So we extend the
1448
+ # parameters for conv1d by prepending their default values.
1449
+ stride_2d = ([1] + stride_list) if len(stride_list) == 1 else stride_list
1450
+ padding_2d = ([0] + padding_list) if len(padding_list) == 1 else padding_list
1451
+ dilation_2d = (
1452
+ ([1] + dilation_list) if len(dilation_list) == 1 else dilation_list
1453
+ )
1454
+ kernel_size = ([1] + kernel_size) if len(kernel_size) == 1 else kernel_size
1455
+ # Assert that kernel size does not have a 0
1456
+ assert 0 not in kernel_size
1457
+
1458
+ graph = node.graph
1459
+
1460
+ # Create an im2row node with the input. This will create a 2d matrix of
1461
+ # shape [out_height*out_weight, X*in_channels]. X is as defined in the
1462
+ # comment above.
1463
+ im2row_args = (
1464
+ in_tensor,
1465
+ kernel_size,
1466
+ dilation_2d,
1467
+ padding_2d,
1468
+ stride_2d,
1469
+ in_zero_point,
1470
+ channel_last,
1471
+ )
1472
+ with graph.inserting_before(node):
1473
+ im2row = graph.call_function(
1474
+ exir_ops.edge.cadence.im2row.per_tensor,
1475
+ args=im2row_args,
1476
+ )
1477
+ im2row.meta = node.meta
1478
+
1479
+ # Get the product of the >2 dims of the weight
1480
+ K = math.prod(weight_shape[1:])
1481
+
1482
+ # For channel-last 2D convolutions (NHWC), the weight layout is
1483
+ # [OC, kH, kW, IC] but im2row outputs patches in channel-first order [C, K].
1484
+ # We need to permute the weight from [OC, kH, kW, IC] to [OC, IC, kH, kW]
1485
+ # before flattening to match the im2row patch ordering.
1486
+ weight_to_flatten = weight
1487
+ if channel_last:
1488
+ # 2D: [OC, kH, kW, IC] -> [OC, IC, kH, kW] via permute
1489
+ with graph.inserting_before(node):
1490
+ weight_to_flatten = graph.call_function(
1491
+ exir_ops.edge.aten.permute_copy.default,
1492
+ args=(weight, [0, 3, 1, 2]),
1493
+ )
1494
+ # NB: The shape in node.meta won't match the permuted weight
1495
+ # shape. We copy it as a placeholder to populate non-shape
1496
+ # metadata (e.g., stack_trace, dtype). ExportPass.call() is
1497
+ # invoked after this pass (via HierarchicalInplacePassInterface)
1498
+ # and will re-run shape propagation to correct all meta['val'].
1499
+ weight_to_flatten.meta = node.meta
1500
+
1501
+ # Weight is always a Node, so we need a view_copy operation
1502
+ with graph.inserting_before(node):
1503
+ linear_weight = graph.call_function(
1504
+ exir_ops.edge.aten.view_copy.default,
1505
+ args=(
1506
+ weight_to_flatten,
1507
+ [weight_shape[0], K],
1508
+ ),
1509
+ )
1510
+ linear_weight.meta = node.meta
1511
+
1512
+ # Create the linear node, which multiplies the 3d input with 2d weight
1513
+ # tensors with bias addition. The outermost dimension of the input is
1514
+ # the batch size for linear op.
1515
+ if quantized_op:
1516
+ (
1517
+ in_zero_point,
1518
+ weight_zero_point,
1519
+ bias_scale,
1520
+ out_scale,
1521
+ out_zero_point,
1522
+ ) = node.args[7:12]
1523
+ # Always compute out_multiplier and out_shift from bias_scale / out_scale.
1524
+ # The conv reference implementations ignore the out_multiplier and out_shift
1525
+ # args and use out_scale directly, but quantized_linear uses the computed
1526
+ # values. So we must always recompute them to ensure numerical consistency.
1527
+ # pyre-ignore[58]: Division operands
1528
+ requantize_scale = bias_scale / out_scale
1529
+ (out_multiplier, out_shift) = quantize_tensor_multiplier(
1530
+ torch.tensor([requantize_scale])
1531
+ )
1532
+ linear_args = (
1533
+ im2row,
1534
+ linear_weight,
1535
+ bias,
1536
+ in_zero_point,
1537
+ weight_zero_point,
1538
+ int(out_multiplier.item()),
1539
+ int(out_shift.item()),
1540
+ out_zero_point,
1541
+ None,
1542
+ )
1543
+ else:
1544
+ linear_args = (im2row, linear_weight, bias)
1545
+
1546
+ with graph.inserting_before(node):
1547
+ linear_res = graph.call_function(
1548
+ self.conv_op_to_linear_op[cast(EdgeOpOverload, node.target)],
1549
+ args=linear_args,
1550
+ )
1551
+ linear_res.meta = node.meta
1552
+
1553
+ # The output of linear is a 3D tensor. However, the output is in NHWC
1554
+ # layout by default, because an input vector of size X is multiplied
1555
+ # with the weight matrix, i.e., column values are contiguous. If the
1556
+ # channel_last is False, we want to transpose this output.
1557
+ if not channel_last:
1558
+ with graph.inserting_before(node):
1559
+ linear_res = graph.call_function(
1560
+ exir_ops.edge.aten.transpose_copy.int,
1561
+ args=(linear_res, 1, 2),
1562
+ )
1563
+ linear_res.meta = node.meta
1564
+
1565
+ # And finally, we want to view the 3D output of linear op as 4D tensor
1566
+ with graph.inserting_before(node):
1567
+ out_res = graph.call_function(
1568
+ exir_ops.edge.aten.view_copy.default,
1569
+ args=(linear_res, list(out_shape)),
1570
+ )
1571
+ out_res.meta = node.meta
1572
+
1573
+ node.replace_all_uses_with(out_res)
1574
+ return True
1575
+
1576
+
1577
+ @register_cadence_pass(CadencePassAttribute(opt_level=2))
1578
+ class ReplaceTransposedConvWithLinearPass(RemoveOrReplacePassInterface):
1579
+ """
1580
+ Replace transposed convolution where groups=1 with transposed_im2row
1581
+ followed by a linear op.
1582
+ """
1583
+
1584
+ # A map from the transposed_convolution op to the linear op that it should
1585
+ # decompose to.
1586
+ transposed_conv_op_to_linear_op: Dict[EdgeOpOverload, EdgeOpOverload] = {
1587
+ exir_ops.edge.cadence.transposed_convolution.default: exir_ops.edge.aten.linear.default,
1588
+ exir_ops.edge.cadence.quantized_transposed_conv.default: exir_ops.edge.cadence.quantized_linear.default,
1589
+ }
1590
+
1591
+ @property
1592
+ def targets(self) -> list[EdgeOpOverload]:
1593
+ return list(self.transposed_conv_op_to_linear_op.keys())
1594
+
1595
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1596
+ # Get the relevant args from transposed_convolution node.
1597
+ assert isinstance(node.target, EdgeOpOverload)
1598
+ quantized_op = (
1599
+ node.target == exir_ops.edge.cadence.quantized_transposed_conv.default
1600
+ )
1601
+ expected_args = 16 if quantized_op else 9
1602
+ if len(node.args) != expected_args:
1603
+ return False
1604
+
1605
+ (
1606
+ in_tensor,
1607
+ weight,
1608
+ bias,
1609
+ stride,
1610
+ padding,
1611
+ dilation,
1612
+ output_padding,
1613
+ groups,
1614
+ ) = node.args[0:8]
1615
+
1616
+ # We do not replace depthwise transposed_convolution with gemm yet.
1617
+ if groups != 1:
1618
+ return False
1619
+
1620
+ # Get the shapes
1621
+ assert isinstance(weight, torch.fx.Node)
1622
+ out_shape = node.meta["val"].shape
1623
+ weight_shape = weight.meta["val"].shape
1624
+ if None in {weight_shape, out_shape}:
1625
+ return False
1626
+
1627
+ # Determine if the transposed_convolution is NCHW or NHWC. The NHWC,
1628
+ # i.e., the channel_last layout is specified by the channel_last arg
1629
+ # of transposed_conv op, which is the last argument.
1630
+ channel_last = node.args[-1]
1631
+ # The weight tensor is [out_channels, in_channels, X] for NCHW layout,
1632
+ # and [out_channels, X, in_channels] for NHWC layout. Here, X is the
1633
+ # kernel_width for conv1d, and X = kernel_height * kernel_width for
1634
+ # conv2d. We extract X as the kernel_size for im2row.
1635
+ kernel_size = list(weight_shape[1:-1] if channel_last else weight_shape[2:])
1636
+ assert isinstance(in_tensor, torch.fx.Node)
1637
+
1638
+ # Cast to appropriate types
1639
+ stride = cast(Sequence[int], stride)
1640
+ padding = cast(Sequence[int], padding)
1641
+ dilation = cast(Sequence[int], dilation)
1642
+ output_padding = cast(Sequence[int], output_padding)
1643
+
1644
+ # transposed_im2row expects every kernel parameter to be 2d. So we extend the
1645
+ # parameters for conv1d by prepending their default values.
1646
+ stride_list = ([1] + list(stride)) if len(stride) == 1 else list(stride)
1647
+ padding_list = ([0] + list(padding)) if len(padding) == 1 else list(padding)
1648
+ dilation_list = ([1] + list(dilation)) if len(dilation) == 1 else list(dilation)
1649
+ output_padding_list = (
1650
+ ([0] + list(output_padding))
1651
+ if len(output_padding) == 1
1652
+ else list(output_padding)
1653
+ )
1654
+ kernel_size = ([1] + kernel_size) if len(kernel_size) == 1 else kernel_size
1655
+ # Check that kernel size does not have a 0
1656
+ if 0 in kernel_size:
1657
+ return False
1658
+
1659
+ graph = node.graph
1660
+
1661
+ # If the transposed_convolution op was quantized, we need the input tensor's
1662
+ # zero_point for im2row. Otherwise in_zero_point defaults to a zero tensor.
1663
+ # We create the tensor as a graph node using aten.full to avoid
1664
+ # DataDependentOutputException during FakeTensor shape propagation.
1665
+ in_zero_point_val = node.args[8] if quantized_op else 0
1666
+ in_zero_point = graph.call_function(
1667
+ exir_ops.edge.aten.full.default,
1668
+ args=([1], in_zero_point_val),
1669
+ kwargs={"dtype": torch.int32},
1670
+ )
1671
+ # Insert the node before the current node
1672
+ node.prepend(in_zero_point)
1673
+ in_zero_point.meta = node.meta
1674
+
1675
+ # Create a transposed_im2row node with the input. This will create a 2d
1676
+ # matrix of shape [out_height*out_weight, X*in_channels]. X is as
1677
+ # defined in the comment above.
1678
+ transposed_im2row_args = (
1679
+ in_tensor,
1680
+ kernel_size,
1681
+ dilation_list,
1682
+ padding_list,
1683
+ stride_list,
1684
+ output_padding_list,
1685
+ in_zero_point,
1686
+ channel_last,
1687
+ )
1688
+ with graph.inserting_before(node):
1689
+ transposed_im2row = graph.call_function(
1690
+ exir_ops.edge.cadence.transposed_im2row.default,
1691
+ args=transposed_im2row_args,
1692
+ )
1693
+ transposed_im2row.meta = node.meta
1694
+
1695
+ # Reshape the weight to [out_channels, in_channels * X]
1696
+ K = math.prod(weight_shape[1:])
1697
+
1698
+ # Weight is always a Node, so we need a view_copy operation
1699
+ with graph.inserting_before(node):
1700
+ linear_weight = graph.call_function(
1701
+ exir_ops.edge.aten.view_copy.default,
1702
+ args=(
1703
+ weight,
1704
+ [weight_shape[0], K],
1705
+ ),
1706
+ )
1707
+ linear_weight.meta = node.meta
1708
+
1709
+ # Create the linear node, which multiplies the 3d input with 2d weight
1710
+ # tensors with bias addition. The outermost dimension of the input is
1711
+ # the batch size for linear op.
1712
+ if quantized_op:
1713
+ (
1714
+ in_zero_point,
1715
+ weight_zero_point,
1716
+ bias_scale,
1717
+ out_scale,
1718
+ out_zero_point,
1719
+ ) = node.args[8:13]
1720
+ # pyre-ignore[58]: Division operands
1721
+ requantize_scale = bias_scale / out_scale
1722
+ (out_multiplier, out_shift) = quantize_tensor_multiplier(requantize_scale)
1723
+ linear_args = (
1724
+ transposed_im2row,
1725
+ linear_weight,
1726
+ bias,
1727
+ in_zero_point,
1728
+ weight_zero_point,
1729
+ out_multiplier,
1730
+ out_shift,
1731
+ out_zero_point,
1732
+ None,
1733
+ )
1734
+ else:
1735
+ linear_args = (transposed_im2row, linear_weight, bias)
1736
+
1737
+ with graph.inserting_before(node):
1738
+ linear_res = graph.call_function(
1739
+ self.transposed_conv_op_to_linear_op[cast(EdgeOpOverload, node.target)],
1740
+ args=linear_args,
1741
+ )
1742
+ linear_res.meta = node.meta
1743
+
1744
+ # The output of linear is a 3D tensor. However, the output is in NHWC
1745
+ # layout by default, because an input vector of size X is multiplied
1746
+ # with the weight matrix, i.e., column values are contiguous. If the
1747
+ # channel_last is False, we want to transpose this output.
1748
+ if not channel_last:
1749
+ with graph.inserting_before(node):
1750
+ linear_res = graph.call_function(
1751
+ exir_ops.edge.aten.transpose_copy.int,
1752
+ args=(linear_res, 1, 2),
1753
+ )
1754
+ linear_res.meta = node.meta
1755
+
1756
+ # And finally, we want to view the 3D output of linear op as 4D tensor
1757
+ with graph.inserting_before(node):
1758
+ out_res = graph.call_function(
1759
+ exir_ops.edge.aten.view_copy.default,
1760
+ args=(linear_res, list(out_shape)),
1761
+ )
1762
+ out_res.meta = node.meta
1763
+
1764
+ node.replace_all_uses_with(out_res)
1765
+ return True
1766
+
1767
+
1768
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
1769
+ class ReplaceNopTransposeOrPermuteWithViewPass(
1770
+ _SharedReplaceNopTransposeOrPermuteWithViewPass
1771
+ ):
1772
+ pass
1773
+
1774
+
1775
+ @register_cadence_pass(CadencePassAttribute(opt_level=2))
1776
+ class ReplaceLinearWithFullyConnectedOpPass(RemoveOrReplacePassInterface):
1777
+ """
1778
+ If the input of linear/quantized_linear op is a vector, replace it with
1779
+ fully_connected op.
1780
+ """
1781
+
1782
+ linear_to_fc_op: Dict[EdgeOpOverload, EdgeOpOverload] = {
1783
+ # Default variants
1784
+ exir_ops.edge.aten.linear.default: exir_ops.edge.cadence.fully_connected.default,
1785
+ exir_ops.edge.cadence.quantized_linear.default: exir_ops.edge.cadence.quantized_fully_connected.default,
1786
+ # Per-tensor variants
1787
+ exir_ops.edge.cadence.quantized_linear.per_tensor: exir_ops.edge.cadence.quantized_fully_connected.per_tensor,
1788
+ # Type-specialized variants (int8)
1789
+ exir_ops.edge.cadence.quantized_linear_asym8sxasym8s_asym8s.per_tensor: exir_ops.edge.cadence.quantized_fully_connected_asym8sxasym8s_asym8s.per_tensor,
1790
+ # Type-specialized variants (uint8)
1791
+ exir_ops.edge.cadence.quantized_linear_asym8uxasym8u_asym8u.per_tensor: exir_ops.edge.cadence.quantized_fully_connected_asym8uxasym8u_asym8u.per_tensor,
1792
+ }
1793
+
1794
+ @property
1795
+ def targets(self) -> list[EdgeOpOverload]:
1796
+ return list(self.linear_to_fc_op.keys())
1797
+
1798
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1799
+ # Extract the input tensor
1800
+ in_tensor_arg = node.args[0]
1801
+ assert isinstance(in_tensor_arg, torch.fx.Node)
1802
+ in_tensor_shape = in_tensor_arg.meta["val"].shape
1803
+ leading_dims = math.prod(in_tensor_shape[:-1])
1804
+ # If the tensor is not a vector, do nothing.
1805
+ if leading_dims != 1:
1806
+ return False
1807
+
1808
+ # Replace the linear with fully connected op
1809
+ assert isinstance(node.target, EdgeOpOverload)
1810
+ with node.graph.inserting_before(node):
1811
+ new_node = node.graph.call_function(
1812
+ self.linear_to_fc_op[cast(EdgeOpOverload, node.target)],
1813
+ args=node.args,
1814
+ kwargs=node.kwargs,
1815
+ )
1816
+ new_node.meta = node.meta
1817
+
1818
+ node.replace_all_uses_with(new_node)
1819
+ return True
1820
+
1821
+
1822
+ register_cadence_pass(CadencePassAttribute(opt_level=0))(ReplaceScalarWithTensorArgPass)
1823
+
1824
+
1825
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
1826
+ class ReplaceScalarTensorWithFullPass(RemoveOrReplacePassInterface):
1827
+ """
1828
+ aten.scalar_tensor can be replaced by aten.full with a shape of [1].
1829
+ scalar_tensor is not supported, so this is an opt_level=0 pass.
1830
+ """
1831
+
1832
+ @property
1833
+ def targets(self) -> list[EdgeOpOverload]:
1834
+ return [
1835
+ torch.ops.aten.scalar_tensor.default,
1836
+ exir_ops.edge.aten.scalar_tensor.default,
1837
+ ]
1838
+
1839
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1840
+ with node.graph.inserting_before(node):
1841
+ new_node = node.graph.call_function(
1842
+ exir_ops.edge.aten.full.default,
1843
+ args=(
1844
+ [1],
1845
+ node.args[0],
1846
+ ),
1847
+ kwargs={"dtype": torch.float32},
1848
+ )
1849
+ new_node.meta = node.meta
1850
+ node.replace_all_uses_with(new_node)
1851
+ return True
1852
+
1853
+
1854
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
1855
+ class ReplaceFullLikeWithFullPass(RemoveOrReplacePassInterface):
1856
+ """
1857
+ aten.full_like can be replaced by aten.full with the shape of the arg tensor.
1858
+ full_like is not supported, so this is an opt_level=0 pass.
1859
+ """
1860
+
1861
+ @property
1862
+ def targets(self) -> list[EdgeOpOverload]:
1863
+ return [exir_ops.edge.aten.full_like.default]
1864
+
1865
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1866
+ input_arg = node.args[0]
1867
+ assert isinstance(input_arg, torch.fx.Node)
1868
+ shape = input_arg.meta["val"].shape
1869
+ fill_value = node.args[1]
1870
+
1871
+ with node.graph.inserting_before(node):
1872
+ new_node = node.graph.call_function(
1873
+ exir_ops.edge.aten.full.default,
1874
+ args=(shape, fill_value),
1875
+ kwargs={},
1876
+ )
1877
+ new_node.meta = node.meta
1878
+ node.replace_all_uses_with(new_node)
1879
+ return True
1880
+
1881
+
1882
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
1883
+ class ReplaceInfArgInFullWithValuePass(RemoveOrReplacePassInterface):
1884
+ """
1885
+ aten.full allows "-inf" and "inf" as inputs. The profiler cannot
1886
+ handle that, so replace them with the maximum value of the type.
1887
+ """
1888
+
1889
+ @property
1890
+ def targets(self) -> list[EdgeOpOverload]:
1891
+ return [exir_ops.edge.aten.full.default]
1892
+
1893
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1894
+ new_args = list(node.args)
1895
+ fill_value = node.args[1]
1896
+ if fill_value == float("-inf"):
1897
+ new_args[1] = torch.finfo(torch.float32).min
1898
+ elif fill_value == float("inf"):
1899
+ new_args[1] = torch.finfo(torch.float32).max
1900
+ else:
1901
+ return False
1902
+
1903
+ new_args = tuple(new_args)
1904
+
1905
+ with node.graph.inserting_before(node):
1906
+ new_node = node.graph.call_function(
1907
+ exir_ops.edge.aten.full.default,
1908
+ args=new_args,
1909
+ kwargs=node.kwargs,
1910
+ )
1911
+ new_node.meta = node.meta
1912
+ node.replace_all_uses_with(new_node)
1913
+ return True
1914
+
1915
+
1916
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
1917
+ class ReplaceAtenAvgPoolWithCadenceAvgPoolPass(RemoveOrReplacePassInterface):
1918
+ """
1919
+ Replace the aten avg_pool op with the cadence custom avg_pool2d op.
1920
+ """
1921
+
1922
+ @property
1923
+ def targets(self) -> list[EdgeOpOverload]:
1924
+ return [
1925
+ exir_ops.edge.aten.avg_pool1d.default,
1926
+ exir_ops.edge.aten.avg_pool2d.default,
1927
+ ]
1928
+
1929
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
1930
+ # Determine if the op is avg_pool1d or avg_pool2d
1931
+ avg_pool1d: bool = node.target == exir_ops.edge.aten.avg_pool1d.default
1932
+
1933
+ # Get the input tensor node
1934
+ in_tensor_node = node.args[0]
1935
+ assert isinstance(in_tensor_node, torch.fx.Node)
1936
+
1937
+ # Replace avg_pool2d with custom avg_pool2d, and if the input tensor is
1938
+ # quantized, pass its zero_point tensor as arg to the custom avg_pool2d.
1939
+ # stride, padding, ceil_mode, count_include_pad, divisor_override, are
1940
+ # the native avg_pool2d args. 'channel_last' denotes NCHW vs NHWC layout,
1941
+ # and is False by default.
1942
+ kernel_size = node.args[1]
1943
+ # When stride is not provided or is empty, PyTorch defaults to kernel_size
1944
+ stride = node.args[2] if len(node.args) >= 3 and node.args[2] else kernel_size
1945
+ padding = node.args[3] if len(node.args) >= 4 else [0, 0]
1946
+ ceil_mode = node.args[4] if len(node.args) >= 5 else False
1947
+ count_include_pad = node.args[5] if len(node.args) >= 6 else True
1948
+ divisor_override = node.args[6] if len(node.args) >= 7 else None
1949
+ zero_point = node.args[7] if len(node.args) >= 8 else None
1950
+
1951
+ graph = node.graph
1952
+ out_shape = node.meta["val"].shape
1953
+
1954
+ kernel_size = cast(Sequence[int], kernel_size)
1955
+ stride = cast(Sequence[int], stride)
1956
+ padding = cast(Sequence[int], padding)
1957
+
1958
+ # If the op is avg_pool1d, then we need to reshape the 3d input to a 4d
1959
+ # tensor.
1960
+ if avg_pool1d:
1961
+ in_shape = list(in_tensor_node.meta["val"].shape)
1962
+ assert len(in_shape) == 3, "Expected 3d input for avg_pool1d"
1963
+ in_shape_4d = in_shape[:2] + [1] + in_shape[2:]
1964
+
1965
+ with graph.inserting_before(node):
1966
+ in_view_node = graph.call_function(
1967
+ exir_ops.edge.aten.view_copy.default,
1968
+ args=(in_tensor_node, in_shape_4d),
1969
+ )
1970
+ in_view_node.meta = node.meta
1971
+
1972
+ # Extend the kernel_size, stride and padding to 2d
1973
+ kernel_size = (
1974
+ [1] + list(kernel_size) if len(kernel_size) == 1 else kernel_size
1975
+ )
1976
+ stride = [1] + list(stride) if len(stride) == 1 else stride
1977
+ padding = [0] + list(padding) if len(padding) == 1 else padding
1978
+
1979
+ input_for_pool = in_view_node
1980
+ else:
1981
+ input_for_pool = in_tensor_node
1982
+
1983
+ # Create a new avg_pool node with the updated args
1984
+ new_args = (
1985
+ input_for_pool,
1986
+ kernel_size,
1987
+ stride,
1988
+ padding,
1989
+ ceil_mode,
1990
+ count_include_pad,
1991
+ divisor_override,
1992
+ zero_point,
1993
+ False,
1994
+ )
1995
+
1996
+ with graph.inserting_before(node):
1997
+ avg_pool2d_node = graph.call_function(
1998
+ exir_ops.edge.cadence.avg_pool2d.default,
1999
+ args=new_args,
2000
+ )
2001
+ avg_pool2d_node.meta = node.meta
2002
+
2003
+ # If the node was avg_pool1d, we again reshape the 4d output to 3d output
2004
+ if avg_pool1d:
2005
+ with graph.inserting_before(node):
2006
+ result_node = graph.call_function(
2007
+ exir_ops.edge.aten.view_copy.default,
2008
+ args=(avg_pool2d_node, list(out_shape)),
2009
+ )
2010
+ result_node.meta = node.meta
2011
+ node.replace_all_uses_with(result_node)
2012
+ else:
2013
+ node.replace_all_uses_with(avg_pool2d_node)
2014
+
2015
+ return True
2016
+
2017
+
2018
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
2019
+ class ReplaceIm2RowWithViewPass(RemoveOrReplacePassInterface):
2020
+ """
2021
+ Replace im2row with view when possible (no padding, no dilation, and output spatial dimensions are 1).
2022
+ """
2023
+
2024
+ @property
2025
+ def targets(self) -> list[EdgeOpOverload]:
2026
+ return [
2027
+ exir_ops.edge.cadence.im2row.default,
2028
+ exir_ops.edge.cadence.im2row.per_tensor,
2029
+ ]
2030
+
2031
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2032
+ # Check if im2row applies padding. If yes, we cannot replace it with view.
2033
+ pad = cast(Sequence[int], node.args[3])
2034
+ if any(p != 0 for p in pad):
2035
+ return False
2036
+
2037
+ # Check if im2row has dilation. If yes, we cannot replace it with view.
2038
+ dilation = cast(Sequence[int], node.args[2])
2039
+ if any(d != 1 for d in dilation):
2040
+ return False
2041
+
2042
+ # When channel_last=True (NHWC layout), im2row rearranges data from
2043
+ # kp-major (NHWC natural order) to channel-major output layout.
2044
+ # A simple view_copy cannot perform this data rearrangement.
2045
+ channel_last = node.args[6] if len(node.args) > 6 else False
2046
+ if channel_last:
2047
+ return False
2048
+
2049
+ # im2row works on 3D or 4D tensors.
2050
+ # Output shape[1:-1] will be unit if input spatial dimensions are the same as kernel spatial dimensions.
2051
+ output_shape = node.meta["val"].shape
2052
+ if math.prod(output_shape[1:-1]) != 1:
2053
+ return False
2054
+
2055
+ # Replace im2row with view_copy
2056
+ with node.graph.inserting_before(node):
2057
+ new_node = node.graph.call_function(
2058
+ exir_ops.edge.aten.view_copy.default,
2059
+ args=(node.args[0], list(output_shape)),
2060
+ )
2061
+ new_node.meta = node.meta
2062
+
2063
+ node.replace_all_uses_with(new_node)
2064
+ return True
2065
+
2066
+
2067
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
2068
+ class ReplaceEmptyTensorsWithFullPass(ExportPass):
2069
+ """Replaces nodes that produce empty tensors with full nodes."""
2070
+
2071
+ def call_operator(self, op, args, kwargs, meta):
2072
+ val = meta.data.get("val", None)
2073
+ if isinstance(val, torch.Tensor) and val.numel() == 0:
2074
+ return super().call_operator(
2075
+ exir_ops.edge.aten.full.default,
2076
+ args=(val.shape, 0),
2077
+ kwargs={"dtype": val.dtype},
2078
+ meta=meta,
2079
+ )
2080
+ return super().call_operator(op, args, kwargs, meta)
2081
+
2082
+ def call(self, graph_module: torch.fx.GraphModule) -> PassResult:
2083
+ changed = False
2084
+ for module in filter(
2085
+ lambda m: isinstance(m, torch.fx.GraphModule), graph_module.modules()
2086
+ ):
2087
+ module = cast(torch.fx.GraphModule, module)
2088
+ for node in module.graph.nodes:
2089
+ if node.op != "call_function":
2090
+ continue
2091
+ val = node.meta.get("val", None)
2092
+ if isinstance(val, torch.Tensor) and val.numel() == 0:
2093
+ with module.graph.inserting_before(node):
2094
+ new_node = module.graph.call_function(
2095
+ exir_ops.edge.aten.full.default,
2096
+ args=(val.shape, 0),
2097
+ kwargs={"dtype": val.dtype},
2098
+ )
2099
+ new_node.meta = node.meta
2100
+ node.replace_all_uses_with(new_node)
2101
+ changed = True
2102
+
2103
+ if changed:
2104
+ graph_module.graph.eliminate_dead_code()
2105
+ graph_module.recompile()
2106
+ return super().call(graph_module)
2107
+
2108
+ return PassResult(graph_module, False)
2109
+
2110
+
2111
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
2112
+ class ReplaceWhereWithFullArgsWithWhereScalar(RemoveOrReplacePassInterface):
2113
+ """Replaces where ops using two full ops as tensors with a scalar
2114
+ version.
2115
+ """
2116
+
2117
+ @property
2118
+ def targets(self) -> list[EdgeOpOverload]:
2119
+ return [exir_ops.edge.aten.where.self]
2120
+
2121
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2122
+ # Check if args[1] and args[2] are full ops
2123
+ arg1 = node.args[1]
2124
+ arg2 = node.args[2]
2125
+
2126
+ if not isinstance(arg1, torch.fx.Node) or not isinstance(arg2, torch.fx.Node):
2127
+ return False
2128
+
2129
+ if (
2130
+ arg1.target != exir_ops.edge.aten.full.default
2131
+ or arg2.target != exir_ops.edge.aten.full.default
2132
+ ):
2133
+ return False
2134
+
2135
+ # Get the condition tensor shape
2136
+ cond_arg = node.args[0]
2137
+ assert isinstance(cond_arg, torch.fx.Node)
2138
+ cond_shape = list(cond_arg.meta["val"].shape)
2139
+
2140
+ # Check if the full ops have the same size as the cond tensor
2141
+ full1_shape = arg1.args[0]
2142
+ full2_shape = arg2.args[0]
2143
+
2144
+ if cond_shape != full1_shape or cond_shape != full2_shape:
2145
+ return False
2146
+
2147
+ # Get the scalar values from the full ops
2148
+ scalar_value_1 = arg1.args[1]
2149
+ scalar_value_2 = arg2.args[1]
2150
+
2151
+ # Replace the where op with a scalar where op
2152
+ with node.graph.inserting_before(node):
2153
+ new_node = node.graph.call_function(
2154
+ exir_ops.edge.cadence.where_Scalar.default,
2155
+ args=(cond_arg, scalar_value_1, scalar_value_2),
2156
+ )
2157
+ new_node.meta = node.meta
2158
+
2159
+ node.replace_all_uses_with(new_node)
2160
+ return True
2161
+
2162
+
2163
+ # Adapted from fbcode/pyspeech/opt_passes/replace_ops.py
2164
+ @register_cadence_pass(CadencePassAttribute(opt_level=2))
2165
+ class ReplaceSplitWithSlicePass(RemoveOrReplacePassInterface):
2166
+ """
2167
+ split_with_sizes() delegates to slice() op, so perform this replacement here.
2168
+ This avoids the expense of delegation from ATen.
2169
+ """
2170
+
2171
+ @property
2172
+ def targets(self) -> list[EdgeOpOverload]:
2173
+ return [exir_ops.edge.aten.split_with_sizes_copy.default]
2174
+
2175
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2176
+ # All the users of this split_with_sizes op must be getitem ops
2177
+ if any(user.target != operator.getitem for user in node.users):
2178
+ return False
2179
+
2180
+ # Get the slice dim and extent for each split
2181
+ slice_ops = self._get_split_sizes(node)
2182
+ if slice_ops is None:
2183
+ return False
2184
+
2185
+ graph = node.graph
2186
+
2187
+ # Go over each getitem user, and replace it with slice op
2188
+ for user in list(node.users.keys()):
2189
+ assert user.target == operator.getitem
2190
+ item_idx = int(user.args[1])
2191
+ assert item_idx < len(slice_ops)
2192
+ cur_slice = slice_ops[item_idx]
2193
+ with graph.inserting_before(user):
2194
+ cur_slice_node = graph.call_function(
2195
+ exir_ops.edge.aten.slice_copy.Tensor,
2196
+ (node.args[0], cur_slice[0], cur_slice[1], cur_slice[2], 1),
2197
+ )
2198
+ # Metadata copy important
2199
+ cur_slice_node.meta = user.meta
2200
+ user.replace_all_uses_with(cur_slice_node)
2201
+
2202
+ # Return True to indicate the split node should be removed
2203
+ return True
2204
+
2205
+ def _get_split_sizes(self, node: torch.fx.Node) -> Optional[list[tuple[int, ...]]]:
2206
+ """For split_with_sizes, return the slice dim and extent for each split."""
2207
+ # Parse the args of the split_with_sizes op
2208
+ tensor_arg, split_sizes = node.args[0:2]
2209
+ assert isinstance(tensor_arg, torch.fx.Node)
2210
+
2211
+ # Get shape from node metadata
2212
+ val = tensor_arg.meta.get("val")
2213
+ if val is None:
2214
+ return None
2215
+ in_shape = val.shape
2216
+
2217
+ split_dim = 0 if len(node.args) < 3 else node.args[2]
2218
+
2219
+ # Canonicalize the split dimension
2220
+ assert isinstance(split_dim, int)
2221
+ split_dim = split_dim if split_dim >= 0 else len(in_shape) + split_dim
2222
+
2223
+ # Create the slice op args corresponding to each split
2224
+ slice_ops = []
2225
+ split_start = 0
2226
+ assert isinstance(split_sizes, list)
2227
+ for split_size in split_sizes:
2228
+ split_end = split_start + split_size
2229
+ slice_args = (split_dim, split_start, split_end)
2230
+ slice_ops.append(slice_args)
2231
+ split_start = split_end
2232
+
2233
+ return slice_ops
2234
+
2235
+
2236
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
2237
+ class ReplacePowWithMulPass(RemoveOrReplacePassInterface):
2238
+ """
2239
+ Replace the pow op with successive mul ops when the exponent is an
2240
+ integer between 2 and 4 (inclusive). Float exponents that are whole
2241
+ numbers (e.g., 2.0, 3.0, 4.0) are also accepted.
2242
+ """
2243
+
2244
+ @property
2245
+ def targets(self) -> list[EdgeOpOverload]:
2246
+ return [exir_ops.edge.aten.pow.Tensor_Scalar]
2247
+
2248
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2249
+ # Check if we have at least 2 args and the exponent is an int or float
2250
+ if len(node.args) < 2 or not isinstance(node.args[1], (int, float)):
2251
+ return False
2252
+
2253
+ exponent_val = node.args[1]
2254
+ if isinstance(exponent_val, float):
2255
+ if not exponent_val.is_integer():
2256
+ return False
2257
+ exponent_val = int(exponent_val)
2258
+ exponent = cast(int, exponent_val)
2259
+
2260
+ # Only replace if exponent is between 2 and 4 (inclusive)
2261
+ if exponent < 2 or exponent > 4:
2262
+ return False
2263
+
2264
+ x = node.args[0]
2265
+ assert isinstance(x, torch.fx.Node)
2266
+
2267
+ graph = node.graph
2268
+ result_node = x
2269
+
2270
+ # Create successive mul operations
2271
+ # For exponent=2: x * x (1 mul)
2272
+ # For exponent=3: (x * x) * x (2 muls)
2273
+ # For exponent=4: ((x * x) * x) * x (3 muls)
2274
+ for _ in range(exponent - 1):
2275
+ with graph.inserting_before(node):
2276
+ result_node = graph.call_function(
2277
+ exir_ops.edge.aten.mul.Tensor,
2278
+ args=(result_node, x),
2279
+ )
2280
+ result_node.meta = node.meta
2281
+
2282
+ node.replace_all_uses_with(result_node)
2283
+ return True
2284
+
2285
+
2286
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
2287
+ class ReplaceMatmulWithTransposedMatmulPass(RemoveOrReplacePassInterface):
2288
+ """
2289
+ For certain backends, we have efficient kernels for transposed matmul. We
2290
+ replace AxB with AxB' for such backends.
2291
+ """
2292
+
2293
+ @property
2294
+ def targets(self) -> list[EdgeOpOverload]:
2295
+ return [exir_ops.edge.cadence.quantized_matmul.default]
2296
+
2297
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2298
+ # If already transposed, bail
2299
+ if len(node.args) >= 9 and node.args[-1] is True:
2300
+ return False
2301
+
2302
+ # Get the args
2303
+ if len(node.args) == 9:
2304
+ (
2305
+ X_arg,
2306
+ X_zero_point,
2307
+ Y_arg,
2308
+ Y_zero_point,
2309
+ bias,
2310
+ out_multiplier,
2311
+ out_shift,
2312
+ out_zero_point,
2313
+ transposed,
2314
+ ) = node.args
2315
+ elif len(node.args) == 8:
2316
+ (
2317
+ X_arg,
2318
+ X_zero_point,
2319
+ Y_arg,
2320
+ Y_zero_point,
2321
+ bias,
2322
+ out_multiplier,
2323
+ out_shift,
2324
+ out_zero_point,
2325
+ ) = node.args
2326
+ transposed = False
2327
+ else:
2328
+ raise AssertionError(
2329
+ f"Unexpected number of args for quantized_matmul: {len(node.args)}"
2330
+ )
2331
+
2332
+ # If the matmul is already transposed, bail
2333
+ if transposed:
2334
+ return False
2335
+
2336
+ # Get the second tensor from metadata
2337
+ assert isinstance(Y_arg, torch.fx.Node)
2338
+ Y_tensor_val = Y_arg.meta.get("val")
2339
+ if Y_tensor_val is None:
2340
+ return False
2341
+
2342
+ graph = node.graph
2343
+
2344
+ # Create zero bias
2345
+ with graph.inserting_before(node):
2346
+ zero_bias = graph.call_function(
2347
+ exir_ops.edge.aten.full.default,
2348
+ args=([Y_tensor_val.size(-1)], 0),
2349
+ kwargs={"dtype": torch.int32},
2350
+ )
2351
+ zero_bias.meta = node.meta
2352
+
2353
+ # Transpose Y_arg
2354
+ with graph.inserting_before(node):
2355
+ Y_arg_t = graph.call_function(
2356
+ exir_ops.edge.aten.transpose_copy.int,
2357
+ args=(Y_arg, -1, -2),
2358
+ )
2359
+ Y_arg_t.meta = node.meta
2360
+
2361
+ # Construct the new args, and create the transposed matmul op
2362
+ new_args = (
2363
+ X_arg,
2364
+ X_zero_point,
2365
+ Y_arg_t,
2366
+ Y_zero_point,
2367
+ zero_bias,
2368
+ out_multiplier,
2369
+ out_shift,
2370
+ out_zero_point,
2371
+ True,
2372
+ )
2373
+
2374
+ with graph.inserting_before(node):
2375
+ new_node = graph.call_function(
2376
+ exir_ops.edge.cadence.quantized_matmul.default,
2377
+ args=new_args,
2378
+ kwargs=node.kwargs,
2379
+ )
2380
+ new_node.meta = node.meta
2381
+
2382
+ node.replace_all_uses_with(new_node)
2383
+ return True
2384
+
2385
+ def call(self, graph_module: torch.fx.GraphModule) -> PassResult:
2386
+ modified = False
2387
+ result = super().call(graph_module)
2388
+ modified = modified or result.modified
2389
+ if modified:
2390
+ # Fuse any inserted transpose node with transpose/permute nodes
2391
+ # surrounding it.
2392
+ result = FuseCascadedTransposeOrPermuteOps().call(result.graph_module)
2393
+ modified = modified or result.modified
2394
+ # Replace permute with transpose.
2395
+ result = ReplacePermuteWithTransposePass().call(result.graph_module)
2396
+ modified = modified or result.modified
2397
+
2398
+ return PassResult(result.graph_module, modified)
2399
+
2400
+
2401
+ @register_cadence_pass(CadencePassAttribute(opt_level=1))
2402
+ class ReplaceMulTensorWithMulAndFullOpsPass(RemoveOrReplacePassInterface):
2403
+ """
2404
+ Extracts a single value argument of mul op to a separate full op.
2405
+ """
2406
+
2407
+ @property
2408
+ def targets(self) -> list[EdgeOpOverload]:
2409
+ return [torch.ops.aten.mul.Tensor]
2410
+
2411
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2412
+ x_arg, const_arg = node.args
2413
+
2414
+ # Swap arguments if the order is wrong
2415
+ if isinstance(const_arg, torch.fx.Node):
2416
+ x_arg, const_arg = const_arg, x_arg
2417
+
2418
+ # Skip if the const_arg is not a scalar
2419
+ if not isinstance(const_arg, (float, int)) or not isinstance(
2420
+ x_arg, torch.fx.Node
2421
+ ):
2422
+ return False
2423
+
2424
+ # Cast the const_arg to the dtype of the x_arg
2425
+ full_arg = self.resolve_full_arg(x_arg, const_arg)
2426
+
2427
+ full_output_dtype = torch.int32 if isinstance(full_arg, int) else torch.float32
2428
+
2429
+ # Extract an argument to a separate full op.
2430
+ with node.graph.inserting_before(node):
2431
+ full_node = node.graph.call_function(
2432
+ torch.ops.aten.full.default,
2433
+ args=([1], full_arg),
2434
+ kwargs={"dtype": full_output_dtype},
2435
+ )
2436
+ full_node.meta = node.meta.copy()
2437
+ # Create a proper FakeTensor for metadata instead of Python list
2438
+ fake_mode = node.meta["val"].fake_mode
2439
+ if fake_mode is not None:
2440
+ with fake_mode:
2441
+ full_node.meta["val"] = torch.full(
2442
+ [1], full_arg, dtype=full_output_dtype
2443
+ )
2444
+ else:
2445
+ full_node.meta["val"] = torch.empty(
2446
+ [1], dtype=full_output_dtype, device="meta"
2447
+ )
2448
+ new_mul_node = node.graph.call_function(
2449
+ torch.ops.aten.mul.Tensor, args=(x_arg, full_node)
2450
+ )
2451
+ new_mul_node.meta = node.meta
2452
+ # Replace the old mul with a newly created mul.
2453
+ node.replace_all_uses_with(new_mul_node)
2454
+ node.graph.erase_node(node)
2455
+ return True
2456
+
2457
+ def resolve_full_arg(
2458
+ self, x_arg: torch.fx.Node, const_arg: float | int
2459
+ ) -> float | int:
2460
+ if x_arg.meta["val"].dtype == torch.float32 and isinstance(const_arg, int):
2461
+ const_arg = float(const_arg)
2462
+ if x_arg.meta["val"].dtype == torch.int32 and isinstance(const_arg, float):
2463
+ const_arg = int(const_arg)
2464
+ return const_arg
2465
+
2466
+
2467
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
2468
+ class ReplaceAdaptiveAvgPoolWithAtenAvgPoolPass(RemoveOrReplacePassInterface):
2469
+ """
2470
+ Replace the aten adaptive avg_pool op with the aten avg_pool2d op.
2471
+ """
2472
+
2473
+ @property
2474
+ def targets(self) -> list[EdgeOpOverload]:
2475
+ return [exir_ops.edge.aten._adaptive_avg_pool2d.default]
2476
+
2477
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2478
+ # Get the input tensor node
2479
+ in_tensor_node = node.args[0]
2480
+ assert isinstance(in_tensor_node, torch.fx.Node)
2481
+
2482
+ # Get input shape (in NCHW format)
2483
+ in_shape = in_tensor_node.meta["val"].shape
2484
+ output_size = cast(Sequence[int], node.args[1])
2485
+ num_dims = len(output_size)
2486
+
2487
+ # Spatial dimensions are at indices [2:] for NCHW format
2488
+ # TODO: If in_tensor_shape is not a multiple of output size,
2489
+ # this pass will not work. T224984800
2490
+ dim_multiples = [
2491
+ (in_shape[i + 2] % output_size[i]) == 0 for i in range(num_dims)
2492
+ ]
2493
+ if not all(dim_multiples):
2494
+ logging.info(
2495
+ f"Unable to replace adaptive average pool with average pool. Input tensor shape of {in_shape} is not a multiple of output size: {output_size}"
2496
+ )
2497
+ return False
2498
+
2499
+ # Compute stride and kernel_size based on spatial dimensions
2500
+ stride = [(in_shape[i + 2] // output_size[i]) for i in range(num_dims)]
2501
+ kernel_size = [
2502
+ in_shape[i + 2] - (output_size[i] - 1) * stride[i] for i in range(num_dims)
2503
+ ]
2504
+ padding = [0] * num_dims
2505
+ ceil_mode = False
2506
+ count_include_pad = True
2507
+ divisor_override = None
2508
+
2509
+ # Create a new avg_pool2d node with the computed args
2510
+ new_args = (
2511
+ in_tensor_node,
2512
+ kernel_size,
2513
+ stride,
2514
+ padding,
2515
+ ceil_mode,
2516
+ count_include_pad,
2517
+ divisor_override,
2518
+ )
2519
+
2520
+ with node.graph.inserting_before(node):
2521
+ new_node = node.graph.call_function(
2522
+ exir_ops.edge.aten.avg_pool2d.default,
2523
+ args=new_args,
2524
+ )
2525
+ new_node.meta = node.meta
2526
+
2527
+ node.replace_all_uses_with(new_node)
2528
+ return True
2529
+
2530
+
2531
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
2532
+ class ReplaceTorchQuantizedEmbeddingWithCadenceQuantizedEmbedding(
2533
+ RemoveOrReplacePassInterface
2534
+ ):
2535
+ """
2536
+ Replace torch.ops.quantized_decomposed.embedding_byte.dtype with
2537
+ torch.ops.cadence.quantized_embedding_byte
2538
+ """
2539
+
2540
+ @property
2541
+ def targets(self) -> list[EdgeOpOverload]:
2542
+ return [
2543
+ exir_ops.edge.quantized_decomposed.embedding_byte.default,
2544
+ exir_ops.edge.quantized_decomposed.embedding_byte.dtype,
2545
+ ]
2546
+
2547
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2548
+ # Replace with cadence.quantized_embedding_byte
2549
+ if len(node.args) < 6:
2550
+ raise AssertionError(
2551
+ f"Expected 6 arguments for embedding_byte, got {len(node.args)}"
2552
+ )
2553
+ embedding = node.args[0]
2554
+ scales = node.args[1]
2555
+ weight_zero_points = node.args[2]
2556
+ indices = node.args[5]
2557
+
2558
+ if node.target == exir_ops.edge.quantized_decomposed.embedding_byte.dtype:
2559
+ dtype = node.kwargs.get("dtype", None)
2560
+ if dtype is not None and dtype != torch.float32:
2561
+ raise AssertionError(
2562
+ f"Unsupported output dtype for embedding_byte: {dtype}"
2563
+ )
2564
+
2565
+ new_args = (embedding, scales, weight_zero_points, indices, False)
2566
+ with node.graph.inserting_before(node):
2567
+ new_node = node.graph.call_function(
2568
+ exir_ops.edge.cadence.quantized_embedding_byte.default,
2569
+ args=new_args,
2570
+ )
2571
+ new_node.meta = node.meta
2572
+
2573
+ node.replace_all_uses_with(new_node)
2574
+ return True
2575
+
2576
+
2577
+ class CommonReplacePasses:
2578
+ passes = [
2579
+ ReplaceScalarWithTensorArgPass,
2580
+ ReplaceSqueezeAndUnsqueezeWithViewPass,
2581
+ ReplaceSplitWithSlicePass,
2582
+ ReplaceSelectWithViewOpPass,
2583
+ ReplaceMMWithAddMMPass,
2584
+ ReplaceRepeatWithCatPass,
2585
+ ReplaceFullLikeWithFullPass,
2586
+ ReplaceAtenConvolutionWithCadenceConvolutionPass,
2587
+ ReplacePT2QuantWithCadenceQuantPass,
2588
+ ReplacePT2DequantWithCadenceDequantPass,
2589
+ ReplacePowWithMulPass,
2590
+ ReplaceTorchQuantizedEmbeddingWithCadenceQuantizedEmbedding,
2591
+ ]
2592
+
2593
+
2594
+ @register_cadence_pass(CadencePassAttribute(opt_level=0))
2595
+ class ReplaceAtenLinalgSvdWithCadenceLinalgSvdPass(RemoveOrReplacePassInterface):
2596
+ """
2597
+ Replace aten linalg svd op with cadence custom op.
2598
+ """
2599
+
2600
+ @property
2601
+ def targets(self) -> list[EdgeOpOverload]:
2602
+ return [exir_ops.edge.aten._linalg_svd.default]
2603
+
2604
+ def maybe_remove_or_replace(self, node: torch.fx.Node) -> bool:
2605
+ with node.graph.inserting_before(node):
2606
+ new_node = node.graph.call_function(
2607
+ exir_ops.edge.cadence.linalg_svd.default,
2608
+ args=node.args,
2609
+ kwargs=node.kwargs,
2610
+ )
2611
+ new_node.meta = node.meta
2612
+ node.replace_all_uses_with(new_node)
2613
+ return True
2614
+
2615
+
2616
+ # This class encapsulates all the functions that replace/switch one op in the
2617
+ # graph with another.
2618
+ class CadenceReplaceOpsInGraph:
2619
+ passes = CommonReplacePasses.passes + [
2620
+ ReplaceAtenLinalgSvdWithCadenceLinalgSvdPass,
2621
+ ReplaceEmptyTensorsWithFullPass,
2622
+ ReplaceFunctionallyEquivalentOpTargets,
2623
+ ReplacePermuteWithTransposePass,
2624
+ ReplaceConvolutionOptionalArgsWithConcreteArgsPass,
2625
+ ReplaceAddMMWithLinearPass,
2626
+ ReplacePadWithCatPass,
2627
+ ReplaceConstantPadNdWithSlicePass,
2628
+ ReplaceConvWithChannelLastConvPass,
2629
+ ReplaceMaxPool2dWithChannelLastMaxPool2dPass,
2630
+ ReplaceTrivialConvWithLinear,
2631
+ ReplaceConvWithIm2RowAndLinear,
2632
+ ReplaceTransposedConvWithLinearPass,
2633
+ # This pass should be after passes that replace conv -> im2row + linear.
2634
+ ReplaceIm2RowWithViewPass,
2635
+ MakeSliceAndCatDimOutermostPass,
2636
+ ReplaceMatmulWithTransposedMatmulPass,
2637
+ ReplaceNopTransposeOrPermuteWithViewPass,
2638
+ ReplaceLinearWithFullyConnectedOpPass,
2639
+ ReplaceScalarTensorWithFullPass,
2640
+ ReplaceInfArgInFullWithValuePass,
2641
+ ReplaceLogicalNotBooleanWhereWithWherePass,
2642
+ ReplaceAdaptiveAvgPoolWithAtenAvgPoolPass,
2643
+ ReplaceAtenAvgPoolWithCadenceAvgPoolPass,
2644
+ ReplaceWhereWithFullArgsWithWhereScalar,
2645
+ ReplaceMulTensorWithMulAndFullOpsPass,
2646
+ ]