executorch 1.1.0__cp313-cp313-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 (2784) hide show
  1. executorch/backends/__init__.py +0 -0
  2. executorch/backends/aoti/aoti_backend.py +271 -0
  3. executorch/backends/aoti/aoti_partitioner.py +109 -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 +588 -0
  7. executorch/backends/apple/coreml/compiler/enumerated_shape_utils.py +244 -0
  8. executorch/backends/apple/coreml/compiler/torch_ops.py +219 -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 +307 -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_partitioner.py +348 -0
  23. executorch/backends/apple/coreml/test/test_coreml_quantizer.py +172 -0
  24. executorch/backends/apple/coreml/test/test_coreml_recipes.py +552 -0
  25. executorch/backends/apple/coreml/test/test_coreml_utils.py +19 -0
  26. executorch/backends/apple/coreml/test/test_enumerated_shapes.py +112 -0
  27. executorch/backends/apple/coreml/test/test_torch_ops.py +333 -0
  28. executorch/backends/apple/coreml/test/tester.py +121 -0
  29. executorch/backends/apple/metal/metal_backend.py +67 -0
  30. executorch/backends/apple/metal/metal_partitioner.py +25 -0
  31. executorch/backends/apple/metal/tests/__init__.py +5 -0
  32. executorch/backends/apple/metal/tests/test_metal_backend.py +79 -0
  33. executorch/backends/apple/metal/tests/test_metal_partitioner.py +172 -0
  34. executorch/backends/apple/mps/__init__.py +10 -0
  35. executorch/backends/apple/mps/mps_preprocess.py +301 -0
  36. executorch/backends/apple/mps/operators/__init__.py +77 -0
  37. executorch/backends/apple/mps/operators/activation_ops.py +99 -0
  38. executorch/backends/apple/mps/operators/binary_ops.py +156 -0
  39. executorch/backends/apple/mps/operators/clamp_ops.py +60 -0
  40. executorch/backends/apple/mps/operators/constant_ops.py +128 -0
  41. executorch/backends/apple/mps/operators/convolution_ops.py +70 -0
  42. executorch/backends/apple/mps/operators/indexing_ops.py +225 -0
  43. executorch/backends/apple/mps/operators/linear_algebra_ops.py +52 -0
  44. executorch/backends/apple/mps/operators/node_visitor.py +473 -0
  45. executorch/backends/apple/mps/operators/normalization_ops.py +97 -0
  46. executorch/backends/apple/mps/operators/op_clone.py +54 -0
  47. executorch/backends/apple/mps/operators/op_getitem.py +29 -0
  48. executorch/backends/apple/mps/operators/op_quant_dequant.py +142 -0
  49. executorch/backends/apple/mps/operators/op_skip_ops.py +24 -0
  50. executorch/backends/apple/mps/operators/pad_ops.py +37 -0
  51. executorch/backends/apple/mps/operators/pooling_ops.py +142 -0
  52. executorch/backends/apple/mps/operators/range_ops.py +53 -0
  53. executorch/backends/apple/mps/operators/reduce_ops.py +39 -0
  54. executorch/backends/apple/mps/operators/shape_ops.py +263 -0
  55. executorch/backends/apple/mps/operators/unary_ops.py +130 -0
  56. executorch/backends/apple/mps/partition/__init__.py +10 -0
  57. executorch/backends/apple/mps/partition/mps_partitioner.py +133 -0
  58. executorch/backends/apple/mps/serialization/mps_graph_schema.py +817 -0
  59. executorch/backends/apple/mps/serialization/mps_graph_serialize.py +34 -0
  60. executorch/backends/apple/mps/serialization/schema.fbs +523 -0
  61. executorch/backends/apple/mps/test/test_mps.py +2507 -0
  62. executorch/backends/apple/mps/test/test_mps_binary_ops.py +296 -0
  63. executorch/backends/apple/mps/test/test_mps_indexing_ops.py +264 -0
  64. executorch/backends/apple/mps/test/test_mps_linear.py +523 -0
  65. executorch/backends/apple/mps/test/test_mps_models.py +416 -0
  66. executorch/backends/apple/mps/test/test_mps_unary_ops.py +26 -0
  67. executorch/backends/apple/mps/test/test_mps_utils.py +326 -0
  68. executorch/backends/apple/mps/utils/mps_utils.py +82 -0
  69. executorch/backends/apple/mps/utils/quant_utils.py +41 -0
  70. executorch/backends/arm/_passes/__init__.py +134 -0
  71. executorch/backends/arm/_passes/_debug_passes.py +58 -0
  72. executorch/backends/arm/_passes/annotate_decomposed_matmul.py +127 -0
  73. executorch/backends/arm/_passes/annotate_output_dim_order_pass.py +26 -0
  74. executorch/backends/arm/_passes/arm_pass.py +95 -0
  75. executorch/backends/arm/_passes/arm_pass_manager.py +462 -0
  76. executorch/backends/arm/_passes/arm_pass_utils.py +281 -0
  77. executorch/backends/arm/_passes/broadcast_args_pass.py +72 -0
  78. executorch/backends/arm/_passes/cast_int64_pass.py +70 -0
  79. executorch/backends/arm/_passes/cast_to_int32_pass.py +67 -0
  80. executorch/backends/arm/_passes/conv1d_unsqueeze_pass.py +98 -0
  81. executorch/backends/arm/_passes/convert_elu_params.py +67 -0
  82. executorch/backends/arm/_passes/convert_expand_copy_to_repeat.py +77 -0
  83. executorch/backends/arm/_passes/convert_full_like_to_full_pass.py +42 -0
  84. executorch/backends/arm/_passes/convert_int64_const_ops_to_int32.py +81 -0
  85. executorch/backends/arm/_passes/convert_int64_output_ops_to_int32.py +155 -0
  86. executorch/backends/arm/_passes/convert_minmax_pass.py +166 -0
  87. executorch/backends/arm/_passes/convert_permute_singleton_to_view_pass.py +64 -0
  88. executorch/backends/arm/_passes/convert_split_to_slice.py +132 -0
  89. executorch/backends/arm/_passes/convert_squeezes_to_view.py +36 -0
  90. executorch/backends/arm/_passes/convert_to_clamp_pass.py +45 -0
  91. executorch/backends/arm/_passes/decompose_acosh_pass.py +77 -0
  92. executorch/backends/arm/_passes/decompose_adaptive_avg_pool2d_pass.py +112 -0
  93. executorch/backends/arm/_passes/decompose_add_sub_alpha_pass.py +98 -0
  94. executorch/backends/arm/_passes/decompose_addmm_pass.py +72 -0
  95. executorch/backends/arm/_passes/decompose_any_pass.py +114 -0
  96. executorch/backends/arm/_passes/decompose_asin_and_acos_pass.py +236 -0
  97. executorch/backends/arm/_passes/decompose_asinh_pass.py +74 -0
  98. executorch/backends/arm/_passes/decompose_atan_pass.py +142 -0
  99. executorch/backends/arm/_passes/decompose_atanh_pass.py +86 -0
  100. executorch/backends/arm/_passes/decompose_avg_pool2d_pass.py +154 -0
  101. executorch/backends/arm/_passes/decompose_batch_norm_no_stats.py +233 -0
  102. executorch/backends/arm/_passes/decompose_cosh_pass.py +72 -0
  103. executorch/backends/arm/_passes/decompose_cosine_similarity_pass.py +92 -0
  104. executorch/backends/arm/_passes/decompose_cumsum_pass.py +167 -0
  105. executorch/backends/arm/_passes/decompose_div_pass.py +61 -0
  106. executorch/backends/arm/_passes/decompose_div_tensor_mode.py +94 -0
  107. executorch/backends/arm/_passes/decompose_elu_pass.py +107 -0
  108. executorch/backends/arm/_passes/decompose_embedding_pass.py +126 -0
  109. executorch/backends/arm/_passes/decompose_expm1_pass.py +163 -0
  110. executorch/backends/arm/_passes/decompose_floor_divide_pass.py +75 -0
  111. executorch/backends/arm/_passes/decompose_gelu_pass.py +172 -0
  112. executorch/backends/arm/_passes/decompose_glu_pass.py +81 -0
  113. executorch/backends/arm/_passes/decompose_grouped_conv_pass.py +185 -0
  114. executorch/backends/arm/_passes/decompose_groupnorm_pass.py +227 -0
  115. executorch/backends/arm/_passes/decompose_int16_activation_conv_pass.py +137 -0
  116. executorch/backends/arm/_passes/decompose_int_pow_pass.py +64 -0
  117. executorch/backends/arm/_passes/decompose_layernorm_pass.py +203 -0
  118. executorch/backends/arm/_passes/decompose_leaky_relu_pass.py +76 -0
  119. executorch/backends/arm/_passes/decompose_linalg_vector_norm_pass.py +92 -0
  120. executorch/backends/arm/_passes/decompose_linear_pass.py +109 -0
  121. executorch/backends/arm/_passes/decompose_log1p_pass.py +97 -0
  122. executorch/backends/arm/_passes/decompose_logit_pass.py +119 -0
  123. executorch/backends/arm/_passes/decompose_masked_fill_pass.py +58 -0
  124. executorch/backends/arm/_passes/decompose_maxpool2d_with_dilation_pass.py +218 -0
  125. executorch/backends/arm/_passes/decompose_meandim_pass.py +283 -0
  126. executorch/backends/arm/_passes/decompose_ne_pass.py +74 -0
  127. executorch/backends/arm/_passes/decompose_quant_nodes.py +156 -0
  128. executorch/backends/arm/_passes/decompose_remainder_pass.py +70 -0
  129. executorch/backends/arm/_passes/decompose_round_pass.py +96 -0
  130. executorch/backends/arm/_passes/decompose_sdpa_pass.py +36 -0
  131. executorch/backends/arm/_passes/decompose_select.py +75 -0
  132. executorch/backends/arm/_passes/decompose_select_scatter_pass.py +143 -0
  133. executorch/backends/arm/_passes/decompose_sign_pass.py +78 -0
  134. executorch/backends/arm/_passes/decompose_silu_pass.py +41 -0
  135. executorch/backends/arm/_passes/decompose_sinh_pass.py +77 -0
  136. executorch/backends/arm/_passes/decompose_softmax_pass.py +93 -0
  137. executorch/backends/arm/_passes/decompose_softmax_unstable_pass.py +86 -0
  138. executorch/backends/arm/_passes/decompose_sqrt_pass.py +53 -0
  139. executorch/backends/arm/_passes/decompose_sum_pass.py +93 -0
  140. executorch/backends/arm/_passes/decompose_tan_pass.py +51 -0
  141. executorch/backends/arm/_passes/decompose_tosa_unsupported_clamp_pass.py +97 -0
  142. executorch/backends/arm/_passes/decompose_var_pass.py +109 -0
  143. executorch/backends/arm/_passes/decorate_fp32_to_int32_casting_pass.py +88 -0
  144. executorch/backends/arm/_passes/fold_qdq_with_annotated_qparams_pass.py +394 -0
  145. executorch/backends/arm/_passes/fuse_batch_norm2d_pass.py +241 -0
  146. executorch/backends/arm/_passes/fuse_constant_ops_pass.py +272 -0
  147. executorch/backends/arm/_passes/fuse_duplicate_users_pass.py +165 -0
  148. executorch/backends/arm/_passes/fuse_equal_placeholders_pass.py +104 -0
  149. executorch/backends/arm/_passes/fuse_quantized_activation_pass.py +80 -0
  150. executorch/backends/arm/_passes/fuse_view_copy_transform_pass.py +14 -0
  151. executorch/backends/arm/_passes/insert_int32_casts_after_int64_placeholders.py +130 -0
  152. executorch/backends/arm/_passes/insert_rescales_pass.py +584 -0
  153. executorch/backends/arm/_passes/insert_table_ops.py +305 -0
  154. executorch/backends/arm/_passes/match_arg_dtype_pass.py +104 -0
  155. executorch/backends/arm/_passes/match_arg_ranks_pass.py +114 -0
  156. executorch/backends/arm/_passes/mm_to_bmm_pass.py +104 -0
  157. executorch/backends/arm/_passes/normalize_while_initial_args_pass.py +113 -0
  158. executorch/backends/arm/_passes/promote_bool_operands_pass.py +88 -0
  159. executorch/backends/arm/_passes/quant_args.py +125 -0
  160. executorch/backends/arm/_passes/remove_getitem_pass.py +14 -0
  161. executorch/backends/arm/_passes/remove_graph_asserts_pass.py +18 -0
  162. executorch/backends/arm/_passes/remove_noop_pass.py +40 -0
  163. executorch/backends/arm/_passes/replace_inf_and_limit_values_pass.py +73 -0
  164. executorch/backends/arm/_passes/replace_scalar_with_tensor_pass.py +121 -0
  165. executorch/backends/arm/_passes/rewrite_bool_bitwise_not_to_logical_not_pass.py +43 -0
  166. executorch/backends/arm/_passes/rewrite_bool_to_fp32_cast_via_int8_pass.py +77 -0
  167. executorch/backends/arm/_passes/rewrite_conv_pass.py +336 -0
  168. executorch/backends/arm/_passes/rewrite_matmul.py +98 -0
  169. executorch/backends/arm/_passes/rewrite_upsample.py +93 -0
  170. executorch/backends/arm/_passes/scalars_to_attribute_pass.py +97 -0
  171. executorch/backends/arm/_passes/size_adjust_input_pass.py +227 -0
  172. executorch/backends/arm/_passes/to_tosa_memory_format_pass.py +454 -0
  173. executorch/backends/arm/_passes/unsqueeze_before_repeat_pass.py +66 -0
  174. executorch/backends/arm/_passes/unsqueeze_scalar_placeholders_pass.py +85 -0
  175. executorch/backends/arm/arm_vela.py +137 -0
  176. executorch/backends/arm/common/__init__.py +4 -0
  177. executorch/backends/arm/common/annotation_meta.py +39 -0
  178. executorch/backends/arm/common/arm_compile_spec.py +271 -0
  179. executorch/backends/arm/common/debug.py +79 -0
  180. executorch/backends/arm/common/pipeline_config.py +59 -0
  181. executorch/backends/arm/common/type.py +28 -0
  182. executorch/backends/arm/constants.py +48 -0
  183. executorch/backends/arm/debug/__init__.py +4 -0
  184. executorch/backends/arm/debug/schema.py +145 -0
  185. executorch/backends/arm/ethosu/__init__.py +11 -0
  186. executorch/backends/arm/ethosu/backend.py +115 -0
  187. executorch/backends/arm/ethosu/compile_spec.py +117 -0
  188. executorch/backends/arm/ethosu/partitioner.py +35 -0
  189. executorch/backends/arm/operator_support/__init__.py +23 -0
  190. executorch/backends/arm/operator_support/clone_dim_order_support.py +82 -0
  191. executorch/backends/arm/operator_support/control_flow_support.py +162 -0
  192. executorch/backends/arm/operator_support/convolution_support.py +160 -0
  193. executorch/backends/arm/operator_support/embedding_support.py +61 -0
  194. executorch/backends/arm/operator_support/ethos_u55_support.py +598 -0
  195. executorch/backends/arm/operator_support/index_select_support.py +60 -0
  196. executorch/backends/arm/operator_support/index_tensor_support.py +160 -0
  197. executorch/backends/arm/operator_support/minmax_support.py +61 -0
  198. executorch/backends/arm/operator_support/pool_2d_support.py +214 -0
  199. executorch/backends/arm/operator_support/reduce_sum_support.py +76 -0
  200. executorch/backends/arm/operator_support/right_shift_support.py +52 -0
  201. executorch/backends/arm/operator_support/slice_copy_support.py +48 -0
  202. executorch/backends/arm/operator_support/to_dim_order_copy_support.py +200 -0
  203. executorch/backends/arm/operator_support/tosa_profile_supported_op_lists.py +251 -0
  204. executorch/backends/arm/operator_support/tosa_supported_operators.py +748 -0
  205. executorch/backends/arm/operator_support/where_support.py +77 -0
  206. executorch/backends/arm/operators/__init__.py +71 -0
  207. executorch/backends/arm/operators/node_visitor.py +177 -0
  208. executorch/backends/arm/operators/op_abs.py +59 -0
  209. executorch/backends/arm/operators/op_add.py +60 -0
  210. executorch/backends/arm/operators/op_amax.py +72 -0
  211. executorch/backends/arm/operators/op_amin.py +73 -0
  212. executorch/backends/arm/operators/op_any.py +61 -0
  213. executorch/backends/arm/operators/op_avg_pool2d.py +142 -0
  214. executorch/backends/arm/operators/op_bitwise_not.py +62 -0
  215. executorch/backends/arm/operators/op_cat.py +68 -0
  216. executorch/backends/arm/operators/op_ceil.py +56 -0
  217. executorch/backends/arm/operators/op_clamp.py +118 -0
  218. executorch/backends/arm/operators/op_cond_if.py +56 -0
  219. executorch/backends/arm/operators/op_constant_pad_nd.py +117 -0
  220. executorch/backends/arm/operators/op_cos.py +49 -0
  221. executorch/backends/arm/operators/op_eq.py +64 -0
  222. executorch/backends/arm/operators/op_erf.py +54 -0
  223. executorch/backends/arm/operators/op_exp.py +54 -0
  224. executorch/backends/arm/operators/op_floor.py +56 -0
  225. executorch/backends/arm/operators/op_ge.py +64 -0
  226. executorch/backends/arm/operators/op_gt.py +64 -0
  227. executorch/backends/arm/operators/op_index_select.py +109 -0
  228. executorch/backends/arm/operators/op_index_tensor.py +247 -0
  229. executorch/backends/arm/operators/op_le.py +64 -0
  230. executorch/backends/arm/operators/op_log.py +50 -0
  231. executorch/backends/arm/operators/op_logical_not.py +62 -0
  232. executorch/backends/arm/operators/op_lt.py +64 -0
  233. executorch/backends/arm/operators/op_max_pool2d.py +108 -0
  234. executorch/backends/arm/operators/op_maximum.py +66 -0
  235. executorch/backends/arm/operators/op_minimum.py +67 -0
  236. executorch/backends/arm/operators/op_mul.py +61 -0
  237. executorch/backends/arm/operators/op_neg.py +92 -0
  238. executorch/backends/arm/operators/op_permute.py +150 -0
  239. executorch/backends/arm/operators/op_pow.py +63 -0
  240. executorch/backends/arm/operators/op_reciprocal.py +51 -0
  241. executorch/backends/arm/operators/op_repeat.py +77 -0
  242. executorch/backends/arm/operators/op_rshift_tensor.py +62 -0
  243. executorch/backends/arm/operators/op_rsqrt.py +51 -0
  244. executorch/backends/arm/operators/op_sigmoid.py +50 -0
  245. executorch/backends/arm/operators/op_sin.py +49 -0
  246. executorch/backends/arm/operators/op_slice.py +151 -0
  247. executorch/backends/arm/operators/op_sub.py +63 -0
  248. executorch/backends/arm/operators/op_sum.py +64 -0
  249. executorch/backends/arm/operators/op_tanh.py +51 -0
  250. executorch/backends/arm/operators/op_to_dim_order_copy.py +49 -0
  251. executorch/backends/arm/operators/op_tosa_conv2d.py +139 -0
  252. executorch/backends/arm/operators/op_tosa_conv3d.py +24 -0
  253. executorch/backends/arm/operators/op_tosa_depthwise_conv2d.py +32 -0
  254. executorch/backends/arm/operators/op_tosa_matmul.py +102 -0
  255. executorch/backends/arm/operators/op_tosa_rescale.py +261 -0
  256. executorch/backends/arm/operators/op_tosa_resize.py +124 -0
  257. executorch/backends/arm/operators/op_tosa_table.py +72 -0
  258. executorch/backends/arm/operators/op_tosa_transpose.py +71 -0
  259. executorch/backends/arm/operators/op_view.py +81 -0
  260. executorch/backends/arm/operators/op_where.py +75 -0
  261. executorch/backends/arm/operators/op_while.py +103 -0
  262. executorch/backends/arm/operators/operator_validation_utils.py +201 -0
  263. executorch/backends/arm/operators/ops_binary.py +112 -0
  264. executorch/backends/arm/operators/ops_identity.py +80 -0
  265. executorch/backends/arm/process_node.py +230 -0
  266. executorch/backends/arm/quantizer/__init__.py +35 -0
  267. executorch/backends/arm/quantizer/arm_quantizer.py +710 -0
  268. executorch/backends/arm/quantizer/arm_quantizer_utils.py +76 -0
  269. executorch/backends/arm/quantizer/quantization_annotator.py +815 -0
  270. executorch/backends/arm/quantizer/quantization_config.py +228 -0
  271. executorch/backends/arm/scripts/collect_testname_resources.py +194 -0
  272. executorch/backends/arm/scripts/parse_test_names.py +271 -0
  273. executorch/backends/arm/test/common.py +302 -0
  274. executorch/backends/arm/test/conftest.py +134 -0
  275. executorch/backends/arm/test/misc/test_bn_relu_folding_qat.py +124 -0
  276. executorch/backends/arm/test/misc/test_call_operator_submodule.py +72 -0
  277. executorch/backends/arm/test/misc/test_compile_spec.py +55 -0
  278. executorch/backends/arm/test/misc/test_conv_relu_residual_add.py +108 -0
  279. executorch/backends/arm/test/misc/test_custom_partition.py +183 -0
  280. executorch/backends/arm/test/misc/test_debug_feats.py +316 -0
  281. executorch/backends/arm/test/misc/test_debug_hook.py +186 -0
  282. executorch/backends/arm/test/misc/test_dim_order.py +127 -0
  283. executorch/backends/arm/test/misc/test_dw_convs_with_shared_weights.py +46 -0
  284. executorch/backends/arm/test/misc/test_extract_io_params_tosa.py +92 -0
  285. executorch/backends/arm/test/misc/test_int64.py +112 -0
  286. executorch/backends/arm/test/misc/test_lifted_tensor.py +146 -0
  287. executorch/backends/arm/test/misc/test_mixed_type_lowering.py +83 -0
  288. executorch/backends/arm/test/misc/test_model_evaluator.py +58 -0
  289. executorch/backends/arm/test/misc/test_multiple_delegates.py +53 -0
  290. executorch/backends/arm/test/misc/test_multiple_outputs.py +70 -0
  291. executorch/backends/arm/test/misc/test_non_persistent_buffers.py +49 -0
  292. executorch/backends/arm/test/misc/test_outputs_order.py +122 -0
  293. executorch/backends/arm/test/misc/test_partition_decomposed_quantized_ops.py +181 -0
  294. executorch/backends/arm/test/misc/test_pass_pipeline_config.py +35 -0
  295. executorch/backends/arm/test/misc/test_pass_required_order.py +97 -0
  296. executorch/backends/arm/test/misc/test_qat_training_loop.py +100 -0
  297. executorch/backends/arm/test/misc/test_quant_custom_meta.py +105 -0
  298. executorch/backends/arm/test/misc/test_rescale_range.py +130 -0
  299. executorch/backends/arm/test/misc/test_save_exported_model.py +64 -0
  300. executorch/backends/arm/test/misc/test_tosa_dialect_conv2d.py +220 -0
  301. executorch/backends/arm/test/misc/test_tosa_dialect_dw_conv2d.py +227 -0
  302. executorch/backends/arm/test/misc/test_tosa_spec.py +169 -0
  303. executorch/backends/arm/test/models/stable_diffusion/stable_diffusion_module_test_configs.py +114 -0
  304. executorch/backends/arm/test/models/stable_diffusion/test_CLIPTextModelWithProjection.py +168 -0
  305. executorch/backends/arm/test/models/stable_diffusion/test_SD3Transformer2DModel.py +193 -0
  306. executorch/backends/arm/test/models/stable_diffusion/test_T5EncoderModel.py +161 -0
  307. executorch/backends/arm/test/models/stable_diffusion/test_vae_AutoencoderKL.py +118 -0
  308. executorch/backends/arm/test/models/test_conformer.py +138 -0
  309. executorch/backends/arm/test/models/test_deit_tiny_arm.py +120 -0
  310. executorch/backends/arm/test/models/test_dl3_arm.py +118 -0
  311. executorch/backends/arm/test/models/test_inception_v3_arm.py +120 -0
  312. executorch/backends/arm/test/models/test_llama.py +227 -0
  313. executorch/backends/arm/test/models/test_lstm_arm.py +194 -0
  314. executorch/backends/arm/test/models/test_mobilenet_v2_arm.py +179 -0
  315. executorch/backends/arm/test/models/test_mobilenet_v3_arm.py +113 -0
  316. executorch/backends/arm/test/models/test_nn_functional.py +124 -0
  317. executorch/backends/arm/test/models/test_nn_modules.py +161 -0
  318. executorch/backends/arm/test/models/test_nss.py +146 -0
  319. executorch/backends/arm/test/models/test_resnet18.py +98 -0
  320. executorch/backends/arm/test/models/test_torch_functions.py +151 -0
  321. executorch/backends/arm/test/models/test_w2l_arm.py +136 -0
  322. executorch/backends/arm/test/ops/test_abs.py +101 -0
  323. executorch/backends/arm/test/ops/test_acos.py +121 -0
  324. executorch/backends/arm/test/ops/test_acosh.py +137 -0
  325. executorch/backends/arm/test/ops/test_adaptive_avg_pool2d.py +242 -0
  326. executorch/backends/arm/test/ops/test_add.py +291 -0
  327. executorch/backends/arm/test/ops/test_addmm.py +285 -0
  328. executorch/backends/arm/test/ops/test_alias_copy.py +114 -0
  329. executorch/backends/arm/test/ops/test_amax.py +191 -0
  330. executorch/backends/arm/test/ops/test_amin.py +205 -0
  331. executorch/backends/arm/test/ops/test_any.py +210 -0
  332. executorch/backends/arm/test/ops/test_arange.py +258 -0
  333. executorch/backends/arm/test/ops/test_asin.py +105 -0
  334. executorch/backends/arm/test/ops/test_asinh.py +104 -0
  335. executorch/backends/arm/test/ops/test_at.py +256 -0
  336. executorch/backends/arm/test/ops/test_atan.py +111 -0
  337. executorch/backends/arm/test/ops/test_atanh.py +112 -0
  338. executorch/backends/arm/test/ops/test_avg_pool2d.py +295 -0
  339. executorch/backends/arm/test/ops/test_batch_norm.py +385 -0
  340. executorch/backends/arm/test/ops/test_bitwise.py +672 -0
  341. executorch/backends/arm/test/ops/test_bitwise_not.py +115 -0
  342. executorch/backends/arm/test/ops/test_bmm.py +193 -0
  343. executorch/backends/arm/test/ops/test_cat.py +219 -0
  344. executorch/backends/arm/test/ops/test_ceil.py +125 -0
  345. executorch/backends/arm/test/ops/test_clamp.py +474 -0
  346. executorch/backends/arm/test/ops/test_clone.py +150 -0
  347. executorch/backends/arm/test/ops/test_cond.py +338 -0
  348. executorch/backends/arm/test/ops/test_constant_pad_nd.py +119 -0
  349. executorch/backends/arm/test/ops/test_conv1d.py +356 -0
  350. executorch/backends/arm/test/ops/test_conv2d.py +592 -0
  351. executorch/backends/arm/test/ops/test_conv3d.py +711 -0
  352. executorch/backends/arm/test/ops/test_conv_combos.py +668 -0
  353. executorch/backends/arm/test/ops/test_conv_constant_pad_nd.py +145 -0
  354. executorch/backends/arm/test/ops/test_copy.py +171 -0
  355. executorch/backends/arm/test/ops/test_cos.py +115 -0
  356. executorch/backends/arm/test/ops/test_cosh.py +111 -0
  357. executorch/backends/arm/test/ops/test_cumsum.py +122 -0
  358. executorch/backends/arm/test/ops/test_depthwise_conv.py +404 -0
  359. executorch/backends/arm/test/ops/test_div.py +151 -0
  360. executorch/backends/arm/test/ops/test_div_tensor_mode.py +151 -0
  361. executorch/backends/arm/test/ops/test_elu.py +133 -0
  362. executorch/backends/arm/test/ops/test_embedding.py +146 -0
  363. executorch/backends/arm/test/ops/test_eq.py +286 -0
  364. executorch/backends/arm/test/ops/test_erf.py +96 -0
  365. executorch/backends/arm/test/ops/test_exp.py +110 -0
  366. executorch/backends/arm/test/ops/test_expand.py +153 -0
  367. executorch/backends/arm/test/ops/test_expm1.py +113 -0
  368. executorch/backends/arm/test/ops/test_eye.py +156 -0
  369. executorch/backends/arm/test/ops/test_fill_scalar.py +108 -0
  370. executorch/backends/arm/test/ops/test_floor.py +125 -0
  371. executorch/backends/arm/test/ops/test_floor_div.py +156 -0
  372. executorch/backends/arm/test/ops/test_full.py +255 -0
  373. executorch/backends/arm/test/ops/test_ge.py +292 -0
  374. executorch/backends/arm/test/ops/test_gelu.py +156 -0
  375. executorch/backends/arm/test/ops/test_glu.py +130 -0
  376. executorch/backends/arm/test/ops/test_group_norm.py +200 -0
  377. executorch/backends/arm/test/ops/test_gt.py +293 -0
  378. executorch/backends/arm/test/ops/test_hardsigmoid.py +114 -0
  379. executorch/backends/arm/test/ops/test_hardswish.py +104 -0
  380. executorch/backends/arm/test/ops/test_hardtanh.py +113 -0
  381. executorch/backends/arm/test/ops/test_index_select.py +177 -0
  382. executorch/backends/arm/test/ops/test_index_tensor.py +467 -0
  383. executorch/backends/arm/test/ops/test_layer_norm.py +186 -0
  384. executorch/backends/arm/test/ops/test_le.py +296 -0
  385. executorch/backends/arm/test/ops/test_leaky_relu.py +125 -0
  386. executorch/backends/arm/test/ops/test_linalg_vector_norm.py +160 -0
  387. executorch/backends/arm/test/ops/test_linear.py +389 -0
  388. executorch/backends/arm/test/ops/test_log.py +100 -0
  389. executorch/backends/arm/test/ops/test_log1p.py +78 -0
  390. executorch/backends/arm/test/ops/test_logical.py +429 -0
  391. executorch/backends/arm/test/ops/test_logit.py +119 -0
  392. executorch/backends/arm/test/ops/test_logsoftmax.py +124 -0
  393. executorch/backends/arm/test/ops/test_lshift.py +220 -0
  394. executorch/backends/arm/test/ops/test_lt.py +293 -0
  395. executorch/backends/arm/test/ops/test_masked_fill.py +171 -0
  396. executorch/backends/arm/test/ops/test_matmul.py +289 -0
  397. executorch/backends/arm/test/ops/test_max_pool.py +329 -0
  398. executorch/backends/arm/test/ops/test_maximum.py +98 -0
  399. executorch/backends/arm/test/ops/test_mean_dim.py +388 -0
  400. executorch/backends/arm/test/ops/test_minimum.py +98 -0
  401. executorch/backends/arm/test/ops/test_mm.py +93 -0
  402. executorch/backends/arm/test/ops/test_mul.py +376 -0
  403. executorch/backends/arm/test/ops/test_multihead_attention.py +138 -0
  404. executorch/backends/arm/test/ops/test_ne.py +243 -0
  405. executorch/backends/arm/test/ops/test_neg.py +99 -0
  406. executorch/backends/arm/test/ops/test_ones.py +152 -0
  407. executorch/backends/arm/test/ops/test_permute.py +256 -0
  408. executorch/backends/arm/test/ops/test_pixel_shuffling.py +237 -0
  409. executorch/backends/arm/test/ops/test_pow.py +204 -0
  410. executorch/backends/arm/test/ops/test_reciprocal.py +112 -0
  411. executorch/backends/arm/test/ops/test_relu.py +166 -0
  412. executorch/backends/arm/test/ops/test_remainder.py +199 -0
  413. executorch/backends/arm/test/ops/test_repeat.py +162 -0
  414. executorch/backends/arm/test/ops/test_round.py +111 -0
  415. executorch/backends/arm/test/ops/test_rshift.py +226 -0
  416. executorch/backends/arm/test/ops/test_rsqrt.py +153 -0
  417. executorch/backends/arm/test/ops/test_rsub.py +126 -0
  418. executorch/backends/arm/test/ops/test_scalar_tensor.py +135 -0
  419. executorch/backends/arm/test/ops/test_scalars.py +438 -0
  420. executorch/backends/arm/test/ops/test_sdpa.py +73 -0
  421. executorch/backends/arm/test/ops/test_select.py +219 -0
  422. executorch/backends/arm/test/ops/test_select_scatter.py +173 -0
  423. executorch/backends/arm/test/ops/test_sigmoid.py +357 -0
  424. executorch/backends/arm/test/ops/test_sigmoid_32bit.py +177 -0
  425. executorch/backends/arm/test/ops/test_sign.py +113 -0
  426. executorch/backends/arm/test/ops/test_silu.py +174 -0
  427. executorch/backends/arm/test/ops/test_sin.py +108 -0
  428. executorch/backends/arm/test/ops/test_sinh.py +103 -0
  429. executorch/backends/arm/test/ops/test_slice.py +217 -0
  430. executorch/backends/arm/test/ops/test_softmax.py +121 -0
  431. executorch/backends/arm/test/ops/test_split.py +312 -0
  432. executorch/backends/arm/test/ops/test_sqrt.py +112 -0
  433. executorch/backends/arm/test/ops/test_squeeze.py +291 -0
  434. executorch/backends/arm/test/ops/test_stack.py +150 -0
  435. executorch/backends/arm/test/ops/test_sub.py +366 -0
  436. executorch/backends/arm/test/ops/test_sum.py +161 -0
  437. executorch/backends/arm/test/ops/test_t_copy.py +115 -0
  438. executorch/backends/arm/test/ops/test_tan.py +146 -0
  439. executorch/backends/arm/test/ops/test_tanh.py +169 -0
  440. executorch/backends/arm/test/ops/test_to_copy.py +296 -0
  441. executorch/backends/arm/test/ops/test_transpose_copy.py +114 -0
  442. executorch/backends/arm/test/ops/test_unary_combos.py +144 -0
  443. executorch/backends/arm/test/ops/test_unbind.py +82 -0
  444. executorch/backends/arm/test/ops/test_unflatten.py +107 -0
  445. executorch/backends/arm/test/ops/test_unsqueeze.py +107 -0
  446. executorch/backends/arm/test/ops/test_upsample_bilinear2d.py +443 -0
  447. executorch/backends/arm/test/ops/test_upsample_nearest2d.py +520 -0
  448. executorch/backends/arm/test/ops/test_var.py +394 -0
  449. executorch/backends/arm/test/ops/test_view.py +253 -0
  450. executorch/backends/arm/test/ops/test_where.py +285 -0
  451. executorch/backends/arm/test/ops/test_while.py +264 -0
  452. executorch/backends/arm/test/ops/test_zeros.py +158 -0
  453. executorch/backends/arm/test/passes/test_broadcast_args_pass.py +57 -0
  454. executorch/backends/arm/test/passes/test_cast_int64_pass.py +49 -0
  455. executorch/backends/arm/test/passes/test_convert_expand_copy_to_repeat.py +51 -0
  456. executorch/backends/arm/test/passes/test_convert_int64_const_ops_to_int32.py +529 -0
  457. executorch/backends/arm/test/passes/test_convert_int64_output_ops_to_int32.py +129 -0
  458. executorch/backends/arm/test/passes/test_convert_permute_singleton_to_view_pass.py +100 -0
  459. executorch/backends/arm/test/passes/test_convert_split_to_slice.py +75 -0
  460. executorch/backends/arm/test/passes/test_convert_to_clamp.py +92 -0
  461. executorch/backends/arm/test/passes/test_decompose_avg_pool2d_pass.py +82 -0
  462. executorch/backends/arm/test/passes/test_decompose_cosine_similarity_pass.py +58 -0
  463. executorch/backends/arm/test/passes/test_decompose_div_pass.py +70 -0
  464. executorch/backends/arm/test/passes/test_decompose_int_pow_pass.py +80 -0
  465. executorch/backends/arm/test/passes/test_decompose_layernorm_pass.py +69 -0
  466. executorch/backends/arm/test/passes/test_decompose_linalg_vector_norm_pass.py +107 -0
  467. executorch/backends/arm/test/passes/test_decompose_meandim_pass.py +137 -0
  468. executorch/backends/arm/test/passes/test_decompose_quant_nodes.py +44 -0
  469. executorch/backends/arm/test/passes/test_decompose_softmax_pass.py +103 -0
  470. executorch/backends/arm/test/passes/test_decompose_tosa_unsupported_clamp_pass.py +73 -0
  471. executorch/backends/arm/test/passes/test_decompose_var_pass.py +89 -0
  472. executorch/backends/arm/test/passes/test_decorate_fp32_to_int32_casting_pass.py +85 -0
  473. executorch/backends/arm/test/passes/test_fold_qdq_pass.py +51 -0
  474. executorch/backends/arm/test/passes/test_fuse_batchnorm_pass.py +164 -0
  475. executorch/backends/arm/test/passes/test_fuse_constant_ops_pass.py +177 -0
  476. executorch/backends/arm/test/passes/test_fuse_duplicate_users_pass.py +70 -0
  477. executorch/backends/arm/test/passes/test_fuse_equal_placeholders_ops_pass.py +170 -0
  478. executorch/backends/arm/test/passes/test_fuse_view_copy.py +82 -0
  479. executorch/backends/arm/test/passes/test_insert_int32_casts_after_int64_placeholders_pass.py +114 -0
  480. executorch/backends/arm/test/passes/test_insert_rescale_i32_pass.py +107 -0
  481. executorch/backends/arm/test/passes/test_insert_table_ops_pass.py +47 -0
  482. executorch/backends/arm/test/passes/test_int32_cast_embedding_pass.py +46 -0
  483. executorch/backends/arm/test/passes/test_ioquantization_pass.py +48 -0
  484. executorch/backends/arm/test/passes/test_promote_bool_operands_pass.py +103 -0
  485. executorch/backends/arm/test/passes/test_remove_clone_pass.py +45 -0
  486. executorch/backends/arm/test/passes/test_replace_inf_values_pass.py +96 -0
  487. executorch/backends/arm/test/passes/test_rescale_pass.py +81 -0
  488. executorch/backends/arm/test/passes/test_rewrite_conv_pass.py +20 -0
  489. executorch/backends/arm/test/passes/test_to_tosa_memory_format.py +213 -0
  490. executorch/backends/arm/test/passes/test_unsqueeze_before_repeat_pass.py +62 -0
  491. executorch/backends/arm/test/quantizer/test_conv_relu_fusing.py +118 -0
  492. executorch/backends/arm/test/quantizer/test_generic_annotater.py +109 -0
  493. executorch/backends/arm/test/quantizer/test_partial_quantization.py +104 -0
  494. executorch/backends/arm/test/quantizer/test_selective_quantization.py +212 -0
  495. executorch/backends/arm/test/quantizer/test_set_module_name.py +158 -0
  496. executorch/backends/arm/test/quantizer/test_tosa_quantizer_validate.py +55 -0
  497. executorch/backends/arm/test/runner_utils.py +840 -0
  498. executorch/backends/arm/test/test_memory_allocator_log.py +170 -0
  499. executorch/backends/arm/test/test_model.py +345 -0
  500. executorch/backends/arm/test/tester/__init__.py +6 -0
  501. executorch/backends/arm/test/tester/analyze_output_utils.py +346 -0
  502. executorch/backends/arm/test/tester/arm_tester.py +1149 -0
  503. executorch/backends/arm/test/tester/quantize.py +43 -0
  504. executorch/backends/arm/test/tester/serialize.py +78 -0
  505. executorch/backends/arm/test/tester/test_pipeline.py +1190 -0
  506. executorch/backends/arm/tosa/__init__.py +9 -0
  507. executorch/backends/arm/tosa/backend.py +443 -0
  508. executorch/backends/arm/tosa/compile_spec.py +48 -0
  509. executorch/backends/arm/tosa/dialect/__init__.py +15 -0
  510. executorch/backends/arm/tosa/dialect/lib.py +85 -0
  511. executorch/backends/arm/tosa/dialect/ops/conv2d.py +111 -0
  512. executorch/backends/arm/tosa/dialect/ops/conv3d.py +74 -0
  513. executorch/backends/arm/tosa/dialect/ops/depthwise_conv2d.py +56 -0
  514. executorch/backends/arm/tosa/dialect/ops/matmul.py +56 -0
  515. executorch/backends/arm/tosa/dialect/ops/rescale.py +53 -0
  516. executorch/backends/arm/tosa/dialect/ops/resize.py +66 -0
  517. executorch/backends/arm/tosa/dialect/ops/table.py +53 -0
  518. executorch/backends/arm/tosa/dialect/ops/transpose.py +35 -0
  519. executorch/backends/arm/tosa/dialect/ops_registration.py +68 -0
  520. executorch/backends/arm/tosa/mapping.py +309 -0
  521. executorch/backends/arm/tosa/partitioner.py +486 -0
  522. executorch/backends/arm/tosa/schemas/__init__.py +4 -0
  523. executorch/backends/arm/tosa/schemas/tosa_1.0.fbs +559 -0
  524. executorch/backends/arm/tosa/specification.py +406 -0
  525. executorch/backends/arm/tosa/utils.py +304 -0
  526. executorch/backends/arm/util/_factory.py +59 -0
  527. executorch/backends/arm/util/arm_model_evaluator.py +388 -0
  528. executorch/backends/arm/vgf/__init__.py +11 -0
  529. executorch/backends/arm/vgf/backend.py +172 -0
  530. executorch/backends/arm/vgf/compile_spec.py +73 -0
  531. executorch/backends/arm/vgf/model_converter.py +34 -0
  532. executorch/backends/arm/vgf/partitioner.py +35 -0
  533. executorch/backends/cadence/aot/__init__.py +5 -0
  534. executorch/backends/cadence/aot/compiler.py +484 -0
  535. executorch/backends/cadence/aot/compiler_funcs.py +203 -0
  536. executorch/backends/cadence/aot/compiler_utils.py +293 -0
  537. executorch/backends/cadence/aot/decompose_ops.py +120 -0
  538. executorch/backends/cadence/aot/export_example.py +117 -0
  539. executorch/backends/cadence/aot/functions.yaml +470 -0
  540. executorch/backends/cadence/aot/functions_fusion_g3.yaml +191 -0
  541. executorch/backends/cadence/aot/functions_hifi.yaml +565 -0
  542. executorch/backends/cadence/aot/functions_vision.yaml +275 -0
  543. executorch/backends/cadence/aot/fuse_ops.py +1138 -0
  544. executorch/backends/cadence/aot/graph_builder.py +142 -0
  545. executorch/backends/cadence/aot/memory_constraints.py +731 -0
  546. executorch/backends/cadence/aot/memory_planning.py +469 -0
  547. executorch/backends/cadence/aot/memory_planning_algo.py +341 -0
  548. executorch/backends/cadence/aot/ops_registrations.py +2841 -0
  549. executorch/backends/cadence/aot/pass_utils.py +274 -0
  550. executorch/backends/cadence/aot/passes.py +132 -0
  551. executorch/backends/cadence/aot/program_builder.py +138 -0
  552. executorch/backends/cadence/aot/quantizer/fusion_pass.py +743 -0
  553. executorch/backends/cadence/aot/quantizer/patterns.py +738 -0
  554. executorch/backends/cadence/aot/quantizer/quantizer.py +409 -0
  555. executorch/backends/cadence/aot/quantizer/utils.py +288 -0
  556. executorch/backends/cadence/aot/ref_implementations.py +2210 -0
  557. executorch/backends/cadence/aot/remove_ops.py +936 -0
  558. executorch/backends/cadence/aot/reorder_ops.py +843 -0
  559. executorch/backends/cadence/aot/replace_ops.py +2526 -0
  560. executorch/backends/cadence/aot/simplify_ops.py +221 -0
  561. executorch/backends/cadence/aot/tests/test_decompose_ops_passes.py +80 -0
  562. executorch/backends/cadence/aot/tests/test_fusion_ops_passes.py +1023 -0
  563. executorch/backends/cadence/aot/tests/test_graph_builder.py +104 -0
  564. executorch/backends/cadence/aot/tests/test_idma_ops.py +31 -0
  565. executorch/backends/cadence/aot/tests/test_memory_passes.py +1367 -0
  566. executorch/backends/cadence/aot/tests/test_pass_filter.py +167 -0
  567. executorch/backends/cadence/aot/tests/test_program_builder.py +222 -0
  568. executorch/backends/cadence/aot/tests/test_quantizer_ops.py +649 -0
  569. executorch/backends/cadence/aot/tests/test_ref_implementations.py +3151 -0
  570. executorch/backends/cadence/aot/tests/test_remove_ops_passes.py +977 -0
  571. executorch/backends/cadence/aot/tests/test_reorder_ops_passes.py +635 -0
  572. executorch/backends/cadence/aot/tests/test_replace_ops_passes.py +2777 -0
  573. executorch/backends/cadence/aot/tests/test_simplify_ops_passes.py +160 -0
  574. executorch/backends/cadence/aot/tests/test_type_dispatch_passes.py +481 -0
  575. executorch/backends/cadence/aot/type_dispatch.py +190 -0
  576. executorch/backends/cadence/aot/typing_stubs.py +22 -0
  577. executorch/backends/cadence/aot/utils.py +363 -0
  578. executorch/backends/cadence/runtime/__init__.py +13 -0
  579. executorch/backends/cadence/runtime/etdump.py +173 -0
  580. executorch/backends/cadence/runtime/executor.py +202 -0
  581. executorch/backends/cadence/runtime/runtime.py +156 -0
  582. executorch/backends/cadence/runtime/utils.py +100 -0
  583. executorch/backends/cadence/utils/FACTO/calibrator/__init__.py +0 -0
  584. executorch/backends/cadence/utils/FACTO/calibrator/runner.py +190 -0
  585. executorch/backends/cadence/utils/FACTO/examples/example.py +59 -0
  586. executorch/backends/cadence/utils/FACTO/examples/minimal_example.py +21 -0
  587. executorch/backends/cadence/utils/FACTO/examples/random_seed.py +35 -0
  588. executorch/backends/cadence/utils/FACTO/facto/__init__.py +0 -0
  589. executorch/backends/cadence/utils/FACTO/facto/inputgen/__init__.py +0 -0
  590. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/__init__.py +0 -0
  591. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/engine.py +129 -0
  592. executorch/backends/cadence/utils/FACTO/facto/inputgen/argtuple/gen.py +47 -0
  593. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/__init__.py +0 -0
  594. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/engine.py +291 -0
  595. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/gen.py +139 -0
  596. executorch/backends/cadence/utils/FACTO/facto/inputgen/argument/type.py +134 -0
  597. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/__init__.py +0 -0
  598. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/engine.py +54 -0
  599. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/model.py +133 -0
  600. executorch/backends/cadence/utils/FACTO/facto/inputgen/attribute/solve.py +125 -0
  601. executorch/backends/cadence/utils/FACTO/facto/inputgen/specs/__init__.py +0 -0
  602. executorch/backends/cadence/utils/FACTO/facto/inputgen/specs/model.py +149 -0
  603. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/__init__.py +0 -0
  604. executorch/backends/cadence/utils/FACTO/facto/inputgen/utils/random_manager.py +33 -0
  605. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/__init__.py +0 -0
  606. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/constants.py +9 -0
  607. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/gen.py +283 -0
  608. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/solve.py +142 -0
  609. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/space.py +450 -0
  610. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/type.py +105 -0
  611. executorch/backends/cadence/utils/FACTO/facto/inputgen/variable/utils.py +30 -0
  612. executorch/backends/cadence/utils/FACTO/facto/specdb/db.py +4361 -0
  613. executorch/backends/cadence/utils/FACTO/facto/specdb/default.py +36 -0
  614. executorch/backends/cadence/utils/FACTO/facto/specdb/dtypes.py +25 -0
  615. executorch/backends/cadence/utils/FACTO/facto/specdb/function.py +537 -0
  616. executorch/backends/cadence/utils/FACTO/setup.py +19 -0
  617. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argtuple_engine.py +53 -0
  618. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argtuple_generator.py +58 -0
  619. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_attributes.py +78 -0
  620. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_generator.py +98 -0
  621. executorch/backends/cadence/utils/FACTO/test/inputgen/test_argument_types.py +28 -0
  622. executorch/backends/cadence/utils/FACTO/test/inputgen/test_attribute_engine.py +62 -0
  623. executorch/backends/cadence/utils/FACTO/test/inputgen/test_attribute_solver.py +83 -0
  624. executorch/backends/cadence/utils/FACTO/test/inputgen/test_constraints.py +27 -0
  625. executorch/backends/cadence/utils/FACTO/test/inputgen/test_meta_arg_engine.py +70 -0
  626. executorch/backends/cadence/utils/FACTO/test/inputgen/test_specs.py +48 -0
  627. executorch/backends/cadence/utils/FACTO/test/inputgen/test_structural_engine.py +32 -0
  628. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_generation.py +80 -0
  629. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_solving.py +193 -0
  630. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_space.py +231 -0
  631. executorch/backends/cadence/utils/FACTO/test/inputgen/test_variable_types.py +71 -0
  632. executorch/backends/cadence/utils/facto_util.py +611 -0
  633. executorch/backends/cadence/utils/gen_header.py +49 -0
  634. executorch/backends/cadence/utils/post_compilation.py +173 -0
  635. executorch/backends/cortex_m/ops/operators.py +890 -0
  636. executorch/backends/cortex_m/ops/operators.yaml +78 -0
  637. executorch/backends/cortex_m/passes/__init__.py +12 -0
  638. executorch/backends/cortex_m/passes/activation_fusion_pass.py +181 -0
  639. executorch/backends/cortex_m/passes/clamp_hardswish_pass.py +37 -0
  640. executorch/backends/cortex_m/passes/convert_to_cortex_m_pass.py +318 -0
  641. executorch/backends/cortex_m/passes/cortex_m_pass_manager.py +73 -0
  642. executorch/backends/cortex_m/passes/decompose_hardswish_pass.py +127 -0
  643. executorch/backends/cortex_m/passes/passes_utils.py +226 -0
  644. executorch/backends/cortex_m/passes/quantized_op_fusion_pass.py +255 -0
  645. executorch/backends/cortex_m/passes/replace_quant_nodes_pass.py +67 -0
  646. executorch/backends/cortex_m/quantizer/__init__.py +18 -0
  647. executorch/backends/cortex_m/quantizer/operator_configs.py +74 -0
  648. executorch/backends/cortex_m/quantizer/quantization_configs.py +121 -0
  649. executorch/backends/cortex_m/quantizer/quantizer.py +500 -0
  650. executorch/backends/cortex_m/test/misc/test_quantization.py +359 -0
  651. executorch/backends/cortex_m/test/models/__init__.py +4 -0
  652. executorch/backends/cortex_m/test/models/test_mobilenet_v3.py +74 -0
  653. executorch/backends/cortex_m/test/ops/__init__.py +4 -0
  654. executorch/backends/cortex_m/test/ops/test_activation.py +528 -0
  655. executorch/backends/cortex_m/test/ops/test_add.py +182 -0
  656. executorch/backends/cortex_m/test/ops/test_avg_pool2d.py +99 -0
  657. executorch/backends/cortex_m/test/ops/test_conv.py +342 -0
  658. executorch/backends/cortex_m/test/ops/test_linear.py +130 -0
  659. executorch/backends/cortex_m/test/ops/test_lstm.py +98 -0
  660. executorch/backends/cortex_m/test/ops/test_maximum.py +83 -0
  661. executorch/backends/cortex_m/test/ops/test_minimum.py +104 -0
  662. executorch/backends/cortex_m/test/ops/test_mul.py +156 -0
  663. executorch/backends/cortex_m/test/ops/test_softmax.py +84 -0
  664. executorch/backends/cortex_m/test/ops/test_transpose.py +102 -0
  665. executorch/backends/cortex_m/test/test_helpers_passes_utils.py +103 -0
  666. executorch/backends/cortex_m/test/test_replace_quant_nodes.py +134 -0
  667. executorch/backends/cortex_m/test/tester.py +113 -0
  668. executorch/backends/cuda/__init__.py +5 -0
  669. executorch/backends/cuda/cuda_backend.py +250 -0
  670. executorch/backends/cuda/cuda_partitioner.py +25 -0
  671. executorch/backends/cuda/passes/__init__.py +0 -0
  672. executorch/backends/cuda/passes/move_cond_predicate_to_cpu.py +90 -0
  673. executorch/backends/cuda/passes/tests/__init__.py +0 -0
  674. executorch/backends/cuda/passes/tests/test_move_cond_predicate_to_cpu.py +553 -0
  675. executorch/backends/cuda/tests/__init__.py +5 -0
  676. executorch/backends/cuda/tests/test_cuda_export.py +327 -0
  677. executorch/backends/cuda/tests/test_cuda_partitioner.py +224 -0
  678. executorch/backends/cuda/triton/__init__.py +17 -0
  679. executorch/backends/cuda/triton/kernels/__init__.py +11 -0
  680. executorch/backends/cuda/triton/kernels/sdpa.py +827 -0
  681. executorch/backends/cuda/triton/replacement_pass.py +130 -0
  682. executorch/backends/example/example_backend.py +41 -0
  683. executorch/backends/example/example_backend_delegate_passes/merge_to_dim_pass.py +83 -0
  684. executorch/backends/example/example_backend_delegate_passes/permute_memory_formats_pass.py +143 -0
  685. executorch/backends/example/example_operators/adaptive_avg_pool2d.py +47 -0
  686. executorch/backends/example/example_operators/add.py +44 -0
  687. executorch/backends/example/example_operators/conv2d.py +52 -0
  688. executorch/backends/example/example_operators/conv_relu.py +49 -0
  689. executorch/backends/example/example_operators/dropout.py +44 -0
  690. executorch/backends/example/example_operators/flatten.py +40 -0
  691. executorch/backends/example/example_operators/linear.py +62 -0
  692. executorch/backends/example/example_operators/op_base.py +15 -0
  693. executorch/backends/example/example_operators/ops.py +32 -0
  694. executorch/backends/example/example_operators/utils.py +34 -0
  695. executorch/backends/example/example_partitioner.py +88 -0
  696. executorch/backends/example/example_quantizer.py +108 -0
  697. executorch/backends/example/test_example_delegate.py +112 -0
  698. executorch/backends/mediatek/__init__.py +5 -0
  699. executorch/backends/mediatek/_passes/__init__.py +0 -0
  700. executorch/backends/mediatek/_passes/decompose_scaled_dot_product_attention.py +79 -0
  701. executorch/backends/mediatek/partitioner.py +107 -0
  702. executorch/backends/mediatek/preprocess.py +214 -0
  703. executorch/backends/mediatek/quantizer/__init__.py +4 -0
  704. executorch/backends/mediatek/quantizer/annotator.py +263 -0
  705. executorch/backends/mediatek/quantizer/qconfig.py +174 -0
  706. executorch/backends/mediatek/quantizer/quantizer.py +45 -0
  707. executorch/backends/nxp/_passes/remove_getitem_pass.py +105 -0
  708. executorch/backends/nxp/aten_passes/convert_unsqueeze_to_view.py +79 -0
  709. executorch/backends/nxp/aten_passes/decompose_split_to_slices_pass.py +193 -0
  710. executorch/backends/nxp/aten_passes/fuse_batch_norm_with_conv_pass.py +146 -0
  711. executorch/backends/nxp/aten_passes/fuse_batch_norm_with_linear_pass.py +150 -0
  712. executorch/backends/nxp/aten_passes/fuse_linear_and_add_pass.py +204 -0
  713. executorch/backends/nxp/aten_passes/move_activation_before_concat.py +102 -0
  714. executorch/backends/nxp/aten_passes/neutron_aten_pass_manager.py +80 -0
  715. executorch/backends/nxp/aten_passes/remove_nodes_with_known_outputs.py +196 -0
  716. executorch/backends/nxp/aten_passes/split_group_convolution.py +281 -0
  717. executorch/backends/nxp/aten_passes/split_gru_based_on_num_layers.py +307 -0
  718. executorch/backends/nxp/backend/custom_delegation_options.py +19 -0
  719. executorch/backends/nxp/backend/edge_helper.py +222 -0
  720. executorch/backends/nxp/backend/edge_program_converter.py +247 -0
  721. executorch/backends/nxp/backend/ir/conversion_config.py +54 -0
  722. executorch/backends/nxp/backend/ir/conversion_context.py +38 -0
  723. executorch/backends/nxp/backend/ir/converter/__init__.py +0 -0
  724. executorch/backends/nxp/backend/ir/converter/builder/__init__.py +0 -0
  725. executorch/backends/nxp/backend/ir/converter/builder/aten_model_builder_director.py +142 -0
  726. executorch/backends/nxp/backend/ir/converter/builder/model_builder.py +1720 -0
  727. executorch/backends/nxp/backend/ir/converter/builder/quantization_verification.py +377 -0
  728. executorch/backends/nxp/backend/ir/converter/conversion/__init__.py +0 -0
  729. executorch/backends/nxp/backend/ir/converter/conversion/aten_translator.py +71 -0
  730. executorch/backends/nxp/backend/ir/converter/conversion/common.py +207 -0
  731. executorch/backends/nxp/backend/ir/converter/conversion/translator.py +689 -0
  732. executorch/backends/nxp/backend/ir/converter/node_converter.py +234 -0
  733. executorch/backends/nxp/backend/ir/converter/node_converters/__init__.py +4 -0
  734. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/__init__.py +101 -0
  735. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/abs_converter.py +35 -0
  736. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/adaptive_avg_pool_2d_converter.py +68 -0
  737. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/add_tensor_converter.py +57 -0
  738. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/addmm_converter.py +62 -0
  739. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/avg_pool_2d_converter.py +101 -0
  740. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/cat_converter.py +209 -0
  741. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/clone_converter.py +118 -0
  742. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/constant_pad_nd_converter.py +160 -0
  743. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/convolution_converter.py +535 -0
  744. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/hardtanh_converter.py +47 -0
  745. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/max_pool_2d_converter.py +104 -0
  746. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mean_dim_converter.py +112 -0
  747. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mm_converter.py +59 -0
  748. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/mul_tensor_converter.py +61 -0
  749. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/permute_copy_converter.py +468 -0
  750. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/qdq_dequantize_converter.py +116 -0
  751. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/qdq_quantize_converter.py +46 -0
  752. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/relu_converter.py +33 -0
  753. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sigmoid_converter.py +35 -0
  754. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/slice_tensor_converter.py +158 -0
  755. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/softmax_converter.py +62 -0
  756. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/sub_tensor_converter.py +59 -0
  757. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/tanh_converter.py +33 -0
  758. executorch/backends/nxp/backend/ir/converter/node_converters/ops_converters/view_copy_converter.py +220 -0
  759. executorch/backends/nxp/backend/ir/converter/node_converters/shared/__init__.py +0 -0
  760. executorch/backends/nxp/backend/ir/converter/node_converters/shared/conv_utils.py +105 -0
  761. executorch/backends/nxp/backend/ir/converter/node_converters/shared/recurrent_utils.py +64 -0
  762. executorch/backends/nxp/backend/ir/converter/node_converters/shared/reduce_utils.py +97 -0
  763. executorch/backends/nxp/backend/ir/converter/node_converters/shared/reshape_transposition.py +233 -0
  764. executorch/backends/nxp/backend/ir/converter/quantization_utils.py +149 -0
  765. executorch/backends/nxp/backend/ir/converter/tensor_utils.py +50 -0
  766. executorch/backends/nxp/backend/ir/lib/__init__.py +0 -0
  767. executorch/backends/nxp/backend/ir/lib/tflite/ATan2Options.py +50 -0
  768. executorch/backends/nxp/backend/ir/lib/tflite/AbsOptions.py +50 -0
  769. executorch/backends/nxp/backend/ir/lib/tflite/ActivationFunctionType.py +12 -0
  770. executorch/backends/nxp/backend/ir/lib/tflite/AddNOptions.py +50 -0
  771. executorch/backends/nxp/backend/ir/lib/tflite/AddOptions.py +82 -0
  772. executorch/backends/nxp/backend/ir/lib/tflite/ArgMaxOptions.py +65 -0
  773. executorch/backends/nxp/backend/ir/lib/tflite/ArgMinOptions.py +65 -0
  774. executorch/backends/nxp/backend/ir/lib/tflite/AssignVariableOptions.py +50 -0
  775. executorch/backends/nxp/backend/ir/lib/tflite/BatchMatMulOptions.py +101 -0
  776. executorch/backends/nxp/backend/ir/lib/tflite/BatchToSpaceNDOptions.py +50 -0
  777. executorch/backends/nxp/backend/ir/lib/tflite/BidirectionalSequenceLSTMOptions.py +160 -0
  778. executorch/backends/nxp/backend/ir/lib/tflite/BidirectionalSequenceRNNOptions.py +126 -0
  779. executorch/backends/nxp/backend/ir/lib/tflite/BitcastOptions.py +50 -0
  780. executorch/backends/nxp/backend/ir/lib/tflite/BitwiseXorOptions.py +50 -0
  781. executorch/backends/nxp/backend/ir/lib/tflite/BroadcastToOptions.py +50 -0
  782. executorch/backends/nxp/backend/ir/lib/tflite/BucketizeOptions.py +98 -0
  783. executorch/backends/nxp/backend/ir/lib/tflite/Buffer.py +132 -0
  784. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOperator.py +212 -0
  785. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOptions.py +133 -0
  786. executorch/backends/nxp/backend/ir/lib/tflite/BuiltinOptions2.py +27 -0
  787. executorch/backends/nxp/backend/ir/lib/tflite/CallOnceOptions.py +65 -0
  788. executorch/backends/nxp/backend/ir/lib/tflite/CallOptions.py +67 -0
  789. executorch/backends/nxp/backend/ir/lib/tflite/CastOptions.py +80 -0
  790. executorch/backends/nxp/backend/ir/lib/tflite/CombinerType.py +9 -0
  791. executorch/backends/nxp/backend/ir/lib/tflite/ConcatEmbeddingsOptions.py +163 -0
  792. executorch/backends/nxp/backend/ir/lib/tflite/ConcatenationOptions.py +80 -0
  793. executorch/backends/nxp/backend/ir/lib/tflite/Conv2DOptions.py +155 -0
  794. executorch/backends/nxp/backend/ir/lib/tflite/Conv3DOptions.py +170 -0
  795. executorch/backends/nxp/backend/ir/lib/tflite/CosOptions.py +50 -0
  796. executorch/backends/nxp/backend/ir/lib/tflite/CumsumOptions.py +84 -0
  797. executorch/backends/nxp/backend/ir/lib/tflite/CustomOptionsFormat.py +7 -0
  798. executorch/backends/nxp/backend/ir/lib/tflite/CustomQuantization.py +98 -0
  799. executorch/backends/nxp/backend/ir/lib/tflite/DensifyOptions.py +50 -0
  800. executorch/backends/nxp/backend/ir/lib/tflite/DepthToSpaceOptions.py +65 -0
  801. executorch/backends/nxp/backend/ir/lib/tflite/DepthwiseConv2DOptions.py +157 -0
  802. executorch/backends/nxp/backend/ir/lib/tflite/DequantizeOptions.py +50 -0
  803. executorch/backends/nxp/backend/ir/lib/tflite/DilateOptions.py +50 -0
  804. executorch/backends/nxp/backend/ir/lib/tflite/DimensionMetadata.py +152 -0
  805. executorch/backends/nxp/backend/ir/lib/tflite/DimensionType.py +8 -0
  806. executorch/backends/nxp/backend/ir/lib/tflite/DivOptions.py +65 -0
  807. executorch/backends/nxp/backend/ir/lib/tflite/DynamicUpdateSliceOptions.py +52 -0
  808. executorch/backends/nxp/backend/ir/lib/tflite/EmbeddingLookupSparseOptions.py +67 -0
  809. executorch/backends/nxp/backend/ir/lib/tflite/EqualOptions.py +50 -0
  810. executorch/backends/nxp/backend/ir/lib/tflite/ExpOptions.py +50 -0
  811. executorch/backends/nxp/backend/ir/lib/tflite/ExpandDimsOptions.py +50 -0
  812. executorch/backends/nxp/backend/ir/lib/tflite/FakeQuantOptions.py +116 -0
  813. executorch/backends/nxp/backend/ir/lib/tflite/FillOptions.py +50 -0
  814. executorch/backends/nxp/backend/ir/lib/tflite/FloorDivOptions.py +50 -0
  815. executorch/backends/nxp/backend/ir/lib/tflite/FloorModOptions.py +50 -0
  816. executorch/backends/nxp/backend/ir/lib/tflite/FullyConnectedOptions.py +129 -0
  817. executorch/backends/nxp/backend/ir/lib/tflite/FullyConnectedOptionsWeightsFormat.py +8 -0
  818. executorch/backends/nxp/backend/ir/lib/tflite/GatherNdOptions.py +50 -0
  819. executorch/backends/nxp/backend/ir/lib/tflite/GatherOptions.py +80 -0
  820. executorch/backends/nxp/backend/ir/lib/tflite/GeluOptions.py +67 -0
  821. executorch/backends/nxp/backend/ir/lib/tflite/GreaterEqualOptions.py +50 -0
  822. executorch/backends/nxp/backend/ir/lib/tflite/GreaterOptions.py +50 -0
  823. executorch/backends/nxp/backend/ir/lib/tflite/HardSwishOptions.py +50 -0
  824. executorch/backends/nxp/backend/ir/lib/tflite/HashtableFindOptions.py +50 -0
  825. executorch/backends/nxp/backend/ir/lib/tflite/HashtableImportOptions.py +52 -0
  826. executorch/backends/nxp/backend/ir/lib/tflite/HashtableOptions.py +95 -0
  827. executorch/backends/nxp/backend/ir/lib/tflite/HashtableSizeOptions.py +50 -0
  828. executorch/backends/nxp/backend/ir/lib/tflite/IfOptions.py +80 -0
  829. executorch/backends/nxp/backend/ir/lib/tflite/Int32Vector.py +98 -0
  830. executorch/backends/nxp/backend/ir/lib/tflite/L2NormOptions.py +65 -0
  831. executorch/backends/nxp/backend/ir/lib/tflite/LSHProjectionOptions.py +65 -0
  832. executorch/backends/nxp/backend/ir/lib/tflite/LSHProjectionType.py +9 -0
  833. executorch/backends/nxp/backend/ir/lib/tflite/LSTMKernelType.py +8 -0
  834. executorch/backends/nxp/backend/ir/lib/tflite/LSTMOptions.py +131 -0
  835. executorch/backends/nxp/backend/ir/lib/tflite/LeakyReluOptions.py +67 -0
  836. executorch/backends/nxp/backend/ir/lib/tflite/LessEqualOptions.py +50 -0
  837. executorch/backends/nxp/backend/ir/lib/tflite/LessOptions.py +50 -0
  838. executorch/backends/nxp/backend/ir/lib/tflite/LocalResponseNormalizationOptions.py +118 -0
  839. executorch/backends/nxp/backend/ir/lib/tflite/LogSoftmaxOptions.py +50 -0
  840. executorch/backends/nxp/backend/ir/lib/tflite/LogicalAndOptions.py +50 -0
  841. executorch/backends/nxp/backend/ir/lib/tflite/LogicalNotOptions.py +50 -0
  842. executorch/backends/nxp/backend/ir/lib/tflite/LogicalOrOptions.py +50 -0
  843. executorch/backends/nxp/backend/ir/lib/tflite/MatrixDiagOptions.py +50 -0
  844. executorch/backends/nxp/backend/ir/lib/tflite/MatrixSetDiagOptions.py +50 -0
  845. executorch/backends/nxp/backend/ir/lib/tflite/MaximumMinimumOptions.py +50 -0
  846. executorch/backends/nxp/backend/ir/lib/tflite/Metadata.py +84 -0
  847. executorch/backends/nxp/backend/ir/lib/tflite/MirrorPadMode.py +8 -0
  848. executorch/backends/nxp/backend/ir/lib/tflite/MirrorPadOptions.py +65 -0
  849. executorch/backends/nxp/backend/ir/lib/tflite/Model.py +352 -0
  850. executorch/backends/nxp/backend/ir/lib/tflite/MulOptions.py +65 -0
  851. executorch/backends/nxp/backend/ir/lib/tflite/NegOptions.py +50 -0
  852. executorch/backends/nxp/backend/ir/lib/tflite/NonMaxSuppressionV4Options.py +52 -0
  853. executorch/backends/nxp/backend/ir/lib/tflite/NonMaxSuppressionV5Options.py +52 -0
  854. executorch/backends/nxp/backend/ir/lib/tflite/NotEqualOptions.py +50 -0
  855. executorch/backends/nxp/backend/ir/lib/tflite/OneHotOptions.py +65 -0
  856. executorch/backends/nxp/backend/ir/lib/tflite/Operator.py +428 -0
  857. executorch/backends/nxp/backend/ir/lib/tflite/OperatorCode.py +112 -0
  858. executorch/backends/nxp/backend/ir/lib/tflite/PackOptions.py +80 -0
  859. executorch/backends/nxp/backend/ir/lib/tflite/PadOptions.py +50 -0
  860. executorch/backends/nxp/backend/ir/lib/tflite/PadV2Options.py +50 -0
  861. executorch/backends/nxp/backend/ir/lib/tflite/Padding.py +8 -0
  862. executorch/backends/nxp/backend/ir/lib/tflite/Pool2DOptions.py +140 -0
  863. executorch/backends/nxp/backend/ir/lib/tflite/PowOptions.py +50 -0
  864. executorch/backends/nxp/backend/ir/lib/tflite/QuantizationDetails.py +8 -0
  865. executorch/backends/nxp/backend/ir/lib/tflite/QuantizationParameters.py +295 -0
  866. executorch/backends/nxp/backend/ir/lib/tflite/QuantizeOptions.py +50 -0
  867. executorch/backends/nxp/backend/ir/lib/tflite/RNNOptions.py +82 -0
  868. executorch/backends/nxp/backend/ir/lib/tflite/RandomOptions.py +80 -0
  869. executorch/backends/nxp/backend/ir/lib/tflite/RangeOptions.py +50 -0
  870. executorch/backends/nxp/backend/ir/lib/tflite/RankOptions.py +50 -0
  871. executorch/backends/nxp/backend/ir/lib/tflite/ReadVariableOptions.py +50 -0
  872. executorch/backends/nxp/backend/ir/lib/tflite/ReduceWindowFunction.py +13 -0
  873. executorch/backends/nxp/backend/ir/lib/tflite/ReduceWindowOptions.py +65 -0
  874. executorch/backends/nxp/backend/ir/lib/tflite/ReducerOptions.py +67 -0
  875. executorch/backends/nxp/backend/ir/lib/tflite/ReshapeOptions.py +98 -0
  876. executorch/backends/nxp/backend/ir/lib/tflite/ResizeBilinearOptions.py +84 -0
  877. executorch/backends/nxp/backend/ir/lib/tflite/ResizeNearestNeighborOptions.py +86 -0
  878. executorch/backends/nxp/backend/ir/lib/tflite/ReverseSequenceOptions.py +82 -0
  879. executorch/backends/nxp/backend/ir/lib/tflite/ReverseV2Options.py +50 -0
  880. executorch/backends/nxp/backend/ir/lib/tflite/Rfft2dOptions.py +50 -0
  881. executorch/backends/nxp/backend/ir/lib/tflite/RightShiftOptions.py +50 -0
  882. executorch/backends/nxp/backend/ir/lib/tflite/RngAlgorithm.py +9 -0
  883. executorch/backends/nxp/backend/ir/lib/tflite/SVDFOptions.py +97 -0
  884. executorch/backends/nxp/backend/ir/lib/tflite/ScatterNdOptions.py +50 -0
  885. executorch/backends/nxp/backend/ir/lib/tflite/SegmentSumOptions.py +50 -0
  886. executorch/backends/nxp/backend/ir/lib/tflite/SelectOptions.py +50 -0
  887. executorch/backends/nxp/backend/ir/lib/tflite/SelectV2Options.py +50 -0
  888. executorch/backends/nxp/backend/ir/lib/tflite/SequenceRNNOptions.py +99 -0
  889. executorch/backends/nxp/backend/ir/lib/tflite/ShapeOptions.py +65 -0
  890. executorch/backends/nxp/backend/ir/lib/tflite/SignOptions.py +50 -0
  891. executorch/backends/nxp/backend/ir/lib/tflite/SignatureDef.py +172 -0
  892. executorch/backends/nxp/backend/ir/lib/tflite/SkipGramOptions.py +97 -0
  893. executorch/backends/nxp/backend/ir/lib/tflite/SliceOptions.py +50 -0
  894. executorch/backends/nxp/backend/ir/lib/tflite/SoftmaxOptions.py +67 -0
  895. executorch/backends/nxp/backend/ir/lib/tflite/SpaceToBatchNDOptions.py +50 -0
  896. executorch/backends/nxp/backend/ir/lib/tflite/SpaceToDepthOptions.py +65 -0
  897. executorch/backends/nxp/backend/ir/lib/tflite/SparseIndexVector.py +10 -0
  898. executorch/backends/nxp/backend/ir/lib/tflite/SparseToDenseOptions.py +67 -0
  899. executorch/backends/nxp/backend/ir/lib/tflite/SparsityParameters.py +190 -0
  900. executorch/backends/nxp/backend/ir/lib/tflite/SplitOptions.py +65 -0
  901. executorch/backends/nxp/backend/ir/lib/tflite/SplitVOptions.py +65 -0
  902. executorch/backends/nxp/backend/ir/lib/tflite/SquareOptions.py +50 -0
  903. executorch/backends/nxp/backend/ir/lib/tflite/SquaredDifferenceOptions.py +52 -0
  904. executorch/backends/nxp/backend/ir/lib/tflite/SqueezeOptions.py +98 -0
  905. executorch/backends/nxp/backend/ir/lib/tflite/StablehloBroadcastInDimOptions.py +102 -0
  906. executorch/backends/nxp/backend/ir/lib/tflite/StablehloCompareOptions.py +86 -0
  907. executorch/backends/nxp/backend/ir/lib/tflite/StablehloComparisonDirection.py +12 -0
  908. executorch/backends/nxp/backend/ir/lib/tflite/StablehloComparisonType.py +11 -0
  909. executorch/backends/nxp/backend/ir/lib/tflite/StablehloConcatenateOptions.py +67 -0
  910. executorch/backends/nxp/backend/ir/lib/tflite/StablehloConvolutionOptions.py +634 -0
  911. executorch/backends/nxp/backend/ir/lib/tflite/StablehloCustomCallOptions.py +214 -0
  912. executorch/backends/nxp/backend/ir/lib/tflite/StablehloDotGeneralOptions.py +304 -0
  913. executorch/backends/nxp/backend/ir/lib/tflite/StablehloDynamicSliceOptions.py +100 -0
  914. executorch/backends/nxp/backend/ir/lib/tflite/StablehloGatherOptions.py +276 -0
  915. executorch/backends/nxp/backend/ir/lib/tflite/StablehloIotaOptions.py +65 -0
  916. executorch/backends/nxp/backend/ir/lib/tflite/StablehloPadOptions.py +194 -0
  917. executorch/backends/nxp/backend/ir/lib/tflite/StablehloPrecisionConfig.py +9 -0
  918. executorch/backends/nxp/backend/ir/lib/tflite/StablehloReduceOptions.py +115 -0
  919. executorch/backends/nxp/backend/ir/lib/tflite/StablehloReduceWindowOptions.py +307 -0
  920. executorch/backends/nxp/backend/ir/lib/tflite/StablehloRngBitGeneratorOptions.py +67 -0
  921. executorch/backends/nxp/backend/ir/lib/tflite/StablehloScatterOptions.py +268 -0
  922. executorch/backends/nxp/backend/ir/lib/tflite/StablehloSliceOptions.py +194 -0
  923. executorch/backends/nxp/backend/ir/lib/tflite/StablehloSortOptions.py +97 -0
  924. executorch/backends/nxp/backend/ir/lib/tflite/StablehloTransposeOptions.py +100 -0
  925. executorch/backends/nxp/backend/ir/lib/tflite/StablehloWhileOptions.py +80 -0
  926. executorch/backends/nxp/backend/ir/lib/tflite/StridedSliceOptions.py +142 -0
  927. executorch/backends/nxp/backend/ir/lib/tflite/SubGraph.py +251 -0
  928. executorch/backends/nxp/backend/ir/lib/tflite/SubOptions.py +82 -0
  929. executorch/backends/nxp/backend/ir/lib/tflite/Tensor.py +317 -0
  930. executorch/backends/nxp/backend/ir/lib/tflite/TensorMap.py +84 -0
  931. executorch/backends/nxp/backend/ir/lib/tflite/TensorType.py +24 -0
  932. executorch/backends/nxp/backend/ir/lib/tflite/TileOptions.py +50 -0
  933. executorch/backends/nxp/backend/ir/lib/tflite/TopKV2Options.py +50 -0
  934. executorch/backends/nxp/backend/ir/lib/tflite/TransposeConvOptions.py +125 -0
  935. executorch/backends/nxp/backend/ir/lib/tflite/TransposeOptions.py +50 -0
  936. executorch/backends/nxp/backend/ir/lib/tflite/Uint16Vector.py +98 -0
  937. executorch/backends/nxp/backend/ir/lib/tflite/Uint8Vector.py +98 -0
  938. executorch/backends/nxp/backend/ir/lib/tflite/UnidirectionalSequenceLSTMOptions.py +164 -0
  939. executorch/backends/nxp/backend/ir/lib/tflite/UniqueOptions.py +65 -0
  940. executorch/backends/nxp/backend/ir/lib/tflite/UnpackOptions.py +80 -0
  941. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentMaxOptions.py +52 -0
  942. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentMinOptions.py +52 -0
  943. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentProdOptions.py +52 -0
  944. executorch/backends/nxp/backend/ir/lib/tflite/UnsortedSegmentSumOptions.py +52 -0
  945. executorch/backends/nxp/backend/ir/lib/tflite/VarHandleOptions.py +84 -0
  946. executorch/backends/nxp/backend/ir/lib/tflite/VariantSubType.py +130 -0
  947. executorch/backends/nxp/backend/ir/lib/tflite/WhereOptions.py +50 -0
  948. executorch/backends/nxp/backend/ir/lib/tflite/WhileOptions.py +80 -0
  949. executorch/backends/nxp/backend/ir/lib/tflite/ZerosLikeOptions.py +50 -0
  950. executorch/backends/nxp/backend/ir/lib/tflite/__init__.py +0 -0
  951. executorch/backends/nxp/backend/ir/logger.py +341 -0
  952. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/__init__.py +0 -0
  953. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/graph_utils.py +115 -0
  954. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/operator_rules.py +102 -0
  955. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/__init__.py +0 -0
  956. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/base_optimization.py +41 -0
  957. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/permute_fully_connected_weights_after_reshape.py +121 -0
  958. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizations/prune_transpose_operators.py +174 -0
  959. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/optimizer.py +127 -0
  960. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/pattern_matcher.py +925 -0
  961. executorch/backends/nxp/backend/ir/neutron_ir_post_processing/tensor_rules.py +710 -0
  962. executorch/backends/nxp/backend/ir/tensor_formatting.py +69 -0
  963. executorch/backends/nxp/backend/ir/tflite_generator/__init__.py +0 -0
  964. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/__init__.py +0 -0
  965. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/abs_options.py +25 -0
  966. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/add_n_options.py +32 -0
  967. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/add_options.py +46 -0
  968. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/arg_max_options.py +28 -0
  969. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/arg_min_options.py +28 -0
  970. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/average_pool_2d_options.py +63 -0
  971. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/batch_mat_mul_options.py +45 -0
  972. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bidirectional_sequence_lstm_options.py +65 -0
  973. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bidirectional_sequence_rnn_options.py +53 -0
  974. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/bitwise_xor_options.py +23 -0
  975. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/broadcast_to_options.py +25 -0
  976. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/cast_options.py +32 -0
  977. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/concatenation_options.py +38 -0
  978. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/conv_2d_options.py +59 -0
  979. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/conv_3d_options.py +67 -0
  980. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/cum_sum_options.py +30 -0
  981. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/depth_to_space_options.py +29 -0
  982. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/depthwise_conv_2d_options.py +69 -0
  983. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/dequantize_options.py +24 -0
  984. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/div_options.py +43 -0
  985. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/equal_options.py +23 -0
  986. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/exp_options.py +23 -0
  987. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/floor_mod_options.py +23 -0
  988. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/fully_connected_options.py +57 -0
  989. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gather_nd_options.py +23 -0
  990. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gather_options.py +30 -0
  991. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/gelu_options.py +27 -0
  992. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/greater_equal_options.py +25 -0
  993. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/greater_options.py +23 -0
  994. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/hard_swish_options.py +23 -0
  995. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/leaky_relu_options.py +32 -0
  996. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/less_equal_options.py +23 -0
  997. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/less_options.py +23 -0
  998. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/log_softmax_options.py +30 -0
  999. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_and_options.py +23 -0
  1000. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_not_options.py +23 -0
  1001. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/logical_or_options.py +23 -0
  1002. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/lrn_options.py +45 -0
  1003. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/lstm_options.py +60 -0
  1004. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/max_pool_2d_options.py +63 -0
  1005. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/maximum_options.py +21 -0
  1006. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mean_options.py +28 -0
  1007. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/minimum_options.py +21 -0
  1008. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mirror_pad_options.py +60 -0
  1009. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/mul_options.py +36 -0
  1010. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/multinomial_options.py +30 -0
  1011. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/neg_options.py +23 -0
  1012. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/not_equal_options.py +23 -0
  1013. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/one_hot_options.py +27 -0
  1014. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pad_options.py +27 -0
  1015. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pad_v2_options.py +27 -0
  1016. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/pow_options.py +24 -0
  1017. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/quantize_options.py +30 -0
  1018. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/range_options.py +23 -0
  1019. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_max_options.py +28 -0
  1020. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_min_options.py +28 -0
  1021. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reduce_prod_options.py +28 -0
  1022. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reshape_options.py +53 -0
  1023. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/resize_bilinear_options.py +34 -0
  1024. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/resize_nearest_neighbor_options.py +37 -0
  1025. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/reverse_sequence_options.py +32 -0
  1026. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/scatter_nd_options.py +23 -0
  1027. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/select_v2_options.py +24 -0
  1028. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/shape_options.py +34 -0
  1029. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sign_options.py +22 -0
  1030. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/slice_options.py +29 -0
  1031. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/softmax_options.py +36 -0
  1032. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/space_to_depth_options.py +29 -0
  1033. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/split_options.py +27 -0
  1034. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/split_v_options.py +27 -0
  1035. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/square_options.py +23 -0
  1036. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/squared_difference_options.py +25 -0
  1037. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/squeeze_options.py +42 -0
  1038. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/strided_slice_options.py +35 -0
  1039. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sub_options.py +46 -0
  1040. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/sum_options.py +28 -0
  1041. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/tile_options.py +23 -0
  1042. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/transpose_conv_options.py +50 -0
  1043. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/transpose_options.py +30 -0
  1044. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/unidirectional_sequence_lstm_options.py +67 -0
  1045. executorch/backends/nxp/backend/ir/tflite_generator/builtin_options/unidirectional_sequence_rnn_options.py +49 -0
  1046. executorch/backends/nxp/backend/ir/tflite_generator/custom_options/flex_transpose_options.py +78 -0
  1047. executorch/backends/nxp/backend/ir/tflite_generator/meta/__init__.py +0 -0
  1048. executorch/backends/nxp/backend/ir/tflite_generator/meta/meta.py +255 -0
  1049. executorch/backends/nxp/backend/ir/tflite_generator/meta/types.py +198 -0
  1050. executorch/backends/nxp/backend/ir/tflite_generator/tflite_model.py +814 -0
  1051. executorch/backends/nxp/backend/neutron_converter_manager.py +117 -0
  1052. executorch/backends/nxp/backend/neutron_operator_support.py +79 -0
  1053. executorch/backends/nxp/backend/neutron_target_spec.py +143 -0
  1054. executorch/backends/nxp/backend/node_format.py +26 -0
  1055. executorch/backends/nxp/backend/node_format_inference.py +294 -0
  1056. executorch/backends/nxp/edge_passes/move_auxiliary_operator_into_separate_qdq_cluster_pass.py +254 -0
  1057. executorch/backends/nxp/edge_passes/neutron_edge_pass.py +55 -0
  1058. executorch/backends/nxp/edge_passes/neutron_edge_pass_manager.py +29 -0
  1059. executorch/backends/nxp/edge_passes/remove_additional_quantize_dequantize_nodes_pass.py +111 -0
  1060. executorch/backends/nxp/edge_passes/remove_as_strided_copy_nodes.py +66 -0
  1061. executorch/backends/nxp/edge_passes/remove_io_quant_ops_pass.py +79 -0
  1062. executorch/backends/nxp/neutron_node_extraction.py +152 -0
  1063. executorch/backends/nxp/neutron_partitioner.py +421 -0
  1064. executorch/backends/nxp/neutron_pass_manager.py +50 -0
  1065. executorch/backends/nxp/nxp_backend.py +391 -0
  1066. executorch/backends/nxp/quantizer/neutron_quantizer.py +353 -0
  1067. executorch/backends/nxp/quantizer/patterns.py +1089 -0
  1068. executorch/backends/nxp/quantizer/utils.py +195 -0
  1069. executorch/backends/nxp/tests/executorch_pipeline.py +184 -0
  1070. executorch/backends/nxp/tests/executors.py +401 -0
  1071. executorch/backends/nxp/tests/ir/__init__.py +0 -0
  1072. executorch/backends/nxp/tests/ir/converter/__init__.py +0 -0
  1073. executorch/backends/nxp/tests/ir/converter/node_converter/__init__.py +0 -0
  1074. executorch/backends/nxp/tests/ir/converter/node_converter/test_abs_converter.py +91 -0
  1075. executorch/backends/nxp/tests/ir/converter/node_converter/test_adaptive_avg_pool2d_converter.py +148 -0
  1076. executorch/backends/nxp/tests/ir/converter/node_converter/test_add_tensor_converter.py +161 -0
  1077. executorch/backends/nxp/tests/ir/converter/node_converter/test_addmm_converter.py +96 -0
  1078. executorch/backends/nxp/tests/ir/converter/node_converter/test_avg_pool2d_converter.py +221 -0
  1079. executorch/backends/nxp/tests/ir/converter/node_converter/test_cat_converter.py +495 -0
  1080. executorch/backends/nxp/tests/ir/converter/node_converter/test_clone_converter.py +344 -0
  1081. executorch/backends/nxp/tests/ir/converter/node_converter/test_constant_pad_nd_converter.py +188 -0
  1082. executorch/backends/nxp/tests/ir/converter/node_converter/test_conv_converter.py +680 -0
  1083. executorch/backends/nxp/tests/ir/converter/node_converter/test_hardtanh_converter.py +107 -0
  1084. executorch/backends/nxp/tests/ir/converter/node_converter/test_max_pool_2d_converter.py +132 -0
  1085. executorch/backends/nxp/tests/ir/converter/node_converter/test_mean_dim_converter.py +280 -0
  1086. executorch/backends/nxp/tests/ir/converter/node_converter/test_mm_converter.py +96 -0
  1087. executorch/backends/nxp/tests/ir/converter/node_converter/test_mul_tensor_converter.py +212 -0
  1088. executorch/backends/nxp/tests/ir/converter/node_converter/test_permute_copy_converter.py +426 -0
  1089. executorch/backends/nxp/tests/ir/converter/node_converter/test_relu_converter.py +114 -0
  1090. executorch/backends/nxp/tests/ir/converter/node_converter/test_sigmoid_converter.py +78 -0
  1091. executorch/backends/nxp/tests/ir/converter/node_converter/test_slice_tensor_converter.py +244 -0
  1092. executorch/backends/nxp/tests/ir/converter/node_converter/test_softmax_converter.py +115 -0
  1093. executorch/backends/nxp/tests/ir/converter/node_converter/test_sub_tensor_converter.py +181 -0
  1094. executorch/backends/nxp/tests/ir/converter/node_converter/test_tanh_converter.py +90 -0
  1095. executorch/backends/nxp/tests/ir/converter/node_converter/test_view_copy_converter.py +509 -0
  1096. executorch/backends/nxp/tests/ir/edge_passes/__init__.py +0 -0
  1097. executorch/backends/nxp/tests/ir/edge_passes/test_remove_io_quant_ops_pass.py +127 -0
  1098. executorch/backends/nxp/tests/models.py +737 -0
  1099. executorch/backends/nxp/tests/test_batch_norm_fusion.py +233 -0
  1100. executorch/backends/nxp/tests/test_context_sensitive_delegation.py +71 -0
  1101. executorch/backends/nxp/tests/test_convert_unsqueeze_to_view.py +143 -0
  1102. executorch/backends/nxp/tests/test_decompose_split_to_slices.py +315 -0
  1103. executorch/backends/nxp/tests/test_edge_passes.py +382 -0
  1104. executorch/backends/nxp/tests/test_gru_splitting.py +267 -0
  1105. executorch/backends/nxp/tests/test_integration.py +53 -0
  1106. executorch/backends/nxp/tests/test_linear_and_add_fusion.py +653 -0
  1107. executorch/backends/nxp/tests/test_move_activation_before_concatenation.py +905 -0
  1108. executorch/backends/nxp/tests/test_neutron_backend.py +63 -0
  1109. executorch/backends/nxp/tests/test_neutron_backend_executor.py +327 -0
  1110. executorch/backends/nxp/tests/test_neutron_converter_manager.py +59 -0
  1111. executorch/backends/nxp/tests/test_node_format_inference.py +91 -0
  1112. executorch/backends/nxp/tests/test_operator_selector.py +24 -0
  1113. executorch/backends/nxp/tests/test_per_channel_conversion.py +182 -0
  1114. executorch/backends/nxp/tests/test_qdq_clustering_conv.py +31 -0
  1115. executorch/backends/nxp/tests/test_quantizer.py +703 -0
  1116. executorch/backends/nxp/tests/test_removing_dead_code.py +65 -0
  1117. executorch/backends/nxp/tests/test_removing_nodes_with_known_outputs.py +116 -0
  1118. executorch/backends/nxp/tests/test_split_group_convolution.py +291 -0
  1119. executorch/backends/nxp/tests/use_qat.py +11 -0
  1120. executorch/backends/openvino/__init__.py +5 -0
  1121. executorch/backends/openvino/partitioner.py +238 -0
  1122. executorch/backends/openvino/preprocess.py +62 -0
  1123. executorch/backends/openvino/quantizer/__init__.py +3 -0
  1124. executorch/backends/openvino/quantizer/observers.py +186 -0
  1125. executorch/backends/openvino/quantizer/quantizer.py +595 -0
  1126. executorch/backends/openvino/tests/models/test_classification.py +38 -0
  1127. executorch/backends/openvino/tests/ops/base_openvino_op_test.py +86 -0
  1128. executorch/backends/openvino/tests/ops/test_add.py +22 -0
  1129. executorch/backends/openvino/tests/ops/test_addmm.py +28 -0
  1130. executorch/backends/openvino/tests/ops/test_arange.py +23 -0
  1131. executorch/backends/openvino/tests/ops/test_batch_norm.py +61 -0
  1132. executorch/backends/openvino/tests/ops/test_convolution.py +296 -0
  1133. executorch/backends/openvino/tests/ops/test_mean.py +90 -0
  1134. executorch/backends/openvino/tests/ops/test_permute.py +33 -0
  1135. executorch/backends/openvino/tests/ops/test_pooling.py +91 -0
  1136. executorch/backends/openvino/tests/ops/test_unary_ops.py +37 -0
  1137. executorch/backends/openvino/tests/ops/test_view.py +35 -0
  1138. executorch/backends/openvino/tests/test_runner.py +76 -0
  1139. executorch/backends/qualcomm/__init__.py +13 -0
  1140. executorch/backends/qualcomm/_passes/__init__.py +101 -0
  1141. executorch/backends/qualcomm/_passes/annotate_adaptive_avg_pool1d.py +43 -0
  1142. executorch/backends/qualcomm/_passes/annotate_quant_attrs.py +145 -0
  1143. executorch/backends/qualcomm/_passes/annotate_stack.py +44 -0
  1144. executorch/backends/qualcomm/_passes/annotate_unbind.py +42 -0
  1145. executorch/backends/qualcomm/_passes/build_quant_io.py +49 -0
  1146. executorch/backends/qualcomm/_passes/canonicalize_conv.py +204 -0
  1147. executorch/backends/qualcomm/_passes/convert_bmm_to_matmul.py +83 -0
  1148. executorch/backends/qualcomm/_passes/convert_linear_to_conv2d.py +232 -0
  1149. executorch/backends/qualcomm/_passes/convert_mha_to_sha.py +626 -0
  1150. executorch/backends/qualcomm/_passes/convert_square_to_pow.py +38 -0
  1151. executorch/backends/qualcomm/_passes/decompose_any.py +64 -0
  1152. executorch/backends/qualcomm/_passes/decompose_binary_alpha.py +61 -0
  1153. executorch/backends/qualcomm/_passes/decompose_cdist.py +69 -0
  1154. executorch/backends/qualcomm/_passes/decompose_col_im.py +121 -0
  1155. executorch/backends/qualcomm/_passes/decompose_einsum.py +51 -0
  1156. executorch/backends/qualcomm/_passes/decompose_expm1.py +46 -0
  1157. executorch/backends/qualcomm/_passes/decompose_floor_divide.py +62 -0
  1158. executorch/backends/qualcomm/_passes/decompose_glu.py +55 -0
  1159. executorch/backends/qualcomm/_passes/decompose_linalg_vector_norm.py +75 -0
  1160. executorch/backends/qualcomm/_passes/decompose_maxpool3d.py +133 -0
  1161. executorch/backends/qualcomm/_passes/decompose_minmaxdim.py +109 -0
  1162. executorch/backends/qualcomm/_passes/decompose_roll.py +78 -0
  1163. executorch/backends/qualcomm/_passes/decompose_silu.py +45 -0
  1164. executorch/backends/qualcomm/_passes/decompose_threshold.py +61 -0
  1165. executorch/backends/qualcomm/_passes/decompose_triu.py +71 -0
  1166. executorch/backends/qualcomm/_passes/decompose_wrap_with_autocast.py +79 -0
  1167. executorch/backends/qualcomm/_passes/expand_broadcast_tensor_shape.py +67 -0
  1168. executorch/backends/qualcomm/_passes/fixed_linear_keep_dim.py +95 -0
  1169. executorch/backends/qualcomm/_passes/fold_qdq.py +80 -0
  1170. executorch/backends/qualcomm/_passes/fuse_consecutive_cast.py +116 -0
  1171. executorch/backends/qualcomm/_passes/fuse_consecutive_transpose.py +109 -0
  1172. executorch/backends/qualcomm/_passes/i64_to_i32.py +184 -0
  1173. executorch/backends/qualcomm/_passes/insert_io_qdq.py +154 -0
  1174. executorch/backends/qualcomm/_passes/insert_requantize.py +111 -0
  1175. executorch/backends/qualcomm/_passes/insert_reshape_for_reduce_ops.py +59 -0
  1176. executorch/backends/qualcomm/_passes/layout_transform.py +355 -0
  1177. executorch/backends/qualcomm/_passes/lift_constant_scalar_operands.py +186 -0
  1178. executorch/backends/qualcomm/_passes/qnn_pass_manager.py +271 -0
  1179. executorch/backends/qualcomm/_passes/recompose_pad_maxpool2d.py +147 -0
  1180. executorch/backends/qualcomm/_passes/recompose_pixel_unshuffle.py +81 -0
  1181. executorch/backends/qualcomm/_passes/recompose_rms_norm.py +155 -0
  1182. executorch/backends/qualcomm/_passes/reduce_dynamic_range.py +60 -0
  1183. executorch/backends/qualcomm/_passes/remove_0d_tensor.py +36 -0
  1184. executorch/backends/qualcomm/_passes/remove_redundancy.py +73 -0
  1185. executorch/backends/qualcomm/_passes/replace_arange_args.py +48 -0
  1186. executorch/backends/qualcomm/_passes/replace_inf_values.py +60 -0
  1187. executorch/backends/qualcomm/_passes/seq_mse.py +224 -0
  1188. executorch/backends/qualcomm/_passes/tag_quant_io.py +45 -0
  1189. executorch/backends/qualcomm/_passes/utils.py +283 -0
  1190. executorch/backends/qualcomm/builders/__init__.py +221 -0
  1191. executorch/backends/qualcomm/builders/node_visitor.py +518 -0
  1192. executorch/backends/qualcomm/builders/node_visitor_manager.py +79 -0
  1193. executorch/backends/qualcomm/builders/op_abs.py +57 -0
  1194. executorch/backends/qualcomm/builders/op_adaptive_avg_pool2d.py +126 -0
  1195. executorch/backends/qualcomm/builders/op_adaptive_max_pool2d.py +151 -0
  1196. executorch/backends/qualcomm/builders/op_add.py +62 -0
  1197. executorch/backends/qualcomm/builders/op_amax.py +90 -0
  1198. executorch/backends/qualcomm/builders/op_amin.py +90 -0
  1199. executorch/backends/qualcomm/builders/op_and.py +60 -0
  1200. executorch/backends/qualcomm/builders/op_arange.py +41 -0
  1201. executorch/backends/qualcomm/builders/op_argmax.py +78 -0
  1202. executorch/backends/qualcomm/builders/op_argmin.py +78 -0
  1203. executorch/backends/qualcomm/builders/op_asin.py +56 -0
  1204. executorch/backends/qualcomm/builders/op_atan.py +55 -0
  1205. executorch/backends/qualcomm/builders/op_avg_pool2d.py +168 -0
  1206. executorch/backends/qualcomm/builders/op_avg_pool3d.py +299 -0
  1207. executorch/backends/qualcomm/builders/op_batch_norm.py +151 -0
  1208. executorch/backends/qualcomm/builders/op_binary.py +84 -0
  1209. executorch/backends/qualcomm/builders/op_bmm.py +59 -0
  1210. executorch/backends/qualcomm/builders/op_cat.py +89 -0
  1211. executorch/backends/qualcomm/builders/op_ceil.py +56 -0
  1212. executorch/backends/qualcomm/builders/op_clamp.py +80 -0
  1213. executorch/backends/qualcomm/builders/op_conv.py +263 -0
  1214. executorch/backends/qualcomm/builders/op_copy.py +112 -0
  1215. executorch/backends/qualcomm/builders/op_cos.py +56 -0
  1216. executorch/backends/qualcomm/builders/op_cum_sum.py +87 -0
  1217. executorch/backends/qualcomm/builders/op_custom_op.py +93 -0
  1218. executorch/backends/qualcomm/builders/op_depth_to_space.py +78 -0
  1219. executorch/backends/qualcomm/builders/op_dequantize.py +72 -0
  1220. executorch/backends/qualcomm/builders/op_div.py +62 -0
  1221. executorch/backends/qualcomm/builders/op_elu.py +68 -0
  1222. executorch/backends/qualcomm/builders/op_embedding.py +79 -0
  1223. executorch/backends/qualcomm/builders/op_eq.py +62 -0
  1224. executorch/backends/qualcomm/builders/op_exp.py +60 -0
  1225. executorch/backends/qualcomm/builders/op_expand.py +84 -0
  1226. executorch/backends/qualcomm/builders/op_flip.py +81 -0
  1227. executorch/backends/qualcomm/builders/op_floor.py +56 -0
  1228. executorch/backends/qualcomm/builders/op_full.py +41 -0
  1229. executorch/backends/qualcomm/builders/op_full_like.py +41 -0
  1230. executorch/backends/qualcomm/builders/op_gather.py +76 -0
  1231. executorch/backends/qualcomm/builders/op_ge.py +62 -0
  1232. executorch/backends/qualcomm/builders/op_gelu.py +57 -0
  1233. executorch/backends/qualcomm/builders/op_grid_sampler_2d.py +162 -0
  1234. executorch/backends/qualcomm/builders/op_group_norm.py +94 -0
  1235. executorch/backends/qualcomm/builders/op_gt.py +62 -0
  1236. executorch/backends/qualcomm/builders/op_hardsigmoid.py +78 -0
  1237. executorch/backends/qualcomm/builders/op_hardswish.py +57 -0
  1238. executorch/backends/qualcomm/builders/op_hardtanh.py +79 -0
  1239. executorch/backends/qualcomm/builders/op_index.py +85 -0
  1240. executorch/backends/qualcomm/builders/op_index_put.py +481 -0
  1241. executorch/backends/qualcomm/builders/op_index_select.py +81 -0
  1242. executorch/backends/qualcomm/builders/op_instance_norm.py +112 -0
  1243. executorch/backends/qualcomm/builders/op_layer_norm.py +113 -0
  1244. executorch/backends/qualcomm/builders/op_le.py +62 -0
  1245. executorch/backends/qualcomm/builders/op_linear.py +98 -0
  1246. executorch/backends/qualcomm/builders/op_log.py +58 -0
  1247. executorch/backends/qualcomm/builders/op_log_softmax.py +81 -0
  1248. executorch/backends/qualcomm/builders/op_logical_not.py +56 -0
  1249. executorch/backends/qualcomm/builders/op_lt.py +62 -0
  1250. executorch/backends/qualcomm/builders/op_matmul.py +59 -0
  1251. executorch/backends/qualcomm/builders/op_max.py +62 -0
  1252. executorch/backends/qualcomm/builders/op_max_dim.py +89 -0
  1253. executorch/backends/qualcomm/builders/op_max_pool2d.py +145 -0
  1254. executorch/backends/qualcomm/builders/op_mean_dim.py +100 -0
  1255. executorch/backends/qualcomm/builders/op_min.py +62 -0
  1256. executorch/backends/qualcomm/builders/op_min_dim.py +89 -0
  1257. executorch/backends/qualcomm/builders/op_mul.py +62 -0
  1258. executorch/backends/qualcomm/builders/op_ne.py +62 -0
  1259. executorch/backends/qualcomm/builders/op_neg.py +54 -0
  1260. executorch/backends/qualcomm/builders/op_or.py +60 -0
  1261. executorch/backends/qualcomm/builders/op_pad.py +97 -0
  1262. executorch/backends/qualcomm/builders/op_pow.py +75 -0
  1263. executorch/backends/qualcomm/builders/op_prelu.py +81 -0
  1264. executorch/backends/qualcomm/builders/op_quantize.py +73 -0
  1265. executorch/backends/qualcomm/builders/op_relu.py +58 -0
  1266. executorch/backends/qualcomm/builders/op_repeat.py +68 -0
  1267. executorch/backends/qualcomm/builders/op_reshape.py +55 -0
  1268. executorch/backends/qualcomm/builders/op_resize.py +85 -0
  1269. executorch/backends/qualcomm/builders/op_rms_norm.py +132 -0
  1270. executorch/backends/qualcomm/builders/op_round.py +58 -0
  1271. executorch/backends/qualcomm/builders/op_rsqrt.py +58 -0
  1272. executorch/backends/qualcomm/builders/op_scalar_tensor.py +51 -0
  1273. executorch/backends/qualcomm/builders/op_select_copy.py +89 -0
  1274. executorch/backends/qualcomm/builders/op_sigmoid.py +58 -0
  1275. executorch/backends/qualcomm/builders/op_sign.py +56 -0
  1276. executorch/backends/qualcomm/builders/op_sin.py +56 -0
  1277. executorch/backends/qualcomm/builders/op_skip_ops.py +55 -0
  1278. executorch/backends/qualcomm/builders/op_slice_copy.py +95 -0
  1279. executorch/backends/qualcomm/builders/op_slice_scatter.py +122 -0
  1280. executorch/backends/qualcomm/builders/op_softmax.py +79 -0
  1281. executorch/backends/qualcomm/builders/op_space_to_depth.py +78 -0
  1282. executorch/backends/qualcomm/builders/op_split_with_sizes.py +99 -0
  1283. executorch/backends/qualcomm/builders/op_sqrt.py +60 -0
  1284. executorch/backends/qualcomm/builders/op_squeeze.py +57 -0
  1285. executorch/backends/qualcomm/builders/op_stack.py +71 -0
  1286. executorch/backends/qualcomm/builders/op_sub.py +62 -0
  1287. executorch/backends/qualcomm/builders/op_sum_int_list.py +84 -0
  1288. executorch/backends/qualcomm/builders/op_tanh.py +57 -0
  1289. executorch/backends/qualcomm/builders/op_to.py +149 -0
  1290. executorch/backends/qualcomm/builders/op_topk.py +112 -0
  1291. executorch/backends/qualcomm/builders/op_transpose.py +76 -0
  1292. executorch/backends/qualcomm/builders/op_unbind.py +73 -0
  1293. executorch/backends/qualcomm/builders/op_unsqueeze.py +57 -0
  1294. executorch/backends/qualcomm/builders/op_upsample_bilinear2d.py +68 -0
  1295. executorch/backends/qualcomm/builders/op_upsample_nearest2d.py +68 -0
  1296. executorch/backends/qualcomm/builders/op_where.py +82 -0
  1297. executorch/backends/qualcomm/builders/op_xor.py +60 -0
  1298. executorch/backends/qualcomm/builders/qnn_constants.py +630 -0
  1299. executorch/backends/qualcomm/builders/utils.py +160 -0
  1300. executorch/backends/qualcomm/debugger/format_outputs.py +221 -0
  1301. executorch/backends/qualcomm/debugger/metrics_evaluator.py +90 -0
  1302. executorch/backends/qualcomm/debugger/qnn_intermediate_debugger.py +328 -0
  1303. executorch/backends/qualcomm/debugger/utils.py +435 -0
  1304. executorch/backends/qualcomm/partition/common_defs.py +44 -0
  1305. executorch/backends/qualcomm/partition/qnn_partitioner.py +225 -0
  1306. executorch/backends/qualcomm/partition/utils.py +75 -0
  1307. executorch/backends/qualcomm/qnn_preprocess.py +218 -0
  1308. executorch/backends/qualcomm/quantizer/annotators.py +1501 -0
  1309. executorch/backends/qualcomm/quantizer/custom_annotation.py +499 -0
  1310. executorch/backends/qualcomm/quantizer/observers/concat_observer.py +74 -0
  1311. executorch/backends/qualcomm/quantizer/observers/per_block_param_observer.py +145 -0
  1312. executorch/backends/qualcomm/quantizer/observers/per_channel_param_observer.py +113 -0
  1313. executorch/backends/qualcomm/quantizer/qconfig.py +802 -0
  1314. executorch/backends/qualcomm/quantizer/quant_recipe.py +402 -0
  1315. executorch/backends/qualcomm/quantizer/quantizer.py +507 -0
  1316. executorch/backends/qualcomm/recipes/__init__.py +16 -0
  1317. executorch/backends/qualcomm/recipes/qnn_recipe_provider.py +179 -0
  1318. executorch/backends/qualcomm/recipes/qnn_recipe_types.py +28 -0
  1319. executorch/backends/qualcomm/scripts/download_qnn_sdk.py +642 -0
  1320. executorch/backends/qualcomm/serialization/qc_compiler_spec.fbs +299 -0
  1321. executorch/backends/qualcomm/serialization/qc_schema.py +240 -0
  1322. executorch/backends/qualcomm/serialization/qc_schema_serialize.py +59 -0
  1323. executorch/backends/qualcomm/tests/models.py +2393 -0
  1324. executorch/backends/qualcomm/tests/test_passes.py +154 -0
  1325. executorch/backends/qualcomm/tests/test_qnn_delegate.py +8454 -0
  1326. executorch/backends/qualcomm/tests/tester.py +113 -0
  1327. executorch/backends/qualcomm/tests/utils.py +770 -0
  1328. executorch/backends/qualcomm/utils/constants.py +49 -0
  1329. executorch/backends/qualcomm/utils/qnn_manager_lifecycle.py +88 -0
  1330. executorch/backends/qualcomm/utils/utils.py +1302 -0
  1331. executorch/backends/samsung/_passes/annotate_qparams.py +201 -0
  1332. executorch/backends/samsung/_passes/annotate_scalar_parameters.py +65 -0
  1333. executorch/backends/samsung/_passes/conv1d_to_conv2d.py +97 -0
  1334. executorch/backends/samsung/_passes/customized_constant_prop.py +40 -0
  1335. executorch/backends/samsung/_passes/fold_qdq.py +36 -0
  1336. executorch/backends/samsung/_passes/fuse_conv_act.py +77 -0
  1337. executorch/backends/samsung/_passes/insert_qdq.py +164 -0
  1338. executorch/backends/samsung/_passes/remove_useless_ops.py +87 -0
  1339. executorch/backends/samsung/_passes/replace_scalar_ops.py +46 -0
  1340. executorch/backends/samsung/_passes/utils.py +15 -0
  1341. executorch/backends/samsung/builders/__init__.py +97 -0
  1342. executorch/backends/samsung/builders/node_visitor.py +116 -0
  1343. executorch/backends/samsung/builders/op_add.py +41 -0
  1344. executorch/backends/samsung/builders/op_avg_pool2d.py +70 -0
  1345. executorch/backends/samsung/builders/op_batch_norm.py +56 -0
  1346. executorch/backends/samsung/builders/op_bmm.py +43 -0
  1347. executorch/backends/samsung/builders/op_cat.py +50 -0
  1348. executorch/backends/samsung/builders/op_clamp.py +47 -0
  1349. executorch/backends/samsung/builders/op_constant_pad_nd.py +56 -0
  1350. executorch/backends/samsung/builders/op_conv2d.py +89 -0
  1351. executorch/backends/samsung/builders/op_dequantize.py +19 -0
  1352. executorch/backends/samsung/builders/op_div.py +42 -0
  1353. executorch/backends/samsung/builders/op_embedding.py +41 -0
  1354. executorch/backends/samsung/builders/op_expand_copy.py +77 -0
  1355. executorch/backends/samsung/builders/op_gelu.py +40 -0
  1356. executorch/backends/samsung/builders/op_getitem.py +32 -0
  1357. executorch/backends/samsung/builders/op_hardsigmoid.py +35 -0
  1358. executorch/backends/samsung/builders/op_hardswish.py +35 -0
  1359. executorch/backends/samsung/builders/op_hardtanh.py +42 -0
  1360. executorch/backends/samsung/builders/op_layer_norm.py +53 -0
  1361. executorch/backends/samsung/builders/op_leaky_relu.py +58 -0
  1362. executorch/backends/samsung/builders/op_linear.py +51 -0
  1363. executorch/backends/samsung/builders/op_log_softmax.py +39 -0
  1364. executorch/backends/samsung/builders/op_max_pool2d.py +96 -0
  1365. executorch/backends/samsung/builders/op_maximum.py +37 -0
  1366. executorch/backends/samsung/builders/op_mean_dim.py +49 -0
  1367. executorch/backends/samsung/builders/op_minimum.py +39 -0
  1368. executorch/backends/samsung/builders/op_mul.py +43 -0
  1369. executorch/backends/samsung/builders/op_permute.py +37 -0
  1370. executorch/backends/samsung/builders/op_pixel_shuffle.py +38 -0
  1371. executorch/backends/samsung/builders/op_quantize.py +60 -0
  1372. executorch/backends/samsung/builders/op_relu.py +37 -0
  1373. executorch/backends/samsung/builders/op_reshape.py +34 -0
  1374. executorch/backends/samsung/builders/op_rsqrt.py +35 -0
  1375. executorch/backends/samsung/builders/op_select.py +52 -0
  1376. executorch/backends/samsung/builders/op_slice_copy.py +59 -0
  1377. executorch/backends/samsung/builders/op_softmax.py +39 -0
  1378. executorch/backends/samsung/builders/op_sqrt.py +35 -0
  1379. executorch/backends/samsung/builders/op_squeeze.py +37 -0
  1380. executorch/backends/samsung/builders/op_sub.py +39 -0
  1381. executorch/backends/samsung/builders/op_to_copy.py +44 -0
  1382. executorch/backends/samsung/builders/op_unsqueeze.py +35 -0
  1383. executorch/backends/samsung/builders/op_upsample_bilinear2d.py +53 -0
  1384. executorch/backends/samsung/builders/op_upsample_nearest2d.py +52 -0
  1385. executorch/backends/samsung/builders/utils.py +64 -0
  1386. executorch/backends/samsung/enn_preprocess.py +110 -0
  1387. executorch/backends/samsung/partition/enn_partitioner.py +134 -0
  1388. executorch/backends/samsung/quantizer/__init__.py +10 -0
  1389. executorch/backends/samsung/quantizer/annotator.py +871 -0
  1390. executorch/backends/samsung/quantizer/qconfig.py +174 -0
  1391. executorch/backends/samsung/quantizer/quantizer.py +65 -0
  1392. executorch/backends/samsung/serialization/compile_options.py +77 -0
  1393. executorch/backends/samsung/serialization/compile_options_def.fbs +23 -0
  1394. executorch/backends/samsung/serialization/enn_graph_schema.py +141 -0
  1395. executorch/backends/samsung/test/models/test_deeplab_v3.py +28 -0
  1396. executorch/backends/samsung/test/models/test_edsr.py +30 -0
  1397. executorch/backends/samsung/test/models/test_inception_v3.py +30 -0
  1398. executorch/backends/samsung/test/models/test_inception_v4.py +30 -0
  1399. executorch/backends/samsung/test/models/test_mobilenet_v2.py +28 -0
  1400. executorch/backends/samsung/test/models/test_mobilenet_v3.py +33 -0
  1401. executorch/backends/samsung/test/models/test_resnet18.py +30 -0
  1402. executorch/backends/samsung/test/models/test_resnet50.py +30 -0
  1403. executorch/backends/samsung/test/models/test_torchvision_vit.py +32 -0
  1404. executorch/backends/samsung/test/models/test_wav2letter.py +28 -0
  1405. executorch/backends/samsung/test/ops/test_add.py +63 -0
  1406. executorch/backends/samsung/test/ops/test_avg_pool2d.py +66 -0
  1407. executorch/backends/samsung/test/ops/test_batch_norm.py +52 -0
  1408. executorch/backends/samsung/test/ops/test_bmm.py +50 -0
  1409. executorch/backends/samsung/test/ops/test_cat.py +49 -0
  1410. executorch/backends/samsung/test/ops/test_clamp.py +50 -0
  1411. executorch/backends/samsung/test/ops/test_constant_pad_nd.py +46 -0
  1412. executorch/backends/samsung/test/ops/test_conv2d.py +95 -0
  1413. executorch/backends/samsung/test/ops/test_div.py +50 -0
  1414. executorch/backends/samsung/test/ops/test_embedding.py +46 -0
  1415. executorch/backends/samsung/test/ops/test_expand_copy.py +46 -0
  1416. executorch/backends/samsung/test/ops/test_gelu.py +65 -0
  1417. executorch/backends/samsung/test/ops/test_leaky_relu.py +47 -0
  1418. executorch/backends/samsung/test/ops/test_linear.py +48 -0
  1419. executorch/backends/samsung/test/ops/test_log_softmax.py +46 -0
  1420. executorch/backends/samsung/test/ops/test_max_pool2d.py +72 -0
  1421. executorch/backends/samsung/test/ops/test_mean_dim.py +51 -0
  1422. executorch/backends/samsung/test/ops/test_minimum.py +45 -0
  1423. executorch/backends/samsung/test/ops/test_mul.py +63 -0
  1424. executorch/backends/samsung/test/ops/test_permute.py +55 -0
  1425. executorch/backends/samsung/test/ops/test_pixel_shuffle.py +49 -0
  1426. executorch/backends/samsung/test/ops/test_relu.py +65 -0
  1427. executorch/backends/samsung/test/ops/test_reshape.py +46 -0
  1428. executorch/backends/samsung/test/ops/test_rsqrt.py +46 -0
  1429. executorch/backends/samsung/test/ops/test_select.py +52 -0
  1430. executorch/backends/samsung/test/ops/test_slice_copy.py +46 -0
  1431. executorch/backends/samsung/test/ops/test_softmax.py +46 -0
  1432. executorch/backends/samsung/test/ops/test_sqrt.py +46 -0
  1433. executorch/backends/samsung/test/ops/test_squeeze.py +47 -0
  1434. executorch/backends/samsung/test/ops/test_sub.py +63 -0
  1435. executorch/backends/samsung/test/ops/test_to_copy.py +49 -0
  1436. executorch/backends/samsung/test/ops/test_unsqueeze.py +47 -0
  1437. executorch/backends/samsung/test/ops/test_upsample_bilinear2d.py +54 -0
  1438. executorch/backends/samsung/test/ops/test_upsample_nearest2d.py +53 -0
  1439. executorch/backends/samsung/test/tester/__init__.py +11 -0
  1440. executorch/backends/samsung/test/tester/samsung_tester.py +125 -0
  1441. executorch/backends/samsung/test/utils/__init__.py +11 -0
  1442. executorch/backends/samsung/test/utils/runtime_executor.py +186 -0
  1443. executorch/backends/samsung/utils/constants.py +45 -0
  1444. executorch/backends/samsung/utils/export_utils.py +93 -0
  1445. executorch/backends/samsung/utils/utils.py +125 -0
  1446. executorch/backends/test/facto/__init__.py +0 -0
  1447. executorch/backends/test/facto/facto_specs.py +59 -0
  1448. executorch/backends/test/facto/test_facto.py +301 -0
  1449. executorch/backends/test/harness/__init__.py +3 -0
  1450. executorch/backends/test/harness/error_statistics.py +99 -0
  1451. executorch/backends/test/harness/stages/__init__.py +23 -0
  1452. executorch/backends/test/harness/stages/export.py +32 -0
  1453. executorch/backends/test/harness/stages/partition.py +26 -0
  1454. executorch/backends/test/harness/stages/quantize.py +128 -0
  1455. executorch/backends/test/harness/stages/run_passes.py +66 -0
  1456. executorch/backends/test/harness/stages/serialize.py +59 -0
  1457. executorch/backends/test/harness/stages/stage.py +91 -0
  1458. executorch/backends/test/harness/stages/to_edge.py +27 -0
  1459. executorch/backends/test/harness/stages/to_edge_transform_and_lower.py +50 -0
  1460. executorch/backends/test/harness/stages/to_executorch.py +54 -0
  1461. executorch/backends/test/harness/tester.py +478 -0
  1462. executorch/backends/test/harness/tests/test_error_statistics.py +67 -0
  1463. executorch/backends/test/suite/__init__.py +74 -0
  1464. executorch/backends/test/suite/conftest.py +182 -0
  1465. executorch/backends/test/suite/context.py +34 -0
  1466. executorch/backends/test/suite/discovery.py +100 -0
  1467. executorch/backends/test/suite/flow.py +150 -0
  1468. executorch/backends/test/suite/flows/__init__.py +7 -0
  1469. executorch/backends/test/suite/flows/arm.py +92 -0
  1470. executorch/backends/test/suite/flows/coreml.py +31 -0
  1471. executorch/backends/test/suite/flows/portable.py +19 -0
  1472. executorch/backends/test/suite/flows/qualcomm.py +61 -0
  1473. executorch/backends/test/suite/flows/vulkan.py +44 -0
  1474. executorch/backends/test/suite/flows/xnnpack.py +79 -0
  1475. executorch/backends/test/suite/generate_markdown_summary.py +249 -0
  1476. executorch/backends/test/suite/generate_markdown_summary_json.py +229 -0
  1477. executorch/backends/test/suite/models/__init__.py +7 -0
  1478. executorch/backends/test/suite/models/test_torchaudio.py +110 -0
  1479. executorch/backends/test/suite/models/test_torchvision.py +194 -0
  1480. executorch/backends/test/suite/operators/__init__.py +108 -0
  1481. executorch/backends/test/suite/operators/test_abs.py +60 -0
  1482. executorch/backends/test/suite/operators/test_adaptive_avgpool1d.py +102 -0
  1483. executorch/backends/test/suite/operators/test_adaptive_avgpool2d.py +112 -0
  1484. executorch/backends/test/suite/operators/test_adaptive_avgpool3d.py +112 -0
  1485. executorch/backends/test/suite/operators/test_adaptive_maxpool1d.py +125 -0
  1486. executorch/backends/test/suite/operators/test_adaptive_maxpool2d.py +135 -0
  1487. executorch/backends/test/suite/operators/test_adaptive_maxpool3d.py +135 -0
  1488. executorch/backends/test/suite/operators/test_add.py +76 -0
  1489. executorch/backends/test/suite/operators/test_amax.py +257 -0
  1490. executorch/backends/test/suite/operators/test_amin.py +259 -0
  1491. executorch/backends/test/suite/operators/test_argmax.py +201 -0
  1492. executorch/backends/test/suite/operators/test_argmin.py +201 -0
  1493. executorch/backends/test/suite/operators/test_avgpool1d.py +155 -0
  1494. executorch/backends/test/suite/operators/test_avgpool2d.py +168 -0
  1495. executorch/backends/test/suite/operators/test_avgpool3d.py +163 -0
  1496. executorch/backends/test/suite/operators/test_cat.py +176 -0
  1497. executorch/backends/test/suite/operators/test_ceil.py +60 -0
  1498. executorch/backends/test/suite/operators/test_clamp.py +86 -0
  1499. executorch/backends/test/suite/operators/test_conv1d.py +144 -0
  1500. executorch/backends/test/suite/operators/test_conv2d.py +167 -0
  1501. executorch/backends/test/suite/operators/test_conv3d.py +162 -0
  1502. executorch/backends/test/suite/operators/test_convtranspose1d.py +144 -0
  1503. executorch/backends/test/suite/operators/test_convtranspose2d.py +171 -0
  1504. executorch/backends/test/suite/operators/test_convtranspose3d.py +166 -0
  1505. executorch/backends/test/suite/operators/test_div.py +110 -0
  1506. executorch/backends/test/suite/operators/test_elu.py +49 -0
  1507. executorch/backends/test/suite/operators/test_embedding.py +89 -0
  1508. executorch/backends/test/suite/operators/test_embedding_bag.py +118 -0
  1509. executorch/backends/test/suite/operators/test_exp.py +65 -0
  1510. executorch/backends/test/suite/operators/test_expand.py +122 -0
  1511. executorch/backends/test/suite/operators/test_floor.py +57 -0
  1512. executorch/backends/test/suite/operators/test_floor_divide.py +216 -0
  1513. executorch/backends/test/suite/operators/test_gelu.py +52 -0
  1514. executorch/backends/test/suite/operators/test_glu.py +52 -0
  1515. executorch/backends/test/suite/operators/test_hardsigmoid.py +50 -0
  1516. executorch/backends/test/suite/operators/test_hardswish.py +50 -0
  1517. executorch/backends/test/suite/operators/test_hardtanh.py +57 -0
  1518. executorch/backends/test/suite/operators/test_index_put.py +455 -0
  1519. executorch/backends/test/suite/operators/test_index_select.py +128 -0
  1520. executorch/backends/test/suite/operators/test_leaky_relu.py +56 -0
  1521. executorch/backends/test/suite/operators/test_linear.py +124 -0
  1522. executorch/backends/test/suite/operators/test_log.py +80 -0
  1523. executorch/backends/test/suite/operators/test_log10.py +60 -0
  1524. executorch/backends/test/suite/operators/test_log1p.py +60 -0
  1525. executorch/backends/test/suite/operators/test_log2.py +60 -0
  1526. executorch/backends/test/suite/operators/test_logsigmoid.py +40 -0
  1527. executorch/backends/test/suite/operators/test_lstm.py +247 -0
  1528. executorch/backends/test/suite/operators/test_masked_fill.py +101 -0
  1529. executorch/backends/test/suite/operators/test_maxpool1d.py +185 -0
  1530. executorch/backends/test/suite/operators/test_maxpool2d.py +191 -0
  1531. executorch/backends/test/suite/operators/test_maxpool3d.py +189 -0
  1532. executorch/backends/test/suite/operators/test_mean.py +305 -0
  1533. executorch/backends/test/suite/operators/test_median.py +188 -0
  1534. executorch/backends/test/suite/operators/test_mul.py +66 -0
  1535. executorch/backends/test/suite/operators/test_neg.py +70 -0
  1536. executorch/backends/test/suite/operators/test_permute.py +109 -0
  1537. executorch/backends/test/suite/operators/test_pow.py +149 -0
  1538. executorch/backends/test/suite/operators/test_prelu.py +55 -0
  1539. executorch/backends/test/suite/operators/test_relu.py +45 -0
  1540. executorch/backends/test/suite/operators/test_reshape.py +82 -0
  1541. executorch/backends/test/suite/operators/test_round.py +129 -0
  1542. executorch/backends/test/suite/operators/test_rsqrt.py +76 -0
  1543. executorch/backends/test/suite/operators/test_select.py +89 -0
  1544. executorch/backends/test/suite/operators/test_sigmoid.py +40 -0
  1545. executorch/backends/test/suite/operators/test_silu.py +50 -0
  1546. executorch/backends/test/suite/operators/test_slice.py +146 -0
  1547. executorch/backends/test/suite/operators/test_split.py +119 -0
  1548. executorch/backends/test/suite/operators/test_sqrt.py +71 -0
  1549. executorch/backends/test/suite/operators/test_square.py +67 -0
  1550. executorch/backends/test/suite/operators/test_squeeze.py +86 -0
  1551. executorch/backends/test/suite/operators/test_stack.py +114 -0
  1552. executorch/backends/test/suite/operators/test_sub.py +84 -0
  1553. executorch/backends/test/suite/operators/test_tanh.py +40 -0
  1554. executorch/backends/test/suite/operators/test_threshold.py +71 -0
  1555. executorch/backends/test/suite/operators/test_transpose.py +143 -0
  1556. executorch/backends/test/suite/operators/test_trunc.py +73 -0
  1557. executorch/backends/test/suite/operators/test_unsqueeze.py +110 -0
  1558. executorch/backends/test/suite/operators/test_upsample_bilinear2d.py +273 -0
  1559. executorch/backends/test/suite/operators/test_upsample_nearest2d.py +137 -0
  1560. executorch/backends/test/suite/operators/test_view.py +82 -0
  1561. executorch/backends/test/suite/reporting.py +421 -0
  1562. executorch/backends/test/suite/runner.py +328 -0
  1563. executorch/backends/test/suite/tests/__init__.py +0 -0
  1564. executorch/backends/test/suite/tests/test_reporting.py +144 -0
  1565. executorch/backends/transforms/__init__.py +12 -0
  1566. executorch/backends/transforms/addmm_mm_to_linear.py +170 -0
  1567. executorch/backends/transforms/convert_dtype_pass.py +103 -0
  1568. executorch/backends/transforms/decompose_sdpa.py +120 -0
  1569. executorch/backends/transforms/duplicate_dynamic_quant_chain.py +195 -0
  1570. executorch/backends/transforms/fuse_batch_norm_with_conv.py +161 -0
  1571. executorch/backends/transforms/fuse_conv_with_clamp.py +77 -0
  1572. executorch/backends/transforms/fuse_view_copy.py +116 -0
  1573. executorch/backends/transforms/mean_to_sum_div.py +41 -0
  1574. executorch/backends/transforms/rank_0_to_rank_1.py +18 -0
  1575. executorch/backends/transforms/remove_clone_ops.py +109 -0
  1576. executorch/backends/transforms/remove_getitem_op.py +109 -0
  1577. executorch/backends/transforms/replace_scalar_with_tensor.py +93 -0
  1578. executorch/backends/transforms/test/test_create_delete_constant_placeholder.py +122 -0
  1579. executorch/backends/transforms/test/test_create_mutable_buffer.py +503 -0
  1580. executorch/backends/transforms/test/test_duplicate_dynamic_quant_chain.py +184 -0
  1581. executorch/backends/transforms/test/test_rank_0_to_rank_1.py +32 -0
  1582. executorch/backends/transforms/test/test_remove_clone_ops.py +197 -0
  1583. executorch/backends/transforms/utils.py +377 -0
  1584. executorch/backends/transforms/view_copy_to_squeeze_unsqueeze.py +135 -0
  1585. executorch/backends/vulkan/__init__.py +14 -0
  1586. executorch/backends/vulkan/_passes/__init__.py +37 -0
  1587. executorch/backends/vulkan/_passes/fold_qdq.py +40 -0
  1588. executorch/backends/vulkan/_passes/fuse_patterns.py +34 -0
  1589. executorch/backends/vulkan/_passes/fuse_quantized_ops.py +236 -0
  1590. executorch/backends/vulkan/_passes/insert_prepack_nodes.py +83 -0
  1591. executorch/backends/vulkan/_passes/remove_asserts.py +52 -0
  1592. executorch/backends/vulkan/_passes/remove_redundant_ops.py +72 -0
  1593. executorch/backends/vulkan/_passes/squeeze_unsqueeze_inputs.py +93 -0
  1594. executorch/backends/vulkan/_passes/tag_memory_meta_pass.py +485 -0
  1595. executorch/backends/vulkan/custom_ops_lib.py +597 -0
  1596. executorch/backends/vulkan/op_registry.py +885 -0
  1597. executorch/backends/vulkan/partitioner/vulkan_partitioner.py +401 -0
  1598. executorch/backends/vulkan/patterns/__init__.py +139 -0
  1599. executorch/backends/vulkan/patterns/pattern_registry.py +116 -0
  1600. executorch/backends/vulkan/patterns/quantized_binary.py +161 -0
  1601. executorch/backends/vulkan/patterns/quantized_convolution.py +260 -0
  1602. executorch/backends/vulkan/patterns/quantized_linear.py +500 -0
  1603. executorch/backends/vulkan/patterns/rope.py +173 -0
  1604. executorch/backends/vulkan/patterns/sdpa.py +155 -0
  1605. executorch/backends/vulkan/patterns/select_as_symint.py +104 -0
  1606. executorch/backends/vulkan/quantizer/vulkan_quantizer.py +183 -0
  1607. executorch/backends/vulkan/quantizer/vulkan_quantizer_utils.py +223 -0
  1608. executorch/backends/vulkan/runtime/gen_vulkan_spv.py +1517 -0
  1609. executorch/backends/vulkan/runtime/graph/ops/glsl/addmm_naive_buffer.yaml +19 -0
  1610. executorch/backends/vulkan/runtime/graph/ops/glsl/addmm_naive_texture3d.yaml +24 -0
  1611. executorch/backends/vulkan/runtime/graph/ops/glsl/addmm_optimized.yaml +43 -0
  1612. executorch/backends/vulkan/runtime/graph/ops/glsl/arange.yaml +19 -0
  1613. executorch/backends/vulkan/runtime/graph/ops/glsl/avg_pool2d.yaml +18 -0
  1614. executorch/backends/vulkan/runtime/graph/ops/glsl/batchnorm.yaml +11 -0
  1615. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_op.yaml +118 -0
  1616. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_q8ta_q8ta_q8to.yaml +19 -0
  1617. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_buffer.yaml +20 -0
  1618. executorch/backends/vulkan/runtime/graph/ops/glsl/binary_scalar_texture.yaml +20 -0
  1619. executorch/backends/vulkan/runtime/graph/ops/glsl/bitw8_image_to_nchw_nobitw8buffer.yaml +22 -0
  1620. executorch/backends/vulkan/runtime/graph/ops/glsl/buffer_to_buffer.yaml +20 -0
  1621. executorch/backends/vulkan/runtime/graph/ops/glsl/buffer_to_nchw.yaml +25 -0
  1622. executorch/backends/vulkan/runtime/graph/ops/glsl/choose_qparams_per_row.yaml +19 -0
  1623. executorch/backends/vulkan/runtime/graph/ops/glsl/clone.yaml +13 -0
  1624. executorch/backends/vulkan/runtime/graph/ops/glsl/col2im.yaml +19 -0
  1625. executorch/backends/vulkan/runtime/graph/ops/glsl/concat_buffer.yaml +15 -0
  1626. executorch/backends/vulkan/runtime/graph/ops/glsl/concat_texture.yaml +15 -0
  1627. executorch/backends/vulkan/runtime/graph/ops/glsl/conv1d.yaml +19 -0
  1628. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d.yaml +19 -0
  1629. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw.yaml +19 -0
  1630. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.yaml +27 -0
  1631. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_prepack_weights.yaml +15 -0
  1632. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_q8ta_q8csw_q8to.yaml +20 -0
  1633. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_sned_output_tile.yaml +25 -0
  1634. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_prepack_weights.yaml +15 -0
  1635. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.yaml +21 -0
  1636. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_q8ta_q8csw_q8to_tiled.yaml +20 -0
  1637. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_pw_s1p0.yaml +19 -0
  1638. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8csw_linear_tiled.yaml +22 -0
  1639. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_linear_tiled.yaml +22 -0
  1640. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to.yaml +21 -0
  1641. executorch/backends/vulkan/runtime/graph/ops/glsl/conv2d_q8ta_q8csw_q8to_linear_tiled.yaml +21 -0
  1642. executorch/backends/vulkan/runtime/graph/ops/glsl/conv_transpose2d.yaml +19 -0
  1643. executorch/backends/vulkan/runtime/graph/ops/glsl/conv_transpose2d_prepack_weights.yaml +16 -0
  1644. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_buffer.yaml +16 -0
  1645. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_legacy.yaml +12 -0
  1646. executorch/backends/vulkan/runtime/graph/ops/glsl/embedding_texture.yaml +15 -0
  1647. executorch/backends/vulkan/runtime/graph/ops/glsl/expand_buffer.yaml +11 -0
  1648. executorch/backends/vulkan/runtime/graph/ops/glsl/flip.yaml +14 -0
  1649. executorch/backends/vulkan/runtime/graph/ops/glsl/full.yaml +20 -0
  1650. executorch/backends/vulkan/runtime/graph/ops/glsl/gather_buffer.yaml +18 -0
  1651. executorch/backends/vulkan/runtime/graph/ops/glsl/gather_texture.yaml +17 -0
  1652. executorch/backends/vulkan/runtime/graph/ops/glsl/grid_priors.yaml +12 -0
  1653. executorch/backends/vulkan/runtime/graph/ops/glsl/group_norm_reduce_texture.yaml +15 -0
  1654. executorch/backends/vulkan/runtime/graph/ops/glsl/group_norm_texture.yaml +15 -0
  1655. executorch/backends/vulkan/runtime/graph/ops/glsl/im2col.yaml +18 -0
  1656. executorch/backends/vulkan/runtime/graph/ops/glsl/im2col_packed_int8.yaml +15 -0
  1657. executorch/backends/vulkan/runtime/graph/ops/glsl/image_to_nchw.yaml +38 -0
  1658. executorch/backends/vulkan/runtime/graph/ops/glsl/index_select.yaml +13 -0
  1659. executorch/backends/vulkan/runtime/graph/ops/glsl/index_select_channel.yaml +13 -0
  1660. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_dq8ca_q4gsw_tiled.yaml +29 -0
  1661. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_coop.yaml +43 -0
  1662. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q4gsw_tiled.yaml +28 -0
  1663. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q8csw_tiled.yaml +28 -0
  1664. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_q8ta_q8csw_tiled.yaml +24 -0
  1665. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw.yaml +23 -0
  1666. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_coop.yaml +38 -0
  1667. executorch/backends/vulkan/runtime/graph/ops/glsl/linear_qcsnw_tiled.yaml +54 -0
  1668. executorch/backends/vulkan/runtime/graph/ops/glsl/max_pool2d.yaml +17 -0
  1669. executorch/backends/vulkan/runtime/graph/ops/glsl/native_layer_norm.yaml +16 -0
  1670. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_bitw8_image_nobitw8buffer.yaml +22 -0
  1671. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_buffer.yaml +25 -0
  1672. executorch/backends/vulkan/runtime/graph/ops/glsl/nchw_to_image.yaml +37 -0
  1673. executorch/backends/vulkan/runtime/graph/ops/glsl/no_op.yaml +24 -0
  1674. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_int4_linear_weight_transposed_interleaved.yaml +19 -0
  1675. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q4_linear_weight.yaml +14 -0
  1676. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_conv2d_dw_weights.yaml +15 -0
  1677. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_conv2d_weights.yaml +15 -0
  1678. executorch/backends/vulkan/runtime/graph/ops/glsl/pack_q8_linear_weight.yaml +14 -0
  1679. executorch/backends/vulkan/runtime/graph/ops/glsl/pad_channel.yaml +14 -0
  1680. executorch/backends/vulkan/runtime/graph/ops/glsl/pad_height_width.yaml +14 -0
  1681. executorch/backends/vulkan/runtime/graph/ops/glsl/permute_buffer.yaml +11 -0
  1682. executorch/backends/vulkan/runtime/graph/ops/glsl/permute_texture.yaml +11 -0
  1683. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4h4w.yaml +25 -0
  1684. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4h4w_with_group_sums.yaml +30 -0
  1685. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_4w4c.yaml +23 -0
  1686. executorch/backends/vulkan/runtime/graph/ops/glsl/quantize_and_pack_im2col.yaml +18 -0
  1687. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce.yaml +29 -0
  1688. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce2d.yaml +29 -0
  1689. executorch/backends/vulkan/runtime/graph/ops/glsl/reduce_per_row_buffer.yaml +42 -0
  1690. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat.yaml +14 -0
  1691. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_channel.yaml +12 -0
  1692. executorch/backends/vulkan/runtime/graph/ops/glsl/repeat_interleave.yaml +12 -0
  1693. executorch/backends/vulkan/runtime/graph/ops/glsl/rotary_embedding.yaml +13 -0
  1694. executorch/backends/vulkan/runtime/graph/ops/glsl/scalar_tensor.yaml +27 -0
  1695. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_attn_weights_softmax.yaml +19 -0
  1696. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_attn_weights_coop.yaml +25 -0
  1697. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_attn_weights_tiled.yaml +26 -0
  1698. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_out_coop.yaml +25 -0
  1699. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_compute_out_tiled.yaml +26 -0
  1700. executorch/backends/vulkan/runtime/graph/ops/glsl/sdpa_kv_cache_update.yaml +23 -0
  1701. executorch/backends/vulkan/runtime/graph/ops/glsl/set_zero.yaml +8 -0
  1702. executorch/backends/vulkan/runtime/graph/ops/glsl/softmax.yaml +21 -0
  1703. executorch/backends/vulkan/runtime/graph/ops/glsl/split_buffer.yaml +18 -0
  1704. executorch/backends/vulkan/runtime/graph/ops/glsl/split_texture.yaml +17 -0
  1705. executorch/backends/vulkan/runtime/graph/ops/glsl/tan.yaml +13 -0
  1706. executorch/backends/vulkan/runtime/graph/ops/glsl/transfer_buffer.yaml +22 -0
  1707. executorch/backends/vulkan/runtime/graph/ops/glsl/transfer_texture.yaml +22 -0
  1708. executorch/backends/vulkan/runtime/graph/ops/glsl/unary_op.yaml +48 -0
  1709. executorch/backends/vulkan/runtime/graph/ops/glsl/unpack_4w4c_and_dequantize.yaml +23 -0
  1710. executorch/backends/vulkan/runtime/graph/ops/glsl/update_concat_offset.yaml +13 -0
  1711. executorch/backends/vulkan/runtime/graph/ops/glsl/upsample_2d.yaml +19 -0
  1712. executorch/backends/vulkan/runtime/graph/ops/glsl/var_buffer.yaml +16 -0
  1713. executorch/backends/vulkan/runtime/graph/ops/glsl/var_texture3d.yaml +16 -0
  1714. executorch/backends/vulkan/runtime/graph/ops/glsl/view.yaml +13 -0
  1715. executorch/backends/vulkan/runtime/graph/ops/glsl/view_buffer.yaml +20 -0
  1716. executorch/backends/vulkan/runtime/graph/ops/glsl/view_convert_buffer.yaml +22 -0
  1717. executorch/backends/vulkan/runtime/graph/ops/glsl/where.yaml +12 -0
  1718. executorch/backends/vulkan/serialization/schema.fbs +158 -0
  1719. executorch/backends/vulkan/serialization/vulkan_graph_builder.py +505 -0
  1720. executorch/backends/vulkan/serialization/vulkan_graph_schema.py +161 -0
  1721. executorch/backends/vulkan/serialization/vulkan_graph_serialize.py +274 -0
  1722. executorch/backends/vulkan/test/custom_ops/glsl/add.yaml +29 -0
  1723. executorch/backends/vulkan/test/glsl/all_shaders.yaml +68 -0
  1724. executorch/backends/vulkan/test/glsl/dynamic_dispatch_test.yaml +7 -0
  1725. executorch/backends/vulkan/test/glsl/reference_matmul_common.yaml +9 -0
  1726. executorch/backends/vulkan/test/op_tests/cases.py +2026 -0
  1727. executorch/backends/vulkan/test/op_tests/generate_op_benchmarks.py +88 -0
  1728. executorch/backends/vulkan/test/op_tests/generate_op_correctness_tests.py +90 -0
  1729. executorch/backends/vulkan/test/op_tests/utils/aten_types.py +31 -0
  1730. executorch/backends/vulkan/test/op_tests/utils/gen_benchmark_vk.py +350 -0
  1731. executorch/backends/vulkan/test/op_tests/utils/gen_computegraph.py +788 -0
  1732. executorch/backends/vulkan/test/op_tests/utils/gen_correctness_base.py +421 -0
  1733. executorch/backends/vulkan/test/op_tests/utils/gen_correctness_vk.py +164 -0
  1734. executorch/backends/vulkan/test/op_tests/utils/test_suite.py +53 -0
  1735. executorch/backends/vulkan/test/test_serialization.py +135 -0
  1736. executorch/backends/vulkan/test/test_vulkan_delegate.py +2696 -0
  1737. executorch/backends/vulkan/test/test_vulkan_delegate_header.py +104 -0
  1738. executorch/backends/vulkan/test/test_vulkan_passes.py +193 -0
  1739. executorch/backends/vulkan/test/tester.py +91 -0
  1740. executorch/backends/vulkan/test/utils.py +1111 -0
  1741. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/apiconventions.py +21 -0
  1742. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/base_generator.py +937 -0
  1743. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/cgenerator.py +549 -0
  1744. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/generator.py +1424 -0
  1745. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/parse_dependency.py +404 -0
  1746. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/reg.py +1885 -0
  1747. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/spec_tools/conventions.py +562 -0
  1748. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/spec_tools/util.py +58 -0
  1749. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/stripAPI.py +42 -0
  1750. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/vkconventions.py +314 -0
  1751. executorch/backends/vulkan/third-party/Vulkan-Headers/registry/vulkan_object.py +484 -0
  1752. executorch/backends/vulkan/third-party/VulkanMemoryAllocator/tools/GpuMemDumpVis/GpuMemDumpVis.py +333 -0
  1753. executorch/backends/vulkan/third-party/volk/generate.py +212 -0
  1754. executorch/backends/vulkan/tools/gpuinfo/glsl/buf_bandwidth.yaml +18 -0
  1755. executorch/backends/vulkan/tools/gpuinfo/glsl/buf_cacheline_size.yaml +12 -0
  1756. executorch/backends/vulkan/tools/gpuinfo/glsl/reg_count.yaml +16 -0
  1757. executorch/backends/vulkan/tools/gpuinfo/glsl/tex_bandwidth.yaml +15 -0
  1758. executorch/backends/vulkan/tools/gpuinfo/glsl/tex_cacheline_concurr.yaml +14 -0
  1759. executorch/backends/vulkan/tools/gpuinfo/glsl/warp_size.yaml +16 -0
  1760. executorch/backends/vulkan/utils.py +1474 -0
  1761. executorch/backends/vulkan/vulkan_preprocess.py +249 -0
  1762. executorch/backends/xnnpack/__init__.py +34 -0
  1763. executorch/backends/xnnpack/_passes/__init__.py +109 -0
  1764. executorch/backends/xnnpack/_passes/channels_last_tagged_reshape_pass.py +561 -0
  1765. executorch/backends/xnnpack/_passes/conv1d_unsqueeze_pass.py +203 -0
  1766. executorch/backends/xnnpack/_passes/convert_to_linear.py +198 -0
  1767. executorch/backends/xnnpack/_passes/convert_to_sdpa.py +114 -0
  1768. executorch/backends/xnnpack/_passes/convert_to_upsample_bilinear2d.py +65 -0
  1769. executorch/backends/xnnpack/_passes/decompose_cat.py +105 -0
  1770. executorch/backends/xnnpack/_passes/fuse_activation_pass.py +87 -0
  1771. executorch/backends/xnnpack/_passes/fuse_batch_norm.py +261 -0
  1772. executorch/backends/xnnpack/_passes/prelu_reshape_pass.py +75 -0
  1773. executorch/backends/xnnpack/_passes/propagate_custom_meta_pass.py +44 -0
  1774. executorch/backends/xnnpack/_passes/remove_redundant_copy_pass.py +158 -0
  1775. executorch/backends/xnnpack/_passes/xnnpack_pass.py +22 -0
  1776. executorch/backends/xnnpack/operators/__init__.py +61 -0
  1777. executorch/backends/xnnpack/operators/node_visitor.py +758 -0
  1778. executorch/backends/xnnpack/operators/op_abs.py +52 -0
  1779. executorch/backends/xnnpack/operators/op_add.py +77 -0
  1780. executorch/backends/xnnpack/operators/op_addmm.py +70 -0
  1781. executorch/backends/xnnpack/operators/op_avg_pooling2d.py +75 -0
  1782. executorch/backends/xnnpack/operators/op_bmm.py +54 -0
  1783. executorch/backends/xnnpack/operators/op_cat.py +121 -0
  1784. executorch/backends/xnnpack/operators/op_ceiling.py +52 -0
  1785. executorch/backends/xnnpack/operators/op_clamp.py +64 -0
  1786. executorch/backends/xnnpack/operators/op_clone.py +60 -0
  1787. executorch/backends/xnnpack/operators/op_conv2d.py +186 -0
  1788. executorch/backends/xnnpack/operators/op_cos.py +52 -0
  1789. executorch/backends/xnnpack/operators/op_div.py +56 -0
  1790. executorch/backends/xnnpack/operators/op_dynamic_dequantize_ops.py +114 -0
  1791. executorch/backends/xnnpack/operators/op_dynamic_quantize_ops.py +179 -0
  1792. executorch/backends/xnnpack/operators/op_elu.py +56 -0
  1793. executorch/backends/xnnpack/operators/op_exp.py +52 -0
  1794. executorch/backends/xnnpack/operators/op_floor.py +48 -0
  1795. executorch/backends/xnnpack/operators/op_gelu.py +52 -0
  1796. executorch/backends/xnnpack/operators/op_hardswish.py +52 -0
  1797. executorch/backends/xnnpack/operators/op_hardtanh.py +67 -0
  1798. executorch/backends/xnnpack/operators/op_leaky_relu.py +65 -0
  1799. executorch/backends/xnnpack/operators/op_linear.py +110 -0
  1800. executorch/backends/xnnpack/operators/op_log.py +52 -0
  1801. executorch/backends/xnnpack/operators/op_matrix_multiplication.py +66 -0
  1802. executorch/backends/xnnpack/operators/op_max_dim.py +206 -0
  1803. executorch/backends/xnnpack/operators/op_max_pool2d.py +136 -0
  1804. executorch/backends/xnnpack/operators/op_maximum.py +55 -0
  1805. executorch/backends/xnnpack/operators/op_mean_dim.py +78 -0
  1806. executorch/backends/xnnpack/operators/op_minimum.py +54 -0
  1807. executorch/backends/xnnpack/operators/op_multiply.py +84 -0
  1808. executorch/backends/xnnpack/operators/op_negate.py +51 -0
  1809. executorch/backends/xnnpack/operators/op_permute.py +87 -0
  1810. executorch/backends/xnnpack/operators/op_prelu.py +56 -0
  1811. executorch/backends/xnnpack/operators/op_quant_dequant.py +196 -0
  1812. executorch/backends/xnnpack/operators/op_relu.py +61 -0
  1813. executorch/backends/xnnpack/operators/op_rsqrt.py +52 -0
  1814. executorch/backends/xnnpack/operators/op_sigmoid.py +48 -0
  1815. executorch/backends/xnnpack/operators/op_sin.py +52 -0
  1816. executorch/backends/xnnpack/operators/op_skip_ops.py +89 -0
  1817. executorch/backends/xnnpack/operators/op_slice_copy.py +100 -0
  1818. executorch/backends/xnnpack/operators/op_softmax.py +58 -0
  1819. executorch/backends/xnnpack/operators/op_square.py +52 -0
  1820. executorch/backends/xnnpack/operators/op_square_root.py +52 -0
  1821. executorch/backends/xnnpack/operators/op_squeeze.py +145 -0
  1822. executorch/backends/xnnpack/operators/op_static_constant_pad.py +145 -0
  1823. executorch/backends/xnnpack/operators/op_static_resize_bilinear_2d.py +61 -0
  1824. executorch/backends/xnnpack/operators/op_sub.py +77 -0
  1825. executorch/backends/xnnpack/operators/op_tanh.py +52 -0
  1826. executorch/backends/xnnpack/operators/op_to_copy.py +91 -0
  1827. executorch/backends/xnnpack/operators/op_view_copy.py +96 -0
  1828. executorch/backends/xnnpack/operators/quant_params.py +412 -0
  1829. executorch/backends/xnnpack/partition/config/__init__.py +127 -0
  1830. executorch/backends/xnnpack/partition/config/gemm_configs.py +575 -0
  1831. executorch/backends/xnnpack/partition/config/generic_node_configs.py +711 -0
  1832. executorch/backends/xnnpack/partition/config/node_configs.py +148 -0
  1833. executorch/backends/xnnpack/partition/config/quant_affine_configs.py +56 -0
  1834. executorch/backends/xnnpack/partition/config/xnnpack_config.py +252 -0
  1835. executorch/backends/xnnpack/partition/graphs/bilinear_2d.py +55 -0
  1836. executorch/backends/xnnpack/partition/graphs/sdpa.py +86 -0
  1837. executorch/backends/xnnpack/partition/xnnpack_partitioner.py +160 -0
  1838. executorch/backends/xnnpack/quantizer/xnnpack_quantizer.py +462 -0
  1839. executorch/backends/xnnpack/quantizer/xnnpack_quantizer_utils.py +1121 -0
  1840. executorch/backends/xnnpack/recipes/__init__.py +19 -0
  1841. executorch/backends/xnnpack/recipes/xnnpack_recipe_provider.py +200 -0
  1842. executorch/backends/xnnpack/recipes/xnnpack_recipe_types.py +37 -0
  1843. executorch/backends/xnnpack/serialization/runtime_schema.fbs +389 -0
  1844. executorch/backends/xnnpack/serialization/schema.fbs +381 -0
  1845. executorch/backends/xnnpack/serialization/xnnpack_graph_schema.py +542 -0
  1846. executorch/backends/xnnpack/serialization/xnnpack_graph_serialize.py +373 -0
  1847. executorch/backends/xnnpack/test/__init__.py +0 -0
  1848. executorch/backends/xnnpack/test/models/deeplab_v3.py +42 -0
  1849. executorch/backends/xnnpack/test/models/edsr.py +55 -0
  1850. executorch/backends/xnnpack/test/models/emformer_rnnt.py +98 -0
  1851. executorch/backends/xnnpack/test/models/inception_v3.py +84 -0
  1852. executorch/backends/xnnpack/test/models/inception_v4.py +62 -0
  1853. executorch/backends/xnnpack/test/models/llama2_et_example.py +49 -0
  1854. executorch/backends/xnnpack/test/models/mobilebert.py +55 -0
  1855. executorch/backends/xnnpack/test/models/mobilenet_v2.py +103 -0
  1856. executorch/backends/xnnpack/test/models/mobilenet_v3.py +84 -0
  1857. executorch/backends/xnnpack/test/models/resnet.py +90 -0
  1858. executorch/backends/xnnpack/test/models/torchvision_vit.py +105 -0
  1859. executorch/backends/xnnpack/test/models/very_big_model.py +45 -0
  1860. executorch/backends/xnnpack/test/models/w2l.py +60 -0
  1861. executorch/backends/xnnpack/test/ops/test_abs.py +87 -0
  1862. executorch/backends/xnnpack/test/ops/test_add.py +266 -0
  1863. executorch/backends/xnnpack/test/ops/test_avgpool2d.py +92 -0
  1864. executorch/backends/xnnpack/test/ops/test_bilinear2d.py +152 -0
  1865. executorch/backends/xnnpack/test/ops/test_bmm.py +49 -0
  1866. executorch/backends/xnnpack/test/ops/test_cat.py +254 -0
  1867. executorch/backends/xnnpack/test/ops/test_ceil.py +63 -0
  1868. executorch/backends/xnnpack/test/ops/test_check_quant_params.py +107 -0
  1869. executorch/backends/xnnpack/test/ops/test_clamp.py +79 -0
  1870. executorch/backends/xnnpack/test/ops/test_clone.py +137 -0
  1871. executorch/backends/xnnpack/test/ops/test_conv1d.py +190 -0
  1872. executorch/backends/xnnpack/test/ops/test_conv2d.py +747 -0
  1873. executorch/backends/xnnpack/test/ops/test_cos.py +87 -0
  1874. executorch/backends/xnnpack/test/ops/test_div.py +78 -0
  1875. executorch/backends/xnnpack/test/ops/test_elu.py +98 -0
  1876. executorch/backends/xnnpack/test/ops/test_exp.py +70 -0
  1877. executorch/backends/xnnpack/test/ops/test_floor.py +58 -0
  1878. executorch/backends/xnnpack/test/ops/test_gelu.py +71 -0
  1879. executorch/backends/xnnpack/test/ops/test_hardswish.py +70 -0
  1880. executorch/backends/xnnpack/test/ops/test_hardtanh.py +83 -0
  1881. executorch/backends/xnnpack/test/ops/test_leaky_relu.py +131 -0
  1882. executorch/backends/xnnpack/test/ops/test_linear.py +1059 -0
  1883. executorch/backends/xnnpack/test/ops/test_log.py +45 -0
  1884. executorch/backends/xnnpack/test/ops/test_lstm.py +67 -0
  1885. executorch/backends/xnnpack/test/ops/test_max_dim.py +66 -0
  1886. executorch/backends/xnnpack/test/ops/test_maximum.py +66 -0
  1887. executorch/backends/xnnpack/test/ops/test_maxpool2d.py +221 -0
  1888. executorch/backends/xnnpack/test/ops/test_mean_dim.py +98 -0
  1889. executorch/backends/xnnpack/test/ops/test_minimum.py +49 -0
  1890. executorch/backends/xnnpack/test/ops/test_multiply.py +149 -0
  1891. executorch/backends/xnnpack/test/ops/test_negate.py +58 -0
  1892. executorch/backends/xnnpack/test/ops/test_permute.py +134 -0
  1893. executorch/backends/xnnpack/test/ops/test_pow.py +62 -0
  1894. executorch/backends/xnnpack/test/ops/test_prelu.py +50 -0
  1895. executorch/backends/xnnpack/test/ops/test_quantize_per_tensor.py +70 -0
  1896. executorch/backends/xnnpack/test/ops/test_relu.py +38 -0
  1897. executorch/backends/xnnpack/test/ops/test_rsqrt.py +45 -0
  1898. executorch/backends/xnnpack/test/ops/test_sigmoid.py +45 -0
  1899. executorch/backends/xnnpack/test/ops/test_sin.py +87 -0
  1900. executorch/backends/xnnpack/test/ops/test_slice_copy.py +170 -0
  1901. executorch/backends/xnnpack/test/ops/test_softmax.py +73 -0
  1902. executorch/backends/xnnpack/test/ops/test_sqrt.py +45 -0
  1903. executorch/backends/xnnpack/test/ops/test_square.py +48 -0
  1904. executorch/backends/xnnpack/test/ops/test_static_constant_pad.py +294 -0
  1905. executorch/backends/xnnpack/test/ops/test_sub.py +178 -0
  1906. executorch/backends/xnnpack/test/ops/test_tanh.py +43 -0
  1907. executorch/backends/xnnpack/test/ops/test_to_copy.py +113 -0
  1908. executorch/backends/xnnpack/test/passes/__init__.py +0 -0
  1909. executorch/backends/xnnpack/test/passes/test_activation_fusion.py +232 -0
  1910. executorch/backends/xnnpack/test/passes/test_batch_norm_fusion.py +179 -0
  1911. executorch/backends/xnnpack/test/passes/test_channels_last_tagged_reshape.py +635 -0
  1912. executorch/backends/xnnpack/test/passes/test_convert_to_linear.py +43 -0
  1913. executorch/backends/xnnpack/test/passes/test_decompose_cat_pass.py +112 -0
  1914. executorch/backends/xnnpack/test/passes/test_propagate_custom_meta_pass.py +185 -0
  1915. executorch/backends/xnnpack/test/passes/test_remove_get_item_pass.py +99 -0
  1916. executorch/backends/xnnpack/test/passes/test_remove_redundant_copy_pass.py +182 -0
  1917. executorch/backends/xnnpack/test/quantizer/test_pt2e_quantization.py +818 -0
  1918. executorch/backends/xnnpack/test/quantizer/test_representation.py +308 -0
  1919. executorch/backends/xnnpack/test/quantizer/test_xnnpack_quantizer.py +1217 -0
  1920. executorch/backends/xnnpack/test/recipes/test_xnnpack_recipes.py +320 -0
  1921. executorch/backends/xnnpack/test/serialization/test_serialization.py +50 -0
  1922. executorch/backends/xnnpack/test/serialization/test_xnnheader.py +104 -0
  1923. executorch/backends/xnnpack/test/test_xnnpack_partitioner.py +163 -0
  1924. executorch/backends/xnnpack/test/test_xnnpack_utils.py +532 -0
  1925. executorch/backends/xnnpack/test/test_xnnpack_utils_classes.py +44 -0
  1926. executorch/backends/xnnpack/test/tester/__init__.py +29 -0
  1927. executorch/backends/xnnpack/test/tester/tester.py +130 -0
  1928. executorch/backends/xnnpack/third-party/FP16/configure.py +65 -0
  1929. executorch/backends/xnnpack/third-party/FP16/confu.yaml +8 -0
  1930. executorch/backends/xnnpack/third-party/FP16/include/fp16/__init__.py +0 -0
  1931. executorch/backends/xnnpack/third-party/FP16/include/fp16/avx.py +50 -0
  1932. executorch/backends/xnnpack/third-party/FP16/include/fp16/avx2.py +53 -0
  1933. executorch/backends/xnnpack/third-party/FP16/test/peachpy/stubs.py +38 -0
  1934. executorch/backends/xnnpack/third-party/FXdiv/configure.py +30 -0
  1935. executorch/backends/xnnpack/third-party/FXdiv/confu.yaml +6 -0
  1936. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/aarch32_template.py +361 -0
  1937. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/aarch64_template.py +347 -0
  1938. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/arm_template.py +78 -0
  1939. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512bf16_template.py +81 -0
  1940. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512f_template.py +390 -0
  1941. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/avx512vnni_template.py +703 -0
  1942. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/base_architecture.py +276 -0
  1943. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/fma3_template.py +260 -0
  1944. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate.py +71 -0
  1945. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_bf16_f32_gemm_microkernels.py +39 -0
  1946. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_f32_gemm_microkernels.py +104 -0
  1947. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_gemm_microkernels_main.py +29 -0
  1948. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qd8_f32_qc4w_gemm_microkernels.py +69 -0
  1949. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qd8_f32_qc8w_gemm_microkernels.py +93 -0
  1950. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qs8_qc4w_gemm_microkernels.py +42 -0
  1951. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/generate_qs8_qc8w_gemm_microkernels.py +43 -0
  1952. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neondot_template.py +463 -0
  1953. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neonfma_template.py +248 -0
  1954. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/neonmlal_aarch32_template.py +394 -0
  1955. executorch/backends/xnnpack/third-party/XNNPACK/gemm_compiler/x64_template.py +512 -0
  1956. executorch/backends/xnnpack/third-party/XNNPACK/scripts/check_config.py +152 -0
  1957. executorch/backends/xnnpack/third-party/XNNPACK/scripts/check_files_changed.py +53 -0
  1958. executorch/backends/xnnpack/third-party/XNNPACK/scripts/generate-build-identifier.py +106 -0
  1959. executorch/backends/xnnpack/third-party/XNNPACK/scripts/sort-filenames.py +134 -0
  1960. executorch/backends/xnnpack/third-party/XNNPACK/test/bf16-f32-gemm-minmax.yaml +121 -0
  1961. executorch/backends/xnnpack/third-party/XNNPACK/test/bf16-gemm-minmax.yaml +106 -0
  1962. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-conv-hwc2chw.yaml +8 -0
  1963. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-dwconv2d-chw.yaml +88 -0
  1964. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-f32acc-gemm-minmax.yaml +42 -0
  1965. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-f32acc-igemm-minmax.yaml +42 -0
  1966. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-gemm-minmax.yaml +219 -0
  1967. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-ibilinear-chw.yaml +9 -0
  1968. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-ibilinear.yaml +12 -0
  1969. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-igemm-minmax.yaml +165 -0
  1970. executorch/backends/xnnpack/third-party/XNNPACK/test/f16-raddstoreexpminusmax.yaml +52 -0
  1971. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-argmaxpool.yaml +15 -0
  1972. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-conv-hwc2chw.yaml +32 -0
  1973. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-dwconv2d-chw.yaml +712 -0
  1974. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm-minmax.yaml +1572 -0
  1975. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm-relu.yaml +123 -0
  1976. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-gemm.yaml +127 -0
  1977. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ibilinear-chw.yaml +22 -0
  1978. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ibilinear.yaml +23 -0
  1979. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm-minmax.yaml +1001 -0
  1980. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm-relu.yaml +123 -0
  1981. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-igemm.yaml +123 -0
  1982. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-ppmm-minmax.yaml +122 -0
  1983. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc4w-gemm-minmax.yaml +277 -0
  1984. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm-minmax.yaml +757 -0
  1985. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm-relu.yaml +67 -0
  1986. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-qc8w-gemm.yaml +115 -0
  1987. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-raddexpminusmax.yaml +38 -0
  1988. executorch/backends/xnnpack/third-party/XNNPACK/test/f32-raddstoreexpminusmax.yaml +88 -0
  1989. executorch/backends/xnnpack/third-party/XNNPACK/test/pf16-gemm-minmax.yaml +20 -0
  1990. executorch/backends/xnnpack/third-party/XNNPACK/test/pf32-gemm-minmax.yaml +34 -0
  1991. executorch/backends/xnnpack/third-party/XNNPACK/test/pqs8-qc8w-gemm-minmax.yaml +20 -0
  1992. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qb4w-gemm-minmax.yaml +260 -0
  1993. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc4w-gemm-minmax.yaml +678 -0
  1994. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc8w-gemm-minmax.yaml +412 -0
  1995. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f16-qc8w-igemm-minmax.yaml +360 -0
  1996. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qb4w-gemm-minmax.yaml +537 -0
  1997. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc4w-gemm-minmax.yaml +2176 -0
  1998. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc8w-gemm-minmax.yaml +1564 -0
  1999. executorch/backends/xnnpack/third-party/XNNPACK/test/qd8-f32-qc8w-igemm-minmax.yaml +1035 -0
  2000. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qb4w-gemm-minmax.yaml +54 -0
  2001. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qc4w-gemm-minmax.yaml +66 -0
  2002. executorch/backends/xnnpack/third-party/XNNPACK/test/qp8-f32-qc8w-gemm-minmax.yaml +42 -0
  2003. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc4w-gemm-minmax-fp32.yaml +345 -0
  2004. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc8w-gemm-minmax-fp32.yaml +2029 -0
  2005. executorch/backends/xnnpack/third-party/XNNPACK/test/qs8-qc8w-igemm-minmax-fp32.yaml +1831 -0
  2006. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-gemm-minmax-fp32.yaml +584 -0
  2007. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-gemm-minmax-rndnu.yaml +148 -0
  2008. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-igemm-minmax-fp32.yaml +581 -0
  2009. executorch/backends/xnnpack/third-party/XNNPACK/test/qu8-igemm-minmax-rndnu.yaml +110 -0
  2010. executorch/backends/xnnpack/third-party/XNNPACK/test/s8-ibilinear.yaml +22 -0
  2011. executorch/backends/xnnpack/third-party/XNNPACK/test/u8-ibilinear.yaml +22 -0
  2012. executorch/backends/xnnpack/third-party/XNNPACK/test/x8-lut.yaml +47 -0
  2013. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-argmaxpool-test.py +692 -0
  2014. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-conv-hwc-test.py +543 -0
  2015. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-conv-hwc2chw-test.py +591 -0
  2016. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-dwconv-test.py +339 -0
  2017. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-dwconv2d-chw-test.py +463 -0
  2018. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-gemm-test.py +1150 -0
  2019. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-ibilinear-chw-test.py +236 -0
  2020. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-ibilinear-test.py +237 -0
  2021. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-lut-test.py +164 -0
  2022. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-raddexpminusmax-test.py +159 -0
  2023. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-raddstoreexpminusmax-test.py +210 -0
  2024. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-reduce-discontiguous-test.py +495 -0
  2025. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-reduce-test.py +289 -0
  2026. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-spmm-test.py +554 -0
  2027. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-vbinary-test.py +179 -0
  2028. executorch/backends/xnnpack/third-party/XNNPACK/tools/generate-vunary-test.py +280 -0
  2029. executorch/backends/xnnpack/third-party/XNNPACK/tools/parse-microkernel-bench.py +189 -0
  2030. executorch/backends/xnnpack/third-party/XNNPACK/tools/primes.py +1015 -0
  2031. executorch/backends/xnnpack/third-party/XNNPACK/tools/update-microkernels.py +567 -0
  2032. executorch/backends/xnnpack/third-party/XNNPACK/tools/xngen.py +159 -0
  2033. executorch/backends/xnnpack/third-party/XNNPACK/tools/xnncommon.py +275 -0
  2034. executorch/backends/xnnpack/third-party/cpuinfo/configure.py +110 -0
  2035. executorch/backends/xnnpack/third-party/cpuinfo/confu.yaml +9 -0
  2036. executorch/backends/xnnpack/third-party/cpuinfo/deps/clog/configure.py +26 -0
  2037. executorch/backends/xnnpack/third-party/cpuinfo/deps/clog/confu.yaml +5 -0
  2038. executorch/backends/xnnpack/third-party/cpuinfo/scripts/android-device-dump.py +400 -0
  2039. executorch/backends/xnnpack/third-party/cpuinfo/scripts/arm-linux-filesystem-dump.py +107 -0
  2040. executorch/backends/xnnpack/third-party/cpuinfo/scripts/parse-x86-cpuid-dump.py +47 -0
  2041. executorch/backends/xnnpack/third-party/generate-cpuinfo-wrappers.py +93 -0
  2042. executorch/backends/xnnpack/third-party/pthreadpool/configure.py +38 -0
  2043. executorch/backends/xnnpack/third-party/pthreadpool/confu.yaml +8 -0
  2044. executorch/backends/xnnpack/utils/configs.py +49 -0
  2045. executorch/backends/xnnpack/utils/quant_utils.py +301 -0
  2046. executorch/backends/xnnpack/utils/utils.py +221 -0
  2047. executorch/backends/xnnpack/utils/xnnpack_constants.py +37 -0
  2048. executorch/backends/xnnpack/xnnpack_preprocess.py +194 -0
  2049. executorch/codegen/__init__.py +0 -0
  2050. executorch/codegen/api/__init__.py +0 -0
  2051. executorch/codegen/api/custom_ops.py +151 -0
  2052. executorch/codegen/api/et_cpp.py +368 -0
  2053. executorch/codegen/api/types/__init__.py +5 -0
  2054. executorch/codegen/api/types/signatures.py +77 -0
  2055. executorch/codegen/api/types/types.py +77 -0
  2056. executorch/codegen/api/unboxing.py +232 -0
  2057. executorch/codegen/gen.py +1049 -0
  2058. executorch/codegen/model.py +220 -0
  2059. executorch/codegen/parse.py +158 -0
  2060. executorch/codegen/test/test_executorch_custom_ops.py +153 -0
  2061. executorch/codegen/test/test_executorch_gen.py +698 -0
  2062. executorch/codegen/test/test_executorch_signatures.py +65 -0
  2063. executorch/codegen/test/test_executorch_types.py +121 -0
  2064. executorch/codegen/test/test_executorch_unboxing.py +165 -0
  2065. executorch/codegen/test/test_selective_build.py +51 -0
  2066. executorch/codegen/tools/__init__.py +0 -0
  2067. executorch/codegen/tools/combine_prim_ops_headers.py +164 -0
  2068. executorch/codegen/tools/gen_all_oplist.py +252 -0
  2069. executorch/codegen/tools/gen_oplist.py +369 -0
  2070. executorch/codegen/tools/gen_ops_def.py +96 -0
  2071. executorch/codegen/tools/gen_selected_op_variants.py +133 -0
  2072. executorch/codegen/tools/gen_selected_prim_ops.py +96 -0
  2073. executorch/codegen/tools/merge_yaml.py +125 -0
  2074. executorch/codegen/tools/selective_build.cp313-win_amd64.pyd +0 -0
  2075. executorch/codegen/tools/selective_build.pyi +23 -0
  2076. executorch/codegen/tools/test/test_gen_all_oplist.py +122 -0
  2077. executorch/codegen/tools/test/test_gen_oplist.py +263 -0
  2078. executorch/codegen/tools/test/test_gen_oplist_real_model.py +50 -0
  2079. executorch/codegen/tools/test/test_gen_selected_op_variants.py +204 -0
  2080. executorch/codegen/tools/test/test_tools_selective_build.py +129 -0
  2081. executorch/codegen/tools/yaml_util.py +15 -0
  2082. executorch/data/bin/__init__.py +44 -0
  2083. executorch/data/bin/flatc.exe +0 -0
  2084. executorch/data/lib/aoti_cuda_shims.lib +0 -0
  2085. executorch/devtools/__init__.py +19 -0
  2086. executorch/devtools/backend_debug/__init__.py +13 -0
  2087. executorch/devtools/backend_debug/delegation_info.py +183 -0
  2088. executorch/devtools/backend_debug/tests/test_delegation_info.py +79 -0
  2089. executorch/devtools/bundled_program/config.py +120 -0
  2090. executorch/devtools/bundled_program/core.py +395 -0
  2091. executorch/devtools/bundled_program/schema/__init__.py +9 -0
  2092. executorch/devtools/bundled_program/schema/bundled_program_schema.fbs +97 -0
  2093. executorch/devtools/bundled_program/schema/bundled_program_schema.py +99 -0
  2094. executorch/devtools/bundled_program/schema/scalar_type.fbs +43 -0
  2095. executorch/devtools/bundled_program/schema/test/test_schema.py +27 -0
  2096. executorch/devtools/bundled_program/serialize/__init__.py +122 -0
  2097. executorch/devtools/bundled_program/serialize/bundled_program_schema.fbs +97 -0
  2098. executorch/devtools/bundled_program/serialize/scalar_type.fbs +43 -0
  2099. executorch/devtools/bundled_program/serialize/test/test_serialize.py +37 -0
  2100. executorch/devtools/bundled_program/test/test_bundle_data.py +168 -0
  2101. executorch/devtools/bundled_program/test/test_config.py +131 -0
  2102. executorch/devtools/bundled_program/test/test_end2end.py +64 -0
  2103. executorch/devtools/bundled_program/util/test_util.py +289 -0
  2104. executorch/devtools/bundled_program/version.py +13 -0
  2105. executorch/devtools/debug_format/base_schema.py +61 -0
  2106. executorch/devtools/debug_format/et_schema.py +432 -0
  2107. executorch/devtools/etdump/etdump_schema_flatcc.fbs +181 -0
  2108. executorch/devtools/etdump/scalar_type.fbs +43 -0
  2109. executorch/devtools/etdump/schema_flatcc.py +154 -0
  2110. executorch/devtools/etdump/serialize.py +108 -0
  2111. executorch/devtools/etdump/tests/serialize_test.py +142 -0
  2112. executorch/devtools/etrecord/__init__.py +13 -0
  2113. executorch/devtools/etrecord/_etrecord.py +776 -0
  2114. executorch/devtools/etrecord/tests/etrecord_test.py +1748 -0
  2115. executorch/devtools/inspector/__init__.py +24 -0
  2116. executorch/devtools/inspector/_inspector.py +1495 -0
  2117. executorch/devtools/inspector/_inspector_utils.py +1286 -0
  2118. executorch/devtools/inspector/_intermediate_output_capturer.py +62 -0
  2119. executorch/devtools/inspector/inspector_cli.py +76 -0
  2120. executorch/devtools/inspector/numerical_comparator/__init__.py +25 -0
  2121. executorch/devtools/inspector/numerical_comparator/l1_numerical_comparator.py +27 -0
  2122. executorch/devtools/inspector/numerical_comparator/mse_numerical_comparator.py +29 -0
  2123. executorch/devtools/inspector/numerical_comparator/numerical_comparator_base.py +26 -0
  2124. executorch/devtools/inspector/numerical_comparator/snr_numerical_comparator.py +39 -0
  2125. executorch/devtools/inspector/tests/event_blocks_test.py +657 -0
  2126. executorch/devtools/inspector/tests/inspector_test.py +1263 -0
  2127. executorch/devtools/inspector/tests/inspector_test_utils.py +189 -0
  2128. executorch/devtools/inspector/tests/inspector_utils_test.py +896 -0
  2129. executorch/devtools/inspector/tests/intermediate_output_capturer_test.py +101 -0
  2130. executorch/devtools/inspector/tests/l1_comparator_test.py +49 -0
  2131. executorch/devtools/inspector/tests/mse_comparator_test.py +49 -0
  2132. executorch/devtools/inspector/tests/snr_comparator_test.py +52 -0
  2133. executorch/devtools/scripts/generate_profiling_csv.py +62 -0
  2134. executorch/devtools/size_analysis_tool/size_analysis_tool.py +180 -0
  2135. executorch/devtools/size_analysis_tool/size_analysis_tool_test.py +101 -0
  2136. executorch/devtools/visualization/__init__.py +13 -0
  2137. executorch/devtools/visualization/visualization_utils.py +301 -0
  2138. executorch/devtools/visualization/visualization_utils_test.py +177 -0
  2139. executorch/examples/apple/coreml/llama/export.py +224 -0
  2140. executorch/examples/apple/coreml/llama/export_static_llm_coreml.py +504 -0
  2141. executorch/examples/apple/coreml/llama/llama_transformer.py +958 -0
  2142. executorch/examples/apple/coreml/llama/run.py +206 -0
  2143. executorch/examples/apple/coreml/llama/run_lookahead.py +284 -0
  2144. executorch/examples/apple/coreml/llama/run_static_llm.py +293 -0
  2145. executorch/examples/apple/coreml/llama/test.py +48 -0
  2146. executorch/examples/apple/coreml/llama/utils.py +116 -0
  2147. executorch/examples/apple/coreml/scripts/debugger_cli.py +181 -0
  2148. executorch/examples/apple/coreml/scripts/export.py +280 -0
  2149. executorch/examples/apple/coreml/scripts/extract_coreml_models.py +77 -0
  2150. executorch/examples/apple/coreml/scripts/inspector_cli.py +64 -0
  2151. executorch/examples/apple/coreml/scripts/inspector_utils.py +425 -0
  2152. executorch/examples/apple/mps/scripts/bench_utils.py +108 -0
  2153. executorch/examples/apple/mps/scripts/mps_example.py +230 -0
  2154. executorch/examples/cuda/scripts/__init__.py +0 -0
  2155. executorch/examples/cuda/scripts/export.py +107 -0
  2156. executorch/examples/llm_pte_finetuning/__init__.py +19 -0
  2157. executorch/examples/llm_pte_finetuning/llama3_config.yaml +96 -0
  2158. executorch/examples/llm_pte_finetuning/model_exporter.py +87 -0
  2159. executorch/examples/llm_pte_finetuning/model_loading_lib.py +143 -0
  2160. executorch/examples/llm_pte_finetuning/phi3_alpaca_code_config.yaml +44 -0
  2161. executorch/examples/llm_pte_finetuning/phi3_config.yaml +42 -0
  2162. executorch/examples/llm_pte_finetuning/qwen_05b_config.yaml +43 -0
  2163. executorch/examples/llm_pte_finetuning/runner.py +135 -0
  2164. executorch/examples/llm_pte_finetuning/training_lib.py +157 -0
  2165. executorch/examples/models/__init__.py +98 -0
  2166. executorch/examples/models/checkpoint.py +114 -0
  2167. executorch/examples/models/codegen/__init__.py +16 -0
  2168. executorch/examples/models/codegen/convert_weight.py +93 -0
  2169. executorch/examples/models/deeplab_v3/__init__.py +9 -0
  2170. executorch/examples/models/deeplab_v3/model.py +50 -0
  2171. executorch/examples/models/deepseek-r1-distill-llama-8B/config/deepseek_xnnpack_q8da4w.yaml +16 -0
  2172. executorch/examples/models/deit_tiny/__init__.py +8 -0
  2173. executorch/examples/models/deit_tiny/model.py +36 -0
  2174. executorch/examples/models/edsr/__init__.py +11 -0
  2175. executorch/examples/models/edsr/model.py +28 -0
  2176. executorch/examples/models/efficient_sam/__init__.py +11 -0
  2177. executorch/examples/models/efficient_sam/efficient_sam_core/build_efficient_sam.py +25 -0
  2178. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam.py +316 -0
  2179. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam_decoder.py +373 -0
  2180. executorch/examples/models/efficient_sam/efficient_sam_core/efficient_sam_encoder.py +261 -0
  2181. executorch/examples/models/efficient_sam/efficient_sam_core/mlp.py +31 -0
  2182. executorch/examples/models/efficient_sam/efficient_sam_core/two_way_transformer.py +268 -0
  2183. executorch/examples/models/efficient_sam/model.py +34 -0
  2184. executorch/examples/models/emformer_rnnt/__init__.py +17 -0
  2185. executorch/examples/models/emformer_rnnt/model.py +132 -0
  2186. executorch/examples/models/gemma/__init__.py +16 -0
  2187. executorch/examples/models/gemma/convert_weights.py +104 -0
  2188. executorch/examples/models/gemma2/__init__.py +19 -0
  2189. executorch/examples/models/gemma2/convert_weights.py +113 -0
  2190. executorch/examples/models/gemma3/__init__.py +16 -0
  2191. executorch/examples/models/gemma3/convert_weights.py +110 -0
  2192. executorch/examples/models/glm/__init__.py +16 -0
  2193. executorch/examples/models/glm/convert_weights.py +79 -0
  2194. executorch/examples/models/granite/__init__.py +16 -0
  2195. executorch/examples/models/granite/convert_weights.py +106 -0
  2196. executorch/examples/models/inception_v3/__init__.py +11 -0
  2197. executorch/examples/models/inception_v3/model.py +28 -0
  2198. executorch/examples/models/inception_v4/__init__.py +11 -0
  2199. executorch/examples/models/inception_v4/model.py +26 -0
  2200. executorch/examples/models/internvl3/__init__.py +16 -0
  2201. executorch/examples/models/internvl3/convert_weights.py +82 -0
  2202. executorch/examples/models/lfm2/__init__.py +10 -0
  2203. executorch/examples/models/lfm2/config/lfm2_xnnpack_fp32.yaml +12 -0
  2204. executorch/examples/models/lfm2/config/lfm2_xnnpack_q8da4w.yaml +15 -0
  2205. executorch/examples/models/lfm2/convert_weights.py +74 -0
  2206. executorch/examples/models/lfm2/short_conv.py +112 -0
  2207. executorch/examples/models/llama/__init__.py +11 -0
  2208. executorch/examples/models/llama/attention.py +536 -0
  2209. executorch/examples/models/llama/config/llama_bf16.yaml +7 -0
  2210. executorch/examples/models/llama/config/llama_q8da4w.yaml +11 -0
  2211. executorch/examples/models/llama/config/llama_xnnpack.yaml +17 -0
  2212. executorch/examples/models/llama/config/llama_xnnpack_qat.yaml +23 -0
  2213. executorch/examples/models/llama/config/llama_xnnpack_spinquant.yaml +22 -0
  2214. executorch/examples/models/llama/config/test_llm_config.py +104 -0
  2215. executorch/examples/models/llama/convert_weights.py +61 -0
  2216. executorch/examples/models/llama/eval_llama.py +38 -0
  2217. executorch/examples/models/llama/eval_llama_lib.py +395 -0
  2218. executorch/examples/models/llama/evaluate/__init__.py +11 -0
  2219. executorch/examples/models/llama/evaluate/eager_eval.py +87 -0
  2220. executorch/examples/models/llama/experimental/__init__.py +0 -0
  2221. executorch/examples/models/llama/experimental/generate.py +190 -0
  2222. executorch/examples/models/llama/experimental/load_gguf_q4_0.py +184 -0
  2223. executorch/examples/models/llama/experimental/subclass.py +413 -0
  2224. executorch/examples/models/llama/experimental/test_subclass.py +28 -0
  2225. executorch/examples/models/llama/export_llama.py +56 -0
  2226. executorch/examples/models/llama/export_llama_lib.py +1560 -0
  2227. executorch/examples/models/llama/fairseq2.py +63 -0
  2228. executorch/examples/models/llama/feed_forward.py +68 -0
  2229. executorch/examples/models/llama/hf_download.py +50 -0
  2230. executorch/examples/models/llama/install_requirement_helper.py +29 -0
  2231. executorch/examples/models/llama/llama_transformer.py +289 -0
  2232. executorch/examples/models/llama/lora.py +48 -0
  2233. executorch/examples/models/llama/model.py +378 -0
  2234. executorch/examples/models/llama/model_args.py +173 -0
  2235. executorch/examples/models/llama/norm.py +55 -0
  2236. executorch/examples/models/llama/rope.py +327 -0
  2237. executorch/examples/models/llama/runner/eager.py +142 -0
  2238. executorch/examples/models/llama/runner/generation.py +234 -0
  2239. executorch/examples/models/llama/runner/native.py +145 -0
  2240. executorch/examples/models/llama/source_transformation/__init__.py +0 -0
  2241. executorch/examples/models/llama/source_transformation/apply_spin_quant_r1_r2.py +202 -0
  2242. executorch/examples/models/llama/source_transformation/attention.py +223 -0
  2243. executorch/examples/models/llama/source_transformation/attention_sink.py +312 -0
  2244. executorch/examples/models/llama/source_transformation/custom_kv_cache.py +598 -0
  2245. executorch/examples/models/llama/source_transformation/lora.py +153 -0
  2246. executorch/examples/models/llama/source_transformation/pre_quantization.py +204 -0
  2247. executorch/examples/models/llama/source_transformation/prune_vocab.py +119 -0
  2248. executorch/examples/models/llama/source_transformation/quantize.py +910 -0
  2249. executorch/examples/models/llama/source_transformation/rms_norm.py +23 -0
  2250. executorch/examples/models/llama/source_transformation/rope.py +39 -0
  2251. executorch/examples/models/llama/source_transformation/sdpa.py +550 -0
  2252. executorch/examples/models/llama/source_transformation/spin_quant.py +88 -0
  2253. executorch/examples/models/llama/source_transformation/test_attention_sink.py +514 -0
  2254. executorch/examples/models/llama/source_transformation/test_quantized_kv_cache.py +126 -0
  2255. executorch/examples/models/llama/source_transformation/test_quantized_sdpa.py +167 -0
  2256. executorch/examples/models/llama/source_transformation/test_sdpa_with_quantized_kv_cache.py +97 -0
  2257. executorch/examples/models/llama/static_attention.py +1257 -0
  2258. executorch/examples/models/llama/tests/test_export_llama_lib.py +50 -0
  2259. executorch/examples/models/llama/tests/test_pre_quantization_transforms.py +188 -0
  2260. executorch/examples/models/llama/tests/test_replace_kv_cache.py +158 -0
  2261. executorch/examples/models/llama/tests/test_ring_attention.py +359 -0
  2262. executorch/examples/models/llama/tests/test_ring_kv_cache.py +492 -0
  2263. executorch/examples/models/llama/tests/test_simple_sdpa.py +66 -0
  2264. executorch/examples/models/llama/tests/test_static_attention.py +363 -0
  2265. executorch/examples/models/llama/tokenizer/tiktoken.py +224 -0
  2266. executorch/examples/models/llama3_2_vision/__init__.py +14 -0
  2267. executorch/examples/models/llama3_2_vision/preprocess/__init__.py +0 -0
  2268. executorch/examples/models/llama3_2_vision/preprocess/export_preprocess.py +45 -0
  2269. executorch/examples/models/llama3_2_vision/preprocess/model.py +72 -0
  2270. executorch/examples/models/llama3_2_vision/preprocess/test_preprocess.py +338 -0
  2271. executorch/examples/models/llama3_2_vision/runner/eager.py +59 -0
  2272. executorch/examples/models/llama3_2_vision/runner/exported.py +94 -0
  2273. executorch/examples/models/llama3_2_vision/runner/generation.py +102 -0
  2274. executorch/examples/models/llama3_2_vision/runner/native.py +143 -0
  2275. executorch/examples/models/llama3_2_vision/text_decoder/model.py +199 -0
  2276. executorch/examples/models/llama3_2_vision/text_decoder/test/__init__.py +0 -0
  2277. executorch/examples/models/llama3_2_vision/text_decoder/test/test_text_decoder.py +89 -0
  2278. executorch/examples/models/llama3_2_vision/vision_encoder/__init__.py +12 -0
  2279. executorch/examples/models/llama3_2_vision/vision_encoder/model.py +96 -0
  2280. executorch/examples/models/llama3_2_vision/vision_encoder/test/__init__.py +0 -0
  2281. executorch/examples/models/llama3_2_vision/vision_encoder/test/test_vision_encoder.py +46 -0
  2282. executorch/examples/models/llava/__init__.py +11 -0
  2283. executorch/examples/models/llava/export_llava.py +365 -0
  2284. executorch/examples/models/llava/model.py +415 -0
  2285. executorch/examples/models/llava/test/test_llava.py +151 -0
  2286. executorch/examples/models/llava/test/test_pte.py +106 -0
  2287. executorch/examples/models/lstm/__init__.py +12 -0
  2288. executorch/examples/models/lstm/model.py +31 -0
  2289. executorch/examples/models/mobilebert/__init__.py +11 -0
  2290. executorch/examples/models/mobilebert/model.py +31 -0
  2291. executorch/examples/models/mobilenet_v2/__init__.py +12 -0
  2292. executorch/examples/models/mobilenet_v2/model.py +73 -0
  2293. executorch/examples/models/mobilenet_v3/__init__.py +11 -0
  2294. executorch/examples/models/mobilenet_v3/model.py +30 -0
  2295. executorch/examples/models/model_base.py +47 -0
  2296. executorch/examples/models/model_factory.py +60 -0
  2297. executorch/examples/models/moshi/mimi/test_mimi.py +245 -0
  2298. executorch/examples/models/parakeet/export_parakeet_tdt.py +591 -0
  2299. executorch/examples/models/phi-3-mini/__init__.py +11 -0
  2300. executorch/examples/models/phi-3-mini-lora/export_model.py +144 -0
  2301. executorch/examples/models/phi_4_mini/__init__.py +16 -0
  2302. executorch/examples/models/phi_4_mini/config/phi_4_mini_xnnpack.yaml +12 -0
  2303. executorch/examples/models/phi_4_mini/convert_weights.py +151 -0
  2304. executorch/examples/models/qwen2_5/__init__.py +16 -0
  2305. executorch/examples/models/qwen2_5/config/qwen2_5_xnnpack_q8da4w.yaml +14 -0
  2306. executorch/examples/models/qwen2_5/convert_weights.py +90 -0
  2307. executorch/examples/models/qwen3/__init__.py +16 -0
  2308. executorch/examples/models/qwen3/config/qwen3_xnnpack.yaml +18 -0
  2309. executorch/examples/models/qwen3/config/qwen3_xnnpack_q8da4w.yaml +20 -0
  2310. executorch/examples/models/qwen3/convert_weights.py +140 -0
  2311. executorch/examples/models/resnet/__init__.py +12 -0
  2312. executorch/examples/models/resnet/model.py +48 -0
  2313. executorch/examples/models/smollm2/__init__.py +16 -0
  2314. executorch/examples/models/smollm2/convert_weights.py +84 -0
  2315. executorch/examples/models/smollm3/__init__.py +16 -0
  2316. executorch/examples/models/smollm3/convert_weights.py +112 -0
  2317. executorch/examples/models/smolvlm/__init__.py +16 -0
  2318. executorch/examples/models/smolvlm/convert_weights.py +77 -0
  2319. executorch/examples/models/stable_diffusion/__init__.py +9 -0
  2320. executorch/examples/models/stable_diffusion/model.py +193 -0
  2321. executorch/examples/models/test/__init__.py +0 -0
  2322. executorch/examples/models/test/test_export.py +161 -0
  2323. executorch/examples/models/torchvision_vit/__init__.py +11 -0
  2324. executorch/examples/models/torchvision_vit/model.py +28 -0
  2325. executorch/examples/models/toy_model/__init__.py +25 -0
  2326. executorch/examples/models/toy_model/model.py +143 -0
  2327. executorch/examples/models/wav2letter/__init__.py +11 -0
  2328. executorch/examples/models/wav2letter/model.py +29 -0
  2329. executorch/examples/models/yolo12/export_and_validate.py +397 -0
  2330. executorch/examples/nxp/experimental/cifar_net/cifar_net.py +270 -0
  2331. executorch/examples/xnnpack/__init__.py +51 -0
  2332. executorch/examples/xnnpack/aot_compiler.py +148 -0
  2333. executorch/examples/xnnpack/quantization/example.py +205 -0
  2334. executorch/examples/xnnpack/quantization/utils.py +42 -0
  2335. executorch/exir/__init__.py +57 -0
  2336. executorch/exir/_serialize/__init__.py +20 -0
  2337. executorch/exir/_serialize/_cord.py +49 -0
  2338. executorch/exir/_serialize/_dataclass.py +145 -0
  2339. executorch/exir/_serialize/_flatbuffer.py +357 -0
  2340. executorch/exir/_serialize/_named_data_store.py +222 -0
  2341. executorch/exir/_serialize/_program.py +767 -0
  2342. executorch/exir/_serialize/_serialize.py +123 -0
  2343. executorch/exir/_serialize/data_serializer.py +80 -0
  2344. executorch/exir/_serialize/padding.py +39 -0
  2345. executorch/exir/_serialize/program.fbs +489 -0
  2346. executorch/exir/_serialize/scalar_type.fbs +43 -0
  2347. executorch/exir/_serialize/test/__init__.py +0 -0
  2348. executorch/exir/_serialize/test/test_cord.py +63 -0
  2349. executorch/exir/_serialize/test/test_flatbuffer.py +331 -0
  2350. executorch/exir/_serialize/test/test_named_data_store.py +212 -0
  2351. executorch/exir/_serialize/test/test_program.py +1265 -0
  2352. executorch/exir/_serialize/test/test_serialize.py +289 -0
  2353. executorch/exir/_warnings.py +44 -0
  2354. executorch/exir/backend/backend_api.py +794 -0
  2355. executorch/exir/backend/backend_details.py +150 -0
  2356. executorch/exir/backend/canonical_partitioners/all_node_partitioner.py +55 -0
  2357. executorch/exir/backend/canonical_partitioners/config_partitioner.py +262 -0
  2358. executorch/exir/backend/canonical_partitioners/duplicate_constant_node_pass.py +156 -0
  2359. executorch/exir/backend/canonical_partitioners/duplicate_dequant_node_pass.py +111 -0
  2360. executorch/exir/backend/canonical_partitioners/group_partitioner.py +371 -0
  2361. executorch/exir/backend/canonical_partitioners/pattern_op_partitioner.py +185 -0
  2362. executorch/exir/backend/compile_spec_schema.py +17 -0
  2363. executorch/exir/backend/operator_support.py +127 -0
  2364. executorch/exir/backend/partitioner.py +114 -0
  2365. executorch/exir/backend/test/backend_with_compiler_demo.py +142 -0
  2366. executorch/exir/backend/test/backend_with_delegate_mapping_demo.py +210 -0
  2367. executorch/exir/backend/test/backend_with_named_data_map.py +115 -0
  2368. executorch/exir/backend/test/backend_with_preprocess_all_demo.py +297 -0
  2369. executorch/exir/backend/test/demo_backend.py +35 -0
  2370. executorch/exir/backend/test/demos/rpc/executor_backend_partitioner.py +77 -0
  2371. executorch/exir/backend/test/demos/rpc/executor_backend_preprocess.py +39 -0
  2372. executorch/exir/backend/test/demos/rpc/test_rpc.py +189 -0
  2373. executorch/exir/backend/test/demos/test_delegate_aten_mode.py +72 -0
  2374. executorch/exir/backend/test/hta_partitioner_demo.py +288 -0
  2375. executorch/exir/backend/test/op_partitioner_demo.py +266 -0
  2376. executorch/exir/backend/test/test_backend_with_named_data_map.py +85 -0
  2377. executorch/exir/backend/test/test_backends.py +1463 -0
  2378. executorch/exir/backend/test/test_backends_lifted.py +1326 -0
  2379. executorch/exir/backend/test/test_backends_nested.py +258 -0
  2380. executorch/exir/backend/test/test_compatibility.py +118 -0
  2381. executorch/exir/backend/test/test_debug_handle_map.py +143 -0
  2382. executorch/exir/backend/test/test_delegate_map_builder.py +257 -0
  2383. executorch/exir/backend/test/test_graph_partition.py +248 -0
  2384. executorch/exir/backend/test/test_group_partitioner.py +1674 -0
  2385. executorch/exir/backend/test/test_lowered_backend_module.py +207 -0
  2386. executorch/exir/backend/test/test_partitioner.py +731 -0
  2387. executorch/exir/backend/test/test_passes.py +48 -0
  2388. executorch/exir/backend/test/test_to_backend_multi_method.py +595 -0
  2389. executorch/exir/backend/test/test_utils.py +264 -0
  2390. executorch/exir/backend/utils.py +661 -0
  2391. executorch/exir/capture/__init__.py +28 -0
  2392. executorch/exir/capture/_capture.py +381 -0
  2393. executorch/exir/capture/_config.py +119 -0
  2394. executorch/exir/capture/_unlift.py +159 -0
  2395. executorch/exir/common.py +140 -0
  2396. executorch/exir/control_flow.py +204 -0
  2397. executorch/exir/debug_handle_utils.py +29 -0
  2398. executorch/exir/delegate.py +171 -0
  2399. executorch/exir/delegate.pyi +21 -0
  2400. executorch/exir/dialects/__init__.py +0 -0
  2401. executorch/exir/dialects/_ops.py +166 -0
  2402. executorch/exir/dialects/backend/_ops.py +123 -0
  2403. executorch/exir/dialects/backend/test/test_backend_ops.py +88 -0
  2404. executorch/exir/dialects/edge/_ops.py +405 -0
  2405. executorch/exir/dialects/edge/arg/model.py +238 -0
  2406. executorch/exir/dialects/edge/arg/type.py +46 -0
  2407. executorch/exir/dialects/edge/dtype/runner.py +122 -0
  2408. executorch/exir/dialects/edge/dtype/supported.py +47 -0
  2409. executorch/exir/dialects/edge/dtype/utils.py +36 -0
  2410. executorch/exir/dialects/edge/edge.yaml +8388 -0
  2411. executorch/exir/dialects/edge/op/api.py +86 -0
  2412. executorch/exir/dialects/edge/op/sample_input.py +1283 -0
  2413. executorch/exir/dialects/edge/op/test/test_api.py +51 -0
  2414. executorch/exir/dialects/edge/spec/gen.py +523 -0
  2415. executorch/exir/dialects/edge/spec/utils.py +328 -0
  2416. executorch/exir/dialects/edge/test/test_edge_ops.py +443 -0
  2417. executorch/exir/dialects/edge/test/test_edge_yaml.py +225 -0
  2418. executorch/exir/dialects/test/test_exir_dialect_ops.py +22 -0
  2419. executorch/exir/dim_order_utils.py +85 -0
  2420. executorch/exir/dynamic_shape.py +12 -0
  2421. executorch/exir/emit/__init__.py +11 -0
  2422. executorch/exir/emit/_emit_program.py +216 -0
  2423. executorch/exir/emit/_emitter.py +2071 -0
  2424. executorch/exir/emit/test/test_emit.py +2405 -0
  2425. executorch/exir/error.py +55 -0
  2426. executorch/exir/graph.py +86 -0
  2427. executorch/exir/graph_module.py +153 -0
  2428. executorch/exir/lowered_backend_module.py +966 -0
  2429. executorch/exir/memory.py +50 -0
  2430. executorch/exir/memory_planning.py +1250 -0
  2431. executorch/exir/operator/convert.py +307 -0
  2432. executorch/exir/operator/manip.py +89 -0
  2433. executorch/exir/operator/test/test_operator.py +84 -0
  2434. executorch/exir/operator/util.py +68 -0
  2435. executorch/exir/pass_base.py +808 -0
  2436. executorch/exir/pass_manager.py +78 -0
  2437. executorch/exir/passes/__init__.py +528 -0
  2438. executorch/exir/passes/_quant_patterns_and_replacements.py +1563 -0
  2439. executorch/exir/passes/const_prop_pass.py +84 -0
  2440. executorch/exir/passes/constant_prop_pass.py +402 -0
  2441. executorch/exir/passes/debug_handle_generator_pass.py +81 -0
  2442. executorch/exir/passes/dim_order_ops_registry.py +90 -0
  2443. executorch/exir/passes/dynamic_shape_prop_pass.py +301 -0
  2444. executorch/exir/passes/executorch_prim_ops_registry.py +166 -0
  2445. executorch/exir/passes/external_constants_pass.py +113 -0
  2446. executorch/exir/passes/init_mutable_pass.py +32 -0
  2447. executorch/exir/passes/insert_write_back_for_buffers_pass.py +185 -0
  2448. executorch/exir/passes/memory_format_ops_pass.py +119 -0
  2449. executorch/exir/passes/memory_planning_pass.py +315 -0
  2450. executorch/exir/passes/normalize_transpose_pass.py +24 -0
  2451. executorch/exir/passes/normalize_view_copy_base_pass.py +62 -0
  2452. executorch/exir/passes/pass_registry.py +89 -0
  2453. executorch/exir/passes/prune_empty_tensors_pass.py +67 -0
  2454. executorch/exir/passes/quant_fusion_pass.py +174 -0
  2455. executorch/exir/passes/quantize_io_pass.py +414 -0
  2456. executorch/exir/passes/reinplace.py +103 -0
  2457. executorch/exir/passes/remove_graph_asserts_pass.py +62 -0
  2458. executorch/exir/passes/remove_mixed_type_operators.py +95 -0
  2459. executorch/exir/passes/remove_noop_pass.py +123 -0
  2460. executorch/exir/passes/remove_unused_parameters_pass.py +55 -0
  2461. executorch/exir/passes/replace_aten_with_edge_pass.py +59 -0
  2462. executorch/exir/passes/replace_broken_ops_with_function_ops_pass.py +42 -0
  2463. executorch/exir/passes/replace_edge_with_backend_pass.py +56 -0
  2464. executorch/exir/passes/replace_sym_size_op_pass.py +35 -0
  2465. executorch/exir/passes/replace_view_copy_with_view_pass.py +325 -0
  2466. executorch/exir/passes/scalar_to_tensor_pass.py +32 -0
  2467. executorch/exir/passes/spec_prop_pass.py +123 -0
  2468. executorch/exir/passes/sym_shape_eval_pass.py +289 -0
  2469. executorch/exir/passes/sym_to_tensor_pass.py +64 -0
  2470. executorch/exir/passes/weights_to_outputs_pass.py +88 -0
  2471. executorch/exir/print_program.py +420 -0
  2472. executorch/exir/program/__init__.py +32 -0
  2473. executorch/exir/program/_fake_program.py +67 -0
  2474. executorch/exir/program/_program.py +1995 -0
  2475. executorch/exir/program/test/__init__.py +0 -0
  2476. executorch/exir/program/test/test_fake_program.py +78 -0
  2477. executorch/exir/program/test/test_program.py +922 -0
  2478. executorch/exir/scalar_type.py +38 -0
  2479. executorch/exir/schema.py +308 -0
  2480. executorch/exir/serde/__init__.py +0 -0
  2481. executorch/exir/serde/export_serialize.py +2872 -0
  2482. executorch/exir/serde/schema.py +406 -0
  2483. executorch/exir/serde/schema_check.py +293 -0
  2484. executorch/exir/serde/serialize.py +838 -0
  2485. executorch/exir/serde/union.py +77 -0
  2486. executorch/exir/sym_util.py +95 -0
  2487. executorch/exir/tensor.py +392 -0
  2488. executorch/exir/tensor_layout.py +39 -0
  2489. executorch/exir/tests/asr_joiner.py +82 -0
  2490. executorch/exir/tests/common.py +96 -0
  2491. executorch/exir/tests/control_flow_models.py +105 -0
  2492. executorch/exir/tests/dynamic_shape_models.py +51 -0
  2493. executorch/exir/tests/models.py +576 -0
  2494. executorch/exir/tests/test_arg_validator.py +72 -0
  2495. executorch/exir/tests/test_capture.py +29 -0
  2496. executorch/exir/tests/test_common.py +81 -0
  2497. executorch/exir/tests/test_delegate.py +308 -0
  2498. executorch/exir/tests/test_dim_order_utils.py +49 -0
  2499. executorch/exir/tests/test_dynamic_shape_propagation.py +174 -0
  2500. executorch/exir/tests/test_error.py +19 -0
  2501. executorch/exir/tests/test_extract_io_quant_params.py +93 -0
  2502. executorch/exir/tests/test_joint_graph.py +124 -0
  2503. executorch/exir/tests/test_memory_format_ops_pass.py +443 -0
  2504. executorch/exir/tests/test_memory_format_ops_pass_aten.py +142 -0
  2505. executorch/exir/tests/test_memory_format_ops_pass_utils.py +266 -0
  2506. executorch/exir/tests/test_memory_planning.py +1135 -0
  2507. executorch/exir/tests/test_op_convert.py +88 -0
  2508. executorch/exir/tests/test_pass_infra.py +179 -0
  2509. executorch/exir/tests/test_passes.py +2332 -0
  2510. executorch/exir/tests/test_print_program.py +54 -0
  2511. executorch/exir/tests/test_prune_empty_tensors_pass.py +77 -0
  2512. executorch/exir/tests/test_quant_fusion_pass.py +528 -0
  2513. executorch/exir/tests/test_quantization.py +120 -0
  2514. executorch/exir/tests/test_quantize_io_pass.py +156 -0
  2515. executorch/exir/tests/test_reinplace_pass.py +124 -0
  2516. executorch/exir/tests/test_remove_unused_parameters_pass.py +205 -0
  2517. executorch/exir/tests/test_remove_view_copy.py +234 -0
  2518. executorch/exir/tests/test_serde.py +352 -0
  2519. executorch/exir/tests/test_tensor.py +341 -0
  2520. executorch/exir/tests/test_tracer.py +497 -0
  2521. executorch/exir/tests/test_verification.py +275 -0
  2522. executorch/exir/tests/test_warnings.py +169 -0
  2523. executorch/exir/tests/transformer.py +122 -0
  2524. executorch/exir/tracer.py +803 -0
  2525. executorch/exir/types.py +18 -0
  2526. executorch/exir/verification/arg_validator.py +140 -0
  2527. executorch/exir/verification/dev_html.py +68 -0
  2528. executorch/exir/verification/interpreter.py +425 -0
  2529. executorch/exir/verification/test/test_verifier.py +177 -0
  2530. executorch/exir/verification/verifier.py +352 -0
  2531. executorch/exir/version.py +9 -0
  2532. executorch/exir/wrap.py +32 -0
  2533. executorch/export/__init__.py +40 -0
  2534. executorch/export/export.py +696 -0
  2535. executorch/export/recipe.py +328 -0
  2536. executorch/export/recipe_provider.py +60 -0
  2537. executorch/export/recipe_registry.py +86 -0
  2538. executorch/export/stages.py +686 -0
  2539. executorch/export/target_recipes.py +185 -0
  2540. executorch/export/tests/test_executorch_export.py +34 -0
  2541. executorch/export/tests/test_export_recipe.py +131 -0
  2542. executorch/export/tests/test_export_session.py +973 -0
  2543. executorch/export/tests/test_export_stages.py +534 -0
  2544. executorch/export/tests/test_recipe_provider.py +98 -0
  2545. executorch/export/tests/test_recipe_registry.py +139 -0
  2546. executorch/export/tests/test_target_recipes.py +539 -0
  2547. executorch/export/types.py +22 -0
  2548. executorch/export/utils.py +51 -0
  2549. executorch/extension/audio/mel_spectrogram.py +279 -0
  2550. executorch/extension/export_util/__init__.py +13 -0
  2551. executorch/extension/export_util/utils.py +147 -0
  2552. executorch/extension/flat_tensor/__init__.py +0 -0
  2553. executorch/extension/flat_tensor/serialize/__init__.py +0 -0
  2554. executorch/extension/flat_tensor/serialize/flat_tensor.fbs +73 -0
  2555. executorch/extension/flat_tensor/serialize/flat_tensor_schema.py +34 -0
  2556. executorch/extension/flat_tensor/serialize/scalar_type.fbs +38 -0
  2557. executorch/extension/flat_tensor/serialize/serialize.py +446 -0
  2558. executorch/extension/flat_tensor/test/test_serialize.py +326 -0
  2559. executorch/extension/gguf_util/convert_main.py +53 -0
  2560. executorch/extension/gguf_util/converter.py +27 -0
  2561. executorch/extension/gguf_util/converters/llama_converter.py +123 -0
  2562. executorch/extension/gguf_util/load_gguf.py +103 -0
  2563. executorch/extension/llm/custom_ops/__init__.py +0 -0
  2564. executorch/extension/llm/custom_ops/custom_ops.py +495 -0
  2565. executorch/extension/llm/custom_ops/custom_ops_aot_lib.dll +0 -0
  2566. executorch/extension/llm/custom_ops/model_sharding.py +117 -0
  2567. executorch/extension/llm/custom_ops/op_tile_crop_aot.py +38 -0
  2568. executorch/extension/llm/custom_ops/preprocess_custom_ops.py +69 -0
  2569. executorch/extension/llm/custom_ops/spinquant/special_hadamard_code_gen.py +279 -0
  2570. executorch/extension/llm/custom_ops/spinquant/third-party/FFHT/example.py +20 -0
  2571. executorch/extension/llm/custom_ops/spinquant/third-party/FFHT/gen.py +869 -0
  2572. executorch/extension/llm/custom_ops/test_preprocess_custom_ops.py +34 -0
  2573. executorch/extension/llm/custom_ops/test_quantized_sdpa.py +550 -0
  2574. executorch/extension/llm/custom_ops/test_sdpa_with_kv_cache.py +651 -0
  2575. executorch/extension/llm/custom_ops/test_update_cache.py +433 -0
  2576. executorch/extension/llm/custom_ops/test_update_cross_attn_cache.py +284 -0
  2577. executorch/extension/llm/export/__init__.py +11 -0
  2578. executorch/extension/llm/export/builder.py +535 -0
  2579. executorch/extension/llm/export/config/llm_config.py +727 -0
  2580. executorch/extension/llm/export/export_llm.py +106 -0
  2581. executorch/extension/llm/export/export_passes.py +202 -0
  2582. executorch/extension/llm/export/partitioner_lib.py +238 -0
  2583. executorch/extension/llm/export/quantizer_lib.py +322 -0
  2584. executorch/extension/llm/export/test/__init__.py +0 -0
  2585. executorch/extension/llm/export/test/test_builder.py +118 -0
  2586. executorch/extension/llm/export/test/test_export_llm.py +155 -0
  2587. executorch/extension/llm/export/test_export_passes.py +209 -0
  2588. executorch/extension/llm/modules/__init__.py +24 -0
  2589. executorch/extension/llm/modules/_position_embeddings.py +729 -0
  2590. executorch/extension/llm/modules/attention.py +415 -0
  2591. executorch/extension/llm/modules/kv_cache.py +151 -0
  2592. executorch/extension/llm/modules/test/__init__.py +0 -0
  2593. executorch/extension/llm/modules/test/test_attention.py +353 -0
  2594. executorch/extension/llm/modules/test/test_kv_cache.py +218 -0
  2595. executorch/extension/llm/modules/test/test_position_embeddings.py +230 -0
  2596. executorch/extension/llm/runner/__init__.py +235 -0
  2597. executorch/extension/llm/runner/_llm_runner.pyi +523 -0
  2598. executorch/extension/llm/runner/test/test_runner_pybindings.py +258 -0
  2599. executorch/extension/llm/tokenizers/pytorch_tokenizers/__init__.py +63 -0
  2600. executorch/extension/llm/tokenizers/pytorch_tokenizers/constants.py +31 -0
  2601. executorch/extension/llm/tokenizers/pytorch_tokenizers/hf_tokenizer.py +57 -0
  2602. executorch/extension/llm/tokenizers/pytorch_tokenizers/llama2c.py +111 -0
  2603. executorch/extension/llm/tokenizers/pytorch_tokenizers/tiktoken.py +198 -0
  2604. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/__init__.py +0 -0
  2605. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/llama2c/__init__.py +0 -0
  2606. executorch/extension/llm/tokenizers/pytorch_tokenizers/tools/llama2c/convert.py +49 -0
  2607. executorch/extension/llm/tokenizers/setup.py +199 -0
  2608. executorch/extension/llm/tokenizers/test/test_hf_tokenizer.py +83 -0
  2609. executorch/extension/llm/tokenizers/test/test_python_bindings.py +131 -0
  2610. executorch/extension/llm/tokenizers/test/test_tekken_python.py +453 -0
  2611. executorch/extension/llm/tokenizers/test/test_tiktoken.py +49 -0
  2612. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/abseil.podspec.gen.py +243 -0
  2613. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/copts/copts.py +194 -0
  2614. executorch/extension/llm/tokenizers/third-party/abseil-cpp/absl/copts/generate_copts.py +109 -0
  2615. executorch/extension/llm/tokenizers/third-party/abseil-cpp/conanfile.py +51 -0
  2616. executorch/extension/llm/tokenizers/third-party/abseil-cpp/create_lts.py +139 -0
  2617. executorch/extension/llm/tokenizers/third-party/json/docs/mkdocs/scripts/check_structure.py +179 -0
  2618. executorch/extension/llm/tokenizers/third-party/json/tests/thirdparty/imapdl/filterbr.py +109 -0
  2619. executorch/extension/llm/tokenizers/third-party/json/tools/amalgamate/amalgamate.py +299 -0
  2620. executorch/extension/llm/tokenizers/third-party/json/tools/gdb_pretty_printer/nlohmann-json.py +32 -0
  2621. executorch/extension/llm/tokenizers/third-party/json/tools/generate_natvis/generate_natvis.py +41 -0
  2622. executorch/extension/llm/tokenizers/third-party/json/tools/serve_header/serve_header.py +410 -0
  2623. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateCommon.py +354 -0
  2624. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateTest.py +186 -0
  2625. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcd.py +999 -0
  2626. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcpHeader.py +98 -0
  2627. executorch/extension/llm/tokenizers/third-party/pcre2/maint/GenerateUcpTables.py +203 -0
  2628. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateCommon.py +31 -0
  2629. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateDates.py +59 -0
  2630. executorch/extension/llm/tokenizers/third-party/pcre2/maint/UpdateRelease.py +29 -0
  2631. executorch/extension/llm/tokenizers/third-party/re2/benchlog/benchplot.py +98 -0
  2632. executorch/extension/llm/tokenizers/third-party/re2/python/re2.py +583 -0
  2633. executorch/extension/llm/tokenizers/third-party/re2/python/re2_test.py +489 -0
  2634. executorch/extension/llm/tokenizers/third-party/re2/python/setup.py +158 -0
  2635. executorch/extension/llm/tokenizers/third-party/re2/python/toolchains/generate.py +98 -0
  2636. executorch/extension/llm/tokenizers/third-party/re2/re2/make_unicode_casefold.py +151 -0
  2637. executorch/extension/llm/tokenizers/third-party/re2/re2/make_unicode_groups.py +117 -0
  2638. executorch/extension/llm/tokenizers/third-party/re2/re2/unicode.py +284 -0
  2639. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/setup.py +215 -0
  2640. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/__init__.py +1224 -0
  2641. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/_version.py +1 -0
  2642. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/sentencepiece_model_pb2.py +44 -0
  2643. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/src/sentencepiece/sentencepiece_pb2.py +30 -0
  2644. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/test/__init__.py +0 -0
  2645. executorch/extension/llm/tokenizers/third-party/sentencepiece/python/test/sentencepiece_test.py +937 -0
  2646. executorch/extension/module/test/resources/gen_bundled_program.py +99 -0
  2647. executorch/extension/pybindings/_portable_lib.cp313-win_amd64.pyd +0 -0
  2648. executorch/extension/pybindings/portable_lib.py +82 -0
  2649. executorch/extension/pybindings/pybindings.pyi +299 -0
  2650. executorch/extension/pybindings/test/make_test.py +186 -0
  2651. executorch/extension/pybindings/test/test_backend_pybinding.py +27 -0
  2652. executorch/extension/pybindings/test/test_pybindings.py +735 -0
  2653. executorch/extension/pytree/__init__.py +54 -0
  2654. executorch/extension/pytree/test/test.py +331 -0
  2655. executorch/extension/training/__init__.py +22 -0
  2656. executorch/extension/training/examples/CIFAR/data_utils.py +389 -0
  2657. executorch/extension/training/examples/CIFAR/export.py +220 -0
  2658. executorch/extension/training/examples/CIFAR/main.py +188 -0
  2659. executorch/extension/training/examples/CIFAR/model.py +73 -0
  2660. executorch/extension/training/examples/CIFAR/train_utils.py +622 -0
  2661. executorch/extension/training/examples/XOR/export_model.py +74 -0
  2662. executorch/extension/training/examples/XOR/model.py +34 -0
  2663. executorch/extension/training/examples/XOR/test/test_export.py +19 -0
  2664. executorch/extension/training/pybindings/_training_lib.pyi +45 -0
  2665. executorch/extension/training/pybindings/_training_module.py +86 -0
  2666. executorch/extension/training/pybindings/test/test.py +67 -0
  2667. executorch/extension/wasm/test/test_model.py +34 -0
  2668. executorch/include/executorch/extension/kernel_util/make_boxed_from_unboxed_functor.h +260 -0
  2669. executorch/include/executorch/extension/kernel_util/meta_programming.h +119 -0
  2670. executorch/include/executorch/extension/kernel_util/type_list.h +145 -0
  2671. executorch/include/executorch/extension/tensor/tensor.h +14 -0
  2672. executorch/include/executorch/extension/tensor/tensor_accessor.h +218 -0
  2673. executorch/include/executorch/extension/tensor/tensor_ptr.h +485 -0
  2674. executorch/include/executorch/extension/tensor/tensor_ptr_maker.h +687 -0
  2675. executorch/include/executorch/extension/threadpool/cpuinfo_utils.h +24 -0
  2676. executorch/include/executorch/extension/threadpool/threadpool.h +124 -0
  2677. executorch/include/executorch/extension/threadpool/threadpool_guard.h +37 -0
  2678. executorch/include/executorch/runtime/core/array_ref.h +262 -0
  2679. executorch/include/executorch/runtime/core/data_loader.h +141 -0
  2680. executorch/include/executorch/runtime/core/defines.h +20 -0
  2681. executorch/include/executorch/runtime/core/error.h +257 -0
  2682. executorch/include/executorch/runtime/core/evalue.h +608 -0
  2683. executorch/include/executorch/runtime/core/event_tracer.h +592 -0
  2684. executorch/include/executorch/runtime/core/event_tracer_hooks.h +340 -0
  2685. executorch/include/executorch/runtime/core/event_tracer_hooks_delegate.h +204 -0
  2686. executorch/include/executorch/runtime/core/exec_aten/exec_aten.h +175 -0
  2687. executorch/include/executorch/runtime/core/exec_aten/testing_util/tensor_factory.h +1133 -0
  2688. executorch/include/executorch/runtime/core/exec_aten/testing_util/tensor_util.h +378 -0
  2689. executorch/include/executorch/runtime/core/exec_aten/util/dim_order_util.h +278 -0
  2690. executorch/include/executorch/runtime/core/exec_aten/util/scalar_type_util.h +1430 -0
  2691. executorch/include/executorch/runtime/core/exec_aten/util/tensor_dimension_limit.h +21 -0
  2692. executorch/include/executorch/runtime/core/exec_aten/util/tensor_shape_to_c_string.h +68 -0
  2693. executorch/include/executorch/runtime/core/exec_aten/util/tensor_util.h +1415 -0
  2694. executorch/include/executorch/runtime/core/freeable_buffer.h +218 -0
  2695. executorch/include/executorch/runtime/core/function_ref.h +108 -0
  2696. executorch/include/executorch/runtime/core/hierarchical_allocator.h +125 -0
  2697. executorch/include/executorch/runtime/core/memory_allocator.h +233 -0
  2698. executorch/include/executorch/runtime/core/named_data_map.h +77 -0
  2699. executorch/include/executorch/runtime/core/portable_type/bfloat16.h +27 -0
  2700. executorch/include/executorch/runtime/core/portable_type/bfloat16_math.h +14 -0
  2701. executorch/include/executorch/runtime/core/portable_type/bits_types.h +83 -0
  2702. executorch/include/executorch/runtime/core/portable_type/c10/c10/macros/Export.h +1 -0
  2703. executorch/include/executorch/runtime/core/portable_type/c10/c10/macros/Macros.h +1 -0
  2704. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-inl.h +1 -0
  2705. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-math.h +299 -0
  2706. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16.h +1 -0
  2707. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/Half-inl.h +1 -0
  2708. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/Half.h +8 -0
  2709. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/TypeSafeSignMath.h +1 -0
  2710. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/bit_cast.h +1 -0
  2711. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex.h +78 -0
  2712. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex_math.h +406 -0
  2713. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/complex_utils.h +46 -0
  2714. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/floating_point_utils.h +1 -0
  2715. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/irange.h +123 -0
  2716. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h +905 -0
  2717. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/overflows.h +100 -0
  2718. executorch/include/executorch/runtime/core/portable_type/c10/c10/util/safe_numerics.h +114 -0
  2719. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Export.h +153 -0
  2720. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h +694 -0
  2721. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/BFloat16.h +480 -0
  2722. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/Half.h +788 -0
  2723. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h +148 -0
  2724. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/bit_cast.h +50 -0
  2725. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/complex.h +616 -0
  2726. executorch/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/floating_point_utils.h +38 -0
  2727. executorch/include/executorch/runtime/core/portable_type/complex.h +21 -0
  2728. executorch/include/executorch/runtime/core/portable_type/device.h +74 -0
  2729. executorch/include/executorch/runtime/core/portable_type/half.h +27 -0
  2730. executorch/include/executorch/runtime/core/portable_type/optional.h +36 -0
  2731. executorch/include/executorch/runtime/core/portable_type/qint_types.h +83 -0
  2732. executorch/include/executorch/runtime/core/portable_type/scalar.h +123 -0
  2733. executorch/include/executorch/runtime/core/portable_type/scalar_type.h +154 -0
  2734. executorch/include/executorch/runtime/core/portable_type/string_view.h +29 -0
  2735. executorch/include/executorch/runtime/core/portable_type/tensor.h +175 -0
  2736. executorch/include/executorch/runtime/core/portable_type/tensor_impl.h +304 -0
  2737. executorch/include/executorch/runtime/core/portable_type/tensor_options.h +60 -0
  2738. executorch/include/executorch/runtime/core/result.h +268 -0
  2739. executorch/include/executorch/runtime/core/span.h +110 -0
  2740. executorch/include/executorch/runtime/core/tag.h +90 -0
  2741. executorch/include/executorch/runtime/core/tensor_layout.h +93 -0
  2742. executorch/include/executorch/runtime/core/tensor_shape_dynamism.h +39 -0
  2743. executorch/include/executorch/runtime/core/testing_util/error_matchers.h +303 -0
  2744. executorch/include/executorch/runtime/kernel/kernel_includes.h +23 -0
  2745. executorch/include/executorch/runtime/kernel/kernel_runtime_context.h +131 -0
  2746. executorch/include/executorch/runtime/kernel/operator_registry.h +298 -0
  2747. executorch/include/executorch/runtime/kernel/test/test_util.h +40 -0
  2748. executorch/include/executorch/runtime/kernel/thread_parallel_interface.h +107 -0
  2749. executorch/include/executorch/runtime/platform/abort.h +36 -0
  2750. executorch/include/executorch/runtime/platform/assert.h +124 -0
  2751. executorch/include/executorch/runtime/platform/clock.h +43 -0
  2752. executorch/include/executorch/runtime/platform/compat_unistd.h +75 -0
  2753. executorch/include/executorch/runtime/platform/compiler.h +214 -0
  2754. executorch/include/executorch/runtime/platform/log.h +216 -0
  2755. executorch/include/executorch/runtime/platform/platform.h +276 -0
  2756. executorch/include/executorch/runtime/platform/profiler.h +289 -0
  2757. executorch/include/executorch/runtime/platform/runtime.h +35 -0
  2758. executorch/include/executorch/runtime/platform/system.h +49 -0
  2759. executorch/include/executorch/runtime/platform/test/pal_spy.h +86 -0
  2760. executorch/include/executorch/runtime/platform/test/stub_platform.h +76 -0
  2761. executorch/include/executorch/runtime/platform/types.h +24 -0
  2762. executorch/kernels/quantized/__init__.py +21 -0
  2763. executorch/kernels/quantized/quantized.yaml +113 -0
  2764. executorch/kernels/quantized/quantized_ops_aot_lib.dll +0 -0
  2765. executorch/kernels/quantized/test/supported_features_def.yaml +1 -0
  2766. executorch/kernels/quantized/test/test_out_variants.py +100 -0
  2767. executorch/kernels/quantized/test/test_quant_dequant_per_token.py +149 -0
  2768. executorch/runtime/__init__.py +334 -0
  2769. executorch/runtime/kernel/test/functions.yaml +13 -0
  2770. executorch/runtime/test/test_runtime.py +106 -0
  2771. executorch/runtime/test/test_runtime_etdump_gen.py +163 -0
  2772. executorch/schema/program.fbs +489 -0
  2773. executorch/schema/scalar_type.fbs +43 -0
  2774. executorch/share/cmake/executorch-config.cmake +77 -0
  2775. executorch/util/activation_memory_profiler.py +177 -0
  2776. executorch/util/collect_env.py +748 -0
  2777. executorch/util/python_profiler.py +96 -0
  2778. executorch/version.py +4 -0
  2779. executorch-1.1.0.dist-info/METADATA +87 -0
  2780. executorch-1.1.0.dist-info/RECORD +2784 -0
  2781. executorch-1.1.0.dist-info/WHEEL +5 -0
  2782. executorch-1.1.0.dist-info/entry_points.txt +2 -0
  2783. executorch-1.1.0.dist-info/licenses/LICENSE +37 -0
  2784. executorch-1.1.0.dist-info/top_level.txt +7 -0
@@ -0,0 +1,2210 @@
1
+ # Copyright (c) Meta Platforms, Inc. and affiliates.
2
+ # All rights reserved.
3
+ #
4
+ # This source code is licensed under the BSD-style license found in the
5
+ # LICENSE file in the root directory of this source tree.
6
+
7
+ # pyre-strict
8
+
9
+ from typing import Callable, Protocol, TypeVar
10
+
11
+ import torch
12
+ import torch.nn as nn
13
+ import torch.nn.functional as F
14
+ from executorch.exir.scalar_type import ScalarType
15
+ from torch.library import impl, Library
16
+
17
+ m = Library("cadence", "IMPL", "CompositeExplicitAutograd")
18
+
19
+ try:
20
+ torch.ops.load_library("//executorch/kernels/quantized:custom_ops_generated_lib")
21
+ except (OSError, RuntimeError):
22
+ # Fall back to path-based loading for CMake/OSS builds
23
+ from pathlib import Path
24
+
25
+ custom_libs: list[Path] = list(
26
+ Path(__file__)
27
+ .parent.parent.parent.resolve()
28
+ .glob("**/kernels/quantized/**/*custom_ops_generated_lib.*")
29
+ )
30
+ if custom_libs:
31
+ torch.ops.load_library(str(custom_libs[0]))
32
+ del Path
33
+
34
+ # Registry to track all ops with reference implementations
35
+ _REGISTERED_REF_IMPLEMENTATIONS: set[str] = set()
36
+
37
+ T = TypeVar("T", bound=Callable[..., torch.Tensor | tuple[torch.Tensor, ...]])
38
+
39
+
40
+ class MyDecorator(Protocol):
41
+ def __call__(self, __f: T) -> T: ...
42
+
43
+
44
+ # Custom impl wrapper that tracks registrations
45
+ def impl_tracked(lib: Library, op_name: str) -> MyDecorator:
46
+ """Wrapper around impl that tracks registered ops."""
47
+ _REGISTERED_REF_IMPLEMENTATIONS.add(op_name)
48
+ return impl(lib, op_name)
49
+
50
+
51
+ def get_registered_ref_implementations() -> set[str]:
52
+ """Get all ops that have reference implementations."""
53
+ return _REGISTERED_REF_IMPLEMENTATIONS.copy()
54
+
55
+
56
+ qdtype_map: dict[ScalarType, torch.dtype] = {
57
+ ScalarType.QINT8: torch.qint8,
58
+ ScalarType.QUINT8: torch.quint8,
59
+ ScalarType.QINT32: torch.qint32,
60
+ }
61
+
62
+
63
+ def quantize_per_tensor_common(
64
+ input_tensor: torch.Tensor,
65
+ scale: float,
66
+ zero_point: int,
67
+ quant_min: int,
68
+ quant_max: int,
69
+ dtype: torch.dtype,
70
+ ) -> torch.Tensor:
71
+ """
72
+ Quantizes a floating-point tensor to an integral tensor.
73
+
74
+ Args:
75
+ - input_tensor (Tensor): input tensor
76
+ - scale (float): Quantization scale. Derived from the ratio
77
+ between the min/max of the floating-point tensor and the
78
+ min/max of the quantized range, and then inverted.
79
+ - zero_point (int): The point which represents 0 in the quantized
80
+ range. For example, consider the floating point range [-1., 2.] and
81
+ quantized integer range [-7, 7]. In this case, 0 is 1/3 of way from
82
+ -1. to 2. So, the point that represents 0 in the quantized range should
83
+ be 1/3 of the way from [-7, 7]. This ends up being -2 in the integer space.
84
+ - quant_min (int): The smallest value in the quantized domain. Unused since scale
85
+ is already provided.
86
+ - quant_max (int): The largest value in the quantized domain. Unused since scale
87
+ is already provided.
88
+ - dtype (torch.dtype): The type of the output tensor
89
+ """
90
+ supported_quant_types = [
91
+ torch.int8,
92
+ torch.int16,
93
+ torch.int32,
94
+ torch.uint8,
95
+ torch.uint16,
96
+ ]
97
+ if dtype not in supported_quant_types:
98
+ raise ValueError(
99
+ f"Unsupported dtype to quantize to {dtype}. Supported dtypes must be one of {supported_quant_types}"
100
+ )
101
+
102
+ return torch.ops.quantized_decomposed.quantize_per_tensor(
103
+ input_tensor,
104
+ scale,
105
+ zero_point,
106
+ quant_min,
107
+ quant_max,
108
+ dtype,
109
+ )
110
+
111
+
112
+ def quantize_per_tensor_variant(
113
+ dtype: torch.dtype | None = None,
114
+ ) -> Callable[[Callable[..., torch.Tensor]], Callable[..., torch.Tensor]]:
115
+ """Create a quantize_per_tensor variant with type checking."""
116
+
117
+ def decorator(_: Callable[..., torch.Tensor]) -> Callable[..., torch.Tensor]:
118
+ def variant(
119
+ input_tensor: torch.Tensor,
120
+ scale: float,
121
+ zero_point: int,
122
+ quant_min: int,
123
+ quant_max: int,
124
+ out_dtype: torch.dtype,
125
+ ) -> torch.Tensor:
126
+ if dtype and out_dtype != dtype:
127
+ raise ValueError(f"dtype must be {dtype}. Got {out_dtype}")
128
+
129
+ return quantize_per_tensor_common(
130
+ input_tensor,
131
+ scale,
132
+ zero_point,
133
+ quant_min,
134
+ quant_max,
135
+ out_dtype,
136
+ )
137
+
138
+ return variant
139
+
140
+ return decorator
141
+
142
+
143
+ @impl_tracked(m, "quantize_per_tensor")
144
+ @quantize_per_tensor_variant()
145
+ def quantize_per_tensor() -> torch.Tensor: ...
146
+
147
+
148
+ @impl_tracked(m, "quantize_per_tensor_asym8u")
149
+ @quantize_per_tensor_variant(torch.uint8)
150
+ def quantize_per_tensor_asym8u() -> torch.Tensor: ...
151
+
152
+
153
+ @impl_tracked(m, "quantize_per_tensor_asym8s")
154
+ @quantize_per_tensor_variant(torch.int8)
155
+ def quantize_per_tensor_asym8s() -> torch.Tensor: ...
156
+
157
+
158
+ @impl_tracked(m, "quantize_per_tensor_asym16u")
159
+ @quantize_per_tensor_variant(torch.uint16)
160
+ def quantize_per_tensor_asym16u() -> torch.Tensor: ...
161
+
162
+
163
+ @impl_tracked(m, "quantize_per_tensor_asym16s")
164
+ @quantize_per_tensor_variant(torch.int16)
165
+ def quantize_per_tensor_asym16s() -> torch.Tensor: ...
166
+
167
+
168
+ @impl_tracked(m, "quantize_per_tensor_asym32s")
169
+ @quantize_per_tensor_variant(torch.int32)
170
+ def quantize_per_tensor_asym32s() -> torch.Tensor: ...
171
+
172
+
173
+ def dequantize_per_tensor_common(
174
+ input_tensor: torch.Tensor,
175
+ scale: float,
176
+ zero_point: int,
177
+ quant_min: int,
178
+ quant_max: int,
179
+ dtype: torch.dtype,
180
+ ) -> torch.Tensor:
181
+ """
182
+ Dequantizes an integral tensor to a floating-point tensor.
183
+
184
+ Args:
185
+ - input (Tensor): input tensor
186
+ - scale (float): Quantization scale. Derived from the ratio
187
+ between the min/max of the floating-point tensor and the
188
+ min/max of the quantized range.
189
+ - zero_point (int): The point which represents 0 in the quantized
190
+ range. For example, consider the floating point range [-1., 2.] and
191
+ quantized integer range [-7, 7]. In this case, 0 is 1/3 of way from
192
+ -1. to 2. So, the point that represents 0 in the quantized range should
193
+ be 1/3 of the way from [-7, 7]. This ends up being -2 in the integer space.
194
+ - quant_min (int): The smallest value in the quantized domain. Unused since scale
195
+ is already provided.
196
+ - quant_max (int): The largest value in the quantized domain. Unused since scale
197
+ is already provided.
198
+ - dtype (torch.dtype): The type of the input tensor.
199
+ """
200
+ supported_quant_types = [
201
+ torch.int8,
202
+ torch.int16,
203
+ torch.int32,
204
+ torch.uint8,
205
+ torch.uint16,
206
+ ]
207
+ if input_tensor.dtype not in supported_quant_types:
208
+ raise ValueError(f"Input dtype must be one of {supported_quant_types}")
209
+ if input_tensor.dtype != dtype:
210
+ raise ValueError("Input dtype must match dtype")
211
+
212
+ return torch.ops.quantized_decomposed.dequantize_per_tensor(
213
+ input_tensor, scale, zero_point, quant_min, quant_max, dtype
214
+ )
215
+
216
+
217
+ def dequantize_per_tensor_variant(
218
+ dtype: torch.dtype | None = None,
219
+ ) -> Callable[[Callable[..., torch.Tensor]], Callable[..., torch.Tensor]]:
220
+ """Create a dequantize_per_tensor variant with type checking."""
221
+
222
+ def decorator(_: Callable[..., torch.Tensor]) -> Callable[..., torch.Tensor]:
223
+ def variant(
224
+ input_tensor: torch.Tensor,
225
+ scale: float,
226
+ zero_point: int,
227
+ quant_min: int,
228
+ quant_max: int,
229
+ in_dtype: torch.dtype,
230
+ ) -> torch.Tensor:
231
+ if dtype and in_dtype != dtype:
232
+ raise ValueError(f"dtype must be {dtype}. Got {in_dtype}")
233
+
234
+ return dequantize_per_tensor_common(
235
+ input_tensor,
236
+ scale,
237
+ zero_point,
238
+ quant_min,
239
+ quant_max,
240
+ in_dtype,
241
+ )
242
+
243
+ return variant
244
+
245
+ return decorator
246
+
247
+
248
+ @impl_tracked(m, "dequantize_per_tensor")
249
+ @dequantize_per_tensor_variant()
250
+ def dequantize_per_tensor() -> torch.Tensor: ...
251
+
252
+
253
+ @impl_tracked(m, "dequantize_per_tensor_asym8u")
254
+ @dequantize_per_tensor_variant(torch.uint8)
255
+ def dequantize_per_tensor_asym8u() -> torch.Tensor: ...
256
+
257
+
258
+ @impl_tracked(m, "dequantize_per_tensor_asym32s")
259
+ @dequantize_per_tensor_variant(torch.int32)
260
+ def dequantize_per_tensor_asym32s() -> torch.Tensor: ...
261
+
262
+
263
+ @impl_tracked(m, "dequantize_per_tensor_asym16u")
264
+ @dequantize_per_tensor_variant(torch.uint16)
265
+ def dequantize_per_tensor_asym16u() -> torch.Tensor: ...
266
+
267
+
268
+ @impl_tracked(m, "dequantize_per_tensor_asym8s")
269
+ @dequantize_per_tensor_variant(torch.int8)
270
+ def dequantize_per_tensor_asym8s() -> torch.Tensor: ...
271
+
272
+
273
+ @impl_tracked(m, "dequantize_per_tensor_asym16s")
274
+ @dequantize_per_tensor_variant(torch.int16)
275
+ def dequantize_per_tensor_asym16s() -> torch.Tensor: ...
276
+
277
+
278
+ @impl_tracked(m, "quantized_add.per_tensor")
279
+ def quantized_add_per_tensor(
280
+ X: torch.Tensor,
281
+ X_scale: float,
282
+ X_zero_point: int,
283
+ Y: torch.Tensor,
284
+ Y_scale: float,
285
+ Y_zero_point: int,
286
+ out_scale: float,
287
+ out_zero_point: int,
288
+ ) -> torch.Tensor:
289
+ """
290
+ Sums up two quantized tensors and returns another quantized tensor. The intuition
291
+ is that we want dequant(out) ~= dequant(X) + dequant(Y)
292
+
293
+ If we do that math, we get
294
+ out_scale(out - out_zero_point) = X_scale(X - X_zero_point) + Y_scale(Y - Y_zero_point)
295
+
296
+ Rearranging, we get
297
+ out = (X_scale(X - X_zero_point) + Y_scale(Y - Y_zero_point)) / out_scale + out_zero_point
298
+
299
+ Args:
300
+ - X: The first operand
301
+ - X_scale: The ratio between the sizes of X's floating point and quantized
302
+ ranges
303
+ - X_zero_point: The quantized mapping of zero for X
304
+ - Y: The second operand
305
+ - Y_scale: The ratio between the sizes of Y's floating point and quantized
306
+ ranges
307
+ - Y_zero_point: The quantized mapping of zero for Y
308
+ - out_scale: The ratio between the sizes of the output's floating point and
309
+ quantized ranges
310
+ - out_zero_point: The quantized mapping of zero for the output
311
+ """
312
+ supported_dtypes = [torch.int8, torch.uint8]
313
+ if X.dtype != Y.dtype:
314
+ raise ValueError("X and Y dtypes need to match")
315
+
316
+ dtype = X.dtype
317
+ if dtype not in supported_dtypes:
318
+ raise ValueError(
319
+ f"X and Y dtypes need to be in {supported_dtypes}. Got {dtype}"
320
+ )
321
+
322
+ if dtype == torch.uint8:
323
+ X = X.to(torch.int8)
324
+ Y = Y.to(torch.int8)
325
+
326
+ # TODO(agrebenisan): This should be done in fixed point arithmetic, but to match the quantized_add_out.cpp
327
+ # reference implementation, we'll do it in floating point.
328
+ dequant_X = X_scale * (X - X_zero_point)
329
+ dequant_Y = Y_scale * (Y - Y_zero_point)
330
+
331
+ # q_min/q_max are unused args
332
+ return quantize_per_tensor(
333
+ dequant_X + dequant_Y,
334
+ out_scale,
335
+ out_zero_point,
336
+ torch.iinfo(dtype).min,
337
+ torch.iinfo(dtype).max,
338
+ dtype,
339
+ )
340
+
341
+
342
+ @impl_tracked(m, "quantized_add")
343
+ def quantized_add(
344
+ X: torch.Tensor,
345
+ X_scale: torch.Tensor,
346
+ X_zero_point: torch.Tensor,
347
+ Y: torch.Tensor,
348
+ Y_scale: torch.Tensor,
349
+ Y_zero_point: torch.Tensor,
350
+ out_scale: float,
351
+ out_zero_point: int,
352
+ ) -> torch.Tensor:
353
+ return quantized_add_per_tensor(
354
+ X,
355
+ float(X_scale.item()),
356
+ int(X_zero_point.item()),
357
+ Y,
358
+ float(Y_scale.item()),
359
+ int(Y_zero_point.item()),
360
+ out_scale,
361
+ out_zero_point,
362
+ )
363
+
364
+
365
+ @impl_tracked(m, "quantized_add_asym8sxasym8s_asym8s.per_tensor")
366
+ def quantized_add_asym8sxasym8s_asym8s_per_tensor(
367
+ X: torch.Tensor,
368
+ X_scale: float,
369
+ X_zero_point: int,
370
+ Y: torch.Tensor,
371
+ Y_scale: float,
372
+ Y_zero_point: int,
373
+ out_scale: float,
374
+ out_zero_point: int,
375
+ ) -> torch.Tensor:
376
+ if X.dtype != torch.int8:
377
+ raise ValueError("X dtype must be torch.int8")
378
+ if Y.dtype != torch.int8:
379
+ raise ValueError("Y dtype must be torch.int8")
380
+
381
+ return quantized_add_per_tensor(
382
+ X, X_scale, X_zero_point, Y, Y_scale, Y_zero_point, out_scale, out_zero_point
383
+ )
384
+
385
+
386
+ @impl_tracked(m, "quantized_add_asym8uxasym8u_asym8u.per_tensor")
387
+ def quantized_add_asym8uxasym8u_asym8u_per_tensor(
388
+ X: torch.Tensor,
389
+ X_scale: float,
390
+ X_zero_point: int,
391
+ Y: torch.Tensor,
392
+ Y_scale: float,
393
+ Y_zero_point: int,
394
+ out_scale: float,
395
+ out_zero_point: int,
396
+ ) -> torch.Tensor:
397
+ if X.dtype != torch.uint8:
398
+ raise ValueError("X dtype must be torch.int8")
399
+ if Y.dtype != torch.uint8:
400
+ raise ValueError("Y dtype must be torch.int8")
401
+
402
+ return quantized_add_per_tensor(
403
+ X, X_scale, X_zero_point, Y, Y_scale, Y_zero_point, out_scale, out_zero_point
404
+ )
405
+
406
+
407
+ def quantized_linear_common(
408
+ src: torch.Tensor,
409
+ weight: torch.Tensor,
410
+ bias: torch.Tensor,
411
+ in_zero_point: int,
412
+ weight_zero_point: torch.Tensor | int,
413
+ out_multiplier: int,
414
+ out_shift: int,
415
+ out_zero_point: int,
416
+ ) -> torch.Tensor:
417
+ """
418
+ Quantized linear (transposed matmul) operation.
419
+
420
+ Args:
421
+ - src (Tensor): The activations tensor
422
+ - weight (Tensor): The weight tensor
423
+ - bias (Tensor): The bias tensor
424
+ - in_zero_point (int): The quantized mapping of zero for the input
425
+ - weight_zero_point (Tensor): The quantized mapping of zero for the weight
426
+ - out_multiplier (Tensor): The multiplier used to scale the output
427
+ - out_shift (Tensor): The shift used to scale the output
428
+ - out_zero_point (int): The quantized mapping of zero for the output
429
+ - offset (Tensor): Unused
430
+ """
431
+ out_scale = 1.0 / (-out_multiplier * (1 / (1 << 31)) * (2**out_shift))
432
+
433
+ N, K = weight.shape
434
+
435
+ leading_dims = src.shape[:-1]
436
+ src = src.view(-1, K)
437
+
438
+ dtype = src.dtype
439
+ supported_dtypes = [torch.int8, torch.uint8, torch.int16, torch.int32]
440
+ if dtype not in supported_dtypes:
441
+ raise ValueError(
442
+ f"Unsupported dtype to quantize to {dtype}. Supported dtypes must be one of {supported_dtypes}"
443
+ )
444
+
445
+ out = torch.nn.functional.linear(
446
+ (src - in_zero_point).float(),
447
+ (weight - weight_zero_point).float(),
448
+ bias.float(),
449
+ )
450
+ return quantize_per_tensor(
451
+ out,
452
+ out_scale,
453
+ out_zero_point,
454
+ torch.iinfo(dtype).min,
455
+ torch.iinfo(dtype).max,
456
+ dtype,
457
+ ).reshape(*leading_dims, N)
458
+
459
+
460
+ def quantized_linear_variant(
461
+ per_tensor: bool,
462
+ fully_connected: bool,
463
+ src_dtype: torch.dtype | None = None,
464
+ weight_dtype: torch.dtype | None = None,
465
+ ) -> Callable[[Callable[..., torch.Tensor]], Callable[..., torch.Tensor]]:
466
+
467
+ def decorator(_: Callable[..., torch.Tensor]) -> Callable[..., torch.Tensor]:
468
+ def variant(
469
+ src: torch.Tensor,
470
+ weight: torch.Tensor,
471
+ bias: torch.Tensor,
472
+ in_zero_point: int,
473
+ weight_zero_point: torch.Tensor | int,
474
+ out_multiplier: torch.Tensor | int,
475
+ out_shift: torch.Tensor | int,
476
+ out_zero_point: int,
477
+ offset: torch.Tensor | None = None,
478
+ ) -> torch.Tensor:
479
+ if fully_connected and src.shape[0] != 1:
480
+ raise ValueError(
481
+ "Fully connected quantized linear only supports batch size of 1"
482
+ )
483
+ if src_dtype and src.dtype != src_dtype:
484
+ raise ValueError(
485
+ f"src dtype must be {src_dtype}. Got {src.dtype} instead"
486
+ )
487
+ if weight_dtype and weight.dtype != weight_dtype:
488
+ raise ValueError(
489
+ f"weight dtype must be {weight_dtype}. Got {weight.dtype} instead"
490
+ )
491
+ if bias.dtype != torch.int32:
492
+ raise ValueError(
493
+ f"bias dtype must be torch.int32. Got {bias.dtype} instead"
494
+ )
495
+
496
+ if per_tensor:
497
+ assert isinstance(weight_zero_point, int)
498
+ assert isinstance(out_multiplier, int)
499
+ assert isinstance(out_shift, int)
500
+ _out_shift = out_shift
501
+ _out_multiplier = out_multiplier
502
+ else:
503
+ assert isinstance(out_shift, torch.Tensor)
504
+ assert isinstance(out_multiplier, torch.Tensor)
505
+ if out_shift.numel() != 1:
506
+ raise ValueError("out_shift must be a scalar")
507
+
508
+ if out_shift.dtype != torch.int32:
509
+ raise ValueError(
510
+ f"out_shift must be an int32. Got {out_shift.dtype} instead"
511
+ )
512
+
513
+ _out_shift = int(out_shift.item())
514
+ _out_multiplier = int(out_multiplier[0].item())
515
+
516
+ return quantized_linear_common(
517
+ src,
518
+ weight,
519
+ bias,
520
+ in_zero_point,
521
+ weight_zero_point,
522
+ _out_multiplier,
523
+ _out_shift,
524
+ out_zero_point,
525
+ )
526
+
527
+ return variant
528
+
529
+ return decorator
530
+
531
+
532
+ @impl_tracked(m, "quantized_linear")
533
+ @quantized_linear_variant(False, False)
534
+ def quantized_linear() -> torch.Tensor: ...
535
+
536
+
537
+ @impl_tracked(m, "quantized_linear.per_tensor")
538
+ @quantized_linear_variant(True, False)
539
+ def quantized_linear_per_tensor() -> torch.Tensor: ...
540
+
541
+
542
+ @impl_tracked(m, "quantized_linear_asym8sxasym8s_asym8s.per_tensor")
543
+ @quantized_linear_variant(True, False, torch.int8, torch.int8)
544
+ def quantized_linear_asym8sxasym8s_asym8s_per_tensor() -> torch.Tensor: ...
545
+
546
+
547
+ @impl_tracked(m, "quantized_linear_asym8uxasym8u_asym8u.per_tensor")
548
+ @quantized_linear_variant(True, False, torch.uint8, torch.uint8)
549
+ def quantized_linear_asym8uxasym8u_asym8u_per_tensor() -> torch.Tensor: ...
550
+
551
+
552
+ @impl_tracked(m, "quantized_fully_connected")
553
+ @quantized_linear_variant(False, True)
554
+ def quantized_fully_connected() -> torch.Tensor: ...
555
+
556
+
557
+ @impl_tracked(m, "quantized_fully_connected.per_tensor")
558
+ @quantized_linear_variant(True, True)
559
+ def quantized_fully_connected_per_tensor() -> torch.Tensor: ...
560
+
561
+
562
+ @impl_tracked(m, "quantized_fully_connected_asym8sxasym8s_asym8s.per_tensor")
563
+ @quantized_linear_variant(True, True, torch.int8, torch.int8)
564
+ def quantized_fully_connected_asym8sxasym8s_asym8s_per_tensor() -> torch.Tensor: ...
565
+
566
+
567
+ @impl_tracked(m, "quantized_fully_connected_asym8uxasym8u_asym8u.per_tensor")
568
+ @quantized_linear_variant(True, True, torch.uint8, torch.uint8)
569
+ def quantized_fully_connected_asym8uxasym8u_asym8u_per_tensor() -> torch.Tensor: ...
570
+
571
+
572
+ @impl_tracked(m, "fully_connected")
573
+ def fully_connected(
574
+ input_tensor: torch.Tensor,
575
+ weight: torch.Tensor,
576
+ bias: torch.Tensor,
577
+ ) -> torch.Tensor:
578
+ if input_tensor.shape[0] != 1:
579
+ raise ValueError("Fully connected linear only supports batch size of 1")
580
+ return F.linear(input_tensor, weight, bias)
581
+
582
+
583
+ @impl_tracked(m, "quantized_matmul")
584
+ def quantized_matmul(
585
+ X: torch.Tensor,
586
+ X_zero_point: int,
587
+ Y: torch.Tensor,
588
+ Y_zero_point: int,
589
+ bias: torch.Tensor | None,
590
+ out_multiplier: int,
591
+ out_shift: int,
592
+ out_zero_point: int,
593
+ transposed: bool = False,
594
+ ) -> torch.Tensor:
595
+ """
596
+ Quantized matmul operation.
597
+
598
+ Args:
599
+ - X (Tensor): The activations tensor
600
+ - X_zero_point (int): The quantized mapping of zero for the input
601
+ - Y (Tensor): The weight tensor
602
+ - Y_zero_point (int): The quantized mapping of zero for the weight
603
+ - bias (Tensor): The bias tensor
604
+ - out_multiplier (int): The multiplier used to scale the output
605
+ - out_shift (int): The shift used to scale the output
606
+ - out_zero_point (int): The quantized mapping of zero for the output
607
+ - transposed (bool): Whether Y is transposed.
608
+ """
609
+ if bias is not None and not torch.all(bias == 0):
610
+ raise ValueError("bias must be None or all zeros since unused in out variant")
611
+
612
+ if transposed:
613
+ Y = Y.transpose(-1, -2)
614
+
615
+ out_scale = 1.0 / (-out_multiplier * (1 / (1 << 31)) * (2**out_shift))
616
+
617
+ out = torch.matmul(
618
+ (X - X_zero_point).float(),
619
+ (Y - Y_zero_point).float(),
620
+ )
621
+ return quantize_per_tensor(
622
+ out,
623
+ out_scale,
624
+ out_zero_point,
625
+ torch.iinfo(X.dtype).min,
626
+ torch.iinfo(X.dtype).max,
627
+ X.dtype,
628
+ )
629
+
630
+
631
+ @impl_tracked(m, "quantized_matmul_asym8sxasym8s_asym8s")
632
+ def quantized_matmul_asym8sxasym8s_asym8s(
633
+ X: torch.Tensor,
634
+ X_zero_point: int,
635
+ Y: torch.Tensor,
636
+ Y_zero_point: int,
637
+ bias: torch.Tensor | None,
638
+ out_multiplier: int,
639
+ out_shift: int,
640
+ out_zero_point: int,
641
+ transposed: bool = False,
642
+ ) -> torch.Tensor:
643
+ if X.dtype != torch.int8:
644
+ raise ValueError("X dtype must be torch.int8")
645
+ if Y.dtype != torch.int8:
646
+ raise ValueError("Y dtype must be torch.int8")
647
+
648
+ return quantized_matmul(
649
+ X,
650
+ X_zero_point,
651
+ Y,
652
+ Y_zero_point,
653
+ bias,
654
+ out_multiplier,
655
+ out_shift,
656
+ out_zero_point,
657
+ transposed,
658
+ )
659
+
660
+
661
+ @impl_tracked(m, "quantized_matmul_asym8uxasym8u_asym8u")
662
+ def quantized_matmul_asym8uxasym8u_asym8u(
663
+ X: torch.Tensor,
664
+ X_zero_point: int,
665
+ Y: torch.Tensor,
666
+ Y_zero_point: int,
667
+ bias: torch.Tensor | None,
668
+ out_multiplier: int,
669
+ out_shift: int,
670
+ out_zero_point: int,
671
+ transposed: bool = False,
672
+ ) -> torch.Tensor:
673
+ if X.dtype != torch.uint8:
674
+ raise ValueError("X dtype must be torch.uint8")
675
+ if Y.dtype != torch.uint8:
676
+ raise ValueError("Y dtype must be torch.uint8")
677
+
678
+ return quantized_matmul(
679
+ X,
680
+ X_zero_point,
681
+ Y,
682
+ Y_zero_point,
683
+ bias,
684
+ out_multiplier,
685
+ out_shift,
686
+ out_zero_point,
687
+ transposed,
688
+ )
689
+
690
+
691
+ @impl_tracked(m, "quantized_layer_norm.per_tensor")
692
+ def quantized_layer_norm_per_tensor(
693
+ input_tensor: torch.Tensor,
694
+ X_scale: float,
695
+ X_zero_point: int,
696
+ normalized_shape: list[int],
697
+ weight: torch.Tensor,
698
+ bias: torch.Tensor,
699
+ eps: float,
700
+ output_scale: float,
701
+ output_zero_point: int,
702
+ ) -> torch.Tensor:
703
+ """
704
+ Quantized layer norm operation.
705
+
706
+ Args:
707
+ - input_tensor (Tensor): The activations tensor
708
+ - X_scale (float): The scale of the input
709
+ - X_zero_point (int): The zero point of the input
710
+ - normalized_shape (int): The shape of the input
711
+ - weight (Tensor): The weight tensor
712
+ - bias (Tensor): The bias tensor
713
+ - eps (float): The epsilon value
714
+ - output_scale (float): The scale of the output
715
+ - output_zero_point (int): The zero point of the output
716
+ """
717
+ supported_dtypes = [torch.int8, torch.uint8]
718
+ if input_tensor.dtype not in supported_dtypes:
719
+ raise ValueError(
720
+ f"Input dtype must be one of {supported_dtypes}. Got {input_tensor.dtype}"
721
+ )
722
+
723
+ float_input_tensor = dequantize_per_tensor(
724
+ input_tensor, X_scale, X_zero_point, -128, 127, input_tensor.dtype
725
+ )
726
+ assert isinstance(float_input_tensor, torch.Tensor)
727
+ out = torch.nn.functional.layer_norm(
728
+ float_input_tensor, normalized_shape, weight, bias, eps=eps
729
+ )
730
+
731
+ return quantize_per_tensor(
732
+ out,
733
+ output_scale,
734
+ output_zero_point,
735
+ torch.iinfo(input_tensor.dtype).min,
736
+ torch.iinfo(input_tensor.dtype).max,
737
+ input_tensor.dtype,
738
+ )
739
+
740
+
741
+ @impl_tracked(m, "quantized_layer_norm")
742
+ def quantized_layer_norm(
743
+ input_tensor: torch.Tensor,
744
+ X_scale: torch.Tensor,
745
+ X_zero_point: torch.Tensor,
746
+ normalized_shape: list[int],
747
+ weight: torch.Tensor,
748
+ bias: torch.Tensor,
749
+ eps: float,
750
+ output_scale: float,
751
+ output_zero_point: int,
752
+ ) -> torch.Tensor:
753
+ return quantized_layer_norm_per_tensor(
754
+ input_tensor,
755
+ float(X_scale.item()),
756
+ int(X_zero_point.item()),
757
+ normalized_shape,
758
+ weight,
759
+ bias,
760
+ eps,
761
+ output_scale,
762
+ output_zero_point,
763
+ )
764
+
765
+
766
+ def quantized_conv_per_tensor(
767
+ input_tensor: torch.Tensor,
768
+ weight: torch.Tensor,
769
+ bias: torch.Tensor,
770
+ stride: tuple[int, int],
771
+ padding: tuple[int, int],
772
+ dilation: tuple[int, int],
773
+ groups: int,
774
+ in_zero_point: int,
775
+ weight_zero_point: int,
776
+ bias_scale: float,
777
+ output_scale: float,
778
+ output_zero_point: int,
779
+ out_multiplier: int,
780
+ out_shift: int,
781
+ ) -> torch.Tensor:
782
+ """
783
+ Quantized convolution operation.
784
+
785
+ Args:
786
+ - input_tensor (Tensor): The activations tensor
787
+ - weight (Tensor): The weight tensor
788
+ - bias (Tensor): The bias tensor
789
+ - stride (Tuple[int]): The stride of the convolution
790
+ - padding (Tuple[int]): The padding of the convolution
791
+ - dilation (Tuple[int]): The dilation of the convolution
792
+ - groups (int): The number of groups
793
+ - in_zero_point (int): The quantized mapping of zero for the input
794
+ - weight_zero_point (int): The quantized mapping of zero for the weight
795
+ - bias_scale (float): The quantized bias scale
796
+ - output_scale (float): The scale of the output
797
+ - output_zero_point (int): The zero point of the output
798
+ - out_multiplier (int): Unused
799
+ - out_shift (int): Unused
800
+ """
801
+ if len(input_tensor.shape) == 3:
802
+ float_out = torch.nn.functional.conv1d(
803
+ (input_tensor - in_zero_point).float(),
804
+ (weight - weight_zero_point).float(),
805
+ (bias * bias_scale).float(),
806
+ stride[-1],
807
+ padding[-1],
808
+ dilation[-1],
809
+ groups,
810
+ )
811
+
812
+ elif len(input_tensor.shape) == 4:
813
+ float_out = torch.nn.functional.conv2d(
814
+ (input_tensor - in_zero_point).float(),
815
+ (weight - weight_zero_point).float(),
816
+ (bias * bias_scale).float(),
817
+ stride,
818
+ padding,
819
+ dilation,
820
+ groups,
821
+ )
822
+ else:
823
+ raise ValueError("Input tensor must be 3D or 4D")
824
+
825
+ return quantize_per_tensor(
826
+ float_out,
827
+ output_scale,
828
+ output_zero_point,
829
+ torch.iinfo(input_tensor.dtype).min,
830
+ torch.iinfo(input_tensor.dtype).max,
831
+ input_tensor.dtype,
832
+ )
833
+
834
+
835
+ @impl_tracked(m, "quantized_conv2d_nchw.per_tensor")
836
+ def quantized_conv2d_nchw_per_tensor(
837
+ input_tensor: torch.Tensor,
838
+ weight: torch.Tensor,
839
+ bias: torch.Tensor,
840
+ stride: tuple[int, int],
841
+ padding: tuple[int, int],
842
+ dilation: tuple[int, int],
843
+ groups: int,
844
+ in_zero_point: int,
845
+ weight_zero_point: int,
846
+ bias_scale: float,
847
+ output_scale: float,
848
+ output_zero_point: int,
849
+ out_multiplier: int,
850
+ out_shift: int,
851
+ ) -> torch.Tensor:
852
+ """
853
+ Quantized convolution operation.
854
+
855
+ Args:
856
+ - input_tensor (Tensor): The activations tensor
857
+ - weight (Tensor): The weight tensor
858
+ - bias (Tensor): The bias tensor
859
+ - stride (Tuple[int]): The stride of the convolution
860
+ - padding (Tuple[int]): The padding of the convolution
861
+ - dilation (Tuple[int]): The dilation of the convolution
862
+ - groups (int): The number of groups
863
+ - in_zero_point (int): The quantized mapping of zero for the input
864
+ - weight_zero_point (int): The quantized mapping of zero for the weight
865
+ - bias_scale (float): The quantized bias scale
866
+ - output_scale (float): The scale of the output
867
+ - output_zero_point (int): The zero point of the output
868
+ - out_multiplier (int): Unused
869
+ - out_shift (int): Unused
870
+ """
871
+ if not input_tensor.is_contiguous(memory_format=torch.contiguous_format):
872
+ raise ValueError("Input tensor must be in NCHW format")
873
+ return quantized_conv_per_tensor(
874
+ input_tensor,
875
+ weight,
876
+ bias,
877
+ stride,
878
+ padding,
879
+ dilation,
880
+ groups,
881
+ in_zero_point,
882
+ weight_zero_point,
883
+ bias_scale,
884
+ output_scale,
885
+ output_zero_point,
886
+ out_multiplier,
887
+ out_shift,
888
+ )
889
+
890
+
891
+ @impl_tracked(m, "quantized_conv2d_nchw")
892
+ def quantized_conv2d_nchw(
893
+ input_tensor: torch.Tensor,
894
+ weight: torch.Tensor,
895
+ bias: torch.Tensor,
896
+ stride: tuple[int, int],
897
+ padding: tuple[int, int],
898
+ dilation: tuple[int, int],
899
+ groups: int,
900
+ in_zero_point: int,
901
+ weight_zero_point: torch.Tensor,
902
+ bias_scale: torch.Tensor,
903
+ output_scale: float,
904
+ output_zero_point: int,
905
+ out_multiplier: torch.Tensor,
906
+ out_shift: torch.Tensor,
907
+ ) -> torch.Tensor:
908
+ return quantized_conv2d_nchw_per_tensor(
909
+ input_tensor,
910
+ weight,
911
+ bias,
912
+ stride,
913
+ padding,
914
+ dilation,
915
+ groups,
916
+ in_zero_point,
917
+ int(weight_zero_point.item()),
918
+ float(bias_scale.item()),
919
+ output_scale,
920
+ output_zero_point,
921
+ int(out_multiplier.item()),
922
+ int(out_shift.item()),
923
+ )
924
+
925
+
926
+ @impl_tracked(m, "quantized_w8a32_conv")
927
+ def quantized_w8a32_conv(
928
+ src: torch.Tensor,
929
+ weight: torch.Tensor,
930
+ w_scale: float,
931
+ bias: torch.Tensor,
932
+ b_scale: float,
933
+ ) -> torch.Tensor:
934
+
935
+ if len(weight.shape) != 3:
936
+ raise ValueError("Weight tensor must be 3D")
937
+
938
+ kernel_size, out_channels, in_channels = weight.shape
939
+ if kernel_size != 3:
940
+ raise ValueError("Kernel size must be 3")
941
+ if (out_channels % 4) != 0:
942
+ raise ValueError("Out channels must be a multiple of 4")
943
+ if (in_channels % 4) != 0:
944
+ raise ValueError("In channels must be a multiple of 4")
945
+
946
+ assert weight.dtype == torch.int8
947
+ assert bias.dtype == torch.int8
948
+
949
+ # To make compliant with torch (LCN -> NCL format)
950
+ weight = weight.permute(1, 2, 0).contiguous()
951
+
952
+ # channels last to channels first
953
+ src = src.permute(0, 2, 1).contiguous()
954
+
955
+ dequant_weight = weight.float() * w_scale
956
+
957
+ # Dequantize bias using scale
958
+ dequant_bias = bias.float() * b_scale
959
+
960
+ # Perform 1D convolution
961
+ # src: [batch, in_channel, in_length]
962
+ # weight: [out_ch, in_ch, kernel_dim]
963
+ # bias: [out_ch]
964
+ output = torch.nn.functional.conv1d(
965
+ src.float(),
966
+ dequant_weight,
967
+ dequant_bias,
968
+ )
969
+
970
+ return output
971
+
972
+
973
+ @impl_tracked(m, "quantized_w8a32_linear")
974
+ def quantized_w8a32_linear(
975
+ src: torch.Tensor,
976
+ weight: torch.Tensor,
977
+ w_scale: float,
978
+ bias: torch.Tensor,
979
+ b_scale: float,
980
+ ) -> torch.Tensor:
981
+ # src comes in shape [leading_dims, in_dim]
982
+ # weight comes in shape [in_dim, out_dim]
983
+ # output comes in empty with shape [leading_dims, out_dim]
984
+ assert weight.dtype == torch.int8
985
+ assert bias.dtype == torch.int8
986
+ if len(src.shape) >= 2:
987
+ assert src.shape[-2] == 1, "Only supporting vector-matrix multiplication"
988
+
989
+ # need to transpose to make compliant with torch linear (in, out -> out, in)
990
+ weight = weight.transpose(1, 0).contiguous()
991
+ dequant_weight = weight.float() * w_scale
992
+ dequant_bias = bias.float() * b_scale
993
+
994
+ output = torch.nn.functional.linear(
995
+ src.float(),
996
+ dequant_weight,
997
+ dequant_bias,
998
+ )
999
+
1000
+ return output
1001
+
1002
+
1003
+ @impl_tracked(m, "quantized_w8a32_gru")
1004
+ def quantized_w8a32_gru(
1005
+ inputs: torch.Tensor,
1006
+ hidden: torch.Tensor,
1007
+ weights_inputs: torch.Tensor,
1008
+ w_i_scale: float,
1009
+ weights_hidden: torch.Tensor,
1010
+ w_h_scale: float,
1011
+ bias_inputs: torch.Tensor,
1012
+ b_i_scale: float,
1013
+ bias_hidden: torch.Tensor,
1014
+ b_h_scale: float,
1015
+ ) -> torch.Tensor:
1016
+ assert weights_inputs.dtype == torch.int8
1017
+ assert weights_hidden.dtype == torch.int8
1018
+ assert bias_inputs.dtype == torch.int8
1019
+ assert bias_hidden.dtype == torch.int8
1020
+ assert inputs.dtype == torch.float32
1021
+ assert hidden.dtype == torch.float32
1022
+
1023
+ if len(hidden.shape) > 2:
1024
+ raise ValueError("Hidden state must be 2D or 1D")
1025
+
1026
+ if len(hidden.shape) == 2 and hidden.shape[0] != 1:
1027
+ raise ValueError("Leading dimension of hidden state must be 1")
1028
+
1029
+ original_hidden_shape = hidden.shape
1030
+ hidden = hidden.view(-1)
1031
+
1032
+ hidden_dim = hidden.shape[0]
1033
+ if (hidden_dim % 4) != 0:
1034
+ raise ValueError(
1035
+ "Hidden dimension must be a multiple of 4 for HiFi SIMD operations"
1036
+ )
1037
+
1038
+ dequant_weights_inputs = weights_inputs.float() * w_i_scale
1039
+ dequant_weights_hidden = weights_hidden.float() * w_h_scale
1040
+
1041
+ # C++ implementation averages the two bias scales
1042
+ avg_bias_scale = (b_i_scale + b_h_scale) / 2
1043
+ dequant_bias_inputs = bias_inputs.float() * avg_bias_scale
1044
+ dequant_bias_hidden = bias_hidden.float() * avg_bias_scale
1045
+
1046
+ gi = F.linear(inputs, dequant_weights_inputs, dequant_bias_inputs)
1047
+ gh = F.linear(hidden, dequant_weights_hidden, dequant_bias_hidden)
1048
+
1049
+ i_r, i_z, i_n = gi.chunk(3, -1)
1050
+ h_r, h_z, h_n = gh.chunk(3, -1)
1051
+
1052
+ reset_gate = torch.sigmoid(i_r + h_r)
1053
+ update_gate = torch.sigmoid(i_z + h_z)
1054
+ new_gate = torch.tanh(i_n + reset_gate * h_n)
1055
+
1056
+ new_hidden = (1 - update_gate) * new_gate + update_gate * hidden
1057
+
1058
+ if new_hidden.shape[0] != 1:
1059
+ raise ValueError("Leading dimension of hidden state must be 1")
1060
+
1061
+ assert new_hidden.shape == original_hidden_shape
1062
+
1063
+ new_hidden = new_hidden.view(-1)
1064
+ return torch.stack([new_hidden, new_hidden], dim=0)
1065
+
1066
+
1067
+ @impl_tracked(m, "quantized_conv2d_nhwc.per_tensor")
1068
+ def quantized_conv2d_nhwc_per_tensor(
1069
+ input_tensor: torch.Tensor,
1070
+ weight: torch.Tensor,
1071
+ bias: torch.Tensor,
1072
+ stride: tuple[int, int],
1073
+ padding: tuple[int, int],
1074
+ dilation: tuple[int, int],
1075
+ groups: int,
1076
+ in_zero_point: int,
1077
+ weight_zero_point: int,
1078
+ bias_scale: float,
1079
+ output_scale: float,
1080
+ output_zero_point: int,
1081
+ out_multiplier: int,
1082
+ out_shift: int,
1083
+ ) -> torch.Tensor:
1084
+ """
1085
+ Quantized convolution operation.
1086
+
1087
+ Args:
1088
+ - input_tensor (Tensor): The activations tensor
1089
+ - weight (Tensor): The weight tensor
1090
+ - bias (Tensor): The bias tensor
1091
+ - stride (Tuple[int]): The stride of the convolution
1092
+ - padding (Tuple[int]): The padding of the convolution
1093
+ - dilation (Tuple[int]): The dilation of the convolution
1094
+ - groups (int): The number of groups
1095
+ - in_zero_point (int): The quantized mapping of zero for the input
1096
+ - weight_zero_point (int): The quantized mapping of zero for the weight
1097
+ - bias_scale (float): The quantized bias scale
1098
+ - output_scale (float): The scale of the output
1099
+ - output_zero_point (int): The zero point of the output
1100
+ - out_multiplier (int): Unused
1101
+ - out_shift (int): Unused
1102
+ """
1103
+
1104
+ # Convert to NCHW format to reuse the existing implementation
1105
+ conv_is_1d = False
1106
+ if len(input_tensor.shape) == 3:
1107
+ conv_is_1d = True
1108
+ input_tensor = input_tensor.movedim(-1, 1).contiguous()
1109
+ if len(weight.shape) != 3:
1110
+ raise ValueError("Weight tensor must be 3D if input is 3D")
1111
+ weight = weight.movedim(-1, 1).contiguous()
1112
+ else:
1113
+ input_tensor = input_tensor.movedim(-1, -3)
1114
+ if len(weight.shape) != 4:
1115
+ raise ValueError("Weight tensor must be 4D if input is nd > 3")
1116
+ weight = torch.permute(weight, (0, -1, 1, 2)).contiguous()
1117
+
1118
+ nchw_out = quantized_conv_per_tensor(
1119
+ input_tensor,
1120
+ weight,
1121
+ bias,
1122
+ stride,
1123
+ padding,
1124
+ dilation,
1125
+ groups,
1126
+ in_zero_point,
1127
+ weight_zero_point,
1128
+ bias_scale,
1129
+ output_scale,
1130
+ output_zero_point,
1131
+ out_multiplier,
1132
+ out_shift,
1133
+ )
1134
+
1135
+ if conv_is_1d:
1136
+ return nchw_out.movedim(1, -1).contiguous()
1137
+ else:
1138
+ return nchw_out.movedim(-3, -1).contiguous()
1139
+
1140
+
1141
+ @impl_tracked(m, "quantized_conv2d_nhwc")
1142
+ def quantized_conv2d_nhwc(
1143
+ input_tensor: torch.Tensor,
1144
+ weight: torch.Tensor,
1145
+ bias: torch.Tensor,
1146
+ stride: tuple[int, int],
1147
+ padding: tuple[int, int],
1148
+ dilation: tuple[int, int],
1149
+ groups: int,
1150
+ in_zero_point: int,
1151
+ weight_zero_point: torch.Tensor,
1152
+ bias_scale: torch.Tensor,
1153
+ output_scale: float,
1154
+ output_zero_point: int,
1155
+ out_multiplier: torch.Tensor,
1156
+ out_shift: torch.Tensor,
1157
+ ) -> torch.Tensor:
1158
+ return quantized_conv2d_nhwc_per_tensor(
1159
+ input_tensor,
1160
+ weight,
1161
+ bias,
1162
+ stride,
1163
+ padding,
1164
+ dilation,
1165
+ groups,
1166
+ in_zero_point,
1167
+ int(weight_zero_point.item()),
1168
+ float(bias_scale.item()),
1169
+ output_scale,
1170
+ output_zero_point,
1171
+ int(out_multiplier.item()),
1172
+ int(out_shift.item()),
1173
+ )
1174
+
1175
+
1176
+ def quantized_conv_variant(
1177
+ layout: str,
1178
+ input_dtype: torch.dtype,
1179
+ weight_dtype: torch.dtype,
1180
+ is_1d: bool = False,
1181
+ ) -> Callable[[Callable[..., torch.Tensor]], Callable[..., torch.Tensor]]:
1182
+ """Create a quantized conv variant with type checking."""
1183
+
1184
+ def decorator(_: Callable[..., torch.Tensor]) -> Callable[..., torch.Tensor]:
1185
+ def variant(
1186
+ input_tensor: torch.Tensor,
1187
+ weight: torch.Tensor,
1188
+ bias: torch.Tensor,
1189
+ stride: tuple[int, int],
1190
+ padding: tuple[int, int],
1191
+ dilation: tuple[int, int],
1192
+ groups: int,
1193
+ in_zero_point: int,
1194
+ weight_zero_point: int,
1195
+ bias_scale: float,
1196
+ output_scale: float,
1197
+ output_zero_point: int,
1198
+ out_multiplier: int,
1199
+ out_shift: int,
1200
+ ) -> torch.Tensor:
1201
+ assert (
1202
+ input_tensor.dtype == input_dtype
1203
+ ), f"Expected input dtype {input_dtype}, got {input_tensor.dtype}"
1204
+ assert (
1205
+ weight.dtype == weight_dtype
1206
+ ), f"Expected weight dtype {weight_dtype}, got {weight.dtype}"
1207
+
1208
+ assert (
1209
+ bias.dtype == torch.int32
1210
+ ), f"Expected bias dtype int32, got {bias.dtype}"
1211
+
1212
+ if is_1d:
1213
+ assert (
1214
+ len(input_tensor.shape) == 3
1215
+ ), f"1D convolution requires 3D input tensor, got {len(input_tensor.shape)}D"
1216
+ assert (
1217
+ len(weight.shape) == 3
1218
+ ), f"1D convolution requires 3D weight tensor, got {len(weight.shape)}D"
1219
+
1220
+ # Call the appropriate base function
1221
+ match layout:
1222
+ case "nchw":
1223
+ return quantized_conv2d_nchw_per_tensor(
1224
+ input_tensor,
1225
+ weight,
1226
+ bias,
1227
+ stride,
1228
+ padding,
1229
+ dilation,
1230
+ groups,
1231
+ in_zero_point,
1232
+ weight_zero_point,
1233
+ bias_scale,
1234
+ output_scale,
1235
+ output_zero_point,
1236
+ out_multiplier,
1237
+ out_shift,
1238
+ )
1239
+ case "nhwc":
1240
+ return quantized_conv2d_nhwc_per_tensor(
1241
+ input_tensor,
1242
+ weight,
1243
+ bias,
1244
+ stride,
1245
+ padding,
1246
+ dilation,
1247
+ groups,
1248
+ in_zero_point,
1249
+ weight_zero_point,
1250
+ bias_scale,
1251
+ output_scale,
1252
+ output_zero_point,
1253
+ out_multiplier,
1254
+ out_shift,
1255
+ )
1256
+ case _:
1257
+ raise ValueError(f"Unknown layout {layout}")
1258
+
1259
+ return variant
1260
+
1261
+ return decorator
1262
+
1263
+
1264
+ @impl_tracked(m, "quantized_conv2d_nchw_asym8sxsym8s_asym8s.per_tensor")
1265
+ @quantized_conv_variant("nchw", torch.int8, torch.int8)
1266
+ def quantized_conv2d_nchw_asym8sxsym8s_asym8s_per_tensor() -> torch.Tensor: ...
1267
+
1268
+
1269
+ @impl_tracked(m, "quantized_conv2d_nchw_asym8uxsym8u_asym8u.per_tensor")
1270
+ @quantized_conv_variant("nchw", torch.uint8, torch.uint8)
1271
+ def quantized_conv2d_nchw_asym8uxsym8u_asym8u_per_tensor() -> torch.Tensor: ...
1272
+
1273
+
1274
+ @impl_tracked(m, "quantized_conv2d_nhwc_asym8sxsym8s_asym8s.per_tensor")
1275
+ @quantized_conv_variant("nhwc", torch.int8, torch.int8)
1276
+ def quantized_conv2d_nhwc_asym8sxsym8s_asym8s_per_tensor() -> torch.Tensor: ...
1277
+
1278
+
1279
+ @impl_tracked(m, "quantized_conv2d_nhwc_asym8uxsym8u_asym8u.per_tensor")
1280
+ @quantized_conv_variant("nhwc", torch.uint8, torch.uint8)
1281
+ def quantized_conv2d_nhwc_asym8uxsym8u_asym8u_per_tensor() -> torch.Tensor: ...
1282
+
1283
+
1284
+ @impl_tracked(m, "quantized_conv2d_nchw_dilated_asym8sxsym8s_asym8s.per_tensor")
1285
+ @quantized_conv_variant("nchw", torch.int8, torch.int8)
1286
+ def quantized_conv2d_nchw_dilated_asym8sxsym8s_asym8s_per_tensor() -> torch.Tensor: ...
1287
+
1288
+
1289
+ @impl_tracked(m, "quantized_conv2d_nchw_dilated_asym8uxsym8u_asym8u.per_tensor")
1290
+ @quantized_conv_variant("nchw", torch.uint8, torch.uint8)
1291
+ def quantized_conv2d_nchw_dilated_asym8uxsym8u_asym8u_per_tensor() -> torch.Tensor: ...
1292
+
1293
+
1294
+ @impl_tracked(m, "quantized_conv2d_nhwc_dilated_asym8sxsym8s_asym8s.per_tensor")
1295
+ @quantized_conv_variant("nhwc", torch.int8, torch.int8)
1296
+ def quantized_conv2d_nhwc_dilated_asym8sxsym8s_asym8s_per_tensor() -> torch.Tensor: ...
1297
+
1298
+
1299
+ @impl_tracked(m, "quantized_conv2d_nhwc_dilated_asym8uxsym8u_asym8u.per_tensor")
1300
+ @quantized_conv_variant("nhwc", torch.uint8, torch.uint8)
1301
+ def quantized_conv2d_nhwc_dilated_asym8uxsym8u_asym8u_per_tensor() -> torch.Tensor: ...
1302
+
1303
+
1304
+ @impl_tracked(m, "quantized_conv2d_nchw_depthwise_asym8sxsym8s_asym8s.per_tensor")
1305
+ @quantized_conv_variant("nchw", torch.int8, torch.int8)
1306
+ def quantized_conv2d_nchw_depthwise_asym8sxsym8s_asym8s_per_tensor() -> (
1307
+ torch.Tensor
1308
+ ): ...
1309
+
1310
+
1311
+ @impl_tracked(m, "quantized_conv2d_nchw_depthwise_asym8uxsym8u_asym8u.per_tensor")
1312
+ @quantized_conv_variant("nchw", torch.uint8, torch.uint8)
1313
+ def quantized_conv2d_nchw_depthwise_asym8uxsym8u_asym8u_per_tensor() -> (
1314
+ torch.Tensor
1315
+ ): ...
1316
+
1317
+
1318
+ @impl_tracked(m, "quantized_conv2d_nhwc_depthwise_asym8sxsym8s_asym8s.per_tensor")
1319
+ @quantized_conv_variant("nhwc", torch.int8, torch.int8)
1320
+ def quantized_conv2d_nhwc_depthwise_asym8sxsym8s_asym8s_per_tensor() -> (
1321
+ torch.Tensor
1322
+ ): ...
1323
+
1324
+
1325
+ @impl_tracked(m, "quantized_conv2d_nhwc_depthwise_asym8uxsym8u_asym8u.per_tensor")
1326
+ @quantized_conv_variant("nhwc", torch.uint8, torch.uint8)
1327
+ def quantized_conv2d_nhwc_depthwise_asym8uxsym8u_asym8u_per_tensor() -> (
1328
+ torch.Tensor
1329
+ ): ...
1330
+
1331
+
1332
+ @impl_tracked(m, "quantized_conv1d_ncl_asym8sxsym8s_asym8s.per_tensor")
1333
+ @quantized_conv_variant("nchw", torch.int8, torch.int8, is_1d=True)
1334
+ def quantized_conv1d_ncl_asym8sxsym8s_asym8s_per_tensor() -> torch.Tensor: ...
1335
+
1336
+
1337
+ @impl_tracked(m, "quantized_conv1d_ncl_asym8uxsym8u_asym8u.per_tensor")
1338
+ @quantized_conv_variant("nchw", torch.uint8, torch.uint8, is_1d=True)
1339
+ def quantized_conv1d_ncl_asym8uxsym8u_asym8u_per_tensor() -> torch.Tensor: ...
1340
+
1341
+
1342
+ @impl_tracked(m, "quantized_conv1d_nlc_asym8sxsym8s_asym8s.per_tensor")
1343
+ @quantized_conv_variant("nhwc", torch.int8, torch.int8, is_1d=True)
1344
+ def quantized_conv1d_nlc_asym8sxsym8s_asym8s_per_tensor() -> torch.Tensor: ...
1345
+
1346
+
1347
+ @impl_tracked(m, "quantized_conv1d_nlc_asym8uxsym8u_asym8u.per_tensor")
1348
+ @quantized_conv_variant("nhwc", torch.uint8, torch.uint8, is_1d=True)
1349
+ def quantized_conv1d_nlc_asym8uxsym8u_asym8u_per_tensor() -> torch.Tensor: ...
1350
+
1351
+
1352
+ @impl_tracked(m, "conv1d")
1353
+ def conv1d(
1354
+ input_tensor: torch.Tensor,
1355
+ weight: torch.Tensor,
1356
+ bias: torch.Tensor,
1357
+ stride: tuple[int],
1358
+ padding: tuple[int],
1359
+ dilation: tuple[int],
1360
+ groups: int,
1361
+ ) -> torch.Tensor:
1362
+ conv_out = torch.nn.functional.conv1d(
1363
+ input_tensor, weight, bias, stride[0], padding[0], dilation[0], groups
1364
+ )
1365
+
1366
+ return conv_out
1367
+
1368
+
1369
+ @impl_tracked(m, "conv2d")
1370
+ def conv2d(
1371
+ input_tensor: torch.Tensor,
1372
+ weight: torch.Tensor,
1373
+ bias: torch.Tensor,
1374
+ stride: tuple[int, int],
1375
+ padding: tuple[int, int],
1376
+ dilation: tuple[int, int],
1377
+ groups: int,
1378
+ ) -> torch.Tensor:
1379
+ conv_out = torch.nn.functional.conv2d(
1380
+ input_tensor, weight, bias, stride, padding, dilation, groups
1381
+ )
1382
+
1383
+ return conv_out
1384
+
1385
+
1386
+ @impl_tracked(m, "conv3d")
1387
+ def conv3d(
1388
+ input_tensor: torch.Tensor,
1389
+ weight: torch.Tensor,
1390
+ bias: torch.Tensor,
1391
+ stride: tuple[int, int, int],
1392
+ padding: tuple[int, int, int],
1393
+ dilation: tuple[int, int, int],
1394
+ groups: int,
1395
+ ) -> torch.Tensor:
1396
+ conv_out = torch.nn.functional.conv3d(
1397
+ input_tensor, weight, bias, stride, padding, dilation, groups
1398
+ )
1399
+
1400
+ return conv_out
1401
+
1402
+
1403
+ @impl_tracked(m, "transposed_convolution")
1404
+ def transposed_convolution(
1405
+ input_tensor: torch.Tensor,
1406
+ weight: torch.Tensor,
1407
+ bias: torch.Tensor,
1408
+ stride: tuple[int, int],
1409
+ padding: tuple[int, int],
1410
+ dilation: tuple[int, int],
1411
+ output_padding: tuple[int, int],
1412
+ groups: int,
1413
+ channel_last: bool = False,
1414
+ ) -> torch.Tensor:
1415
+
1416
+ # Cadence transposed conv receives weights that have been transformed by the pass:
1417
+ # 1. Transposed (dims 0 and 1 swapped): [out_channels, in_channels, *kernel]
1418
+ # 2. Flipped (spatial dimensions reversed)
1419
+ # We need to reverse both transformations to call PyTorch's conv_transpose
1420
+
1421
+ conv_is_1d = len(input_tensor.shape) == 3
1422
+
1423
+ # Determine flip dimensions based on weight dimensionality
1424
+ weight_dim = len(weight.shape)
1425
+ flip_dims = [-1] if weight_dim == 3 else [-1, -2]
1426
+
1427
+ # Reverse transformation step 1: Unflip the spatial dimensions
1428
+ weight = torch.flip(weight, dims=flip_dims)
1429
+
1430
+ # Reverse transformation step 2: Transpose back to PyTorch format [in, out, *kernel]
1431
+ weight = weight.transpose(0, 1).contiguous()
1432
+ if channel_last:
1433
+ if conv_is_1d:
1434
+ input_tensor = input_tensor.movedim(-1, 1).contiguous()
1435
+ if len(weight.shape) != 3:
1436
+ raise ValueError("Weight tensor must be 3D if input is 3D")
1437
+ weight = weight.movedim(-1, 1).contiguous()
1438
+ else:
1439
+ input_tensor = input_tensor.movedim(-1, -3)
1440
+ if len(weight.shape) != 4:
1441
+ raise ValueError("Weight tensor must be 4D if input is nd > 3")
1442
+ weight = torch.permute(weight, (0, -1, 1, 2)).contiguous()
1443
+
1444
+ _stride: tuple[int, int] | int = stride
1445
+ _padding: tuple[int, int] | int = padding
1446
+ _dilation: tuple[int, int] | int = dilation
1447
+ _output_padding: tuple[int, int] | int = output_padding
1448
+ if conv_is_1d:
1449
+ conv = torch.nn.functional.conv_transpose1d
1450
+ _stride = stride[0]
1451
+ _padding = padding[0]
1452
+ _dilation = dilation[0]
1453
+ _output_padding = output_padding[0]
1454
+ else:
1455
+ conv = torch.nn.functional.conv_transpose2d
1456
+
1457
+ conv_out = conv(
1458
+ input_tensor,
1459
+ weight,
1460
+ bias,
1461
+ _stride,
1462
+ _padding,
1463
+ _output_padding,
1464
+ groups,
1465
+ _dilation,
1466
+ )
1467
+ if channel_last:
1468
+ if conv_is_1d:
1469
+ conv_out = conv_out.movedim(1, -1).contiguous()
1470
+ else:
1471
+ conv_out = conv_out.movedim(-3, -1).contiguous()
1472
+
1473
+ return conv_out
1474
+
1475
+
1476
+ @impl_tracked(m, "avg_pool2d")
1477
+ def avg_pool2d(
1478
+ input_tensor: torch.Tensor,
1479
+ kernel_size: tuple[int, int],
1480
+ stride: tuple[int, int],
1481
+ padding: tuple[int, int],
1482
+ ceil_mode: bool = False,
1483
+ count_include_pad: bool = False,
1484
+ divisor_override: int | None = None,
1485
+ in_zero_point: torch.Tensor | None = None,
1486
+ channel_last: bool = False,
1487
+ ) -> torch.Tensor:
1488
+ if channel_last:
1489
+ raise NotImplementedError("Channel last is not yet supported for avg_pool2d")
1490
+
1491
+ in_dtype = input_tensor.dtype
1492
+ pad_h, pad_w = padding
1493
+ if in_zero_point is not None:
1494
+ # Avg pool2d does not allow non-0 padding,
1495
+ # so we manually pad the input
1496
+ pad_value = in_zero_point.item()
1497
+ if not count_include_pad:
1498
+ # To simulate this, just pad with 0s
1499
+ pad_value = 0
1500
+
1501
+ input_tensor = torch.nn.functional.pad(
1502
+ input_tensor,
1503
+ (pad_w, pad_w, pad_h, pad_h),
1504
+ mode="constant",
1505
+ value=pad_value,
1506
+ ).float()
1507
+
1508
+ padding = (0, 0)
1509
+
1510
+ out = torch.nn.functional.avg_pool2d(
1511
+ input_tensor,
1512
+ kernel_size,
1513
+ stride,
1514
+ padding,
1515
+ ceil_mode,
1516
+ count_include_pad,
1517
+ divisor_override,
1518
+ )
1519
+
1520
+ if in_zero_point is not None:
1521
+ min_val = torch.iinfo(in_dtype).min
1522
+ max_val = torch.iinfo(in_dtype).max
1523
+ out = torch.clamp(torch.round(out), min_val, max_val)
1524
+
1525
+ return out.to(in_dtype)
1526
+
1527
+
1528
+ def quantized_relu_common(
1529
+ X: torch.Tensor,
1530
+ X_zero_point: torch.Tensor | int,
1531
+ out_zero_point: int,
1532
+ out_multiplier: int,
1533
+ out_shift: int,
1534
+ ) -> torch.Tensor:
1535
+ """
1536
+ Quantized ReLU operation followed by requantization.
1537
+
1538
+ Args:
1539
+ - X (Tensor): The input tensor
1540
+ - X_zero_point (Tensor): The quantized mapping of zero for the input
1541
+ - out_zero_point (int): The quantized mapping of zero for the output
1542
+ - out_multiplier (Tensor): The multiplier used to scale the output
1543
+ - out_shift (Tensor): The shift used to scale the output
1544
+ """
1545
+ supported_dtypes = [torch.int8, torch.int16, torch.uint8, torch.uint16]
1546
+ if X.dtype not in supported_dtypes:
1547
+ raise ValueError(f"X dtype must be one of {supported_dtypes}. Got {X.dtype}")
1548
+
1549
+ out_scale = 1.0 / (-out_multiplier * (1 / (1 << 31)) * (2**out_shift))
1550
+ dequantized_X = torch.where(
1551
+ X > X_zero_point, X - X_zero_point, torch.zeros_like(X)
1552
+ ).to(torch.float32)
1553
+ out = quantize_per_tensor(
1554
+ dequantized_X,
1555
+ out_scale,
1556
+ out_zero_point,
1557
+ torch.iinfo(X.dtype).min,
1558
+ torch.iinfo(X.dtype).max,
1559
+ X.dtype,
1560
+ )
1561
+ assert isinstance(out, torch.Tensor)
1562
+ return out
1563
+
1564
+
1565
+ def quantized_relu_variant(
1566
+ dtype: torch.dtype | None = None,
1567
+ ) -> Callable[[Callable[..., torch.Tensor]], Callable[..., torch.Tensor]]:
1568
+ """Create a quantized relu variant with type checking."""
1569
+
1570
+ def decorator(_: Callable[..., torch.Tensor]) -> Callable[..., torch.Tensor]:
1571
+ def variant(
1572
+ X: torch.Tensor,
1573
+ X_zero_point: int,
1574
+ out_zero_point: int,
1575
+ out_multiplier: int,
1576
+ out_shift: int,
1577
+ ) -> torch.Tensor:
1578
+ if dtype and X.dtype != dtype:
1579
+ raise ValueError(f"X dtype must be {dtype}. Got {X.dtype}")
1580
+
1581
+ return quantized_relu_common(
1582
+ X,
1583
+ X_zero_point,
1584
+ out_zero_point,
1585
+ out_multiplier,
1586
+ out_shift,
1587
+ )
1588
+
1589
+ return variant
1590
+
1591
+ return decorator
1592
+
1593
+
1594
+ @impl_tracked(m, "quantized_relu.per_tensor")
1595
+ @quantized_relu_variant()
1596
+ def quantized_relu_per_tensor() -> torch.Tensor: ...
1597
+
1598
+
1599
+ @impl_tracked(m, "quantized_relu_asym8s_asym8s.per_tensor")
1600
+ @quantized_relu_variant(torch.int8)
1601
+ def quantized_relu_asym8s_asym8s_per_tensor() -> torch.Tensor: ...
1602
+
1603
+
1604
+ @impl_tracked(m, "quantized_relu_asym8u_asym8u.per_tensor")
1605
+ @quantized_relu_variant(torch.uint8)
1606
+ def quantized_relu_asym8u_asym8u_per_tensor() -> torch.Tensor: ...
1607
+
1608
+
1609
+ @impl_tracked(m, "quantized_relu")
1610
+ def quantized_relu(
1611
+ X: torch.Tensor,
1612
+ X_zero_point: torch.Tensor,
1613
+ out_zero_point: int,
1614
+ out_multiplier: torch.Tensor,
1615
+ out_shift: torch.Tensor,
1616
+ ) -> torch.Tensor:
1617
+ return quantized_relu_per_tensor(
1618
+ X, X_zero_point.item(), out_zero_point, out_multiplier.item(), out_shift.item()
1619
+ )
1620
+
1621
+
1622
+ @impl_tracked(m, "requantize.per_tensor")
1623
+ def requantize_per_tensor(
1624
+ input: torch.Tensor,
1625
+ in_scale: float,
1626
+ in_zero_point: int,
1627
+ out_scale: float,
1628
+ out_zero_point: int,
1629
+ dtype: ScalarType,
1630
+ ) -> torch.Tensor:
1631
+ if dtype in qdtype_map:
1632
+ # Old quantization mechanism
1633
+ return torch.quantize_per_tensor(
1634
+ torch.dequantize(input), out_scale, out_zero_point, qdtype_map[dtype]
1635
+ )
1636
+
1637
+ quant_min = torch.iinfo(input.dtype).min
1638
+ quant_max = torch.iinfo(input.dtype).max
1639
+ # pyre-fixme[6]: This dtype is actually the right one.
1640
+ out_quant_min = torch.iinfo(dtype).min
1641
+ # pyre-fixme[6]: This dtype is actually the right one.
1642
+ out_quant_max = torch.iinfo(dtype).max
1643
+ return torch.ops.quantized_decomposed.quantize_per_tensor(
1644
+ torch.ops.quantized_decomposed.dequantize_per_tensor(
1645
+ input,
1646
+ in_scale,
1647
+ in_zero_point,
1648
+ quant_min,
1649
+ quant_max,
1650
+ input.dtype,
1651
+ ),
1652
+ out_scale,
1653
+ out_zero_point,
1654
+ out_quant_min,
1655
+ out_quant_max,
1656
+ dtype,
1657
+ )
1658
+
1659
+
1660
+ @impl_tracked(m, "requantize")
1661
+ def requantize(
1662
+ input_tensor: torch.Tensor,
1663
+ in_scale: torch.Tensor,
1664
+ in_zero_point: torch.Tensor,
1665
+ out_scale: torch.Tensor,
1666
+ out_zero_point: torch.Tensor,
1667
+ dtype: ScalarType,
1668
+ ) -> torch.Tensor:
1669
+ return requantize_per_tensor(
1670
+ input_tensor,
1671
+ float(in_scale.item()),
1672
+ int(in_zero_point.item()),
1673
+ float(out_scale.item()),
1674
+ int(out_zero_point.item()),
1675
+ dtype,
1676
+ )
1677
+
1678
+
1679
+ @impl_tracked(m, "rms_norm")
1680
+ def rms_norm(
1681
+ X: torch.Tensor,
1682
+ normalized_shape: tuple[int],
1683
+ W: torch.Tensor,
1684
+ eps: float,
1685
+ ) -> torch.Tensor:
1686
+ return W * nn.RMSNorm(list(normalized_shape), eps=eps, dtype=X.dtype)(X)
1687
+
1688
+
1689
+ @impl_tracked(m, "where_Scalar")
1690
+ def where_Scalar(
1691
+ condition: torch.Tensor,
1692
+ if_true: float,
1693
+ if_false: float,
1694
+ ) -> torch.Tensor:
1695
+ if condition.dtype != torch.bool:
1696
+ raise ValueError("condition must be a bool tensor")
1697
+
1698
+ return torch.where(condition, if_true, if_false)
1699
+
1700
+
1701
+ @impl_tracked(m, "rope")
1702
+ def rope(
1703
+ input_tensor: torch.Tensor,
1704
+ sin_tensor: torch.Tensor,
1705
+ cos_tensor: torch.Tensor,
1706
+ pos: torch.Tensor | None,
1707
+ ) -> torch.Tensor:
1708
+ original_shape = input_tensor.shape
1709
+
1710
+ if len(original_shape) not in [4, 5]:
1711
+ raise ValueError(
1712
+ f"Input tensor must be 4D or 5D. Got {len(original_shape)}D tensor"
1713
+ )
1714
+ if original_shape[0] != 1:
1715
+ raise ValueError("Input tensor must have batch size 1")
1716
+ if len(original_shape) == 5:
1717
+ input_tensor = input_tensor.view(
1718
+ input_tensor.shape[0], input_tensor.shape[1], input_tensor.shape[2], -1
1719
+ )
1720
+
1721
+ _, seq, _, hd = input_tensor.shape
1722
+
1723
+ if hd % 2:
1724
+ raise ValueError("Hidden dimension must be divisible by 2")
1725
+
1726
+ if (
1727
+ sin_tensor.size(-1) * 2 != hd
1728
+ or cos_tensor.size(-1) * 2 != hd
1729
+ or sin_tensor.size(0) < seq
1730
+ or cos_tensor.size(0) < seq
1731
+ ):
1732
+ raise ValueError(
1733
+ f"sin_tensor and cos_tensor must have shape <kvseq (> {seq}) x {hd // 2}>. Got {sin_tensor.shape} and {cos_tensor.shape}"
1734
+ )
1735
+
1736
+ if pos is not None:
1737
+ if pos.shape != (seq,):
1738
+ raise ValueError(
1739
+ f"pos must have shape {input_tensor.shape[1]}. Got {pos.shape}"
1740
+ )
1741
+ sin_tensor = sin_tensor[pos]
1742
+ cos_tensor = cos_tensor[pos]
1743
+
1744
+ # seq x 1 x hd
1745
+ sin_tensor = sin_tensor.unsqueeze(1)
1746
+ cos_tensor = cos_tensor.unsqueeze(1)
1747
+
1748
+ # batch x seq x num_heads x head_dim_by_two
1749
+ x0, x1 = input_tensor[..., ::2], input_tensor[..., 1::2]
1750
+ o0 = x0 * cos_tensor - x1 * sin_tensor
1751
+ o1 = x0 * sin_tensor + x1 * cos_tensor
1752
+ rotated = torch.cat([o0.view(-1, 1), o1.view(-1, 1)], dim=-1)
1753
+ return rotated.view(original_shape)
1754
+
1755
+
1756
+ @impl_tracked(m, "im2row")
1757
+ def im2row(
1758
+ input_tensor: torch.Tensor,
1759
+ kernel_size: tuple[int, int],
1760
+ dilation: tuple[int, int],
1761
+ padding: tuple[int, int],
1762
+ stride: tuple[int, int],
1763
+ in_zero_point: torch.Tensor,
1764
+ channel_last: bool = False,
1765
+ ) -> torch.Tensor:
1766
+ """
1767
+ Converts an input tensor into a 2D matrix where each row is a flattened sliding window (patch)
1768
+ from the input, suitable for use in convolution as a matrix multiplication (im2row).
1769
+
1770
+ Args:
1771
+ - input_tensor: Input tensor of shape (N, C, H, W) or (N, H, W, C) if channel_last.
1772
+ - kernel_size: Size of the convolution kernel.
1773
+ - dilation: Dilation of the convolution kernel.
1774
+ - padding: Padding to apply to the input.
1775
+ - stride: Stride of the convolution.
1776
+ - in_zero_point : Zero point for input quantization (broadcastable to input).
1777
+ - channel_last: If True, input is in NHWC format, else NCHW.
1778
+
1779
+ Returns:
1780
+ - Tensor of shape (N, num_patches, patch_size)
1781
+ """
1782
+ if len(input_tensor.shape) == 3:
1783
+ height_dim = 1 if channel_last else 2
1784
+ input_tensor = input_tensor.unsqueeze(height_dim)
1785
+
1786
+ if in_zero_point is not None:
1787
+ if in_zero_point.numel() != 1 and in_zero_point.shape != (
1788
+ input_tensor.shape[0],
1789
+ ):
1790
+ raise ValueError(
1791
+ f"Input zero point must be a scalar or broadcastable to input shape {input_tensor.shape}"
1792
+ )
1793
+ if in_zero_point.dtype != torch.int32:
1794
+ raise ValueError("Input zero point must be an int32 tensor")
1795
+
1796
+ if channel_last:
1797
+ input_tensor = input_tensor.movedim(-1, -3).contiguous() # NHWC -> NCHW
1798
+
1799
+ N, C, H, W = input_tensor.shape
1800
+ kH, kW = kernel_size
1801
+ dH, dW = dilation
1802
+ pH, pW = padding
1803
+ sH, sW = stride
1804
+
1805
+ # Handle padding with zero point values
1806
+ if in_zero_point is not None and (pH > 0 or pW > 0):
1807
+ # Expand zero point to (N, 1, 1, 1) for broadcasting
1808
+ in_zero_point = in_zero_point.expand(N)
1809
+
1810
+ # Pad input with the per-batch zero point values
1811
+ input_tensor = torch.stack(
1812
+ [
1813
+ torch.nn.functional.pad(
1814
+ input_tensor[i],
1815
+ (pW, pW, pH, pH),
1816
+ mode="constant",
1817
+ value=in_zero_point[i].item(),
1818
+ )
1819
+ for i in range(len(input_tensor))
1820
+ ]
1821
+ )
1822
+
1823
+ padding = (0, 0) # Already padded manually
1824
+
1825
+ # Use unfold to extract sliding local blocks
1826
+ # Unfold: (N, C, H, W) -> (N, C, L, kH, kW), where L = number of sliding windows
1827
+ # torch.nn.functional.unfold returns (N, C*kH*kW, L)
1828
+ patches = torch.nn.functional.unfold(
1829
+ input_tensor.float(), # unfold not implemented for int
1830
+ kernel_size=(kH, kW),
1831
+ dilation=(dH, dW),
1832
+ padding=padding,
1833
+ stride=(sH, sW),
1834
+ ).to(
1835
+ input_tensor.dtype
1836
+ ) # (N, C*kH*kW, L)
1837
+
1838
+ # Transpose to (N, L, C*kH*kW)
1839
+ patches = patches.transpose(1, 2).contiguous()
1840
+
1841
+ # Reshape to (N*L, C*kH*kW)
1842
+ patches = patches.view(N, -1, C * kH * kW)
1843
+
1844
+ # If channel_last, output should be in NHWC patch order (but im2row is always row-major)
1845
+ return patches
1846
+
1847
+
1848
+ @impl_tracked(m, "im2row.per_tensor")
1849
+ def im2row_per_tensor(
1850
+ input_tensor: torch.Tensor,
1851
+ kernel_size: tuple[int, int],
1852
+ dilation: tuple[int, int],
1853
+ padding: tuple[int, int],
1854
+ stride: tuple[int, int],
1855
+ in_zero_point: int,
1856
+ channel_last: bool = False,
1857
+ ) -> torch.Tensor:
1858
+ out = im2row(
1859
+ input_tensor,
1860
+ kernel_size,
1861
+ dilation,
1862
+ padding,
1863
+ stride,
1864
+ torch.tensor(in_zero_point, dtype=torch.int32),
1865
+ channel_last,
1866
+ )
1867
+ assert isinstance(out, torch.Tensor)
1868
+ return out
1869
+
1870
+
1871
+ @impl_tracked(m, "transposed_im2row")
1872
+ def transposed_im2row(
1873
+ input_tensor: torch.Tensor,
1874
+ kernel_size: tuple[int, int],
1875
+ dilation: tuple[int, int],
1876
+ padding: tuple[int, int],
1877
+ stride: tuple[int, int],
1878
+ output_padding: tuple[int, int],
1879
+ in_zero_point: torch.Tensor,
1880
+ channel_last: bool = False,
1881
+ ) -> torch.Tensor:
1882
+ """
1883
+ Converts input tensor into im2row format for transposed convolutions.
1884
+ For each output position, extracts the kernel-sized patch of input values that
1885
+ contribute to that position in a transposed convolution.
1886
+
1887
+ Args:
1888
+ - input_tensor: Input spatial tensor, NCHW or NHWC format (3D or 4D).
1889
+ - kernel_size: Size of the convolution kernel (kernel_h, kernel_w).
1890
+ - dilation: Dilation of the convolution kernel.
1891
+ - padding: Padding to apply to the input.
1892
+ - stride: Stride of the convolution.
1893
+ - output_padding: Additional output padding for transposed convolution.
1894
+ - in_zero_point: Zero point for input quantization (broadcastable to input).
1895
+ - channel_last: If True, input is in NHWC format, else NCHW.
1896
+
1897
+ Returns:
1898
+ - 3D tensor of shape (N, output_h * output_w, kernel_h * kernel_w * in_c)
1899
+ """
1900
+ # Handle 1D convolution case by adding height dimension
1901
+ if len(input_tensor.shape) == 3:
1902
+ height_dim = 1 if channel_last else 2
1903
+ input_tensor = input_tensor.unsqueeze(height_dim)
1904
+
1905
+ if in_zero_point is not None:
1906
+ if in_zero_point.dtype != torch.int32:
1907
+ raise ValueError("Input zero point must be an int32 tensor")
1908
+
1909
+ # Move to NCHW for processing if needed
1910
+ if channel_last:
1911
+ input_tensor = input_tensor.movedim(-1, -3).contiguous() # NHWC -> NCHW
1912
+
1913
+ N, C, H_in, W_in = input_tensor.shape
1914
+ K_h, K_w = kernel_size
1915
+ device = input_tensor.device
1916
+
1917
+ # Calculate output spatial size
1918
+ H_out = (
1919
+ (H_in - 1) * stride[0]
1920
+ - 2 * padding[0]
1921
+ + dilation[0] * (K_h - 1)
1922
+ + output_padding[0]
1923
+ + 1
1924
+ )
1925
+ W_out = (
1926
+ (W_in - 1) * stride[1]
1927
+ - 2 * padding[1]
1928
+ + dilation[1] * (K_w - 1)
1929
+ + output_padding[1]
1930
+ + 1
1931
+ )
1932
+
1933
+ # Create meshgrids for all output positions and kernel positions
1934
+ h_out_grid = torch.arange(H_out, device=device).view(
1935
+ -1, 1, 1, 1
1936
+ ) # [H_out, 1, 1, 1]
1937
+ w_out_grid = torch.arange(W_out, device=device).view(
1938
+ 1, -1, 1, 1
1939
+ ) # [1, W_out, 1, 1]
1940
+ kh_grid = torch.arange(K_h, device=device).view(1, 1, -1, 1) # [1, 1, K_h, 1]
1941
+ kw_grid = torch.arange(K_w, device=device).view(1, 1, 1, -1) # [1, 1, 1, K_w]
1942
+
1943
+ # Compute input positions for all (h_out, w_out, kh, kw) combinations
1944
+ # From C++ reference: h_im = _h - ((kernel_h - 1) * dilation_h) + _kh * dilation_h + pad_h
1945
+ h_im = h_out_grid - (K_h - 1) * dilation[0] + kh_grid * dilation[0] + padding[0]
1946
+ w_im = w_out_grid - (K_w - 1) * dilation[1] + kw_grid * dilation[1] + padding[1]
1947
+
1948
+ # Check which positions are valid (divisible by stride and within bounds)
1949
+ # From C++ reference: if (h_im < 0 || h_im >= stride_h * height || h_im % stride_h != 0)
1950
+ h_valid = (h_im % stride[0] == 0) & (h_im >= 0) & (h_im < stride[0] * H_in)
1951
+ w_valid = (w_im % stride[1] == 0) & (w_im >= 0) & (w_im < stride[1] * W_in)
1952
+ valid = h_valid & w_valid # [H_out, W_out, K_h, K_w]
1953
+
1954
+ # Actual input indices (h_im / stride_h from C++ reference)
1955
+ h_in = h_im // stride[0]
1956
+ w_in = w_im // stride[1]
1957
+
1958
+ # Clamp indices to valid range (will be masked out anyway)
1959
+ h_in_safe = h_in.clamp(0, H_in - 1)
1960
+ w_in_safe = w_in.clamp(0, W_in - 1)
1961
+
1962
+ # Initialize output patches with zero points (vectorized across batches)
1963
+ # Layout depends on channel_last: NHWC uses [K_h, K_w, C], NCHW uses [C, K_h, K_w]
1964
+ if channel_last:
1965
+ # NHWC: patches layout [N, H_out, W_out, K_h, K_w, C]
1966
+ patches = torch.zeros(
1967
+ (N, H_out, W_out, K_h, K_w, C),
1968
+ dtype=input_tensor.dtype,
1969
+ device=device,
1970
+ )
1971
+ else:
1972
+ # NCHW: patches layout [N, H_out, W_out, C, K_h, K_w]
1973
+ patches = torch.zeros(
1974
+ (N, H_out, W_out, C, K_h, K_w),
1975
+ dtype=input_tensor.dtype,
1976
+ device=device,
1977
+ )
1978
+
1979
+ # Initialize patches with zero points (vectorized)
1980
+ if in_zero_point is not None:
1981
+ if in_zero_point.numel() == 1:
1982
+ # Scalar zero point - fill all patches
1983
+ patches.fill_(in_zero_point.item())
1984
+ else:
1985
+ # Per-batch zero points - expand and fill
1986
+ # in_zero_point: [N] -> [N, 1, 1, 1, 1, 1] or [N, 1, 1, 1, 1, 1]
1987
+ zp_shape = [N] + [1] * (patches.ndim - 1)
1988
+ patches = patches + in_zero_point.view(*zp_shape)
1989
+
1990
+ # Flatten the spatial and kernel dimensions for efficient gathering
1991
+ # h_in_safe, w_in_safe: [H_out, W_out, K_h, K_w] (broadcast shape)
1992
+ h_flat = h_in_safe.expand(H_out, W_out, K_h, K_w).contiguous().view(-1)
1993
+ w_flat = w_in_safe.expand(H_out, W_out, K_h, K_w).contiguous().view(-1)
1994
+
1995
+ # Vectorized gathering across all batches and channels using advanced indexing
1996
+ # Create index tensors with appropriate broadcasting shapes
1997
+ num_positions = h_flat.shape[0]
1998
+
1999
+ # batch_indices: [N, 1, 1] -> broadcasts to [N, C, num_positions]
2000
+ batch_indices = torch.arange(N, device=device).view(N, 1, 1)
2001
+
2002
+ # channel_indices: [1, C, 1] -> broadcasts to [N, C, num_positions]
2003
+ channel_indices = torch.arange(C, device=device).view(1, C, 1)
2004
+
2005
+ # h_flat, w_flat: [1, 1, num_positions] -> broadcasts to [N, C, num_positions]
2006
+ h_indices = h_flat.view(1, 1, num_positions)
2007
+ w_indices = w_flat.view(1, 1, num_positions)
2008
+
2009
+ # Advanced indexing gathers all values at once: [N, C, num_positions]
2010
+ gathered = input_tensor[batch_indices, channel_indices, h_indices, w_indices]
2011
+
2012
+ # Reshape based on channel_last flag
2013
+ if channel_last:
2014
+ # NHWC: Reshape to [N, H_out, W_out, K_h, K_w, C]
2015
+ # gathered: [N, C, H_out*W_out*K_h*K_w] -> [N, H_out*W_out*K_h*K_w, C] -> [N, H_out, W_out, K_h, K_w, C]
2016
+ gathered = gathered.transpose(1, 2).contiguous() # [N, num_positions, C]
2017
+ gathered = gathered.view(N, H_out, W_out, K_h, K_w, C)
2018
+ else:
2019
+ # NCHW: Reshape to [N, H_out, W_out, C, K_h, K_w]
2020
+ # gathered: [N, C, H_out*W_out*K_h*K_w] -> [N, C, H_out, W_out, K_h, K_w] -> [N, H_out, W_out, C, K_h, K_w]
2021
+ gathered = gathered.view(N, C, H_out, W_out, K_h, K_w)
2022
+ gathered = gathered.permute(0, 2, 3, 1, 4, 5).contiguous()
2023
+
2024
+ # Apply validity mask (vectorized across batches)
2025
+ # valid: [H_out, W_out, K_h, K_w] -> expand to match gathered shape
2026
+ if channel_last:
2027
+ # gathered: [N, H_out, W_out, K_h, K_w, C]
2028
+ valid_exp = valid.unsqueeze(0).unsqueeze(-1) # [1, H_out, W_out, K_h, K_w, 1]
2029
+ else:
2030
+ # gathered: [N, H_out, W_out, C, K_h, K_w]
2031
+ valid_exp = valid.unsqueeze(0).unsqueeze(3) # [1, H_out, W_out, 1, K_h, K_w]
2032
+
2033
+ patches = torch.where(valid_exp, gathered, patches)
2034
+
2035
+ # Reshape to final output format: [N, H_out * W_out, K_h * K_w * C]
2036
+ # The reshaping will preserve the correct dimension ordering
2037
+ if channel_last:
2038
+ # patches: [N, H_out, W_out, K_h, K_w, C] -> [N, H_out*W_out, K_h*K_w*C]
2039
+ patches = patches.view(N, H_out * W_out, K_h * K_w * C)
2040
+ else:
2041
+ # patches: [N, H_out, W_out, C, K_h, K_w] -> [N, H_out*W_out, C*K_h*K_w]
2042
+ patches = patches.view(N, H_out * W_out, C * K_h * K_w)
2043
+
2044
+ return patches
2045
+
2046
+
2047
+ @impl_tracked(m, "quantized_embedding_byte")
2048
+ def quantized_embedding_byte(
2049
+ weight: torch.Tensor,
2050
+ weight_scales: torch.Tensor,
2051
+ weight_zero_points: torch.Tensor | None,
2052
+ indices: torch.Tensor,
2053
+ pruned_weights: bool = False,
2054
+ ) -> torch.Tensor:
2055
+ if pruned_weights:
2056
+ raise NotImplementedError("Pruned weights not supported")
2057
+
2058
+ # Cannot use torch.ops.quantized_decomposed.embedding_byte.dtype because
2059
+ # it doesn't support num_groups == 1
2060
+ num_groups = 1
2061
+ if len(weight_scales.shape) == 2:
2062
+ num_groups = weight_scales.shape[1]
2063
+
2064
+ group_size = weight.shape[1] // num_groups
2065
+ weight = torch.ops.torchao.dequantize_affine.default(
2066
+ input=weight,
2067
+ block_size=(1, group_size),
2068
+ scale=weight_scales,
2069
+ zero_point=weight_zero_points,
2070
+ input_dtype=weight.dtype,
2071
+ quant_min=torch.iinfo(weight.dtype).min,
2072
+ quant_max=torch.iinfo(weight.dtype).max,
2073
+ )
2074
+
2075
+ return weight[indices]
2076
+
2077
+
2078
+ @impl_tracked(m, "idma_copy")
2079
+ def idma_copy(src: torch.Tensor, task_num: int = 0, channel: int = 0) -> torch.Tensor:
2080
+ return src.clone()
2081
+
2082
+
2083
+ @impl_tracked(m, "idma_store")
2084
+ def idma_store(src: torch.Tensor, task_num: int = 0, channel: int = 0) -> torch.Tensor:
2085
+ return src.clone()
2086
+
2087
+
2088
+ @impl_tracked(m, "idma_load")
2089
+ def idma_load(src: torch.Tensor, task_num: int = 0, channel: int = 0) -> torch.Tensor:
2090
+ return src.clone()
2091
+
2092
+
2093
+ @impl_tracked(m, "idma_wait")
2094
+ def idma_wait(src: torch.Tensor, task_num: int = 0, channel: int = 0) -> torch.Tensor:
2095
+ return src.clone()
2096
+
2097
+
2098
+ @impl_tracked(m, "linalg_svd")
2099
+ def linalg_svd(
2100
+ A: torch.Tensor,
2101
+ full_matrices: bool = False,
2102
+ compute_uv: bool = True,
2103
+ driver: str | None = None,
2104
+ ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
2105
+ assert compute_uv
2106
+ U, S, Vh = torch.linalg.svd(A, full_matrices=full_matrices, driver=driver)
2107
+ return U.contiguous(), S.contiguous(), Vh.contiguous()
2108
+
2109
+
2110
+ @impl_tracked(m, "_softmax_f32_f32")
2111
+ def softmax_f32_f32(
2112
+ input_tensor: torch.Tensor,
2113
+ dim: int,
2114
+ half_to_float: bool | None = None,
2115
+ ) -> torch.Tensor:
2116
+ assert input_tensor.dtype == torch.float32, "input_tensor must be float32"
2117
+ assert not half_to_float, "half_to_float is not supported"
2118
+ return torch.nn.functional.softmax(input_tensor, dim=dim, dtype=torch.float32)
2119
+
2120
+
2121
+ def quantized_softmax_per_tensor_common(
2122
+ input_tensor: torch.Tensor,
2123
+ mask: torch.Tensor | None,
2124
+ dim: int,
2125
+ in_scale: float,
2126
+ in_zero_point: int,
2127
+ out_scale: float,
2128
+ out_zero_point: int,
2129
+ ) -> torch.Tensor:
2130
+ """
2131
+ Quantized softmax operation.
2132
+
2133
+ Args:
2134
+ - input_tensor (Tensor): The quantized input tensor
2135
+ - mask (Tensor): Mask tensor
2136
+ - dim (int): The dimension along which softmax is computed
2137
+ - in_scale (float): The scale of the input quantization
2138
+ - in_zero_point (int): The zero point of the input quantization
2139
+ - out_scale (float): The scale of the output quantization
2140
+ - out_zero_point (int): The zero point of the output quantization
2141
+ """
2142
+ # TODO: T228751479 - Add support for mask parameter in softmax
2143
+ assert mask is None
2144
+ supported_dtypes = [torch.int8, torch.uint8, torch.int16]
2145
+ if input_tensor.dtype not in supported_dtypes:
2146
+ raise ValueError(
2147
+ f"Input dtype must be one of {supported_dtypes}. Got {input_tensor.dtype}"
2148
+ )
2149
+
2150
+ float_input_tensor = dequantize_per_tensor(
2151
+ input_tensor,
2152
+ in_scale,
2153
+ in_zero_point,
2154
+ torch.iinfo(input_tensor.dtype).min,
2155
+ torch.iinfo(input_tensor.dtype).max,
2156
+ input_tensor.dtype,
2157
+ )
2158
+
2159
+ softmax_output = torch.nn.functional.softmax(float_input_tensor, dim=dim)
2160
+
2161
+ return quantize_per_tensor(
2162
+ softmax_output,
2163
+ out_scale,
2164
+ out_zero_point,
2165
+ torch.iinfo(input_tensor.dtype).min,
2166
+ torch.iinfo(input_tensor.dtype).max,
2167
+ input_tensor.dtype,
2168
+ )
2169
+
2170
+
2171
+ @impl_tracked(m, "quantized_softmax.per_tensor")
2172
+ def quantized_softmax_per_tensor(
2173
+ input_tensor: torch.Tensor,
2174
+ mask: torch.Tensor | None,
2175
+ dim: int,
2176
+ in_scale: float,
2177
+ in_zero_point: int,
2178
+ out_scale: float,
2179
+ out_zero_point: int,
2180
+ ) -> torch.Tensor:
2181
+ return quantized_softmax_per_tensor_common(
2182
+ input_tensor,
2183
+ mask,
2184
+ dim,
2185
+ in_scale,
2186
+ in_zero_point,
2187
+ out_scale,
2188
+ out_zero_point,
2189
+ )
2190
+
2191
+
2192
+ @impl_tracked(m, "quantized_softmax")
2193
+ def quantized_softmax(
2194
+ input_tensor: torch.Tensor,
2195
+ mask: torch.Tensor | None,
2196
+ dim: int,
2197
+ in_scale: torch.Tensor,
2198
+ in_zero_point: torch.Tensor,
2199
+ out_scale: float,
2200
+ out_zero_point: int,
2201
+ ) -> torch.Tensor:
2202
+ return quantized_softmax_per_tensor_common(
2203
+ input_tensor,
2204
+ mask,
2205
+ dim,
2206
+ float(in_scale.item()),
2207
+ int(in_zero_point.item()),
2208
+ out_scale,
2209
+ out_zero_point,
2210
+ )