react-native-executorch 0.5.15 → 0.6.0-nightly-897eae9-20251213
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.
- package/README.md +42 -36
- package/android/CMakeLists.txt +13 -25
- package/android/build.gradle +2 -3
- package/android/libs/classes.jar +0 -0
- package/android/src/main/cpp/CMakeLists.txt +2 -1
- package/common/rnexecutorch/RnExecutorchInstaller.cpp +18 -0
- package/common/rnexecutorch/TokenizerModule.cpp +3 -3
- package/common/rnexecutorch/data_processing/Numerical.cpp +31 -23
- package/common/rnexecutorch/data_processing/Numerical.h +6 -1
- package/common/rnexecutorch/data_processing/dsp.cpp +0 -46
- package/common/rnexecutorch/host_objects/JsiConversions.h +16 -0
- package/common/rnexecutorch/host_objects/ModelHostObject.h +26 -11
- package/common/rnexecutorch/jsi/OwningArrayBuffer.h +19 -2
- package/common/rnexecutorch/metaprogramming/TypeConcepts.h +0 -20
- package/common/rnexecutorch/models/BaseModel.cpp +12 -11
- package/common/rnexecutorch/models/BaseModel.h +18 -10
- package/common/rnexecutorch/models/embeddings/BaseEmbeddings.cpp +3 -11
- package/common/rnexecutorch/models/embeddings/text/TextEmbeddings.cpp +0 -1
- package/common/rnexecutorch/models/image_segmentation/ImageSegmentation.cpp +6 -12
- package/common/rnexecutorch/models/llm/LLM.cpp +25 -8
- package/common/rnexecutorch/models/llm/LLM.h +4 -4
- package/common/rnexecutorch/models/ocr/CTCLabelConverter.h +1 -1
- package/common/rnexecutorch/models/ocr/utils/RecognitionHandlerUtils.cpp +7 -4
- package/common/rnexecutorch/models/speech_to_text/SpeechToText.cpp +8 -13
- package/common/rnexecutorch/models/speech_to_text/SpeechToText.h +1 -3
- package/common/rnexecutorch/models/speech_to_text/asr/ASR.cpp +12 -19
- package/common/rnexecutorch/models/speech_to_text/asr/ASR.h +4 -5
- package/common/rnexecutorch/models/text_to_image/Constants.h +9 -0
- package/common/rnexecutorch/models/text_to_image/Decoder.cpp +32 -0
- package/common/rnexecutorch/models/text_to_image/Decoder.h +24 -0
- package/common/rnexecutorch/models/text_to_image/Encoder.cpp +44 -0
- package/common/rnexecutorch/models/text_to_image/Encoder.h +32 -0
- package/common/rnexecutorch/models/text_to_image/Scheduler.cpp +152 -0
- package/common/rnexecutorch/models/text_to_image/Scheduler.h +41 -0
- package/common/rnexecutorch/models/text_to_image/TextToImage.cpp +141 -0
- package/common/rnexecutorch/models/text_to_image/TextToImage.h +64 -0
- package/common/rnexecutorch/models/text_to_image/UNet.cpp +38 -0
- package/common/rnexecutorch/models/text_to_image/UNet.h +28 -0
- package/common/rnexecutorch/models/voice_activity_detection/Constants.h +27 -0
- package/common/rnexecutorch/models/voice_activity_detection/Types.h +12 -0
- package/common/rnexecutorch/models/voice_activity_detection/Utils.cpp +15 -0
- package/common/rnexecutorch/models/voice_activity_detection/Utils.h +13 -0
- package/common/rnexecutorch/models/voice_activity_detection/VoiceActivityDetection.cpp +160 -0
- package/common/rnexecutorch/models/voice_activity_detection/VoiceActivityDetection.h +36 -0
- package/common/rnexecutorch/tests/CMakeLists.txt +30 -0
- package/common/rnexecutorch/tests/NumericalTest.cpp +110 -0
- package/common/rnexecutorch/tests/README.md +30 -13
- package/common/rnexecutorch/threads/GlobalThreadPool.h +4 -0
- package/common/runner/arange_util.cpp +44 -0
- package/common/runner/arange_util.h +37 -0
- package/common/runner/constants.h +28 -0
- package/common/runner/io_manager.h +240 -0
- package/common/runner/irunner.h +87 -16
- package/common/runner/kernel_includes.h +23 -0
- package/common/runner/runner.cpp +151 -66
- package/common/runner/runner.h +39 -22
- package/common/runner/sampler.cpp +8 -1
- package/common/runner/sampler.h +4 -2
- package/common/runner/stats.h +1 -4
- package/common/runner/text_decoder_runner.cpp +26 -12
- package/common/runner/text_decoder_runner.h +52 -31
- package/common/runner/text_prefiller.cpp +46 -12
- package/common/runner/text_prefiller.h +38 -4
- package/common/runner/text_token_generator.h +51 -26
- package/common/runner/util.h +53 -8
- package/ios/RnExecutorch.xcodeproj/project.pbxproj +0 -23
- package/lib/module/Error.js +1 -0
- package/lib/module/Error.js.map +1 -1
- package/lib/module/constants/directories.js +1 -1
- package/lib/module/constants/directories.js.map +1 -1
- package/lib/module/constants/modelUrls.js +32 -1
- package/lib/module/constants/modelUrls.js.map +1 -1
- package/lib/module/constants/ocr/models.js +7 -7
- package/lib/module/constants/ocr/models.js.map +1 -1
- package/lib/module/constants/ocr/symbols.js +3 -2
- package/lib/module/constants/ocr/symbols.js.map +1 -1
- package/lib/module/controllers/LLMController.js +10 -1
- package/lib/module/controllers/LLMController.js.map +1 -1
- package/lib/module/controllers/OCRController.js +3 -3
- package/lib/module/controllers/OCRController.js.map +1 -1
- package/lib/module/controllers/VerticalOCRController.js +2 -2
- package/lib/module/controllers/VerticalOCRController.js.map +1 -1
- package/lib/module/hooks/computer_vision/useOCR.js +3 -3
- package/lib/module/hooks/computer_vision/useOCR.js.map +1 -1
- package/lib/module/hooks/{useNonStaticModule.js → computer_vision/useTextToImage.js} +21 -16
- package/lib/module/hooks/computer_vision/useTextToImage.js.map +1 -0
- package/lib/module/hooks/computer_vision/useVerticalOCR.js +3 -3
- package/lib/module/hooks/computer_vision/useVerticalOCR.js.map +1 -1
- package/lib/module/hooks/natural_language_processing/useLLM.js +3 -3
- package/lib/module/hooks/natural_language_processing/useLLM.js.map +1 -1
- package/lib/module/hooks/natural_language_processing/useTokenizer.js +5 -5
- package/lib/module/hooks/natural_language_processing/useTokenizer.js.map +1 -1
- package/lib/module/hooks/natural_language_processing/useVAD.js +13 -0
- package/lib/module/hooks/natural_language_processing/useVAD.js.map +1 -0
- package/lib/module/index.js +7 -2
- package/lib/module/index.js.map +1 -1
- package/lib/module/modules/computer_vision/OCRModule.js +2 -2
- package/lib/module/modules/computer_vision/OCRModule.js.map +1 -1
- package/lib/module/modules/computer_vision/TextToImageModule.js +48 -0
- package/lib/module/modules/computer_vision/TextToImageModule.js.map +1 -0
- package/lib/module/modules/computer_vision/VerticalOCRModule.js +2 -2
- package/lib/module/modules/computer_vision/VerticalOCRModule.js.map +1 -1
- package/lib/module/modules/natural_language_processing/SpeechToTextModule.js +7 -4
- package/lib/module/modules/natural_language_processing/SpeechToTextModule.js.map +1 -1
- package/lib/module/modules/natural_language_processing/VADModule.js +19 -0
- package/lib/module/modules/natural_language_processing/VADModule.js.map +1 -0
- package/lib/module/types/llm.js.map +1 -1
- package/lib/module/types/vad.js +2 -0
- package/lib/module/types/vad.js.map +1 -0
- package/lib/module/utils/ResourceFetcher.js +2 -1
- package/lib/module/utils/ResourceFetcher.js.map +1 -1
- package/lib/module/utils/ResourceFetcherUtils.js +6 -6
- package/lib/module/utils/ResourceFetcherUtils.js.map +1 -1
- package/lib/typescript/Error.d.ts +1 -0
- package/lib/typescript/Error.d.ts.map +1 -1
- package/lib/typescript/constants/modelUrls.d.ts +23 -0
- package/lib/typescript/constants/modelUrls.d.ts.map +1 -1
- package/lib/typescript/constants/ocr/symbols.d.ts +1 -1
- package/lib/typescript/constants/ocr/symbols.d.ts.map +1 -1
- package/lib/typescript/controllers/LLMController.d.ts.map +1 -1
- package/lib/typescript/controllers/OCRController.d.ts +1 -1
- package/lib/typescript/controllers/OCRController.d.ts.map +1 -1
- package/lib/typescript/controllers/VerticalOCRController.d.ts +1 -1
- package/lib/typescript/controllers/VerticalOCRController.d.ts.map +1 -1
- package/lib/typescript/hooks/computer_vision/useOCR.d.ts +1 -1
- package/lib/typescript/hooks/computer_vision/useOCR.d.ts.map +1 -1
- package/lib/typescript/hooks/computer_vision/useTextToImage.d.ts +22 -0
- package/lib/typescript/hooks/computer_vision/useTextToImage.d.ts.map +1 -0
- package/lib/typescript/hooks/computer_vision/useVerticalOCR.d.ts +1 -1
- package/lib/typescript/hooks/computer_vision/useVerticalOCR.d.ts.map +1 -1
- package/lib/typescript/hooks/natural_language_processing/useLLM.d.ts.map +1 -1
- package/lib/typescript/hooks/natural_language_processing/useSpeechToText.d.ts +2 -2
- package/lib/typescript/hooks/natural_language_processing/useVAD.d.ts +16 -0
- package/lib/typescript/hooks/natural_language_processing/useVAD.d.ts.map +1 -0
- package/lib/typescript/index.d.ts +8 -1
- package/lib/typescript/index.d.ts.map +1 -1
- package/lib/typescript/modules/computer_vision/OCRModule.d.ts +1 -1
- package/lib/typescript/modules/computer_vision/OCRModule.d.ts.map +1 -1
- package/lib/typescript/modules/computer_vision/TextToImageModule.d.ts +16 -0
- package/lib/typescript/modules/computer_vision/TextToImageModule.d.ts.map +1 -0
- package/lib/typescript/modules/computer_vision/VerticalOCRModule.d.ts +1 -1
- package/lib/typescript/modules/computer_vision/VerticalOCRModule.d.ts.map +1 -1
- package/lib/typescript/modules/natural_language_processing/SpeechToTextModule.d.ts +3 -2
- package/lib/typescript/modules/natural_language_processing/SpeechToTextModule.d.ts.map +1 -1
- package/lib/typescript/modules/natural_language_processing/VADModule.d.ts +10 -0
- package/lib/typescript/modules/natural_language_processing/VADModule.d.ts.map +1 -0
- package/lib/typescript/types/llm.d.ts +2 -0
- package/lib/typescript/types/llm.d.ts.map +1 -1
- package/lib/typescript/types/vad.d.ts +5 -0
- package/lib/typescript/types/vad.d.ts.map +1 -0
- package/lib/typescript/utils/ResourceFetcher.d.ts +29 -0
- package/lib/typescript/utils/ResourceFetcher.d.ts.map +1 -1
- package/lib/typescript/utils/ResourceFetcherUtils.d.ts +2 -2
- package/lib/typescript/utils/ResourceFetcherUtils.d.ts.map +1 -1
- package/package.json +11 -8
- package/react-native-executorch.podspec +9 -9
- package/src/Error.ts +1 -0
- package/src/constants/directories.ts +1 -1
- package/src/constants/modelUrls.ts +36 -1
- package/src/constants/ocr/models.ts +7 -7
- package/src/constants/ocr/symbols.ts +3 -2
- package/src/controllers/LLMController.ts +12 -1
- package/src/controllers/OCRController.ts +3 -3
- package/src/controllers/VerticalOCRController.ts +2 -2
- package/src/hooks/computer_vision/useOCR.ts +4 -5
- package/src/hooks/computer_vision/useTextToImage.ts +92 -0
- package/src/hooks/computer_vision/useVerticalOCR.ts +4 -5
- package/src/hooks/natural_language_processing/useLLM.ts +3 -4
- package/src/hooks/natural_language_processing/useTokenizer.ts +5 -5
- package/src/hooks/natural_language_processing/useVAD.ts +15 -0
- package/src/index.ts +20 -1
- package/src/modules/computer_vision/OCRModule.ts +2 -2
- package/src/modules/computer_vision/TextToImageModule.ts +93 -0
- package/src/modules/computer_vision/VerticalOCRModule.ts +2 -2
- package/src/modules/natural_language_processing/SpeechToTextModule.ts +8 -4
- package/src/modules/natural_language_processing/VADModule.ts +27 -0
- package/src/types/llm.ts +2 -0
- package/src/types/vad.ts +4 -0
- package/src/utils/ResourceFetcher.ts +2 -1
- package/src/utils/ResourceFetcherUtils.ts +8 -8
- package/third-party/android/libs/cpuinfo/arm64-v8a/libcpuinfo.so +0 -0
- package/third-party/android/libs/executorch/arm64-v8a/libexecutorch.so +0 -0
- package/third-party/android/libs/executorch/x86_64/libexecutorch.so +0 -0
- package/third-party/android/libs/pthreadpool/arm64-v8a/libpthreadpool.so +0 -0
- package/third-party/include/c10/macros/Export.h +0 -78
- package/third-party/include/c10/macros/Macros.h +1 -520
- package/third-party/include/c10/util/BFloat16-inl.h +1 -339
- package/third-party/include/c10/util/BFloat16.h +1 -122
- package/third-party/include/c10/util/Half-inl.h +1 -347
- package/third-party/include/c10/util/Half.h +6 -419
- package/third-party/include/c10/util/TypeSafeSignMath.h +1 -133
- package/third-party/include/c10/util/bit_cast.h +1 -43
- package/third-party/include/c10/util/complex.h +1 -568
- package/third-party/include/c10/util/floating_point_utils.h +1 -33
- package/third-party/include/c10/util/irange.h +1 -1
- package/third-party/include/c10/util/llvmMathExtras.h +866 -0
- package/third-party/include/c10/util/safe_numerics.h +97 -0
- package/third-party/include/executorch/ExecuTorchError.h +6 -7
- package/third-party/include/executorch/ExecuTorchLLM/ExecuTorchLLM.h +12 -0
- package/third-party/include/executorch/ExecuTorchLLM/ExecuTorchLLMConfig.h +56 -0
- package/third-party/include/executorch/ExecuTorchLLM/ExecuTorchLLMError.h +16 -0
- package/third-party/include/executorch/ExecuTorchLLM/ExecuTorchLLMMultimodalRunner.h +227 -0
- package/third-party/include/executorch/ExecuTorchLLM/ExecuTorchLLMTextRunner.h +97 -0
- package/third-party/include/executorch/ExecuTorchLLM/module.modulemap +4 -0
- package/third-party/include/executorch/ExecuTorchLog.h +1 -0
- package/third-party/include/executorch/ExecuTorchModule.h +177 -4
- package/third-party/include/executorch/ExecuTorchTensor.h +3 -4
- package/third-party/include/executorch/ExecuTorchValue.h +1 -7
- package/third-party/include/executorch/extension/module/module.h +139 -8
- package/third-party/include/executorch/extension/tensor/tensor.h +1 -0
- package/third-party/include/executorch/extension/tensor/tensor_ptr.h +88 -26
- package/third-party/include/executorch/extension/threadpool/threadpool.h +4 -1
- package/third-party/include/executorch/runtime/backend/backend_init_context.h +6 -0
- package/third-party/include/executorch/runtime/backend/interface.h +1 -1
- package/third-party/include/executorch/runtime/core/error.h +76 -49
- package/third-party/include/executorch/runtime/core/exec_aten/util/scalar_type_util.h +18 -4
- package/third-party/include/executorch/runtime/core/memory_allocator.h +12 -2
- package/third-party/include/executorch/runtime/core/named_data_map.h +1 -11
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/macros/Export.h +0 -78
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/macros/Macros.h +1 -520
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16-inl.h +1 -339
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/BFloat16.h +1 -122
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/Half-inl.h +1 -347
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/Half.h +6 -419
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/TypeSafeSignMath.h +1 -133
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/bit_cast.h +1 -43
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/complex.h +1 -568
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/floating_point_utils.h +1 -33
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/irange.h +1 -1
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/llvmMathExtras.h +866 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/c10/util/safe_numerics.h +97 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Export.h +66 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/macros/Macros.h +553 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/BFloat16.h +477 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/Half.h +781 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/TypeSafeSignMath.h +141 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/bit_cast.h +49 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/complex.h +593 -0
- package/third-party/include/executorch/runtime/core/portable_type/c10/torch/headeronly/util/floating_point_utils.h +38 -0
- package/third-party/include/executorch/runtime/core/tensor_layout.h +1 -1
- package/third-party/include/executorch/runtime/executor/merged_data_map.h +142 -0
- package/third-party/include/executorch/runtime/executor/method.h +21 -8
- package/third-party/include/executorch/runtime/executor/method_meta.h +20 -2
- package/third-party/include/executorch/runtime/executor/program.h +0 -10
- package/third-party/include/executorch/runtime/kernel/operator_registry.h +1 -1
- package/third-party/include/executorch/runtime/platform/compiler.h +2 -0
- package/third-party/include/executorch/schema/extended_header.h +10 -1
- package/third-party/include/torch/headeronly/macros/Export.h +66 -0
- package/third-party/include/torch/headeronly/macros/Macros.h +553 -0
- package/third-party/include/torch/headeronly/util/BFloat16.h +477 -0
- package/third-party/include/torch/headeronly/util/Half.h +781 -0
- package/third-party/include/torch/headeronly/util/TypeSafeSignMath.h +141 -0
- package/third-party/include/torch/headeronly/util/bit_cast.h +49 -0
- package/third-party/include/torch/headeronly/util/complex.h +593 -0
- package/third-party/include/torch/headeronly/util/floating_point_utils.h +38 -0
- package/third-party/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/ExecutorchLib +0 -0
- package/third-party/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/Info.plist +0 -0
- package/third-party/ios/ExecutorchLib.xcframework/ios-arm64-simulator/ExecutorchLib.framework/ExecutorchLib +0 -0
- package/third-party/ios/ExecutorchLib.xcframework/ios-arm64-simulator/ExecutorchLib.framework/Info.plist +0 -0
- package/common/rnexecutorch/tests/run_all_tests.sh +0 -14
- package/common/rnexecutorch/tests/run_test.sh +0 -18
- package/ios/RnExecutorch/utils/Conversions.h +0 -14
- package/ios/RnExecutorch/utils/ETError.h +0 -26
- package/ios/RnExecutorch/utils/ImageProcessor.h +0 -15
- package/ios/RnExecutorch/utils/ImageProcessor.mm +0 -147
- package/ios/RnExecutorch/utils/Numerical.h +0 -3
- package/ios/RnExecutorch/utils/Numerical.mm +0 -18
- package/ios/RnExecutorch/utils/ScalarType.h +0 -14
- package/ios/RnExecutorch/utils/ScalarType.mm +0 -21
- package/lib/module/hooks/useNonStaticModule.js.map +0 -1
- package/lib/typescript/hooks/useNonStaticModule.d.ts +0 -21
- package/lib/typescript/hooks/useNonStaticModule.d.ts.map +0 -1
- package/src/hooks/useNonStaticModule.ts +0 -74
- package/third-party/include/executorch/extension/kernel_util/make_boxed_from_unboxed_functor.h +0 -181
- package/third-party/include/executorch/extension/kernel_util/meta_programming.h +0 -108
- package/third-party/include/executorch/extension/kernel_util/type_list.h +0 -137
- package/third-party/include/executorch/extension/threadpool/threadpool_guard.h +0 -35
|
@@ -0,0 +1,593 @@
|
|
|
1
|
+
#pragma once
|
|
2
|
+
|
|
3
|
+
#include <complex>
|
|
4
|
+
|
|
5
|
+
#include <torch/headeronly/macros/Macros.h>
|
|
6
|
+
#include <torch/headeronly/util/Half.h>
|
|
7
|
+
|
|
8
|
+
#if defined(__CUDACC__) || defined(__HIPCC__)
|
|
9
|
+
#include <thrust/complex.h>
|
|
10
|
+
#endif
|
|
11
|
+
|
|
12
|
+
C10_CLANG_DIAGNOSTIC_PUSH()
|
|
13
|
+
#if C10_CLANG_HAS_WARNING("-Wimplicit-float-conversion")
|
|
14
|
+
C10_CLANG_DIAGNOSTIC_IGNORE("-Wimplicit-float-conversion")
|
|
15
|
+
#endif
|
|
16
|
+
#if C10_CLANG_HAS_WARNING("-Wfloat-conversion")
|
|
17
|
+
C10_CLANG_DIAGNOSTIC_IGNORE("-Wfloat-conversion")
|
|
18
|
+
#endif
|
|
19
|
+
|
|
20
|
+
namespace c10 {
|
|
21
|
+
|
|
22
|
+
// c10::complex is an implementation of complex numbers that aims
|
|
23
|
+
// to work on all devices supported by PyTorch
|
|
24
|
+
//
|
|
25
|
+
// Most of the APIs duplicates std::complex
|
|
26
|
+
// Reference: https://en.cppreference.com/w/cpp/numeric/complex
|
|
27
|
+
//
|
|
28
|
+
// [NOTE: Complex Operator Unification]
|
|
29
|
+
// Operators currently use a mix of std::complex, thrust::complex, and
|
|
30
|
+
// c10::complex internally. The end state is that all operators will use
|
|
31
|
+
// c10::complex internally. Until then, there may be some hacks to support all
|
|
32
|
+
// variants.
|
|
33
|
+
//
|
|
34
|
+
//
|
|
35
|
+
// [Note on Constructors]
|
|
36
|
+
//
|
|
37
|
+
// The APIs of constructors are mostly copied from C++ standard:
|
|
38
|
+
// https://en.cppreference.com/w/cpp/numeric/complex/complex
|
|
39
|
+
//
|
|
40
|
+
// Since C++14, all constructors are constexpr in std::complex
|
|
41
|
+
//
|
|
42
|
+
// There are three types of constructors:
|
|
43
|
+
// - initializing from real and imag:
|
|
44
|
+
// `constexpr complex( const T& re = T(), const T& im = T() );`
|
|
45
|
+
// - implicitly-declared copy constructor
|
|
46
|
+
// - converting constructors
|
|
47
|
+
//
|
|
48
|
+
// Converting constructors:
|
|
49
|
+
// - std::complex defines converting constructor between float/double/long
|
|
50
|
+
// double,
|
|
51
|
+
// while we define converting constructor between float/double.
|
|
52
|
+
// - For these converting constructors, upcasting is implicit, downcasting is
|
|
53
|
+
// explicit.
|
|
54
|
+
// - We also define explicit casting from std::complex/thrust::complex
|
|
55
|
+
// - Note that the conversion from thrust is not constexpr, because
|
|
56
|
+
// thrust does not define them as constexpr ????
|
|
57
|
+
//
|
|
58
|
+
//
|
|
59
|
+
// [Operator =]
|
|
60
|
+
//
|
|
61
|
+
// The APIs of operator = are mostly copied from C++ standard:
|
|
62
|
+
// https://en.cppreference.com/w/cpp/numeric/complex/operator%3D
|
|
63
|
+
//
|
|
64
|
+
// Since C++20, all operator= are constexpr. Although we are not building with
|
|
65
|
+
// C++20, we also obey this behavior.
|
|
66
|
+
//
|
|
67
|
+
// There are three types of assign operator:
|
|
68
|
+
// - Assign a real value from the same scalar type
|
|
69
|
+
// - In std, this is templated as complex& operator=(const T& x)
|
|
70
|
+
// with specialization `complex& operator=(T x)` for float/double/long
|
|
71
|
+
// double Since we only support float and double, on will use `complex&
|
|
72
|
+
// operator=(T x)`
|
|
73
|
+
// - Copy assignment operator and converting assignment operator
|
|
74
|
+
// - There is no specialization of converting assignment operators, which type
|
|
75
|
+
// is
|
|
76
|
+
// convertible is solely dependent on whether the scalar type is convertible
|
|
77
|
+
//
|
|
78
|
+
// In addition to the standard assignment, we also provide assignment operators
|
|
79
|
+
// with std and thrust
|
|
80
|
+
//
|
|
81
|
+
//
|
|
82
|
+
// [Casting operators]
|
|
83
|
+
//
|
|
84
|
+
// std::complex does not have casting operators. We define casting operators
|
|
85
|
+
// casting to std::complex and thrust::complex
|
|
86
|
+
//
|
|
87
|
+
//
|
|
88
|
+
// [Operator ""]
|
|
89
|
+
//
|
|
90
|
+
// std::complex has custom literals `i`, `if` and `il` defined in namespace
|
|
91
|
+
// `std::literals::complex_literals`. We define our own custom literals in the
|
|
92
|
+
// namespace `c10::complex_literals`. Our custom literals does not follow the
|
|
93
|
+
// same behavior as in std::complex, instead, we define _if, _id to construct
|
|
94
|
+
// float/double complex literals.
|
|
95
|
+
//
|
|
96
|
+
//
|
|
97
|
+
// [real() and imag()]
|
|
98
|
+
//
|
|
99
|
+
// In C++20, there are two overload of these functions, one it to return the
|
|
100
|
+
// real/imag, another is to set real/imag, they are both constexpr. We follow
|
|
101
|
+
// this design.
|
|
102
|
+
//
|
|
103
|
+
//
|
|
104
|
+
// [Operator +=,-=,*=,/=]
|
|
105
|
+
//
|
|
106
|
+
// Since C++20, these operators become constexpr. In our implementation, they
|
|
107
|
+
// are also constexpr.
|
|
108
|
+
//
|
|
109
|
+
// There are two types of such operators: operating with a real number, or
|
|
110
|
+
// operating with another complex number. For the operating with a real number,
|
|
111
|
+
// the generic template form has argument type `const T &`, while the overload
|
|
112
|
+
// for float/double/long double has `T`. We will follow the same type as
|
|
113
|
+
// float/double/long double in std.
|
|
114
|
+
//
|
|
115
|
+
// [Unary operator +-]
|
|
116
|
+
//
|
|
117
|
+
// Since C++20, they are constexpr. We also make them expr
|
|
118
|
+
//
|
|
119
|
+
// [Binary operators +-*/]
|
|
120
|
+
//
|
|
121
|
+
// Each operator has three versions (taking + as example):
|
|
122
|
+
// - complex + complex
|
|
123
|
+
// - complex + real
|
|
124
|
+
// - real + complex
|
|
125
|
+
//
|
|
126
|
+
// [Operator ==, !=]
|
|
127
|
+
//
|
|
128
|
+
// Each operator has three versions (taking == as example):
|
|
129
|
+
// - complex == complex
|
|
130
|
+
// - complex == real
|
|
131
|
+
// - real == complex
|
|
132
|
+
//
|
|
133
|
+
// Some of them are removed on C++20, but we decide to keep them
|
|
134
|
+
//
|
|
135
|
+
// [Operator <<, >>]
|
|
136
|
+
//
|
|
137
|
+
// These are implemented by casting to std::complex
|
|
138
|
+
//
|
|
139
|
+
//
|
|
140
|
+
//
|
|
141
|
+
// TODO(@zasdfgbnm): c10::complex<c10::Half> is not currently supported,
|
|
142
|
+
// because:
|
|
143
|
+
// - lots of members and functions of c10::Half are not constexpr
|
|
144
|
+
// - thrust::complex only support float and double
|
|
145
|
+
|
|
146
|
+
template <typename T> struct alignas(sizeof(T) * 2) complex {
|
|
147
|
+
using value_type = T;
|
|
148
|
+
|
|
149
|
+
T real_ = T(0);
|
|
150
|
+
T imag_ = T(0);
|
|
151
|
+
|
|
152
|
+
constexpr complex() = default;
|
|
153
|
+
C10_HOST_DEVICE constexpr complex(const T &re, const T &im = T())
|
|
154
|
+
: real_(re), imag_(im) {}
|
|
155
|
+
template <typename U>
|
|
156
|
+
explicit constexpr complex(const std::complex<U> &other)
|
|
157
|
+
: complex(other.real(), other.imag()) {}
|
|
158
|
+
#if defined(__CUDACC__) || defined(__HIPCC__)
|
|
159
|
+
template <typename U>
|
|
160
|
+
explicit C10_HOST_DEVICE complex(const thrust::complex<U> &other)
|
|
161
|
+
: real_(other.real()), imag_(other.imag()) {}
|
|
162
|
+
// NOTE can not be implemented as follow due to ROCm bug:
|
|
163
|
+
// explicit C10_HOST_DEVICE complex(const thrust::complex<U> &other):
|
|
164
|
+
// complex(other.real(), other.imag()) {}
|
|
165
|
+
#endif
|
|
166
|
+
|
|
167
|
+
// Use SFINAE to specialize casting constructor for c10::complex<float> and
|
|
168
|
+
// c10::complex<double>
|
|
169
|
+
template <typename U = T>
|
|
170
|
+
C10_HOST_DEVICE explicit constexpr complex(
|
|
171
|
+
const std::enable_if_t<std::is_same_v<U, float>, complex<double>> &other)
|
|
172
|
+
: real_(other.real_), imag_(other.imag_) {}
|
|
173
|
+
template <typename U = T>
|
|
174
|
+
C10_HOST_DEVICE constexpr complex(
|
|
175
|
+
const std::enable_if_t<std::is_same_v<U, double>, complex<float>> &other)
|
|
176
|
+
: real_(other.real_), imag_(other.imag_) {}
|
|
177
|
+
|
|
178
|
+
constexpr complex<T> &operator=(T re) {
|
|
179
|
+
real_ = re;
|
|
180
|
+
imag_ = 0;
|
|
181
|
+
return *this;
|
|
182
|
+
}
|
|
183
|
+
|
|
184
|
+
constexpr complex<T> &operator+=(T re) {
|
|
185
|
+
real_ += re;
|
|
186
|
+
return *this;
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
constexpr complex<T> &operator-=(T re) {
|
|
190
|
+
real_ -= re;
|
|
191
|
+
return *this;
|
|
192
|
+
}
|
|
193
|
+
|
|
194
|
+
constexpr complex<T> &operator*=(T re) {
|
|
195
|
+
real_ *= re;
|
|
196
|
+
imag_ *= re;
|
|
197
|
+
return *this;
|
|
198
|
+
}
|
|
199
|
+
|
|
200
|
+
constexpr complex<T> &operator/=(T re) {
|
|
201
|
+
real_ /= re;
|
|
202
|
+
imag_ /= re;
|
|
203
|
+
return *this;
|
|
204
|
+
}
|
|
205
|
+
|
|
206
|
+
template <typename U> constexpr complex<T> &operator=(const complex<U> &rhs) {
|
|
207
|
+
real_ = rhs.real();
|
|
208
|
+
imag_ = rhs.imag();
|
|
209
|
+
return *this;
|
|
210
|
+
}
|
|
211
|
+
|
|
212
|
+
template <typename U>
|
|
213
|
+
constexpr complex<T> &operator+=(const complex<U> &rhs) {
|
|
214
|
+
real_ += rhs.real();
|
|
215
|
+
imag_ += rhs.imag();
|
|
216
|
+
return *this;
|
|
217
|
+
}
|
|
218
|
+
|
|
219
|
+
template <typename U>
|
|
220
|
+
constexpr complex<T> &operator-=(const complex<U> &rhs) {
|
|
221
|
+
real_ -= rhs.real();
|
|
222
|
+
imag_ -= rhs.imag();
|
|
223
|
+
return *this;
|
|
224
|
+
}
|
|
225
|
+
|
|
226
|
+
template <typename U>
|
|
227
|
+
constexpr complex<T> &operator*=(const complex<U> &rhs) {
|
|
228
|
+
// (a + bi) * (c + di) = (a*c - b*d) + (a * d + b * c) i
|
|
229
|
+
T a = real_;
|
|
230
|
+
T b = imag_;
|
|
231
|
+
U c = rhs.real();
|
|
232
|
+
U d = rhs.imag();
|
|
233
|
+
real_ = a * c - b * d;
|
|
234
|
+
imag_ = a * d + b * c;
|
|
235
|
+
return *this;
|
|
236
|
+
}
|
|
237
|
+
|
|
238
|
+
#ifdef __APPLE__
|
|
239
|
+
#define FORCE_INLINE_APPLE __attribute__((always_inline))
|
|
240
|
+
#else
|
|
241
|
+
#define FORCE_INLINE_APPLE
|
|
242
|
+
#endif
|
|
243
|
+
template <typename U>
|
|
244
|
+
constexpr FORCE_INLINE_APPLE complex<T> &
|
|
245
|
+
operator/=(const complex<U> &rhs) __ubsan_ignore_float_divide_by_zero__ {
|
|
246
|
+
// (a + bi) / (c + di) = (ac + bd)/(c^2 + d^2) + (bc - ad)/(c^2 + d^2) i
|
|
247
|
+
// the calculation below follows numpy's complex division
|
|
248
|
+
T a = real_;
|
|
249
|
+
T b = imag_;
|
|
250
|
+
U c = rhs.real();
|
|
251
|
+
U d = rhs.imag();
|
|
252
|
+
|
|
253
|
+
#if defined(__GNUC__) && !defined(__clang__)
|
|
254
|
+
// std::abs is already constexpr by gcc
|
|
255
|
+
auto abs_c = std::abs(c);
|
|
256
|
+
auto abs_d = std::abs(d);
|
|
257
|
+
#else
|
|
258
|
+
auto abs_c = c < 0 ? -c : c;
|
|
259
|
+
auto abs_d = d < 0 ? -d : d;
|
|
260
|
+
#endif
|
|
261
|
+
|
|
262
|
+
if (abs_c >= abs_d) {
|
|
263
|
+
if (abs_c == U(0) && abs_d == U(0)) {
|
|
264
|
+
/* divide by zeros should yield a complex inf or nan */
|
|
265
|
+
real_ = a / abs_c;
|
|
266
|
+
imag_ = b / abs_d;
|
|
267
|
+
} else {
|
|
268
|
+
auto rat = d / c;
|
|
269
|
+
auto scl = U(1.0) / (c + d * rat);
|
|
270
|
+
real_ = (a + b * rat) * scl;
|
|
271
|
+
imag_ = (b - a * rat) * scl;
|
|
272
|
+
}
|
|
273
|
+
} else {
|
|
274
|
+
auto rat = c / d;
|
|
275
|
+
auto scl = U(1.0) / (d + c * rat);
|
|
276
|
+
real_ = (a * rat + b) * scl;
|
|
277
|
+
imag_ = (b * rat - a) * scl;
|
|
278
|
+
}
|
|
279
|
+
return *this;
|
|
280
|
+
}
|
|
281
|
+
#undef FORCE_INLINE_APPLE
|
|
282
|
+
|
|
283
|
+
template <typename U>
|
|
284
|
+
constexpr complex<T> &operator=(const std::complex<U> &rhs) {
|
|
285
|
+
real_ = rhs.real();
|
|
286
|
+
imag_ = rhs.imag();
|
|
287
|
+
return *this;
|
|
288
|
+
}
|
|
289
|
+
|
|
290
|
+
#if defined(__CUDACC__) || defined(__HIPCC__)
|
|
291
|
+
template <typename U>
|
|
292
|
+
C10_HOST_DEVICE complex<T> &operator=(const thrust::complex<U> &rhs) {
|
|
293
|
+
real_ = rhs.real();
|
|
294
|
+
imag_ = rhs.imag();
|
|
295
|
+
return *this;
|
|
296
|
+
}
|
|
297
|
+
#endif
|
|
298
|
+
|
|
299
|
+
template <typename U> explicit constexpr operator std::complex<U>() const {
|
|
300
|
+
return std::complex<U>(std::complex<T>(real(), imag()));
|
|
301
|
+
}
|
|
302
|
+
|
|
303
|
+
#if defined(__CUDACC__) || defined(__HIPCC__)
|
|
304
|
+
template <typename U>
|
|
305
|
+
C10_HOST_DEVICE explicit operator thrust::complex<U>() const {
|
|
306
|
+
return static_cast<thrust::complex<U>>(thrust::complex<T>(real(), imag()));
|
|
307
|
+
}
|
|
308
|
+
#endif
|
|
309
|
+
|
|
310
|
+
// consistent with NumPy behavior
|
|
311
|
+
explicit constexpr operator bool() const { return real() || imag(); }
|
|
312
|
+
|
|
313
|
+
C10_HOST_DEVICE constexpr T real() const { return real_; }
|
|
314
|
+
constexpr void real(T value) { real_ = value; }
|
|
315
|
+
C10_HOST_DEVICE constexpr T imag() const { return imag_; }
|
|
316
|
+
constexpr void imag(T value) { imag_ = value; }
|
|
317
|
+
};
|
|
318
|
+
|
|
319
|
+
namespace complex_literals {
|
|
320
|
+
|
|
321
|
+
constexpr complex<float> operator""_if(long double imag) {
|
|
322
|
+
return complex<float>(0.0f, static_cast<float>(imag));
|
|
323
|
+
}
|
|
324
|
+
|
|
325
|
+
constexpr complex<double> operator""_id(long double imag) {
|
|
326
|
+
return complex<double>(0.0, static_cast<double>(imag));
|
|
327
|
+
}
|
|
328
|
+
|
|
329
|
+
constexpr complex<float> operator""_if(unsigned long long imag) {
|
|
330
|
+
return complex<float>(0.0f, static_cast<float>(imag));
|
|
331
|
+
}
|
|
332
|
+
|
|
333
|
+
constexpr complex<double> operator""_id(unsigned long long imag) {
|
|
334
|
+
return complex<double>(0.0, static_cast<double>(imag));
|
|
335
|
+
}
|
|
336
|
+
|
|
337
|
+
} // namespace complex_literals
|
|
338
|
+
|
|
339
|
+
template <typename T> constexpr complex<T> operator+(const complex<T> &val) {
|
|
340
|
+
return val;
|
|
341
|
+
}
|
|
342
|
+
|
|
343
|
+
template <typename T> constexpr complex<T> operator-(const complex<T> &val) {
|
|
344
|
+
return complex<T>(-val.real(), -val.imag());
|
|
345
|
+
}
|
|
346
|
+
|
|
347
|
+
template <typename T>
|
|
348
|
+
constexpr complex<T> operator+(const complex<T> &lhs, const complex<T> &rhs) {
|
|
349
|
+
complex<T> result = lhs;
|
|
350
|
+
return result += rhs;
|
|
351
|
+
}
|
|
352
|
+
|
|
353
|
+
template <typename T>
|
|
354
|
+
constexpr complex<T> operator+(const complex<T> &lhs, const T &rhs) {
|
|
355
|
+
complex<T> result = lhs;
|
|
356
|
+
return result += rhs;
|
|
357
|
+
}
|
|
358
|
+
|
|
359
|
+
template <typename T>
|
|
360
|
+
constexpr complex<T> operator+(const T &lhs, const complex<T> &rhs) {
|
|
361
|
+
return complex<T>(lhs + rhs.real(), rhs.imag());
|
|
362
|
+
}
|
|
363
|
+
|
|
364
|
+
template <typename T>
|
|
365
|
+
constexpr complex<T> operator-(const complex<T> &lhs, const complex<T> &rhs) {
|
|
366
|
+
complex<T> result = lhs;
|
|
367
|
+
return result -= rhs;
|
|
368
|
+
}
|
|
369
|
+
|
|
370
|
+
template <typename T>
|
|
371
|
+
constexpr complex<T> operator-(const complex<T> &lhs, const T &rhs) {
|
|
372
|
+
complex<T> result = lhs;
|
|
373
|
+
return result -= rhs;
|
|
374
|
+
}
|
|
375
|
+
|
|
376
|
+
template <typename T>
|
|
377
|
+
constexpr complex<T> operator-(const T &lhs, const complex<T> &rhs) {
|
|
378
|
+
complex<T> result = -rhs;
|
|
379
|
+
return result += lhs;
|
|
380
|
+
}
|
|
381
|
+
|
|
382
|
+
template <typename T>
|
|
383
|
+
constexpr complex<T> operator*(const complex<T> &lhs, const complex<T> &rhs) {
|
|
384
|
+
complex<T> result = lhs;
|
|
385
|
+
return result *= rhs;
|
|
386
|
+
}
|
|
387
|
+
|
|
388
|
+
template <typename T>
|
|
389
|
+
constexpr complex<T> operator*(const complex<T> &lhs, const T &rhs) {
|
|
390
|
+
complex<T> result = lhs;
|
|
391
|
+
return result *= rhs;
|
|
392
|
+
}
|
|
393
|
+
|
|
394
|
+
template <typename T>
|
|
395
|
+
constexpr complex<T> operator*(const T &lhs, const complex<T> &rhs) {
|
|
396
|
+
complex<T> result = rhs;
|
|
397
|
+
return result *= lhs;
|
|
398
|
+
}
|
|
399
|
+
|
|
400
|
+
template <typename T>
|
|
401
|
+
constexpr complex<T> operator/(const complex<T> &lhs, const complex<T> &rhs) {
|
|
402
|
+
complex<T> result = lhs;
|
|
403
|
+
return result /= rhs;
|
|
404
|
+
}
|
|
405
|
+
|
|
406
|
+
template <typename T>
|
|
407
|
+
constexpr complex<T> operator/(const complex<T> &lhs, const T &rhs) {
|
|
408
|
+
complex<T> result = lhs;
|
|
409
|
+
return result /= rhs;
|
|
410
|
+
}
|
|
411
|
+
|
|
412
|
+
template <typename T>
|
|
413
|
+
constexpr complex<T> operator/(const T &lhs, const complex<T> &rhs) {
|
|
414
|
+
complex<T> result(lhs, T());
|
|
415
|
+
return result /= rhs;
|
|
416
|
+
}
|
|
417
|
+
|
|
418
|
+
// Define operators between integral scalars and c10::complex. std::complex does
|
|
419
|
+
// not support this when T is a floating-point number. This is useful because it
|
|
420
|
+
// saves a lot of "static_cast" when operate a complex and an integer. This
|
|
421
|
+
// makes the code both less verbose and potentially more efficient.
|
|
422
|
+
#define COMPLEX_INTEGER_OP_TEMPLATE_CONDITION \
|
|
423
|
+
typename std::enable_if_t< \
|
|
424
|
+
std::is_floating_point_v<fT> && std::is_integral_v<iT>, int> = 0
|
|
425
|
+
|
|
426
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
427
|
+
constexpr c10::complex<fT> operator+(const c10::complex<fT> &a, const iT &b) {
|
|
428
|
+
return a + static_cast<fT>(b);
|
|
429
|
+
}
|
|
430
|
+
|
|
431
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
432
|
+
constexpr c10::complex<fT> operator+(const iT &a, const c10::complex<fT> &b) {
|
|
433
|
+
return static_cast<fT>(a) + b;
|
|
434
|
+
}
|
|
435
|
+
|
|
436
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
437
|
+
constexpr c10::complex<fT> operator-(const c10::complex<fT> &a, const iT &b) {
|
|
438
|
+
return a - static_cast<fT>(b);
|
|
439
|
+
}
|
|
440
|
+
|
|
441
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
442
|
+
constexpr c10::complex<fT> operator-(const iT &a, const c10::complex<fT> &b) {
|
|
443
|
+
return static_cast<fT>(a) - b;
|
|
444
|
+
}
|
|
445
|
+
|
|
446
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
447
|
+
constexpr c10::complex<fT> operator*(const c10::complex<fT> &a, const iT &b) {
|
|
448
|
+
return a * static_cast<fT>(b);
|
|
449
|
+
}
|
|
450
|
+
|
|
451
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
452
|
+
constexpr c10::complex<fT> operator*(const iT &a, const c10::complex<fT> &b) {
|
|
453
|
+
return static_cast<fT>(a) * b;
|
|
454
|
+
}
|
|
455
|
+
|
|
456
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
457
|
+
constexpr c10::complex<fT> operator/(const c10::complex<fT> &a, const iT &b) {
|
|
458
|
+
return a / static_cast<fT>(b);
|
|
459
|
+
}
|
|
460
|
+
|
|
461
|
+
template <typename fT, typename iT, COMPLEX_INTEGER_OP_TEMPLATE_CONDITION>
|
|
462
|
+
constexpr c10::complex<fT> operator/(const iT &a, const c10::complex<fT> &b) {
|
|
463
|
+
return static_cast<fT>(a) / b;
|
|
464
|
+
}
|
|
465
|
+
|
|
466
|
+
#undef COMPLEX_INTEGER_OP_TEMPLATE_CONDITION
|
|
467
|
+
|
|
468
|
+
template <typename T>
|
|
469
|
+
constexpr bool operator==(const complex<T> &lhs, const complex<T> &rhs) {
|
|
470
|
+
return (lhs.real() == rhs.real()) && (lhs.imag() == rhs.imag());
|
|
471
|
+
}
|
|
472
|
+
|
|
473
|
+
template <typename T>
|
|
474
|
+
constexpr bool operator==(const complex<T> &lhs, const T &rhs) {
|
|
475
|
+
return (lhs.real() == rhs) && (lhs.imag() == T());
|
|
476
|
+
}
|
|
477
|
+
|
|
478
|
+
template <typename T>
|
|
479
|
+
constexpr bool operator==(const T &lhs, const complex<T> &rhs) {
|
|
480
|
+
return (lhs == rhs.real()) && (T() == rhs.imag());
|
|
481
|
+
}
|
|
482
|
+
|
|
483
|
+
template <typename T>
|
|
484
|
+
constexpr bool operator!=(const complex<T> &lhs, const complex<T> &rhs) {
|
|
485
|
+
return !(lhs == rhs);
|
|
486
|
+
}
|
|
487
|
+
|
|
488
|
+
template <typename T>
|
|
489
|
+
constexpr bool operator!=(const complex<T> &lhs, const T &rhs) {
|
|
490
|
+
return !(lhs == rhs);
|
|
491
|
+
}
|
|
492
|
+
|
|
493
|
+
template <typename T>
|
|
494
|
+
constexpr bool operator!=(const T &lhs, const complex<T> &rhs) {
|
|
495
|
+
return !(lhs == rhs);
|
|
496
|
+
}
|
|
497
|
+
|
|
498
|
+
template <typename T, typename CharT, typename Traits>
|
|
499
|
+
std::basic_ostream<CharT, Traits> &
|
|
500
|
+
operator<<(std::basic_ostream<CharT, Traits> &os, const complex<T> &x) {
|
|
501
|
+
return (os << static_cast<std::complex<T>>(x));
|
|
502
|
+
}
|
|
503
|
+
|
|
504
|
+
template <typename T, typename CharT, typename Traits>
|
|
505
|
+
std::basic_istream<CharT, Traits> &
|
|
506
|
+
operator>>(std::basic_istream<CharT, Traits> &is, complex<T> &x) {
|
|
507
|
+
std::complex<T> tmp;
|
|
508
|
+
is >> tmp;
|
|
509
|
+
x = tmp;
|
|
510
|
+
return is;
|
|
511
|
+
}
|
|
512
|
+
|
|
513
|
+
template <typename T>
|
|
514
|
+
C10_HOST_DEVICE complex<T> polar(const T &r, const T &theta = T()) {
|
|
515
|
+
#if defined(__CUDACC__) || defined(__HIPCC__)
|
|
516
|
+
return static_cast<complex<T>>(thrust::polar(r, theta));
|
|
517
|
+
#else
|
|
518
|
+
// std::polar() requires r >= 0, so spell out the explicit implementation to
|
|
519
|
+
// avoid a branch.
|
|
520
|
+
return complex<T>(r * std::cos(theta), r * std::sin(theta));
|
|
521
|
+
#endif
|
|
522
|
+
}
|
|
523
|
+
|
|
524
|
+
template <> struct alignas(4) complex<Half> {
|
|
525
|
+
Half real_;
|
|
526
|
+
Half imag_;
|
|
527
|
+
|
|
528
|
+
// Constructors
|
|
529
|
+
complex() = default;
|
|
530
|
+
// Half constructor is not constexpr so the following constructor can't
|
|
531
|
+
// be constexpr
|
|
532
|
+
C10_HOST_DEVICE explicit inline complex(const Half &real, const Half &imag)
|
|
533
|
+
: real_(real), imag_(imag) {}
|
|
534
|
+
C10_HOST_DEVICE inline complex(const c10::complex<float> &value)
|
|
535
|
+
: real_(value.real()), imag_(value.imag()) {}
|
|
536
|
+
|
|
537
|
+
// Conversion operator
|
|
538
|
+
inline C10_HOST_DEVICE operator c10::complex<float>() const {
|
|
539
|
+
return {real_, imag_};
|
|
540
|
+
}
|
|
541
|
+
|
|
542
|
+
constexpr C10_HOST_DEVICE Half real() const { return real_; }
|
|
543
|
+
constexpr C10_HOST_DEVICE Half imag() const { return imag_; }
|
|
544
|
+
|
|
545
|
+
C10_HOST_DEVICE complex<Half> &operator+=(const complex<Half> &other) {
|
|
546
|
+
real_ = static_cast<float>(real_) + static_cast<float>(other.real_);
|
|
547
|
+
imag_ = static_cast<float>(imag_) + static_cast<float>(other.imag_);
|
|
548
|
+
return *this;
|
|
549
|
+
}
|
|
550
|
+
|
|
551
|
+
C10_HOST_DEVICE complex<Half> &operator-=(const complex<Half> &other) {
|
|
552
|
+
real_ = static_cast<float>(real_) - static_cast<float>(other.real_);
|
|
553
|
+
imag_ = static_cast<float>(imag_) - static_cast<float>(other.imag_);
|
|
554
|
+
return *this;
|
|
555
|
+
}
|
|
556
|
+
|
|
557
|
+
C10_HOST_DEVICE complex<Half> &operator*=(const complex<Half> &other) {
|
|
558
|
+
auto a = static_cast<float>(real_);
|
|
559
|
+
auto b = static_cast<float>(imag_);
|
|
560
|
+
auto c = static_cast<float>(other.real());
|
|
561
|
+
auto d = static_cast<float>(other.imag());
|
|
562
|
+
real_ = a * c - b * d;
|
|
563
|
+
imag_ = a * d + b * c;
|
|
564
|
+
return *this;
|
|
565
|
+
}
|
|
566
|
+
};
|
|
567
|
+
|
|
568
|
+
} // namespace c10
|
|
569
|
+
|
|
570
|
+
namespace torch::headeronly {
|
|
571
|
+
using c10::complex;
|
|
572
|
+
using c10::operator+;
|
|
573
|
+
using c10::operator-;
|
|
574
|
+
using c10::operator*;
|
|
575
|
+
using c10::operator/;
|
|
576
|
+
using c10::operator+=;
|
|
577
|
+
using c10::operator-=;
|
|
578
|
+
using c10::operator*=;
|
|
579
|
+
using c10::operator/=;
|
|
580
|
+
using c10::operator==;
|
|
581
|
+
using c10::operator!=;
|
|
582
|
+
using c10::operator<<;
|
|
583
|
+
using c10::operator>>;
|
|
584
|
+
using c10::polar;
|
|
585
|
+
|
|
586
|
+
namespace complex_literals {
|
|
587
|
+
using c10::complex_literals::operator""_if;
|
|
588
|
+
using c10::complex_literals::operator""_id;
|
|
589
|
+
} // namespace complex_literals
|
|
590
|
+
|
|
591
|
+
} // namespace torch::headeronly
|
|
592
|
+
|
|
593
|
+
C10_CLANG_DIAGNOSTIC_POP()
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
#pragma once
|
|
2
|
+
|
|
3
|
+
#include <cstdint>
|
|
4
|
+
#include <torch/headeronly/macros/Macros.h>
|
|
5
|
+
#include <torch/headeronly/util/bit_cast.h>
|
|
6
|
+
|
|
7
|
+
namespace torch::headeronly::detail {
|
|
8
|
+
|
|
9
|
+
C10_HOST_DEVICE inline float fp32_from_bits(uint32_t w) {
|
|
10
|
+
#if defined(__OPENCL_VERSION__)
|
|
11
|
+
return as_float(w);
|
|
12
|
+
#elif defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
|
|
13
|
+
return __uint_as_float((unsigned int)w);
|
|
14
|
+
#elif defined(__INTEL_COMPILER)
|
|
15
|
+
return _castu32_f32(w);
|
|
16
|
+
#else
|
|
17
|
+
return torch::headeronly::bit_cast<float>(w);
|
|
18
|
+
#endif
|
|
19
|
+
}
|
|
20
|
+
|
|
21
|
+
C10_HOST_DEVICE inline uint32_t fp32_to_bits(float f) {
|
|
22
|
+
#if defined(__OPENCL_VERSION__)
|
|
23
|
+
return as_uint(f);
|
|
24
|
+
#elif defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
|
|
25
|
+
return (uint32_t)__float_as_uint(f);
|
|
26
|
+
#elif defined(__INTEL_COMPILER)
|
|
27
|
+
return _castf32_u32(f);
|
|
28
|
+
#else
|
|
29
|
+
return torch::headeronly::bit_cast<uint32_t>(f);
|
|
30
|
+
#endif
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
} // namespace torch::headeronly::detail
|
|
34
|
+
|
|
35
|
+
namespace c10::detail {
|
|
36
|
+
using torch::headeronly::detail::fp32_from_bits;
|
|
37
|
+
using torch::headeronly::detail::fp32_to_bits;
|
|
38
|
+
} // namespace c10::detail
|
package/third-party/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/ExecutorchLib
CHANGED
|
Binary file
|
package/third-party/ios/ExecutorchLib.xcframework/ios-arm64/ExecutorchLib.framework/Info.plist
CHANGED
|
Binary file
|
|
Binary file
|
|
Binary file
|
|
@@ -1,14 +0,0 @@
|
|
|
1
|
-
#!/bin/bash
|
|
2
|
-
|
|
3
|
-
# Get the directory where the script is located
|
|
4
|
-
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" &> /dev/null && pwd)"
|
|
5
|
-
|
|
6
|
-
# Loop through all .cpp files in that directory
|
|
7
|
-
for file in "$SCRIPT_DIR"/*.cpp
|
|
8
|
-
do
|
|
9
|
-
if [ -f "$file" ]; then
|
|
10
|
-
file_name=$(basename "$file")
|
|
11
|
-
echo "Processing $file_name"
|
|
12
|
-
"$SCRIPT_DIR"/run_test.sh "$file"
|
|
13
|
-
fi
|
|
14
|
-
done
|
|
@@ -1,18 +0,0 @@
|
|
|
1
|
-
#!/bin/bash
|
|
2
|
-
|
|
3
|
-
# Check if a file name is provided
|
|
4
|
-
if [ "$#" -ne 1 ]; then
|
|
5
|
-
echo "Usage: $0 <cpp_file>"
|
|
6
|
-
exit 1
|
|
7
|
-
fi
|
|
8
|
-
|
|
9
|
-
file_name="$1"
|
|
10
|
-
|
|
11
|
-
# Compile the file with specified libraries
|
|
12
|
-
g++ -std=c++20 -o test_executable "$file_name" -lgtest -lgtest_main -lpthread
|
|
13
|
-
|
|
14
|
-
# Execute the binary
|
|
15
|
-
./test_executable
|
|
16
|
-
|
|
17
|
-
# Remove the executable
|
|
18
|
-
rm -f test_executable
|
|
@@ -1,14 +0,0 @@
|
|
|
1
|
-
#import "opencv2/opencv.hpp"
|
|
2
|
-
|
|
3
|
-
// Convert a matrix containing a single value per cell to a NSArray
|
|
4
|
-
template <typename T> NSArray *simpleMatToNSArray(const cv::Mat &mat) {
|
|
5
|
-
std::size_t numPixels = mat.rows * mat.cols;
|
|
6
|
-
NSMutableArray *arr = [[NSMutableArray alloc] initWithCapacity:numPixels];
|
|
7
|
-
|
|
8
|
-
for (std::size_t x = 0; x < mat.rows; ++x) {
|
|
9
|
-
for (std::size_t y = 0; y < mat.cols; ++y) {
|
|
10
|
-
arr[x * mat.cols + y] = @(mat.at<T>(x, y));
|
|
11
|
-
}
|
|
12
|
-
}
|
|
13
|
-
return arr;
|
|
14
|
-
}
|