diff --git a/binaries/dnn_benchmark.cpp b/binaries/dnn_benchmark.cpp index 1a834a6..304d438 100644 --- a/binaries/dnn_benchmark.cpp +++ b/binaries/dnn_benchmark.cpp @@ -39,7 +39,7 @@ bool hasEnding(std::string const &fullString, std::string const &ending) { } auto GetModel(css &daqName, const bool allow_fp16, - const PreferenceCode &compile_preference) { + const int compile_preference) { std::unique_ptr model; ModelBuilder builder; if (hasEnding(daqName, ".daq")) { @@ -64,7 +64,7 @@ auto GetModel(css &daqName, const bool allow_fp16, return model; } -auto PrefCodeToStr(const PreferenceCode &preference_code) { +auto PrefCodeToStr(const int &preference_code) { if (preference_code == ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER) { return "fast single"; } @@ -129,13 +129,13 @@ int main(int argc, char **argv) { } \ } - const std::vector preference_candidates{ + const std::vector preference_candidates{ ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER, ANEURALNETWORKS_PREFER_SUSTAINED_SPEED, ANEURALNETWORKS_PREFER_LOW_POWER}; if (quant) { uint8_t data[input_len]; - uint8_t output[output_len]; + float output[output_len]; WARM_UP; const std::vector fp16_candidates{false}; BENCHMARK(fp16_candidates, preference_candidates); @@ -149,7 +149,7 @@ int main(int argc, char **argv) { WARM_UP; const std::vector fp16_candidates = - ModelBuilder::GetAndroidSdkVersion() >= __ANDROID_API_P__ + GetAndroidSdkVersion() >= __ANDROID_API_P__ ? std::vector{false, true} : std::vector{false}; BENCHMARK(fp16_candidates, preference_candidates); diff --git a/dnnlibrary/CMakeLists.txt b/dnnlibrary/CMakeLists.txt index bb6adf8..0042cc0 100644 --- a/dnnlibrary/CMakeLists.txt +++ b/dnnlibrary/CMakeLists.txt @@ -4,7 +4,7 @@ set(dnnlibrary_src ${PROJECT_SOURCE_DIR}/include/dnnlibrary/ModelBuilder.h ${PROJECT_SOURCE_DIR}/include/dnnlibrary/Model.h ${PROJECT_SOURCE_DIR}/include/dnnlibrary/DaqReader.h - ${PROJECT_SOURCE_DIR}/include/dnnlibrary/NeuralNetworksWrapper.h + ${PROJECT_SOURCE_DIR}/include/dnnlibrary/nnapi_implementation.h android_log_helper.h operand_helper.h flatbuffers_helper.h @@ -12,6 +12,7 @@ set(dnnlibrary_src Model.cpp DaqReader.cpp NeuralNetworksWrapper.cpp + nnapi_implementation.cc ${PROJECT_SOURCE_DIR}/include/common/Shaper.h ${PROJECT_SOURCE_DIR}/common/Shaper.cpp ${PROJECT_SOURCE_DIR}/include/common/StrKeyMap.h diff --git a/dnnlibrary/DaqReader.cpp b/dnnlibrary/DaqReader.cpp index fd7869b..abbae5f 100644 --- a/dnnlibrary/DaqReader.cpp +++ b/dnnlibrary/DaqReader.cpp @@ -59,13 +59,13 @@ std::string layer_type_to_str(DNN::LayerType type) { int convert_fuse_code_to_nnapi(const DNN::FuseCode fuse_code) { switch (fuse_code) { case DNN::FuseCode::None: - return FuseCode::ANEURALNETWORKS_FUSED_NONE; + return ANEURALNETWORKS_FUSED_NONE; case DNN::FuseCode::Relu: - return FuseCode::ANEURALNETWORKS_FUSED_RELU; + return ANEURALNETWORKS_FUSED_RELU; case DNN::FuseCode::Relu1: - return FuseCode::ANEURALNETWORKS_FUSED_RELU1; + return ANEURALNETWORKS_FUSED_RELU1; case DNN::FuseCode::Relu6: - return FuseCode::ANEURALNETWORKS_FUSED_RELU6; + return ANEURALNETWORKS_FUSED_RELU6; } throw std::invalid_argument("Invalid fuse_code"); } diff --git a/dnnlibrary/Model.cpp b/dnnlibrary/Model.cpp index 299790a..be564b4 100644 --- a/dnnlibrary/Model.cpp +++ b/dnnlibrary/Model.cpp @@ -29,15 +29,19 @@ void Model::PrepareForExecution() { throw std::invalid_argument( "Error in PrepareForExecution, compilation_ == nullptr"); } - THROW_ON_ERROR(ANeuralNetworksExecution_create(compilation_, &execution_)); + THROW_ON_ERROR( + nnapi_->ANeuralNetworksExecution_create(compilation_, &execution_)); prepared_for_exe_ = true; } +Model::Model() : nnapi_(NnApiImplementation()) { +} + Model::~Model() { munmap(data_, data_size_); - ANeuralNetworksModel_free(model_); - ANeuralNetworksCompilation_free(compilation_); - ANeuralNetworksMemory_free(memory_); + nnapi_->ANeuralNetworksModel_free(model_); + nnapi_->ANeuralNetworksCompilation_free(compilation_); + nnapi_->ANeuralNetworksMemory_free(memory_); } void Model::SetInputBuffer(const int32_t index, const float *buffer) { @@ -52,8 +56,8 @@ void Model::SetInputBuffer(const int32_t index, const void *buffer, const size_t elemsize) { if (!prepared_for_exe_) PrepareForExecution(); auto size = shaper_.GetSize(input_names_[index]) * elemsize; - THROW_ON_ERROR(ANeuralNetworksExecution_setInput(execution_, index, nullptr, - buffer, size)) + THROW_ON_ERROR(nnapi_->ANeuralNetworksExecution_setInput( + execution_, index, nullptr, buffer, size)) } void Model::SetOutputBuffer(const int32_t index, float *buffer) { @@ -72,17 +76,18 @@ void Model::SetOutputBuffer(const int32_t index, void *buffer, const size_t elemsize) { if (!prepared_for_exe_) PrepareForExecution(); auto size = shaper_.GetSize(output_names_[index]) * elemsize; - THROW_ON_ERROR(ANeuralNetworksExecution_setOutput(execution_, index, - nullptr, buffer, size)) + THROW_ON_ERROR(nnapi_->ANeuralNetworksExecution_setOutput( + execution_, index, nullptr, buffer, size)) } void Model::PredictAfterSetInputBuffer() { ANeuralNetworksEvent *event = nullptr; - THROW_ON_ERROR(ANeuralNetworksExecution_startCompute(execution_, &event)); - THROW_ON_ERROR(ANeuralNetworksEvent_wait(event)); + THROW_ON_ERROR( + nnapi_->ANeuralNetworksExecution_startCompute(execution_, &event)); + THROW_ON_ERROR(nnapi_->ANeuralNetworksEvent_wait(event)); - ANeuralNetworksEvent_free(event); - ANeuralNetworksExecution_free(execution_); + nnapi_->ANeuralNetworksEvent_free(event); + nnapi_->ANeuralNetworksExecution_free(execution_); prepared_for_exe_ = false; } diff --git a/dnnlibrary/ModelBuilder.cpp b/dnnlibrary/ModelBuilder.cpp index 728c18e..09ba49e 100644 --- a/dnnlibrary/ModelBuilder.cpp +++ b/dnnlibrary/ModelBuilder.cpp @@ -30,41 +30,6 @@ using std::stringstream; using std::vector; using namespace android::nn::wrapper; -// Copy from -// https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/nnapi/nnapi_implementation.cc -int32_t ModelBuilder::GetAndroidSdkVersion() { - const char *sdkProp = "ro.build.version.sdk"; - char sdkVersion[PROP_VALUE_MAX]; - int length = __system_property_get(sdkProp, sdkVersion); - if (length != 0) { - int32_t result = 0; - for (int i = 0; i < length; ++i) { - int digit = sdkVersion[i] - '0'; - if (digit < 0 || digit > 9) { - // Non-numeric SDK version, assume it's higher than expected; - return 0xffff; - } - result = result * 10 + digit; - } - // TODO(levp): remove once SDK gets updated to 29th level - // Upgrade SDK version for pre-release Q to be able to test - // functionality available from SDK level 29. - if (result == 28) { - char versionCodename[PROP_VALUE_MAX]; - const char *versionCodenameProp = "ro.build.version.codename"; - length = - __system_property_get(versionCodenameProp, versionCodename); - if (length != 0) { - if (versionCodename[0] == 'Q') { - return 29; - } - } - } - return result; - } - return 0; -} - void ModelBuilder::RegisterOperand(const std::string &name, ModelBuilder::Index index, const OperandType &operand_type) { @@ -539,7 +504,7 @@ ModelBuilder::Index ModelBuilder::AddDequantize(const std::string &input, input_indexes.push_back(input_idx); shaper_.Identity(input, output); const OperandType operand_type = - GetOperandType(Type::FLOAT32, shaper_[output]); + GetOperandType(Type::TENSOR_FLOAT32, shaper_[output]); const auto output_idx = AddOperation(ANEURALNETWORKS_DEQUANTIZE, input_indexes, operand_type)[0]; RegisterOperand(output, output_idx, operand_type); @@ -696,7 +661,7 @@ OperandType ModelBuilder::GetOperandType(const QuantInfo &quant_info, map_type##_operand_map_.end()) { \ const auto index = AddNewOperand({Type::op_type}); \ THROW_ON_ERROR_WITH_NOTE( \ - ANeuralNetworksModel_setOperandValue( \ + nnapi_->ANeuralNetworksModel_setOperandValue( \ dnn_model_->model_, index, &value, sizeof(value)), \ "value: " + std::to_string(value)); \ map_type##_operand_map_[value] = index; \ @@ -713,15 +678,15 @@ DEFINE_OPERAND_FROM_SCALAR(float, float32, FLOAT32); ModelBuilder::Index ModelBuilder::AddMissingOperand( const OperandType &operand_type) { const auto index = AddNewOperand(operand_type); - THROW_ON_ERROR(ANeuralNetworksModel_setOperandValue(dnn_model_->model_, - index, nullptr, 0)); + THROW_ON_ERROR(nnapi_->ANeuralNetworksModel_setOperandValue( + dnn_model_->model_, index, nullptr, 0)); return index; } ModelBuilder::Index ModelBuilder::AddNewOperand( const OperandType &operand_type) { - THROW_ON_ERROR(ANeuralNetworksModel_addOperand(dnn_model_->model_, - &operand_type.operandType)); + THROW_ON_ERROR(nnapi_->ANeuralNetworksModel_addOperand( + dnn_model_->model_, &operand_type.operandType)); return next_index_++; } @@ -732,7 +697,7 @@ ModelBuilder::Index ModelBuilder::AddTensorFromMemory(const string &name, throw std::invalid_argument(""); DNN_ASSERT(!dimen.empty(), ""); const auto index = AddNewOperand({Type::TENSOR_FLOAT32, dimen}); - THROW_ON_ERROR(ANeuralNetworksModel_setOperandValueFromMemory( + THROW_ON_ERROR(nnapi_->ANeuralNetworksModel_setOperandValueFromMemory( dnn_model_->model_, index, dnn_model_->memory_, addr - dnn_model_->data_, Product(dimen) * sizeof(float))); shaper_.AddShape(name, dimen); @@ -775,7 +740,7 @@ ModelBuilder::Index ModelBuilder::AddTensorFromBuffer( typeToStr(operand_type.type)); } uint32_t index = AddNewOperand(operand_type); - THROW_ON_ERROR(ANeuralNetworksModel_setOperandValue( + THROW_ON_ERROR(nnapi_->ANeuralNetworksModel_setOperandValue( dnn_model_->model_, index, buffer, Product(operand_type.dimensions) * element_size)); shaper_.AddShape(name, operand_type.dimensions); @@ -797,27 +762,28 @@ std::unique_ptr ModelBuilder::Compile(uint32_t preference) { } } THROW_ON_ERROR_WITH_NOTE( - ANeuralNetworksModel_identifyInputsAndOutputs( + nnapi_->ANeuralNetworksModel_identifyInputsAndOutputs( dnn_model_->model_, static_cast(input_index_vec_.size()), &input_index_vec_[0], static_cast(output_index_vec_.size()), &output_index_vec_[0]), "on identifyInputsAndOutputs"); - THROW_ON_ERROR_WITH_NOTE(ANeuralNetworksModel_finish(dnn_model_->model_), - "on model finish"); + THROW_ON_ERROR_WITH_NOTE( + nnapi_->ANeuralNetworksModel_finish(dnn_model_->model_), + "on model finish"); ; - THROW_ON_ERROR_WITH_NOTE(ANeuralNetworksCompilation_create( + THROW_ON_ERROR_WITH_NOTE(nnapi_->ANeuralNetworksCompilation_create( dnn_model_->model_, &dnn_model_->compilation_), "on create"); - THROW_ON_ERROR_WITH_NOTE(ANeuralNetworksCompilation_setPreference( + THROW_ON_ERROR_WITH_NOTE(nnapi_->ANeuralNetworksCompilation_setPreference( dnn_model_->compilation_, preference), "on setPreference"); THROW_ON_ERROR_WITH_NOTE( - ANeuralNetworksCompilation_finish(dnn_model_->compilation_), + nnapi_->ANeuralNetworksCompilation_finish(dnn_model_->compilation_), "on compilation finish"); VLOG(5) << "Finishing.. Here are operands in the model:"; @@ -904,7 +870,7 @@ ModelBuilder::IndexSeq ModelBuilder::AddOperation( } THROW_ON_ERROR_WITH_NOTE( - ANeuralNetworksModel_addOperation( + nnapi_->ANeuralNetworksModel_addOperation( dnn_model_->model_, op, input_indexes.size(), &input_indexes[0], output_indexes.size(), &output_indexes[0]), "op = " + std::to_string(op)); @@ -914,7 +880,7 @@ ModelBuilder::IndexSeq ModelBuilder::AddOperation( void ModelBuilder::Prepare() { dnn_model_ = std::unique_ptr(new Model()); - const auto ret = ANeuralNetworksModel_create(&dnn_model_->model_); + const auto ret = nnapi_->ANeuralNetworksModel_create(&dnn_model_->model_); if (ret == ANEURALNETWORKS_OUT_OF_MEMORY) { throw std::bad_alloc(); } @@ -922,8 +888,8 @@ void ModelBuilder::Prepare() { void ModelBuilder::SetMemory(int fd, size_t size, size_t offset) { ANeuralNetworksMemory *mem = nullptr; - THROW_ON_ERROR( - ANeuralNetworksMemory_createFromFd(size, PROT_READ, fd, offset, &mem)); + THROW_ON_ERROR(nnapi_->ANeuralNetworksMemory_createFromFd( + size, PROT_READ, fd, offset, &mem)); dnn_model_->memory_ = mem; } @@ -938,10 +904,14 @@ ModelBuilder &ModelBuilder::AddOutput(const std::string &name) { } ModelBuilder &ModelBuilder::AllowFp16(const bool allowed) { - if (GetAndroidSdkVersion() >= __ANDROID_API_P__) { - ANeuralNetworksModel_relaxComputationFloat32toFloat16( + if (nnapi_->ANeuralNetworksModel_relaxComputationFloat32toFloat16 != + nullptr) { + nnapi_->ANeuralNetworksModel_relaxComputationFloat32toFloat16( dnn_model_->model_, allowed); } return *this; } + +ModelBuilder::ModelBuilder() : nnapi_(NnApiImplementation()) { +} } // namespace dnn diff --git a/dnnlibrary/NeuralNetworksWrapper.cpp b/dnnlibrary/NeuralNetworksWrapper.cpp index 5638539..57be2c9 100644 --- a/dnnlibrary/NeuralNetworksWrapper.cpp +++ b/dnnlibrary/NeuralNetworksWrapper.cpp @@ -14,7 +14,7 @@ OperandType::OperandType(Type type, std::vector d, float scale, dimensions = {1}; } } else { - DNN_ASSERT(!isScalarType(type), typeToStr(type)); + DNN_ASSERT(!isScalarType(type), typeToStr(type), " ", dimensions); } operandType = { .type = static_cast(type), diff --git a/dnnlibrary/nnapi_helper.h b/dnnlibrary/nnapi_helper.h index 0d07aaf..a85c2ec 100644 --- a/dnnlibrary/nnapi_helper.h +++ b/dnnlibrary/nnapi_helper.h @@ -3,7 +3,7 @@ #include -#include +#include #define THROW_ON_ERROR(val) \ { \ diff --git a/dnnlibrary/nnapi_implementation.cc b/dnnlibrary/nnapi_implementation.cc new file mode 100644 index 0000000..f5a4d8a --- /dev/null +++ b/dnnlibrary/nnapi_implementation.cc @@ -0,0 +1,225 @@ +/* Copyright 2017 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ +#include + +#include +#include +#include +#include +#include + +#include + +#ifdef __ANDROID__ +#include +#endif // __ANDROID__ + +#define NNAPI_LOG(format, ...) fprintf(stderr, format "\n", __VA_ARGS__); + +#ifdef __ANDROID__ +int32_t GetAndroidSdkVersion() { + const char* sdkProp = "ro.build.version.sdk"; + char sdkVersion[PROP_VALUE_MAX]; + int length = __system_property_get(sdkProp, sdkVersion); + if (length != 0) { + int32_t result = 0; + for (int i = 0; i < length; ++i) { + int digit = sdkVersion[i] - '0'; + if (digit < 0 || digit > 9) { + // Non-numeric SDK version, assume it's higher than expected; + return 0xffff; + } + result = result * 10 + digit; + } + // TODO(levp): remove once SDK gets updated to 29th level + // Upgrade SDK version for pre-release Q to be able to test functionality + // available from SDK level 29. + if (result == 28) { + char versionCodename[PROP_VALUE_MAX]; + const char* versionCodenameProp = "ro.build.version.codename"; + length = __system_property_get(versionCodenameProp, versionCodename); + if (length != 0) { + if (versionCodename[0] == 'Q') { + return 29; + } + } + } + return result; + } + return 0; +} +#endif // __ANDROID__ + +namespace { + +void* LoadFunction(void* handle, const char* name, bool optional) { + if (handle == nullptr) { + return nullptr; + } + void* fn = dlsym(handle, name); + if (fn == nullptr && !optional) { + NNAPI_LOG("nnapi error: unable to open function %s", name); + } + return fn; +} + +#ifndef __ANDROID__ +// Add /dev/shm implementation of shared memory for non-Android platforms +int ASharedMemory_create(const char* name, size_t size) { + int fd = shm_open(name, O_RDWR | O_CREAT, 0644); + if (fd < 0) { + return fd; + } + int result = ftruncate(fd, size); + if (result < 0) { + close(fd); + return -1; + } + return fd; +} +#endif // __ANDROID__ + +#define LOAD_FUNCTION(handle, name) \ + nnapi.name = reinterpret_cast( \ + LoadFunction(handle, #name, /*optional*/ false)); + +#define LOAD_FUNCTION_OPTIONAL(handle, name) \ + nnapi.name = reinterpret_cast( \ + LoadFunction(handle, #name, /*optional*/ true)); + +#define LOAD_FUNCTION_RENAME(handle, name, symbol) \ + nnapi.name = reinterpret_cast( \ + LoadFunction(handle, symbol, /*optional*/ false)); + +const NnApi LoadNnApi() { + NnApi nnapi = {}; + nnapi.android_sdk_version = 0; + +#ifdef __ANDROID__ + nnapi.android_sdk_version = GetAndroidSdkVersion(); + if (nnapi.android_sdk_version < 27) { + NNAPI_LOG("nnapi error: requires android sdk version to be at least %d", + 27); + nnapi.nnapi_exists = false; + return nnapi; + } +#endif // __ANDROID__ + + void* libneuralnetworks = nullptr; + // TODO(b/123243014): change RTLD_LOCAL? Assumes there can be multiple + // instances of nn api RT + libneuralnetworks = dlopen("libneuralnetworks.so", RTLD_LAZY | RTLD_LOCAL); + if (libneuralnetworks == nullptr) { + NNAPI_LOG("nnapi error: unable to open library %s", "libneuralnetworks.so"); + } + + nnapi.nnapi_exists = libneuralnetworks != nullptr; + + // API 27 (NN 1.0) methods. + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksMemory_createFromFd); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksMemory_free); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksModel_create); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksModel_free); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksModel_finish); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksModel_addOperand); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksModel_setOperandValue); + LOAD_FUNCTION_OPTIONAL( + libneuralnetworks, + ANeuralNetworksModel_setOperandSymmPerChannelQuantParams); + LOAD_FUNCTION(libneuralnetworks, + ANeuralNetworksModel_setOperandValueFromMemory); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksModel_addOperation); + LOAD_FUNCTION(libneuralnetworks, + ANeuralNetworksModel_identifyInputsAndOutputs); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksCompilation_create); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksCompilation_free); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksCompilation_setPreference); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksCompilation_finish); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksExecution_create); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksExecution_free); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksExecution_setInput); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksExecution_setInputFromMemory); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksExecution_setOutput); + LOAD_FUNCTION(libneuralnetworks, + ANeuralNetworksExecution_setOutputFromMemory); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksExecution_startCompute); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksEvent_wait); + LOAD_FUNCTION(libneuralnetworks, ANeuralNetworksEvent_free); + + // ASharedMemory_create has different implementations in Android depending on + // the partition. Generally it can be loaded from libandroid.so but in vendor + // partition (e.g. if a HAL wants to use NNAPI) it is only accessible through + // libcutils. +#ifdef __ANDROID__ + void* libandroid = nullptr; + libandroid = dlopen("libandroid.so", RTLD_LAZY | RTLD_LOCAL); + if (libandroid != nullptr) { + LOAD_FUNCTION(libandroid, ASharedMemory_create); + } else { + void* cutils_handle = dlopen("libcutils.so", RTLD_LAZY | RTLD_LOCAL); + if (cutils_handle != nullptr) { + LOAD_FUNCTION_RENAME(cutils_handle, ASharedMemory_create, + "ashmem_create_region"); + } else { + NNAPI_LOG("nnapi error: unable to open neither libraries %s and %s", + "libandroid.so", "libcutils.so"); + } + } +#else + nnapi.ASharedMemory_create = ASharedMemory_create; +#endif // __ANDROID__ + + // API 28 (NN 1.1) methods. + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksModel_relaxComputationFloat32toFloat16); + + // API 29 (NN 1.2) methods. + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworks_getDeviceCount); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworks_getDevice); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworksDevice_getName); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworksDevice_getVersion); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksDevice_getFeatureLevel); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworksDevice_getType); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksModel_getSupportedOperationsForDevices); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksCompilation_createForDevices); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksCompilation_setCaching); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworksExecution_compute); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksExecution_getOutputOperandRank); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksExecution_getOutputOperandDimensions); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworksBurst_create); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, ANeuralNetworksBurst_free); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksExecution_burstCompute); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksMemory_createFromAHardwareBuffer); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksExecution_setMeasureTiming); + LOAD_FUNCTION_OPTIONAL(libneuralnetworks, + ANeuralNetworksExecution_getDuration); + return nnapi; +} + +} // namespace + +const NnApi* NnApiImplementation() { + static const NnApi nnapi = LoadNnApi(); + return &nnapi; +} diff --git a/include/dnnlibrary/Model.h b/include/dnnlibrary/Model.h index 3b320b4..91761a2 100644 --- a/include/dnnlibrary/Model.h +++ b/include/dnnlibrary/Model.h @@ -18,12 +18,12 @@ class Model { friend class ModelBuilder; private: - ANeuralNetworksModel *model_; - ANeuralNetworksCompilation *compilation_; - ANeuralNetworksExecution *execution_; - ANeuralNetworksMemory *memory_; - unsigned char *data_; - size_t data_size_; + ANeuralNetworksModel *model_ = nullptr; + ANeuralNetworksCompilation *compilation_ = nullptr; + ANeuralNetworksExecution *execution_ = nullptr; + ANeuralNetworksMemory *memory_ = nullptr; + unsigned char *data_ = nullptr; + size_t data_size_ = 0; std::vector> uint8_buf_pointers_; std::vector> int8_buf_pointers_; std::vector> float_buf_pointers_; @@ -39,8 +39,9 @@ class Model { const size_t elemsize); void PrepareForExecution(); void PredictAfterSetInputBuffer(); - bool prepared_for_exe_; - Model() = default; + bool prepared_for_exe_ = false; + const NnApi *nnapi_ = nullptr; + Model(); public: template diff --git a/include/dnnlibrary/ModelBuilder.h b/include/dnnlibrary/ModelBuilder.h index fd04699..2cdc92c 100644 --- a/include/dnnlibrary/ModelBuilder.h +++ b/include/dnnlibrary/ModelBuilder.h @@ -86,7 +86,11 @@ class ModelBuilder { android::nn::wrapper::OperandType GetOperandType( const QuantInfo &quant_info, const Shape &dims); + + const NnApi* nnapi_ = nullptr; + public: + ModelBuilder(); enum class PoolingType { MAX_POOL, AVE_POOL }; static const int32_t ACTIVATION_NONE = ANEURALNETWORKS_FUSED_NONE; diff --git a/include/dnnlibrary/NeuralNetworksMock.h b/include/dnnlibrary/NeuralNetworksMock.h deleted file mode 100644 index fe253cc..0000000 --- a/include/dnnlibrary/NeuralNetworksMock.h +++ /dev/null @@ -1,5820 +0,0 @@ -/* - * Copyright (C) 2017 The Android Open Source Project - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ -/** - * @addtogroup NeuralNetworks - * @{ - */ -/** - * @file NeuralNetworks.h - */ -#ifndef ANDROID_ML_NN_RUNTIME_NEURAL_NETWORKS_MOCK_H -#define ANDROID_ML_NN_RUNTIME_NEURAL_NETWORKS_MOCK_H -/****************************************************************** - * - * IMPORTANT NOTICE: - * - * This file is part of Android's set of stable system headers - * exposed by the Android NDK (Native Development Kit). - * - * Third-party source AND binary code relies on the definitions - * here to be FROZEN ON ALL UPCOMING PLATFORM RELEASES. - * - * - DO NOT MODIFY ENUMS (EXCEPT IF YOU ADD NEW 32-BIT VALUES) - * - DO NOT MODIFY CONSTANTS OR FUNCTIONAL MACROS - * - DO NOT CHANGE THE SIGNATURE OF FUNCTIONS IN ANY WAY - * - DO NOT CHANGE THE LAYOUT OR SIZE OF STRUCTURES - */ -#include -#include -#include -#include -__BEGIN_DECLS -/** - * Operand types. - * - * The type of operands that can be added to a model. - * - * Although we define many types, most operators accept just a few - * types. Most used are {@link ANEURALNETWORKS_TENSOR_FLOAT32}, - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * and {@link ANEURALNETWORKS_INT32}. - * - * Available since API level 27. - */ -typedef enum { - /** A 32 bit floating point scalar value. */ - ANEURALNETWORKS_FLOAT32 = 0, - /** A signed 32 bit integer scalar value. */ - ANEURALNETWORKS_INT32 = 1, - /** An unsigned 32 bit integer scalar value. */ - ANEURALNETWORKS_UINT32 = 2, - /** A tensor of 32 bit floating point values. */ - ANEURALNETWORKS_TENSOR_FLOAT32 = 3, - /** A tensor of 32 bit integer values. */ - ANEURALNETWORKS_TENSOR_INT32 = 4, - /** - * A tensor of 8 bit unsigned integers that represent real numbers. - * - * Attached to this tensor are two numbers that be used to convert the - * 8 bit integer to the real value and vice versa. These two numbers are: - * - scale: a 32 bit floating point value greater than zero. - * - zeroPoint: a 32 bit integer, in range [0, 255]. - * - * The formula is: - * real_value = (integer_value - zeroPoint) * scale. - */ - ANEURALNETWORKS_TENSOR_QUANT8_ASYMM = 5, - /** - * An 8 bit boolean scalar value. - * - * Values of this operand type are either true or false. A zero value - * represents false; any other value represents true. - * - * Available since API level 29. - */ - ANEURALNETWORKS_BOOL = 6, - /** - * A tensor of 16 bit signed integers that represent real numbers. - * - * Attached to this tensor is a number representing real value scale that is - * used to convert the 16 bit number to a real value in the following way: - * realValue = integerValue * scale. - * - * scale is a 32 bit floating point with value greater then zero. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TENSOR_QUANT16_SYMM = 7, - /** - * A tensor of IEEE 754 16 bit floating point values. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TENSOR_FLOAT16 = 8, - /** - * A tensor of 8 bit boolean values. - * - * Values of this operand type are either true or false. A zero value - * represents false; any other value represents true. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TENSOR_BOOL8 = 9, - /** - * An IEEE 754 16 bit floating point scalar value. - * - * Available since API level 29. - */ - ANEURALNETWORKS_FLOAT16 = 10, - /** - * A tensor of 8 bit signed integers that represent real numbers. - * - * This tensor is associated with additional fields that can - * be used to convert the 8 bit signed integer to the real value and vice versa. - * These fields are: - * - channelDim: a 32 bit unsigned integer indicating channel dimension. - * - scales: an array of positive 32 bit floating point values. - * The size of the scales array must be equal to dimensions[channelDim]. - * - * {@link ANeuralNetworksModel_setOperandSymmPerChannelQuantParams} must be used - * to set the parameters for an Operand of this type. - * - * The channel dimension of this tensor must not be unknown (dimensions[channelDim] != 0). - * - * The formula is: - * realValue[..., C, ...] = - * integerValue[..., C, ...] * scales[C] - * where C is an index in the Channel dimension. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL = 11, - /** - * A tensor of 16 bit unsigned integers that represent real numbers. - * - * Attached to this tensor are two numbers that can be used to convert the - * 16 bit integer to the real value and vice versa. These two numbers are: - * - scale: a 32 bit floating point value greater than zero. - * - zeroPoint: a 32 bit integer, in range [0, 65535]. - * - * The formula is: - * real_value = (integer_value - zeroPoint) * scale. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TENSOR_QUANT16_ASYMM = 12, - /** - * A tensor of 8 bit signed integers that represent real numbers. - * - * Attached to this tensor is a number representing real value scale that is - * used to convert the 8 bit number to a real value in the following way: - * realValue = integerValue * scale. - * - * scale is a 32 bit floating point with value greater then zero. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TENSOR_QUANT8_SYMM = 13, -} OperandCode; -/** - * Operation types. - * - * The type of operations that can be added to a model. - * - * Available since API level 27. - */ -typedef enum { - // Operations below are available since API level 27. - /** - * Adds two tensors, element-wise. - * - * Takes two input tensors of identical {@link OperandCode} and compatible - * dimensions. The output is the sum of both input tensors, optionally - * modified by an activation function. - * - * Two dimensions are compatible when: - * 1. they are equal, or - * 2. one of them is 1 - * - * The size of the output is the maximum size along each dimension of the - * input operands. It starts with the trailing dimensions, and works its - * way forward. - * - * Example: - * - * input1.dimension = {4, 1, 2} - * input2.dimension = {5, 4, 3, 1} - * output.dimension = {5, 4, 3, 2} - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode}, and compatible dimensions - * as input0. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * - * Outputs: - * * 0: The sum, a tensor of the same {@link OperandCode} as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_ADD = 0, - /** - * Performs a 2-D average pooling operation. - * - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * The values in the output tensor are computed as: - * - * output[b, i, j, channel] = - * sum_{di, dj}( - * input[b, strides[1] * i + di, strides[2] * j + dj, channel] - * ) / sum(1) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * width. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * height. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 10: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * width. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * height. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 7: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_AVERAGE_POOL_2D = 1, - /** - * Concatenates the input tensors along the given dimension. - * - * The input tensors must have identical {@link OperandCode} and the same - * dimensions except the dimension along the concatenation axis. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} (full support since API - * level 29, see the input section) - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0 ~ n-1: The list of n input tensors, of shape - * [D0, D1, ..., Daxis(i), ..., Dm]. - * Before API level 29, all input tensors of - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * must have the same scale and zeroPoint as the output tensor. - * * n: An {@link ANEURALNETWORKS_INT32} scalar, specifying the - * concatenation axis. - * - * Outputs: - * * 0: The output, a tensor of the same {@link OperandCode} as the input - * tensors. The output shape is [D0, D1, ..., sum(Daxis(i)), ..., Dm]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_CONCATENATION = 2, - /** - * Performs an 2-D convolution operation. - * - * The CONV_2D op sweeps a 2-D filter that can mix channels together over a - * batch of images, applying the filter to each window of each image of the - * appropriate size. - * - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * The values in the output tensor are computed as: - * - * output[b, i, j, channel] = - * sum_{di, dj, k} ( - * input[b, strides[1] * i + di, strides[2] * j + dj, k] * - * filter[channel, di, dj, k] - * ) + bias[channel] - * - * Supported tensor {@link OperandCode} configurations: - * * 32 bit Floating point : - * * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} for input, filter, output, and bias. - * - * * Quantized: - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} for input, filter, and output. - * * * {@link ANEURALNETWORKS_TENSOR_INT32} for bias (with scale set to - * * * input.scale * filter.scale). - * - * Available since API level 29: - * * Quantized with symetric per channel quantization for the filter: - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} for input, and output. - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} for filter. - * * * {@link ANEURALNETWORKS_TENSOR_INT32} for bias (scale set to 0.0, - * * * each value scaling is separate and equal to input.scale * filter.scales[channel]). - * - * * 16 bit Floating point: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} for input, filter, output, and bias. - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: A 4-D tensor, of shape - * [depth_out, filter_height, filter_width, depth_in], specifying the - * filter. For tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} the channel - * dimension (extraParams.channelQuant.channelDim) must be set to 0. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias must be of the same - * type. For filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the bias should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint - * of 0 and bias_scale == input_scale * filter_scale. For filter tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}, the bias - * should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of - * 0 and bias_scale of 0. The actual scale of each value 'i' is equal to - * bias_scale[i] = input_scale * filter_scale[i]. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 10: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * * 11: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for width. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on width dimension. If this input is set, - * input 12 (dilation factor for height) must be specified as well. - * Available since API level 29. - * * 12: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for height. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on height dimension. If this input is set, - * input 11 (dilation factor for width) must be specified as well. - * Available since API level 29. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: A 4-D tensor, of shape - * [depth_out, filter_height, filter_width, depth_in], specifying the - * filter. For tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} the channel - * dimension (extraParams.channelQuant.channelDim) must be set to 0. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias must be of the same - * type. For filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the bias should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint - * of 0 and bias_scale == input_scale * filter_scale. For filter tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}, the bias - * should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of - * 0 and bias_scale of 0. The actual scale of each value 'i' is equal to - * bias_scale[i] = input_scale * filter_scale[i]. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 7: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * * 8: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for width. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on width dimension. If this input is set, - * input 9 (dilation factor for height) must be specified as well. - * Available since API level 29. - * * 9: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for height. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on height dimension. If this input is set, - * input 8 (dilation factor for width) must be specified as well. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth_out]. For output tensor of - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the following condition - * must be satisfied: output_scale > input_scale * filter_scale (for - * filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} - * this condition must be true for all filter scales). - * - * Available since API level 27. - */ - ANEURALNETWORKS_CONV_2D = 3, - /** - * Performs a depthwise 2-D convolution operation. - * - * Given an input tensor of shape [batches, height, width, depth_in] and a - * filter tensor of shape [1, filter_height, filter_width, depth_out] - * containing depth_out convolutional filters of depth 1, DEPTHWISE_CONV - * applies a different filter to each input channel (expanding from 1 - * channel to channel_multiplier channels for each), then concatenates the - * results together. - * - * The output has depth_out = depth_in * depth_multiplier channels. - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * The values in the output tensor are computed as: - * - * output[b, i, j, k * channel_multiplier + q] = - * sum_{di, dj} ( - * input[b, strides[1] * i + di, strides[2] * j + dj, k] * - * filter[1, di, dj, k * channel_multiplier + q] - * ) + bias[k * channel_multiplier + q] - * - * Supported tensor {@link OperandCode} configurations: - * * 32 bit Floating point : - * * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} for input, filter, output, and bias. - * - * * Quantized: - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} for input, filter, and output. - * * * {@link ANEURALNETWORKS_TENSOR_INT32} for bias (with scale set to - * * * input.scale * filter.scale). - * - * Available since API level 29: - * * Quantized with symetric per channel quantization for the filter: - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} for input, and output. - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} for filter. - * * * {@link ANEURALNETWORKS_TENSOR_INT32} for bias (scale set to 0.0, - * * * each value scaling is separate and equal to input.scale * filter.scales[channel]). - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: A 4-D tensor, of shape [1, filter_height, filter_width, depth_out], - * specifying the filter. For tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} the channel - * dimension (extraParams.channelQuant.channelDim) must be set to 3. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias must be of the same - * type. For filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the bias should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint - * of 0 and bias_scale == input_scale * filter_scale. For filter tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}, the bias - * should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of - * 0 and bias_scale of 0. The actual scale of each value 'i' is equal to - * bias_scale[i] = input_scale * filter_scale[i]. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, specifying the depthwise - * multiplier. - * * 10: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 11: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * * 12: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for width. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on width dimension. If this input is set, - * input 13 (dilation factor for height) must be specified as well. - * Available since API level 29. - * * 13: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for height. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on height dimension. If this input is set, - * input 12 (dilation factor for width) must be specified as well. - * Available since API level 29. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: A 4-D tensor, of shape [1, filter_height, filter_width, depth_out], - * specifying the filter. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias must be of the same - * type. For filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the bias should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint - * of 0 and bias_scale == input_scale * filter_scale. For filter tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}, the bias - * should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of - * 0 and bias_scale of 0. The actual scale of each value 'i' is equal to - * bias_scale[i] = input_scale * filter_scale[i]. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the depthwise - * multiplier. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 8: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * * 9: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for width. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on width dimension. If this input is set, - * input 10 (dilation factor for height) must be specified as well. - * Available since API level 29. - * * 10: An optional {@link ANEURALNETWORKS_INT32} scalar, specifying the dilation - * factor for height. Defaults to 1. If set to k > 1, there will be k-1 skipped - * cells between each filter element on height dimension. If this input is set, - * input 9 (dilation factor for width) must be specified as well. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth_out]. For output tensor of - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the following condition - * must be satisfied: output_scale > input_scale * filter_scale (for - * filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} - * this condition must be true for all filter scales). - * - * Available since API level 27. - */ - ANEURALNETWORKS_DEPTHWISE_CONV_2D = 4, - /** - * Rearranges data from depth into blocks of spatial data. - * - * More specifically, this op outputs a copy of the input tensor where - * values from the depth dimension are moved in spatial blocks to the height - * and width dimensions. The value block_size indicates the input block size - * and how the data is moved. - * - * Chunks of data of size block_size * block_size from depth are rearranged - * into non-overlapping blocks of size block_size x block_size. - * - * The width of the output tensor is input_depth * block_size, whereas the - * height is input_height * block_size. The depth of the input tensor must - * be divisible by block_size * block_size - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the block_size. - * block_size must be >=1 and block_size * block_size must be a divisor - * of the input depth. - * * 2: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape [batch, height*block_size, - * width*block_size, depth/(block_size*block_size)]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_DEPTH_TO_SPACE = 5, - /** - * Dequantizes the input tensor. - * - * The formula is: - * - * output = (input - zeroPoint) * scale. - * - * Supported input tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} (since API level 29) - * - * Supported output tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32}. - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: A tensor with the same shape as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_DEQUANTIZE = 6, - /** - * Looks up sub-tensors in the input tensor. - * - * This operator takes for input a tensor of values (Values) and - * a one-dimensional tensor of selection indices (Lookups). - * The output tensor is the concatenation of sub-tensors of Values as - * selected by Lookups. - * - * Think of Values as being sliced along its first dimension: - * The entries in Lookups select which slices are concatenated together - * to create the output tensor. - * - * For example, if Values has shape of [40, 200, 300] and - * Lookups has shape of [3], all three values found in Lookups are - * expected to be between 0 and 39. The resulting tensor must - * have shape of [3, 200, 300]. - * - * If a value in Lookups is out of bounds, the operation must fail - * and an error must be reported. - * - * Supported value tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported value tensor rank: from 2 - * - * Inputs: - * * 0: Lookups. A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. - * The values are indices into the first dimension of Values. - * * 1: Values. An n-D tensor, where n >= 2, from which sub-tensors are - * extracted. - * - * Output: - * * 0: A n-D tensor with the same rank and shape as the Values - * tensor, except for the first dimension which has the same size - * as Lookups' only dimension. - * - * Available since API level 27. - */ - ANEURALNETWORKS_EMBEDDING_LOOKUP = 7, - /** - * Computes element-wise floor() on the input tensor. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor, of the same {@link OperandCode} and dimensions as - * the input tensor. - * - * Available since API level 27. - */ - ANEURALNETWORKS_FLOOR = 8, - /** - * Denotes a fully (densely) connected layer, which connects all elements - * in the input tensor with each element in the output tensor. - * - * This layer implements the operation: - * - * outputs = activation(inputs * weights’ + bias) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor of at least rank 2, specifying the input. If rank is - * greater than 2, then it gets flattened to a 2-D Tensor. The - * (flattened) 2-D Tensor is reshaped (if necessary) to - * [batch_size, input_size], where "input_size" corresponds to the - * number of inputs to the layer, matching the second dimension of - * weights, and "batch_size" is calculated by dividing the number of - * elements by "input_size". - * * 1: A 2-D tensor, specifying the weights, of shape - * [num_units, input_size], where "num_units" corresponds to the number - * of output nodes. - * * 2: A 1-D tensor, of shape [num_units], specifying the bias. For input - * tensor of {@link ANEURALNETWORKS_TENSOR_FLOAT32}, the bias should - * also be of {@link ANEURALNETWORKS_TENSOR_FLOAT32}. For input tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the bias should be - * of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of 0 and - * bias_scale == input_scale * filter_scale. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * - * Outputs: - * * 0: The output tensor, of shape [batch_size, num_units]. For output - * tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the following - * condition must be satisfied: - * output_scale > input_scale * filter_scale. - * - * Available since API level 27. - */ - ANEURALNETWORKS_FULLY_CONNECTED = 9, - /** - * Looks up sub-tensors in the input tensor using a key-value map. - * - * This operator takes for input a tensor of values (Values), - * a one-dimensional tensor of selection values (Lookups) and - * a one-dimensional tensor that maps these values to Values - * indexes. The output tensor is the concatenation of sub-tensors of - * Values as selected by Lookups via Keys. - * - * Think of Values as being sliced along its outer-most dimension. - * The output is a concatenation of selected slices, with one slice - * for each entry of Lookups. The slice selected is the one at the - * same index as the Maps entry that matches the value in Lookups. - * - * For a hit, the corresponding sub-tensor of Values is included - * in the Output tensor. For a miss, the corresponding sub-tensor in - * Output must have zero values. - * - * For example, if Values has shape of [40, 200, 300], - * Keys should have a shape of [40]. If Lookups tensor has shape - * of [3], three slices are being concatenated, so the resulting tensor - * must have the shape of [3, 200, 300]. If the first entry in Lookups - * has the value 123456, that value must be located in Keys tensor. - * If the sixth entry of Keys contains 123456, the sixth slice of Values - * must be selected. If no entry in Keys has 123456, a slice of zeroes - * must be concatenated. - * - * Supported value tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported value tensor rank: from 2 - * - * Inputs: - * * 0: Lookups. A 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor with - * shape [ k ]. - * * 1: Keys. A 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor with shape - * [ n ]; Keys and Values pair represent a map, i.e., the ith element - * in Keys (Keys[i]) is the key to select the ith sub-tensor in Values - * (Values[i]), where 0 <= i <= n-1. Keys tensor *MUST* be sorted in - * ascending order. - * * 2: Values. A tensor with shape of [ n, … ]; i.e., the first dimension - * must be n. - * - * Outputs: - * * 0: Output. A tensor with shape [ k …]. - * * 1: Hits. A boolean tensor with shape [ k ] indicates whether the lookup - * hits (True) or not (False). - * Stored as {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} with offset 0 - * and scale 1.0f. - * A non-zero byte represents True, a hit. A zero indicates otherwise. - * - * Available since API level 27. - */ - ANEURALNETWORKS_HASHTABLE_LOOKUP = 10, - /** - * Applies L2 normalization along the depth dimension. - * - * The values in the output tensor are computed as: - * - * output[batch, row, col, channel] = - * input[batch, row, col, channel] / - * sqrt(sum_{c} pow(input[batch, row, col, c], 2)) - * - * For input tensor with rank less than 4, independently normalizes each - * 1-D slice along dimension dim. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: up to 4 - * Tensors with rank less than 4 are only supported since API level 29. - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be normalized. - * * 1: An optional {@link ANEURALNETWORKS_INT32} scalar, default to -1, - * specifying the dimension normalization would be performed on. - * Negative index is used to specify axis from the end (e.g. -1 for - * the last axis). Must be in the range [-n, n). - * Available since API level 29. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} and same shape as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_L2_NORMALIZATION = 11, - /** - * Performs an 2-D L2 pooling operation. - * - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * The values in the output tensor are computed as: - * - * output[b, i, j, c] = - * sqrt(sum_{di, dj} pow(input[b, strides[1] * i + di, strides[2] * j + dj, c], 2) / - * sum(1)) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * width. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * height. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 10: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * width. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * height. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 7: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_L2_POOL_2D = 12, - /** - * Applies Local Response Normalization along the depth dimension. - * - * The 4-D input tensor is treated as a 3-D array of 1-D vectors (along the - * last dimension), and each vector is normalized independently. Within a - * given vector, each component is divided by the weighted, squared sum of - * inputs within depth_radius. - * - * The output is calculated using this formula: - * - * sqr_sum[a, b, c, d] = sum( - * pow(input[a, b, c, d - depth_radius : d + depth_radius + 1], 2)) - * output = input / pow((bias + alpha * sqr_sum), beta) - * - * For input tensor with rank less than 4, independently normalizes each - * 1-D slice along specified dimension. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: up to 4 - * Tensors with rank less than 4 are only supported since API level 29. - * - * Inputs: - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the radius of - * the normalization window. - * * 2: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the bias, must - * not be zero. - * * 3: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the scale - * factor, alpha. - * * 4: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the exponent, - * beta. - * * 5: An optional {@link ANEURALNETWORKS_INT32} scalar, default to -1, - * specifying the dimension normalization would be performed on. - * Negative index is used to specify axis from the end (e.g. -1 for - * the last axis). Must be in the range [-n, n). - * Available since API level 29. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_LOCAL_RESPONSE_NORMALIZATION = 13, - /** - * Computes sigmoid activation on the input tensor element-wise. - * - * The output is calculated using this formula: - * - * output = 1 / (1 + exp(-input)) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor, specifying the input. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * For {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the scale must be 1.f / 256 and the zeroPoint must be 0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_LOGISTIC = 14, - /** - * Projects an input to a bit vector via locality senstive hashing. - * - * Supported input tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported input tensor rank: from 1 - * - * Inputs: - * * 0: Hash functions. Dim.size == 2, DataType: Float. - * Tensor[0].Dim[0]: Number of hash functions. - * Tensor[0].Dim[1]: Number of projected output bits generated by each - * hash function. - * If the projection type is Sparse: - * Tensor[0].Dim[1] + ceil(log2(Tensor[0].Dim[0])) <= 32 - * - * * 1: Input. Dim.size >= 1, no restriction on DataType. - * * 2: Weight. Optional. Dim.size == 1, DataType: Float. - * If not set, each input element is considered to have the same weight - * of 1.0. - * Tensor[1].Dim[0] == Tensor[2].Dim[0] - * * 3: Type: - * Sparse: - * Value LSHProjectionType_SPARSE(=3) (since API level 29). - * Computed bit vector is considered to be sparse. - * Each output element is an int32 made up of multiple bits - * computed from hash functions. - * - * NOTE: To avoid collisions across hash functions, an offset value - * of k * (1 << Tensor[0].Dim[1]) will be added to each signature, - * where k is the index of the hash function. - * - * Value LSHProjectionType_SPARSE_DEPRECATED(=1). - * Legacy behavior that does not include the offset value. - * - * Dense: - * Value LSHProjectionType_DENSE(=2). - * Computed bit vector is considered to be dense. Each output - * element represents a bit and can take the value of either - * 0 or 1. - * - * Outputs: - * * 0: If the projection type is Sparse: - * Output.Dim == { Tensor[0].Dim[0] } - * A tensor of int32 that represents hash signatures, - * - * If the projection type is Dense: - * Output.Dim == { Tensor[0].Dim[0] * Tensor[0].Dim[1] } - * A flattened tensor that represents projected bit vectors. - * - * Available since API level 27. - * The offset value for sparse projections was added in API level 29. - */ - ANEURALNETWORKS_LSH_PROJECTION = 15, - /** - * Performs a single time step in a Long Short-Term Memory (LSTM) layer - * - * The LSTM operation is described by the following equations. - * - * \f{eqnarray*}{ - * i_t =& \sigma(W_{xi}x_t+W_{hi}h_{t-1}+W_{ci}C_{t-1}+b_i) & \\ - * f_t =& \sigma(W_{xf}x_t+W_{hf}h_{t-1}+W_{cf}C_{t-1}+b_f) & \\ - * C_t =& clip(f_t \odot C_{t-1} + i_t \odot - * g(W_{xc}x_t+W_{hc}h_{t-1}+b_c),\ t_{cell}) & \\ - * o_t =& \sigma(W_{xo}x_t+W_{ho}h_{t-1}+W_{co}C_t+b_o) & \\ - * & & \\ - * & clip(W_{proj}(o_t \odot g(C_t))+b_{proj},\ t_{proj}) - * & if\ there\ is\ a\ projection; \\ - * h_t =& & \\ - * & o_t \odot g(C_t) & otherwise. \\ - * \f} - * Where: - * * \f$x_t\f$ is the input, - * * \f$i_t\f$ is the input gate, - * * \f$f_t\f$ is the forget gate, - * * \f$C_t\f$ is the cell state, - * * \f$o_t\f$ is the output, - * * \f$h_t\f$ is the output state, - * * \f$\sigma\f$ is the logistic sigmoid function, - * * \f$g\f$ is the cell input and cell output activation function, usually - * \f$tahn\f$, - * * \f$W_{xi}\f$ is the input-to-input weight matrix, - * * \f$W_{hi}\f$ is the recurrent to input weight matrix, - * * \f$W_{ci}\f$ is the cell-to-input weight matrix, - * * \f$b_i\f$ is the input gate bias, - * * \f$W_{xf}\f$ is the input-to-forget weight matrix, - * * \f$W_{hf}\f$ is the recurrent-to-forget weight matrix, - * * \f$W_{cf}\f$ is the cell-to-forget weight matrix, - * * \f$b_f\f$ is the forget gate bias, - * * \f$W_{xc}\f$ is the input-to-cell weight matrix, - * * \f$W_{hc}\f$ is the recurrent-to-cell weight matrix, - * * \f$b_c\f$ is the cell bias, - * * \f$W_{xo}\f$ is the input-to-output weight matrix, - * * \f$W_{ho}\f$ is the recurrent-to-output weight matrix, - * * \f$W_{co}\f$ is the cell-to-output weight matrix, - * * \f$b_o\f$ is the output gate bias, - * * \f$W_{proj}\f$ is the projection weight matrix, - * * \f$b_{proj}\f$ is the projection bias, - * * \f$t_{cell}\f$ is the threshold for clipping the cell state, and - * * \f$t_{proj}\f$ is the threshold for clipping the projected output. - * * \f$\odot\f$ is the - * - * Hadamard product that takes two matrices and produces another - * matrix, each element of which is the product of the corresponding - * elements of the input matrices. - * - * Since API level 29 LSTM supports layer normalization. - * In case layer normalization is used, the inputs to internal activation - * functions (sigmoid and \f$g\f$) are normalized, rescaled and recentered - * following an approach from section 3.1 from - * https://arxiv.org/pdf/1607.06450.pdf - * - * The operation has the following independently optional inputs: - * * The input-to-input weights (\f$W_{xi}\f$), recurrent-to-input weights - * (\f$W_{hi}\f$), cell-to-input (\f$W_{ci}\f$) weights, and input gate - * bias (\f$b_i\f$) either all have values, or none of them have values - * (i.e., all set to null). If they have no values, coupling of input and - * forget gates (CIFG) is used, in which case the input gate (\f$i_t\f$) - * is calculated using the following equation instead. - * \f{eqnarray*}{ - * i_t = 1 - f_t - * \f} - * * The cell-to-forget weights (\f$W_{cf}\f$) and cell-to-output weights - * (\f$W_{co}\f$) either both have values or neither of them have values. - * If they have values, the peephole optimization is used. Additionally, - * if CIFG is not used, cell-to-input weights (\f$W_{ci}\f$) is also - * required to have values for peephole optimization. - * * The projection weights (\f$W_{proj}\f$) is required only for the - * recurrent projection layer, and should otherwise have no value. - * * The projection bias (\f$b_{proj}\f$) may (but not required to) have a - * value if the recurrent projection layer exists, and should otherwise - * have no value. - * * (API level >= 29) The four layer normalization weights either all have - * values or none of them have values. Layer normalization is used when - * values are present. - * - * References: - * - * The default non-peephole non-CIFG implementation is based on: - * http://www.bioinf.jku.at/publications/older/2604.pdf - * S. Hochreiter and J. Schmidhuber. "Long Short-Term Memory". Neural - * Computation, 9(8):1735-1780, 1997. - * - * The peephole implementation and projection layer is based on: - * https://research.google.com/pubs/archive/43905.pdf - * Hasim Sak, Andrew Senior, and Francoise Beaufays. "Long short-term memory - * recurrent neural network architectures for large scale acoustic - * modeling." INTERSPEECH, 2014. - * (However, the concept of peephole optimization was introduced in work - * prior to this paper.) - * - * The coupling of input and forget gate (CIFG) is based on: - * http://arxiv.org/pdf/1503.04069.pdf - * Greff et al. "LSTM: A Search Space Odyssey" - * - * The layer normalization is based on: - * https://arxiv.org/pdf/1607.06450.pdf - * Jimmy Ba et al. "Layer Normalization" - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * All input and output tensors must be of the same type. - * - * Inputs: - * * 0: The input (\f$x_t\f$). - * A 2-D tensor of shape [batch_size, input_size], where “batch_size” - * corresponds to the batching dimension, and “input_size” is the size - * of the input. - * * 1: The input-to-input weights (\f$W_{xi}\f$). Optional. - * A 2-D tensor of shape [num_units, input_size], where “num_units” - * corresponds to the number of cell units. - * * 2: The input-to-forget weights (\f$W_{xf}\f$). - * A 2-D tensor of shape [num_units, input_size]. - * * 3: The input-to-cell weights (\f$W_{xc}\f$). - * A 2-D tensor of shape [num_units, input_size]. - * * 4: The input-to-output weights (\f$W_{xo}\f$). - * A 2-D tensor of shape [num_units, input_size]. - * * 5: The recurrent-to-input weights (\f$W_{hi}\f$). Optional. - * A 2-D tensor of shape [num_units, output_size], where “output_size” - * corresponds to either the number of cell units (i.e., “num_units”), - * or the second dimension of the “projection_weights”, if defined. - * * 6: The recurrent-to-forget weights (\f$W_{hf}\f$). - * A 2-D tensor of shape [num_units, output_size]. - * * 7: The recurrent-to-cell weights (\f$W_{hc}\f$). - * A 2-D tensor of shape [num_units, output_size]. - * * 8: The recurrent-to-output weights (\f$W_{ho}\f$). - * A 2-D tensor of shape [num_units, output_size]. - * * 9: The cell-to-input weights (\f$W_{ci}\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 10:The cell-to-forget weights (\f$W_{cf}\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 11:The cell-to-output weights (\f$W_{co}\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 12:The input gate bias (\f$b_i\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 13:The forget gate bias (\f$b_f\f$). - * A 1-D tensor of shape [num_units]. - * * 14:The cell bias (\f$b_c\f$). - * A 1-D tensor of shape [num_units]. - * * 15:The output gate bias (\f$b_o\f$). - * A 1-D tensor of shape [num_units]. - * * 16:The projection weights (\f$W_{proj}\f$). Optional. - * A 2-D tensor of shape [output_size, num_units]. - * * 17:The projection bias (\f$b_{proj}\f$). Optional. - * A 1-D tensor of shape [output_size]. - * * 18:The output state (in) (\f$h_{t-1}\f$). - * A 2-D tensor of shape [batch_size, output_size]. - * * 19:The cell state (in) (\f$C_{t-1}\f$). - * A 2-D tensor of shape [batch_size, num_units]. - * * 20:The activation function (\f$g\f$). - * A value indicating the activation function: - *
    - *
  • 0: None; - *
  • 1: Relu; - *
  • 3: Relu6; - *
  • 4: Tanh; - *
  • 6: Sigmoid. - *
- * * 21:The clipping threshold (\f$t_{cell}\f$) for the cell state, such - * that values are bound within [-cell_clip, cell_clip]. If set to 0.0 - * then clipping is disabled. - * Until API level 29 this scalar must be of type {@link - * ANEURALNETWORKS_FLOAT32}. Since API level 29, if all the input - * tensors have type {@link ANEURALNETWORKS_TENSOR_FLOAT32}, this - * scalar must be of the type {@link ANEURALNETWORKS_FLOAT32}, - * otherwise if all the input tensors have the type {@link - * ANEURALNETWORKS_TENSOR_FLOAT16}, this scalar must be of type {@link - * ANEURALNETWORKS_FLOAT16}. - * * 22:The clipping threshold (\f$t_{proj}\f$) for the output from the - * projection layer, such that values are bound within - * [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. - * Until API level 29 this scalar must be of type {@link - * ANEURALNETWORKS_FLOAT32}. Since API level 29, if all the input - * tensors have type {@link ANEURALNETWORKS_TENSOR_FLOAT32}, this - * scalar must be of the type {@link ANEURALNETWORKS_FLOAT32}, - * otherwise if all the input tensors have the type {@link - * ANEURALNETWORKS_TENSOR_FLOAT16}, this scalar must be of type {@link - * ANEURALNETWORKS_FLOAT16}. - * Since API level 29 there are additional inputs to this op: - * * 23:The input layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at input gate. - * * 24:The forget layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at forget gate. - * * 25:The cell layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at cell gate. - * * 26:The output layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at output gate. - * - * Outputs: - * * 0: The scratch buffer. - * A 2-D tensor of shape [batch_size, num_units * 3] with CIFG, or - * [batch_size, num_units * 4] without CIFG. - * * 1: The output state (out) (\f$h_t\f$). - * A 2-D tensor of shape [batch_size, output_size]. - * * 2: The cell state (out) (\f$C_t\f$). - * A 2-D tensor of shape [batch_size, num_units]. - * * 3: The output (\f$o_t\f$). - * A 2-D tensor of shape [batch_size, output_size]. This is effectively - * the same as the current “output state (out)” value. - * - * Available since API level 27. - */ - ANEURALNETWORKS_LSTM = 16, - /** - * Performs an 2-D max pooling operation. - * - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * The values in the output tensor are computed as: - * - * output[b, i, j, channel] = - * max_{di, dj} ( - * input[b, strides[1] * i + di, strides[2] * j + dj, channel] - * ) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * width. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * height. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 10: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * width. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the filter - * height. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 7: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_MAX_POOL_2D = 17, - /** - * Multiplies two tensors, element-wise. - * - * Takes two input tensors of identical {@link OperandCode} and compatible - * dimensions. The output is the product of both input tensors, optionally - * modified by an activation function. - * - * Two dimensions are compatible when: - * 1. they are equal, or - * 2. one of them is 1 - * - * The size of the resulting output is the maximum size along each dimension - * of the input operands. It starts with the trailing dimensions, and works - * its way forward. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode}, and compatible dimensions - * as input0. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * - * Outputs: - * * 0: The product, a tensor of the same {@link OperandCode} as input0. - * For output tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the following condition must be satisfied: - * output_scale > input1_scale * input2_scale. - * - * Available since API level 27. - */ - ANEURALNETWORKS_MUL = 18, - /** - * Computes rectified linear activation on the input tensor element-wise. - * - * The output is calculated using this formula: - * - * output = max(0, input) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor, specifying the input. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_RELU = 19, - /** - * Computes rectified linear 1 activation on the input tensor element-wise. - * - * The output is calculated using this formula: - * - * output = min(1.f, max(-1.f, input)) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor, specifying the input. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_RELU1 = 20, - /** - * Computes rectified linear 6 activation on the input tensor element-wise. - * - * The output is calculated using this formula: - * - * output = min(6, max(0, input)) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor, specifying the input. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_RELU6 = 21, - /** - * Reshapes a tensor. - * - * Given tensor, this operation returns a tensor that has the same values as - * tensor, but with a newly specified shape. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor, specifying the tensor to be reshaped. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, defining the - * shape of the output tensor. The number of elements implied by shape - * must be the same as the number of elements in the input tensor. - * - * Outputs: - * * 0: The output tensor, of shape specified by the input shape. - * - * Available since API level 27. - */ - ANEURALNETWORKS_RESHAPE = 22, - /** - * Resizes images to given size using the bilinear interpretation. - * - * Resized images must be distorted if their output aspect ratio is not the - * same as input aspect ratio. The corner pixels of output may not be the - * same as corner pixels of input. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: A 4-D tensor, of shape [batches, height, width, depth], specifying - * the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the output - * height of the output tensor. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the output - * width of the output tensor. - * * 3: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, new_height, new_width, depth]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_RESIZE_BILINEAR = 23, - /** - * A basic recurrent neural network layer. - * - * This layer implements the operation: - * outputs = state = activation(inputs * input_weights + - * state * recurrent_weights + bias) - * - * Where: - * * “input_weights” is a weight matrix that multiplies the inputs; - * * “recurrent_weights” is a weight matrix that multiplies the current - * “state” which itself is the output from the previous time step - * computation; - * * “bias” is a bias vector (added to each output vector in the batch); - * * “activation” is the function passed as the “fused_activation_function” - * argument (if not “NONE”). - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * The input tensors must all be the same type. - * - * Inputs: - * * 0: input. - * A 2-D tensor of shape [batch_size, input_size], where “batch_size” - * corresponds to the batching dimension, and “input_size” is the size - * of the input. - * * 1: weights. - * A 2-D tensor of shape [num_units, input_size], where “num_units” - * corresponds to the number of units. - * * 2: recurrent_weights. - * A 2-D tensor of shape [num_units, num_units], with columns - * corresponding to the weights from each unit. - * * 3: bias. - * A 1-D tensor of shape [num_units]. - * * 4: hidden state (in). - * A 2-D tensor of shape [batch_size, num_units]. - * * 5: fused_activation_function. - * An optional {@link FuseCode} value indicating the - * activation function. If “NONE” is specified then it results in a - * linear activation. - * - * Outputs: - * * 0: hidden state (out). - * A 2-D tensor of shape [batch_size, num_units]. - * - * * 1: output. - * A 2-D tensor of shape [batch_size, num_units]. This is effectively - * the same as the current state value. - * - * Available since API level 27. - */ - ANEURALNETWORKS_RNN = 24, - /** - * Computes the softmax activation on the input tensor element-wise, per - * batch, by normalizing the input vector so the maximum coefficient is - * zero. - * - * The output is calculated using this formula: - * - * output[batch, i] = - * exp((input[batch, i] - max(input[batch, :])) * beta) / - * sum_{k}{exp((input[batch, k] - max(input[batch, :])) * beta)} - * - * For input tensor with rank other than 2, the activation will be applied - * independently on each 1-D slice along specified dimension. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4. - * Tensors with rank other than 2 or 4 are only supported since API level 29. - * - * Inputs: - * * 0: A 2-D or 4-D tensor, specifying the tensor to be reshaped. - * * 1: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the positive - * scaling factor for the exponent, beta. - * * 2: An optional {@link ANEURALNETWORKS_INT32} scalar, default to -1, - * specifying the dimension the activation would be performed on. - * Negative index is used to specify axis from the end (e.g. -1 for - * the last axis). Must be in the range [-n, n). - * Available since API level 29. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * For {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the scale must be 1.f / 256 and the zeroPoint must be 0. - * - * Available since API level 27. - */ - ANEURALNETWORKS_SOFTMAX = 25, - /** - * Rearranges blocks of spatial data, into depth. - * - * More specifically, this op outputs a copy of the input tensor where - * values from the height and width dimensions are moved to the depth - * dimension. The value block_size indicates the input block size and how - * the data is moved. - * - * Chunks of data of size block_size * block_size from depth are rearranged - * into non-overlapping blocks of size block_size x block_size. - * - * The depth of the output tensor is input_depth * block_size * block_size. - * The input tensor's height and width must be divisible by block_size. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the block_size. - * block_size must be >=1 and block_size must be a divisor of both the - * input height and width. - * * 2: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: The output 4-D tensor, of shape [batches, height/block_size, - * width/block_size, depth_in*block_size*block_size]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_SPACE_TO_DEPTH = 26, - /** - * SVDF op is a kind of stateful layer derived from the notion that a - * densely connected layer that's processing a sequence of input frames can - * be approximated by using a singular value decomposition of each of its - * nodes. The implementation is based on: - * - * https://research.google.com/pubs/archive/43813.pdf - * - * P. Nakkiran, R. Alvarez, R. Prabhavalkar, C. Parada. - * “Compressing Deep Neural Networks using a Rank-Constrained Topology”. - * INTERSPEECH, 2015. - * - * It processes the incoming input using a 2-stage filtering mechanism: - * * stage 1 performs filtering on the "features" dimension, whose outputs - * get pushed into a memory of fixed-size memory_size. - * * stage 2 performs filtering on the "time" dimension of the memory_size - * memoized outputs of stage 1. - * - * Specifically, for rank 1, this layer implements the operation: - * - * memory = push(conv1d(inputs, weights_feature, feature_dim, - * "ANEURALNETWORKS_PADDING_VALID")); - * outputs = activation(memory * weights_time + bias); - * - * Where: - * * “weights_feature” is a weights matrix that processes the inputs (by - * convolving the input with every “feature filter”), and whose outputs - * get pushed, stacked in order, into the fixed-size “memory” (the oldest - * entry gets dropped); - * * “weights_time” is a weights matrix that processes the “memory” (by a - * batched matrix multiplication on the num_units); - * * “bias” is an optional bias vector (added to each output vector in the - * batch); and - * * “activation” is the function passed as the “fused_activation_function” - * argument (if not “NONE”). - * - * Each rank adds a dimension to the weights matrices by means of stacking - * the filters. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * All input tensors must be the same type. - * - * Inputs: - * * 0: input. - * A 2-D tensor of shape [batch_size, input_size], where “batch_size” - * corresponds to the batching dimension, and “input_size” is the size - * of the input. - * * 1: weights_feature. - * A 2-D tensor of shape [num_units, input_size], where “num_units” - * corresponds to the number of units. - * * 2: weights_time. - * A 2-D tensor of shape [num_units, memory_size], where “memory_size” - * corresponds to the fixed-size of the memory. - * * 3: bias. - * An optional 1-D tensor of shape [num_units]. - * * 4: state (in). - * A 2-D tensor of shape [batch_size, (memory_size - 1) * num_units * rank]. - * * 5: rank. - * The rank of the SVD approximation. - * * 6: fused_activation_function. - * An optional {@link FuseCode} value indicating the - * activation function. If “NONE” is specified then it results in a - * linear activation. - * - * Outputs: - * * 0: state (out). - * A 2-D tensor of the same {@link OperandCode} as the inputs, with shape - * [batch_size, (memory_size - 1) * num_units * rank]. - * * 1: output. - * A 2-D tensor of the same {@link OperandCode} as the inputs, with shape - * [batch_size, num_units]. - * - * Available since API level 27. - */ - ANEURALNETWORKS_SVDF = 27, - /** - * Computes hyperbolic tangent of input tensor element-wise. - * - * The output is calculated using this formula: - * - * output = tanh(input) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} (since API level 29) - * - * Supported tensor rank: up to 4. - * - * Inputs: - * * 0: A tensor, specifying the input. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * For {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the scale must be 1.f / 128 and the zeroPoint must be 128. - * - * Available since API level 27. - */ - ANEURALNETWORKS_TANH = 28, - // Operations below are available since API level 28. - // TODO: make the description easier to understand. - /** - * BatchToSpace for N-dimensional tensors. - * - * This operation reshapes the batch dimension (dimension 0) into M + 1 - * dimensions of shape block_shape + [batch], interleaves these blocks back - * into the grid defined by the spatial dimensions [1, ..., M], to obtain a - * result with the same rank as the input. - * - * This is the reverse of SpaceToBatch. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be reshaped - * * 1: A 1-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, the block - * sizes for each spatial dimension of the input tensor. All values - * must be >= 1. - * * 2: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 28. - */ - ANEURALNETWORKS_BATCH_TO_SPACE_ND = 29, - /** - * Element-wise division of two tensors. - * - * Takes two input tensors of identical {@link OperandCode} and compatible - * dimensions. The output is the result of dividing the first input tensor - * by the second, optionally modified by an activation function. - * - * Two dimensions are compatible when: - * 1. they are equal, or - * 2. one of them is 1 - * - * The size of the output is the maximum size along each dimension of the - * input operands. It starts with the trailing dimensions, and works its way - * forward. - * - * Example: - * input1.dimension = {4, 1, 2} - * input2.dimension = {5, 4, 3, 1} - * output.dimension = {5, 4, 3, 2} - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the first input. - * * 1: A tensor of the same {@link OperandCode}, and compatible dimensions - * as input0. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 28. - */ - ANEURALNETWORKS_DIV = 30, - /** - * Computes the mean of elements across dimensions of a tensor. - * - * Reduces the input tensor along the given dimensions to reduce. Unless - * keep_dims is true, the rank of the tensor is reduced by 1 for each entry - * in axis. If keep_dims is true, the reduced dimensions are retained with - * length 1. - * - * If dimensions to reduce have no entries, all dimensions are reduced, and - * a tensor with a single element is returned. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: A tensor, specifying the input. - * * 1: A 1-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. If None (the default), reduces all dimensions. Must be in - * the range [-rank(input_tensor), rank(input_tensor)). - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, keep_dims. If positive, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 28. - */ - ANEURALNETWORKS_MEAN = 31, - /** - * Pads a tensor with zeros. - * - * This operation pads a tensor according to the specified paddings. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be padded. - * * 1: A 2-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, the paddings - * for each spatial dimension of the input tensor. The shape of the - * tensor must be {rank(input0), 2}. - * padding[i, 0] specifies the number of elements to be padded in the - * front of dimension i. - * padding[i, 1] specifies the number of elements to be padded after the - * end of dimension i. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. The - * output tensor has the same rank as input0, and each - * dimension of the output tensor has the same size as the - * corresponding dimension of the input tensor plus the size - * of the padding: - * output0.dimension[i] = - * padding[i, 0] + input0.dimension[i] + padding[i, 1] - * - * Available since API level 28. - */ - ANEURALNETWORKS_PAD = 32, - // TODO: make the description easier to understand. - /** - * SpaceToBatch for N-Dimensional tensors. - * - * This operation divides "spatial" dimensions [1, ..., M] of the input into - * a grid of blocks of shape block_shape, and interleaves these blocks with - * the "batch" dimension (0) such that in the output, the spatial dimensions - * [1, ..., M] correspond to the position within the grid, and the batch - * dimension combines both the position within a spatial block and the - * original batch position. Prior to division into blocks, the spatial - * dimensions of the input are optionally zero padded according to paddings. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: An n-D tensor, specifying the input. - * * 1: A 1-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, the block - * sizes for each spatial dimension of the input tensor. All values - * must be >= 1. - * * 2: A 2-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, the paddings - * for each spatial dimension of the input tensor. All values must be - * >= 0. The shape of the tensor must be {rank(input0), 2}. - * padding[i, 0] specifies the number of element to be padded in the - * front of dimension i. - * padding[i, 1] specifies the number of element to be padded after the - * end of dimension i. - * * 3: An optional {@link ANEURALNETWORKS_BOOL} scalar, default to false. - * Set to true to specify NCHW data layout for input0 and output0. - * Available since API level 29. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 28. - */ - ANEURALNETWORKS_SPACE_TO_BATCH_ND = 33, - /** - * Removes dimensions of size 1 from the shape of a tensor. - * - * Given a tensor input, this operation returns a tensor of the same - * {@link OperandCode} with all dimensions of size 1 removed. If you don't - * want to remove all size 1 dimensions, you can remove specific size 1 - * dimensions by specifying the axes (input1). - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, the tensor to be squeezed. - * * 1: An optional 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The - * dimensions to squeeze. If specified only squeezes the dimensions - * listed. Otherwise, squeezes all dimensions. The dimension index - * starts at 0. An error must be reported if squeezing a dimension that - * is not 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. Contains the - * same data as input, but has one or more dimensions of size 1 - * removed. - * - * Available since API level 28. - */ - ANEURALNETWORKS_SQUEEZE = 34, - /** - * Extracts a strided slice of a tensor. - * - * Roughly speaking, this op extracts a slice of size (end - begin) / stride - * from the given input tensor. Starting at the location specified by begin - * the slice continues by adding stride to the index until all dimensions - * are not less than end. Note that a stride can be negative, which causes a - * reverse slice. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be sliced. - * * 1: begin, a 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The - * starts of the dimensions of the input tensor to be sliced. The - * length must be of rank(input0). - * * 2: end, a 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The - * ends of the dimensions of the input tensor to be sliced. The length - * must be of rank(input0). - * * 3: strides, a 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The - * strides of the dimensions of the input tensor to be sliced. The - * length must be of rank(input0). The entries must be non-zero. - * * 4: begin_mask, an {@link ANEURALNETWORKS_INT32} scalar. If the ith bit - * of begin_mask is set, begin[i] is ignored and the fullest possible - * range in that dimension is used instead. - * * 5: end_mask, an {@link ANEURALNETWORKS_INT32} scalar. If the ith bit of - * end_mask is set, end[i] is ignored and the fullest possible range in - * that dimension is used instead. - * * 6: shrink_axis_mask, an {@link ANEURALNETWORKS_INT32} scalar. If the - * ith bit of shrink_axis_mask is set, the ith dimension specification - * shrinks the dimensionality by 1, taking on the value at index - * begin[i]. In this case, the ith specification must define a - * slice of size 1, e.g. begin[i] = x, end[i] = x + 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0 and rank (n - k), - * where k is the number of bits set in shrink_axis_mask. - * - * Available since API level 28. - */ - ANEURALNETWORKS_STRIDED_SLICE = 35, - /** - * Element-wise subtraction of two tensors. - * - * Takes two input tensors of identical {@link OperandCode} and compatible - * dimensions. The output is the result of subtracting the second input - * tensor from the first one, optionally modified by an activation function. - * - * Two dimensions are compatible when: - * 1. they are equal, or - * 2. one of them is 1 - * - * The size of the output is the maximum size along each dimension of the - * input operands. It starts with the trailing dimensions, and works its way - * forward. - * - * Example: - * input1.dimension = {4, 1, 2} - * input2.dimension = {5, 4, 3, 1} - * output.dimension = {5, 4, 3, 2} - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} (since API level 29) - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the first input. - * * 1: A tensor of the same {@link OperandCode}, and compatible dimensions - * as input0. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 28. - */ - ANEURALNETWORKS_SUB = 36, - /** - * Transposes the input tensor, permuting the dimensions according to the - * perm tensor. - * - * The returned tensor's dimension i corresponds to the input dimension - * perm[i]. If perm is not given, it is set to (n-1...0), where n is the - * rank of the input tensor. Hence by default, this operation performs a - * regular matrix transpose on 2-D input Tensors. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be transposed. - * * 1: An optional 1-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, - * the permutation of the dimensions of the input tensor. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 28. - */ - ANEURALNETWORKS_TRANSPOSE = 37, - // Operations below are available since API level 29. - /** - * Computes the absolute value of a tensor, element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_ABS = 38, - /** - * Returns the index of the largest element along an axis. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: An n-D tensor specifying the input. Must be non-empty. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar specifying the axis to - * reduce across. Negative index is used to specify axis from the - * end (e.g. -1 for the last axis). Must be in the range [-n, n). - * - * Outputs: - * * 0: An (n - 1)-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor. - * - * Available since API level 29. - */ - // There is no underscore in ARG_MAX to avoid name conflict with - // the macro defined in libc/kernel/uapi/linux/limits.h. - ANEURALNETWORKS_ARGMAX = 39, - /** - * Returns the index of the smallest element along an axis. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: An n-D tensor specifying the input. Must be non-empty. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar specifying the axis to - * reduce across. Negative index is used to specify axis from the - * end (e.g. -1 for the last axis). Must be in the range [-n, n). - * - * Outputs: - * * 0: An (n - 1)-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor. - * - * Available since API level 29. - */ - ANEURALNETWORKS_ARGMIN = 40, // See ARGMAX for naming discussion. - /** - * Transform axis-aligned bounding box proposals using bounding box deltas. - * - * Given the positions of bounding box proposals and the corresponding - * bounding box deltas for each class, return the refined bounding box - * regions. The resulting bounding boxes are cliped against the edges of - * the image. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Inputs: - * * 0: A 2-D Tensor of shape [num_rois, 4], specifying the locations of the - * bounding box proposals, each line with format [x1, y1, x2, y2]. - * For tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, - * the zeroPoint must be 0 and the scale must be 0.125. - * * 1: A 2-D Tensor of shape [num_rois, num_classes * 4], specifying the - * bounding box delta for each region of interest and each class. The - * bounding box deltas are organized in the following order - * [dx, dy, dw, dh], where dx and dy is the relative correction factor - * for the center position of the bounding box with respect to the width - * and height, dw and dh is the log-scale relative correction factor - * for the width and height. For input0 of type - * {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, this tensor should be - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}. - * * 2: An 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [batches], specifying the number of output boxes for each batch. - * * 3: A 2-D Tensor of shape [batches, 2], specifying the information of - * each image in the batch, each line with format - * [image_height, image_width]. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0, with shape - * [num_rois, num_classes * 4], specifying the coordinates of each - * output bounding box for each class, with format [x1, y1, x2, y2]. - * - * Available since API level 29. - */ - ANEURALNETWORKS_AXIS_ALIGNED_BBOX_TRANSFORM = 41, - /** - * Performs a forward LSTM on the input followed by a backward LSTM. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: 3, either time-major or batch-major. - * - * All input and output tensors must be of the same type. - * - * - * Inputs: - * * 0: The input. - * A 3-D tensor of shape: - * If time-major: [max_time, batch_size, output_size] - * If batch-major: [batch_size, max_time, output_size] - * where "max_time" is the number of timesteps (sequence length), - * "batch_size" corresponds to the batching dimension, and - * "input_size" is the size of the input. - * * 1: The forward input-to-input weights. Optional. - * A 2-D tensor of shape [num_units, input_size], where “num_units” - * corresponds to the number of cell units. - * * 2: The forward input-to-forget weights. - * A 2-D tensor of shape [num_units, input_size]. - * * 3: The forward input-to-cell weights. - * A 2-D tensor of shape [num_units, input_size]. - * * 4: The forward input-to-output weights. - * A 2-D tensor of shape [num_units, input_size]. - * * 5: The forward recurrent-to-input weights. Optional. - * A 2-D tensor of shape [num_units, output_size], where “output_size” - * corresponds to either the number of cell units (i.e., “num_units”), - * or the second dimension of the “projection_weights”, if defined. - * * 6: The forward recurrent-to-forget weights. - * A 2-D tensor of shape [num_units, output_size]. - * * 7: The forward recurrent-to-cell weights. - * A 2-D tensor of shape [num_units, output_size]. - * * 8: The forward recurrent-to-output weights. - * A 2-D tensor of shape [num_units, output_size]. - * * 9: The forward cell-to-input weights. Optional. - * A 1-D tensor of shape [num_units]. - * * 10: The forward cell-to-forget weights. Optional. - * A 1-D tensor of shape [num_units]. - * * 11: The forward cell-to-output weights. Optional. - * A 1-D tensor of shape [num_units]. - * * 12: The forward input gate bias. Optional. - * A 1-D tensor of shape [num_units]. - * * 13: The forward forget gate bias. - * A 1-D tensor of shape [num_units]. - * * 14: The forward cell gate bias. - * A 1-D tensor of shape [num_units]. - * * 15: The forward output gate bias. - * A 1-D tensor of shape [num_units]. - * * 16: The forward projection weights. Optional. - * A 2-D tensor of shape [output_size, num_units]. - * * 17: The forward projection bias. Optional. - * A 1-D tensor of shape [output_size]. - * * 18: The backward input-to-input weights. Optional. - * A 2-D tensor of shape [num_units, input_size], where “num_units” - * corresponds to the number of cell units. - * * 19: The backward input-to-forget weights. - * A 2-D tensor of shape [num_units, input_size]. - * * 20: The backward input-to-cell weights. - * A 2-D tensor of shape [num_units, input_size]. - * * 21: The backward input-to-output weights. - * A 2-D tensor of shape [num_units, input_size]. - * * 22: The backward recurrent-to-input weights. Optional. - * A 2-D tensor of shape [num_units, output_size], where “output_size” - * corresponds to either the number of cell units (i.e., “num_units”), - * or the second dimension of the “projection_weights”, if defined. - * * 23: The backward recurrent-to-forget weights. - * A 2-D tensor of shape [num_units, output_size]. - * * 24: The backward recurrent-to-cell weights. - * A 2-D tensor of shape [num_units, output_size]. - * * 25: The backward recurrent-to-output weights. - * A 2-D tensor of shape [num_units, output_size]. - * * 26: The backward cell-to-input weights. Optional. - * A 1-D tensor of shape [num_units]. - * * 27: The backward cell-to-forget weights. Optional. - * A 1-D tensor of shape [num_units]. - * * 28: The backward cell-to-output weights. Optional. - * A 1-D tensor of shape [num_units]. - * * 29: The backward input gate bias. Optional. - * A 1-D tensor of shape [num_units]. - * * 30: The backward forget gate bias. - * A 1-D tensor of shape [num_units]. - * * 31: The backward cell gate bias. - * A 1-D tensor of shape [num_units]. - * * 32: The backward output gate bias. - * A 1-D tensor of shape [num_units]. - * * 33: The backward projection weights. Optional. - * A 2-D tensor of shape [output_size, num_units]. - * * 34: The backward projection bias. Optional. - * A 1-D tensor of shape [output_size]. - * * 35: The forward input activation state. - * A 2-D tensor of shape [batch_size, output_size]. - * * 36: The forward input cell state. - * A 2-D tensor of shape [batch_size, num_units]. - * * 37: The backward input activation state. - * A 2-D tensor of shape [batch_size, output_size]. - * * 38: The backward input cell state. - * A 2-D tensor of shape [batch_size, num_units]. - * * 39: The auxiliary input. Optional. - * A 3-D tensor of shape [max_time, batch_size, input_size], where “batch_size” - * corresponds to the batching dimension, and “input_size” is the size - * of the input. - * * 40: The forward auxiliary input-to-input weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 41: The forward auxiliary input-to-forget weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 42: The forward auxiliary input-to-cell weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 43: The forward auxiliary input-to-output weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 44: The backward auxiliary input-to-input weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 45: The backward auxiliary input-to-forget weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 46: The backward auxiliary input-to-cell weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 47: The backward auxiliary input-to-output weights. Optional. - * A 2-D tensor of shape [num_units, input_size]. - * * 48: The activation function. - * A value indicating the activation function: - *
    - *
  • 0: None; - *
  • 1: Relu; - *
  • 3: Relu6; - *
  • 4: Tanh; - *
  • 6: Sigmoid. - *
- * * 49: The clipping threshold for the cell state, such - * that values are bound within [-cell_clip, cell_clip]. If set to 0.0 - * then clipping is disabled. - * If all the input tensors have type {@link ANEURALNETWORKS_TENSOR_FLOAT32}, - * this scalar must be of the type {@link ANEURALNETOWORKS_FLOAT32}, - * otherwise if all the input tensors have the type {@link - * ANEURALNETWORKS_TENSOR_FLOAT16}, this scalar must be of type {@link - * ANEURALNETWORKS_FLOAT16}. - * * 50: The clipping threshold for the output from the - * projection layer, such that values are bound within - * [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. - * If all the input tensors have type {@link ANEURALNETWORKS_TENSOR_FLOAT32}, - * this scalar must be of the type {@link ANEURALNETOWORKS_FLOAT32}, - * otherwise if all the input tensors have the type {@link - * ANEURALNETWORKS_TENSOR_FLOAT16}, this scalar must be of type {@link - * ANEURALNETWORKS_FLOAT16}. - * * 51: merge_outputs - * An {@link ANEURALNETWORKS_BOOL} scalar specifying if the outputs - * from forward and backward cells should be merged. - * * 52: time_major - * An {@link ANEURALNETWORKS_BOOL} scalar specifying the shape format - * of input and output tensors. - * - * Outputs: - * * 0: The forward output. - * A 3-D tensor of shape: - * If time-major: [max_time, batch_size, output_size] - * If batch-major: [batch_size, max_time, output_size] - * * 1: The backward output. Unused if merge_outputs is true. - * A 3-D tensor of shape: - * If time-major: [max_time, batch_size, output_size] - * If batch-major: [batch_size, max_time, output_size] - * - * Available since API level 29. - */ - ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_LSTM = 42, - /** - * A recurrent neural network layer that applies a basic RNN cell to a - * sequence of inputs in forward and backward directions. - * - * This Op unrolls the input along the sequence dimension, and implements - * the following operation for each element in the sequence s = - * 1...sequence_length: - * fw_outputs[s] = fw_state = activation(inputs[s] * fw_input_weights’ + - * fw_state * fw_recurrent_weights’ + fw_bias) - * - * And for each element in sequence t = sequence_length : 1 - * bw_outputs[t] = bw_state = activation(inputs[t] * bw_input_weights’ + - * bw_state * bw_recurrent_weights’ + bw_bias) - * - * Where: - * * “{fw,bw}_input_weights” is a weight matrix that multiplies the inputs; - * * “{fw,bw}_recurrent_weights” is a weight matrix that multiplies the - * current “state” which itself is the output from the previous time step - * computation; - * * “{fw,bw}_bias” is a bias vector (added to each output vector in the - * batch); - * * “activation” is the function passed as the “fused_activation_function” - * argument (if not “NONE”). - * - * The op also supports an auxiliary input. Regular cell feeds one input - * into the two RNN cells in the following way: - * - * INPUT (INPUT_REVERSED) - * | | - * --------------------- - * | FW_RNN BW_RNN | - * --------------------- - * | | - * FW_OUT BW_OUT - * - * An op with an auxiliary input takes two inputs and feeds them into the - * RNN cells in the following way: - * - * AUX_INPUT (AUX_INPUT_REVERSED) - * | | - * INPUT | (INPUT_R'D.)| - * | | | | - * ----------------------- - * | \ / \ / | - * | FW_RNN BW_RNN | - * ----------------------- - * | | - * FW_OUT BW_OUT - * - * While stacking this op on top of itself, this allows to connect both - * forward and backward outputs from previous cell to the next cell's - * inputs. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * The input tensors must all be the same type. - * - * Inputs: - * * 0: input. - * A 3-D tensor. The shape is defined by the input 6 (timeMajor). If - * it is set to true, then the input has a shape [maxTime, batchSize, - * inputSize], otherwise the input has a shape [batchSize, maxTime, - * inputSize]. - * * 1: fwWeights. - * A 2-D tensor of shape [fwNumUnits, inputSize]. - * * 2: fwRecurrentWeights. - * A 2-D tensor of shape [fwNumUnits, fwNumUnits]. - * * 3: fwBias. - * A 1-D tensor of shape [fwNumUnits]. - * * 4: fwHiddenState. - * A 2-D tensor of shape [batchSize, fwNumUnits]. Specifies a hidden - * state input for the first time step of the computation. - * * 5: bwWeights. - * A 2-D tensor of shape [bwNumUnits, inputSize]. - * * 6: bwRecurrentWeights. - * A 2-D tensor of shape [bwNumUnits, bwNumUnits]. - * * 7: bwBias. - * A 1-D tensor of shape [bwNumUnits]. - * * 8: bwHiddenState - * A 2-D tensor of shape [batchSize, bwNumUnits]. Specifies a hidden - * state input for the first time step of the computation. - * * 9: auxInput. - * A 3-D tensor. The shape is the same as of the input 0. - * * 10:fwAuxWeights. - * A 2-D tensor of shape [fwNumUnits, inputSize]. - * * 11:bwAuxWeights. - * A 2-D tensor of shape [bwNumUnits, inputSize]. - * * 12:fusedActivationFunction. - * A {@link FuseCode} value indicating the activation function. If - * “NONE” is specified then it results in a linear activation. - * * 13:timeMajor - * An {@link ANEURALNETWORKS_BOOL} scalar specifying the shape format - * of input and output tensors. - * * 14:mergeOutputs - * An {@link ANEURALNETWORKS_BOOL} scalar specifying if the outputs - * from forward and backward cells are separate (if set to false) or - * concatenated (if set to true). - * Outputs: - * * 0: fwOutput. - * A 3-D tensor. The first two dimensions of the shape are defined by - * the input 6 (timeMajor) and the third dimension is defined by the - * input 14 (mergeOutputs). If timeMajor is set to true, then the first - * two dimensions are [maxTime, batchSize], otherwise they are set to - * [batchSize, maxTime]. If mergeOutputs is set to true, then the third - * dimension is equal to (fwNumUnits + bwNumUnits), otherwise it is set - * to fwNumUnits. - * * 1: bwOutput. - * A 3-D tensor. If the input 14 (mergeOutputs) is set to true, then - * this tensor is not produced. The shape is defined by the input 6 - * (timeMajor). If it is set to true, then the shape is set to - * [maxTime, batchSize, bwNumUnits], otherwise the shape is set to - * [batchSize, maxTime, bwNumUnits]. - * - * Available since API level 29. - */ - ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_RNN = 43, - /** - * Greedily selects a subset of bounding boxes in descending order of score. - * - * This op applies hard NMS algorithm to each class. In each loop of - * execution, the box with maximum score gets selected, and any boxes with - * the intersection-over-union (IOU) greater than a threshold are removed - * from the pending set. - * - * Axis-aligned bounding boxes are represented by its upper-left corner - * coordinate (x1,y1) and lower-right corner coordinate (x2,y2). A valid - * bounding box should satisfy x1 <= x2 and y1 <= y2. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Inputs: - * * 0: A 2-D Tensor of shape [num_rois, num_classes], specifying the score - * of each bounding box proposal. The boxes are grouped by batches in the - * first dimension. - * * 1: A 2-D Tensor specifying the bounding boxes of shape - * [num_rois, num_classes * 4], organized in the order [x1, y1, x2, y2]. - * The boxes are grouped by batches in the first dimension. The sequential - * order of the boxes corresponds with input0. For input0 of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, this tensor should be of - * {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, with zeroPoint of 0 and - * scale of 0.125. - * * 2: A 1-D Tensor of shape [batches], specifying the number of boxes - * for each image in the batch. - * * 3: An {@link ANEURALNETWORKS_FLOAT32} scalar, score_threshold. Boxes - * with scores lower than the threshold are filtered before sending - * to the NMS algorithm. - * * 4: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the IoU - * threshold. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the maximum - * number of selected bounding boxes for each image. Set to a negative - * value for unlimited number of output bounding boxes. - * - * Outputs: - * * 0: A 1-D Tensor of the same {@link OperandCode} as input0, with shape - * [num_output_rois], specifying the score of each output box. The boxes - * are grouped by batches, but the sequential order in each batch is not - * guaranteed. For type of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the scale and zero point must be the same as input0. - * * 1: A 2-D Tensor of the same {@link OperandCode} as input1, with shape - * [num_output_rois, 4], specifying the coordinates of each - * output bounding box with the same format as input1. The sequential - * order of the boxes corresponds with output0. For type of - * {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, the scale must be - * 0.125 and the zero point must be 0. - * * 2: A 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [num_output_rois], specifying the class of each output box. The - * sequential order of the boxes corresponds with output0. - * * 3: A 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [batches], specifying the number of output boxes for each image. - * - * Available since API level 29. - */ - ANEURALNETWORKS_BOX_WITH_NMS_LIMIT = 44, - /** - * Casts a tensor to a new type. - * - * This operation ignores the scale and zeroPoint of quanized tensors, - * e.g. it treats a {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} input - * as a tensor of uint8 values. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: A tensor with the same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_CAST = 45, - /** - * Shuffle the channels of the input tensor. - * - * Given an input tensor and a integer value of num_groups, CHANNEL_SHUFFLE - * divide the channel dimension into num_groups groups, and reorganize the - * channels by grouping channels with the same index in each group. - * - * Along the channel dimension, the output is calculated using this formula: - * - * output_channel[k * num_groups + g] = input_channel[g * group_size + k] - * - * where group_size = num_channels / num_groups - * - * The number of channels must be divisible by num_groups. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be shuffled. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar, specifying the number of - * groups. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar, specifying the dimension - * channel shuffle would be performed on. Negative index is used to - * specify axis from the end (e.g. -1 for the last axis). Must be in - * the range [-n, n). - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} and same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_CHANNEL_SHUFFLE = 46, - /** - * Apply postprocessing steps to bounding box detections. - * - * Bounding box detections are generated by applying transformation on a set - * of predefined anchors with the bounding box deltas from bounding box - * regression. A final step of hard NMS is applied to limit the number of - * returned boxes. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Inputs: - * * 0: A 3-D Tensor of shape [batches, num_anchors, num_classes], specifying - * the score of each anchor with each class. Class 0 for each - * [batches, num_anchors, 0] is background and will be ignored. - * * 1: A 3-D Tensor of shape [batches, num_anchors, length_box_encoding], with - * the first four values in length_box_encoding specifying the bounding - * box deltas. The box deltas are encoded in the order of [dy, dx, dh, dw], - * where dy and dx is the linear-scale relative correction factor for the - * center position of the bounding box with respect to the width and height, - * dh and dw is the log-scale relative correction factor for the width and - * height. All the entries in length_box_encoding beyond the first four - * values are ignored in this operation. - * * 2: A 2-D Tensor of shape [num_anchors, 4], specifying the shape of each - * predefined anchor, with format [ctr_y, ctr_x, h, w], where ctr_y and - * ctr_x are the center position of the box, and h and w are the height - * and the width. - * * 3: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the scaling - * factor for dy in bounding box deltas. - * * 4: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the scaling - * factor for dx in bounding box deltas. - * * 5: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the scaling - * factor for dh in bounding box deltas. - * * 6: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the scaling - * factor for dw in bounding box deltas. - * * 7: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to use regular - * multi-class NMS algorithm that do NMS separately for each class, - * set to false for a faster algorithm that only do one single NMS - * using the highest class score.. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, max_num_detections, specifying - * the maximum number of boxes for the output. Boxes with the lowest - * scores are discarded to meet the limit. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, only used when input7 is - * set to false, specifying the maximum number of classes per detection. - * * 10: An {@link ANEURALNETWORKS_INT32} scalar, only used when input7 is - * set to true, specifying the maximum number of detections when - * applying NMS algorithm for each single class. - * * 11: An {@link ANEURALNETWORKS_FLOAT32} scalar, score_threshold. Boxes - * with scores lower than the threshold are filtered before sending - * to the NMS algorithm. - * * 12: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the IoU - * threshold for hard NMS. - * * 13: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to include - * background class in the list of label map for the output, set - * to false to not include the background. When the background - * class is included, it has label 0 and the output classes start - * at 1 in the label map, otherwise, the output classes start at 0. - * - * Outputs: - * * 0: A 2-D tensor of the same {@link OperandCode} as input0, with shape - * [batches, max_num_detections], specifying the score of each output - * detections. - * * 1: A 3-D tensor of shape [batches, max_num_detections, 4], specifying the - * coordinates of each output bounding box, with format - * [y1, x1, y2, x2]. - * * 2: A 2-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [batches, max_num_detections], specifying the class label for each - * output detection. - * * 3: An 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape [batches], - * specifying the number of valid output detections for each batch. - * - * Available since API level 29. - */ - ANEURALNETWORKS_DETECTION_POSTPROCESSING = 47, - /** - * For input tensors x and y, computes x == y elementwise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and dimensions compatible - * with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_EQUAL = 48, - /** - * Computes exponential of x element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_EXP = 49, - /** - * Inserts a dimension of 1 into a tensor's shape. - * - * Given a tensor input, this operation inserts a dimension of 1 at the - * given dimension index of input's shape. The dimension index starts at - * zero; if you specify a negative dimension index, it is counted backward - * from the end. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar specifying the dimension - * index to expand. Must be in the range [-(n + 1), (n + 1)). - * - * Outputs: - * * 0: An (n + 1)-D tensor with the same {@link OperandCode} and data as - * input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_EXPAND_DIMS = 50, - /** - * Gathers values along an axis. - * - * Produces an output tensor with shape - * input0.dimension[:axis] + indices.dimension + input0.dimension[axis + 1:] - * where: - * # Vector indices (output is rank(input0)). - * output[a_0, ..., a_n, i, b_0, ..., b_n] = - * input0[a_0, ..., a_n, indices[i], b_0, ..., b_n] - * - * # Higher rank indices (output is rank(input0) + rank(indices) - 1). - * output[a_0, ..., a_n, i, ..., j, b_0, ... b_n] = - * input0[a_0, ..., a_n, indices[i, ..., j], b_0, ..., b_n] - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: An n-D tensor from which to gather values. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar specifying the axis. - * Negative index is used to specify axis from the end - * (e.g. -1 for the last axis). Must be in the range [-n, n). - * * 2: A k-D tensor {@link ANEURALNETWORKS_TENSOR_INT32} of indices. - * The values must be in the bounds of the corresponding dimensions - * of input0. - * - * Outputs: - * * 0: An (n + k - 1)-D tensor with the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_GATHER = 51, - /** - * Generate aixs-aligned bounding box proposals. - * - * Bounding box proposals are generated by applying transformation on a set - * of predefined anchors with the bounding box deltas from bounding box - * regression. A final step of hard NMS is applied to limit the number of - * returned boxes. - * - * Axis-aligned bounding boxes are represented by its upper-left corner - * coordinate (x1,y1) and lower-right corner coordinate (x2,y2). A valid - * bounding box should satisfy x1 <= x2 and y1 <= y2. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Inputs: - * * 0: A 4-D Tensor specifying the score of each anchor at each - * location. With "NHWC" data layout, the tensor shape is - * [batches, height, width, num_anchors]. With "NCHW" data layout, - * the tensor shape is [batches, num_anchors, height, width]. - * * 1: A 4-D Tensor specifying the bounding box deltas. With "NHWC" data - * layout, the tensor shape is [batches, height, width, num_anchors * 4]. - * With "NCHW" data layout, the tensor shape is - * [batches, num_anchors * 4, height, width]. The box deltas are encoded - * in the order of [dx, dy, dw, dh], where dx and dy is the linear-scale - * relative correction factor for the center position of the bounding box - * with respect to the width and height, dw and dh is the log-scale - * relative correction factor for the width and height. The last - * dimensions is the channel dimension. - * * 2: A 2-D Tensor of shape [num_anchors, 4], specifying the shape of each - * predefined anchor, with format [x1, y1, x2, y2]. For input0 of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, this tensor should be of - * {@link ANEURALNETWORKS_TENSOR_QUANT16_SYMM}, with scale of 0.125. - * * 3: A 2-D Tensor of shape [batches, 2], specifying the size of - * each image in the batch, with format [image_height, image_width]. - * * 4: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the ratio - * from the height of original image to the height of feature map. - * * 5: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the ratio - * from the width of original image to the width of feature map. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the maximum - * number of boxes before going into the hard NMS algorithm. Boxes - * with the lowest scores are discarded to meet the limit. Set to - * a non-positive value for unlimited number. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the maximum - * number of boxes returning from the hard NMS algorithm. Boxes - * with the lowest scores are discarded to meet the limit. Set to - * a non-positive value for unlimited number. - * * 8: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the IoU - * threshold for hard NMS. - * * 9: An {@link ANEURALNETWORKS_FLOAT32} scalar, min_size. Boxes with - * height or width lower than the absolute threshold are filtered out. - * * 10: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and input1. Set to false for NHWC. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0, of shape - * [num_output_rois], specifying the score of each output box. - * The boxes are grouped by batches, but the sequential order in - * each batch is not guaranteed. For type of - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the scale and zero - * point must be the same as input0. - * * 1: A tensor of the same {@link OperandCode} as input1, of shape - * [num_output_rois, 4], specifying the coordinates of each output - * bounding box for each class, with format [x1, y1, x2, y2]. - * The sequential order of the boxes corresponds with output0. - * For type of {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, the - * scale must be 0.125 and the zero point must be 0. - * * 2: A 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [batches], specifying the number of output boxes for each image. - * - * Available since API level 29. - */ - ANEURALNETWORKS_GENERATE_PROPOSALS = 52, - /** - * For input tensors x and y, computes x > y elementwise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and dimensions compatible - * with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_GREATER = 53, - /** - * For input tensors x and y, computes x >= y elementwise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and dimensions compatible - * with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_GREATER_EQUAL = 54, - /** - * Performs a grouped 2-D convolution operation. - * - * Given an input tensor of shape [batches, height, width, depth_in] and a - * filter tensor of shape [depth_out, filter_height, filter_width, depth_group] - * containing depth_out convolutional filters of depth depth_group, GROUPED_CONV - * applies a group of different filters to each input channel group, then - * concatenates the results together. - * - * Specifically, the input channels are divided into num_groups groups, each with - * depth depth_group, i.e. depth_in = num_groups * depth_group. The convolutional - * filters are also divided into num_groups groups, i.e. depth_out is divisible - * by num_groups. GROUPED_CONV applies each group of filters to the corresponding - * input channel group, and the result are concatenated together. - * - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * The values in the output tensor are computed as: - * - * output[b, i, j, g * channel_multiplier + q] = - * sum_{di, dj, dk} ( - * input[b, strides[1] * i + di, strides[2] * j + dj, - * g * depth_group + dk] * - * filter[g * channel_multiplier + q, di, dj, dk] - * ) + bias[channel] - * - * where channel_multiplier = depth_out / num_groups - * - * Supported tensor {@link OperandCode} configurations: - * * 32 bit Floating point : - * * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} for input, filter, output, and bias. - * - * * 16 bit Floating point: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} for input, filter, output, and bias. - * - * * Quantized: - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} for input, filter, and output. - * * * {@link ANEURALNETWORKS_TENSOR_INT32} for bias (with scale set to - * * * input.scale * filter.scale). - * - * * Quantized with symetric per channel quantization for the filter: - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} for input, and output. - * * * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} for filter. - * * * {@link ANEURALNETWORKS_TENSOR_INT32} for bias (scale set to 0.0, - * * * each value scaling is separate and equal to input.scale * filter.scales[channel]). - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input, where depth_in = num_groups * depth_group. - * * 1: A 4-D tensor, of shape - * [depth_out, filter_height, filter_width, depth_group], specifying - * the filter, where depth_out must be divisible by num_groups. For - * tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} - * the channel dimension (channelDim at - * {@link ANeuralNetworksSymmPerChannelQuantParams}) must be set to 0. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias must be of the same - * type. For filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the bias should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint - * of 0 and bias_scale == input_scale * filter_scale. For filter tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}, the bias - * should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of - * 0 and bias_scale of 0. The actual scale of each value 'i' is equal to - * bias_scale[i] = input_scale * filter_scale[i]. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, specifying the number of - groups. - * * 10: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 11: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input, where depth_in = num_groups * depth_group. - * * 1: A 4-D tensor, of shape - * [depth_out, filter_height, filter_width, depth_group], specifying - * the filter, where depth_out must be divisible by num_groups. For - * tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} - * the channel dimension (channelDim at - * {@link ANeuralNetworksSymmPerChannelQuantParams}) must be set to 0. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias must be of the same - * type. For filter tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the bias should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint - * of 0 and bias_scale == input_scale * filter_scale. For filter tensor - * of {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}, the bias - * should be of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of - * 0 and bias_scale of 0. The actual scale of each value 'i' is equal to - * bias_scale[i] = input_scale * filter_scale[i]. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the number of - * groups. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 8: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth_out]. For output tensor of - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the following condition - * must be satisfied: output_scale > input_scale * filter_scale (for - * filter tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} - * this condition must be true for all filter scales). - * - * Available since API level 29. - */ - ANEURALNETWORKS_GROUPED_CONV_2D = 55, - /** - * Localize the maximum keypoints from heatmaps. - * - * This operation approximates the accurate maximum keypoint scores and - * indices after bicubic upscaling by using Taylor expansion up to the - * quadratic term. - * - * The bounding box is represented by its upper-left corner coordinate - * (x1,y1) and lower-right corner coordinate (x2,y2) in the original image. - * A valid bounding box should satisfy x1 <= x2 and y1 <= y2. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: A 4-D Tensor of shape - * [num_boxes, heatmap_size, heatmap_size, num_keypoints], - * specifying the heatmaps, the height and width of heatmaps should - * be the same, and must be greater than or equal to 2. - * * 1: A 2-D Tensor of shape [num_boxes, 4], specifying the bounding boxes, - * each with format [x1, y1, x2, y2]. For input0 of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, this tensor should - * be of {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, with zeroPoint - * of 0 and scale of 0.125. - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0. Set to false for NHWC. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0, with shape - * [num_boxes, num_keypoints], specifying score of the keypoints. - * * 1: A tensor of the same {@link OperandCode} as input1, with shape - * [num_boxes, num_keypoints, 2], specifying the location of - * the keypoints, the second dimension is organized as - * [keypoint_x, keypoint_y]. - * - * Available since API level 29. - */ - ANEURALNETWORKS_HEATMAP_MAX_KEYPOINT = 56, - /** - * Applies instance normalization to the input tensor. - * - * The values in the output tensor are computed as: - * - * output[b, h, w, c] = - * (input[b, h, w, c] - mean[b, c]) * gamma / - * sqrt(var[b, c] + epsilon) + beta - * - * Where the mean and variance are computed across the spatial dimensions: - * - * mean[b, c] = - * sum_{h, w}(input[b, h, w, c]) / sum(1) - * - * var[b, c] = - * sum_{h, w}(pow(input[b, h, w, c] - mean[b, c], 2)) / sum(1) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be normalized. - * * 1: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying gamma, the - * scale applied to the normalized tensor. - * * 2: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying beta, the - * offset applied to the normalized tensor. - * * 3: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying epsilon, the - * small value added to variance to avoid dividing by zero. - * * 4: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} and same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_INSTANCE_NORMALIZATION = 57, - /** - * For input tensors x and y, computes x < y elementwise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and dimensions compatible - * with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LESS = 58, - /** - * For input tensors x and y, computes x <= y elementwise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and dimensions compatible - * with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LESS_EQUAL = 59, - /** - * Computes natural logarithm of x element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LOG = 60, - /** - * Returns the truth value of x AND y element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * * 1: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8} and dimensions - * compatible with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LOGICAL_AND = 61, - /** - * Computes the truth value of NOT x element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LOGICAL_NOT = 62, - /** - * Returns the truth value of x OR y element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * * 1: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8} and dimensions - * compatible with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LOGICAL_OR = 63, - /** - * Computes the log softmax activations given logits. - * - * The output is calculated using this formula: - * - * output = logits * beta - log(reduce_sum(exp(logits * beta), axis)) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor specifying the input logits. - * * 1: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the positive - * scaling factor for the exponent, beta. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar specifying the axis to - * reduce across. Negative index is used to specify axis from the - * end (e.g. -1 for the last axis). Must be in the range [-n, n). - * - * Outputs: - * * 0: The output tensor of the same {@link OperandCode} and shape as - * input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_LOG_SOFTMAX = 64, - /** - * Returns the element-wise maximum of two tensors. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and compatible dimensions - * with input0. - * - * Outputs: - * * 0: The sum, a tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_MAXIMUM = 65, - /** - * Returns the element-wise minimum of two tensors. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and compatible dimensions - * with input0. - * - * Outputs: - * * 0: The sum, a tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_MINIMUM = 66, - /** - * Computes numerical negative value element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_NEG = 67, - /** - * For input tensors x and y, computes x != y elementwise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * This operation supports broadcasting. - * - * Inputs: - * * 0: A tensor. - * * 1: A tensor of the same {@link OperandCode} and dimensions compatible - * with input0. - * - * Outputs: - * * 0: A tensor of {@link ANEURALNETWORKS_TENSOR_BOOL8}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_NOT_EQUAL = 68, - /** - * Pads a tensor with the given constant value according to the specified - * paddings. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor, specifying the tensor to be padded. - * * 1: A 2-D Tensor of {@link ANEURALNETWORKS_TENSOR_INT32}, the paddings - * for each spatial dimension of the input tensor. The shape of the - * tensor must be {rank(input0), 2}. - * padding[i, 0] specifies the number of elements to be padded in the - * front of dimension i. - * padding[i, 1] specifies the number of elements to be padded after - * the end of dimension i. - * * 2: An scalar specifying the value to use for padding input0. - * For input tensor of {@link ANEURALNETWORKS_TENSOR_FLOAT32}, the - * pad value should be of {@link ANEURALNETWORKS_FLOAT32}. - * For input tensor of {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * the pad value should be of {@link ANEURALNETWORKS_INT32}. The - * scale and zeroPoint are assumed to be the same as in input0. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. The - * output tensor has the same rank as input0, and each - * dimension of the output tensor has the same size as the - * corresponding dimension of the input tensor plus the size - * of the padding: - * output0.dimension[i] = - * padding[i, 0] + input0.dimension[i] + padding[i, 1] - * - * Available since API level 29. - */ - ANEURALNETWORKS_PAD_V2 = 69, - /** - * Computes the power of one value to another. - * - * Given a tensor base and a tensor exponent, this operation computes - * base^exponent elementwise. - * - * This operations supports broadcasting. The size of the output is the - * maximum size along each dimension of the input operands. It starts with - * the trailing dimensions, and works its way forward. - * - * For example: - * base.dimension = {4, 1, 2} - * exponent.dimension = {5, 4, 3, 1} - * output.dimension = {5, 4, 3, 2} - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: A tensor specifying the base. - * * 1: A tensor specifying the exponent. - * - * Outputs: - * * 0: An output tensor. - * - * Available since API level 29. - */ - ANEURALNETWORKS_POW = 70, - /** - * Parametric Rectified Linear Unit. - * - * It follows: f(x) = alpha * x for x < 0, f(x) = x for x >= 0, where alpha - * is a learned array with the same {@link OperandCode} and compatible - * dimensions as input x. - * - * Two dimensions are compatible when: - * 1. they are equal, or - * 2. one of them is 1 - * - * The size of the output is the maximum size along each dimension of the - * input operands. It starts with the trailing dimensions, and works its way - * forward. - * - * Example: - * input.dimension = {4, 1, 2} - * alpha.dimension = {5, 4, 3, 1} - * output.dimension = {5, 4, 3, 2} - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: A tensor, specifying the input. - * * 1: A tensor of the same {@link OperandCode}, and compatible dimensions - * as input0, specifying the alpha. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_PRELU = 71, - /** - * Quantizes the input tensor. - * - * The formula is: - * - * output = max(0, min(255, round(input / scale) + zeroPoint) - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0, but with - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}. - * - * Available since API level 29. - */ - ANEURALNETWORKS_QUANTIZE = 72, - /** - * A version of quantized LSTM, using 16 bit quantization for internal - * state. - * - * There is no projection layer, so cell state size is equal to the output - * size. - * - * Inputs: - * * 0: A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [numBatches, inputSize] specifying the input to the LSTM - * cell. Tensor is quantized with a fixed quantization range of - * [-1, 127/128] (scale = 1/128, zeroPoint = 128). - * * 1: The input-to-input weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying input-to-input part of - * weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 2: The input-to-forget weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying input-to-forget part of - * weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 3: The input-to-cell weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying input-to-cell part of - * weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 4: The input-to-output weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying input-to-output part of - * weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 5: The recurrent-to-input weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying recurrent-to-input part - * of weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 6: The recurrent-to-forget weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying recurrent-to-forget - * part of weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 7: The recurrent-to-cell weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying recurrent-to-cell part - * of weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 8: The recurrent-to-output weights. - * A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [outputSize, inputSize] specifying recurrent-to-output - * part of weights for fully-connected layer inside the LSTM cell. - * Quantization zero point and scale must be the same across all the - * weights. - * * 9: The input gate bias. - * A 1-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} and shape - * [outputSize] specifying the bias for the fully-connected layer - * inside the LSTM cell. Bias is quantized with scale being a product - * of input and weights scales and zeroPoint equal to 0. - * * 10:The forget gate bias. - * A 1-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} and shape - * [outputSize] specifying the bias for the fully-connected layer - * inside the LSTM cell. Bias is quantized with scale being a product - * of input and weights scales and zeroPoint equal to 0. - * * 11:The cell bias. - * A 1-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} and shape - * [outputSize] specifying the bias for the fully-connected layer - * inside the LSTM cell. Bias is quantized with scale being a product - * of input and weights scales and zeroPoint equal to 0. - * * 12:The output gate bias. - * A 1-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} and shape - * [outputSize] specifying the bias for the fully-connected layer - * inside the LSTM cell. Bias is quantized with scale being a product - * of input and weights scales and zeroPoint equal to 0. - * * 13: A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT16_SYMM} - * and shape [numBatches, outputSize] specifying the cell state from the - * previous time step of the LSTM cell. It is quantized using a - * quantization range of [-2^4, 2^4 * 32767/32768] (scale = 2^4 / - * 32768, zeroPoint = 0). - * * 14: A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [numBathes, outputSize] specifying the output of the LSTM - * cell from previous time-step. Tensor is quantized with a fixed - * quantization range of [-1, 127/128] (scale = 1/128, zeroPoint = - * 128). - * - * - * Outputs: - * * 0: A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT16_SYMM} - * and shape [numBatches, outputSize] which contains a cell state from - * the current time step. Tensor is quantized using a quantization - * range of [-2^4, 2^4 * 32767/32768] (scale = 2^4 / 32768, zeroPoint = - * 0). - * * 1: A 2-D tensor of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * and shape [numBathes, outputSize] which contains the output value. - * Tensor is quantized with a fixed quantization range of [-1, 127/128] - * (scale = 1/128, zeroPoint = 128). - */ - ANEURALNETWORKS_QUANTIZED_16BIT_LSTM = 73, - /** - * Draws samples from a multinomial distribution. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Inputs: - * * 0: A 2-D tensor with shape [batches, classes], specifying the - * unnormalized log-probabilities for all classes. - * * 1: A scalar {@link ANEURALNETWORKS_INT32}, specifying the number of - * independent samples to draw for each row slice. - * * 2: A 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor with shape [2], - * specifying seeds used to initialize the random distribution. - * Outputs: - * * 0: A 2-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor with shape - * [batches, samples], containing the drawn samples. - * - * Available since API level 29. - */ - ANEURALNETWORKS_RANDOM_MULTINOMIAL = 74, - /** - * Reduces a tensor by computing the "logical and" of elements along given - * dimensions. - * - * If keep_dims is true, the reduced dimensions are - * retained with length 1. Otherwise, the rank of the tensor is reduced by - * 1 for each entry in dimensions. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. Dimension values must be in the range [-n, n). - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, keep_dims. If true, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_REDUCE_ALL = 75, - /** - * Reduces a tensor by computing the "logical or" of elements along given - * dimensions. - * - * If keep_dims is true, the reduced dimensions are - * retained with length 1. Otherwise, the rank of the tensor is reduced by - * 1 for each entry in dimensions. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_BOOL8} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. Dimension values must be in the range [-n, n). - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, keep_dims. If true, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_REDUCE_ANY = 76, - /** - * Reduces a tensor by computing the maximum of elements along given - * dimensions. - * - * If keep_dims is true, the reduced dimensions are - * retained with length 1. Otherwise, the rank of the tensor is reduced by - * 1 for each entry in dimensions. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. Dimension values must be in the range [-n, n). - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, keep_dims. If true, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_REDUCE_MAX = 77, - /** - * Reduces a tensor by computing the minimum of elements along given - * dimensions. - * - * If keep_dims is true, the reduced dimensions are - * retained with length 1. Otherwise, the rank of the tensor is reduced by - * 1 for each entry in dimensions. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. Dimension values must be in the range [-n, n). - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, keep_dims. If true, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_REDUCE_MIN = 78, - /** - * Reduces a tensor by multiplying elements along given dimensions. - * - * If keep_dims is true, the reduced dimensions are - * retained with length 1. Otherwise, the rank of the tensor is reduced by - * 1 for each entry in dimensions. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. Dimension values must be in the range [-n, n). - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, keep_dims. If true, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_REDUCE_PROD = 79, - /** - * Reduces a tensor by summing elements along given dimensions. - * - * If keep_dims is true, the reduced dimensions are - * retained with length 1. Otherwise, the rank of the tensor is reduced by - * 1 for each entry in dimensions. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: up to 4 - * - * Inputs: - * * 0: An n-D tensor. - * * 1: A 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. The dimensions - * to reduce. Dimension values must be in the range [-n, n). - * * 2: An {@link ANEURALNETWORKS_BOOL} scalar, keep_dims. If true, - * retains reduced dimensions with length 1. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_REDUCE_SUM = 80, - /** - * Select and scale the feature map of each region of interest to a unified - * output size by average pooling sampling points from bilinear interpolation. - * - * The region of interest is represented by its upper-left corner coordinate - * (x1,y1) and lower-right corner coordinate (x2,y2) in the original image. - * A spatial scaling factor is applied to map into feature map coordinate. - * A valid region of interest should satisfy x1 <= x2 and y1 <= y2. - * - * No rounding is applied in this operation. The sampling points are unified - * distributed in the pooling bin and their values are calculated by bilinear - * interpolation. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} (since API level 29) - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: A 4-D tensor, specifying the feature map. - * * 1: A 2-D Tensor of shape [num_rois, 4], specifying the locations of - * the regions of interest, each line with format [x1, y1, x2, y2]. - * For input0 of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * this tensor should be of {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, - * with zeroPoint of 0 and scale of 0.125. - * * 2: An 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [batches], specifying the number of output boxes for each batch. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the output - * height of the output tensor. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the output - * width of the output tensor. - * * 5: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the ratio - * from the height of original image to the height of feature map. - * * 6: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the ratio - * from the width of original image to the width of feature map. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the number of - * sampling points in height dimension used to compute the output. - * Set to 0 for adaptive value of ceil(roi_height/out_height). - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the number of - * sampling points in width dimension used to compute the output. - * Set to 0 for adaptive value of ceil(roi_width/out_width). - * * 9: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. The output - * shape is [num_rois, out_height, out_width, depth]. - * - * Available since API level 29. - */ - ANEURALNETWORKS_ROI_ALIGN = 81, - /** - * Select and scale the feature map of each region of interest to a unified - * output size by max-pooling. - * - * The region of interest is represented by its upper-left corner coordinate - * (x1,y1) and lower-right corner coordinate (x2,y2) in the original image. - * A spatial scaling factor is applied to map into feature map coordinate. - * A valid region of interest should satisfy x1 <= x2 and y1 <= y2. - * - * Rounding is applied in this operation to ensure integer boundary for - * regions of interest and pooling bins. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Inputs: - * * 0: A 4-D tensor, specifying the feature map. - * * 1: A 2-D Tensor of shape [num_rois, 4], specifying the locations of - * the regions of interest, each line with format [x1, y1, x2, y2]. - * For input0 of type {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, - * this tensor should be of {@link ANEURALNETWORKS_TENSOR_QUANT16_ASYMM}, - * with zeroPoint of 0 and scale of 0.125. - * * 2: An 1-D {@link ANEURALNETWORKS_TENSOR_INT32} tensor, of shape - * [batches], specifying the number of output boxes for each batch. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the output - * height of the output tensor. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the output - * width of the output tensor. - * * 5: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the ratio - * from the height of original image to the height of feature map. - * * 6: An {@link ANEURALNETWORKS_FLOAT32} scalar, specifying the ratio - * from the width of original image to the width of feature map. - * * 7: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Outputs: - * * 0: A tensor of the same {@link OperandCode} as input0. The output - * shape is [num_rois, out_height, out_width, depth]. - * - * Available since API level 29. - */ - ANEURALNETWORKS_ROI_POOLING = 82, - /** - * Computes reciprocal of square root of x element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_RSQRT = 83, - /** - * Using a tensor of booleans c and input tensors x and y select values - * elementwise from both input tensors: - * - * O[i] = C[i] ? x[i] : y[i]. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: A tensor of type {@link ANEURALNETWORKS_TENSOR_BOOL8} acting as a - * mask that chooses, based on the value at each element, whether the - * corresponding element in the output should be taken from input1 (if - * true) or input2 (if false). - * * 1: An input tensor of the same shape as input0. - * * 2: An input tensor of the same shape and type as input1. - * - * Outputs: - * * 0: A tensor of the same type and shape as input1 and input2. - * - */ - ANEURALNETWORKS_SELECT = 84, - /** - * Computes sin of x element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_SIN = 85, - /** - * Extracts a slice of specified size from the input tensor starting at a - * specified location. - * - * The starting location is specified as a 1-D tensor containing offsets - * for each dimension. The size is specified as a 1-D tensor containing - * either size of a slice along corresponding dimension or -1. In the latter - * case, all the remaining elements in dimension are included in the slice. - * Slice size in each dimension cannot be zero. - * - * A sum of begin offset and a size of a slice must not exceed size of a - * corresponding dimension. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: An n-D tensor to take slice from. - * * 1: A 1-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} specifying - * the beginning indices of the slice in each dimension. - * * 2: A 1-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} specifying - * the size of the slice in each dimension. - * - * Outputs: - * * 0: An n-D tensor of the same type as the input containing the slice. - * - * Available since API level 29. - */ - ANEURALNETWORKS_SLICE = 86, - /** - * Splits a tensor along a given axis into num_splits subtensors. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: An n-D tensor to split. - * * 1: An {@link ANEURALNETWORKS_INT32} scalar specifying the axis along - * which to split. - * * 2: An {@link ANEURALNETWORKS_INT32} scalar indicating the number of - * splits along given axis. Must evenly divide axis size. - * - * Outputs: - * * 0 ~ (num_splits - 1): Resulting subtensors. - * - * Available since API level 29. - */ - ANEURALNETWORKS_SPLIT = 87, - /** - * Computes square root of x element-wise. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: from 1. - * - * Inputs: - * * 0: A tensor. - * - * Outputs: - * * 0: The output tensor of same shape as input0. - * - * Available since API level 29. - */ - ANEURALNETWORKS_SQRT = 88, - /** - * Constructs a tensor by tiling a given tensor. - * - * This operation creates a new tensor by replicating `input` `multiples` - * times. The output tensor's i-th dimension has `input.dims(i) * multiples[i]` - * elements, and the values of `input` are replicated `multiples[i]` times - * along the i-th dimension. - * For example, tiling `[a b c d]` by `[2]` produces `[a b c d a b c d]`. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: input, an n-D tensor specifying the input. - * * 1: multiples, a 1-D tensor of {@link ANEURALNETWORKS_TENSOR_INT32}. - * The length of multiples must be n. - * - * Outputs: - * * 0: A tiled tensor of the same {@link OperandCode} and rank as `input`. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TILE = 89, - /** - * Finds values and indices of the k largest entries for the last dimension. - * - * Resulting values in each dimensions are sorted in descending order. If - * two values are equal, the one with larger index appears first. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_INT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: from 1 - * - * Inputs: - * * 0: input, an n-D tensor specifying the input. - * * 1: k, an {@link ANEURALNETWORKS_INT32} scalar, specifying the number of - * top elements to look for along the last dimension. - * - * Outputs: - * * 0: An n-D tensor of the same type as the input, containing the k - * largest elements along each last dimensional slice. - * * 1: An n-D tensor of type {@link ANEURALNETWORKS_TENSOR_INT32} - * containing the indices of values within the last dimension of input. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TOPK_V2 = 90, - /** - * Performs the tranpose of 2-D convolution operation. - * - * This operation is sometimes called "deconvolution" after Deconvolutional - * Networks, but is actually the transpose (gradient) of - * {@link ANEURALNETWORKS_CONV_2D} rather than an actual deconvolution. - * - * The output dimensions are functions of the filter dimensions, stride, and - * padding. - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM} - * - * Supported tensor rank: 4, with "NHWC" or "NCHW" data layout. - * With the default data layout NHWC, the data is stored in the order of: - * [batch, height, width, channels]. Alternatively, the data layout could - * be NCHW, the data storage order of: [batch, channels, height, width]. - * - * Both explicit padding and implicit padding are supported. - * - * Inputs (explicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: A 4-D tensor, of shape - * [depth_out, filter_height, filter_width, depth_in], specifying the - * filter. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias should be of the - * same type. For input tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the bias should be - * of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of 0 and - * bias_scale == input_scale * filter_scale. - * * 3: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the left, in the ‘width’ dimension. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the right, in the ‘width’ dimension. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the top, in the ‘height’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the padding on - * the bottom, in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 8: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 9: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 10: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Inputs (implicit padding): - * * 0: A 4-D tensor, of shape [batches, height, width, depth_in], - * specifying the input. - * * 1: A 4-D tensor, of shape - * [depth_out, filter_height, filter_width, depth_in], specifying the - * filter. - * * 2: A 1-D tensor, of shape [depth_out], specifying the bias. For input - * tensor of type {@link ANEURALNETWORKS_TENSOR_FLOAT32} or - * {@link ANEURALNETWORKS_TENSOR_FLOAT16}, the bias should be of the - * same type. For input tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the bias should be - * of {@link ANEURALNETWORKS_TENSOR_INT32}, with zeroPoint of 0 and - * bias_scale == input_scale * filter_scale. - * * 3: An {@link ANEURALNETWORKS_TENSOR_INT32} tensor, specifying the output - * tensor shape. - * * 4: An {@link ANEURALNETWORKS_INT32} scalar, specifying the implicit - * padding scheme, has to be one of the - * {@link PaddingCode} values. - * * 5: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘width’ dimension. - * * 6: An {@link ANEURALNETWORKS_INT32} scalar, specifying the stride when - * walking through input in the ‘height’ dimension. - * * 7: An {@link ANEURALNETWORKS_INT32} scalar, and has to be one of the - * {@link FuseCode} values. Specifies the activation to - * invoke on the result. - * * 8: An {@link ANEURALNETWORKS_BOOL} scalar, set to true to specify - * NCHW data layout for input0 and output0. Set to false for NHWC. - * - * Outputs: - * * 0: The output 4-D tensor, of shape - * [batches, out_height, out_width, depth_out]. For output tensor of - * {@link ANEURALNETWORKS_TENSOR_QUANT8_ASYMM}, the following condition - * must be satisfied: output_scale > input_scale * filter_scale. - * - * Available since API level 29. - */ - ANEURALNETWORKS_TRANSPOSE_CONV_2D = 91, - /** - * A recurrent neural network specified by an LSTM cell. - * - * Performs (fully) dynamic unrolling of input. - * - * This Op unrolls the input along the time dimension, and implements the - * following operation for each element in the sequence - * s = 1...sequence_length: - * outputs[s] = projection(state = activation(LSTMOp(inputs[s]))) - * - * Where LSTMOp is the LSTM op as in {@link ANEURALNETWORKS_LSTM}, - * the "projection" is an optional projection layer from state and output - * and the “activation” is the function passed as the - * “fused_activation_function” argument (if not “NONE”). - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * Supported tensor rank: 3, either time-major or batch-major. - * - * All input and output tensors must be of the same type. - * - * Inputs: - * * 0: The input (\f$x_t\f$). - * A 3-D tensor of shape: - * If time-major: [max_time, batch_size, output_size] - * If batch-major: [batch_size, max_time, output_size] - * where “max_size” is the number of timesteps (sequence length), - * “batch_size” corresponds to the batching dimension, and - * “input_size” is the size of the input. - * * 1: The input-to-input weights (\f$W_{xi}\f$). Optional. - * A 2-D tensor of shape [num_units, input_size], where “num_units” - * corresponds to the number of cell units. - * * 2: The input-to-forget weights (\f$W_{xf}\f$). - * A 2-D tensor of shape [num_units, input_size]. - * * 3: The input-to-cell weights (\f$W_{xc}\f$). - * A 2-D tensor of shape [num_units, input_size]. - * * 4: The input-to-output weights (\f$W_{xo}\f$). - * A 2-D tensor of shape [num_units, input_size]. - * * 5: The recurrent-to-input weights (\f$W_{hi}\f$). Optional. - * A 2-D tensor of shape [num_units, output_size], where “output_size” - * corresponds to either the number of cell units (i.e., “num_units”), - * or the second dimension of the “projection_weights”, if defined. - * * 6: The recurrent-to-forget weights (\f$W_{hf}\f$). - * A 2-D tensor of shape [num_units, output_size]. - * * 7: The recurrent-to-cell weights (\f$W_{hc}\f$). - * A 2-D tensor of shape [num_units, output_size]. - * * 8: The recurrent-to-output weights (\f$W_{ho}\f$). - * A 2-D tensor of shape [num_units, output_size]. - * * 9: The cell-to-input weights (\f$W_{ci}\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 10:The cell-to-forget weights (\f$W_{cf}\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 11:The cell-to-output weights (\f$W_{co}\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 12:The input gate bias (\f$b_i\f$). Optional. - * A 1-D tensor of shape [num_units]. - * * 13:The forget gate bias (\f$b_f\f$). - * A 1-D tensor of shape [num_units]. - * * 14:The cell bias (\f$b_c\f$). - * A 1-D tensor of shape [num_units]. - * * 15:The output gate bias (\f$b_o\f$). - * A 1-D tensor of shape [num_units]. - * * 16:The projection weights (\f$W_{proj}\f$). Optional. - * A 2-D tensor of shape [output_size, num_units]. - * * 17:The projection bias (\f$b_{proj}\f$). Optional. - * A 1-D tensor of shape [output_size]. - * * 18:The output state (in) (\f$h_{t-1}\f$). - * A 2-D tensor of shape [batch_size, output_size]. - * * 19:The cell state (in) (\f$C_{t-1}\f$). - * A 2-D tensor of shape [batch_size, num_units]. - * * 20:The activation function (\f$g\f$). - * A value indicating the activation function: - *
    - *
  • 0: None; - *
  • 1: Relu; - *
  • 3: Relu6; - *
  • 4: Tanh; - *
  • 6: Sigmoid. - *
- * * 21:The clipping threshold (\f$t_{cell}\f$) for the cell state, such - * that values are bound within [-cell_clip, cell_clip]. If set to 0.0 - * then clipping is disabled. - * * 22:The clipping threshold (\f$t_{proj}\f$) for the output from the - * projection layer, such that values are bound within - * [-proj_clip, proj_clip]. If set to 0.0 then clipping is disabled. - * * 23:Time-major if true, batch-major if false. - * * 24:The input layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at input gate. - * * 25:The forget layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at forget gate. - * * 26:The cell layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at cell gate. - * * 27:The output layer normalization weights. - * A 1-D tensor of shape [num_units]. Used to rescale normalized inputs - * to activation at output gate. - * - * Outputs: - * * 0: The output (\f$o_t\f$). - * A 3-D tensor of shape: - * If time-major: [max_time, batch_size, output_size] - * If batch-major: [batch_size, max_time, output_size] - * - * Available since API level 29. - */ - ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_LSTM = 92, - /** - * A recurrent neural network layer that applies a basic RNN cell to a - * sequence of inputs. - * - * This layer unrolls the input along the sequence dimension, and implements - * the following operation - * for each element in the sequence s = 1...sequence_length: - * outputs[s] = state = activation(inputs[s] * input_weights’ + state * - * recurrent_weights’ + bias) - * - * Where: - * * “input_weights” is a weight matrix that multiplies the inputs; - * * “recurrent_weights” is a weight matrix that multiplies the current - * “state” which itself is the output from the previous time step - * computation; - * * “bias” is a bias vector (added to each output vector in the batch); - * * “activation” is the function passed as the “fused_activation_function” - * argument (if not “NONE”). - * - * Supported tensor {@link OperandCode}: - * * {@link ANEURALNETWORKS_TENSOR_FLOAT16} - * * {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * - * The input tensors must all be the same type. - * - * Inputs: - * * 0: input. - * A 3-D tensor. The shape is defined by the input 6 (timeMajor). If - * it is set to 1, then the input has a shape [maxTime, batchSize, - * inputSize], otherwise the input has a shape [batchSize, maxTime, - * inputSize]. - * * 1: weights. - * A 2-D tensor of shape [numUnits, inputSize]. - * * 2: recurrent_weights. - * A 2-D tensor of shape [numUnits, numUnits]. - * * 3: bias. - * A 1-D tensor of shape [numUnits]. - * * 4: hidden state - * A 2-D tensor of shape [batchSize, numUnits]. Specifies a hidden - * state input for the first time step of the computation. - * * 5: fusedActivationFunction. - * A {@link FuseCode} value indicating the activation function. If - * “NONE” is specified then it results in a linear activation. - * * 6: timeMajor - * An {@link ANEURALNETWORKS_INT32} scalar specifying the shape format - * of input and output tensors. Must be set to either 0 or 1. - * Outputs: - * * 0: output. - * A 3-D tensor. The shape is defined by the input 6 (timeMajor). If - * it is set to 1, then the output has a shape [maxTime, batchSize, - * numUnits], otherwise the output has a shape [batchSize, maxTime, - * numUnits]. - * - * Available since API level 29. - */ - ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_RNN = 93, -} OperationCode; -/** - * Fused activation function types. - * - * - * Available since API level 27. - */ -typedef enum { - /** NO fused activation function. */ - ANEURALNETWORKS_FUSED_NONE = 0, - /** Fused ReLU activation function. */ - ANEURALNETWORKS_FUSED_RELU = 1, - /** Fused ReLU1 activation function. */ - ANEURALNETWORKS_FUSED_RELU1 = 2, - /** Fused ReLU6 activation function. */ - ANEURALNETWORKS_FUSED_RELU6 = 3, -} FuseCode; -/** - * Implicit padding algorithms. - * - * - * Available since API level 27. - */ -typedef enum { - /** - * SAME padding. - * Padding on both ends are the "same": - * padding_to_beginning = total_padding / 2 - * padding_to_end = (total_padding + 1)/2. - * i.e., for even number of padding, padding to both ends are exactly - * the same; for odd number of padding, padding to the ending is bigger - * than the padding to the beginning by 1. - * - * total_padding is a function of input, stride and filter size. - * It could be computed as follows: - * out_size = (input + stride - 1) / stride; - * needed_input = (out_size - 1) * stride + filter_size - * total_padding = max(0, needed_input - input_size) - * The computation is the same for the horizontal and vertical directions. - */ - ANEURALNETWORKS_PADDING_SAME = 1, - /** - * VALID padding. - * No padding. When the input size is not evenly divisible by - * the filter size, the input at the end that could not fill - * the whole filter tile will simply be ignored. - */ - ANEURALNETWORKS_PADDING_VALID = 2, -} PaddingCode; -/** - * Execution preferences. - * - * Available since API level 27. - */ -typedef enum { - /** - * Prefer executing in a way that minimizes battery drain. - * This is desirable for compilations that will be executed often. - */ - ANEURALNETWORKS_PREFER_LOW_POWER = 0, - /** - * Prefer returning a single answer as fast as possible, even if this causes - * more power consumption. - */ - ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER = 1, - /** - * Prefer maximizing the throughput of successive frames, for example when - * processing successive frames coming from the camera. - */ - ANEURALNETWORKS_PREFER_SUSTAINED_SPEED = 2, -} PreferenceCode; -/** - * Device types. - * - * The type of NNAPI device. - */ -typedef enum { - /** The device type cannot be provided. */ - ANEURALNETWORKS_DEVICE_UNKNOWN = 0, - /** The device does not fall into any category below. */ - ANEURALNETWORKS_DEVICE_OTHER = 1, - /** The device runs NNAPI models on single or multi-core CPU. */ - ANEURALNETWORKS_DEVICE_CPU = 2, - /** The device can run NNAPI models and also accelerate graphics APIs such - * as OpenGL ES and Vulkan. */ - ANEURALNETWORKS_DEVICE_GPU = 3, - /** Dedicated accelerator for Machine Learning workloads. */ - ANEURALNETWORKS_DEVICE_ACCELERATOR = 4, -} DeviceTypeCode; -/** - * Result codes. - * - *

Any NNAPI function can return any result code, including result codes not - * currently documented. Any value other than {@link ANEURALNETWORKS_NO_ERROR} - * indicates a failure of some kind.

- * - *

Additional information about the nature of a failure can be obtained from - * the device log after enabling NNAPI debugging by setting the debug.nn.vlog - * property to 1, e.g., by calling "adb shell setprop debug.nn.vlog 1".

- * - * Available since API level 27. - */ -typedef enum { - /** - * Operation was succesful. - */ - ANEURALNETWORKS_NO_ERROR = 0, - /** - * Failure caused by not enough available memory. - */ - ANEURALNETWORKS_OUT_OF_MEMORY = 1, - ANEURALNETWORKS_INCOMPLETE = 2, - /** - * Failure caused by unexpected null argument. - */ - ANEURALNETWORKS_UNEXPECTED_NULL = 3, - /** - * Failure caused by invalid function arguments, invalid model definition, - * invalid execution definition or invalid data at execution time. - */ - ANEURALNETWORKS_BAD_DATA = 4, - /** - * Failure caused by failed model execution. - */ - ANEURALNETWORKS_OP_FAILED = 5, - /** - * Failure caused by object being in the wrong state. - */ - ANEURALNETWORKS_BAD_STATE = 6, - /** - * Failure caused by not being able to map a file into memory. - * This may be caused by a file descriptor not being mappable, or an AHardwareBuffer - * not supported by the device. - * Mitigate by reading its content into memory. - */ - ANEURALNETWORKS_UNMAPPABLE = 7, - /** - * Failure caused by insufficient buffer size provided to a model output. - */ - ANEURALNETWORKS_OUTPUT_INSUFFICIENT_SIZE = 8, - /** - * Failure caused by a device not being available. - */ - ANEURALNETWORKS_UNAVAILABLE_DEVICE = 9, -} ResultCode; -/** - * For {@link ANeuralNetworksModel_setOperandValue}, values with a - * length smaller or equal to this will be immediately copied into - * the model. The size is in bytes. - * - * Available since API level 27. - */ -enum { ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES = 128 }; -/** - * For {@link ANeuralNetworksCompilation_setCaching}, specify the size - * of the cache token required from the application. The size is in bytes. - * - * Available since API level 29. - */ -enum { ANEURALNETWORKS_BYTE_SIZE_OF_CACHE_TOKEN = 32 }; -/** - * ANeuralNetworksMemory is an opaque type that represents memory. - * - * This type is used to represent shared memory, memory mapped files, - * and similar memories. - * - * By using shared memory, a program can efficiently communicate to the - * runtime and drivers the tensors that define a model. See - * {@link ANeuralNetworksModel_setOperandValueFromMemory}. An application - * should typically create one shared memory object that contains every constant tensor - * needed to define a model. {@link ANeuralNetworksMemory_createFromFd} can be - * used to create shared memory from a file handle. - * {@link ANeuralNetworksMemory_createFromAHardwareBuffer} can be used to - * create shared memory from an AHardwareBuffer handle. - * - * Memory objects can also be used to specify the input and output arguments of - * an execution. See {@link ANeuralNetworksExecution_setInputFromMemory} - * and {@link ANeuralNetworksExecution_setOutputFromMemory}. - * - * Available since API level 27. - */ -typedef struct ANeuralNetworksMemory ANeuralNetworksMemory; -/** - * ANeuralNetworksModel is an opaque type that contains a description of the - * mathematical operations that constitute the model. - * - *

Build the model by calling

    - *
  • {@link ANeuralNetworksModel_create}
    • - *
  • {@link ANeuralNetworksModel_addOperation}
    • - *
  • {@link ANeuralNetworksModel_addOperand}
    • - *
- * - * This forms a graph in which each operation and operand is a node, a - * directed edge from an operand to an operation indicates that the - * operand is an input to the operation, and a directed edge from an - * operation to an operand indicates that the operand is an output - * from the operation. This graph must be acyclic. - * - * A model is completed by calling {@link ANeuralNetworksModel_finish}. - * A model is destroyed by calling {@link ANeuralNetworksModel_free}. - * - *

A model cannot be modified once {@link ANeuralNetworksModel_finish} - * has been called on it.

- * - *

It is the application's responsibility to make sure that only one thread - * modifies a model at a given time. It is however safe for more than one - * thread to use the model once {@link ANeuralNetworksModel_finish} has returned.

- * - *

It is also the application's responsibility to ensure that there are no other - * uses of the model after calling {@link ANeuralNetworksModel_free}. - * This includes any compilation or execution object created using the model.

- * - * Available since API level 27. - */ -typedef struct ANeuralNetworksModel ANeuralNetworksModel; -/** - * ANeuralNetworksCompilation is an opaque type that can be used to compile - * a machine learning model. - * - *

To use:

    - *
  • Create a new compilation instance by calling the - * {@link ANeuralNetworksCompilation_create} function or - * {@link ANeuralNetworksCompilation_createForDevices}.
    • - *
  • Set any desired properties on the compilation (for example, - * {@link ANeuralNetworksCompilation_setPreference}).
    • - *
  • Optionally, set the caching signature and the cache directory on the - * compilation by calling {@link ANeuralNetworksCompilation_setCaching}.
    • - *
  • Complete the compilation with {@link ANeuralNetworksCompilation_finish}.
    • - *
  • Use the compilation as many times as needed - * with {@link ANeuralNetworksExecution_create} and - * {@link ANeuralNetworksBurst_create}.
    • - *
  • Destroy the compilation with {@link ANeuralNetworksCompilation_free} - * once all executions using the compilation have completed.

- * - * A compilation is completed by calling {@link ANeuralNetworksCompilation_finish}. - * A compilation is destroyed by calling {@link ANeuralNetworksCompilation_free}. - * - *

A compilation cannot be modified once {@link ANeuralNetworksCompilation_finish} - * has been called on it.

- * - *

It is the application's responsibility to make sure that only - * one thread modifies a compilation at a given time. It is however - * safe for more than one thread to use the compilation once - * {@link ANeuralNetworksCompilation_finish} has returned.

- * - *

It is also the application's responsibility to ensure that there are no other - * uses of the compilation after calling {@link ANeuralNetworksCompilation_free}. - * This includes any execution object created using the compilation.

- * - * Available since API level 27. - */ -typedef struct ANeuralNetworksCompilation ANeuralNetworksCompilation; -/** - * ANeuralNetworksExecution is an opaque type that can be used to apply a machine - * learning model to a set of inputs. - * - *

To use:

    - *
  • Create a new execution instance by calling the - * {@link ANeuralNetworksExecution_create} function.
    • - *
  • Associate input buffers or memory regions to the model inputs with - * {@link ANeuralNetworksExecution_setInput} or - * {@link ANeuralNetworksExecution_setInputFromMemory}.
    • - *
  • Associate output buffers or memory regions to the model outputs with - * {@link ANeuralNetworksExecution_setOutput} or - * {@link ANeuralNetworksExecution_setOutputFromMemory}.
    • - *
  • Apply the model with one of the following:
      • - *
    • Asynchronously with {@link ANeuralNetworksExecution_startCompute}, - * waiting for the execution to complete with - * {@link ANeuralNetworksEvent_wait}.
      • - *
    • Synchronously with {@link ANeuralNetworksExecution_compute}.
      • - *
    • Synchronously as part of an execution burst with - * {@link ANeuralNetworksExecution_burstCompute}.
    - *
  • Destroy the execution with - * {@link ANeuralNetworksExecution_free}.

- * - *

An output buffer or memory region must not overlap with any - * other output buffer or memory region, with an input buffer or - * memory region, or with an operand value in a memory object - * ({@link ANeuralNetworksModel_setOperandValueFromMemory}).

- * - *

An execution cannot be modified once - * {@link ANeuralNetworksExecution_compute} or - * {@link ANeuralNetworksExecution_startCompute} has been called on it.

- * - *

An execution can be applied to a model with - * {@link ANeuralNetworksExecution_compute} or - * {@link ANeuralNetworksExecution_startCompute} only once. Create new - * executions to do new evaluations of the model.

- * - *

It is the application's responsibility to make sure that only one thread - * modifies an execution at a given time. It is however safe for more than one - * thread to use {@link ANeuralNetworksEvent_wait} at the same time.

- * - *

It is also the application's responsibility to ensure that there are no other - * uses of the execution after calling {@link ANeuralNetworksExecution_free}.

- * - *

Multiple executions can be scheduled and evaluated concurrently, either by - * means of {@link ANeuralNetworksExecution_compute} (which is synchronous) in - * different threads or by means of - * {@link ANeuralNetworksExecution_startCompute} (which is asynchronous). The - * runtime makes no guarantee on the ordering of completion of executions. If - * it's important to the application, the application should enforce the - * ordering by ensuring that one execution completes before the next is - * scheduled (for example, by scheduling all executions synchronously within a - * single thread, or by scheduling all executions asynchronously and using - * {@link ANeuralNetworksEvent_wait} between calls to - * {@link ANeuralNetworksExecution_startCompute}).

- * - * Available since API level 27. - */ -typedef struct ANeuralNetworksExecution ANeuralNetworksExecution; -/** - * Parameters for ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL operand. - */ -typedef struct ANeuralNetworksSymmPerChannelQuantParams { - /* The index of the channel dimension. */ - uint32_t channelDim; - /** The size of the scale array. Should be equal to dimension[channelDim] of the Operand. */ - uint32_t scaleCount; - /** The array of scaling values for each channel. Each value must be greater than zero. */ - const float* scales; -} ANeuralNetworksSymmPerChannelQuantParams; -/** - * ANeuralNetworksBurst is an opaque type that can be used to reduce the latency - * of a rapid sequence of executions. It will likely cause overhead if only used - * for a single execution. - * - * ANeuralNetworksBurst serves as a context object for any number of inferences - * using {@link ANeuralNetworksExecution} objects. An ANeuralNetworksBurst - * object and the {@link ANeuralNetworksExecution} objects used with it must all - * have been created from the same {@link ANeuralNetworksCompilation} object. - * - * This object is also used as a hint to drivers, providing insight to the - * lifetime of a rapid sequence of executions. For example, a driver may choose - * to increase the clock frequency of its accelerator for the lifetime of a - * burst object. - * - *

To use:

    - *
  • Create a new burst object by calling the - * {@link ANeuralNetworksBurst_create} function.
    • - *
  • For each execution:
      • - *
    • Create {@link ANeuralNetworksExecution} and configure its - * properties (see {@link ANeuralNetworksExecution} for details).
      • - *
    • Apply the model synchronously with - * {@link ANeuralNetworksExecution_burstCompute}, reusing the same - * {@link ANeuralNetworksBurst} with the new - * {@link ANeuralNetworksExecution}.
      • - *
    • Use and free the {@link ANeuralNetworksExecution}.
    - *
  • Destroy the burst with - * {@link ANeuralNetworksBurst_free}.

- * - * Available since API level 29. - */ -typedef struct ANeuralNetworksBurst ANeuralNetworksBurst; -/** - * ANeuralNetworksOperandType describes the type of an operand. - * This structure is used to describe both scalars and tensors. - * - * A tensor operand type must have a specified rank (number of - * dimensions) but may have any of its dimensions unspecified. - * - * A tensor operand type with all dimensions specified is "fully - * specified". Whenever possible (i.e., whenever the dimensions are - * known at model construction time), a tensor operand type should be - * (but is not required to be) fully specified, in order to enable the - * best possible performance. - * - * If a tensor operand's type is not fully specified, the dimensions - * of the operand are deduced from the operand types and values of the - * operation for which that operand is an output. - * - *

In the following situations, a tensor operand type must be fully - * specified:

    - *
  • The operand has a constant value, set by - * {@link ANeuralNetworksModel_setOperandValue} (with a - * non-nullptr buffer) or - * {@link ANeuralNetworksModel_setOperandValueFromMemory}.
    • - *
  • The operand is a model input or model output (see - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}). A - * fully specified tensor operand type must either be provided - * to {@link ANeuralNetworksModel_addOperand}; or it must be - * provided to the corresponding - * {@link ANeuralNetworksExecution_setInput}, - * {@link ANeuralNetworksExecution_setInputFromMemory}, - * {@link ANeuralNetworksExecution_setOutput}, or - * {@link ANeuralNetworksModel_setOperandValueFromMemory}. - * EXCEPTION: If the input or output is optional and omitted - * (by passing nullptr for buffer to - * {@link ANeuralNetworksExecution_setInput} or - * {@link ANeuralNetworksExecution_setOutput}) then it need - * not have a fully specified tensor operand type.
- * - * A tensor operand type with some number of unspecified dimensions is - * represented by setting each unspecified dimension to 0. - * - * Available since API level 27. - */ -typedef struct ANeuralNetworksOperandType { - /** The data type, e.g ANEURALNETWORKS_INT8. */ - int32_t type; - /** The number of dimensions (rank). It should be 0 for scalars. */ - uint32_t dimensionCount; - /** The dimensions of the tensor. It should be nullptr for scalars. */ - const uint32_t* dimensions; - /** These two fields are only used for quantized tensors. - * They should be zero for scalars and non-fixed point tensors. - * The dequantized value of each entry is (value - zeroPoint) * scale. - */ - float scale; - int32_t zeroPoint; -} ANeuralNetworksOperandType; -typedef int32_t ANeuralNetworksOperationType; -/** - * ANeuralNetworksEvent is an opaque type that represents an event - * that will be signaled once an execution completes. - * - * Available since API level 27. - */ -typedef struct ANeuralNetworksEvent ANeuralNetworksEvent; -/** - * ANeuralNetworksDevice is an opaque type that represents a device. - * - * This type is used to query basic properties and supported operations of the corresponding - * device, and control which device(s) a model is to be run on. - * - * Available since API level 29. - */ -typedef struct ANeuralNetworksDevice ANeuralNetworksDevice; -/** - * Get the number of available devices. - * - * @param numDevices Used to return the number of devices. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworks_getDeviceCount(uint32_t* numDevices); -/** - * Get the representation of the specified device. - * - * @param devIndex The index of the specified device. Must be less than the - number of available devices. - * @param device The representation of the specified device. - * The same representation will always be returned for the specified - * device. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworks_getDevice(uint32_t devIndex, ANeuralNetworksDevice** device); -/** - * Get the name of the specified device. - * - * @param device The representation of the specified device. - * @param name The returned name of the specified device. The name will be in UTF-8 - * and will be null-terminated. It will be recognizable as a known device name - * rather than a cryptic string. For devices with feature level 29 and above, the - * format of the name is {VENDOR}-{DEVICE}, e.g. “google-ipu”. For devices with - * feature level 28 or lower, the name will always be “unknown-device”. - * The name will remain valid for the duration of the application. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworksDevice_getName(const ANeuralNetworksDevice* device, const char** name); -/** - * Get the type of a given device. - * - * The device type can be used to help application developers to distribute Machine Learning - * workloads and other workloads such as graphical rendering. - * E.g., for an app which renders AR scenes based on real time object detection results, - * the developer could choose an ACCELERATOR type device for ML workloads, and reserve GPU - * for graphical rendering. - * - * @param device The representation of the specified device. - * @param type The returned {@link DeviceTypeCode} of the specified device. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworksDevice_getType(const ANeuralNetworksDevice* device, int32_t* type); -/** - * Get the version of the driver implementation of the specified device. - * - * It’s the responsibility of the driver implementor to insure that this version string - * uniquely distinguishes this implementation from all previous implementations. - * - * This version string must not be confused with the feature level which is solely defined - * by {@link ANeuralNetworksDevice_getFeatureLevel}. There is no implicit ordering of the versions. - * For example, it is not possible to filter all drivers older than a certain version. - * - * Application developers may use this version string to avoid or prefer specific driver - * implementations. For example, an application may want to do so because: - * - A specific version of the driver does not provide the required performance, - * perhaps because of a performance regression. - * - A specific version of the driver has a bug or returns results that don’t match - * the minimum precision requirement for the application. - * - * @param device The representation of the specified device. - * @param version The returned version string of the driver for the specified device. The - * string will be in UTF-8 and will be null-terminated. For devices with feature - * level 28 or lower, "UNKOWN" will be returned. The version string will remain - * valid for the duration of the application. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworksDevice_getVersion(const ANeuralNetworksDevice* device, const char** version); -/** - * Get the supported NNAPI version of the specified device. - * - * Each device has a supported feature level, which is the most advanced feature this driver - * implements. For example, if the driver implements the features introduced in Android P, - * but does not implement the features introduced after Android P, the value would be 28. - * Developers could decide whether or not the specified device should be used for a Model that - * has certain feature requirements. - * - * @param device The representation of the specified device. - * @param featureLevel The API level of the most advanced feature this driver implements. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworksDevice_getFeatureLevel(const ANeuralNetworksDevice* device, - int64_t* featureLevel); -/** - * Get the supported operations for a specified set of devices. If multiple devices - * are selected, the supported operation list is a union of supported operations of all - * selected devices. - * - * @param model The model to be queried. - * @param devices The set of devices. Must not contain duplicates. - * @param numDevices The number of devices in the set. - * @param supportedOps The boolean array to be filled. True means supported. The size of the - * boolean array must be at least as large as the number of operations - * in the model. The order of elements in the supportedOps array matches - * the order in which the corresponding operations were added to the model. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworksModel_getSupportedOperationsForDevices( - const ANeuralNetworksModel* model, const ANeuralNetworksDevice* const* devices, - uint32_t numDevices, bool* supportedOps); -/** - * Create a {@link ANeuralNetworksCompilation} to compile the given model for a specified set - * of devices. If more than one device is specified, the compilation will - * distribute the workload automatically across the devices. The model must be fully - * supported by the specified set of devices. This means that - * ANeuralNetworksModel_getSupportedOperationsForDevices() must have returned true for every - * operation for that model/devices pair. - * - * @param model The {@link ANeuralNetworksModel} to be compiled. - * @param devices The set of devices. Must not contain duplicates. - * @param numDevices The number of devices in the set. - * @param compilation The newly created object or NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA - * if the model is invalid. - * - * Available since API level 29. - */ -int ANeuralNetworksCompilation_createForDevices(ANeuralNetworksModel* model, - const ANeuralNetworksDevice* const* devices, - uint32_t numDevices, - ANeuralNetworksCompilation** compilation); -/** - * Sets the compilation caching signature and the cache directory. - * - * Provides optional caching information to the runtime for faster repeated - * compilation. - * - * See {@link ANeuralNetworksCompilation} for information on multithreaded usage. - * - * @param compilation The compilation to be modified. - * @param cacheDir The cache directory for the runtime to store and retrieve caching - * data. It is recommended to use the code cache directory provided - * by the Android runtime. If not using the code cache directory, the - * user should choose a directory local to the application, and is - * responsible to managing the cache entries. - * @param token The token provided by the user to specify a model must be of length - * ANEURALNETWORKS_BYTE_SIZE_OF_CACHE_TOKEN. The user should ensure that - * the token is unique to a model within the application. The NNAPI - * runtime cannot detect token collisions; a collision will result in a - * failed execution or in a successful execution that produces incorrect - * output values. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - * - * Available since API level 29. - */ -int ANeuralNetworksCompilation_setCaching(ANeuralNetworksCompilation* compilation, - const char* cacheDir, const uint8_t* token); -/** - * Schedule synchronous evaluation of the execution. - * - *

Schedules synchronous evaluation of the execution. Returns once the - * execution has completed and the outputs are ready to be consumed. - *

- * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * See {@link ANeuralNetworksExecution_startCompute} for asynchronous execution. - * Synchronous execution incurs lower overhead than asynchronous execution. - * - * Available since API level 29. - * - * @param execution The execution to be scheduled and executed. - * - * @return ANEURALNETWORKS_NO_ERROR if the execution completed normally. - * ANEURALNETWORKS_UNMAPPABLE if the execution input or output memory cannot - * be properly mapped. - */ -int ANeuralNetworksExecution_compute(ANeuralNetworksExecution* execution); -/** - * Get the dimensional information of the specified output operand of the model of the - * {@link ANeuralNetworksExecution}. - * - * On asynchronous execution initiated by {@link ANeuralNetworksExecution_startCompute}, - * {@link ANeuralNetworksEvent_wait} must be called prior to this function to recuperate - * the resources used by the execution. - * - * @param execution The execution to be queried. - * @param index The index of the output argument we are querying. It is - * an index into the lists passed to - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is not - * the index associated with {@link ANeuralNetworksModel_addOperand}. - * @param rank The rank of the output operand. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_OUTPUT_INSUFFICIENT_SIZE - * if the target output is provided an insufficient buffer at execution time, - * ANEURALNETWORKS_BAD_DATA if the index is invalid. - * - * Available since API level 29. - */ -int ANeuralNetworksExecution_getOutputOperandRank(ANeuralNetworksExecution* execution, - int32_t index, uint32_t* rank); -/** - * Get the dimensional information of the specified output operand of the model of the - * {@link ANeuralNetworksExecution}. The target output operand cannot be a scalar. - * - * On asynchronous execution initiated by {@link ANeuralNetworksExecution_startCompute}, - * {@link ANeuralNetworksEvent_wait} must be called prior to this function to recuperate - * the resources used by the execution. - * - * @param execution The execution to be queried. - * @param index The index of the output argument we are querying. It is an index into the lists - * passed to {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is not - * the index associated with {@link ANeuralNetworksModel_addOperand}. - * @param dimensions The dimension array to be filled. The size of the array must be exactly as - * large as the rank of the output operand to be queried in the model. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_OUTPUT_INSUFFICIENT_SIZE - * if the target output is provided an insufficient buffer at execution time, - * ANEURALNETWORKS_BAD_DATA if the index is invalid or if the target is a scalar. - * - * Available since API level 29. - */ -int ANeuralNetworksExecution_getOutputOperandDimensions(ANeuralNetworksExecution* execution, - int32_t index, uint32_t* dimensions); -/** - * Create a {@link ANeuralNetworksBurst} to apply the given compilation. - * This only creates the burst object. Computation is only performed once - * {@link ANeuralNetworksExecution_burstCompute} is invoked with a valid - * {@link ANeuralNetworksExecution} and {@link ANeuralNetworksBurst}. - * - *

The provided compilation must outlive the burst object.

- * - * Available since API level 29. - * - * @param compilation The {@link ANeuralNetworksCompilation} to be evaluated. - * @param burst The newly created object or NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA - * if the compilation is invalid. - */ -int ANeuralNetworksBurst_create(ANeuralNetworksCompilation* compilation, - ANeuralNetworksBurst** burst) __INTRODUCED_IN(29); -/** - * Destroys the burst object. - * - * Available since API level 29. - * - * @param burst The burst object to be destroyed. Passing NULL is acceptable and - * results in no operation. - */ -void ANeuralNetworksBurst_free(ANeuralNetworksBurst* burst) __INTRODUCED_IN(29); -/** - * Schedule synchronous evaluation of the execution on a burst object. - * - *

Schedules synchronous evaluation of the execution. Returns once the - * execution has completed and the outputs are ready to be consumed.

- * - *

There must be at most one {@link ANeuralNetworksExecution} processing at - * any given time for any given burst object. Any - * {@link ANeuralNetworksExecution} launched before the previous has finished - * will result in ANEURALNETWORKS_BAD_STATE.

- * - * Available since API level 29. - * - * @param burst The burst object to execute on. - * @param execution The execution to be scheduled and executed. The execution - * must be created from the same {@link - * ANeuralNetworksCompilation} as the burst object. - * - * @return ANEURALNETWORKS_NO_ERROR if the execution completed normally. - */ -int ANeuralNetworksExecution_burstCompute(ANeuralNetworksExecution* execution, - ANeuralNetworksBurst* burst) __INTRODUCED_IN(29); -/** - * Creates a shared memory object from an AHardwareBuffer handle. - * - * If the shared memory is backed by an AHardwareBuffer of AHARDWAREBUFFER_FORMAT_BLOB - * format, it can be used the same way as shared memory created from a file handle. See - * {@link ANeuralNetworksMemory} for a description on how to use this shared memory. - * - * If the shared memory is backed by an AHardwareBuffer of a format other than - * AHARDWAREBUFFER_FORMAT_BLOB, it can only be used for Model inputs and outputs. - * When calling {@link ANeuralNetworksExecution_setInputFromMemory} or - * {@link ANeuralNetworksExecution_setOutputFromMemory} with the shared memory, both - * offset and length must be set to zero and the entire memory region will be - * associated with the specified input or output operand. There is no guarantee - * that an arbitrary AHardwareBuffer_Format and AHardwareBuffer_UsageFlags combination - * can be used by arbitrary devices. The execution will fail if selected set of devices - * cannot consume the buffer. - * - * Calling {@link ANeuralNetworksModel_setOperandValueFromMemory} with shared memory - * backed by an AHardwareBuffer of a format other than AHARDWAREBUFFER_FORMAT_BLOB is - * disallowed. - * - * TODO(miaowang): add documentation about intended usage with introspection API. - * - * Available since API level 29. - * - * @param ahwb The AHardwareBuffer handle. - * @param memory The memory object to be created. - * Set to NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if the request completed normally. - * - * @see AHardwareBuffer - */ -int ANeuralNetworksMemory_createFromAHardwareBuffer(const AHardwareBuffer* ahwb, - ANeuralNetworksMemory** memory); -/** - * Specifies whether duration of the {@link ANeuralNetworksExecution} is to be measured. - * By default, duration is not measured. - * - * The {@link ANeuralNetworksExecution} must have been created with - * {@link ANeuralNetworksCompilation_createForDevices} with numDevices = 1. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 29. - * - * @param execution The execution to be modified. - * @param measure 'true' if duration is to be measured, 'false' if not. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksExecution_setMeasureTiming(ANeuralNetworksExecution* execution, bool measure); -/** - * Different duration measurements. - * - * Durations are measured in nanoseconds. - * - * Available since API level 29. - */ -typedef enum { - // Execution time on hardware (not driver, which runs on host processor). - ANEURALNETWORKS_DURATION_ON_HARDWARE = 0, - // Execution time in driver (including time on hardware). Excludes overhead - // such as that of the runtime itself and the IPC needed for the runtime to - // communicate with the driver. - ANEURALNETWORKS_DURATION_IN_DRIVER = 1, -} DurationCode; -/** - * Get the time spent in the specified {@link ANeuralNetworksExecution}, in nanoseconds. - * The execution must have completed. - * - * @param execution The execution to be queried. - * @param durationCode The measurement to be queried, specified by {@link DurationCode}. - * @param duration The returned duration. If no measurement was requested by - * {@link ANeuralNetworksExecution_setMeasureTiming}, or for some other - * reason the duration is not available, UINT64_MAX will be returned. - * A particular device need not support any given measurement. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksExecution_getDuration(const ANeuralNetworksExecution* execution, - int32_t durationCode, uint64_t* duration); -/** - * Creates a shared memory object from a file descriptor. - * - * The shared memory is backed by a file descriptor via mmap. - * See {@link ANeuralNetworksMemory} for a description on how to use - * this shared memory. - * - * Available since API level 27. - * - * @param size The requested size in bytes. - * Must not be larger than the file size. - * @param prot The desired memory protection for the mapping. - * It is either PROT_NONE or the bitwise OR of one or - * more of the following flags: PROT_READ, PROT_WRITE. - * @param fd The requested file descriptor. - * The file descriptor has to be mmap-able. The file - * descriptor will be duplicated. - * @param offset The offset to the beginning of the file of the area to map. - * The offset has to be aligned to a page size. - * @param memory The memory object to be created. - * Set to NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if the request completed normally. - */ -int ANeuralNetworksMemory_createFromFd(size_t size, int protect, int fd, size_t offset, - ANeuralNetworksMemory** memory) __INTRODUCED_IN(27); -/** - * Delete a memory object. - * - * Destroys the object used by the run time to keep track of the memory. - * This will free the underlying actual memory if no other code has open - * handles to this memory. - * - * Available since API level 27. - * - * @param memory The memory object to be freed. - */ -void ANeuralNetworksMemory_free(ANeuralNetworksMemory* memory) __INTRODUCED_IN(27); -/** - * Create an empty {@link ANeuralNetworksModel}. - * - *

This only creates the object. Computation is performed once - * {@link ANeuralNetworksExecution_compute} or - * {@link ANeuralNetworksExecution_startCompute} is invoked. - * - * The model should be constructed with calls to - * {@link ANeuralNetworksModel_addOperation} and - * {@link ANeuralNetworksModel_addOperand} - * - *

{@link ANeuralNetworksModel_finish} should be called once the model - * has been fully constructed.

- * - *

{@link ANeuralNetworksModel_free} should be called once the model - * is no longer needed.

- * - * Available since API level 27. - * - * @param model The {@link ANeuralNetworksModel} to be created. - * Set to NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_create(ANeuralNetworksModel** model) __INTRODUCED_IN(27); -/** - * Destroy a model. - * - * The model need not have been finished by a call to - * {@link ANeuralNetworksModel_finish}. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param model The model to be destroyed. Passing NULL is acceptable and - * results in no operation. - */ -void ANeuralNetworksModel_free(ANeuralNetworksModel* model) __INTRODUCED_IN(27); -/** - * Indicate that we have finished modifying a model. Required before - * calling {@link ANeuralNetworksCompilation_create} and - * {@link ANeuralNetworksCompilation_createForDevices}. - * - * An application is responsible to make sure that no other thread uses - * the model at the same time. - * - * This function must only be called once for a given model. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param model The model to be finished. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_finish(ANeuralNetworksModel* model) __INTRODUCED_IN(27); -/** - * Add an operand to a model. - * - * The order in which the operands are added is important. The first one added - * to a model will have the index value 0, the second 1, etc. These indexes are - * used as operand identifiers in - * {@link ANeuralNetworksModel_addOperation}, - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}, - * {@link ANeuralNetworksModel_setOperandValue}, - * {@link ANeuralNetworksModel_setOperandValueFromMemory}, - * {@link ANeuralNetworksExecution_setInput}, - * {@link ANeuralNetworksExecution_setInputFromMemory}, - * {@link ANeuralNetworksExecution_setOutput}, - * {@link ANeuralNetworksExecution_setOutputFromMemory} and - * {@link ANeuralNetworksExecution_setOperandValue}. - * - *

Every operand must be referenced in exactly one of the following - * ways:

    - *
  • It is identified as a model input with - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}.
    • - *
  • It is identified as a constant with - * {@link ANeuralNetworksModel_setOperandValue} or - * {@link ANeuralNetworksModel_setOperandValueFromMemory}.
    • - *
  • It is identified as an output of exactly one operation with - * {@link ANeuralNetworksModel_addOperation}.

    • - *

    An operand that is identified as a model input or as a constant - * must not also be identified as a model output with - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}.

    - * - * To build a model that can accommodate inputs of various sizes, as - * you may want to do for a CNN, leave unspecified the dimensions that - * will vary at run time. If you do so, fully specify dimensions - * when calling {@link ANeuralNetworksExecution_setInput} or - * {@link ANeuralNetworksExecution_setInputFromMemory}. - * - * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been - * called will return an error. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param model The model to be modified. - * @param type The {@link ANeuralNetworksOperandType} that describes the shape - * of the operand. Neither the {@link ANeuralNetworksOperandType} - * nor the dimensions it points to need to outlive the call to - * {@link ANeuralNetworksModel_addOperand}. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_addOperand(ANeuralNetworksModel* model, - const ANeuralNetworksOperandType* type) __INTRODUCED_IN(27); -/** - * Sets an operand to a constant value. - * - * Values of length smaller or equal to - * {@link ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES} - * are immediately copied into the model. - * - * For values of length greater than {@link ANEURALNETWORKS_MAX_SIZE_OF_IMMEDIATELY_COPIED_VALUES}, - * a pointer to the buffer is stored within the model. The application is responsible - * for not changing the content of this region until all executions using this model - * have completed. As the data may be copied during processing, modifying the data - * after this call yields undefined results. - * - * For large tensors, using {@link ANeuralNetworksModel_setOperandValueFromMemory} - * is likely to be more efficient. - * - * To indicate that an optional operand should be considered missing, - * pass nullptr for buffer and 0 for length. - * - * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been - * called will return an error. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param model The model to be modified. - * @param index The index of the model operand we're setting. - * @param buffer A pointer to the data to use. - * @param length The size in bytes of the data value. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_setOperandValue(ANeuralNetworksModel* model, int32_t index, - const void* buffer, size_t length) __INTRODUCED_IN(27); -/** - * Sets an operand's per channel quantization parameters. - * - * Sets parameters required by a tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}. - * This function must be called for every tensor of type - * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} before - * calling {@link ANeuralNetworksModel_finish}. - * - * Available since API level 29. - * - * @param model The model to be modified. - * @param index The index of the model operand we're setting. - * @param channelQuant The per channel quantization parameters for the operand. - * No memory in this struct needs to outlive the call to - * this function. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_setOperandSymmPerChannelQuantParams( - ANeuralNetworksModel* model, int32_t index, - const ANeuralNetworksSymmPerChannelQuantParams* channelQuant) __INTRODUCED_IN(29); -/** - * Sets an operand to a value stored in a memory object. - * - * The content of the memory is not copied. A reference to that memory is stored - * inside the model. The application is responsible for not changing the content - * of the memory region until all executions using this model have completed. - * As the data may be copied during processing, modifying the data after this call - * yields undefined results. - * - * To indicate that an optional operand should be considered missing, - * use {@link ANeuralNetworksModel_setOperandValue} instead, passing nullptr for buffer. - * - * Is disallowed to set an operand value with shared memory backed by an AHardwareBuffer - * of a format other than AHARDWAREBUFFER_FORMAT_BLOB. - * - * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been - * called will return an error. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * See {@link ANeuralNetworksMemory_createFromAHardwarBuffer} for information on - * AHardwareBuffer usage. - * - * Available since API level 27. - * - * @param model The model to be modified. - * @param index The index of the model operand we're setting. - * @param buffer A pointer to the data to use. - * @param memory The memory containing the data. - * @param offset This specifies the location of the data within the memory. - * The offset is in bytes from the start of memory. - * @param length The size in bytes of the data value. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_setOperandValueFromMemory(ANeuralNetworksModel* model, int32_t index, - const ANeuralNetworksMemory* memory, - size_t offset, size_t length) - __INTRODUCED_IN(27); -/** - * Add an operation to a model. - * - * @param model The model to be modified. - * @param type The {@link ANeuralNetworksOperationType} of the operation. - * @param inputCount The number of entries in the inputs array. - * @param inputs An array of indexes identifying each operand. - * @param outputCount The number of entries in the outputs array. - * @param outputs An array of indexes identifying each operand. - * - * The operands specified by inputs and outputs must have been - * previously added by calls to {@link ANeuralNetworksModel_addOperand}. - * - * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been - * called will return an error. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * - * Available since API level 27. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksModel_addOperation(ANeuralNetworksModel* model, - ANeuralNetworksOperationType type, uint32_t inputCount, - const uint32_t* inputs, uint32_t outputCount, - const uint32_t* outputs) __INTRODUCED_IN(27); -/** - * Specifies which operands will be the model's inputs and - * outputs. Every model must have at least one input and one output. - * - * An operand cannot be used for both input and output. Doing so will - * return an error. - * - * @param model The model to be modified. - * @param inputCount The number of entries in the inputs array. - * @param inputs An array of indexes identifying the input operands. - * @param outputCount The number of entries in the outputs array. - * @param outputs An array of indexes identifying the output operands. - * - * The operands specified by inputs and outputs must have been - * previously added by calls to {@link ANeuralNetworksModel_addOperand}. - * - * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been - * called will return an error. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - * - * Available since API level 27. - * - */ -int ANeuralNetworksModel_identifyInputsAndOutputs(ANeuralNetworksModel* model, uint32_t inputCount, - const uint32_t* inputs, uint32_t outputCount, - const uint32_t* outputs) __INTRODUCED_IN(27); -/** - * Specifies whether {@link ANEURALNETWORKS_TENSOR_FLOAT32} is allowed to be - * calculated with range and/or precision as low as that of the IEEE 754 16-bit - * floating-point format. By default, {@link ANEURALNETWORKS_TENSOR_FLOAT32} - * must be calculated using at least the range and precision of the IEEE 754 - * 32-bit floating-point format. - * - * @param model The model to be modified. - * @param allow 'true' indicates {@link ANEURALNETWORKS_TENSOR_FLOAT32} may be - * calculated with range and/or precision as low as that of the - * IEEE 754 16-bit floating point format. 'false' indicates - * {@link ANEURALNETWORKS_TENSOR_FLOAT32} must be calculated using - * at least the range and precision of the IEEE 754 32-bit floating - * point format. - * - * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has been - * called will return an error. - * - * Available since API level 28. - * - * See {@link ANeuralNetworksModel} for information on multithreaded usage. - */ -int ANeuralNetworksModel_relaxComputationFloat32toFloat16(ANeuralNetworksModel* model, bool allow) - __INTRODUCED_IN(28); -/** - * Create a {@link ANeuralNetworksCompilation} to compile the given model. - * - *

    This only creates the object. Compilation is only performed once - * {@link ANeuralNetworksCompilation_finish} is invoked.

    - * - *

    {@link ANeuralNetworksCompilation_finish} should be called once - * all desired properties have been set on the compilation.

    - * - *

    {@link ANeuralNetworksModel_free} should be called once the compilation - * is no longer needed.

    - * - *

    The provided model must outlive the compilation.

    - * - * The model must already have been finished by a call to - * {@link ANeuralNetworksModel_finish}. - * - * See {@link ANeuralNetworksCompilation} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param model The {@link ANeuralNetworksModel} to be compiled. - * @param compilation The newly created object or NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA - * if the model is invalid. - */ -int ANeuralNetworksCompilation_create(ANeuralNetworksModel* model, - ANeuralNetworksCompilation** compilation) __INTRODUCED_IN(27); -/** - * Destroy a compilation. - * - * The compilation need not have been finished by a call to - * {@link ANeuralNetworksModel_finish}. - * - * See {@link ANeuralNetworksCompilation} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param compilation The compilation to be destroyed. Passing NULL is acceptable and - * results in no operation. - */ -void ANeuralNetworksCompilation_free(ANeuralNetworksCompilation* compilation) __INTRODUCED_IN(27); -/** - * Sets the execution preference. - * - *

    Provides guidance to the runtime when trade-offs are possible.

    - * - * See {@link ANeuralNetworksCompilation} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param compilation The compilation to be modified. - * @param preference Either {@link PREFER_LOW_POWER}, - * {@link PREFER_SINGLE_FAST_ANSWER}, or - * {@link PREFER_SUSTAINED_SPEED}. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksCompilation_setPreference(ANeuralNetworksCompilation* compilation, - int32_t preference) __INTRODUCED_IN(27); -/** - * Indicate that we have finished modifying a compilation. Required before - * calling {@link ANeuralNetworksExecution_create}. - * - * An application is responsible to make sure that no other thread uses - * the compilation at the same time. - * - * This function must only be called once for a given compilation. - * - * See {@link ANeuralNetworksCompilation} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param compilation The compilation to be finished. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksCompilation_finish(ANeuralNetworksCompilation* compilation) __INTRODUCED_IN(27); -/** - * Create a {@link ANeuralNetworksExecution} to apply the given compilation. - * This only creates the object. Computation is only performed once - * {@link ANeuralNetworksExecution_compute} or - * {@link ANeuralNetworksExecution_startCompute} is invoked. - * - *

    The provided compilation must outlive the execution.

    - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param compilation The {@link ANeuralNetworksCompilation} to be evaluated. - * @param execution The newly created object or NULL if unsuccessful. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA - * if the compilation is invalid. - */ -int ANeuralNetworksExecution_create(ANeuralNetworksCompilation* compilation, - ANeuralNetworksExecution** execution) __INTRODUCED_IN(27); -/** - * Destroy an execution. - * - *

    If called on an execution for which - * {@link ANeuralNetworksExecution_startCompute} has been called, the - * function will return immediately but will mark the execution to be deleted - * once the computation completes. The related {@link ANeuralNetworksEvent} - * will be signaled and the {@link ANeuralNetworksEvent_wait} will return - * ANEURALNETWORKS_ERROR_DELETED. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param execution The execution to be destroyed. Passing NULL is acceptable and - * results in no operation. - */ -void ANeuralNetworksExecution_free(ANeuralNetworksExecution* execution) __INTRODUCED_IN(27); -/** - * Associate a user buffer with an input of the model of the - * {@link ANeuralNetworksExecution}. - * - *

    The provided buffer must outlive the execution.

    - * - * If the input is optional, you can indicate that it is omitted by - * passing nullptr for buffer and 0 for length. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param execution The execution to be modified. - * @param index The index of the input argument we are setting. It is - * an index into the lists passed to - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is not - * the index associated with - * {@link ANeuralNetworksModel_addOperand}. - * @param type The {@link ANeuralNetworksOperandType} of the - * operand. Unless the input is omitted, this should be - * used to specify the dimensions that were left - * unspecified when the operand was added to the - * model. All other properties of the type must be the - * same as specified in the model. If the type is the same - * as specified when the model was built, NULL can be - * passed. Neither the {@link ANeuralNetworksOperandType} - * nor the dimensions it points to need to outlive the call - * to {@link ANeuralNetworksExecution_setInput}. - * @param buffer The buffer containing the data. - * @param length The length in bytes of the buffer. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the - * name is not recognized or the buffer is too small for the input. - */ -int ANeuralNetworksExecution_setInput(ANeuralNetworksExecution* execution, int32_t index, - const ANeuralNetworksOperandType* type, const void* buffer, - size_t length) __INTRODUCED_IN(27); -/** - * Associate part of a memory object with an input of the model of the - * {@link ANeuralNetworksExecution}. - * - *

    The provided memory must outlive the execution.

    - * - * If the input is optional, you can indicate that it is omitted by - * using {@link ANeuralNetworks_setInput} instead, passing nullptr for buffer - * and 0 for length. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * See {@link ANeuralNetworksMemory_createFromAHardwarBuffer} for information on - * AHardwareBuffer usage. - * - * Available since API level 27. - * - * @param execution The execution to be modified. - * @param index The index of the input argument we are setting. It is - * an index into the lists passed to - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is not - * the index associated with {@link ANeuralNetworksModel_addOperand}. - * @param type The {@link ANeuralNetworksOperandType} of the - * operand. This should be used to specify the dimensions - * that were left unspecified when the operand was added - * to the model. All other properties of the type must be - * the same as specified in the model. If the type is the - * same as specified when the model was built, NULL can be - * passed. Neither the {@link ANeuralNetworksOperandType} - * nor the dimensions it points to need to outlive the call - * to {@link ANeuralNetworksExecution_setInputFromMemory}. - * @param memory The memory containing the data. - * @param offset This specifies the location of the data within the memory. - * The offset is in bytes from the start of memory. - * @param length The size in bytes of the data value. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the - * name is not recognized or the buffer is too small for the input. - */ -int ANeuralNetworksExecution_setInputFromMemory(ANeuralNetworksExecution* execution, int32_t index, - const ANeuralNetworksOperandType* type, - const ANeuralNetworksMemory* memory, size_t offset, - size_t length) __INTRODUCED_IN(27); -/** - * Associate a user buffer with an output of the model of the - * {@link ANeuralNetworksExecution}. - * - * If the output is optional, you can indicate that it is omitted by - * passing nullptr for buffer and 0 for length. - * - *

    The provided buffer must outlive the execution.

    - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 27. - * - * @param execution The execution to be modified. - * @param index The index of the output argument we are setting. It is - * an index into the lists passed to - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is not - * the index associated with {@link ANeuralNetworksModel_addOperand}. - * @param type The {@link ANeuralNetworksOperandType} of the - * operand. Unless the output is omitted, this should be - * used to specify the dimensions that were left - * unspecified when the operand was added to the - * model. All other properties of the type must be the - * same as specified in the model. If the type is the same - * as specified when the model was built, NULL can be - * passed. Neither the {@link ANeuralNetworksOperandType} - * nor the dimensions it points to need to outlive the call - * to {@link ANeuralNetworksExecution_setOutput}. - * @param buffer The buffer where the data is to be written. - * @param length The length in bytes of the buffer. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the - * name is not recognized or the buffer is too small for the output. - */ -int ANeuralNetworksExecution_setOutput(ANeuralNetworksExecution* execution, int32_t index, - const ANeuralNetworksOperandType* type, void* buffer, - size_t length) __INTRODUCED_IN(27); -/** - * Associate part of a memory object with an output of the model of the - * {@link ANeuralNetworksExecution}. - * - * If the output is optional, you can indicate that it is omitted by - * using {@link ANeuralNetworks_setOutput} instead, passing nullptr for buffer - * and 0 for length. - * - *

    The provided memory must outlive the execution.

    - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * See {@link ANeuralNetworksMemory_createFromAHardwarBuffer} for information on - * AHardwareBuffer usage. - * - * Available since API level 27. - * - * @param execution The execution to be modified. - * @param index The index of the output argument we are setting. It is - * an index into the lists passed to - * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is not - * the index associated with {@link ANeuralNetworksModel_addOperand}. - * @param type The {@link ANeuralNetworksOperandType} of the operand. This should be - * used to specify the dimensions that were left - * unspecified when the operand was added to the - * model. All other properties of the type must be the - * same as specified in the model. If the type is the same - * as specified when the model was built, NULL can be - * passed. Neither the {@link ANeuralNetworksOperandType} - * nor the dimensions it points to need to outlive the call - * to {@link ANeuralNetworksExecution_setOutputFromMemory}. - * @param memory The memory where the data is to be stored. - * @param offset This specifies the location of the data within the memory. - * The offset is in bytes from the start of memory. - * @param length The length in bytes of the data value. - * - * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if the - * name is not recognized or the buffer is too small for the output. - */ -int ANeuralNetworksExecution_setOutputFromMemory(ANeuralNetworksExecution* execution, int32_t index, - const ANeuralNetworksOperandType* type, - const ANeuralNetworksMemory* memory, size_t offset, - size_t length) __INTRODUCED_IN(27); -/** - * Schedule asynchronous evaluation of the execution. - * - *

    Schedules asynchronous evaluation of the execution. Once the model has - * been applied and the outputs are ready to be consumed, the returned event - * will be signaled. Use {@link ANeuralNetworksEvent_wait} to wait for that - * event. - *

    - * - * ANeuralNetworksEvent_wait must be called to recuperate the resources used - * by the execution. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * See {@link ANeuralNetworksExecution_compute} for synchronous execution. - * Synchronous execution incurs lower overhead than asynchronous execution. - * - * Available since API level 27. - * - * @param execution The execution to be scheduled and executed. - * @param event The event that will be signaled on completion. event is set to - * NULL if there's an error. - * - * @return ANEURALNETWORKS_NO_ERROR if successful. - */ -int ANeuralNetworksExecution_startCompute(ANeuralNetworksExecution* execution, - ANeuralNetworksEvent** event) __INTRODUCED_IN(27); -/** - * Waits until the execution completes. - * - * More than one thread can wait on an event. When the execution completes, - * all threads will be released. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 27. - * - * @return ANEURALNETWORKS_NO_ERROR if the execution completed normally. - * ANEURALNETWORKS_UNMAPPABLE if the execution input or output memory cannot - * be properly mapped. - */ -int ANeuralNetworksEvent_wait(ANeuralNetworksEvent* event) __INTRODUCED_IN(27); -/** - * Destroys the event. - * - * See {@link ANeuralNetworksExecution} for information on multithreaded usage. - * - * Available since API level 27. - */ -void ANeuralNetworksEvent_free(ANeuralNetworksEvent* event) __INTRODUCED_IN(27); -__END_DECLS -#endif // ANDROID_ML_NN_RUNTIME_NEURAL_NETWORKS_MOCK_H -/** @} */ diff --git a/include/dnnlibrary/NeuralNetworksTypes.h b/include/dnnlibrary/NeuralNetworksTypes.h new file mode 100644 index 0000000..69253f1 --- /dev/null +++ b/include/dnnlibrary/NeuralNetworksTypes.h @@ -0,0 +1,552 @@ +/* Copyright 2017 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ +#ifndef TENSORFLOW_LITE_NNAPI_NEURALNETWORKSTYPES_H_ +#define TENSORFLOW_LITE_NNAPI_NEURALNETWORKSTYPES_H_ + +#include +#include + +typedef struct AHardwareBuffer AHardwareBuffer; + +// NN api types based on NNAPI header file +// https://developer.android.com/ndk/reference/group/neural-networks + +/** + * Operand types. + * + * The type of operands that can be added to a model. + * + * Although we define many types, most operators accept just a few + * types. Most used are ANEURALNETWORKS_TENSOR_FLOAT32, + * ANEURALNETWORKS_TENSOR_QUANT8_ASYMM, and ANEURALNETWORKS_INT32. + */ +enum { + ANEURALNETWORKS_FLOAT32 = 0, + ANEURALNETWORKS_INT32 = 1, + ANEURALNETWORKS_UINT32 = 2, + ANEURALNETWORKS_TENSOR_FLOAT32 = 3, + ANEURALNETWORKS_TENSOR_INT32 = 4, + ANEURALNETWORKS_TENSOR_QUANT8_ASYMM = 5, + ANEURALNETWORKS_BOOL = 6, + ANEURALNETWORKS_TENSOR_QUANT16_SYMM = 7, + ANEURALNETWORKS_TENSOR_FLOAT16 = 8, + ANEURALNETWORKS_TENSOR_BOOL8 = 9, + ANEURALNETWORKS_FLOAT16 = 10, + ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL = 11, + ANEURALNETWORKS_TENSOR_QUANT16_ASYMM = 12, + ANEURALNETWORKS_TENSOR_QUANT8_SYMM = 13, +}; + +/** + * Operation types. + * + * The type of operations that can be added to a model. + */ +enum { + ANEURALNETWORKS_ADD = 0, + ANEURALNETWORKS_AVERAGE_POOL_2D = 1, + ANEURALNETWORKS_CONCATENATION = 2, + ANEURALNETWORKS_CONV_2D = 3, + ANEURALNETWORKS_DEPTHWISE_CONV_2D = 4, + ANEURALNETWORKS_DEPTH_TO_SPACE = 5, + ANEURALNETWORKS_DEQUANTIZE = 6, + ANEURALNETWORKS_EMBEDDING_LOOKUP = 7, + ANEURALNETWORKS_FLOOR = 8, + ANEURALNETWORKS_FULLY_CONNECTED = 9, + ANEURALNETWORKS_HASHTABLE_LOOKUP = 10, + ANEURALNETWORKS_L2_NORMALIZATION = 11, + ANEURALNETWORKS_L2_POOL_2D = 12, + ANEURALNETWORKS_LOCAL_RESPONSE_NORMALIZATION = 13, + ANEURALNETWORKS_LOGISTIC = 14, + ANEURALNETWORKS_LSH_PROJECTION = 15, + ANEURALNETWORKS_LSTM = 16, + ANEURALNETWORKS_MAX_POOL_2D = 17, + ANEURALNETWORKS_MUL = 18, + ANEURALNETWORKS_RELU = 19, + ANEURALNETWORKS_RELU1 = 20, + ANEURALNETWORKS_RELU6 = 21, + ANEURALNETWORKS_RESHAPE = 22, + ANEURALNETWORKS_RESIZE_BILINEAR = 23, + ANEURALNETWORKS_RNN = 24, + ANEURALNETWORKS_SOFTMAX = 25, + ANEURALNETWORKS_SPACE_TO_DEPTH = 26, + ANEURALNETWORKS_SVDF = 27, + ANEURALNETWORKS_TANH = 28, + ANEURALNETWORKS_BATCH_TO_SPACE_ND = 29, + ANEURALNETWORKS_DIV = 30, + ANEURALNETWORKS_MEAN = 31, + ANEURALNETWORKS_PAD = 32, + ANEURALNETWORKS_SPACE_TO_BATCH_ND = 33, + ANEURALNETWORKS_SQUEEZE = 34, + ANEURALNETWORKS_STRIDED_SLICE = 35, + ANEURALNETWORKS_SUB = 36, + ANEURALNETWORKS_TRANSPOSE = 37, + ANEURALNETWORKS_ABS = 38, + ANEURALNETWORKS_ARGMAX = 39, + ANEURALNETWORKS_ARGMIN = 40, + ANEURALNETWORKS_BIDIRECTIONAL_SEQUENCE_LSTM = 42, + ANEURALNETWORKS_CAST = 45, + ANEURALNETWORKS_EQUAL = 48, + ANEURALNETWORKS_EXP = 49, + ANEURALNETWORKS_EXPAND_DIMS = 50, + ANEURALNETWORKS_GATHER = 51, + ANEURALNETWORKS_GREATER = 53, + ANEURALNETWORKS_GREATER_EQUAL = 54, + ANEURALNETWORKS_LESS = 58, + ANEURALNETWORKS_LESS_EQUAL = 59, + ANEURALNETWORKS_LOG = 60, + ANEURALNETWORKS_LOGICAL_AND = 61, + ANEURALNETWORKS_LOGICAL_NOT = 62, + ANEURALNETWORKS_LOGICAL_OR = 63, + ANEURALNETWORKS_LOG_SOFTMAX = 64, + ANEURALNETWORKS_MAXIMUM = 65, + ANEURALNETWORKS_MINIMUM = 66, + ANEURALNETWORKS_NEG = 67, + ANEURALNETWORKS_NOT_EQUAL = 68, + ANEURALNETWORKS_PAD_V2 = 69, + ANEURALNETWORKS_POW = 70, + ANEURALNETWORKS_PRELU = 71, + ANEURALNETWORKS_QUANTIZE = 72, + ANEURALNETWORKS_QUANTIZED_16BIT_LSTM = 73, + ANEURALNETWORKS_REDUCE_ANY = 76, + ANEURALNETWORKS_REDUCE_MAX = 77, + ANEURALNETWORKS_REDUCE_MIN = 78, + ANEURALNETWORKS_REDUCE_PROD = 79, + ANEURALNETWORKS_REDUCE_SUM = 80, + ANEURALNETWORKS_RSQRT = 83, + ANEURALNETWORKS_SELECT = 84, + ANEURALNETWORKS_SIN = 85, + ANEURALNETWORKS_SLICE = 86, + ANEURALNETWORKS_SPLIT = 87, + ANEURALNETWORKS_SQRT = 88, + ANEURALNETWORKS_TILE = 89, + ANEURALNETWORKS_TOPK_V2 = 90, + ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_LSTM = 92, + ANEURALNETWORKS_UNIDIRECTIONAL_SEQUENCE_RNN = 93, +}; + +/** + * Fused activation function types. + * + */ +enum { + ANEURALNETWORKS_FUSED_NONE = 0, + ANEURALNETWORKS_FUSED_RELU = 1, + ANEURALNETWORKS_FUSED_RELU1 = 2, + ANEURALNETWORKS_FUSED_RELU6 = 3, +}; + +/** + * Execution preferences. + */ +enum { + ANEURALNETWORKS_PREFER_LOW_POWER = 0, + ANEURALNETWORKS_PREFER_FAST_SINGLE_ANSWER = 1, + ANEURALNETWORKS_PREFER_SUSTAINED_SPEED = 2, +}; + +/** + * Result codes. + */ +enum { + ANEURALNETWORKS_NO_ERROR = 0, + ANEURALNETWORKS_OUT_OF_MEMORY = 1, + ANEURALNETWORKS_INCOMPLETE = 2, + ANEURALNETWORKS_UNEXPECTED_NULL = 3, + ANEURALNETWORKS_BAD_DATA = 4, + ANEURALNETWORKS_OP_FAILED = 5, + ANEURALNETWORKS_BAD_STATE = 6, + ANEURALNETWORKS_UNMAPPABLE = 7, + ANEURALNETWORKS_OUTPUT_INSUFFICIENT_SIZE = 8, + ANEURALNETWORKS_UNAVAILABLE_DEVICE = 9, +}; + +/** + * Implicit padding algorithms. + */ +enum { + ANEURALNETWORKS_PADDING_SAME = 1, + ANEURALNETWORKS_PADDING_VALID = 2, +}; + +/** + * Device types. + * + * The type of NNAPI device. + */ +enum { + /** The device type cannot be provided. */ + ANEURALNETWORKS_DEVICE_UNKNOWN = 0, + /** The device does not fall into any category below. */ + ANEURALNETWORKS_DEVICE_OTHER = 1, + /** The device runs NNAPI models on single or multi-core CPU. */ + ANEURALNETWORKS_DEVICE_CPU = 2, + /** The device can run NNAPI models and also accelerate graphics APIs such + * as OpenGL ES and Vulkan. */ + ANEURALNETWORKS_DEVICE_GPU = 3, + /** Dedicated accelerator for Machine Learning workloads. */ + ANEURALNETWORKS_DEVICE_ACCELERATOR = 4, +}; + +/** + * ANeuralNetworksMemory is an opaque type that represents memory. + * + * This type is used to represent shared memory, memory mapped files, + * and similar memories. + * + * By using shared memory, a program can efficiently communicate to the + * runtime and drivers the tensors that define a model. See + * {@link ANeuralNetworksModel_setOperandValueFromMemory}. An application + * should typically create one shared memory object that contains every tensor + * needed to define a model. {@link ANeuralNetworksMemory_createFromFd} can be + * used to create shared memory from a file handle. {@link + * ANeuralNetworksMemory_createShared} can be used to directly created shared + * memory. + * + * Memory objects can also be used to specify the input and output arguments of + * an execution. See {@link ANeuralNetworksExecution_setInputFromMemory} + * and {@link ANeuralNetworksExecution_setOutputFromMemory}. + */ +typedef struct ANeuralNetworksMemory ANeuralNetworksMemory; + +/** + * ANeuralNetworksModel is an opaque type that contains a description of the + * mathematical operations that constitute the model. + * + *

    The model will be built by calling

      + *
    • {@link ANeuralNetworksModel_create},
      • + *
    • {@link ANeuralNetworksModel_addOperation},
      • + *
    • {@link ANeuralNetworksModel_addOperand},
      • + *
    + * + * A model is completed by calling {@link ANeuralNetworksModel_finish}. + * A model is destroyed by calling {@link ANeuralNetworksModel_free}. + * + *

    It is the application's responsibility to make sure that only one thread + * modifies a model at a given time. It is however safe for more than one + * thread to use the model once {@link ANeuralNetworksModel_finish} has + * returned.

    + * + *

    It is also the application's responsibility to ensure that there are no + * other uses of the model after calling {@link ANeuralNetworksModel_free}. This + * includes any compilation or execution object created using the model.

    + */ +typedef struct ANeuralNetworksModel ANeuralNetworksModel; + +/** + * ANeuralNetworksCompilation is an opaque type that can be used to compile + * a machine learning model. + * + *

    To use:

      + *
    • Create a new compilation instance by calling the + * {@link ANeuralNetworksCompilation_create} function.
      • + *
    • Perform the compilation with {@link + * ANeuralNetworksCompilation_start}.
    • Wait for the compilation to + * complete with {@link ANeuralNetworksCompilation_wait}.
    • Use the + * compilation as many times as needed with {@link + * ANeuralNetworksExecution_create}.
    • Destroy the compilation with + * {@link ANeuralNetworksCompilation_free} once all executions using the + * compilation have completed.

    + * + *

    A compilation cannot be modified once {@link + * ANeuralNetworksCompilation_start} has been called on it.

    + * + *

    It is the application's responsibility to make sure that only one thread + * modifies a compilation at a given time. It is however safe for more than one + * thread to use {@link ANeuralNetworksCompilation_wait} at the same time. + * It is also safe for multiple threads to use a compilation object once + * {@link ANeuralNetworksCompilation_wait} has completed.

    + * + *

    It is also the application's responsibility to ensure that there are no + * other uses of the compilation after calling {@link + * ANeuralNetworksCompilation_free}. This includes any execution object created + * using the compilation.

    + */ +typedef struct ANeuralNetworksCompilation ANeuralNetworksCompilation; + +/** + * ANeuralNetworksExecution is an opaque type that can be used to apply a + * machine learning model to a set of inputs. + * + *

    To use:

      + *
    • Create a new execution instance by calling the + * {@link ANeuralNetworksExecution_create} function.
      • + *
    • Associate data to the model inputs with + * {@link ANeuralNetworksExecution_setInput} or + * {@link ANeuralNetworksExecution_setInputFromMemory}.
      • + *
    • Associate output buffers to the model outputs with + * {@link ANeuralNetworksExecution_setOutput} or + * {@link ANeuralNetworksExecution_setOutputFromMemory}.
      • + *
    • Apply the model with {@link + * ANeuralNetworksExecution_startCompute}.
    • Wait for the execution to + * complete with {@link ANeuralNetworksExecution_wait}.
    • Destroy the + * execution with + * {@link ANeuralNetworksExecution_free}.

    + * + *

    An execution cannot be modified once {@link + * ANeuralNetworksExecution_start} has been called on it.

    + * + *

    An execution can be applied to a model with + * {@link ANeuralNetworksExecution_startCompute} only once. Create new + * executions to do new evaluations of the model.

    + * + *

    It is the application's responsibility to make sure that only one thread + * modifies an execution at a given time. It is however safe for more than one + * thread to use {@link ANeuralNetworksExecution_wait} at the same time.

    + * + *

    It is also the application's responsibility to ensure that there are no + * other uses of the request after calling {@link + * ANeuralNetworksRequest_free}.

    + */ +typedef struct ANeuralNetworksExecution ANeuralNetworksExecution; + +/** + * Parameters for ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL operand. + */ +typedef struct ANeuralNetworksSymmPerChannelQuantParams { + /* The index of the channel dimension. */ + uint32_t channelDim; + /** The size of the scale array. Should be equal to dimension[channelDim] of + * the Operand. */ + uint32_t scaleCount; + /** The array of scaling values for each channel. Each value must be greater + * than zero. */ + const float* scales; +} ANeuralNetworksSymmPerChannelQuantParams; + +/** + * ANeuralNetworksBurst is an opaque type that can be used to reduce the latency + * of a rapid sequence of executions. It will likely cause overhead if only used + * for a single execution. + * + * ANeuralNetworksBurst serves as a context object for any number of inferences + * using {@link ANeuralNetworksExecution} objects. An ANeuralNetworksBurst + * object and the {@link ANeuralNetworksExecution} objects used with it must all + * have been created from the same {@link ANeuralNetworksCompilation} object. + * + * This object is also used as a hint to drivers, providing insight to the + * lifetime of a rapid sequence of executions. For example, a driver may choose + * to increase the clock frequency of its accelerator for the lifetime of a + * burst object. + * + *

    To use:

      + *
    • Create a new burst object by calling the + * {@link ANeuralNetworksBurst_create} function.
      • + *
    • For each execution:
        • + *
      • Create {@link ANeuralNetworksExecution} and configure its + * properties (see {@link ANeuralNetworksExecution} for + * details).
      • Apply the model synchronously with + * {@link ANeuralNetworksExecution_burstCompute}, reusing the same + * {@link ANeuralNetworksBurst} with the new + * {@link ANeuralNetworksExecution}.
        • + *
      • Use and free the {@link ANeuralNetworksExecution}.
      + *
    • Destroy the burst with + * {@link ANeuralNetworksBurst_free}.

    + * + * Available since API level 29. + */ +typedef struct ANeuralNetworksBurst ANeuralNetworksBurst; + +/** + * ANeuralNetworksOperandType describes the type of an operand. + * This structure is used to describe both scalars and tensors. + */ +typedef struct ANeuralNetworksOperandType { + /** The data type, e.g ANEURALNETWORKS_INT8. */ + int32_t type; + /** The number of dimensions. It should be 0 for scalars. */ + uint32_t dimensionCount; + /** The dimensions of the tensor. It should be nullptr for scalars. */ + const uint32_t* dimensions; + /** These two fields are only used for quantized tensors. + * They should be zero for scalars and non-fixed point tensors. + * The dequantized value of each entry is (value - offset) * scale. + */ + float scale; + int32_t zeroPoint; +} ANeuralNetworksOperandType; + +/** + * ANeuralNetworksEvent is an opaque type that represents an event + * that will be signaled once an execution completes. + */ +typedef struct ANeuralNetworksEvent ANeuralNetworksEvent; + +typedef int32_t ANeuralNetworksOperationType; + +/** + * ANeuralNetworksDevice is an opaque type that represents a device. + * + * This type is used to query basic properties and supported operations of the + * corresponding device, and control which device(s) a model is to be run on. + * + * Available since API level 29. + */ +typedef struct ANeuralNetworksDevice ANeuralNetworksDevice; + +// nn api function types + +typedef int (*ANeuralNetworksMemory_createFromFd_fn)( + size_t size, int protect, int fd, size_t offset, + ANeuralNetworksMemory** memory); + +typedef void (*ANeuralNetworksMemory_free_fn)(ANeuralNetworksMemory* memory); + +typedef int (*ANeuralNetworksModel_create_fn)(ANeuralNetworksModel** model); + +typedef int (*ANeuralNetworksModel_finish_fn)(ANeuralNetworksModel* model); + +typedef void (*ANeuralNetworksModel_free_fn)(ANeuralNetworksModel* model); + +typedef int (*ANeuralNetworksCompilation_create_fn)( + ANeuralNetworksModel* model, ANeuralNetworksCompilation** compilation); + +typedef void (*ANeuralNetworksCompilation_free_fn)( + ANeuralNetworksCompilation* compilation); + +typedef int (*ANeuralNetworksCompilation_setPreference_fn)( + ANeuralNetworksCompilation* compilation, int32_t preference); + +typedef int (*ANeuralNetworksCompilation_finish_fn)( + ANeuralNetworksCompilation* compilation); + +typedef int (*ANeuralNetworksModel_addOperand_fn)( + ANeuralNetworksModel* model, const ANeuralNetworksOperandType* type); + +typedef int (*ANeuralNetworksModel_setOperandValue_fn)( + ANeuralNetworksModel* model, int32_t index, const void* buffer, + size_t length); + +typedef int (*ANeuralNetworksModel_setOperandSymmPerChannelQuantParams_fn)( + ANeuralNetworksModel* model, int32_t index, + const ANeuralNetworksSymmPerChannelQuantParams* channelQuant); + +typedef int (*ANeuralNetworksModel_setOperandValueFromMemory_fn)( + ANeuralNetworksModel* model, int32_t index, + const ANeuralNetworksMemory* memory, size_t offset, size_t length); + +typedef int (*ANeuralNetworksModel_addOperation_fn)( + ANeuralNetworksModel* model, ANeuralNetworksOperationType type, + uint32_t inputCount, const uint32_t* inputs, uint32_t outputCount, + const uint32_t* outputs); + +typedef int (*ANeuralNetworksModel_identifyInputsAndOutputs_fn)( + ANeuralNetworksModel* model, uint32_t inputCount, const uint32_t* inputs, + uint32_t outputCount, const uint32_t* outputs); + +typedef int (*ANeuralNetworksModel_relaxComputationFloat32toFloat16_fn)( + ANeuralNetworksModel* model, bool allow); + +typedef int (*ANeuralNetworksExecution_create_fn)( + ANeuralNetworksCompilation* compilation, + ANeuralNetworksExecution** execution); + +typedef void (*ANeuralNetworksExecution_free_fn)( + ANeuralNetworksExecution* execution); + +typedef int (*ANeuralNetworksExecution_setInput_fn)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, const void* buffer, size_t length); + +typedef int (*ANeuralNetworksExecution_setInputFromMemory_fn)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, const ANeuralNetworksMemory* memory, + size_t offset, size_t length); + +typedef int (*ANeuralNetworksExecution_setOutput_fn)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, void* buffer, size_t length); + +typedef int (*ANeuralNetworksExecution_setOutputFromMemory_fn)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, const ANeuralNetworksMemory* memory, + size_t offset, size_t length); + +typedef int (*ANeuralNetworksExecution_startCompute_fn)( + ANeuralNetworksExecution* execution, ANeuralNetworksEvent** event); + +typedef int (*ANeuralNetworksEvent_wait_fn)(ANeuralNetworksEvent* event); + +typedef void (*ANeuralNetworksEvent_free_fn)(ANeuralNetworksEvent* event); + +typedef int (*ASharedMemory_create_fn)(const char* name, size_t size); + +typedef int (*ANeuralNetworks_getDeviceCount_fn)(uint32_t* numDevices); + +typedef int (*ANeuralNetworks_getDevice_fn)(uint32_t devIndex, + ANeuralNetworksDevice** device); + +typedef int (*ANeuralNetworksDevice_getName_fn)( + const ANeuralNetworksDevice* device, const char** name); + +typedef int (*ANeuralNetworksDevice_getType_fn)( + const ANeuralNetworksDevice* device, int32_t* type); + +typedef int (*ANeuralNetworksDevice_getVersion_fn)( + const ANeuralNetworksDevice* device, const char** version); + +typedef int (*ANeuralNetworksDevice_getFeatureLevel_fn)( + const ANeuralNetworksDevice* device, int64_t* featureLevel); + +typedef int (*ANeuralNetworksModel_getSupportedOperationsForDevices_fn)( + const ANeuralNetworksModel* model, + const ANeuralNetworksDevice* const* devices, uint32_t numDevices, + bool* supportedOps); + +typedef int (*ANeuralNetworksCompilation_createForDevices_fn)( + ANeuralNetworksModel* model, const ANeuralNetworksDevice* const* devices, + uint32_t numDevices, ANeuralNetworksCompilation** compilation); + +typedef int (*ANeuralNetworksCompilation_setCaching_fn)( + ANeuralNetworksCompilation* compilation, const char* cacheDir, + const uint8_t* token); + +typedef int (*ANeuralNetworksExecution_compute_fn)( + ANeuralNetworksExecution* execution); + +typedef int (*ANeuralNetworksExecution_getOutputOperandRank_fn)( + ANeuralNetworksExecution* execution, int32_t index, uint32_t* rank); + +typedef int (*ANeuralNetworksExecution_getOutputOperandDimensions_fn)( + ANeuralNetworksExecution* execution, int32_t index, uint32_t* dimensions); + +typedef int (*ANeuralNetworksBurst_create_fn)( + ANeuralNetworksCompilation* compilation, ANeuralNetworksBurst** burst); + +typedef void (*ANeuralNetworksBurst_free_fn)(ANeuralNetworksBurst* burst); + +typedef int (*ANeuralNetworksExecution_burstCompute_fn)( + ANeuralNetworksExecution* execution, ANeuralNetworksBurst* burst); + +typedef int (*ANeuralNetworksMemory_createFromAHardwareBuffer_fn)( + const AHardwareBuffer* ahwb, ANeuralNetworksMemory** memory); + +typedef int (*ANeuralNetworksExecution_setMeasureTiming_fn)( + ANeuralNetworksExecution* execution, bool measure); + +typedef enum { + // Execution time on hardware (not driver, which runs on host processor). + ANEURALNETWORKS_DURATION_ON_HARDWARE = 0, + // Execution time in driver (including time on hardware). Excludes overhead + // such as that of the runtime itself and the IPC needed for the runtime to + // communicate with the driver. + ANEURALNETWORKS_DURATION_IN_DRIVER = 1, +} DurationCode; + +typedef int (*ANeuralNetworksExecution_getDuration_fn)( + const ANeuralNetworksExecution* execution, int32_t durationCode, + uint64_t* duration); + +#endif // TENSORFLOW_LITE_NNAPI_NEURALNETWORKSTYPES_H_ diff --git a/include/dnnlibrary/NeuralNetworksWrapper.h b/include/dnnlibrary/NeuralNetworksWrapper.h index ee4cfa5..7fac537 100644 --- a/include/dnnlibrary/NeuralNetworksWrapper.h +++ b/include/dnnlibrary/NeuralNetworksWrapper.h @@ -16,7 +16,7 @@ // Provides C++ classes to more easily use the Neural Networks API. #ifndef ANDROID_ML_NN_RUNTIME_NEURAL_NETWORKS_WRAPPER_H #define ANDROID_ML_NN_RUNTIME_NEURAL_NETWORKS_WRAPPER_H -#include "NeuralNetworksMock.h" +#include #include #include #include @@ -122,72 +122,6 @@ struct OperandType { } operator ANeuralNetworksOperandType() const {return operandType; } }; -class Memory { -public: - Memory(size_t size, int protect, int fd, size_t offset) { - mValid = ANeuralNetworksMemory_createFromFd(size, protect, fd, offset, &mMemory) == - ANEURALNETWORKS_NO_ERROR; - } - Memory(AHardwareBuffer* buffer) { - mValid = ANeuralNetworksMemory_createFromAHardwareBuffer(buffer, &mMemory) == - ANEURALNETWORKS_NO_ERROR; - } - ~Memory() { ANeuralNetworksMemory_free(mMemory); } - // Disallow copy semantics to ensure the runtime object can only be freed - // once. Copy semantics could be enabled if some sort of reference counting - // or deep-copy system for runtime objects is added later. - Memory(const Memory&) = delete; - Memory& operator=(const Memory&) = delete; - // Move semantics to remove access to the runtime object from the wrapper - // object that is being moved. This ensures the runtime object will be - // freed only once. - Memory(Memory&& other) { *this = std::move(other); } - Memory& operator=(Memory&& other) { - if (this != &other) { - ANeuralNetworksMemory_free(mMemory); - mMemory = other.mMemory; - mValid = other.mValid; - other.mMemory = nullptr; - other.mValid = false; - } - return *this; - } - ANeuralNetworksMemory* get() const { return mMemory; } - bool isValid() const { return mValid; } -private: - ANeuralNetworksMemory* mMemory = nullptr; - bool mValid = true; -}; -class Event { -public: - Event() {} - ~Event() { ANeuralNetworksEvent_free(mEvent); } - // Disallow copy semantics to ensure the runtime object can only be freed - // once. Copy semantics could be enabled if some sort of reference counting - // or deep-copy system for runtime objects is added later. - Event(const Event&) = delete; - Event& operator=(const Event&) = delete; - // Move semantics to remove access to the runtime object from the wrapper - // object that is being moved. This ensures the runtime object will be - // freed only once. - Event(Event&& other) { *this = std::move(other); } - Event& operator=(Event&& other) { - if (this != &other) { - ANeuralNetworksEvent_free(mEvent); - mEvent = other.mEvent; - other.mEvent = nullptr; - } - return *this; - } - Result wait() { return static_cast(ANeuralNetworksEvent_wait(mEvent)); } - // Only for use by Execution - void set(ANeuralNetworksEvent* newEvent) { - ANeuralNetworksEvent_free(mEvent); - mEvent = newEvent; - } -private: - ANeuralNetworksEvent* mEvent = nullptr; -}; } // namespace wrapper } // namespace nn } // namespace android diff --git a/include/dnnlibrary/nnapi_implementation.h b/include/dnnlibrary/nnapi_implementation.h new file mode 100644 index 0000000..0df10e2 --- /dev/null +++ b/include/dnnlibrary/nnapi_implementation.h @@ -0,0 +1,1022 @@ +/* Copyright 2017 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ +#ifndef TENSORFLOW_LITE_NNAPI_NNAPI_IMPLEMENTATION_H_ +#define TENSORFLOW_LITE_NNAPI_NNAPI_IMPLEMENTATION_H_ + +#include +#include +#include + +#include "NeuralNetworksTypes.h" + +struct NnApi { + bool nnapi_exists; + int32_t android_sdk_version; + + /** + * Creates a shared memory object from a file descriptor. + * + * The shared memory is backed by a file descriptor via mmap. + * See {@link ANeuralNetworksMemory} for a description on how to use + * this shared memory. + * + * @param size The requested size in bytes. + * Must not be larger than the file size. + * @param prot The desired memory protection for the mapping. + * It is either PROT_NONE or the bitwise OR of one or + * more of the following flags: PROT_READ, PROT_WRITE. + * @param fd The requested file descriptor. + * The file descriptor has to be mmap-able. The file + * descriptor will be duplicated. + * @param offset The offset to the beginning of the file of the area to map. + * The offset has to be aligned to a page size. + * @param memory The memory object to be created. + * Set to NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if the request completed normally. + */ + int (*ANeuralNetworksMemory_createFromFd)(size_t size, int protect, int fd, + size_t offset, + ANeuralNetworksMemory** memory); + + /** + * Delete a memory object. + * + * Destroys the object used by the run time to keep track of the memory. + * This will free the underlying actual memory if no other code has open + * handles to this memory. + * + * @param memory The memory object to be freed. + */ + void (*ANeuralNetworksMemory_free)(ANeuralNetworksMemory* memory); + + /** + * Create an empty {@link ANeuralNetworksModel}. + * + *

    This only creates the object. Computation is performed once + * {@link ANeuralNetworksExecution_startCompute} is invoked. + * + * The model should be constructed with calls to + * {@link ANeuralNetworksModel_addOperation} and + * {@link ANeuralNetworksModel_addOperand} + * + *

    {@link ANeuralNetworksModel_finish} should be called once the model + * has been fully constructed.

    + * + *

    {@link ANeuralNetworksModel_free} should be called once the model + * is no longer needed.

    + * + * @param model The {@link ANeuralNetworksModel} to be created. + * Set to NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_create)(ANeuralNetworksModel** model); + + /** + * Destroy a model. + * + * The model need not have been finished by a call to + * {@link ANeuralNetworksModel_finish}. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + * @param model The model to be destroyed. Passing NULL is acceptable and + * results in no operation. + */ + void (*ANeuralNetworksModel_free)(ANeuralNetworksModel* model); + + /** + * Indicate that we have finished modifying a model. Required before + * calling {@link ANeuralNetworksCompilation_compile}. + * + * An application is responsible to make sure that no other thread uses + * the model at the same time. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + * @param model The model to be finished. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_finish)(ANeuralNetworksModel* model); + + /** + * Add an operand to a model. + * + * The order in which the operands are added is important. The first one added + * to a model will have the index value 0, the second 1, etc. These indexes + * are used as operand identifiers in + * {@link ANeuralNetworksModel_addOperation}, + * {@link ANeuralNetworksExecution_setInput}, + * {@link ANeuralNetworksExecution_setInputFromMemory}, + * {@link ANeuralNetworksExecution_setOutput}, + * {@link ANeuralNetworksExecution_setOutputFromMemory} and + * {@link ANeuralNetworksExecution_setOperandValue}. + * + * To build a model that can accommodate inputs of various sizes, as you may + * want to do for a CNN, set the size of the dimensions that will vary at run + * time to 0. If you do so, provide the full dimensions when calling + * {@link ANeuralNetworksExecution_setInput} or {@link + * ANeuralNetworksExecution_setInputFromMemory}. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has + * been called will return an error. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + * @param model The model to be modified. + * @param type The {@link ANeuralNetworksOperandType} that describes the shape + * of the operand. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_addOperand)( + ANeuralNetworksModel* model, const ANeuralNetworksOperandType* type); + + /** + * Sets an operand to a constant value. + * + * For scalar values, the content of buffer is copied into the model. + * + * For tensor values, a pointer to the buffer is stored within the model. + * The application is responsible for not changing the content of this region + * until all executions using this model have completed. As the data may + * be copied during processing, modifying the data after this call yields + * undefined results. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has + * been called will return an error. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + * @param model The model to be modified. + * @param index The index of the model operand we're setting. + * @param buffer A pointer to the data to use. + * @param length The size in bytes of the data value. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_setOperandValue)(ANeuralNetworksModel* model, + int32_t index, const void* buffer, + size_t length); + + /** + * Sets an operand's per channel quantization parameters. + * + * Sets parameters required by a tensor of type + * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL}. + * This function must be called for every tensor of type + * {@link ANEURALNETWORKS_TENSOR_QUANT8_SYMM_PER_CHANNEL} before + * calling {@link ANeuralNetworksModel_finish}. + * + * Available since API level 29. + * + * @param model The model to be modified. + * @param index The index of the model operand we're setting. + * @param channelQuant The per channel quantization parameters for the + * operand. No memory in this struct needs to outlive the + * call to this function. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_setOperandSymmPerChannelQuantParams)( + ANeuralNetworksModel* model, int32_t index, + const ANeuralNetworksSymmPerChannelQuantParams* channelQuant); + + /** + * Sets an operand to a value stored in a memory object. + * + * The content of the memory is not copied. A reference to that memory is + * stored inside the model. The application is responsible for not changing + * the content of the memory region until all executions using this model have + * completed. + * As the data may be copied during processing, modifying the data after this + * call yields undefined results. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has + * been called will return an error. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + * @param model The model to be modified. + * @param index The index of the model operand we're setting. + * @param buffer A pointer to the data to use. + * @param memory The memory containing the data. + * @param offset This specifies the location of the data within the memory. + * The offset is in bytes from the start of memory. + * @param length The size in bytes of the data value. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_setOperandValueFromMemory)( + ANeuralNetworksModel* model, int32_t index, + const ANeuralNetworksMemory* memory, size_t offset, size_t length); + + /** + * Add an operation to a model. + * + * @param model The model to be modified. + * @param type The type of the operation. + * @param inputCount The number of entries in the inputs array. + * @param inputs An array of indexes identifying each operand. + * @param outputCount The number of entries in the outputs array. + * @param outputs An array of indexes identifying each operand. + * + * The operands specified by inputs and outputs must have been + * previously added by calls to {@link ANeuralNetworksModel_addOperand}. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has + * been called will return an error. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksModel_addOperation)(ANeuralNetworksModel* model, + ANeuralNetworksOperationType type, + uint32_t inputCount, + const uint32_t* inputs, + uint32_t outputCount, + const uint32_t* outputs); + + /** + * Specifies which operands will be the model's inputs and outputs. + * + * An operand cannot be used for both input and output. Doing so will + * return an error. + * + * @param model The model to be modified. + * @param inputCount The number of entries in the inputs array. + * @param inputs An array of indexes identifying the input operands. + * @param outputCount The number of entries in the outputs array. + * @param outputs An array of indexes identifying the output operands. + * + * The operands specified by inputs and outputs must have been + * previously added by calls to {@link ANeuralNetworksModel_addOperand}. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has + * been called will return an error. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + * + */ + int (*ANeuralNetworksModel_identifyInputsAndOutputs)( + ANeuralNetworksModel* model, uint32_t inputCount, const uint32_t* inputs, + uint32_t outputCount, const uint32_t* outputs); + + /** + * Specifies whether {@link ANEURALNETWORKS_TENSOR_FLOAT32} is allowed to be + * calculated with range and/or precision as low as that of the + * IEEE 754 16-bit floating-point format. By default, + * {@link ANEURALNETWORKS_TENSOR_FLOAT32} must be calculated using at least + * the range and precision of the IEEE 754 32-bit floating-point format. + * + * @param model The model to be modified. + * @param allow 'true' indicates {@link ANEURALNETWORKS_TENSOR_FLOAT32} may be + * calculated with range and/or precision as low as that of the + * IEEE 754 16-bit floating point format. 'false' indicates + * {@link ANEURALNETWORKS_TENSOR_FLOAT32} must be calculated + * using at least the range and precision of the IEEE 754 32-bit + * floating point format. + * + * Attempting to modify a model once {@link ANeuralNetworksModel_finish} has + * been called will return an error. + * + * Available since API level 28. + * + * See {@link ANeuralNetworksModel} for information on multithreaded usage. + */ + int (*ANeuralNetworksModel_relaxComputationFloat32toFloat16)( + ANeuralNetworksModel* model, bool allow); + + /** + * Create a {@link ANeuralNetworksCompilation} to compile the given model. + * This only creates the object. Compilation is only performed once + * {@link ANeuralNetworksCompilation_start} is invoked. + * + *

    The provided model must outlive the compilation.

    + * + * The model must already have been finished by a call to + * {@link ANeuralNetworksModel_finish}. + * + * See {@link ANeuralNetworksCompilation} for information on multithreaded + * usage. + * + * @param model The {@link ANeuralNetworksModel} to be compiled. + * @param compilation The newly created object or NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA + * if the model is invalid. + */ + int (*ANeuralNetworksCompilation_create)( + ANeuralNetworksModel* model, ANeuralNetworksCompilation** compilation); + + /** + * Destroy a compilation. + * + *

    If called on a compilation for which + * {@link ANeuralNetworksCompilation_start} has been called, the + * function will return immediately but will mark the compilation to be + * deleted once the compilation completes. The + * {@link ANeuralNetworksCompilation_wait} will return ERROR_DELETED. + * + * See {@link ANeuralNetworksCompilation} for information on multithreaded + * usage. + * + * @param compilation The compilation to be destroyed. Passing NULL is + * acceptable and results in no operation. + */ + void (*ANeuralNetworksCompilation_free)( + ANeuralNetworksCompilation* compilation); + + /** + * Sets the execution preference. + * + *

    Provides guidance to the runtime when trade-offs are possible.

    + * + * See {@link ANeuralNetworksCompilation} for information on multithreaded + * usage. + * + * @param compilation The compilation to be modified. + * @param preference Either {@link PREFER_LOW_POWER}, + * {@link PREFER_SINGLE_FAST_ANSWER}, or + * {@link PREFER_SUSTAINED_SPEED}. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksCompilation_setPreference)( + ANeuralNetworksCompilation* compilation, int32_t preference); + + /** + * Waits until the compilation completes. + * + * More than one thread can wait on a compilation. When the compilation + * completes, all threads will be released. + * + * See {@link ANeuralNetworksCompilation} for information on multithreaded + * usage. + * + * @return ANEURALNETWORKS_NO_ERROR if the compilation completed normally. + */ + int (*ANeuralNetworksCompilation_finish)( + ANeuralNetworksCompilation* compilation); + + /** + * Create a {@link ANeuralNetworksExecution} to apply the given compilation. + * This only creates the object. Computation is only performed once + * {@link ANeuralNetworksExecution_startCompute} is invoked. + * + *

    The provided compilation must outlive the execution.

    + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param compilation The {@link ANeuralNetworksCompilation} to be evaluated. + * @param execution The newly created object or NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA + * if the compilation is invalid. + */ + int (*ANeuralNetworksExecution_create)( + ANeuralNetworksCompilation* compilation, + ANeuralNetworksExecution** execution); + + /** + * Destroy an execution. + * + *

    If called on an execution for which + * {@link ANeuralNetworksExecution_startCompute} has been called, the + * function will return immediately but will mark the execution to be deleted + * once the computation completes. The {link ANeuralNetworksExecution_wait} + * will return ANEURALNETWORKS_ERROR_DELETED. + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param execution The execution to be destroyed. Passing NULL is acceptable + * and results in no operation. + */ + void (*ANeuralNetworksExecution_free)(ANeuralNetworksExecution* execution); + + /** + * Associate a user buffer with an input of the model of the + * {@link ANeuralNetworksExecution}. + * + *

    The provided buffer must outlive the execution.

    + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param execution The execution to be modified. + * @param index The index of the input argument we are setting. It is + * an index into the lists passed to + * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is + * not the index associated with {@link + * ANeuralNetworksModel_addOperand}. + * @param type The type of the operand. This should be used to specify the + * dimensions that were set to 0 when the operand was added to the + * model. All other properties of the type must be the same as + * specified in the model. If the type is the same as specified + * when the model was built, NULL can be passed. + * @param buffer The buffer containing the data. + * @param length The length in bytes of the buffer. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if + * the name is not recognized or the buffer is too small for the input. + */ + int (*ANeuralNetworksExecution_setInput)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, const void* buffer, + size_t length); + + /** + * Associate part of a memory object with an input of the model of the + * {@link ANeuralNetworksExecution}. + * + *

    The provided memory must outlive the execution.

    + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param execution The execution to be modified. + * @param index The index of the input argument we are setting. It is + * an index into the lists passed to + * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is + * not the index associated with {@link + * ANeuralNetworksModel_addOperand}. + * @param type The type of the operand. This can be used to specify the + * dimensions that were set to 0 when the operand was added to the + * model. All other values must be the same as specified in the + * model. If the type is the same as specified when the model + * was built, NULL can be passed. + * @param memory The memory containing the data. + * @param offset This specifies the location of the data within the memory. + * The offset is in bytes from the start of memory. + * @param length The size in bytes of the data value. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if + * the name is not recognized or the buffer is too small for the input. + */ + int (*ANeuralNetworksExecution_setInputFromMemory)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, + const ANeuralNetworksMemory* memory, size_t offset, size_t length); + + /** + * Associate a user buffer with an output of the model of the + * {@link ANeuralNetworksExecution}. + * + *

    The provided buffer must outlive the execution.

    + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param execution The execution to be modified. + * @param index The index of the output argument we are setting. It is + * an index into the lists passed to + * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is + * not the index associated with {@link + * ANeuralNetworksModel_addOperand}. + * @param type The type of the operand. This can be used to specify the + * dimensions that were set to 0 when the operand was added to the + * model. All other values must be the same as specified in the + * model. If the type is the same as specified when the model + * was built, NULL can be passed. + * @param buffer The buffer where the data is to be written. + * @param length The length in bytes of the buffer. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if + * the name is not recognized or the buffer is too small for the output. + */ + int (*ANeuralNetworksExecution_setOutput)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, void* buffer, size_t length); + + /** + * Associate part of a memory object with an output of the model of the + * {@link ANeuralNetworksExecution}. + * + *

    The provided memory must outlive the execution.

    + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param execution The execution to be modified. + * @param index The index of the output argument we are setting. It is + * an index into the lists passed to + * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is + * not the index associated with {@link + * ANeuralNetworksModel_addOperand}. + * @param type The type of the operand. This can be used to specify the + * dimensions that were set to 0 when the operand was added to the + * model. All other values must be the same as specified in the + * model. If the type is the same as specified when the model + * was built, NULL can be passed. + * @param memory The memory where the data is to be stored. + * @param offset This specifies the location of the data within the memory. + * The offset is in bytes from the start of memory. + * @param length The length in bytes of the data value. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA if + * the name is not recognized or the buffer is too small for the output. + */ + int (*ANeuralNetworksExecution_setOutputFromMemory)( + ANeuralNetworksExecution* execution, int32_t index, + const ANeuralNetworksOperandType* type, + const ANeuralNetworksMemory* memory, size_t offset, size_t length); + + /** + * Schedule evaluation of the execution. + * + *

    Schedules evaluation of the execution. Once the model has been + * applied and the outputs are ready to be consumed, the execution will be + * signaled. Use {@link ANeuralNetworksExecution_wait} to wait for that + * signal. + *

    + * + * Multiple executions can be scheduled and evaluated concurrently, and + * compilations can be performed concurrently with executions. The runtime + * makes no guarantee on the ordering of the completion of compilations and + * executions. If it's important to the application, the application should + * enforce the ordering by using {@link ANeuralNetworksCompilation_wait} and + * {@link ANeuralNetworksExecution_wait}. + * + * ANeuralNetworksExecution_wait must be called to recuperate the resources + * used by the execution. + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @param execution The execution to be scheduled and executed. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksExecution_startCompute)( + ANeuralNetworksExecution* execution, ANeuralNetworksEvent** event); + + /** + * Waits until the execution completes. + * + * More than one thread can wait on an event. When the execution completes, + * all threads will be released. + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * @return ANEURALNETWORKS_NO_ERROR if the execution completed normally. + */ + int (*ANeuralNetworksEvent_wait)(ANeuralNetworksEvent* event); + + /** + * Destroys the event. + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + */ + void (*ANeuralNetworksEvent_free)(ANeuralNetworksEvent* event); + + // ASharedMemory_create was added in Android 8.0, so safe to use with NNAPI + // which was added in 8.1. + int (*ASharedMemory_create)(const char* name, size_t size); + + /** + * Get the number of available devices. + * + * @param numDevices Used to return the number of devices. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworks_getDeviceCount)(uint32_t* numDevices); + + /** + * Get the representation of the specified device. + * + * @param devIndex The index of the specified device. Must be less than the + * number of available devices. + * @param device The representation of the specified device. + * The same representation will always be returned for the + * specified device. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + + int (*ANeuralNetworks_getDevice)(uint32_t devIndex, + ANeuralNetworksDevice** device); + + /** + * Get the name of the specified device. + * + * @param device The representation of the specified device. + * @param name The returned name of the specified device. The name will be + * in UTF-8 and will be null-terminated. It will be recognizable + * as a known device name rather than a cryptic string. For + * devices with API level 29 and above, the format of the name is + * {VENDOR}-{DEVICE}, e.g. “google-ipu”. For devices with feature + * level 28 or lower, the name will always be “unknown-device”. + * The name will remain valid for the duration of the application. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworksDevice_getName)(const ANeuralNetworksDevice* device, + const char** name); + + /** + * Get the version of the driver implementation of the specified device. + * + * It’s the responsibility of the driver implementor to insure that this + * version string uniquely distinguishes this implementation from all previous + * implementations. + * + * This version string must not be confused with the feature level which is + * solely defined by {@link ANeuralNetworksDevice_getFeatureLevel}. There is + * no implicit ordering of the versions. For example, it is not possible to + * filter all drivers older than a certain version. + * + * Application developers may use this version string to avoid or prefer + * specific driver implementations. For example, an application may want to do + * so because: + * - A specific version of the driver does not provide the required + * performance, perhaps because of a performance regression. + * - A specific version of the driver has a bug or returns results that + * don’t match the minimum precision requirement for the application. + * + * @param device The representation of the specified device. + * @param version The returned version string of the driver for the specified + * device. The string will be in UTF-8 and will be + * null-terminated. For devices with feature level 28 or lower, + * "UNKNOWN" will be returned. The version string will remain + * valid for the duration of the application. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworksDevice_getVersion)(const ANeuralNetworksDevice* device, + const char** version); + + /** + * Get the supported NNAPI version of the specified device. + * + * Each device has a supported feature level, which is the most advanced + * feature this driver implements. For example, if the driver implements the + * features introduced in Android P, but does not implement the features + * introduced after Android P, the value would be 28. Developers could decide + * whether or not the specified device should be used for a Model that has + * certain feature requirements. + * + * @param device The representation of the specified device. + * @param featureLevel The API level of the most advanced feature this driver + * implements. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworksDevice_getFeatureLevel)( + const ANeuralNetworksDevice* device, int64_t* featureLevel); + + /** + * Get the type of a given device. + * + * The device type can be used to help application developers to distribute + * Machine Learning workloads and other workloads such as graphical rendering. + * E.g., for an app which renders AR scenes based on real time object + * detection results, the developer could choose an ACCELERATOR type device + * for ML workloads, and reserve GPU for graphical rendering. + * + * @param device The representation of the specified device. + * @param type The returned {@link DeviceTypeCode} of the specified device. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworksDevice_getType)(const ANeuralNetworksDevice* device, + int32_t* type); + + /** + * Get the supported operations for a specified set of devices. If multiple + * devices are selected, the supported operation list is a union of supported + * operations of all selected devices. + * + * @param model The model to be queried. + * @param devices The set of devices. Must not contain duplicates. + * @param numDevices The number of devices in the set. + * @param supportedOps The boolean array to be filled. True means supported. + * The size of the boolean array must be at least as large + * as the number of operations in the model. The order of + * elements in the supportedOps array matches the order in + * which the corresponding operations were added to the + * model. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworksModel_getSupportedOperationsForDevices)( + const ANeuralNetworksModel* model, + const ANeuralNetworksDevice* const* devices, uint32_t numDevices, + bool* supportedOps); + + /** + * Create a {@link ANeuralNetworksCompilation} to compile the given model for + * a specified set of devices. If more than one device is specified, the + * compilation will distribute the workload automatically across the devices. + * The model must be fully supported by the specified set of devices. This + * means that ANeuralNetworksModel_getSupportedOperationsForDevices() must + * have returned true for every operation for that model/devices pair. + * + * @param model The {@link ANeuralNetworksModel} to be compiled. + * @param devices The set of devices. Must not contain duplicates. + * @param numDevices The number of devices in the set. + * @param compilation The newly created object or NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA + * if the model is invalid. + * + * Available since API level 29. + */ + int (*ANeuralNetworksCompilation_createForDevices)( + ANeuralNetworksModel* model, const ANeuralNetworksDevice* const* devices, + uint32_t numDevices, ANeuralNetworksCompilation** compilation); + + /** + * Sets the compilation caching signature and the cache directory. + * + * Provides optional caching information to the runtime for faster repeated + * compilation. + * + * See {@link ANeuralNetworksCompilation} for information on multithreaded + * usage. + * + * @param compilation The compilation to be modified. + * @param cacheDir The cache directory to store and retrieve caching data. It + * is recommended to use the code_cache provided by the + * Android runtime. If not using the code_cache, the user + * should choose a directory local to the application, and is + * responsible to manage and clean the cache entries. + * @param token The token provided by the user to specify a model, must be of + * length ANEURALNETWORKS_BYTE_SIZE_OF_CACHE_TOKEN. The user + * should ensure that the token is unique to a model within the + * application. The NNAPI runtime will not detected token + * collisions. If there is a collision, the compilation outcome + * may be incorrect without notifying with error. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + * + * Available since API level 29. + */ + int (*ANeuralNetworksCompilation_setCaching)( + ANeuralNetworksCompilation* compilation, const char* cacheDir, + const uint8_t* token); + + /** + * Schedule synchronous evaluation of the execution. + * + *

    Schedules synchronous evaluation of the execution. Returns once the + * execution has completed and the outputs are ready to be consumed. + *

    + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * See {@link ANeuralNetworksExecution_startCompute} for asynchronous + * execution. Synchronous execution incurs lower overhead than asynchronous + * execution. + * + * Available since API level 29. + * + * @param execution The execution to be scheduled and executed. + * + * @return ANEURALNETWORKS_NO_ERROR if the execution completed normally. + * ANEURALNETWORKS_UNMAPPABLE if the execution input or output memory + * cannot be properly mapped. + */ + int (*ANeuralNetworksExecution_compute)(ANeuralNetworksExecution* execution); + + /** + * Get the dimensional information of the specified output operand of the + * model of the + * {@link ANeuralNetworksExecution}. + * + * On asynchronous execution initiated by {@link + * ANeuralNetworksExecution_startCompute}, + * {@link ANeuralNetworksEvent_wait} must be called prior to this function to + * recuperate the resources used by the execution. + * + * @param execution The execution to be queried. + * @param index The index of the output argument we are querying. It is + * an index into the lists passed to + * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is + * not the index associated with + * {@link ANeuralNetworksModel_addOperand}. + * @param rank The rank of the output operand. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, + * ANEURALNETWORKS_OUTPUT_INSUFFICIENT_SIZE if the target output is + * provided an insufficient buffer at execution time, + * ANEURALNETWORKS_BAD_DATA if the index is invalid. + * + * Available since API level 29. + */ + int (*ANeuralNetworksExecution_getOutputOperandRank)( + ANeuralNetworksExecution* execution, int32_t index, uint32_t* rank); + + /** + * Get the dimensional information of the specified output operand of the + * model of the + * {@link ANeuralNetworksExecution}. The target output operand cannot be a + * scalar. + * + * On asynchronous execution initiated by {@link + * ANeuralNetworksExecution_startCompute}, + * {@link ANeuralNetworksEvent_wait} must be called prior to this function to + * recuperate the resources used by the execution. + * + * @param execution The execution to be queried. + * @param index The index of the output argument we are querying. It is an + * index into the lists passed to + * {@link ANeuralNetworksModel_identifyInputsAndOutputs}. It is + * not the index associated with + * {@link ANeuralNetworksModel_addOperand}. + * @param dimensions The dimension array to be filled. The size of the array + * must be exactly as large as the rank of the output + * operand to be queried in the model. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, + * ANEURALNETWORKS_OUTPUT_INSUFFICIENT_SIZE if the target output is + * provided an insufficient buffer at execution time, + * ANEURALNETWORKS_BAD_DATA if the index is invalid or if the target + * is a scalar. + * + * Available since API level 29. + */ + int (*ANeuralNetworksExecution_getOutputOperandDimensions)( + ANeuralNetworksExecution* execution, int32_t index, uint32_t* dimensions); + + /** + * Create a {@link ANeuralNetworksBurst} to apply the given compilation. + * This only creates the burst object. Computation is only performed once + * {@link ANeuralNetworksExecution_burstCompute} is invoked with a valid + * {@link ANeuralNetworksExecution} and {@link ANeuralNetworksBurst}. + * + *

    The provided compilation must outlive the burst object.

    + * + * Available since API level 29. + * + * @param compilation The {@link ANeuralNetworksCompilation} to be evaluated. + * @param burst The newly created object or NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if successful, ANEURALNETWORKS_BAD_DATA + * if the compilation is invalid. + */ + int (*ANeuralNetworksBurst_create)(ANeuralNetworksCompilation* compilation, + ANeuralNetworksBurst** burst); + + /** + * Destroys the burst object. + * + * Available since API level 29. + * + * @param burst The burst object to be destroyed. Passing NULL is acceptable + * and results in no operation. + */ + void (*ANeuralNetworksBurst_free)(ANeuralNetworksBurst* burst); + + /** + * Schedule synchronous evaluation of the execution on a burst object. + * + *

    Schedules synchronous evaluation of the execution. Returns once the + * execution has completed and the outputs are ready to be consumed.

    + * + *

    There must be at most one {@link ANeuralNetworksExecution} processing at + * any given time for any given burst object. Any + * {@link ANeuralNetworksExecution} launched before the previous has finished + * will result in ANEURALNETWORKS_BAD_STATE.

    + * + * Available since API level 29. + * + * @param burst The burst object to execute on. + * @param execution The execution to be scheduled and executed. The execution + * must be created from the same {@link + * ANeuralNetworksCompilation} as the burst object. + * + * @return ANEURALNETWORKS_NO_ERROR if the execution completed normally. + */ + int (*ANeuralNetworksExecution_burstCompute)( + ANeuralNetworksExecution* execution, ANeuralNetworksBurst* burst); + + /** + * Creates a shared memory object from an AHardwareBuffer handle. + * + * If the shared memory is backed by an AHardwareBuffer of + * AHARDWAREBUFFER_FORMAT_BLOB format, it can be used the same way as + * shared memory created from a file handle. See + * {@link ANeuralNetworksMemory} for a description on how to use this + * shared memory. + * + * If the shared memory is backed by an AHardwareBuffer of a format other + * than AHARDWAREBUFFER_FORMAT_BLOB, it can only be used for Model inputs + * and outputs. When calling + * {@link ANeuralNetworksExecution_setInputFromMemory} or + * {@link ANeuralNetworksExecution_setOutputFromMemory} with the shared + * memory, both offset and length must be set to zero and the entire + * memory region will be associated with the specified input or output + * operand. There is no guarantee that an arbitrary AHardwareBuffer_Format + * and AHardwareBuffer_UsageFlags combination can be used by arbitrary + * devices. The execution will fail if selected set of devices cannot + * consume the buffer. + * + * Calling {@link ANeuralNetworksModel_setOperandValueFromMemory} with + * shared memory backed by an AHardwareBuffer of a format other than + * AHARDWAREBUFFER_FORMAT_BLOB is disallowed. + * + * TODO(miaowang): add documentation about intended usage with + * introspection API. + * + * Available since API level 29. + * + * @param ahwb The AHardwareBuffer handle. + * @param memory The memory object to be created. + * Set to NULL if unsuccessful. + * + * @return ANEURALNETWORKS_NO_ERROR if the request completed normally. + * + * @see AHardwareBuffer + */ + int (*ANeuralNetworksMemory_createFromAHardwareBuffer)( + const AHardwareBuffer* ahwb, ANeuralNetworksMemory** memory); + + /** + * Specifies whether duration of the {@link ANeuralNetworksExecution} is to be + * measured. By default, duration is not measured. + * + * The {@link ANeuralNetworksExecution} must have been created with + * {@link ANeuralNetworksCompilation_createForDevices} with numDevices = 1. + * + * See {@link ANeuralNetworksExecution} for information on multithreaded + * usage. + * + * Available since API level 29. + * + * @param execution The execution to be modified. + * @param measure 'true' if duration is to be measured, 'false' if not. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksExecution_setMeasureTiming)( + ANeuralNetworksExecution* execution, bool measure); + + /** + * Get the time spent in the specified {@link ANeuralNetworksExecution}, in + * nanoseconds. The execution must have completed. + * + * @param execution The execution to be queried. + * @param durationCode The measurement to be queried, specified by {@link + * DurationCode}. + * @param duration The returned duration. If no measurement was requested by + * {@link ANeuralNetworksExecution_setMeasureTiming}, or for + * some other reason the duration is not available, UINT64_MAX will be + * returned. A particular device need not support any given measurement. + * + * @return ANEURALNETWORKS_NO_ERROR if successful. + */ + int (*ANeuralNetworksExecution_getDuration)( + const ANeuralNetworksExecution* execution, int32_t durationCode, + uint64_t* duration); + + /**/ + +}; + +/** + * Load the NNAPI implementation from the shared libraries. + * The NnApi structure is filled with all the pointers. If one function doesn't + * exist, a null pointer is stored. + */ +const NnApi* NnApiImplementation(); + +#ifdef __ANDROID__ +int32_t GetAndroidSdkVersion(); +#endif + +#endif // TENSORFLOW_LITE_NNAPI_NNAPI_IMPLEMENTATION_H_