load("@bazel_skylib//lib:paths.bzl", "paths") load("@pybind11_bazel//:build_defs.bzl", "pybind_extension") load("@rules_cc//cc:defs.bzl", "cc_binary", "cc_library", "cc_test") load("@rules_python//python:defs.bzl", "py_library", "py_test") load("@pytorch//third_party:substitution.bzl", "header_template_rule", "template_rule") load("@pytorch//:tools/bazel.bzl", "rules") load("@pytorch//tools/rules:cu.bzl", "cu_library") load("@pytorch//tools/config:defs.bzl", "if_cuda") load("@pytorch//:aten.bzl", "generate_aten", "intern_build_aten_ops") load(":build.bzl", "GENERATED_AUTOGRAD_CPP", "GENERATED_AUTOGRAD_PYTHON", "define_targets") load(":build_variables.bzl", "jit_core_sources", "lazy_tensor_ts_sources", "libtorch_core_sources", "libtorch_cuda_sources", "libtorch_distributed_sources", "libtorch_extra_sources", "libtorch_nvfuser_generated_headers", "libtorch_nvfuser_runtime_sources", "libtorch_python_core_sources", "torch_cpp_srcs", "libtorch_python_cuda_sources", "libtorch_python_distributed_sources") load(":ufunc_defs.bzl", "aten_ufunc_generated_cpu_kernel_sources", "aten_ufunc_generated_cpu_sources", "aten_ufunc_generated_cuda_sources") load("//:tools/bazel.bzl", "rules") define_targets(rules = rules) COMMON_COPTS = [ "-DHAVE_MALLOC_USABLE_SIZE=1", "-DHAVE_MMAP=1", "-DHAVE_SHM_OPEN=1", "-DHAVE_SHM_UNLINK=1", "-D_FILE_OFFSET_BITS=64", "-DTH_HAVE_THREAD", "-DUSE_FBGEMM", "-DUSE_DISTRIBUTED", "-DAT_PER_OPERATOR_HEADERS", "-DATEN_THREADING=NATIVE", "-DNO_CUDNN_DESTROY_HANDLE", ] + if_cuda([ "-DUSE_CUDA", "-DUSE_CUDNN", # TODO: This should be passed only when building for CUDA-11.5 or newer # use cub in a safe manner, see: # https://github.com/pytorch/pytorch/pull/55292 "-DCUB_WRAPPED_NAMESPACE=at_cuda_detail", ]) aten_generation_srcs = ["aten/src/ATen/native/native_functions.yaml"] + ["aten/src/ATen/native/tags.yaml"] + glob(["aten/src/ATen/templates/**"]) generated_cpu_cpp = [ "aten/src/ATen/RegisterBackendSelect.cpp", "aten/src/ATen/RegisterCPU.cpp", "aten/src/ATen/RegisterFunctionalization_0.cpp", "aten/src/ATen/RegisterFunctionalization_1.cpp", "aten/src/ATen/RegisterFunctionalization_2.cpp", "aten/src/ATen/RegisterFunctionalization_3.cpp", # "aten/src/ATen/RegisterFunctionalizationEverything.cpp", "aten/src/ATen/RegisterMkldnnCPU.cpp", "aten/src/ATen/RegisterNestedTensorCPU.cpp", "aten/src/ATen/RegisterQuantizedCPU.cpp", "aten/src/ATen/RegisterSparseCPU.cpp", "aten/src/ATen/RegisterSparseCsrCPU.cpp", "aten/src/ATen/RegisterZeroTensor.cpp", "aten/src/ATen/RegisterCompositeImplicitAutograd.cpp", "aten/src/ATen/RegisterCompositeImplicitAutogradNestedTensor.cpp", "aten/src/ATen/RegisterCompositeExplicitAutograd.cpp", "aten/src/ATen/RegisterCompositeExplicitAutogradNonFunctional.cpp", "aten/src/ATen/RegisterMeta.cpp", "aten/src/ATen/RegisterSparseMeta.cpp", "aten/src/ATen/RegisterQuantizedMeta.cpp", "aten/src/ATen/RegisterNestedTensorMeta.cpp", "aten/src/ATen/RegisterSchema.cpp", "aten/src/ATen/CPUFunctions.h", "aten/src/ATen/CPUFunctions_inl.h", "aten/src/ATen/CompositeExplicitAutogradFunctions.h", "aten/src/ATen/CompositeExplicitAutogradFunctions_inl.h", "aten/src/ATen/CompositeExplicitAutogradNonFunctionalFunctions.h", "aten/src/ATen/CompositeExplicitAutogradNonFunctionalFunctions_inl.h", "aten/src/ATen/CompositeImplicitAutogradFunctions.h", "aten/src/ATen/CompositeImplicitAutogradFunctions_inl.h", "aten/src/ATen/CompositeImplicitAutogradNestedTensorFunctions.h", "aten/src/ATen/CompositeImplicitAutogradNestedTensorFunctions_inl.h", "aten/src/ATen/CompositeViewCopyKernels.cpp", "aten/src/ATen/FunctionalInverses.h", "aten/src/ATen/Functions.h", "aten/src/ATen/Functions.cpp", "aten/src/ATen/RedispatchFunctions.h", "aten/src/ATen/Operators.h", "aten/src/ATen/Operators_0.cpp", "aten/src/ATen/Operators_1.cpp", "aten/src/ATen/Operators_2.cpp", "aten/src/ATen/Operators_3.cpp", "aten/src/ATen/Operators_4.cpp", "aten/src/ATen/NativeFunctions.h", "aten/src/ATen/MetaFunctions.h", "aten/src/ATen/MetaFunctions_inl.h", "aten/src/ATen/MethodOperators.h", "aten/src/ATen/NativeMetaFunctions.h", "aten/src/ATen/RegistrationDeclarations.h", "aten/src/ATen/VmapGeneratedPlumbing.h", "aten/src/ATen/core/aten_interned_strings.h", "aten/src/ATen/core/enum_tag.h", "aten/src/ATen/core/TensorBody.h", "aten/src/ATen/core/TensorMethods.cpp", "aten/src/ATen/core/ATenOpList.cpp", ] generated_cuda_cpp = [ "aten/src/ATen/CUDAFunctions.h", "aten/src/ATen/CUDAFunctions_inl.h", "aten/src/ATen/RegisterCUDA.cpp", "aten/src/ATen/RegisterNestedTensorCUDA.cpp", "aten/src/ATen/RegisterQuantizedCUDA.cpp", "aten/src/ATen/RegisterSparseCUDA.cpp", "aten/src/ATen/RegisterSparseCsrCUDA.cpp", ] generate_aten( name = "generated_aten_cpp", srcs = aten_generation_srcs, outs = ( generated_cpu_cpp + generated_cuda_cpp + aten_ufunc_generated_cpu_sources("aten/src/ATen/{}") + aten_ufunc_generated_cpu_kernel_sources("aten/src/ATen/{}") + aten_ufunc_generated_cuda_sources("aten/src/ATen/{}") + [ "aten/src/ATen/Declarations.yaml", ] ), generator = "//torchgen:gen", ) filegroup( name = "cpp_generated_code", srcs = GENERATED_AUTOGRAD_CPP, data = [":generate-code"], ) exports_files( srcs = ["aten/src/ATen/cpu/tbb/extra/version_string.ver.in"], ) # ATen filegroup( name = "aten_base_cpp", srcs = glob([ "aten/src/ATen/*.cpp", "aten/src/ATen/functorch/*.cpp", "aten/src/ATen/detail/*.cpp", "aten/src/ATen/cpu/*.cpp", ]), ) filegroup( name = "ATen_CORE_SRCS", srcs = glob( [ "aten/src/ATen/core/**/*.cpp", ], exclude = [ "aten/src/ATen/core/**/*_test.cpp", ], ), ) filegroup( name = "aten_native_cpp", srcs = glob(["aten/src/ATen/native/*.cpp"]), ) filegroup( name = "aten_native_sparse_cpp", srcs = glob(["aten/src/ATen/native/sparse/*.cpp"]), ) filegroup( name = "aten_native_nested_cpp", srcs = glob(["aten/src/ATen/native/nested/*.cpp"]), ) filegroup( name = "aten_native_quantized_cpp", srcs = glob( [ "aten/src/ATen/native/quantized/*.cpp", "aten/src/ATen/native/quantized/cpu/*.cpp", ], ), ) filegroup( name = "aten_native_transformers_cpp", srcs = glob(["aten/src/ATen/native/transformers/*.cpp"]), ) filegroup( name = "aten_native_mkl_cpp", srcs = glob([ "aten/src/ATen/native/mkl/*.cpp", "aten/src/ATen/mkl/*.cpp", ]), ) filegroup( name = "aten_native_mkldnn_cpp", srcs = glob(["aten/src/ATen/native/mkldnn/*.cpp"]), ) filegroup( name = "aten_native_xnnpack", srcs = glob(["aten/src/ATen/native/xnnpack/*.cpp"]), ) filegroup( name = "aten_base_vulkan", srcs = glob(["aten/src/ATen/vulkan/*.cpp"]), ) filegroup( name = "aten_base_metal", srcs = glob(["aten/src/ATen/metal/*.cpp"]), ) filegroup( name = "ATen_QUANTIZED_SRCS", srcs = glob( [ "aten/src/ATen/quantized/**/*.cpp", ], exclude = [ "aten/src/ATen/quantized/**/*_test.cpp", ], ), ) filegroup( name = "aten_cuda_cpp_srcs", srcs = glob( [ "aten/src/ATen/cuda/*.cpp", "aten/src/ATen/cuda/detail/*.cpp", "aten/src/ATen/cudnn/*.cpp", "aten/src/ATen/native/cuda/*.cpp", "aten/src/ATen/native/cuda/linalg/*.cpp", "aten/src/ATen/native/cudnn/*.cpp", "aten/src/ATen/native/miopen/*.cpp", "aten/src/ATen/native/nested/cuda/*.cpp", "aten/src/ATen/native/quantized/cuda/*.cpp", "aten/src/ATen/native/quantized/cudnn/*.cpp", "aten/src/ATen/native/sparse/cuda/*.cpp", "aten/src/ATen/native/transformers/cuda/*.cpp", ], ), ) filegroup( name = "aten_cu_srcs", srcs = glob([ "aten/src/ATen/cuda/*.cu", "aten/src/ATen/cuda/detail/*.cu", "aten/src/ATen/native/cuda/*.cu", "aten/src/ATen/native/nested/cuda/*.cu", "aten/src/ATen/native/quantized/cuda/*.cu", "aten/src/ATen/native/sparse/cuda/*.cu", "aten/src/ATen/native/transformers/cuda/*.cu", ]) + aten_ufunc_generated_cuda_sources("aten/src/ATen/{}"), # It's a bit puzzling to me why it's not necessary to declare the # target that generates these sources... ) header_template_rule( name = "aten_src_ATen_config", src = "aten/src/ATen/Config.h.in", out = "aten/src/ATen/Config.h", include = "aten/src", substitutions = { "@AT_MKLDNN_ENABLED@": "1", "@AT_MKLDNN_ACL_ENABLED@": "0", "@AT_MKL_ENABLED@": "1", "@AT_MKL_SEQUENTIAL@": "0", "@AT_FFTW_ENABLED@": "0", "@AT_POCKETFFT_ENABLED@": "0", "@AT_NNPACK_ENABLED@": "0", "@CAFFE2_STATIC_LINK_CUDA_INT@": "0", "@AT_BUILD_WITH_BLAS@": "1", "@AT_BUILD_WITH_LAPACK@": "1", "@AT_PARALLEL_OPENMP@": "0", "@AT_PARALLEL_NATIVE@": "1", "@AT_PARALLEL_NATIVE_TBB@": "0", "@AT_BLAS_F2C@": "0", "@AT_BLAS_USE_CBLAS_DOT@": "1", }, ) header_template_rule( name = "aten_src_ATen_cuda_config", src = "aten/src/ATen/cuda/CUDAConfig.h.in", out = "aten/src/ATen/cuda/CUDAConfig.h", include = "aten/src", substitutions = { "@AT_CUDNN_ENABLED@": "1", "@AT_CUSPARSELT_ENABLED@": "0", "@AT_ROCM_ENABLED@": "0", "@AT_MAGMA_ENABLED@": "0", "@NVCC_FLAGS_EXTRA@": "", }, ) cc_library( name = "aten_headers", hdrs = [ "torch/csrc/Export.h", "torch/csrc/jit/frontend/function_schema_parser.h", ] + glob( [ "aten/src/**/*.h", "aten/src/**/*.hpp", "aten/src/ATen/cuda/**/*.cuh", "aten/src/ATen/native/**/*.cuh", "aten/src/THC/*.cuh", ], ) + [ ":aten_src_ATen_config", ":generated_aten_cpp", ], includes = [ "aten/src", ], deps = [ "//c10", ], ) ATEN_COPTS = COMMON_COPTS + [ "-DCAFFE2_BUILD_MAIN_LIBS", "-DHAVE_AVX_CPU_DEFINITION", "-DHAVE_AVX2_CPU_DEFINITION", "-fvisibility-inlines-hidden", "-fno-math-errno", "-fno-trapping-math", ] intern_build_aten_ops( copts = ATEN_COPTS, extra_impls = aten_ufunc_generated_cpu_kernel_sources("aten/src/ATen/{}"), deps = [ ":aten_headers", "@fbgemm", "@mkl", "@sleef", ], ) cc_library( name = "aten", srcs = [ ":ATen_CORE_SRCS", ":ATen_QUANTIZED_SRCS", ":aten_base_cpp", ":aten_base_metal", ":aten_base_vulkan", ":aten_native_cpp", ":aten_native_mkl_cpp", ":aten_native_mkldnn_cpp", ":aten_native_nested_cpp", ":aten_native_quantized_cpp", ":aten_native_sparse_cpp", ":aten_native_transformers_cpp", ":aten_native_xnnpack", ":aten_src_ATen_config", ] + generated_cpu_cpp + aten_ufunc_generated_cpu_sources("aten/src/ATen/{}"), copts = ATEN_COPTS, data = if_cuda( [":libcaffe2_nvrtc.so"], [], ), visibility = ["//visibility:public"], deps = [ ":ATen_CPU", ":aten_headers", ":caffe2_for_aten_headers", ":torch_headers", "@fbgemm", "@ideep", ], alwayslink = True, ) cc_library( name = "aten_nvrtc", srcs = glob([ "aten/src/ATen/cuda/nvrtc_stub/*.cpp", ]), copts = ATEN_COPTS, linkstatic = True, visibility = ["//visibility:public"], deps = [ ":aten_headers", "//c10", "@cuda", "@cuda//:cuda_driver", "@cuda//:nvrtc", ], alwayslink = True, ) cc_binary( name = "libcaffe2_nvrtc.so", linkshared = True, visibility = ["//visibility:public"], deps = [ ":aten_nvrtc", ], ) cc_library( name = "aten_cuda_cpp", srcs = [":aten_cuda_cpp_srcs"] + generated_cuda_cpp, hdrs = [":aten_src_ATen_cuda_config"], copts = ATEN_COPTS, visibility = ["//visibility:public"], deps = [ ":aten", "@cuda", "@cuda//:cusolver", "@cuda//:nvrtc", "@cudnn", ], alwayslink = True, ) torch_cuda_half_options = [ "-DCUDA_HAS_FP16=1", "-D__CUDA_NO_HALF_OPERATORS__", "-D__CUDA_NO_HALF_CONVERSIONS__", "-D__CUDA_NO_BFLOAT16_CONVERSIONS__", "-D__CUDA_NO_HALF2_OPERATORS__", ] cu_library( name = "aten_cuda", srcs = [":aten_cu_srcs"], copts = ATEN_COPTS + torch_cuda_half_options, visibility = ["//visibility:public"], deps = [ ":aten_cuda_cpp", "//c10/util:bit_cast", "@cuda//:cublas", "@cuda//:cufft", "@cuda//:cusparse", "@cutlass", ], alwayslink = True, ) # caffe2 CAFFE2_COPTS = COMMON_COPTS + [ "-Dcaffe2_EXPORTS", "-DCAFFE2_USE_GLOO", "-DCAFFE2_USE_CUDNN", "-DCAFFE2_BUILD_MAIN_LIB", "-fvisibility-inlines-hidden", "-fno-math-errno", "-fno-trapping-math", ] filegroup( name = "caffe2_contrib_srcs", srcs = [ "caffe2/contrib/aten/aten_op.cc", "caffe2/contrib/gloo/allgather_ops.cc", "caffe2/contrib/gloo/allreduce_ops.cc", "caffe2/contrib/gloo/barrier_ops.cc", "caffe2/contrib/gloo/broadcast_ops.cc", "caffe2/contrib/gloo/common.cc", "caffe2/contrib/gloo/common_world_ops.cc", "caffe2/contrib/gloo/context.cc", "caffe2/contrib/gloo/reduce_scatter_ops.cc", "caffe2/contrib/gloo/store_handler.cc", ], ) filegroup( name = "caffe2_core_srcs", srcs = [ "caffe2/core/allocator.cc", "caffe2/core/blob_serialization.cc", "caffe2/core/blob_stats.cc", "caffe2/core/common.cc", "caffe2/core/context.cc", "caffe2/core/context_base.cc", "caffe2/core/db.cc", "caffe2/core/event.cc", "caffe2/core/export_c10_op_to_caffe2.cc", "caffe2/core/graph.cc", "caffe2/core/init.cc", "caffe2/core/init_denormals.cc", "caffe2/core/init_intrinsics_check.cc", "caffe2/core/init_omp.cc", "caffe2/core/int8_serialization.cc", "caffe2/core/memonger.cc", "caffe2/core/module.cc", "caffe2/core/net.cc", "caffe2/core/net_async_base.cc", "caffe2/core/net_async_scheduling.cc", "caffe2/core/net_async_task.cc", "caffe2/core/net_async_task_future.cc", "caffe2/core/net_async_task_graph.cc", "caffe2/core/net_async_tracing.cc", "caffe2/core/net_dag_utils.cc", "caffe2/core/net_parallel.cc", "caffe2/core/net_simple.cc", "caffe2/core/net_simple_refcount.cc", "caffe2/core/nomnigraph/Representations/NeuralNet.cc", "caffe2/core/nomnigraph/tests/test_util.cc", "caffe2/core/numa.cc", "caffe2/core/operator.cc", "caffe2/core/operator_schema.cc", "caffe2/core/plan_executor.cc", "caffe2/core/prof_dag_counters.cc", "caffe2/core/qtensor.cc", "caffe2/core/qtensor_serialization.cc", "caffe2/core/stats.cc", "caffe2/core/tensor.cc", "caffe2/core/tensor_int8.cc", "caffe2/core/test_utils.cc", "caffe2/core/transform.cc", "caffe2/core/types.cc", "caffe2/core/workspace.cc", ], ) filegroup( name = "caffe2_distributed_srcs", srcs = [ "caffe2/distributed/file_store_handler.cc", "caffe2/distributed/file_store_handler_op.cc", "caffe2/distributed/store_handler.cc", "caffe2/distributed/store_ops.cc", ], ) filegroup( name = "caffe2_ideep_srcs", srcs = [ "caffe2/ideep/operators/adam_op.cc", "caffe2/ideep/operators/channel_shuffle_op.cc", "caffe2/ideep/operators/concat_split_op.cc", "caffe2/ideep/operators/conv_op.cc", "caffe2/ideep/operators/conv_transpose_op.cc", "caffe2/ideep/operators/dropout_op.cc", "caffe2/ideep/operators/elementwise_sum_op.cc", "caffe2/ideep/operators/expand_squeeze_dims_op.cc", "caffe2/ideep/operators/fully_connected_op.cc", "caffe2/ideep/operators/local_response_normalization_op.cc", "caffe2/ideep/operators/momentum_sgd_op.cc", "caffe2/ideep/operators/operator_fallback_ideep.cc", "caffe2/ideep/operators/order_switch_ops.cc", "caffe2/ideep/operators/pool_op.cc", "caffe2/ideep/operators/quantization/int8_add_op.cc", "caffe2/ideep/operators/quantization/int8_conv_op.cc", "caffe2/ideep/operators/quantization/int8_dequantize_op.cc", "caffe2/ideep/operators/quantization/int8_fully_connected_op.cc", "caffe2/ideep/operators/quantization/int8_given_tensor_fill_op.cc", "caffe2/ideep/operators/quantization/int8_pool_op.cc", "caffe2/ideep/operators/quantization/int8_quantize_op.cc", "caffe2/ideep/operators/quantization/int8_relu_op.cc", "caffe2/ideep/operators/queue_ops.cc", "caffe2/ideep/operators/relu_op.cc", "caffe2/ideep/operators/reshape_op.cc", "caffe2/ideep/operators/shape_op.cc", "caffe2/ideep/operators/sigmoid_op.cc", "caffe2/ideep/operators/spatial_batch_norm_op.cc", "caffe2/ideep/operators/transpose_op.cc", "caffe2/ideep/operators/utility_ops.cc", "caffe2/ideep/utils/ideep_register.cc", ], ) filegroup( name = "caffe2_onnx_srcs", srcs = [ "caffe2/onnx/backend.cc", "caffe2/onnx/backend_rep.cc", "caffe2/onnx/device.cc", "caffe2/onnx/helper.cc", "caffe2/onnx/offline_tensor.cc", "caffe2/onnx/onnx_exporter.cc", "caffe2/onnx/onnxifi_graph_info.cc", "caffe2/onnx/onnxifi_init.cc", ], ) filegroup( name = "caffe2_operators_srcs", srcs = [ "caffe2/operators/abs_op.cc", "caffe2/operators/accumulate_op.cc", "caffe2/operators/accuracy_op.cc", "caffe2/operators/acos_op.cc", "caffe2/operators/affine_channel_op.cc", "caffe2/operators/alias_with_name.cc", "caffe2/operators/apmeter_op.cc", "caffe2/operators/arg_ops.cc", "caffe2/operators/asin_op.cc", "caffe2/operators/assert_op.cc", "caffe2/operators/atan_op.cc", "caffe2/operators/atomic_ops.cc", "caffe2/operators/batch_box_cox_op.cc", "caffe2/operators/batch_bucketize_op.cc", "caffe2/operators/batch_gather_ops.cc", "caffe2/operators/batch_matmul_op.cc", "caffe2/operators/batch_moments_op.cc", "caffe2/operators/batch_permutation_op.cc", "caffe2/operators/batch_sparse_to_dense_op.cc", "caffe2/operators/bbox_transform_op.cc", "caffe2/operators/bisect_percentile_op.cc", "caffe2/operators/boolean_mask_ops.cc", "caffe2/operators/boolean_unmask_ops.cc", "caffe2/operators/box_with_nms_limit_op.cc", "caffe2/operators/bucketize_op.cc", "caffe2/operators/byte_weight_dequant_op.cc", "caffe2/operators/cast_op.cc", "caffe2/operators/cbrt_op.cc", "caffe2/operators/cc_bmm_bg_op.cc", "caffe2/operators/ceil_op.cc", "caffe2/operators/channel_backprop_stats_op.cc", "caffe2/operators/channel_shuffle_op.cc", "caffe2/operators/channel_stats_op.cc", "caffe2/operators/clip_op.cc", "caffe2/operators/collect_and_distribute_fpn_rpn_proposals_op.cc", "caffe2/operators/communicator_op.cc", "caffe2/operators/concat_split_op.cc", "caffe2/operators/conditional_op.cc", "caffe2/operators/conv_gradient_op.cc", "caffe2/operators/conv_op.cc", "caffe2/operators/conv_op_eigen.cc", "caffe2/operators/conv_op_shared.cc", "caffe2/operators/conv_transpose_gradient_op.cc", "caffe2/operators/conv_transpose_op.cc", "caffe2/operators/conv_transpose_op_mobile.cc", "caffe2/operators/copy_op.cc", "caffe2/operators/copy_rows_to_tensor_op.cc", "caffe2/operators/cos_op.cc", "caffe2/operators/cosh_op.cc", "caffe2/operators/cosine_embedding_criterion_op.cc", "caffe2/operators/counter_ops.cc", "caffe2/operators/crash_op.cc", "caffe2/operators/create_scope_op.cc", "caffe2/operators/crf_viterbi_op.cc", "caffe2/operators/cross_entropy_op.cc", "caffe2/operators/ctc_beam_search_decoder_op.cc", "caffe2/operators/ctc_greedy_decoder_op.cc", "caffe2/operators/cube_op.cc", "caffe2/operators/data_couple.cc", "caffe2/operators/dataset_ops.cc", "caffe2/operators/deform_conv_gradient_op.cc", "caffe2/operators/deform_conv_op.cc", "caffe2/operators/dense_vector_to_id_list_op.cc", "caffe2/operators/distance_op.cc", "caffe2/operators/do_op.cc", "caffe2/operators/dropout_op.cc", "caffe2/operators/elementwise_add_gradient_op.cc", "caffe2/operators/elementwise_add_op.cc", "caffe2/operators/elementwise_div_gradient_op.cc", "caffe2/operators/elementwise_div_op.cc", "caffe2/operators/elementwise_linear_op.cc", "caffe2/operators/elementwise_logical_ops.cc", "caffe2/operators/elementwise_mul_gradient_op.cc", "caffe2/operators/elementwise_mul_op.cc", "caffe2/operators/elementwise_ops.cc", "caffe2/operators/elementwise_ops_schema.cc", "caffe2/operators/elementwise_ops_utils.cc", "caffe2/operators/elementwise_sub_gradient_op.cc", "caffe2/operators/elementwise_sub_op.cc", "caffe2/operators/elementwise_sum_op.cc", "caffe2/operators/elu_op.cc", "caffe2/operators/enforce_finite_op.cc", "caffe2/operators/ensure_clipped_op.cc", "caffe2/operators/ensure_cpu_output_op.cc", "caffe2/operators/erf_op.cc", "caffe2/operators/exp_op.cc", "caffe2/operators/expand_op.cc", "caffe2/operators/expand_squeeze_dims_op.cc", "caffe2/operators/fc_inference.cc", "caffe2/operators/feature_maps_ops.cc", "caffe2/operators/feed_blob_op.cc", "caffe2/operators/filler_op.cc", "caffe2/operators/find_duplicate_elements_op.cc", "caffe2/operators/find_op.cc", "caffe2/operators/flatten_op.cc", "caffe2/operators/flexible_top_k.cc", "caffe2/operators/floor_op.cc", "caffe2/operators/free_op.cc", "caffe2/operators/fully_connected_op.cc", "caffe2/operators/fused_rowwise_8bit_conversion_ops.cc", "caffe2/operators/fused_rowwise_random_quantization_ops.cc", "caffe2/operators/gather_fused_8bit_rowwise_op.cc", "caffe2/operators/gather_op.cc", "caffe2/operators/gather_ranges_to_dense_op.cc", "caffe2/operators/gelu_op.cc", "caffe2/operators/generate_proposals_op.cc", "caffe2/operators/given_tensor_byte_string_to_uint8_fill_op.cc", "caffe2/operators/given_tensor_fill_op.cc", "caffe2/operators/glu_op.cc", "caffe2/operators/group_norm_op.cc", "caffe2/operators/gru_unit_op.cc", "caffe2/operators/h_softmax_op.cc", "caffe2/operators/half_float_ops.cc", "caffe2/operators/hard_sigmoid_op.cc", "caffe2/operators/heatmap_max_keypoint_op.cc", "caffe2/operators/if_op.cc", "caffe2/operators/im2col_op.cc", "caffe2/operators/index_hash_ops.cc", "caffe2/operators/index_ops.cc", "caffe2/operators/inference_lstm_op.cc", "caffe2/operators/instance_norm_gradient_op.cc", "caffe2/operators/instance_norm_op.cc", "caffe2/operators/integral_image_op.cc", "caffe2/operators/is_empty_op.cc", "caffe2/operators/jsd_op.cc", "caffe2/operators/key_split_ops.cc", "caffe2/operators/last_n_window_collector.cc", "caffe2/operators/layer_norm_op.cc", "caffe2/operators/leaky_relu_op.cc", "caffe2/operators/length_split_op.cc", "caffe2/operators/lengths_pad_op.cc", "caffe2/operators/lengths_reducer_fused_8bit_rowwise_ops.cc", "caffe2/operators/lengths_reducer_ops.cc", "caffe2/operators/lengths_reducer_rowwise_8bit_ops.cc", "caffe2/operators/lengths_tile_op.cc", "caffe2/operators/lengths_top_k_op.cc", "caffe2/operators/listwise_l2r_op.cc", "caffe2/operators/load_save_op.cc", "caffe2/operators/load_save_op_util.cc", "caffe2/operators/local_response_normalization_op.cc", "caffe2/operators/locally_connected_op.cc", "caffe2/operators/locally_connected_op_util.cc", "caffe2/operators/log_op.cc", "caffe2/operators/logit_op.cc", "caffe2/operators/loss_op.cc", "caffe2/operators/lp_pool_op.cc", "caffe2/operators/lpnorm_op.cc", "caffe2/operators/lstm_unit_op.cc", "caffe2/operators/map_ops.cc", "caffe2/operators/margin_ranking_criterion_op.cc", "caffe2/operators/matmul_op.cc", "caffe2/operators/mean_op.cc", "caffe2/operators/merge_id_lists_op.cc", "caffe2/operators/minmax_gradient_ops.cc", "caffe2/operators/minmax_ops.cc", "caffe2/operators/mod_op.cc", "caffe2/operators/moments_op.cc", "caffe2/operators/multi_class_accuracy_op.cc", "caffe2/operators/negate_gradient_op.cc", "caffe2/operators/negative_op.cc", "caffe2/operators/ngram_ops.cc", "caffe2/operators/norm_planar_yuv_op.cc", "caffe2/operators/normalize_l1_op.cc", "caffe2/operators/normalize_op.cc", "caffe2/operators/numpy_tile_op.cc", "caffe2/operators/one_hot_ops.cc", "caffe2/operators/onnx_while_op.cc", "caffe2/operators/order_switch_ops.cc", "caffe2/operators/pack_rnn_sequence_op.cc", "caffe2/operators/pack_segments.cc", "caffe2/operators/pad_op.cc", "caffe2/operators/partition_ops.cc", "caffe2/operators/percentile_op.cc", "caffe2/operators/perplexity_op.cc", "caffe2/operators/piecewise_linear_transform_op.cc", "caffe2/operators/pool_gradient_op.cc", "caffe2/operators/pool_op.cc", "caffe2/operators/pool_op_util.cc", "caffe2/operators/pow_op.cc", "caffe2/operators/prelu_op.cc", "caffe2/operators/prepend_dim_op.cc", "caffe2/operators/quant_decode_op.cc", "caffe2/operators/rank_loss_op.cc", "caffe2/operators/reciprocal_gradient_op.cc", "caffe2/operators/reciprocal_op.cc", "caffe2/operators/reduce_front_back_max_ops.cc", "caffe2/operators/reduce_front_back_mean_ops.cc", "caffe2/operators/reduce_front_back_sum_ops.cc", "caffe2/operators/reduce_ops.cc", "caffe2/operators/reduction_ops.cc", "caffe2/operators/relu_n_op.cc", "caffe2/operators/relu_op.cc", "caffe2/operators/remove_data_blocks_op.cc", "caffe2/operators/replace_nan_op.cc", "caffe2/operators/reservoir_sampling.cc", "caffe2/operators/reshape_op.cc", "caffe2/operators/resize_3d_op.cc", "caffe2/operators/resize_op.cc", "caffe2/operators/reverse_packed_segs_op.cc", "caffe2/operators/rmac_regions_op.cc", "caffe2/operators/rnn/recurrent_network_blob_fetcher_op.cc", "caffe2/operators/rnn/recurrent_network_executor.cc", "caffe2/operators/rnn/recurrent_network_op.cc", "caffe2/operators/roi_align_gradient_op.cc", "caffe2/operators/roi_align_op.cc", "caffe2/operators/roi_align_rotated_gradient_op.cc", "caffe2/operators/roi_align_rotated_op.cc", "caffe2/operators/roi_pool_op.cc", "caffe2/operators/rowmul_op.cc", "caffe2/operators/rsqrt_op.cc", "caffe2/operators/scale_blobs_op.cc", "caffe2/operators/scale_op.cc", "caffe2/operators/segment_reduction_op.cc", "caffe2/operators/selu_op.cc", "caffe2/operators/sequence_ops.cc", "caffe2/operators/shape_op.cc", "caffe2/operators/sigmoid_gradient_op.cc", "caffe2/operators/sigmoid_op.cc", "caffe2/operators/sin_op.cc", "caffe2/operators/sinh_op.cc", "caffe2/operators/sinusoid_position_encoding_op.cc", "caffe2/operators/slice_op.cc", "caffe2/operators/softmax_op.cc", "caffe2/operators/softmax_utils.cc", "caffe2/operators/softmax_with_loss_op.cc", "caffe2/operators/softplus_op.cc", "caffe2/operators/softsign_op.cc", "caffe2/operators/space_batch_op.cc", "caffe2/operators/sparse_dropout_with_replacement_op.cc", "caffe2/operators/sparse_normalize_op.cc", "caffe2/operators/sparse_to_dense_mask_op.cc", "caffe2/operators/sparse_to_dense_op.cc", "caffe2/operators/spatial_batch_norm_gradient_op.cc", "caffe2/operators/spatial_batch_norm_op.cc", "caffe2/operators/spatial_softmax_with_loss_op.cc", "caffe2/operators/sqr_op.cc", "caffe2/operators/sqrt_op.cc", "caffe2/operators/square_root_divide_op.cc", "caffe2/operators/stats_ops.cc", "caffe2/operators/stats_put_ops.cc", "caffe2/operators/stop_gradient.cc", "caffe2/operators/string_ops.cc", "caffe2/operators/stump_func_op.cc", "caffe2/operators/stylizer_ops.cc", "caffe2/operators/summarize_op.cc", "caffe2/operators/swish_op.cc", "caffe2/operators/tan_op.cc", "caffe2/operators/tanh_gradient_op.cc", "caffe2/operators/tanh_op.cc", "caffe2/operators/tensor_protos_db_input.cc", "caffe2/operators/text_file_reader.cc", "caffe2/operators/text_file_reader_utils.cc", "caffe2/operators/thresholded_relu_op.cc", "caffe2/operators/tile_op.cc", "caffe2/operators/top_k.cc", "caffe2/operators/transpose_op.cc", "caffe2/operators/tt_linear_op.cc", "caffe2/operators/unique_ops.cc", "caffe2/operators/upsample_op.cc", "caffe2/operators/utility_ops.cc", "caffe2/operators/variable_length_sequence_padding.cc", "caffe2/operators/weighted_multi_sampling_op.cc", "caffe2/operators/weighted_sample_op.cc", "caffe2/operators/while_op.cc", "caffe2/operators/workspace_ops.cc", "caffe2/operators/zero_gradient_op.cc", ], ) filegroup( name = "caffe2_opt_srcs", srcs = [ "caffe2/opt/annotations.cc", "caffe2/opt/backend_cutting.cc", "caffe2/opt/backend_transformer_base.cc", "caffe2/opt/bound_shape_inferencer.cc", "caffe2/opt/converter.cc", "caffe2/opt/dead_code_elim.cc", "caffe2/opt/device.cc", "caffe2/opt/distributed.cc", "caffe2/opt/distributed_converter.cc", "caffe2/opt/fusion.cc", "caffe2/opt/mobile.cc", "caffe2/opt/onnxifi_op.cc", "caffe2/opt/onnxifi_transformer.cc", "caffe2/opt/optimize_ideep.cc", "caffe2/opt/optimizer.cc", "caffe2/opt/passes.cc", "caffe2/opt/shape_info.cc", "caffe2/opt/tvm_transformer.cc", ], ) filegroup( name = "caffe2_perfkernels_srcs", srcs = [ "caffe2/perfkernels/adagrad.cc", "caffe2/perfkernels/embedding_lookup.cc", "caffe2/perfkernels/embedding_lookup_idx.cc", "caffe2/perfkernels/fused_8bit_rowwise_embedding_lookup.cc", "caffe2/perfkernels/fused_8bit_rowwise_embedding_lookup_idx.cc", "caffe2/perfkernels/fused_nbit_rowwise_conversion.cc", "caffe2/perfkernels/lstm_unit_cpu_common.cc", "caffe2/perfkernels/math_cpu_base.cc", "caffe2/perfkernels/typed_axpy.cc", ], ) filegroup( name = "caffe2_predictor_srcs", srcs = [ "caffe2/predictor/emulator/data_filler.cc", "caffe2/predictor/emulator/data_filler.h", "caffe2/predictor/predictor.cc", "caffe2/predictor/predictor_config.cc", "caffe2/predictor/predictor_utils.cc", ], ) filegroup( name = "caffe2_quantization_srcs", srcs = [ "caffe2/quantization/server/activation_distribution_observer.cc", "caffe2/quantization/server/batch_matmul_dnnlowp_op.cc", "caffe2/quantization/server/caffe2_dnnlowp_utils.cc", "caffe2/quantization/server/channel_shuffle_dnnlowp_op.cc", "caffe2/quantization/server/concat_dnnlowp_op.cc", "caffe2/quantization/server/conv_dnnlowp_acc16_op.cc", "caffe2/quantization/server/conv_dnnlowp_op.cc", "caffe2/quantization/server/conv_relu_op.cc", "caffe2/quantization/server/dequantize_dnnlowp_op.cc", "caffe2/quantization/server/dnnlowp.cc", "caffe2/quantization/server/dnnlowp_partition.cc", "caffe2/quantization/server/dynamic_histogram.cc", "caffe2/quantization/server/elementwise_add_dnnlowp_op.cc", "caffe2/quantization/server/elementwise_linear_dnnlowp_op.cc", "caffe2/quantization/server/elementwise_mul_dnnlowp_op.cc", "caffe2/quantization/server/elementwise_sum_dnnlowp_op.cc", "caffe2/quantization/server/elementwise_sum_relu_op.cc", "caffe2/quantization/server/fbgemm_pack_matrix_cache.cc", "caffe2/quantization/server/fbgemm_pack_op.cc", "caffe2/quantization/server/fully_connected_dnnlowp_acc16_op.cc", "caffe2/quantization/server/fully_connected_dnnlowp_op.cc", "caffe2/quantization/server/fully_connected_fake_lowp_op.cc", "caffe2/quantization/server/group_norm_dnnlowp_op.cc", "caffe2/quantization/server/int8_gen_quant_params.cc", "caffe2/quantization/server/kl_minimization.cc", "caffe2/quantization/server/lstm_unit_dnnlowp_op.cc", "caffe2/quantization/server/norm_minimization.cc", "caffe2/quantization/server/p99.cc", "caffe2/quantization/server/pool_dnnlowp_op.cc", "caffe2/quantization/server/quantize_dnnlowp_op.cc", "caffe2/quantization/server/relu_dnnlowp_op.cc", "caffe2/quantization/server/sigmoid.cc", "caffe2/quantization/server/sigmoid_dnnlowp_op.cc", "caffe2/quantization/server/spatial_batch_norm_dnnlowp_op.cc", "caffe2/quantization/server/tanh.cc", "caffe2/quantization/server/tanh_dnnlowp_op.cc", "caffe2/quantization/server/utility_dnnlowp_ops.cc", ], ) filegroup( name = "caffe2_queue_srcs", srcs = [ "caffe2/queue/blobs_queue.cc", "caffe2/queue/blobs_queue_db.cc", "caffe2/queue/queue_ops.cc", "caffe2/queue/rebatching_queue.cc", "caffe2/queue/rebatching_queue_ops.cc", ], ) filegroup( name = "caffe2_serialize_srcs", srcs = [ "caffe2/serialize/file_adapter.cc", "caffe2/serialize/inline_container.cc", "caffe2/serialize/istream_adapter.cc", "caffe2/serialize/read_adapter_interface.cc", ], ) filegroup( name = "caffe2_sgd_srcs", srcs = [ "caffe2/sgd/adadelta_op.cc", "caffe2/sgd/adagrad_op.cc", "caffe2/sgd/adam_op.cc", "caffe2/sgd/clip_tensor_op.cc", "caffe2/sgd/ftrl_op.cc", "caffe2/sgd/gftrl_op.cc", "caffe2/sgd/iter_op.cc", "caffe2/sgd/lars_op.cc", "caffe2/sgd/learning_rate_adaption_op.cc", "caffe2/sgd/learning_rate_op.cc", "caffe2/sgd/momentum_sgd_op.cc", "caffe2/sgd/rmsprop_op.cc", "caffe2/sgd/wngrad_op.cc", "caffe2/sgd/yellowfin_op.cc", ], ) filegroup( name = "caffe2_transforms_srcs", srcs = [ "caffe2/transforms/common_subexpression_elimination.cc", "caffe2/transforms/conv_to_nnpack_transform.cc", "caffe2/transforms/pattern_net_transform.cc", "caffe2/transforms/single_op_transform.cc", ], ) filegroup( name = "caffe2_utils_srcs", srcs = [ "caffe2/utils/bench_utils.cc", "caffe2/utils/cpuid.cc", "caffe2/utils/math/broadcast.cc", "caffe2/utils/math/elementwise.cc", "caffe2/utils/math/reduce.cc", "caffe2/utils/math/transpose.cc", "caffe2/utils/math/utils.cc", "caffe2/utils/math_cpu.cc", "caffe2/utils/murmur_hash3.cc", "caffe2/utils/proto_utils.cc", "caffe2/utils/proto_wrap.cc", "caffe2/utils/signal_handler.cc", "caffe2/utils/smart_tensor_printer.cc", "caffe2/utils/string_utils.cc", "caffe2/utils/threadpool/ThreadPool.cc", "caffe2/utils/threadpool/pthreadpool.cc", "caffe2/utils/threadpool/pthreadpool_impl.cc", "caffe2/utils/threadpool/thread_pool_guard.cpp", ], ) filegroup( name = "caffe2_cuda_cpp_srcs", srcs = [ "caffe2/contrib/aten/aten_op_gpu.cc", "caffe2/contrib/gloo/allreduce_ops_gpu.cc", "caffe2/contrib/gloo/broadcast_ops_gpu.cc", "caffe2/contrib/gloo/common_world_ops_gpu.cc", "caffe2/core/blob_serialization_gpu.cc", "caffe2/core/common_cudnn.cc", "caffe2/core/common_gpu.cc", "caffe2/core/event_gpu.cc", "caffe2/db/create_db_op_gpu.cc", "caffe2/distributed/file_store_handler_op_gpu.cc", "caffe2/operators/communicator_op_gpu.cc", "caffe2/operators/concat_split_op_gpu.cc", "caffe2/operators/conv_op_cache_cudnn.cc", "caffe2/operators/conv_op_cudnn.cc", "caffe2/operators/conv_op_gpu.cc", "caffe2/operators/conv_op_shared_gpu.cc", "caffe2/operators/conv_transpose_op_cudnn.cc", "caffe2/operators/conv_transpose_op_gpu.cc", "caffe2/operators/counter_ops_gpu.cc", "caffe2/operators/do_op_gpu.cc", "caffe2/operators/dropout_op_cudnn.cc", "caffe2/operators/elementwise_add_op_gpu.cc", "caffe2/operators/elementwise_sub_op_gpu.cc", "caffe2/operators/elu_op_cudnn.cc", "caffe2/operators/exp_op_gpu.cc", "caffe2/operators/expand_op_gpu.cc", "caffe2/operators/expand_squeeze_dims_op_gpu.cc", "caffe2/operators/free_op_gpu.cc", "caffe2/operators/fully_connected_op_gpu.cc", "caffe2/operators/if_op_gpu.cc", "caffe2/operators/im2col_op_gpu.cc", "caffe2/operators/load_save_op_gpu.cc", "caffe2/operators/local_response_normalization_op_cudnn.cc", "caffe2/operators/locally_connected_op_gpu.cc", "caffe2/operators/log_op_gpu.cc", "caffe2/operators/matmul_op_gpu.cc", "caffe2/operators/negate_gradient_op_gpu.cc", "caffe2/operators/negative_op_gpu.cc", "caffe2/operators/order_switch_ops_cudnn.cc", "caffe2/operators/order_switch_ops_gpu.cc", "caffe2/operators/pool_op_cudnn.cc", "caffe2/operators/prepend_dim_op_gpu.cc", "caffe2/operators/reshape_op_gpu.cc", "caffe2/operators/rnn/recurrent_network_blob_fetcher_op_gpu.cc", "caffe2/operators/rnn/recurrent_network_executor_gpu.cc", "caffe2/operators/rnn/recurrent_op_cudnn.cc", "caffe2/operators/scale_op_gpu.cc", "caffe2/operators/shape_op_gpu.cc", "caffe2/operators/sigmoid_op_cudnn.cc", "caffe2/operators/softmax_op_cudnn.cc", "caffe2/operators/sqr_op_gpu.cc", "caffe2/operators/sqrt_op_gpu.cc", "caffe2/operators/stop_gradient_gpu.cc", "caffe2/operators/tanh_op_cudnn.cc", "caffe2/operators/tensor_protos_db_input_gpu.cc", "caffe2/operators/transpose_op_cudnn.cc", "caffe2/operators/while_op_gpu.cc", "caffe2/operators/zero_gradient_op_gpu.cc", "caffe2/queue/queue_ops_gpu.cc", "caffe2/sgd/iter_op_gpu.cc", "caffe2/sgd/learning_rate_op_gpu.cc", ], ) filegroup( name = "caffe2_cu_srcs", srcs = [ "caffe2/core/context_gpu.cu", "caffe2/operators/abs_op.cu", "caffe2/operators/accumulate_op.cu", "caffe2/operators/accuracy_op.cu", "caffe2/operators/acos_op.cu", "caffe2/operators/affine_channel_op.cu", "caffe2/operators/alias_with_name.cu", "caffe2/operators/arg_ops.cu", "caffe2/operators/asin_op.cu", "caffe2/operators/assert_op.cu", "caffe2/operators/atan_op.cu", "caffe2/operators/batch_gather_ops.cu", "caffe2/operators/batch_matmul_op.cu", "caffe2/operators/batch_moments_op.cu", "caffe2/operators/batch_permutation_op.cu", "caffe2/operators/batch_sparse_to_dense_op.cu", "caffe2/operators/boolean_mask_ops.cu", "caffe2/operators/boolean_unmask_ops.cu", "caffe2/operators/bucketize_op.cu", "caffe2/operators/cast_op.cu", "caffe2/operators/cbrt_op.cu", "caffe2/operators/ceil_op.cu", "caffe2/operators/channel_backprop_stats_op.cu", "caffe2/operators/channel_shuffle_op.cu", "caffe2/operators/channel_stats_op.cu", "caffe2/operators/channelwise_conv3d_op_cudnn.cu", "caffe2/operators/clip_op.cu", "caffe2/operators/copy_op.cu", "caffe2/operators/cos_op.cu", "caffe2/operators/cosh_op.cu", "caffe2/operators/cosine_embedding_criterion_op.cu", "caffe2/operators/cross_entropy_op.cu", "caffe2/operators/cube_op.cu", "caffe2/operators/data_couple_gpu.cu", "caffe2/operators/deform_conv_op.cu", "caffe2/operators/depthwise_3x3_conv_op_cudnn.cu", "caffe2/operators/distance_op.cu", "caffe2/operators/dropout_op.cu", "caffe2/operators/elementwise_div_op.cu", "caffe2/operators/elementwise_linear_op.cu", "caffe2/operators/elementwise_mul_op.cu", "caffe2/operators/elementwise_ops.cu", "caffe2/operators/elu_op.cu", "caffe2/operators/enforce_finite_op.cu", "caffe2/operators/ensure_cpu_output_op.cu", "caffe2/operators/erf_op.cu", "caffe2/operators/filler_op.cu", "caffe2/operators/find_op.cu", "caffe2/operators/floor_op.cu", "caffe2/operators/gather_op.cu", "caffe2/operators/gelu_op.cu", "caffe2/operators/generate_proposals_op.cu", "caffe2/operators/generate_proposals_op_util_nms_gpu.cu", "caffe2/operators/given_tensor_byte_string_to_uint8_fill_op.cu", "caffe2/operators/given_tensor_fill_op.cu", "caffe2/operators/glu_op.cu", "caffe2/operators/group_norm_op.cu", "caffe2/operators/gru_unit_op_gpu.cu", "caffe2/operators/half_float_ops.cu", "caffe2/operators/hard_sigmoid_op.cu", "caffe2/operators/instance_norm_op.cu", "caffe2/operators/integral_image_op.cu", "caffe2/operators/layer_norm_op.cu", "caffe2/operators/leaky_relu_op.cu", "caffe2/operators/lengths_pad_op.cu", "caffe2/operators/lengths_tile_op.cu", "caffe2/operators/local_response_normalization_op.cu", "caffe2/operators/logit_op.cu", "caffe2/operators/loss_op.cu", "caffe2/operators/lp_pool_op.cu", "caffe2/operators/lstm_unit_op_gpu.cu", "caffe2/operators/margin_ranking_criterion_op.cu", "caffe2/operators/max_pool_with_index.cu", "caffe2/operators/mean_op.cu", "caffe2/operators/mem_query_op.cu", "caffe2/operators/minmax_ops.cu", "caffe2/operators/moments_op.cu", "caffe2/operators/multi_class_accuracy_op.cu", "caffe2/operators/normalize_ops.cu", "caffe2/operators/one_hot_ops.cu", "caffe2/operators/pack_segments.cu", "caffe2/operators/pad_op_gpu.cu", "caffe2/operators/perplexity_op.cu", "caffe2/operators/piecewise_linear_transform_op.cu", "caffe2/operators/pool_op.cu", "caffe2/operators/pow_op.cu", "caffe2/operators/prelu_op.cu", "caffe2/operators/reciprocal_op.cu", "caffe2/operators/reduce_front_back_max_ops.cu", "caffe2/operators/reduce_front_back_sum_mean_ops.cu", "caffe2/operators/reduce_ops.cu", "caffe2/operators/reduction_ops.cu", "caffe2/operators/relu_n_op.cu", "caffe2/operators/relu_op.cu", "caffe2/operators/replace_nan_op.cu", "caffe2/operators/resize_3d_op.cu", "caffe2/operators/resize_op.cu", "caffe2/operators/reverse_packed_segs_op.cu", "caffe2/operators/rmac_regions_op.cu", "caffe2/operators/rnn/recurrent_network_op_gpu.cu", "caffe2/operators/roi_align_gradient_op.cu", "caffe2/operators/roi_align_op.cu", "caffe2/operators/roi_align_rotated_gradient_op.cu", "caffe2/operators/roi_align_rotated_op.cu", "caffe2/operators/roi_pool_op.cu", "caffe2/operators/rsqrt_op.cu", "caffe2/operators/scale_blobs_op.cu", "caffe2/operators/segment_reduction_op_gpu.cu", "caffe2/operators/selu_op.cu", "caffe2/operators/sequence_ops.cu", "caffe2/operators/sigmoid_op.cu", "caffe2/operators/sin_op.cu", "caffe2/operators/sinh_op.cu", "caffe2/operators/slice_op.cu", "caffe2/operators/softmax_ops.cu", "caffe2/operators/softplus_op.cu", "caffe2/operators/softsign_op.cu", "caffe2/operators/space_batch_op_gpu.cu", "caffe2/operators/sparse_normalize_op_gpu.cu", "caffe2/operators/sparse_to_dense_op.cu", "caffe2/operators/spatial_batch_norm_op.cu", "caffe2/operators/spatial_batch_norm_op_cudnn.cu", "caffe2/operators/stump_func_op.cu", "caffe2/operators/summarize_op.cu", "caffe2/operators/swish_op.cu", "caffe2/operators/tan_op.cu", "caffe2/operators/tanh_op.cu", "caffe2/operators/thresholded_relu_op.cu", "caffe2/operators/tile_op.cu", "caffe2/operators/top_k.cu", "caffe2/operators/transpose_op.cu", "caffe2/operators/unique_ops.cu", "caffe2/operators/upsample_op.cu", "caffe2/operators/utility_ops.cu", "caffe2/operators/weighted_sample_op.cu", "caffe2/sgd/adadelta_op_gpu.cu", "caffe2/sgd/adagrad_op_gpu.cu", "caffe2/sgd/adam_op_gpu.cu", "caffe2/sgd/fp16_momentum_sgd_op.cu", "caffe2/sgd/fp32_momentum_sgd_op.cu", "caffe2/sgd/lars_op_gpu.cu", "caffe2/sgd/momentum_sgd_op_gpu.cu", "caffe2/sgd/rmsprop_op_gpu.cu", "caffe2/sgd/yellowfin_op_gpu.cu", "caffe2/utils/math/broadcast.cu", "caffe2/utils/math/elementwise.cu", "caffe2/utils/math/reduce.cu", "caffe2/utils/math/transpose.cu", "caffe2/utils/math_gpu.cu", ], ) # To achieve finer granularity and make debug easier, caffe2 is split into three libraries: # ATen, caffe2 and caffe2_for_aten_headers. ATen lib group up source codes under # aten/ directory and caffe2 contains most files under `caffe2/` directory. Since the # ATen lib and the caffe2 lib would depend on each other, `caffe2_for_aten_headers` is splitted # out from `caffe2` to avoid dependency cycle. cc_library( name = "caffe2_for_aten_headers", hdrs = [ "caffe2/core/common.h", "caffe2/core/logging.h", "caffe2/core/types.h", "caffe2/perfkernels/common.h", "caffe2/perfkernels/embedding_lookup.h", "caffe2/perfkernels/embedding_lookup_idx.h", "caffe2/utils/cpuid.h", "caffe2/utils/fixed_divisor.h", ] + glob([ "caffe2/utils/threadpool/*.h", ]), copts = CAFFE2_COPTS, visibility = ["//visibility:public"], deps = [ ":caffe2_core_macros", "//c10", "//caffe2/proto:caffe2_pb", ], ) py_binary( name = "gen_op", srcs = ["caffe2/contrib/aten/gen_op.py"], deps = ["//torchgen"], ) genrule( name = "generated_caffe2_aten_op_headers", srcs = [ "caffe2/contrib/aten/aten_op_template.h", "aten/src/ATen/Declarations.yaml", ], outs = ["caffe2/caffe2/contrib/aten/gen_aten_op.h"], cmd = """ $(location :gen_op) \ --output_prefix gen_ \ --install_dir $(@D) \ --aten_root `dirname $(location aten/src/ATen/Declarations.yaml)`/../.. \ --template_dir `dirname $(location caffe2/contrib/aten/aten_op_template.h)` \ --yaml_dir `dirname $(location aten/src/ATen/Declarations.yaml)`""", tools = [":gen_op"], ) cc_library( name = "caffe2_headers", hdrs = glob( [ "caffe2/contrib/aten/*.h", "caffe2/contrib/gloo/*.h", "caffe2/core/*.h", "caffe2/core/nomnigraph/include/nomnigraph/Converters/*.h", "caffe2/core/nomnigraph/include/nomnigraph/Generated/*.h", "caffe2/core/nomnigraph/include/nomnigraph/Graph/*.h", "caffe2/core/nomnigraph/include/nomnigraph/Representations/*.h", "caffe2/core/nomnigraph/include/nomnigraph/Support/*.h", "caffe2/core/nomnigraph/include/nomnigraph/Transformations/*.h", "caffe2/core/nomnigraph/tests/*.h", "caffe2/db/*.h", "caffe2/distributed/*.h", "caffe2/ideep/*.h", "caffe2/ideep/operators/*.h", "caffe2/ideep/operators/quantization/*.h", "caffe2/ideep/utils/*.h", "caffe2/onnx/*.h", "caffe2/operators/*.h", "caffe2/operators/rnn/*.h", "caffe2/opt/*.h", "caffe2/perfkernels/*.h", "caffe2/predictor/*.h", "caffe2/predictor/emulator/*.h", "caffe2/quantization/server/*.h", "caffe2/queue/*.h", "caffe2/serialize/*.h", "caffe2/sgd/*.h", "caffe2/share/contrib/depthwise/*.h", "caffe2/transforms/*.h", "caffe2/utils/*.h", "caffe2/utils/math/*.h", "caffe2/utils/threadpool/*.h", "modules/**/*.h", ], exclude = [ "caffe2/core/macros.h", ], ) + if_cuda(glob([ "caffe2/**/*.cuh", "caffe2/image/*.h", ])) + [":generated_caffe2_aten_op_headers"], copts = CAFFE2_COPTS, includes = [ "caffe2/contrib/aten", "caffe2/core/nomnigraph/include", ], visibility = ["//visibility:public"], deps = [ ":caffe2_core_macros", ":caffe2_for_aten_headers", "//caffe2/proto:caffe2_pb", "//caffe2/proto:cc_proto", ], ) cc_library( name = "caffe2_dnnlowp_avx2_ops", srcs = [ "caffe2/quantization/server/elementwise_sum_dnnlowp_op_avx2.cc", "caffe2/quantization/server/fully_connected_fake_lowp_op_avx2.cc", "caffe2/quantization/server/group_norm_dnnlowp_op_avx2.cc", "caffe2/quantization/server/norm_minimization_avx2.cc", "caffe2/quantization/server/pool_dnnlowp_op_avx2.cc", "caffe2/quantization/server/relu_dnnlowp_op_avx2.cc", "caffe2/quantization/server/spatial_batch_norm_dnnlowp_op_avx2.cc", "caffe2/quantization/server/transpose.cc", ], copts = CAFFE2_COPTS + [ "-mf16c", "-mavx2", "-mfma", "-mxsave", ], visibility = ["//visibility:public"], deps = [ ":caffe2_headers", "@fbgemm", ], alwayslink = True, ) cc_library( name = "caffe2", srcs = [ "caffe2/db/create_db_op.cc", "caffe2/db/protodb.cc", "caffe2/share/contrib/depthwise/depthwise3x3_conv_op.cc", ":caffe2_contrib_srcs", ":caffe2_core_srcs", ":caffe2_distributed_srcs", ":caffe2_ideep_srcs", ":caffe2_onnx_srcs", ":caffe2_operators_srcs", ":caffe2_opt_srcs", ":caffe2_perfkernels_srcs", ":caffe2_predictor_srcs", ":caffe2_quantization_srcs", ":caffe2_queue_srcs", ":caffe2_serialize_srcs", ":caffe2_sgd_srcs", ":caffe2_transforms_srcs", ":caffe2_utils_srcs", ], copts = CAFFE2_COPTS + ["-mf16c"], linkstatic = 1, visibility = ["//visibility:public"], deps = [ ":caffe2_core_macros", ":caffe2_dnnlowp_avx2_ops", ":caffe2_headers", ":caffe2_perfkernels_avx", ":caffe2_perfkernels_avx2", ":caffe2_perfkernels_avx512", "//caffe2/proto:caffe2_pb", "//caffe2/proto:cc_proto", "//third_party/miniz-2.1.0:miniz", "@com_google_protobuf//:protobuf", "@eigen", "@fbgemm//:fbgemm_src_headers", "@fmt", "@foxi", "@gloo", "@onnx", ] + if_cuda( [ ":caffe2_cuda_cpp", ":aten_cuda", "@tensorpipe//:tensorpipe_cuda", ], [ ":aten", "@tensorpipe//:tensorpipe_cpu", ], ), alwayslink = True, ) cc_library( name = "caffe2_cuda_cpp", srcs = [":caffe2_cuda_cpp_srcs"], copts = CAFFE2_COPTS, visibility = ["//visibility:public"], deps = [ ":caffe2_cuda", ":caffe2_headers", ], alwayslink = True, ) cu_library( name = "caffe2_cuda", # one may think that `quantization_gpu.cu` could be a separate kernel, # however that leads to de-registration problem that's described in # https://github.com/pytorch/pytorch/issues/79236 # To solve it we add it into the `caffe2_cuda`, # this is also aligned with the CMake build. srcs = [":caffe2_cu_srcs"] + ["torch/csrc/distributed/c10d/quantization/quantization_gpu.cu"], copts = CAFFE2_COPTS + torch_cuda_half_options, visibility = ["//visibility:public"], deps = [ ":aten", ":caffe2_headers", "@cuda//:cublas", "@cuda//:curand", "@cudnn", "@eigen", "@gloo", "@tensorpipe//:tensorpipe_cuda", ], alwayslink = True, ) PERF_COPTS = [ "-DTH_HAVE_THREAD", "-DHAVE_AVX_CPU_DEFINITION", "-DHAVE_AVX2_CPU_DEFINITION", "-DENABLE_ALIAS=1", "-DHAVE_MALLOC_USABLE_SIZE=1", "-DHAVE_MMAP=1", "-DHAVE_SHM_OPEN=1", "-DHAVE_SHM_UNLINK=1", "-DSLEEF_STATIC_LIBS=1", "-DTH_BALS_MKL", "-D_FILE_OFFSET_BITS=64", "-DUSE_FBGEMM", "-fvisibility-inlines-hidden", "-Wunused-parameter", "-fno-math-errno", "-fno-trapping-math", "-mf16c", ] PERF_HEADERS = glob([ "caffe2/perfkernels/*.h", "caffe2/core/*.h", ]) cc_library( name = "caffe2_perfkernels_avx", srcs = glob([ "caffe2/perfkernels/*_avx.cc", ]), hdrs = PERF_HEADERS, copts = PERF_COPTS + [ "-mavx", ], visibility = ["//visibility:public"], deps = [ ":caffe2_headers", "//c10", ], alwayslink = True, ) cc_library( name = "caffe2_perfkernels_avx2", srcs = glob([ "caffe2/perfkernels/*_avx2.cc", ]), hdrs = PERF_HEADERS, copts = PERF_COPTS + [ "-mavx2", "-mfma", "-mavx", ], visibility = ["//visibility:public"], deps = [ ":caffe2_headers", "//c10", ], alwayslink = True, ) cc_library( name = "caffe2_perfkernels_avx512", srcs = [ "caffe2/perfkernels/common_avx512.cc", ], hdrs = PERF_HEADERS, copts = PERF_COPTS + [ "-mavx512f", "-mavx512dq", "-mavx512vl", "-mavx2", "-mfma", "-mavx", ], visibility = ["//visibility:public"], deps = [ ":caffe2_headers", "//c10", ], alwayslink = True, ) # torch torch_cuda_headers = glob(["torch/csrc/cuda/*.h"]) cc_library( name = "torch_headers", hdrs = if_cuda( torch_cuda_headers, ) + glob( [ "torch/*.h", "torch/csrc/**/*.h", "torch/csrc/distributed/c10d/*.hpp", "torch/lib/libshm/*.h", ], exclude = [ "torch/csrc/*/generated/*.h", ] + torch_cuda_headers, ) + GENERATED_AUTOGRAD_CPP + [":version_h"], includes = [ "third_party/kineto/libkineto/include", "torch/csrc", "torch/csrc/api/include", "torch/csrc/distributed", "torch/lib", "torch/lib/libshm", ], visibility = ["//visibility:public"], deps = [ ":aten_headers", ":caffe2_headers", "//c10", "@com_github_google_flatbuffers//:flatbuffers", "@local_config_python//:python_headers", "@onnx", ], alwayslink = True, ) TORCH_COPTS = COMMON_COPTS + [ "-Dtorch_EXPORTS", "-DHAVE_AVX_CPU_DEFINITION", "-DHAVE_AVX2_CPU_DEFINITION", "-DCAFFE2_USE_GLOO", "-fvisibility-inlines-hidden", "-fno-math-errno ", "-fno-trapping-math", "-Wno-error=unused-function", ] torch_sources = { k: "" for k in ( libtorch_core_sources + libtorch_distributed_sources + torch_cpp_srcs + libtorch_extra_sources + jit_core_sources + lazy_tensor_ts_sources + GENERATED_AUTOGRAD_CPP ) }.keys() cc_library( name = "torch", srcs = if_cuda(glob( libtorch_cuda_sources, exclude = [ "torch/csrc/cuda/python_nccl.cpp", "torch/csrc/cuda/nccl.cpp", "torch/csrc/distributed/c10d/quantization/quantization_gpu.cu", ], )) + torch_sources, copts = TORCH_COPTS, defines = [ "CAFFE2_NIGHTLY_VERSION=20200115", ], visibility = ["//visibility:public"], deps = [ ":caffe2", ":torch_headers", "//caffe2/proto:torch_cc_proto", "@kineto", ] + if_cuda([ "@cuda//:nvToolsExt", "@cutlass", ]), alwayslink = True, ) cc_library( name = "shm", srcs = glob(["torch/lib/libshm/*.cpp"]), deps = [ ":torch", ], ) cc_library( name = "libtorch_headers", hdrs = glob([ "**/*.h", "**/*.cuh", ]) + [ # We need the filegroup here because the raw list causes Bazel # to see duplicate files. It knows how to deduplicate with the # filegroup. ":cpp_generated_code", ], includes = [ "torch/csrc/api/include", "torch/csrc/distributed", "torch/lib", "torch/lib/libshm", ], visibility = ["//visibility:public"], deps = [ ":torch_headers", ], ) cc_library( name = "torch_python", srcs = libtorch_python_core_sources + if_cuda(libtorch_python_cuda_sources) + if_cuda(libtorch_python_distributed_sources) + GENERATED_AUTOGRAD_PYTHON, hdrs = glob([ "torch/csrc/generic/*.cpp", ]), copts = COMMON_COPTS + if_cuda(["-DUSE_CUDA=1"]), deps = [ ":torch", ":shm", "@pybind11", ], ) pybind_extension( name = "torch/_C", srcs = ["torch/csrc/stub.c"], deps = [ ":torch_python", ":aten_nvrtc", ], ) cc_library( name = "functorch", hdrs = glob([ "functorch/csrc/dim/*.h", ]), srcs = glob([ "functorch/csrc/dim/*.cpp", ]), deps = [ ":aten_nvrtc", ":torch_python", "@pybind11", ], ) pybind_extension( name = "functorch/_C", copts=[ "-DTORCH_EXTENSION_NAME=_C" ], srcs = [ "functorch/csrc/init_dim_only.cpp", ], deps = [ ":functorch", ":torch_python", ":aten_nvrtc", ], ) cc_binary( name = "torch/bin/torch_shm_manager", srcs = [ "torch/lib/libshm/manager.cpp", ], deps = [ ":shm", ], linkstatic = False, ) template_rule( name = "gen_version_py", src = ":torch/version.py.tpl", out = "torch/version.py", substitutions = if_cuda({ # Set default to 11.2. Otherwise Torchvision complains about incompatibility. "{{CUDA_VERSION}}": "11.2", "{{VERSION}}": "2.0.0", }, { "{{CUDA_VERSION}}": "None", "{{VERSION}}": "2.0.0", }), ) py_library( name = "pytorch_py", visibility = ["//visibility:public"], srcs = glob(["torch/**/*.py"], exclude = ["torch/version.py"]) + [":torch/version.py"] + glob(["functorch/**/*.py"]), deps = [ rules.requirement("future"), rules.requirement("numpy"), rules.requirement("pyyaml"), rules.requirement("requests"), rules.requirement("setuptools"), rules.requirement("six"), rules.requirement("sympy"), rules.requirement("typing_extensions"), "//torchgen", ], data = [ ":torch/_C.so", ":functorch/_C.so", ":torch/bin/torch_shm_manager", ], ) # cpp api tests cc_library( name = "test_support", testonly = True, srcs = [ "test/cpp/api/support.cpp", ], hdrs = [ "test/cpp/api/init_baseline.h", "test/cpp/api/optim_baseline.h", "test/cpp/api/support.h", "test/cpp/common/support.h", ], deps = [ ":torch", "@com_google_googletest//:gtest_main", ], ) # Torch integration tests rely on a labeled data set from the MNIST database. # http://yann.lecun.com/exdb/mnist/ cpp_api_tests = glob( ["test/cpp/api/*.cpp"], exclude = [ "test/cpp/api/imethod.cpp", "test/cpp/api/integration.cpp", ], ) cc_test( name = "integration_test", size = "medium", srcs = ["test/cpp/api/integration.cpp"], data = [ ":download_mnist", ], tags = [ "gpu-required", ], deps = [ ":test_support", "@com_google_googletest//:gtest_main", ], ) [ cc_test( name = paths.split_extension(paths.basename(filename))[0].replace("-", "_") + "_test", size = "medium", srcs = [filename], deps = [ ":test_support", "@com_google_googletest//:gtest_main", ], ) for filename in cpp_api_tests ] test_suite( name = "api_tests", tests = [ "any_test", "autograd_test", "dataloader_test", "enum_test", "expanding_array_test", "functional_test", "init_test", "integration_test", "jit_test", "memory_test", "misc_test", "module_test", "modulelist_test", "modules_test", "nn_utils_test", "optim_test", "ordered_dict_test", "rnn_test", "sequential_test", "serialize_test", "static_test", "tensor_options_test", "tensor_test", "torch_include_test", ], ) # dist autograd tests cc_test( name = "torch_dist_autograd_test", size = "small", srcs = ["test/cpp/dist_autograd/test_dist_autograd.cpp"], tags = [ "exclusive", "gpu-required", ], deps = [ ":torch", "@com_google_googletest//:gtest_main", ], ) # jit tests # Because these individual unit tests require custom registering, # it is easier to mimic the cmake build by globing together a single test. cc_test( name = "jit_tests", size = "small", srcs = glob( [ "test/cpp/jit/*.cpp", "test/cpp/jit/*.h", "test/cpp/tensorexpr/*.cpp", "test/cpp/tensorexpr/*.h", ], exclude = [ # skip this since is not found in OSS build "test/cpp/jit/test_exception.cpp", ], ), linkstatic = True, tags = [ "exclusive", "gpu-required", ], deps = [ ":torch", "@com_google_googletest//:gtest_main", ], ) cc_test( name = "lazy_tests", size = "small", srcs = glob( [ "test/cpp/lazy/*.cpp", "test/cpp/lazy/*.h", ], exclude = [ # skip these since they depend on generated LazyIr.h which isn't available in bazel yet "test/cpp/lazy/test_ir.cpp", "test/cpp/lazy/test_lazy_ops.cpp", "test/cpp/lazy/test_lazy_ops_util.cpp", ], ), linkstatic = True, tags = [ "exclusive", ], deps = [ ":torch", "@com_google_googletest//:gtest_main", ], ) # python api tests py_test( name = "test_bazel", srcs = ["test/_test_bazel.py"], main = "test/_test_bazel.py", deps = [":pytorch_py"], ) # all tests test_suite( name = "all_tests", tests = [ "api_tests", "jit_tests", "torch_dist_autograd_test", "//c10/test:tests", ], ) # An internal genrule that we are converging with refers to these file # as if they are from this package, so we alias them for # compatibility. [ alias( name = paths.basename(path), actual = path, ) for path in [ "aten/src/ATen/templates/DispatchKeyNativeFunctions.cpp", "aten/src/ATen/templates/DispatchKeyNativeFunctions.h", "aten/src/ATen/templates/LazyIr.h", "aten/src/ATen/templates/LazyNonNativeIr.h", "aten/src/ATen/templates/RegisterDispatchKey.cpp", "aten/src/ATen/templates/RegisterDispatchDefinitions.ini", "aten/src/ATen/native/native_functions.yaml", "aten/src/ATen/native/tags.yaml", "aten/src/ATen/native/ts_native_functions.yaml", "torch/csrc/lazy/core/shape_inference.h", "torch/csrc/lazy/ts_backend/ts_native_functions.cpp", ] ] genrule( name = "download_mnist", srcs = ["//:tools/download_mnist.py"], outs = [ "mnist/train-images-idx3-ubyte", "mnist/train-labels-idx1-ubyte", "mnist/t10k-images-idx3-ubyte", "mnist/t10k-labels-idx1-ubyte", ], cmd = "python3 tools/download_mnist.py -d $(RULEDIR)/mnist", )