/usr/local/lib64/python3.6/site-packages/torch/include/THC
NameSizeModeActions
generic/-0755rm
THC.h3280644editdlrm
THCAllocator.h3700644editdlrm
THCAsmUtils.cuh34440644editdlrm
THCAtomics.cuh130940644editdlrm
THCCachingHostAllocator.h12540644editdlrm
THCDeviceTensor-inl.cuh115150644editdlrm
THCDeviceTensor.cuh161600644editdlrm
THCDeviceTensorUtils-inl.cuh44970644editdlrm
THCDeviceTensorUtils.cuh27200644editdlrm
THCDeviceUtils.cuh9420644editdlrm
THCGeneral.h27410644editdlrm
THCGeneral.hpp7430644editdlrm
THCGenerateAllTypes.h9580644editdlrm
THCGenerateBFloat16Type.h5150644editdlrm
THCGenerateBoolType.h4490644editdlrm
THCGenerateByteType.h4190644editdlrm
THCGenerateCharType.h4180644editdlrm
THCGenerateComplexDoubleType.h5330644editdlrm
THCGenerateComplexFloatType.h5260644editdlrm
THCGenerateComplexTypes.h2980644editdlrm
THCGenerateDoubleType.h4640644editdlrm
THCGenerateFloatType.h5500644editdlrm
THCGenerateFloatTypes.h7790644editdlrm
THCGenerateHalfType.h4810644editdlrm
THCGenerateIntType.h4140644editdlrm
THCGenerateLongType.h4190644editdlrm
THCGenerateShortType.h4240644editdlrm
THCIntegerDivider.cuh40950644editdlrm
THCNumerics.cuh198130644editdlrm
THCScanUtils.cuh47890644editdlrm
THCSleep.h2320644editdlrm
THCStorage.h4870644editdlrm
THCStorage.hpp8460644editdlrm
THCStorageCopy.h4660644editdlrm
THCTensor.h6230644editdlrm
THCTensor.hpp10700644editdlrm
THCTensorCopy.h4670644editdlrm
THCTensorCopy.hpp6000644editdlrm
THCTensorMathReduce.cuh6640644editdlrm
THCThrustAllocator.cuh6180644editdlrm
Edit: /usr/local/lib64/python3.6/site-packages/torch/include/THC/THCCachingHostAllocator.h (1254B)
#ifndef THC_CACHING_HOST_ALLOCATOR_INC #define THC_CACHING_HOST_ALLOCATOR_INC #include #include // // A caching allocator for CUDA host allocations (pinned memory). // // This provides a drop-in replacement for THCudaHostAllocator, which re-uses // freed pinned (page-locked) memory allocations. This avoids device // synchronizations due to cudaFreeHost calls. // // To ensure correct behavior, THCCachingHostAllocator_recordEvent must be // called anytime a pointer from this allocator is used in a cudaMemcpyAsync // call between host and device. We implement this for storages and tensors in // copy_from_cpu_async_ and copy_to_cpu_async_. // // Note that this allocator does not split larger allocations into smaller // blocks, unlike the caching device allocator. // TORCH_CUDA_CPP_API c10::Allocator* getTHCCachingHostAllocator(void); // Records an event in the specified stream. The allocation 'ptr' will not be // re-used until the event has occurred. TORCH_CUDA_CPP_API cudaError_t THCCachingHostAllocator_recordEvent(void* ptr, at::cuda::CUDAStream stream); // Releases cached pinned memory allocations via cudaHostFree TORCH_CUDA_CPP_API void THCCachingHostAllocator_emptyCache(void); #endif