FM / c10d

/usr/local/lib64/python3.6/site-packages/torch/include/c10d/

/usr/local/lib64/python3.6/site-packages/torch/include/c10d

Name	Size	Mode	Actions
comm.hpp	4184	0644	edit dl rm
default_comm_hooks.hpp	1479	0644	edit dl rm
FileStore.hpp	1403	0644	edit dl rm
frontend.hpp	9252	0644	edit dl rm
frontend_cuda.hpp	142	0644	edit dl rm
GlooDeviceFactory.hpp	771	0644	edit dl rm
HashStore.hpp	1131	0644	edit dl rm
logger.hpp	3981	0644	edit dl rm
NCCLUtils.hpp	8046	0644	edit dl rm
ParamCommsUtils.hpp	1814	0644	edit dl rm
PrefixStore.hpp	1404	0644	edit dl rm
ProcessGroup.hpp	12435	0644	edit dl rm
ProcessGroupGloo.hpp	13395	0644	edit dl rm
ProcessGroupMPI.hpp	8824	0644	edit dl rm
ProcessGroupNCCL.hpp	22041	0644	edit dl rm
ProcessGroupRoundRobin.hpp	3992	0644	edit dl rm
ProcessGroupWrapper.hpp	4971	0644	edit dl rm
reducer.hpp	26121	0644	edit dl rm
sequence_num.hpp	1683	0644	edit dl rm
Store.hpp	2589	0644	edit dl rm
TCPStore.hpp	3213	0644	edit dl rm
Types.hpp	1370	0644	edit dl rm
UnixSockUtils.hpp	2840	0644	edit dl rm
Utils.hpp	21601	0644	edit dl rm
WinSockUtils.hpp	2701	0644	edit dl rm

Edit: /usr/local/lib64/python3.6/site-packages/torch/include/c10d/ParamCommsUtils.hpp (1814B)

#pragma once

#include <string>
#include <vector>
#include <c10/macros/Macros.h>
#include <c10/util/ThreadLocalDebugInfo.h>
#include <ATen/record_function.h>
#include <ATen/core/ivalue.h>

namespace torch {

extern TORCH_API const std::string kParamCommsCallName;

class TORCH_API ParamCommsDebugInfo
    : public c10::DebugInfoBase {

public:
  ParamCommsDebugInfo() = default;
  ParamCommsDebugInfo(
    int rank,
    std::string&& colName,
    int inSize,
    int outSize,
    at::ScalarType dType,
    std::vector<int64_t> inSplitSizes,
    std::vector<int64_t> outSplitSizes);

~ParamCommsDebugInfo() override = default;

int getRank() const {
    return rank_;
  }

const std::string getColumnName() const {
    return columnName_;
  }

int getInMessageSize() const {
    return inMessageSize_;
  }

int getOutMessageSize() const {
    return outMessageSize_;
  }

at::ScalarType getDType() const {
    return dType_;
  }

const std::vector<int64_t>& getInputSplitSizes() const {
    return inputSplitSizes_;
  }

const std::vector<int64_t>& getOutputSplitSizes() const {
    return outputSplitSizes_;
  }

private:
  int rank_{};
  std::string columnName_;
  int inMessageSize_{};
  int outMessageSize_{};
  at::ScalarType dType_ = at::kByte;
  std::vector<int64_t> inputSplitSizes_;
  std::vector<int64_t> outputSplitSizes_;
};

#define RECORD_PARAM_COMMS(rank, colName, inSize, outSize, dType, inSplitSizes, outSplitSizes) \
  auto paramCommsInfo = std::make_shared<torch::ParamCommsDebugInfo>( \
    rank, \
    colName, \
    inSize, \
    outSize, \
    dType, \
    inSplitSizes, \
    outSplitSizes); \
  c10::DebugInfoGuard g(c10::DebugInfoKind::PARAM_COMMS_INFO, paramCommsInfo); \
  RECORD_FUNCTION(torch::kParamCommsCallName, std::vector<c10::IValue>());

} // namespace torch

#pragma once #include #include #include #include #include #include namespace torch { extern TORCH_API const std::string kParamCommsCallName; class TORCH_API ParamCommsDebugInfo : public c10::DebugInfoBase { public: ParamCommsDebugInfo() = default; ParamCommsDebugInfo( int rank, std::string&& colName, int inSize, int outSize, at::ScalarType dType, std::vector inSplitSizes, std::vector outSplitSizes); ~ParamCommsDebugInfo() override = default; int getRank() const { return rank_; } const std::string getColumnName() const { return columnName_; } int getInMessageSize() const { return inMessageSize_; } int getOutMessageSize() const { return outMessageSize_; } at::ScalarType getDType() const { return dType_; } const std::vector& getInputSplitSizes() const { return inputSplitSizes_; } const std::vector& getOutputSplitSizes() const { return outputSplitSizes_; } private: int rank_{}; std::string columnName_; int inMessageSize_{}; int outMessageSize_{}; at::ScalarType dType_ = at::kByte; std::vector inputSplitSizes_; std::vector outputSplitSizes_; }; #define RECORD_PARAM_COMMS(rank, colName, inSize, outSize, dType, inSplitSizes, outSplitSizes) \ auto paramCommsInfo = std::make_shared( \ rank, \ colName, \ inSize, \ outSize, \ dType, \ inSplitSizes, \ outSplitSizes); \ c10::DebugInfoGuard g(c10::DebugInfoKind::PARAM_COMMS_INFO, paramCommsInfo); \ RECORD_FUNCTION(torch::kParamCommsCallName, std::vector()); } // namespace torch