File: mpi_ops.cc

package info (click to toggle)
pytorch 1.13.1%2Bdfsg-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 139,252 kB
  • sloc: cpp: 1,100,274; python: 706,454; ansic: 83,052; asm: 7,618; java: 3,273; sh: 2,841; javascript: 612; makefile: 323; xml: 269; ruby: 185; yacc: 144; objc: 68; lex: 44
file content (24 lines) | stat: -rw-r--r-- 998 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
#include "caffe2/mpi/mpi_ops.h"

namespace caffe2 {

OPERATOR_SCHEMA(MPICreateCommonWorld).NumInputs(0).NumOutputs(1);
OPERATOR_SCHEMA(MPIBroadcast)
    .NumInputs(2)
    .NumOutputs(1)
    .EnforceInplace({{1, 0}});
OPERATOR_SCHEMA(MPIReduce).NumInputs(2).NumOutputs(1);
OPERATOR_SCHEMA(MPIAllgather).NumInputs(2).NumOutputs(1);
OPERATOR_SCHEMA(MPIAllreduce).NumInputs(2).NumOutputs(1).AllowInplace({{1, 0}});
OPERATOR_SCHEMA(MPISendTensor);
OPERATOR_SCHEMA(MPIReceiveTensor);

REGISTER_CPU_OPERATOR(MPICreateCommonWorld, MPICreateCommonWorldOp<CPUContext>);
REGISTER_CPU_OPERATOR(MPIBroadcast, MPIBroadcastOp<CPUContext>);
REGISTER_CPU_OPERATOR(MPIReduce, MPIReduceOp<float, CPUContext>);
REGISTER_CPU_OPERATOR(MPIAllgather, MPIAllgatherOp<float, CPUContext>);
REGISTER_CPU_OPERATOR(MPIAllreduce, MPIAllreduceOp<float, CPUContext>);
REGISTER_CPU_OPERATOR(MPISendTensor, MPISendTensorOp<CPUContext>);
REGISTER_CPU_OPERATOR(MPIReceiveTensor, MPIReceiveTensorOp<CPUContext>);

} // namespace caffe2