Skip to content

[AMP]register bf16 for communication ops #52555

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Apr 7, 2023
Merged

[AMP]register bf16 for communication ops #52555

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
8b91dbd
register bf16 for communication ops
shaojiewang Apr 5, 2023
c9c8216
formatting
shaojiewang Apr 5, 2023
1f73855
fix bfloat16 type finding compile error in c_allreduce_max_op
shaojiewang Apr 5, 2023
0d3f32b
merge remote develop
shaojiewang Apr 7, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion paddle/fluid/operators/collective/c_allreduce_max_op.cu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,11 @@ PD_REGISTER_STRUCT_KERNEL(c_allreduce_max,
ALL_LAYOUT,
ops::CAllReduceMaxCUDAKernel,
float,
#if NCCL_VERSION_CODE >= 21000
plat::bfloat16,
#endif
double,
int,
int64_t,
plat::float16) {}
plat::float16) {
}
6 changes: 5 additions & 1 deletion paddle/fluid/operators/collective/c_split_op.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,4 +123,8 @@ PD_REGISTER_STRUCT_KERNEL(c_split,
double,
int,
int64_t,
plat::float16) {}
#if NCCL_VERSION_CODE >= 21000
plat::bfloat16,
#endif
plat::float16) {
}
3 changes: 3 additions & 0 deletions paddle/fluid/operators/collective/partial_allgather_op.cu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -104,6 +104,9 @@ namespace plat = paddle::platform;

REGISTER_OP_CUDA_KERNEL(partial_allgather,
ops::PartialAllGatherOpCUDAKernel<float>,
#if NCCL_VERSION_CODE >= 21000
ops::PartialAllGatherOpCUDAKernel<plat::bfloat16>,
#endif
ops::PartialAllGatherOpCUDAKernel<double>,
ops::PartialAllGatherOpCUDAKernel<int>,
ops::PartialAllGatherOpCUDAKernel<int64_t>,
Expand Down
3 changes: 3 additions & 0 deletions paddle/fluid/operators/collective/partial_recv_op.cu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,6 +120,9 @@ namespace plat = paddle::platform;

REGISTER_OP_CUDA_KERNEL(partial_recv,
ops::PartialRecvOpCUDAKernel<float>,
#if NCCL_VERSION_CODE >= 21000
ops::PartialRecvOpCUDAKernel<plat::bfloat16>,
#endif
ops::PartialRecvOpCUDAKernel<double>,
ops::PartialRecvOpCUDAKernel<int>,
ops::PartialRecvOpCUDAKernel<int64_t>,
Expand Down
3 changes: 3 additions & 0 deletions paddle/fluid/operators/collective/partial_send_op.cu.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,6 +120,9 @@ namespace plat = paddle::platform;
REGISTER_OP_CUDA_KERNEL(partial_send,
ops::PartialSendCUDAKernel<float>,
ops::PartialSendCUDAKernel<double>,
#if NCCL_VERSION_CODE >= 21000
ops::PartialSendCUDAKernel<plat::bfloat16>,
#endif
ops::PartialSendCUDAKernel<int>,
ops::PartialSendCUDAKernel<int64_t>,
ops::PartialSendCUDAKernel<plat::float16>);