Skip to content

Add grid-stride + ROCm cap to split optimizer update kernel #6086

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

Open
q10 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add grid-stride + ROCm cap to split optimizer update kernel #6086

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 12 additions & 0 deletions fbgemm_gpu/codegen/training/optimizer/embedding_optimizer_split_kernel_template.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,10 +37,19 @@ void split_{{ optimizer }}_update_kernel(
at::PhiloxCudaState stochastic_rounding_philox_args,
{{ args.split_kernel_args | replace_pta_namespace() | join(",\n ") }}
) {
// On ROCm the launch caps the grid to stay within the HIP 2^32
// threads-per-launch limit, so we grid-stride to cover the full workload.
// On CUDA the grid is not capped and the loop body runs once per warp.
#ifdef USE_ROCM
for (auto run_id = blockIdx.x * blockDim.y + threadIdx.y;
run_id < grad_dev_indices.size(0);
run_id += blockDim.y * gridDim.x) {
#else
const auto run_id = blockIdx.x * blockDim.y + threadIdx.y;
if (run_id >= grad_dev_indices.size(0)) {
return;
}
#endif

#ifdef FBGEMM_USE_SUBWARP_SHUFFLE
const unsigned int shfl_sync_mask =
Expand Down Expand Up @@ -99,6 +108,9 @@ void split_{{ optimizer }}_update_kernel(
shfl_sync_mask,
kMaxVecsPerThread,
{{ args.split_kernel_arg_names | join(", ") }});
#ifdef USE_ROCM
} // for run_id (grid-stride loop, ROCm only)
#endif
}

{%- for use_subwarp in [True, False] %}
Expand Down
5 changes: 4 additions & 1 deletion fbgemm_gpu/codegen/training/optimizer/embedding_optimizer_split_template.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -192,7 +192,10 @@ void split_embedding_{{ optimizer }}_update(
kMaxVecsPerThread,
kThreadGroupSize,
4>),
div_round_up(grad_dev_indices.numel(), kMaxThreads / kThreadGroupSize),
utils::cuda::cap_grid_dim_x(
div_round_up(grad_dev_indices.numel(), kMaxThreads / kThreadGroupSize),
kMaxThreads,
at::cuda::getCurrentCUDAStream()),
dim3(kThreadGroupSize, kMaxThreads / kThreadGroupSize, 1),
0, // Shared memory is not needed because uint8_t is not supported
at::cuda::getCurrentCUDAStream(),
Expand Down
Loading