Skip to content

Fixes bug in one specialized HIP instantiation of the warp-per-row kernel #134

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
aryaman-gupta wants to merge 2 commits into
base: main_12052025_upstream
Choose a base branch
from
+25 −1
Open

Fixes bug in one specialized HIP instantiation of the warp-per-row kernel #134

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
915af9d
split_embeddings_common: fixes pointer arithmetic error in HIP load f…
aryaman-gupta Dec 5, 2025
3a983a0
backward_adagrad_test: adds test for specific config: fp16, D=320
aryaman-gupta Dec 9, 2025
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
2 changes: 1 addition & 1 deletion fbgemm_gpu/include/fbgemm_gpu/rocm/split_embeddings_common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -226,7 +226,7 @@ struct load_row_per_warp<half, 320, index_t> {
llvm_amdgcn_raw_buffer_load_fp16x2(
emb_res, (lane_id + 64) * sizeof(half2));
emb_data[4] = llvm_amdgcn_raw_buffer_load_fp16(
emb_res, (lane_id + 128) * sizeof(half));
emb_res, (lane_id + 256) * sizeof(half));
}
};

Expand Down
24 changes: 24 additions & 0 deletions fbgemm_gpu/test/tbe/training/backward_adagrad_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@
from .backward_adagrad_common import (
additional_decorators,
adjust_mixed_B_st,
CacheAlgorithm,
common_settings,
common_strategy,
execute_backward_adagrad,
Expand Down Expand Up @@ -221,6 +222,29 @@ def test_backward_adagrad_fp16_pmSUM_with_max_norm( # noqa C901
**kwargs,
)

@unittest.skipIf(*gpu_unavailable)
def test_backward_adagrad_fp16_pmSUM_D320(self) -> None:
execute_backward_adagrad(
T=2,
# using D=80 since the test harness multiplies D by 4, so 80*4=320
D=80,
B=16,
log_E=4,
L=4,
D_gradcheck=1,
weights_precision=SparseType.FP16,
stochastic_rounding=False,
weighted=False,
row_wise=True,
mixed=False,
mixed_B=False,
use_cache=False,
cache_algorithm=CacheAlgorithm.LRU,
pooling_mode=PoolingMode.SUM,
use_cpu=False,
output_dtype=SparseType.FP16,
)


if __name__ == "__main__":
unittest.main()