Skip to content

fix a datacopy bug in 910B3 #868

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
leideng wants to merge 4 commits into
base: develop
Choose a base branch
from
Open

fix a datacopy bug in 910B3 #868

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7b54b1f
fix a datacopy bug in 910B3
leideng Mar 25, 2026
f3f4c58
pass cpp linter
leideng Mar 25, 2026
87a95d6
disable inner sort
leideng Mar 26, 2026
de29532
Merge branch 'develop' into 910b3-ops-bug-fix
Infinite666 Mar 26, 2026
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
21 changes: 12 additions & 9 deletions ...parse/gsa_on_device/csrc/ascend/hamming_dist_top_k/op_kernel/hamming_dist_top_k_split_s.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1011,13 +1011,14 @@ class HammingDistTopKSplitSKernel {
if ASCEND_IS_AIC { return; }

// topK小于32，不使用内部API排序
if (curKScalar < 32) { useInnerSort = false; }
// (TODO) do not enable inner sort for now as 910B3 has bug
// if (curKScalar < 32) { useInnerSort = false; }

if (useInnerSort) {
CustomSort(topKIndexOutTensor, curKScalar);
SelectBlockTableFromTopK(curBatchIdx, curKScalar, outGmOffset);
return;
}
// if (useInnerSort) {
// CustomSort(topKIndexOutTensor, curKScalar);
// SelectBlockTableFromTopK(curBatchIdx, curKScalar, outGmOffset);
// return;
// }

// 将 TopK 的“chunk索引”映射成 block_id，并写回 GM(indices)
WriteBlockTableFromTopK(curBatchIdx, topKIndexOutTensor, curKScalar, outGmOffset);
Expand Down Expand Up @@ -1509,9 +1510,11 @@ class HammingDistTopKSplitSKernel {

LocalTensor<int32_t> blockIdUb = topKIndexInnerInQueue_.AllocTensor<int32_t>();
LocalTensor<int32_t> tableBlockTensor = tableBlockBuf_.template Get<int32_t>();
DataCopyParams copyParams{1, static_cast<uint16_t>(param_.blockCount * sizeof(int32_t)), 0,
0};
DataCopy(tableBlockTensor, keyBlockTableGm_[curBatchIdx * param_.blockCount], copyParams);
DataCopyExtParams copyParams{1, static_cast<uint16_t>(param_.blockCount * sizeof(int32_t)),
0, 0, 0};
DataCopyPadExtParams<int32_t> copyPadParams{false, 0, 0, 0};
DataCopyPad(tableBlockTensor, keyBlockTableGm_[curBatchIdx * param_.blockCount], copyParams,
copyPadParams);

::AscendC::WriteBlockTableFromTopK(curBatchIdx, topKIndexUb, blockIdUb, curKScalar,
outGmOffset, tableBlockTensor, indicesGm_,
Expand Down
Loading