Skip to content

issue/1519 - fix Ascend paged attention block index name collision #1520

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
ETOwang wants to merge 1 commit into
base: main
Choose a base branch
from
+7 −7
Open

issue/1519 - fix Ascend paged attention block index name collision #1520

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
6 changes: 3 additions & 3 deletions src/infiniop/ops/paged_attention/ascend/paged_attention_ascend_kernel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,15 +121,15 @@ class PagedAttentionKernel {
float max_score = -3.4028234663852886e38f;
float sum_exp = 0.0f;

for (size_t block_idx = 0; block_idx < _max_num_blocks_per_seq; ++block_idx) {
size_t token_start = block_idx * _page_block_size;
for (size_t page_block_idx = 0; page_block_idx < _max_num_blocks_per_seq; ++page_block_idx) {
size_t token_start = page_block_idx * _page_block_size;
if (token_start >= seq_len) {
break;
}

const int64_t physical_block = loadIndex(
_block_tables_gm,
static_cast<ptrdiff_t>(seq_idx) * _block_table_batch_stride + static_cast<ptrdiff_t>(block_idx));
static_cast<ptrdiff_t>(seq_idx) * _block_table_batch_stride + static_cast<ptrdiff_t>(page_block_idx));

size_t token_end = token_start + _page_block_size;
if (token_end > seq_len) {
Expand Down
8 changes: 4 additions & 4 deletions src/infiniop/ops/paged_attention_prefill/ascend/paged_attention_prefill_ascend_kernel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,22 +197,22 @@ class PagedAttentionPrefillKernel {
}

__aicore__ inline ptrdiff_t keyBase(size_t seq_idx, size_t kv_head_idx, size_t token_idx) {
const size_t block_idx = token_idx / _page_block_size;
const size_t page_block_idx = token_idx / _page_block_size;
const size_t token_offset = token_idx % _page_block_size;
const int64_t physical_block = loadPrefillIndex(
_block_tables_gm,
static_cast<ptrdiff_t>(seq_idx) * _block_table_batch_stride + static_cast<ptrdiff_t>(block_idx));
static_cast<ptrdiff_t>(seq_idx) * _block_table_batch_stride + static_cast<ptrdiff_t>(page_block_idx));
return static_cast<ptrdiff_t>(physical_block) * _k_batch_stride
+ static_cast<ptrdiff_t>(kv_head_idx) * _k_head_stride
+ static_cast<ptrdiff_t>(token_offset) * _k_row_stride;
}

__aicore__ inline ptrdiff_t valueBase(size_t seq_idx, size_t kv_head_idx, size_t token_idx) {
const size_t block_idx = token_idx / _page_block_size;
const size_t page_block_idx = token_idx / _page_block_size;
const size_t token_offset = token_idx % _page_block_size;
const int64_t physical_block = loadPrefillIndex(
_block_tables_gm,
static_cast<ptrdiff_t>(seq_idx) * _block_table_batch_stride + static_cast<ptrdiff_t>(block_idx));
static_cast<ptrdiff_t>(seq_idx) * _block_table_batch_stride + static_cast<ptrdiff_t>(page_block_idx));
return static_cast<ptrdiff_t>(physical_block) * _v_batch_stride
+ static_cast<ptrdiff_t>(kv_head_idx) * _v_head_stride
+ static_cast<ptrdiff_t>(token_offset) * _v_row_stride;
Expand Down