Skip to content

[MLA] fake non persistent fix dp causal #1677

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
valarLip merged 3 commits into from
Dec 20, 2025
Merged

[MLA] fake non persistent fix dp causal #1677

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3746bf3
fix dp causal
Zzz9990 Dec 18, 2025
87e7716
Merge branch 'main' into mla_fix_fake_non_ps
Zzz9990 Dec 19, 2025
021a560
Merge branch 'main' into mla_fix_fake_non_ps
Zzz9990 Dec 19, 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
15 changes: 6 additions & 9 deletions csrc/kernels/mla/metadata/v1_0_device.cuh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -47,20 +47,19 @@ void kn_get_mla_metadata_v1_0(MlaMetadataV1KernelParameter params)
const int32_t bid_ori = bid / params.qk_batch_ratio;

const int32_t kv_begin = params.p_seqlens_kv_indptr[bid_ori];
const int32_t kv_end = params.p_seqlens_kv_indptr[bid_ori + 1];

int32_t kv_tail = [&](){
if constexpr(DP_MODE)
{
// max(*, 0) for cuda graph capture: kvlen < mtp+1
return max(bid % params.ori_seqlen_qo - params.ori_seqlen_qo + 1, 0);
return bid % params.ori_seqlen_qo - params.ori_seqlen_qo + 1;
}
else
{
return 0;
}
}();
const int32_t seqlen_kv = kv_end - kv_begin + kv_tail;
const int32_t kv_end = max(params.p_seqlens_kv_indptr[bid_ori + 1] + kv_tail, kv_begin + 1);

const int32_t seqlen_kv = kv_end - kv_begin;

const int32_t num_blocks = integer_divide_ceil_power2(
seqlen_kv, params.kv_granularity, params.kv_granularity_log2);
Expand Down Expand Up @@ -98,19 +97,17 @@ void kn_get_mla_metadata_v1_0(MlaMetadataV1KernelParameter params)
const int32_t bid_ori = bid / params.qk_batch_ratio;

const int32_t kv_begin = p_lds_kv_seqlen[bid_ori];
int32_t kv_end = p_lds_kv_seqlen[bid_ori + 1];
int32_t kv_tail = [&](){
if constexpr(DP_MODE)
{
// max(*, 0) for cuda graph capture: kvlen < mtp+1
return max(bid % params.ori_seqlen_qo - params.ori_seqlen_qo + 1, 0);
return bid % params.ori_seqlen_qo - params.ori_seqlen_qo + 1;
}
else
{
return 0;
}
}();
kv_end += kv_tail;
const int32_t kv_end = max(p_lds_kv_seqlen[bid_ori + 1] + kv_tail, kv_begin + 1);
MlaWorkInfo work_info{};
const int32_t split_start = p_lds_shift[bid];
const int32_t split_local = p_lds_split[bid];
Expand Down