Skip to content

refactor #664

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 1 commit into from
Jul 16, 2025
Merged

refactor #664

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
23 changes: 12 additions & 11 deletions csrc/py_itfs_ck/mha_varlen_fwd_kernels.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,7 +93,7 @@ fmha_fwd_args get_ck_fmha_varlen_fwd_args(bool has_lse,
bias_ptr = alibi_slopes.data_ptr();
stride_bias = alibi_slopes.dim() == 2 ? alibi_slopes.stride(0) : 0;
}

return fmha_fwd_args{q.data_ptr(),
k.data_ptr(),
v.data_ptr(),
Expand Down Expand Up @@ -468,19 +468,11 @@ mha_varlen_fwd(at::Tensor &q, // [total_q, hq, d]
if (return_dropout_randval) {p.zero_();}
}

int num_splits = 0;
num_splits = aiter::override_num_splits_if_necessary(batch_size, num_heads, max_seqlen_q, head_size_v, 0, num_splits);
TORCH_CHECK(num_splits > 0, "num_splits should greater than 0");
TORCH_CHECK(num_splits <= 128, "num_splits greater than 128 is not supported");

auto softmax_lse_accum = torch::empty({num_heads, num_splits, total_q}, opts.dtype(at::kFloat));
auto out_accum = torch::empty({num_heads, num_splits, total_q, head_size_v}, opts.dtype(at::kFloat));

int64_t counter_offset = batch_size * num_heads * ck_tile::get_warp_size();
auto rng_state = torch::empty({2}, opts.dtype(torch::kInt64));
auto rng_state_ptr = reinterpret_cast<uint64_t*>(rng_state.data_ptr());

if (p_dropout > 0.0) {
int64_t counter_offset = batch_size * num_heads * ck_tile::get_warp_size();
auto gen = at::get_generator_or_default<at::CUDAGeneratorImpl>(
gen_, at::cuda::detail::getDefaultCUDAGenerator());
// See Note [Acquire lock when using random generators]
Expand All @@ -490,12 +482,21 @@ mha_varlen_fwd(at::Tensor &q, // [total_q, hq, d]
aiter::ParsePhiloxCudaState, dim3(1), dim3(64), 0, 0, philox_args, rng_state_ptr);
}
std::optional<const at::Tensor> seqlens_k = std::nullopt;

if (max_seqlen_k > 0) {
auto stream = at::cuda::getCurrentHIPStream().stream();
ck_tile::stream_config stream_config{stream};

if (paged_KV)
{
int num_splits = 0;
num_splits = aiter::override_num_splits_if_necessary(batch_size, num_heads, max_seqlen_q, head_size_v, 0, num_splits);
TORCH_CHECK(num_splits > 0, "num_splits should greater than 0");
TORCH_CHECK(num_splits <= 128, "num_splits greater than 128 is not supported");

auto softmax_lse_accum = torch::empty({num_heads, num_splits, total_q}, opts.dtype(at::kFloat));
auto out_accum = torch::empty({num_heads, num_splits, total_q, head_size_v}, opts.dtype(at::kFloat));

auto args =
get_ck_fmha_varlen_fwd_splitkv_args(
has_lse,
Expand Down Expand Up @@ -581,7 +582,7 @@ mha_varlen_fwd(at::Tensor &q, // [total_q, hq, d]
out.zero_();
softmax_lse.fill_(std::numeric_limits<float>::infinity());
}

return {out, softmax_lse, p, rng_state};
}

Expand Down