Skip to content

Fix potential memory leak in sparse_attn_decode_interface #174

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
NJX-njx wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix potential memory leak in sparse_attn_decode_interface #174

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
14 changes: 7 additions & 7 deletions csrc/api/sparse_decode.h
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,7 @@
#pragma once

#include <memory>

#include "common.h"

#include "params.h"
Expand Down Expand Up @@ -359,23 +361,23 @@ sparse_attn_decode_interface(
features.push_back(DecodeFeatures::EXTRA_TOPK_LENGTH);
}

DecodeImplBase* impl;
std::unique_ptr<DecodeImplBase> impl;
if (arch.is_sm100f()) {
if (h_q == 64) {
impl = new Decode_Sm100_Head64_Impl();
impl = std::make_unique<Decode_Sm100_Head64_Impl>();
} else if (h_q == 128) {
if (d_qk == 576) {
impl = new Decode_Sm100_Head64x2_Impl();
impl = std::make_unique<Decode_Sm100_Head64x2_Impl>();
} else if (d_qk == 512) {
impl = new Decode_Sm100_Head128_Impl();
impl = std::make_unique<Decode_Sm100_Head128_Impl>();
} else {
TORCH_CHECK(false, "Unsupported d_qk: ", d_qk);
}
} else {
TORCH_CHECK(false, "Unsupported h_q: ", h_q);
}
} else if (arch.is_sm90a()) {
impl = new Decode_Sm90_Impl();
impl = std::make_unique<Decode_Sm90_Impl>();
} else {
TORCH_CHECK(false, "Unsupported architecture for sparse decode fwd");
}
Expand Down Expand Up @@ -489,7 +491,5 @@ sparse_attn_decode_interface(
};
smxx::decode::run_flash_mla_combine_kernel<bf16>(combine_params);

delete impl;

return {out, lse.transpose(1, 2), tile_scheduler_metadata, num_splits};
}
Loading