Skip to content

CANN: optimize rope cache #15436

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

Closed
wants to merge 1 commit into from
Closed

CANN: optimize rope cache #15436

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
12 changes: 5 additions & 7 deletions ggml/src/ggml-cann/aclnn_ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2216,16 +2216,14 @@ static void aclnn_cache_init(ggml_backend_cann_context& ctx, ggml_tensor* dst,
ggml_cann_release_resources(ctx, acl_theta_scale_tensor,acl_theta_scale);
}

if(ctx.sin_ptr == nullptr) {
int64_t theta_length = theta_scale_length * ctx.max_prompt_length;
ACL_CHECK(aclrtMalloc(&ctx.sin_ptr, theta_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMalloc(&ctx.cos_ptr, theta_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST));
}
if(position_length > ctx.max_prompt_length) {
ctx.max_prompt_length = position_length;
int64_t theta_length = theta_scale_length * ctx.max_prompt_length;
ACL_CHECK(aclrtFree(ctx.sin_ptr));
ACL_CHECK(aclrtFree(ctx.cos_ptr));

if (ctx.sin_ptr != nullptr)
ACL_CHECK(aclrtFree(ctx.sin_ptr));
if (ctx.cos_ptr != nullptr)
ACL_CHECK(aclrtFree(ctx.cos_ptr));
Copy link

Copilot AI Aug 20, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The conditional check for ctx.cos_ptr != nullptr is missing braces around the ACL_CHECK(aclrtFree(ctx.cos_ptr)) call on line 2226. This creates a logic error where the free operation is always executed regardless of the null check.

Suggested change
ACL_CHECK(aclrtFree(ctx.cos_ptr));
if (ctx.cos_ptr != nullptr) {
ACL_CHECK(aclrtFree(ctx.cos_ptr));
}

Copilot uses AI. Check for mistakes.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It's fine because there's only one line in if.

ACL_CHECK(aclrtMalloc(&ctx.sin_ptr, theta_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST));
ACL_CHECK(aclrtMalloc(&ctx.cos_ptr, theta_length * sizeof(float_t), ACL_MEM_MALLOC_HUGE_FIRST));
}
Expand Down
2 changes: 1 addition & 1 deletion ggml/src/ggml-cann/common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -371,7 +371,7 @@ struct ggml_backend_cann_context {
void* init_ptr = nullptr;
void* sin_ptr = nullptr;
void* cos_ptr = nullptr;
int64_t max_prompt_length = 65536;
int64_t max_prompt_length = 0;
#ifdef USE_ACL_GRAPH
/// Cached CANN ACL graph used for executing the current ggml computation graph.
std::unique_ptr<ggml_cann_graph> cann_graph;
Expand Down
Loading