Skip to content

sycl:remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor #12734

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
Rbiessy merged 1 commit into from
Apr 7, 2025
Merged

sycl:remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor #12734

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
5 changes: 1 addition & 4 deletions ggml/src/ggml-sycl/ggml-sycl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -372,12 +372,9 @@ static void ggml_backend_sycl_buffer_set_tensor(ggml_backend_buffer_t buffer,
auto stream = &(dpct::dev_mgr::instance().get_device(ctx->device).default_queue());
SYCL_CHECK(
CHECK_TRY_ERROR(dpct::dev_mgr::instance().get_device(ctx->device).queues_wait_and_throw()));
Comment on lines 373 to 374

@Rbiessy Rbiessy Apr 3, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

That makes sense to me. I think this wait before copying should also not be needed. Looking at the other backends it seems to be llama's responsibility to ensure the data is available on the host before calling buffer_set_tensor. Do you think we could remove that as part of this PR too?

@Alcpz Alcpz Apr 3, 2025

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

I think you are right, we discussed a similar issue in #12580 and the tensor memset API didn't need the wait. Memcpy should be equivalent.

@Rbiessy Rbiessy Apr 7, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Merging as is and we'll try to investigate the waits separately.

char* host_buf = (char*)malloc(size);
memcpy(host_buf, data, size);
SYCL_CHECK(
CHECK_TRY_ERROR((*stream).memcpy((char *)tensor->data + offset, host_buf, size)
CHECK_TRY_ERROR((*stream).memcpy((char *)tensor->data + offset, data, size)
.wait()));
free(host_buf);
}
catch (sycl::exception const &exc) {
std::cerr << exc.what() << "Exception caught at file:" << __FILE__
Expand Down