Skip to content
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.

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

Numerous hipblaslt related fixes #2772

Merged
merged 2 commits into from
Dec 4, 2024

Conversation

pemeliya
Copy link

  1. fixing multi-threading/multi-GPU issue in gpublaslt matmul_op kernel
  2. added better cache gpublas_lt_matmul thunk
  3. added #define LEGACY_HIPBLAS_DIRECT to hipblaslt wrapper to fix potential problems with new hipblaslt versions

Copy link

@i-chaochen i-chaochen left a comment

Choose a reason for hiding this comment

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

Thanks for the work and comments!

static std::deque< BlasLtMatmulPlanCache > meta(8);
absl::MutexLock lock(&m);
size_t dev_id = stream->parent()->device_ordinal();
if (dev_id >= meta.size()) meta.resize(dev_id + 1);
Copy link

@i-chaochen i-chaochen Nov 28, 2024

Choose a reason for hiding this comment

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

If it's multi-nodes and each node at least has 8 gpus, I think we should meta.resize(dev_id + 8); ?

@pemeliya pemeliya force-pushed the r2.17-rocm-enhanced-hipblaslt-fixes branch from 7cd6ea8 to f229ec4 Compare November 29, 2024 20:36
@jayfurmanek
Copy link

We need this for 2.18 as well.

@pemeliya pemeliya merged commit e49be7f into r2.17-rocm-enhanced Dec 4, 2024
3 of 6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants