Skip to content

Barrier before reusing GQA reduction scratch #63

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
quantz8a wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Barrier before reusing GQA reduction scratch #63

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
4 changes: 4 additions & 0 deletions h3_shaders.metal
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4003,6 +4003,10 @@ kernel void h3_gqa_causal_bf16(
threadgroup_barrier(mem_flags::mem_threadgroup);
}
float maximum = reductions[0];
/* WAR hazard: reductions[] is reused for the sum reduction. Without
* this barrier one thread can overwrite reductions[0] before another
* has read the max (antirez/h3.c#52). */
threadgroup_barrier(mem_flags::mem_threadgroup);
float local_sum = 0.0f;
for (uint key_row = tid; key_row < key_count; key_row += threads) {
float probability = exp(scores[key_row] - maximum);
Expand Down