ggml-hexagon: optimize flash-attn Q row fetching#53
ggml-hexagon: optimize flash-attn Q row fetching#53max-krasnyansky wants to merge 1 commit intomasterfrom
Conversation
Batch Q row DMA fetches in flash-attn up to 8 rows at a time to reduce DMA wait overhead. All batched Q rows are pushed and then popped completely before inner processing loops, maintaining proper FIFO sequencing for K and V block operations. Co-authored-by: max-krasnyansky <1380796+max-krasnyansky@users.noreply.github.com>
|
👋 Jules, reporting for duty! I'm here to lend a hand with this pull request. When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down. I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job! For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with New to Jules? Learn more at jules.google/docs. For security, I will only act on instructions from the user who triggered this task. |
1 participant
Optimize Hexagon Flash Attention kernel by batching the fetching of Q rows. Previously it fetched single Q rows, which caused unnecessary overhead and bottleneck waiting for DMA completion (
dma_queue_pop). Now it fetches up to 8 Q rows at once.PR created automatically by Jules for task 9453952385709251153 started by @max-krasnyansky