et-backend: F32 decode GEMV rewrite for register-resident vecdot#13
Closed
RehanQasim-dev wants to merge 1 commit into
Closed
et-backend: F32 decode GEMV rewrite for register-resident vecdot#13RehanQasim-dev wants to merge 1 commit into
RehanQasim-dev wants to merge 1 commit into
Conversation
Stripe output elements across every hart of all 32 shires instead of blocking work into 16-element chunks, which only filled 8 shires for a typical decode GEMV. Adds a register-resident f32 row-dot helper with L2 prefetch for the streamed weight row, and stages the reused B activation vector into per-shire L2 SCP. Also fixes the matrix-engine GEMV dispatch check, which compared src1->ne[0] instead of src1->ne[1] and so never actually caught the n=1 decode case, sending it to the matrix engine kernel where it stalls on a single-column matmul.
Author
|
Superseded by the upstream PR against ggml-org/llama.cpp. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Performance (Llama-3.2-1B-Instruct, ET-SoC-1)
Decode improved roughly 1.9x. Prefill is the same as et since it already uses the matrix-engine path, untouched by this PR.
Decode t/s (@context=32)
Testing
Verified with llama-cli on ET-SoC-1 hardware (Llama-3.2-1B-Instruct), coherent output.