sgemm: reuse loaded vector in AVX dot product calculation #17648
+9
−12
Conversation
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
This change optimizes the AVX-based `sgemm` (single-precision general matrix multiplication) kernel by introducing a local `__m256i` variable, `avec`, to cache the result of `load(A + lda * (ii + i) + l)`. Previously, this memory load was redundantly performed four times for each iteration within the `updot` calls for `Cv[0][i]` through `Cv[3][i]`. By loading vector once and reusing it, the code eliminates these redundant memory accesses, reducing memory latency and improving instruction-level parallelism. This is a common subexpression elimination (CSE) optimization, crucial for performance in tight loops of vectorized kernels. References: * [Common Subexpression Elimination - Wikipedia](https://en.wikipedia.org/wiki/Common_subexpression_elimination) * [Optimizing with Intel AVX2 - Intel Developer Zone](https://www.intel.com/content/www/us/en/developer/articles/technical/optimizing-with-intel-avx2.html) * [SIMD performance: data alignment and memory access - Daniel Lemire's Blog](https://lemire.me/blog/2012/05/31/simd-performance-data-alignment-and-memory-access/) * [Loop Optimization in Compiler Design - GeeksforGeeks](https://www.geeksforgeeks.org/loop-optimization-in-compiler-design/) * [Performance Optimization - CPU Caches and Memory Hierarchy - Princeton University](https://www.cs.princeton.edu/courses/archive/fall09/cos333/lectures/17_perf.pdf)
github-actions
bot
added
the
ggml
pwilkin
added
the
vibe-coded
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.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.
This change optimizes the AVX-based
sgemm(single-precision general matrix multiplication) kernel by introducing a local__m256ivariable,avec, to cache the result ofload(A + lda * (ii + i) + l). Previously, this memory load was redundantly performed four times for each iteration within theupdotcalls forCv[0][i]throughCv[3][i].By loading vector once and reusing it, the code eliminates these redundant memory accesses, reducing memory latency and improving instruction-level parallelism. This is a common subexpression elimination (CSE) optimization, crucial for performance in tight loops of vectorized kernels.
References:
Co-Authored-By: Gemini 2.5 Pro (References and desc commit changes)