Leverage the existing GGML_F32_VEC helpers to vectorize ggml_vec_set_f32 for faster fills #16522
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
…e across SIMD registers and store in vector-sized chunks, while retaining the scalar tail for leftover elements and non-SIMD builds.
github-actions
bot
added
the
ggml
|
microbenchmarks show sometimes very little change, but sometimes a nice bump |
ggerganov
reviewed
Oct 13, 2025
Comment on lines
80
to
102
| inline static void ggml_vec_add1_f32(const int n, float * z, const float * x, const float v) { | ||
| #if defined(GGML_SIMD) | ||
| const int np = (n & ~(GGML_F32_STEP - 1)); | ||
|
|
||
| GGML_F32_VEC vv = GGML_F32_VEC_SET1(v); | ||
|
|
||
| for (int i = 0; i < np; i += GGML_F32_STEP) { | ||
| for (int j = 0; j < GGML_F32_ARR; ++j) { | ||
| GGML_F32_VEC ax = GGML_F32_VEC_LOAD(x + i + j*GGML_F32_EPR); | ||
| GGML_F32_VEC az = GGML_F32_VEC_ADD(ax, vv); | ||
| GGML_F32_VEC_STORE(z + i + j*GGML_F32_EPR, az); | ||
| } | ||
| } | ||
|
|
||
| for (int i = np; i < n; ++i) { | ||
| z[i] = x[i] + v; | ||
| } | ||
| #else | ||
| for (int i = 0; i < n; ++i) { | ||
| z[i] = x[i] + v; | ||
| } | ||
| #endif | ||
| } |
There was a problem hiding this comment.
We should make the code consistent about how it handles the leftovers. Here we duplicate the scalar code, while in ggml_vec_add_f32 above we use a common loop iterator. I think we should do the same as in ggml_vec_add_f32.
There was a problem hiding this comment.
sure thing. latest brings simd/scalar functions inline with each other.
ggerganov
approved these changes
Oct 14, 2025
|
Looks ready to merge, forgotten? |
slaren
approved these changes
Oct 22, 2025
|
Looks like we got some issues due to this PR being somewhat out of sync with recent CI additions: Edit: this one looks like it was already failing before merge: |
|
I am not sure what's the issue with the failing systems, so I have reverted this. Feel free to resubmit this change once the issue is fixed. |
FMayran
pushed a commit
to FMayran/llama.cpp
that referenced
this pull request
Oct 23, 2025
…ec_set_f32 for faster fills (ggml-org#16522) * Leverage the existing GGML_F32_VEC helpers to broadcast the fill value across SIMD registers and store in vector-sized chunks, while retaining the scalar tail for leftover elements and non-SIMD builds. * Vectorize additional f32 helper loops * Normalize f32 helper tails for ggml vec ops --------- Co-authored-by: Aaron <[email protected]>
pwilkin
pushed a commit
to pwilkin/llama.cpp
that referenced
this pull request
Oct 23, 2025
…ec_set_f32 for faster fills (ggml-org#16522) * Leverage the existing GGML_F32_VEC helpers to broadcast the fill value across SIMD registers and store in vector-sized chunks, while retaining the scalar tail for leftover elements and non-SIMD builds. * Vectorize additional f32 helper loops * Normalize f32 helper tails for ggml vec ops --------- Co-authored-by: Aaron <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 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.
Add a SIMD path to
ggml_vec_set_f32, broadcasting the fill value with the existing GGML_F32_VEC helpersKeep the scalar tail for leftover elements and non-SIMD builds