metal: per-op source split + parallel compile (replay) - #26561
Merged
Conversation
github-actions
Bot
added
ggml
forforever73
force-pushed
the
stack/metal-split
branch
2 times, most recently
from
August 5, 2026 08:22
2934b5b to
5522498
Compare
* preliminary extract common header * op source split * split metallib into 8 libs && load in parallel * derive kernel->library routing from functionNames * x-macro lib list + underscore filenames, dedup QK_NL, MRC fixes * op source split 8 to 20 * improve robustness of source fallback * clean up * change bool -> atomic_bool * only prepend headers that source actually includes * no semaphore, use GCD global queue * dedup library compile path, fix NSError lifetime, rename gla * relocate upstream concat/rope_back/repeat kernel changes into split files * move ggml-common.h from common.h into dequantize.h to shrink binary size --------- Co-authored-by: lvyichen <lvyichen@stepfun.com>
Move the kernels added on master after the split (lightning indexer, DSv4 hyper-connections, silu_back, f16 bin ops, TQ2_0, the flash-attn KV dequantization pass, rope offset/inplace, ssm_scan rollback, packed q8_0 dequantization and the tensor-API mat-mat K clamp) into the corresponding kernels/*.metal sources. Copied verbatim, no functional change.
forforever73
force-pushed
the
stack/metal-split
branch
from
August 22, 2026 05:54
5522498 to
89eea16
Compare
ggerganov
approved these changes
Aug 24, 2026
ggerganov
left a comment
ggerganov
left a comment
Member
There was a problem hiding this comment.
@forforever73 The stack is good to merge. Anything left on your end?
Contributor
Author
|
@ggerganov no, ready to go :) |
therealkenc
pushed a commit
to therealkenc/llama.cpp
that referenced
this pull request
Aug 24, 2026
* metal : per-op source split + parallel compile (ggml-org#24021) * preliminary extract common header * op source split * split metallib into 8 libs && load in parallel * derive kernel->library routing from functionNames * x-macro lib list + underscore filenames, dedup QK_NL, MRC fixes * op source split 8 to 20 * improve robustness of source fallback * clean up * change bool -> atomic_bool * only prepend headers that source actually includes * no semaphore, use GCD global queue * dedup library compile path, fix NSError lifetime, rename gla * relocate upstream concat/rope_back/repeat kernel changes into split files * move ggml-common.h from common.h into dequantize.h to shrink binary size --------- Co-authored-by: lvyichen <lvyichen@stepfun.com> * metal: add col2im_1d op (f32/f16/bf16) (ggml-org#25176) * metal : add set_rows with src0 f16 (ggml-org#25434) * metal : add CONV_2D_DW (depthwise convolution) support (ggml-org#21565) * metal : add Q2_0 support (ggml-org#25419) * metal: fuse snake activation (mul, sin, sqr, mul, add) (ggml-org#25459) * ggml-metal: FWHT kernel for metal backend (ggml-org#25924) * metal : port new kernels into the split sources Move the kernels added on master after the split (lightning indexer, DSv4 hyper-connections, silu_back, f16 bin ops, TQ2_0, the flash-attn KV dequantization pass, rope offset/inplace, ssm_scan rollback, packed q8_0 dequantization and the tensor-API mat-mat K clamp) into the corresponding kernels/*.metal sources. Copied verbatim, no functional change. --------- Co-authored-by: lvyichen <lvyichen@stepfun.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
srossitto79
added a commit
to srossitto79/llama.cpp
that referenced
this pull request
Aug 25, 2026
Conflicts resolved: - common/CMakeLists.txt: keep both the fork's jsonl.cpp/.h and upstream's new json.cpp/.h (common_json wrapper, ggml-org#27511). - ggml/src/ggml-cuda/out-prod.cu: take upstream's removal of the redundant cublasSetStream (ggml-org#26574 binds the cuBLAS handle to its stream at creation). The fork's conditional lda for dequantized (quantized) src0 is unchanged; upstream's unconditional lda decl is dropped as it would redeclare it. - ggml/src/ggml-metal/ggml-metal.metal: upstream split the monolithic source into kernels/*.metal (ggml-org#26561). The fork's kernels were ported into the new layout: * quantize_q3_K / quantize_q4_K / quantize_mxfp4 -> kernels/quantize.h * kernel_cpy_* q3_K/q4_K/mxfp4 instantiations -> kernels/quantize.metal * adamw gclip, out_prod, out_prod_id, get_rows_back, repeat_back, cross_entropy_loss{,_back} -> kernels/misc.metal * rms_norm_back -> kernels/norm.metal * soft_max_back -> kernels/softmax.metal No CMake or enum changes are needed: kernel->library routing is built from each compiled library's functionNames. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
1 task
ravel7524
pushed a commit
to ravel7524/llama.cpp
that referenced
this pull request
Aug 30, 2026
* metal : per-op source split + parallel compile (ggml-org#24021) * preliminary extract common header * op source split * split metallib into 8 libs && load in parallel * derive kernel->library routing from functionNames * x-macro lib list + underscore filenames, dedup QK_NL, MRC fixes * op source split 8 to 20 * improve robustness of source fallback * clean up * change bool -> atomic_bool * only prepend headers that source actually includes * no semaphore, use GCD global queue * dedup library compile path, fix NSError lifetime, rename gla * relocate upstream concat/rope_back/repeat kernel changes into split files * move ggml-common.h from common.h into dequantize.h to shrink binary size --------- Co-authored-by: lvyichen <lvyichen@stepfun.com> * metal: add col2im_1d op (f32/f16/bf16) (ggml-org#25176) * metal : add set_rows with src0 f16 (ggml-org#25434) * metal : add CONV_2D_DW (depthwise convolution) support (ggml-org#21565) * metal : add Q2_0 support (ggml-org#25419) * metal: fuse snake activation (mul, sin, sqr, mul, add) (ggml-org#25459) * ggml-metal: FWHT kernel for metal backend (ggml-org#25924) * metal : port new kernels into the split sources Move the kernels added on master after the split (lightning indexer, DSv4 hyper-connections, silu_back, f16 bin ops, TQ2_0, the flash-attn KV dequantization pass, rope offset/inplace, ssm_scan rollback, packed q8_0 dequantization and the tensor-API mat-mat K clamp) into the corresponding kernels/*.metal sources. Copied verbatim, no functional change. --------- Co-authored-by: lvyichen <lvyichen@stepfun.com> Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
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
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.
Overview
First layer of the
dev-metal-> stacked PR migration proposed in #26520. cc @ggerganovstack/metal-splitmastersince thenstack/metal-fa-vecstack/metal-tunerThe first seven commits are a replay of #24021 together with the six follow-up commits needed to adapt it to the kernels that have since been merged into
master.The final commit ports four additional kernels that landed on
masterafter the original split:kernels/fa.metalkernels/misc.metalsilu_backkernels/unary.metalkernels/binbcast.metalRequirements