Repository navigation
metal : clamp K extent in tensor API mat-mat kernel - #27450
Merged
Merged
Conversation
github-actions
Bot
added
testing
…ple of 32 The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op on every iteration. On the last, partial K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the tensor, and the op reads those out-of-bounds elements (undefined behavior per the MSL specification, section 2.22.2). Depending on stale memory contents, this corrupted the result or produced NaN. Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range on every iteration (mirroring the tail handling of the MPP matmul2d examples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration). Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path. Assisted-by: pi:llama.cpp/Qwen3.8-27B
ggerganov
force-pushed
the
gg/metal-tensor-mm-k-oob
branch
from
August 20, 2026 17:15
29f939e to
398c331
Compare
ggerganov
changed the title
Collaborator
|
Both mtmd + test-backend-ops pass on my machine with this fix |
Member
Author
|
Awesome |
ggerganov
marked this pull request as ready for review
August 20, 2026 17:24
therealkenc
pushed a commit
to therealkenc/llama.cpp
that referenced
this pull request
Aug 24, 2026
…ple of 32 (ggml-org#27450) The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op on every iteration. On the last, partial K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the tensor, and the op reads those out-of-bounds elements (undefined behavior per the MSL specification, section 2.22.2). Depending on stale memory contents, this corrupted the result or produced NaN. Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range on every iteration (mirroring the tail handling of the MPP matmul2d examples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration). Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path. Assisted-by: pi:llama.cpp/Qwen3.8-27B
thecodacus
pushed a commit
to thecodacus/llama.cpp
that referenced
this pull request
Sep 7, 2026
…ple of 32 (ggml-org#27450) The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op on every iteration. On the last, partial K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the tensor, and the op reads those out-of-bounds elements (undefined behavior per the MSL specification, section 2.22.2). Depending on stale memory contents, this corrupted the result or produced NaN. Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range on every iteration (mirroring the tail handling of the MPP matmul2d examples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration). Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path. Assisted-by: pi:llama.cpp/Qwen3.8-27B
zbrad
pushed a commit
to zbrad/llama.cpp
that referenced
this pull request
Sep 10, 2026
…ple of 32 (ggml-org#27450) The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op on every iteration. On the last, partial K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the tensor, and the op reads those out-of-bounds elements (undefined behavior per the MSL specification, section 2.22.2). Depending on stale memory contents, this corrupted the result or produced NaN. Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range on every iteration (mirroring the tail handling of the MPP matmul2d examples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration). Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path. Assisted-by: pi:llama.cpp/Qwen3.8-27B
pl752
pushed a commit
to pl752/llama.cpp
that referenced
this pull request
Sep 15, 2026
…ple of 32 (ggml-org#27450) The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op on every iteration. On the last, partial K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the tensor, and the op reads those out-of-bounds elements (undefined behavior per the MSL specification, section 2.22.2). Depending on stale memory contents, this corrupted the result or produced NaN. Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range on every iteration (mirroring the tail handling of the MPP matmul2d examples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration). Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path. Assisted-by: pi:llama.cpp/Qwen3.8-27B
frostyautumnleaf
pushed a commit
to frostyautumnleaf/llama.cpp
that referenced
this pull request
Oct 5, 2026
…ple of 32 (ggml-org#27450) The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a static K=32 tile to the matmul2d op on every iteration. On the last, partial K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the tensor, and the op reads those out-of-bounds elements (undefined behavior per the MSL specification, section 2.22.2). Depending on stale memory contents, this corrupted the result or produced NaN. Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range on every iteration (mirroring the tail handling of the MPP matmul2d examples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration). Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path. Assisted-by: pi:llama.cpp/Qwen3.8-27B
4 tasks done
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
The Tensor API mat-mat path of
kernel_mul_mm(GGML_METAL_HAS_TENSOR) feeds a static K=32 tile to thematmul2dop on every K-loop iteration. When K is not a multiple of 32, the last, partial K tile extends past the K extent of the src1 tensor, and the op reads those out-of-bounds elements — undefined behavior per the MSL spec (section 2.22.2). Depending on stale arena memory contents, this silently corrupted results or produced NaN (e.g. #25652, K=588, M5).Fix the kernel itself: make the op use
dynamic_extentfor K, and clamp the K extent of both operand tensor views to the remaining valid K range (min(32, K - loop_k)) per iteration, so the op reads exactly the valid K range (mirroring the tail handling of the MPPmatmul2dexamples). On K-aligned inputs the clamp degenerates to the full 32-wide tile: the only difference from the static-K op is that the dynamic-K op derives K from the operand extents and edge-checks the tile against the tensor extents (a handful of integer ops per iteration).Also adds
test-backend-opsMUL_MAT cases with K not a multiple of 32 to exercise the unaligned K path.Additional information
has_tensor): fullMUL_MATsuite 1161/1161; unaligned K cases (k=65/80/588, f16/f16, f16/f32, f32/f32, batched) pass.Requirements