SG-Amadeus 0b9fb0f26c vulkan: make MUL_MAT_ID BN/2 tail unconditional (llama/28923) ...

Use BN/2 as the default for BNover2 and as the disabled fallback for BNover4, and remove the enable gate from the MUL_MAT_ID BN/2 branch. The BN/4 branch remains gated by enable_smaller_matrices, while the p.N path is unchanged.

2026-09-23 20:46:47 +03:00

..

ggml-blas

llama: add default load-mode auto, which avoids mmap on iGPUs (llama/26081)

2026-08-14 22:16:06 +03:00

ggml-cann

ggml: add SWIGLU_CLAMP (llama/27930)

2026-09-04 13:39:35 +03:00

ggml-cpu

tests(s390x): add non-vxe build to tests (llama/28776)

2026-09-14 20:45:06 +03:00

ggml-cuda

HIP: Enable AllReduce for ROCm (llama/27825)

2026-09-23 20:46:47 +03:00

ggml-et

ggml: add SWIGLU_CLAMP (llama/27930)

2026-09-04 13:39:35 +03:00

ggml-hexagon

hexagon: add back missing contiguous fast-path and hvx_copy_uu for each run (llama/28886)

2026-09-23 20:46:47 +03:00

ggml-hip

CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled (llama/28079)

2026-09-14 20:45:06 +03:00

ggml-metal

metal: fix NaN in mul_mm_id when activations exceed f16 range (llama/26223)

2026-09-23 20:46:47 +03:00

ggml-musa

CUDA: replace GGML_FA_ALL_QUANTS with GGML_FA_QUANTS, more control over what is compiled (llama/28079)

2026-09-14 20:45:06 +03:00

ggml-opencl

opencl: choose the MoE expert matmul by batch size for speculative decoding/MTP (llama/27637)

2026-09-23 20:46:47 +03:00

ggml-openvino

OpenVINO: optimize stateful decode and GPU MoE inference (llama/28638)

2026-09-23 20:46:47 +03:00

ggml-rpc

rpc : hash-cache only weights (llama/28789)

2026-09-23 20:46:47 +03:00

ggml-sycl

sycl: rfc: Use radix select for top_k (llama/28670)

2026-09-14 20:45:06 +03:00

ggml-virtgpu

ggml: allow passing alloc dependencies in graph_optimize (llama/27301)

2026-09-04 13:39:33 +03:00

ggml-vulkan

vulkan: make MUL_MAT_ID BN/2 tail unconditional (llama/28923)

2026-09-23 20:46:47 +03:00

ggml-webgpu

webgpu: align tensor bindings to the type block size (llama/28382)

2026-09-14 20:45:06 +03:00

ggml-zdnn

llama: add default load-mode auto, which avoids mmap on iGPUs (llama/26081)

2026-08-14 22:16:06 +03:00

ggml-zendnn

CI: Use LLVM's OpenMP over MSVC_DEBUG_non_redist on Windows (llama/26678)

2026-08-21 19:23:26 +03:00

CMakeLists.txt

ggml : replace compile definitions with version.h.in (llama/28364)

2026-09-04 13:39:43 +03:00

ggml-alloc.c

ggml : fix ggml_clamp (llama/27644)

2026-08-25 15:28:56 +03:00

ggml-backend-dl.cpp

…

ggml-backend-dl.h

…

ggml-backend-impl.h

ggml : rename and make private ggml_op_alloc_size_may_expand() (ggml/0)

2026-09-04 13:39:43 +03:00

ggml-backend-meta.cpp

ggml-meta: propagate buffer usage and call init on the new tensors (llama/27586)

2026-09-04 13:39:29 +03:00

ggml-backend-reg.cpp

ggml : don't crash when backend search path can't be read (llama/28271)

2026-09-04 13:39:43 +03:00

ggml-backend.cpp

ggml: skip 0-sized ids tensor when offloading selected experts (llama/28739)

2026-09-14 20:45:06 +03:00

ggml-common.h

AVX2: Speed up large batch size prompt processing of IQ models (llama/27402)

2026-09-04 13:39:37 +03:00

ggml-feats.h

ggml : fix arm builds, unused var (llama/26991)

2026-08-14 22:16:06 +03:00

ggml-impl.h

ggml : update ggml_prec specification (llama/26675)

2026-09-14 20:45:06 +03:00

ggml-opt.cpp

…

ggml-quants.c

…

ggml-quants.h

…

ggml-threading.cpp

…

ggml-threading.h

…

ggml-version.h.in

ggml : replace compile definitions with version.h.in (llama/28364)

2026-09-04 13:39:43 +03:00

ggml.c

ggml : update ggml_prec specification (llama/26675)

2026-09-14 20:45:06 +03:00

ggml.cpp

…

gguf.cpp

llama : add missing headers (llama/28566)

2026-09-14 20:45:06 +03:00