sync : ggml#2953

Merged

ggerganov merged 59 commits intomasterfrom

sync-ggml-25-03-27

Mar 27, 2025

Commits on Mar 27, 2025

scripts : update sync
ggerganov
committed
cmake: Comment out GGML_BIN_DIR for now (ggml/1139)

ckastner
authored and
ggerganov
committed
cmake: Enable specifying exact PowerPC CPU architecture (ggml/1138)

ckastner
authored and
ggerganov
committed
ggml : skip intermediate .air file when compiling .metallib (llama/12247)

danbev
authored and
ggerganov
committed
ggml-backend : make path_str compatible with C++20 (llama/12269)

ctrysbita
authored and
ggerganov
committed
opencl: use OpenCL C standard supported by the device (llama/12221)

Henry Linjamäki
authored and
ggerganov
committed
musa: support new arch mp_31 and update doc (llama/12296)

yeahdongcn
authored and
ggerganov
committed
mat vec double buffer (llama/12188)

netrunnereve
authored and
ggerganov
committed
metal : Cache the Metal library at the device context level (llama/12265)

BB-fat
authored and
ggerganov
committed
ggml-backend : fix backend search path (llama/12330)

jklincn
authored and
ggerganov
committed
CUDA/HIP: refractor mmqv to unify the calculation of nwarps and rows per block between host and device code. (llama/12177)

authored and
ggerganov
committed
vulkan: fix bug in coopmat1 mul_mat_id (llama/12316)

jeffbolznv
authored and
ggerganov
committed
CUDA/HIP: Fix fattn-vec-* when device warp size is not 32 (llama/12315)

IMbackK
authored and
ggerganov
committed
sycl : variable sg_size support for mmvq kernels (llama/12336)

Alberto Cabrera Pérez
authored and
ggerganov
committed
MUL_MAT optimization (llama/12382)

noemotiovon
authored and
ggerganov
committed
SYCL : support non-contiguous tensors in binary ops (add, sub, etc) (llama/12399)

authored and
ggerganov
committed
SYCL: Delete redundant plus sign and space (llama/12391)

aubreyli
authored and
ggerganov
committed
SYCL: set extras only on GGML_TYPE_Q4_0 (llama/12366)

qnixsynapse
authored and
ggerganov
committed
cmake : enable building llama.cpp using system libggml (llama/12321)

ckastner
authored and
ggerganov
committed
vulkan: Adjust coopmat2 tile sizes and selection heuristic (llama/12258)

jeffbolznv
authored and
ggerganov
committed
vulkan: Pad N dimension of B matrix for coopmat2 perf, to avoid bounds checking (llama/12273)

jeffbolznv
authored and
ggerganov
committed
vulkan: use fp32 in coopmat2 q4_k dequant function (llama/12309)

jeffbolznv
authored and
ggerganov
committed
vulkan: subgroup size tuning (llama/12087)

authored and
ggerganov
committed
vulkan: Add N/2 and N/4 optimized paths in coopmat2 shader (llama/12312)

jeffbolznv
authored and
ggerganov
committed
ggml-vulkan: remove unused find_program(glslc) (llama/12416)

guusw
authored and
ggerganov
committed
cuda : enable CUDA Graph on CUDA Toolkit < 12.x (llama/12394)

gaugarg-nv
authored and
ggerganov
committed
llama: Add support for RWKV v7 architecture (llama/12412)

MollySophia
authored and
ggerganov
committed
fixed compilation warnings in ggml-sycl (llama/12424)

lslusarczyk
authored and
ggerganov
committed
Vulkan: Default to 1GB allocations instead of 4GB to avoid fragmentation and driver issues (llama/12434)

0cc4m
authored and
ggerganov
committed
ggml : add SVE support for q6_K_q8_K (llama/12361)

fj-y-saito
authored and
ggerganov
committed
SYCL: using graphs is configurable by environment variable and compile option (llama/12371)

authored and
ggerganov
committed
musa: override warp_size of musa device to 32 (llama/12445)

yeahdongcn
authored and
ggerganov
committed
opencl: improve profiling (llama/12442)

lhez
authored and
ggerganov
committed
vulkan: Submit once enough matmul work has been recorded (llama/12406)

jeffbolznv
authored and
ggerganov
committed
Fix visionOS build and add CI (llama/12415)

authored and
ggerganov
committed
vulkan: optimize iq1 coopmat2 dequant functions (llama/12427)

jeffbolznv
authored and
ggerganov
committed
CUDA: Improve flash decoding kernel GPU occupancy for BS=1 case (llama/12183)

authored and
ggerganov
committed
ggml : block interleaving support for Q4_K quantization for x86 AVX2 architecture (llama/12332)

Srihari-mcw
authored and
ggerganov
committed
sycl: cleanup oneDNN related code (llama/12097)

sgeor255
authored and
ggerganov
committed
Fix build on Windows when ccache enabled (ggml/9954) (llama/9976)

authored and
ggerganov
committed
vulkan: workaround for AMD Windows driver 16 bit unpack8 bug (llama/12472)

netrunnereve
authored and
ggerganov
committed
Vulkan: RTE rounding for cpy to quant (llama/12480)

authored and
ggerganov
committed
vulkan: Optimize mul_mat_vec p021 and nc shaders (llama/12505)

jeffbolznv
authored and
ggerganov
committed
musa: refine compute capability (llama/12493)

yeahdongcn
authored and
ggerganov
committed
ggml : fix quantized cpy op (llama/12310)
ggerganov
committed
vulkan: fix mul_mat_vec failure in backend tests (llama/12529)

jeffbolznv
authored and
ggerganov
committed
CUDA: Fix clang warnings (llama/12540)

yeahdongcn
authored and
ggerganov
committed
opencl: simplify kernel embedding logic in cmakefile (llama/12503)

authored and
ggerganov
committed
SYCL: disable Q4_0 reorder optimization (llama/12560)

qnixsynapse
authored and
ggerganov
committed
ggml-cpu : update KleidiAI to v1.5.0 (llama/12568)

eddnjjn
authored and
ggerganov
committed
ggml : fix MUL_MAT_ID repack with Q8_K (llama/12544)
ggerganov
committed
metal : refactor mat-vec code (llama/12569)
ggerganov
committed
HIP: Add support for RDNA4 targets (llama/12372)

slojosic-amd
authored and
ggerganov
committed
SYCL: implement memset ggml backend buffer interface (llama/12580)

qnixsynapse
authored and
ggerganov
committed
llamafile : ppc64le MMA implementation for Q4_0. (llama/12489)

amritahs-ibm
authored and
ggerganov
committed
ggml : sync/merge cmake,riscv,powerpc, add common.cmake (ggml/0)
ggerganov
committed
sync : ggml
ggerganov
committed
files : remove old wkv6 (#0)
ggerganov
committed
xcf : fix visionOS build
ggerganov
committed