sync : ggml#3383

Merged

ggerganov merged 86 commits intomasterfrom

sync-ggml-25-08-18

Aug 18, 2025

Commits on Aug 18, 2025

scripts : update sync scripts
ggerganov
committed
vulkan : fix 32-bit builds (ggml/1313)

dg0yt
authored and
ggerganov
committed
cmake : Fix BLAS link interface (ggml/1316)

dg0yt
authored and
ggerganov
committed
sycl: refactor quantization to q8_1 (llama/14815)

Alcpz
authored and
ggerganov
committed
CUDA: fix pointer incrementation in FA (llama/14916)

JohannesGaessler
authored and
ggerganov
committed
SYCL: Add set_rows support for quantized types (llama/14883)

qnixsynapse
authored and
ggerganov
committed
ggml-cpu : deduplicate scalar implementations (llama/14897)

xctan
authored and
ggerganov
committed
CUDA: add roll (llama/14919)

am17an
authored and
ggerganov
committed
cuda : add softcap fusion (llama/14907)

CISC
authored and
ggerganov
committed
CANN: Add ggml_set_rows (llama/14943)

hipudding
authored and
ggerganov
committed
HIP: Ignore unsupported unroll transformation in fattn-vec (llama/14931)

IMbackK
authored and
ggerganov
committed
HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (llama/14930)

IMbackK
authored and
ggerganov
committed
HIP: remove the use of __HIP_PLATFORM_AMD__, explicitly support only AMD targets (llama/14945)

IMbackK
authored and
ggerganov
committed
CUDA: skip masked KV slices for all FA kernels (llama/14924)

JohannesGaessler
authored and
ggerganov
committed
HIP: enable mfma mmq on gfx908 and gfx90a for select datatypes and shapes (llama/14949)

IMbackK
authored and
ggerganov
committed
opencl: add `mul_mat_f32_f32_l4_lm` and `mul_mat_f16_f32_l4_lm` (llama/14809)

lhez
authored and
ggerganov
committed
CANN: Improve loading efficiency after converting weights to NZ format. (llama/14985)

hipudding
authored and
ggerganov
committed
Vulkan: Fix minor debug mode issues (llama/14899)

0cc4m
authored and
ggerganov
committed
docker : add cann build pipline (llama/14591)

committed
ggml : Q2k interleaving implementation - x86/x64 SIMD (llama/14373)

authored and
ggerganov
committed
opencl: add f16 for `add`, `sub`, `mul`, `div` (llama/14984)

lhez
authored and
ggerganov
committed
CUDA: fix MMQ nwarps for AMD with warp_size==32 (llama/15014)

JohannesGaessler
authored and
ggerganov
committed
vulkan: optimizations for direct convolution (llama/14933)

authored and
ggerganov
committed
vulkan: Support ne[3]>1 in noncontig matrix-vector multiply (llama/15015)

jeffbolznv
authored and
ggerganov
committed
vulkan: coopmat2 mul_mat optimizations (llama/14934)

jeffbolznv
authored and
ggerganov
committed
cuda, sycl : fix batched gemm when ne02 == 1 && ne03 > 1 (llama/15038)
ggerganov
committed
cuda: make im2col a little faster (llama/15025)

leejet
authored and
ggerganov
committed
CUDA: use mma FA kernel for gqa > 4 on RTX 4000 (llama/15035)

JohannesGaessler
authored and
ggerganov
committed
opencl: fix adreno compiler detection logic (llama/15029)

lhez
authored and
ggerganov
committed
vulkan: Use coopmat2 for conv2d (llama/14982)

jeffbolznv
authored and
ggerganov
committed
vulkan: fix build when using glslang that does not support coopmat2 (llama/15062)

jeffbolznv
authored and
ggerganov
committed
cmake: Add GGML_BACKEND_DIR option (llama/15074)

ckastner
authored and
ggerganov
committed
sycl: fix mul_mat selection (llama/15092)

Rbiessy
authored and
ggerganov
committed
llama : add gpt-oss (llama/15091)

committed
CANN: add support for ACL Graph (llama/15065)

noemotiovon
authored and
ggerganov
committed
ggml : fix fallback to CPU for ununsupported ops (llama/15118)

slaren
authored and
ggerganov
committed
opencl: add `swiglu_oai` and `add_id` (llama/15121)

lhez
authored and
ggerganov
committed
fix profiling crash (llama/15072)

rmatif
authored and
ggerganov
committed
CUDA: GEMM for FP32/FP16/BF16 and ne11 <= 16 (llama/15131)

JohannesGaessler
authored and
ggerganov
committed
ggml: Skip backend library linking code when GGML_BACKEND_DL=ON (llama/15094)

ckastner
authored and
ggerganov
committed
HIP: add cmake option to enable compiler output of kernel resource usage metrics (llama/15103)

IMbackK
authored and
ggerganov
committed
vulkan: Add env var to disable host visible vidmem (llama/15109)

jeffbolznv
authored and
ggerganov
committed
vulkan: support fattn sinks (llama/15126)

jeffbolznv
authored and
ggerganov
committed
opencl: support sink in `soft_max` (attn sinks) (llama/15152)

lhez
authored and
ggerganov
committed
CUDA: attention sinks for mma FlashAttention (llama/15157)

JohannesGaessler
authored and
ggerganov
committed
ggml : fix field name when new ggml_backend (llama/14944)

aisk
authored and
ggerganov
committed
gguf-py : add Numpy MXFP4 de/quantization support (llama/15111)

compilade
authored and
ggerganov
committed
CUDA: add attention sinks for tile and wmma (llama/15178)

am17an
authored and
ggerganov
committed
cuda: refactored ssm_scan and use CUB (llama/13291)

Your-Cheese
authored and
ggerganov
committed
kleidiai: fix unsigned overflow bug (llama/15150)

chaxu01
authored and
ggerganov
committed
CANN: Add broadcast for softmax and FA (llama/15208)

hipudding
authored and
ggerganov
committed
musa: fix failures in test-backend-ops for mul_mat_id op (llama/15236)

yeahdongcn
authored and
ggerganov
committed
CANN: GGML_OP_CPY optimization (llama/15070)

noemotiovon
authored and
ggerganov
committed
CUDA cmake: add `-lineinfo` for easier debug (llama/15260)

am17an
authored and
ggerganov
committed
opencl: allow mixed f16/f32 `add` (llama/15140)

rmatif
authored and
ggerganov
committed
sycl: Fix and disable more configurations of mul_mat (llama/15151)

Rbiessy
authored and
ggerganov
committed
HIP: disable sync warp shuffel operators from clr amd_warp_sync_functions.h (llama/15273)

IMbackK
authored and
ggerganov
committed
ggml-rpc: chunk send()/recv() to avoid EINVAL for very large tensors over RPC (macOS & others) (llama/15188)

authored and
ggerganov
committed
CUDA: Optimize `reduce_rows_f32` kernel, leading up to 25x perf improvement on kernel-level and 10% perf increase for Gemma3n (llama/15132)

ORippler
authored and
ggerganov
committed
ggml : repack block_iq4_nlx8 (llama/14904)
ggerganov
committed
ggml : update `ggml_rope_multi` (llama/12665)

foldl
and
ggerganov
committed
HIP: bump requirement to rocm 6.1 (llama/15296)

IMbackK
authored and
ggerganov
committed
finetune: SGD optimizer, more CLI args (llama/13873)

authored and
ggerganov
committed
cuda : fix GGML_CUDA_GRAPHS=OFF (llama/15300)

CISC
authored and
ggerganov
committed
ggml: fix ggml_conv_1d_dw bug (ggml/1323)

jason-ni
authored and
ggerganov
committed
vulkan: perf_logger improvements (llama/15246)

jeffbolznv
authored and
ggerganov
committed
HIP: Cleanup hipification header (llama/15285)

authored and
ggerganov
committed
CUDA: fix negative KV_max values in FA (llama/15321)

JohannesGaessler
authored and
ggerganov
committed
ggml: initial IBM zDNN backend (llama/14975)

taronaeo
authored and
ggerganov
committed
vulkan : fix compile warnings on macos (llama/15340)
ggerganov
committed
vulkan : fix out-of-bounds access in argmax kernel (llama/15342)
ggerganov
committed
opencl: add initial mxfp4 support via mv (llama/15270)

authored and
ggerganov
committed
OpenCL: add initial FA support (llama/14987)

rmatif
authored and
ggerganov
committed
vulkan: Add missing bounds checking to scalar/coopmat1 mul_mat_id (llama/15334)

jeffbolznv
authored and
ggerganov
committed
vulkan: Support mul_mat_id with f32 accumulators (llama/15337)

jeffbolznv
authored and
ggerganov
committed
vulkan: fuse adds (llama/15252)

jeffbolznv
authored and
ggerganov
committed
vulkan: Optimize argsort (llama/15354)

jeffbolznv
authored and
ggerganov
committed
vulkan: support sqrt (llama/15370)

ddwkim
authored and
ggerganov
committed
vulkan: Use larger workgroups for mul_mat_vec when M is small (llama/15355)

authored and
ggerganov
committed
vulkan: disable spirv-opt for bfloat16 shaders (llama/15352)

jeffbolznv
authored and
ggerganov
committed
ggml-quants : fix make_qp_quants NANs and IQ1 assertion errors (llama/15379)

compilade
authored and
ggerganov
committed
common : handle mxfp4 enum
ggerganov
committed
ggml : initial zDNN backend (llama/14975)

taronaeo
authored and
ggerganov
committed
ggml: Add initial WebGPU backend (llama/14521)

reeselevine
authored and
ggerganov
committed
sync : ggml
ggerganov
committed
talk-llama : sync llama.cpp
ggerganov
committed