sync : ggml

ggml : fix and optimize ppc64le (ggml/849)

ggml : remove duplicate include of ggml-common.h (ggml/853)

remove global variables (llama/7710)

CUDA: faster q2_K, q3_K MMQ + int8 tensor cores (llama/7921)

metal : utilize max shared memory for mul_mat_id (llama/7935)

rpc : fix ggml_backend_rpc_supports_buft() (llama/7918)

move BLAS to a separate backend (llama/6210)

CUDA: fix broken oob check for FA vec f32 kernel (llama/7904)

tests : add non-cont unary tests (llama/7857)

ggml : improve ggml_is_contiguous logic (llama/7856)

vulkan: select only one device for single gpu with multiple drivers (llama/7582)

Update Vulkan RoPE implementation (llama/7818)

CUDA: int8 tensor cores for MMQ (q4_K, q5_K, q6_K) (llama/7860)

CUDA: use tensor cores for MMQ (llama/7676)

use the correct SYCL context for host USM allocations (llama/7777)

CUDA: revise q8_1 data layout for mul_mat_q (llama/7824)

vulkan : reuse parent extra for views (llama/7806)

fix softmax r2r result wrong issue (llama/7811)

CUDA: refactor mmq, dmmv, mmvq (llama/7716)

ggml : refactor rope norm/neox (llama/7634)

Allow number of nodes in CUDA graph to change (llama/7738)

ggml : remove OpenCL (llama/7735)

ggml : prevent builds with -ffinite-math-only (llama/7726)

llama : offload to RPC in addition to other backends (llama/7640)

ggml : use OpenMP as a thread pool (llama/7606)

Vulkan Mixture of Experts (MoE) support (llama/7628)

kompute : implement op_getrows_f32 (llama/6403)

fix bug introduced in using calloc (llama/7701)

Fix FlashAttention debug test, FP32 assert (llama/7684)

CUDA: fix Pascal FA, deq. KV to FP16 for batch > 8 (llama/7681)

CUDA: quantized KV support for FA vec (llama/7527)

ggml : fix loongson compile warnings (llama/7537)

faster avx512 exp implementation (llama/7551)

ggml : fix loongarch build (O2 issue) (llama/7636)

metal : remove invalid asserts (llama/7617)

metal : add missing asserts (llama/7617)

ggml : fix YARN + add tests + add asserts (llama/7617)

cuda : non-cont concat support (llama/7610)

llama-bench : add support for the RPC backend (llama/7435)

ggml : use atomic_flag for critical section (llama/7598)

examples : adapt to new ggml_concat (ggml/0)

ggml : fix typo in ggml.c (llama/7603)

Align GEMM dispatch (llama/7566)

sycl : fix assert (llama/7563)

vulkan: properly initialize vulkan devices for LLAMA_SPLIT_MODE_NONE (llama/7552)

rpc : resource management rework (llama/7562)

fix ggml_sycl_mul_mat_id() to match the change of api (llama/7436)

ggml : generalize GGML_OP_CONCAT (llama/7563)

update HIP_UMA #7399 (llama/7414)