ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

ggml : simplify Arm fp16 CPU logic (ggml/1177)

Browse source 
* ggml : simlpify Arm fp16 CPU logic

ggml-ci

* cont : bring back CUDA/MUSA checks

ggml-ci
This commit is contained in:
Georgi Gerganov 2025-04-07 12:25:15 +03:00
parent 36ca8b3628
commit ff067dbcb9
3 changed files with 23 additions and 42 deletions
Download patch file Download diff file Expand all files Collapse all files

21
ggml/src/ggml-cpu/ggml-cpu-impl.h
View file

@ -4,13 +4,13 @@
#include "ggml.h"
#include "ggml-impl.h"
#include <stdlib.h> // load `stdlib.h` before other headers to work around MinGW bug: https://sourceforge.net/p/mingw-w64/bugs/192/
//#include <stddef.h>
#include <stdbool.h>
#include <string.h> // memcpy
#include <math.h> // fabsf
#ifdef __cplusplus
extern "C" {
#endif
@ -69,33 +69,16 @@ struct ggml_compute_params {
#endif
#if defined(__ARM_FEATURE_SVE)
#include <arm_sve.h>
#include <sys/prctl.h>
#endif
// 16-bit float
// on Arm, we use __fp16
// on x86, we use uint16_t
#if defined(__ARM_NEON)
// if YCM cannot find <arm_neon.h>, make a symbolic link to it, for example:
//
// $ ln -sfn /Library/Developer/CommandLineTools/usr/lib/clang/13.1.6/include/arm_neon.h ./src/
//
#include <arm_neon.h>
// ref: https://github.com/ggml-org/llama.cpp/pull/5404
#ifdef _MSC_VER
typedef uint16_t ggml_fp16_internal_t;
#define ggml_vld1q_u32(w,x,y,z) { ((w) + ((uint64_t)(x) << 32)), ((y) + ((uint64_t)(z) << 32)) }
#else
typedef __fp16 ggml_fp16_internal_t;
#define ggml_vld1q_u32(w,x,y,z) { (w), (x), (y), (z) }
#endif // _MSC_VER
#if !defined(__aarch64__)