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

ggml : always define ggml_fp16_t as uint16_t (#5666)

Browse source 
* ggml : always define ggml_fp16_t as uint16_t

ggml-ci

* ggml : cont

ggml-ci

* ggml : cont

* ggml : cont

ggml-ci

* ggml : cont

ggml-ci

* cuda : no longer ggml headers last

ggml-ci

* ggml : fix q6_K FP16 -> FP32 conversion

ggml-ci

* ggml : more FP16 -> FP32 conversion fixes

ggml-ci
This commit is contained in:
Georgi Gerganov 2024-02-22 23:21:39 +02:00 • committed by GitHub
parent 334f76fa38
commit 7e4f339c40
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 42 additions and 36 deletions
Download patch file Download diff file Expand all files Collapse all files

27
ggml-impl.h
View file

@ -53,11 +53,23 @@ extern "C" {
//
#include <arm_neon.h>
#define GGML_COMPUTE_FP16_TO_FP32(x) ((float) (x))
#define GGML_COMPUTE_FP32_TO_FP16(x) (x)
#define GGML_COMPUTE_FP16_TO_FP32(x) ggml_compute_fp16_to_fp32(x)
#define GGML_COMPUTE_FP32_TO_FP16(x) ggml_compute_fp32_to_fp16(x)
#define GGML_FP16_TO_FP32(x) ((float) (x))
#define GGML_FP32_TO_FP16(x) (x)
#define GGML_FP16_TO_FP32(x) ggml_compute_fp16_to_fp32(x)
static inline float ggml_compute_fp16_to_fp32(ggml_fp16_t h) {
__fp16 tmp;
memcpy(&tmp, &h, sizeof(ggml_fp16_t));
return (float)tmp;
}
static inline ggml_fp16_t ggml_compute_fp32_to_fp16(float f) {
ggml_fp16_t res;
__fp16 tmp = f;
memcpy(&res, &tmp, sizeof(ggml_fp16_t));
return res;
}
#else
@ -214,8 +226,7 @@ extern float ggml_table_f32_f16[1 << 16];
// On ARM NEON, it's quicker to directly convert x -> x instead of calling into ggml_lookup_fp16_to_fp32,
// so we define GGML_FP16_TO_FP32 and GGML_FP32_TO_FP16 elsewhere for NEON.
// This is also true for POWER9.
#if !defined(GGML_FP16_TO_FP32) || !defined(GGML_FP32_TO_FP16)
#if !defined(GGML_FP16_TO_FP32)
inline static float ggml_lookup_fp16_to_fp32(ggml_fp16_t f) {
uint16_t s;
memcpy(&s, &f, sizeof(uint16_t));
@ -223,8 +234,10 @@ inline static float ggml_lookup_fp16_to_fp32(ggml_fp16_t f) {
}
#define GGML_FP16_TO_FP32(x) ggml_lookup_fp16_to_fp32(x)
#define GGML_FP32_TO_FP16(x) GGML_COMPUTE_FP32_TO_FP16(x)
#endif
#if !defined(GGML_FP32_TO_FP16)
#define GGML_FP32_TO_FP16(x) GGML_COMPUTE_FP32_TO_FP16(x)
#endif
#define GGML_HASHTABLE_FULL ((size_t)-1)