vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 09:34:37 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

up to date merge, without vulkan-gen-shaders. They will be built before each release from now on, as they are very large

Browse source
This commit is contained in:
Concedo 2024-12-13 17:18:01 +08:00
commit ed75f8a741
24 changed files with 655 additions and 573 deletions
Download patch file Download diff file Expand all files Collapse all files

37
ggml/src/ggml-cpu/ggml-cpu.c
View file

@ -129,8 +129,7 @@ struct ggml_arm_arch_features_type {
#endif
#include <windows.h>
#if !defined(__clang__)
#if defined(_MSC_VER) && !defined(__clang__)
#define GGML_CACHE_ALIGN __declspec(align(GGML_CACHE_LINE))
typedef volatile LONG atomic_int;
@ -458,21 +457,21 @@ const struct ggml_type_traits_cpu * ggml_get_type_traits_cpu(enum ggml_type type
#define GGML_F32x4_ADD vaddq_f32
#define GGML_F32x4_MUL vmulq_f32
#define GGML_F32x4_REDUCE_ONE(x) vaddvq_f32(x)
#define GGML_F32x4_REDUCE(res, x) \
{ \
int offset = GGML_F32_ARR >> 1; \
for (int i = 0; i < offset; ++i) { \
(x)[i] = vaddq_f32((x)[i], (x)[offset+i]); \
} \
offset >>= 1; \
for (int i = 0; i < offset; ++i) { \
(x)[i] = vaddq_f32((x)[i], (x)[offset+i]); \
} \
offset >>= 1; \
for (int i = 0; i < offset; ++i) { \
(x)[i] = vaddq_f32((x)[i], (x)[offset+i]); \
} \
(res) = GGML_F32x4_REDUCE_ONE((x)[0]); \
#define GGML_F32x4_REDUCE(res, x) \
{ \
int offset = GGML_F32_ARR >> 1; \
for (int i = 0; i < offset; ++i) { \
(x)[i] = vaddq_f32((x)[i], (x)[offset+i]); \
} \
offset >>= 1; \
for (int i = 0; i < offset; ++i) { \
(x)[i] = vaddq_f32((x)[i], (x)[offset+i]); \
} \
offset >>= 1; \
for (int i = 0; i < offset; ++i) { \
(x)[i] = vaddq_f32((x)[i], (x)[offset+i]); \
} \
(res) = (ggml_float) GGML_F32x4_REDUCE_ONE((x)[0]); \
}
#define GGML_F32_VEC GGML_F32x4
@ -2400,7 +2399,7 @@ static void ggml_init_arm_arch_features(void) {
uint32_t hwcap2 = getauxval(AT_HWCAP2);
ggml_arm_arch_features.has_neon = !!(hwcap & HWCAP_ASIMD);
ggml_arm_arch_features.has_dotprod = !!(hwcap && HWCAP_ASIMDDP);
ggml_arm_arch_features.has_dotprod = !!(hwcap & HWCAP_ASIMDDP);
ggml_arm_arch_features.has_i8mm = !!(hwcap2 & HWCAP2_I8MM);
ggml_arm_arch_features.has_sve = !!(hwcap & HWCAP_SVE);
@ -12982,7 +12981,7 @@ static thread_ret_t ggml_graph_compute_secondary_thread(void* data);
#include "windows.h"
// TODO: support > 64 CPUs
bool ggml_thread_apply_affinity(bool * mask) {
static bool ggml_thread_apply_affinity(bool * mask) {
HANDLE h = GetCurrentThread();
uint64_t bitmask = 0ULL;