CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml : posixify madvise and pagesize (#3037)

Browse Source 
* llama : use posix_madvise() instead of madvise() derived from BSD

sed -i 's,\<madvise\>,posix_&,g;s,\<MADV_,POSIX_&,g' llama.cpp

* ggml : use sysconf(_SC_PAGESIZE) instead of getpagesize() derived from BSD

sed -i 's,getpagesize(),sysconf(_SC_PAGESIZE),g' ggml.c

* metal : use sysconf(_SC_PAGESIZE) instead of getpagesize() derived from BSD

sed -i 's,getpagesize(),sysconf(_SC_PAGESIZE),g' ggml-metal.m
This commit is contained in:
Przemysław Pawełczyk
2023-09-07 10:15:06 +02:00
committed by GitHub
parent 178b1850eb
commit fec2fb19e4
3 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml-metal.m
View File

@@ -327,7 +327,7 @@ void ggml_metal_free(struct ggml_metal_context * ctx) {
void * ggml_metal_host_malloc(size_t n) { void * ggml_metal_host_malloc(size_t n) {
void * data = NULL; void * data = NULL;
const int result = posix_memalign((void **) &data, getpagesize(), n); const int result = posix_memalign((void **) &data, sysconf(_SC_PAGESIZE), n);
if (result != 0) { if (result != 0) {
metal_printf("%s: error: posix_memalign failed\n", __func__); metal_printf("%s: error: posix_memalign failed\n", __func__);
return NULL; return NULL;
@@ -401,7 +401,7 @@ bool ggml_metal_add_buffer(
} }
} }
const size_t size_page = getpagesize(); const size_t size_page = sysconf(_SC_PAGESIZE);
size_t size_aligned = size; size_t size_aligned = size;
if ((size_aligned % size_page) != 0) { if ((size_aligned % size_page) != 0) {

2
ggml.c
View File

@@ -194,7 +194,7 @@ typedef void * thread_ret_t;
inline static void * ggml_aligned_malloc(size_t size) { inline static void * ggml_aligned_malloc(size_t size) {
void * aligned_memory = NULL; void * aligned_memory = NULL;
#ifdef GGML_USE_METAL #ifdef GGML_USE_METAL
int result = posix_memalign(&aligned_memory, getpagesize(), size); int result = posix_memalign(&aligned_memory, sysconf(_SC_PAGESIZE), size);
#else #else
int result = posix_memalign(&aligned_memory, GGML_MEM_ALIGN, size); int result = posix_memalign(&aligned_memory, GGML_MEM_ALIGN, size);
#endif #endif

8
llama.cpp
View File

@@ -606,16 +606,16 @@ struct llama_mmap {
if (prefetch > 0) { if (prefetch > 0) {
// Advise the kernel to preload the mapped memory // Advise the kernel to preload the mapped memory
if (madvise(addr, std::min(file->size, prefetch), MADV_WILLNEED)) { if (posix_madvise(addr, std::min(file->size, prefetch), POSIX_MADV_WILLNEED)) {
fprintf(stderr, "warning: madvise(.., MADV_WILLNEED) failed: %s\n", fprintf(stderr, "warning: posix_madvise(.., POSIX_MADV_WILLNEED) failed: %s\n",
strerror(errno)); strerror(errno));
} }
} }
if (numa) { if (numa) {
// advise the kernel not to use readahead // advise the kernel not to use readahead
// (because the next page might not belong on the same node) // (because the next page might not belong on the same node)
if (madvise(addr, file->size, MADV_RANDOM)) { if (posix_madvise(addr, file->size, POSIX_MADV_RANDOM)) {
fprintf(stderr, "warning: madvise(.., MADV_RANDOM) failed: %s\n", fprintf(stderr, "warning: posix_madvise(.., POSIX_MADV_RANDOM) failed: %s\n",
strerror(errno)); strerror(errno));
} }
} }