Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 22:08:46 +01:00
Code Issues Packages Projects Releases Wiki Activity

ggml : android and old glibc NUMA incompatibility bugfixes (#5557)

Browse Source 
* #ifdef out some code NUMA blocks for Android due to lack of support

* added in some __ANDROID__ if def gates around numa code and forced GLIBC prior to 2.29 to use a syscall for getcpu instead of the wrapper

* Changed gates on numa platform specific stuff to __gnu_linux__ to skip any platforms without glibc

* harmonizing #if defined blocks for numa code to __gnu_linux__ since that's the only model that's being followed anyways

---------

Co-authored-by: root <root@nenya.lothlorien.ca>
This commit is contained in:
bmwl 2024-02-18 23:38:32 -08:00 committed by GitHub
parent a0c2dad9d4
commit f0d1fafc02
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 14 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
ggml.c
View File

@ -23,6 +23,9 @@
#include <limits.h> #include <limits.h>
#include <stdarg.h> #include <stdarg.h>
#include <signal.h> #include <signal.h>
#if defined(__gnu_linux__)
#include <syscall.h>
#endif
#ifdef GGML_USE_METAL #ifdef GGML_USE_METAL
#include <unistd.h> #include <unistd.h>
@ -1971,7 +1974,7 @@ struct ggml_numa_nodes {
uint32_t n_nodes; uint32_t n_nodes;
uint32_t total_cpus; // hardware threads on system uint32_t total_cpus; // hardware threads on system
uint32_t current_node; // node on which main process is execting uint32_t current_node; // node on which main process is execting
#ifdef __linux__ #if defined(__gnu_linux__)
cpu_set_t cpuset; // cpuset from numactl cpu_set_t cpuset; // cpuset from numactl
#else #else
uint32_t cpuset; // no NUMA support outside of Linux at this time. Use a portable datatype uint32_t cpuset; // no NUMA support outside of Linux at this time. Use a portable datatype
@ -2009,7 +2012,7 @@ inline static void ggml_critical_section_end(void) {
atomic_fetch_sub(&g_state_barrier, 1); atomic_fetch_sub(&g_state_barrier, 1);
} }
#ifdef __linux__ #if defined(__gnu_linux__)
static cpu_set_t ggml_get_numa_affinity(void) { static cpu_set_t ggml_get_numa_affinity(void) {
cpu_set_t cpuset; cpu_set_t cpuset;
pthread_t thread; pthread_t thread;
@ -2031,7 +2034,7 @@ void ggml_numa_init(enum ggml_numa_strategy numa_flag) {
return; return;
} }
#ifdef __linux__ #if defined(__gnu_linux__)
struct stat st; struct stat st;
char path[256]; char path[256];
int rv; int rv;
@ -2063,7 +2066,13 @@ void ggml_numa_init(enum ggml_numa_strategy numa_flag) {
// figure out which node we're on // figure out which node we're on
uint current_cpu; uint current_cpu;
int getcpu_ret = getcpu(&current_cpu, &g_state.numa.current_node); int getcpu_ret = 0;
#if __GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ > 28)
getcpu_ret = getcpu(&current_cpu, &g_state.numa.current_node);
#else
// old glibc doesn't have a wrapper for this call. Fall back on direct syscall
getcpu_ret = syscall(SYS_getcpu,&current_cpu,&g_state.numa.current_node);
#endif
if (g_state.numa.n_nodes < 1 || g_state.numa.total_cpus < 1 || getcpu_ret != 0) { if (g_state.numa.n_nodes < 1 || g_state.numa.total_cpus < 1 || getcpu_ret != 0) {
g_state.numa.n_nodes = 0; g_state.numa.n_nodes = 0;
@ -16734,7 +16743,7 @@ typedef pthread_t ggml_thread_t;
#endif #endif
// Android's libc implementation "bionic" does not support setting affinity // Android's libc implementation "bionic" does not support setting affinity
#if defined(__linux__) && !defined(__BIONIC__) #if defined(__gnu_linux__)
static void set_numa_thread_affinity(int thread_n) { static void set_numa_thread_affinity(int thread_n) {
if (!ggml_is_numa()) { if (!ggml_is_numa()) {
return; return;