From: Przemysław Pawełczyk <redacted>
Date: Thu, 7 Sep 2023 08:15:06 +0000 (+0200)
Subject: ggml : posixify madvise and pagesize (#3037)
X-Git-Tag: gguf-v0.4.0~118
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=fec2fb19e4229aac58c98171c46e77144b99f8a3;p=pkg%2Fggml%2Fsources%2Fllama.cpp

ggml : posixify madvise and pagesize (#3037)

* 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
---

diff --git a/ggml-metal.m b/ggml-metal.m
index d0d23442..521ca180 100644
--- a/ggml-metal.m
+++ b/ggml-metal.m
@@ -327,7 +327,7 @@ void ggml_metal_free(struct ggml_metal_context * ctx) {
 
 void * ggml_metal_host_malloc(size_t n) {
     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) {
         metal_printf("%s: error: posix_memalign failed\n", __func__);
         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;
         if ((size_aligned % size_page) != 0) {
diff --git a/ggml.c b/ggml.c
index 38b1155c..50adf18e 100644
--- a/ggml.c
+++ b/ggml.c
@@ -194,7 +194,7 @@ typedef void * thread_ret_t;
 inline static void * ggml_aligned_malloc(size_t size) {
     void * aligned_memory = NULL;
 #ifdef GGML_USE_METAL
-    int result = posix_memalign(&aligned_memory, getpagesize(), size);
+    int result = posix_memalign(&aligned_memory, sysconf(_SC_PAGESIZE), size);
 #else
     int result = posix_memalign(&aligned_memory, GGML_MEM_ALIGN, size);
 #endif
diff --git a/llama.cpp b/llama.cpp
index 3413288f..2c9071a8 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -606,16 +606,16 @@ struct llama_mmap {
 
         if (prefetch > 0) {
             // Advise the kernel to preload the mapped memory
-            if (madvise(addr, std::min(file->size, prefetch), MADV_WILLNEED)) {
-                fprintf(stderr, "warning: madvise(.., MADV_WILLNEED) failed: %s\n",
+            if (posix_madvise(addr, std::min(file->size, prefetch), POSIX_MADV_WILLNEED)) {
+                fprintf(stderr, "warning: posix_madvise(.., POSIX_MADV_WILLNEED) failed: %s\n",
                         strerror(errno));
             }
         }
         if (numa) {
             // advise the kernel not to use readahead
             // (because the next page might not belong on the same node)
-            if (madvise(addr, file->size, MADV_RANDOM)) {
-                fprintf(stderr, "warning: madvise(.., MADV_RANDOM) failed: %s\n",
+            if (posix_madvise(addr, file->size, POSIX_MADV_RANDOM)) {
+                fprintf(stderr, "warning: posix_madvise(.., POSIX_MADV_RANDOM) failed: %s\n",
                         strerror(errno));
             }
         }