vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-07 21:49:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

disable force feteching

Browse source
This commit is contained in:
Lizonghang 2025-02-12 16:55:21 +04:00
parent b163918b46
commit ea0e655a8b
1 changed files with 8 additions and 8 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
src/llama.cpp
View file

@ -17827,14 +17827,14 @@ static void manage_graph_tensors(struct ggml_cgraph * cgraph, int advice, bool f
size_t len = std::max(segment.end - segment.start, static_cast<size_t>(page_size));
posix_madvise(reinterpret_cast<void *>(segment.start), len, advice); // hint to load into memory
// force to prefetch data
if (force && advice == POSIX_MADV_WILLNEED) {
volatile char * ptr = reinterpret_cast<volatile char *>(segment.start);
for (size_t off = 0; off < len; off += prefetch_dense * page_size) {
for (size_t i = 0; i < prefetch_dense; i++) {
if (off + i * page_size < len) (void)ptr[off + i * page_size];
}
}
}
// if (force && advice == POSIX_MADV_WILLNEED && false) {
// volatile char * ptr = reinterpret_cast<volatile char *>(segment.start);
// for (size_t off = 0; off < len; off += prefetch_dense * page_size) {
// for (size_t i = 0; i < prefetch_dense; i++) {
// if (off + i * page_size < len) (void)ptr[off + i * page_size];
// }
// }
// }
}
}