From 708b1d8c897e83d6030227520387c995c173b5d9 Mon Sep 17 00:00:00 2001 From: Lizonghang <870644199@qq.com> Date: Wed, 12 Feb 2025 16:55:44 +0400 Subject: [PATCH] disable force fetching --- src/llama.cpp | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/src/llama.cpp b/src/llama.cpp index 99bb6bc3..55df72c5 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -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(page_size)); posix_madvise(reinterpret_cast(segment.start), len, advice); // hint to load into memory // force to prefetch data - // if (force && advice == POSIX_MADV_WILLNEED && false) { - // volatile char * ptr = reinterpret_cast(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(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]; + } + } + } } }