root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 12:24:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Updated preloader to use multithreading

Browse Source 
Tested on Windows - a small performance hit during loading is not avoidable but this is the fastest method I found
On Linux - madvise needs a test if it's working. otherwise readahead() needs to be implemented in the TODO region
This commit is contained in:
John 2023-04-11 00:36:00 +02:00
parent f4c1c6b97a
commit 6e65b8a817
1 changed files with 7 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
llama_util.h
View File

@ -225,13 +225,19 @@ struct llama_mmap {
perror("sysconf");
return;
}
#ifdef _WIN32
HANDLE hProcess = GetCurrentProcess();
WIN32_MEMORY_RANGE_ENTRY range;
range.VirtualAddress = addr;
range.NumberOfBytes = length;
// if (!VirtualLock(addr, length)) { }; // no benefit. for systems with too little RAM we should lock a part and restrict the preload to that new length
if (!PrefetchVirtualMemory(hProcess, 1, &range, 0)) { }; // Prefetches part of the data and signals readahead to the file system
#else
// todo
//if (posix_madvise(addr, length, POSIX_MADV_WILLNEED) == -1) { };
// readahead() should be the equivalent method for Linux. I don't think madvise will cause a full fetch
// the multi threaded read below is pseudo sequential, it also needs a test without OS level readahead in place (worst case set threads to 1 in linux or return)
#endif
if (n_threads > 32)
n_threads = 32;