CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-28 08:31:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: use a more appropriate amount of threads when generating shaders (#16418)

Browse Source 
* use a more flexible amount of threads

* fix windows compile and 0 thread case

* nominmax
This commit is contained in:
Eve
2025-10-04 20:04:27 +00:00
committed by GitHub
parent f39283960b
commit 86df2c9ae4
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
View File

@@ -1,5 +1,3 @@
#include <iostream> #include <iostream>
#include <fstream> #include <fstream>
#include <sstream> #include <sstream>
@@ -22,6 +20,7 @@
#include <sys/types.h> #include <sys/types.h>
#ifdef _WIN32 #ifdef _WIN32
#define NOMINMAX
#include <windows.h> #include <windows.h>
#include <direct.h> // For _mkdir on Windows #include <direct.h> // For _mkdir on Windows
#else #else
@@ -306,7 +305,7 @@ using compile_count_guard = std::unique_ptr<uint32_t, decltype(&decrement_compil
compile_count_guard acquire_compile_slot() { compile_count_guard acquire_compile_slot() {
// wait until fewer than N compiles are in progress. // wait until fewer than N compiles are in progress.
// 16 is an arbitrary limit, the goal is to avoid "failed to create pipe" errors. // 16 is an arbitrary limit, the goal is to avoid "failed to create pipe" errors.
uint32_t N = 16; uint32_t N = std::max(1u, std::min(16u, std::thread::hardware_concurrency()));
std::unique_lock<std::mutex> guard(compile_count_mutex); std::unique_lock<std::mutex> guard(compile_count_mutex);
compile_count_cond.wait(guard, [N] { return compile_count < N; }); compile_count_cond.wait(guard, [N] { return compile_count < N; });
compile_count++; compile_count++;