Use thread_local RNG to save 10% CPU when loading libraries.

This commit is contained in:
Alex Shvartzkop
2026-02-11 00:27:02 +03:00
committed by Seth Hillbrand
parent 18432a6499
commit 605ea60bb2
+3 -6
View File
@@ -42,15 +42,13 @@
#include <wx/log.h>
// boost:mt19937 is not thread-safe
static std::mutex rng_mutex;
// Use thread_local because boost:mt19937 is not thread-safe
// Static rng and generators are used because the overhead of constant seeding is expensive
// We rely on the default non-arg constructor of basic_random_generator to provide a random seed.
// We use a separate rng object for cases where we want to control the basic_random_generator
// initial seed by calling SeedGenerator from unit tests and other special cases.
static boost::mt19937 rng;
static boost::uuids::basic_random_generator<boost::mt19937> randomGenerator;
static thread_local boost::mt19937 rng;
static thread_local boost::uuids::basic_random_generator<boost::mt19937> randomGenerator;
// These don't have the same performance penalty, but we might as well be consistent
static boost::uuids::string_generator stringGenerator;
@@ -87,7 +85,6 @@ KIID::KIID()
}
else
{
std::lock_guard<std::mutex> lock( rng_mutex );
m_uuid = randomGenerator();
}