Float to int is slow on ppc. It requires a store, a big stall, and then a load. It's even worse on a G5 if the compiler is dumb (and let's face it, I'm using gcc here...). Just use lroundf when we want an int. It will certainly be no slower.

2007-01-02 23:36:44 +00:00
parent 7a6eabd616
commit eb0b26cd82
17 changed files with 37 additions and 37 deletions
@@ -58,7 +58,7 @@ void SetThreadPrecedence( float prec )
 	// Real values are between 0 and 63.
 	if( CLAMP(prec, 0.0f, 1.0f) )
 		LOG->Warn( "Thread precedence clamped to %f.", prec );
-	thread_precedence_policy po = { integer_t( roundf(prec * 63) ) };
+	thread_precedence_policy po = { integer_t( lroundf(prec * 63) ) };
 	kern_return_t ret = thread_policy_set( mach_thread_self(), THREAD_PRECEDENCE_POLICY,
 					       (thread_policy_t)&po, THREAD_PRECEDENCE_POLICY_COUNT );