//---------------------------------------------------------------------
OptimisedUtil* OptimisedUtil::_detectImplementation(void)
{
//
// Some speed test results (averaged number of CPU timestamp (RDTSC) per-function call):
//
// Dagon SkeletonAnimation sample - softwareVertexSkinning:
//
// Pentium 4 3.0G HT Athlon XP 2500+ Athlon 64 X2 Dual Core 3800+
//
// Shared Buffers, General C 763677 462903 473038
// Shared Buffers, Unrolled SSE 210030 *best* 369762 228328 *best*
// Shared Buffers, General SSE 286202 352412 *best* 302796
//
// Separated Buffers, General C 762640 464840 478740
// Separated Buffers, Unrolled SSE 219222 *best* 287992 *best* 238770 *best*
// Separated Buffers, General SSE 290129 341614 307262
//
// PosOnly, General C 388663 257350 262831
// PosOnly, Unrolled SSE 139814 *best* 200323 *best* 168995 *best*
// PosOnly, General SSE 172693 213704 175447
//
// Another my own test scene - softwareVertexSkinning:
//
// Pentium P4 3.0G HT Athlon XP 2500+
//
// Shared Buffers, General C 74527 -
// Shared Buffers, Unrolled SSE 22743 *best* -
// Shared Buffers, General SSE 28527 -
//
//
// Note that speed test appears unaligned load/store instruction version
// loss performance 5%-10% than aligned load/store version, even if both
// of them access to aligned data. Thus, we should use aligned load/store
// as soon as possible.
//
//
// We are pick up the implementation based on test results above.
//
#ifdef __DO_PROFILE__
{
static OptimisedUtilProfiler msOptimisedUtilProfiler;
return &msOptimisedUtilProfiler;
}
#else // !__DO_PROFILE__
#if __OGRE_HAVE_SSE
if (PlatformInformation::getCpuFeatures() & PlatformInformation::CPU_FEATURE_SSE)
{
return _getOptimisedUtilSSE();
}
else
#endif // __OGRE_HAVE_SSE
{
return _getOptimisedUtilGeneral();
}
#endif // __DO_PROFILE__
}
OptimisedUtilProfiler(void)
{
mOptimisedUtils.push_back(_getOptimisedUtilGeneral());
#if __OGRE_HAVE_SSE
if (PlatformInformation::getCpuFeatures() & PlatformInformation::CPU_FEATURE_SSE)
{
mOptimisedUtils.push_back(_getOptimisedUtilSSE());
}
#endif
}
OptimisedUtilProfiler(void)
{
mOptimisedUtils.push_back(_getOptimisedUtilGeneral());
#if __OGRE_HAVE_SSE
if (PlatformInformation::getCpuFeatures() & PlatformInformation::CPU_FEATURE_SSE)
{
mOptimisedUtils.push_back(_getOptimisedUtilSSE());
}
//#elif __OGRE_HAVE_VFP
// if (PlatformInformation::getCpuFeatures() & PlatformInformation::CPU_FEATURE_VFP)
// {
// mOptimisedUtils.push_back(_getOptimisedUtilVFP());
// }
//#elif __OGRE_HAVE_NEON
// if (PlatformInformation::getCpuFeatures() & PlatformInformation::CPU_FEATURE_NEON)
// {
// mOptimisedUtils.push_back(_getOptimisedUtilNEON());
// }
#endif
}
