csgo/cstrike15_src/mathlib/randsse.cpp

//========= Copyright � 1996-2006, Valve Corporation, All rights reserved. ============//
//
// Purpose: generates 4 randum numbers in the range 0..1 quickly, using SIMD
//
//=====================================================================================//

#include <math.h>
#include <float.h>	// needed for flt_epsilon
#include "basetypes.h"
#ifndef _PS3
#include <memory.h>
#endif
#include "tier0/dbg.h"
#include "mathlib/mathlib.h"
#include "mathlib/vector.h"
#include "mathlib/ssemath.h"

// memdbgon must be the last include file in a .cpp file!!!
#include "tier0/memdbgon.h"

// see knuth volume 3 for insight.

class SIMDRandStreamContext
{
	fltx4 m_RandY[55];

	fltx4 *m_pRand_J, *m_pRand_K;


public:
	void Seed( uint32 seed )
	{
		m_pRand_J=m_RandY+23; m_pRand_K=m_RandY+54;
		for(int i=0;i<55;i++)
		{
			for(int j=0;j<4;j++)
			{
				SubFloat( m_RandY[i], j) = (seed>>16)/65536.0;
				seed=(seed+1)*3141592621u;
			}
		}
	}

	inline fltx4 RandSIMD( void )
	{
		// ret= rand[k]+rand[j]
		fltx4 retval=AddSIMD( *m_pRand_K, *m_pRand_J );
		
		// if ( ret>=1.0) ret-=1.0
		bi32x4 overflow_mask=CmpGeSIMD( retval, Four_Ones );
		retval=SubSIMD( retval, AndSIMD( Four_Ones, overflow_mask ) );
		
		*m_pRand_K = retval;
		
		// update pointers w/ wrap-around
		if ( --m_pRand_J < m_RandY )
			m_pRand_J=m_RandY+54;
		if ( --m_pRand_K < m_RandY )
			m_pRand_K=m_RandY+54;
		
		return retval;
	}
};

#define MAX_SIMULTANEOUS_RANDOM_STREAMS 32

static SIMDRandStreamContext s_SIMDRandContexts[MAX_SIMULTANEOUS_RANDOM_STREAMS];

static volatile int s_nRandContextsInUse[MAX_SIMULTANEOUS_RANDOM_STREAMS];

void SeedRandSIMD(uint32 seed)
{
	for( int i = 0; i<MAX_SIMULTANEOUS_RANDOM_STREAMS; i++)
		s_SIMDRandContexts[i].Seed( seed+i );
}

fltx4 RandSIMD( int nContextIndex )
{
	return s_SIMDRandContexts[nContextIndex].RandSIMD();
}

int GetSIMDRandContext( void )
{
	for(;;)
	{
		for(int i=0; i < NELEMS( s_SIMDRandContexts ); i++)
		{
			if ( ! s_nRandContextsInUse[i] )				// available?
			{
				// try to take it!
				if ( ThreadInterlockedAssignIf( &( s_nRandContextsInUse[i]), 1, 0 ) )
				{
					ThreadMemoryBarrier();
					return i;								// done!
				}
			}
		}
		Assert(0);											// why don't we have enough buffers?
		ThreadSleep();
	}
}

void ReleaseSIMDRandContext( int nContext )
{
	ThreadMemoryBarrier();
	s_nRandContextsInUse[ nContext ] = 0;
}


fltx4 RandSIMD( void )
{
	return s_SIMDRandContexts[0].RandSIMD();
}