Backups/hl2sdk
1
0
Fork 0
mirror of https://github.com/alliedmodders/hl2sdk.git synced 2024-12-23 01:59:43 +08:00
Code

Removed 3DNow support from mathlib.

Browse Source
This commit is contained in:
Scott Ehlert 2013-04-27 10:34:41 -05:00
parent 94844b1d99
commit e0fe50f9f5
7 changed files with 0 additions and 1434 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
linux_sdk/Makefile.mathlib
View File

@ -30,7 +30,6 @@ DO_CC += -o $@ -c $<
#####################################################################
LIB_OBJS= \
$(LIB_OBJ_DIR)/3dnow.o \
$(LIB_OBJ_DIR)/anorms.o \
$(LIB_OBJ_DIR)/bumpvects.o \
$(LIB_OBJ_DIR)/color_conversion.o \

193
mathlib/3dnow.cpp
View File

@ -1,193 +0,0 @@
//========= Copyright © 1996-2005, Valve Corporation, All rights reserved. ============//
//
// Purpose: 3DNow Math primitives.
//
//=====================================================================================//
#include <math.h>
#include <float.h> // Needed for FLT_EPSILON
#include "basetypes.h"
#include <memory.h>
#include "tier0/dbg.h"
#include "mathlib/mathlib.h"
#include "mathlib/amd3dx.h"
#include "mathlib/vector.h"
// memdbgon must be the last include file in a .cpp file!!!
#include "tier0/memdbgon.h"
#ifdef _MSC_VER
#pragma warning(disable:4244) // "conversion from 'const int' to 'float', possible loss of data"
#pragma warning(disable:4730) // "mixing _m64 and floating point expressions may result in incorrect code"
#endif
//-----------------------------------------------------------------------------
// 3D Now Implementations of optimized routines:
//-----------------------------------------------------------------------------
float _3DNow_Sqrt(float x)
{
Assert( s_bMathlibInitialized );
float root = 0.f;
#ifdef _WIN32
_asm
{
femms
movd mm0, x
PFRSQRT (mm1,mm0)
punpckldq mm0, mm0
PFMUL (mm0, mm1)
movd root, mm0
femms
}
#elif defined _LINUX || defined __APPLE__
__asm __volatile__( "femms" );
__asm __volatile__
(
"pfrsqrt %y0, %y1 \n\t"
"punpckldq %y1, %y1 \n\t"
"pfmul %y1, %y0 \n\t"
: "=y" (root), "=y" (x)
:"0" (x)
);
__asm __volatile__( "femms" );
#else
#error
#endif
return root;
}
// NJS FIXME: Need to test Recripricol squareroot performance and accuraccy
// on AMD's before using the specialized instruction.
float _3DNow_RSqrt(float x)
{
Assert( s_bMathlibInitialized );
return 1.f / _3DNow_Sqrt(x);
}
float FASTCALL _3DNow_VectorNormalize (Vector& vec)
{
Assert( s_bMathlibInitialized );
float *v = &vec[0];
float radius = 0.f;
if ( v[0] || v[1] || v[2] )
{
#ifdef _WIN32
_asm
{
mov eax, v
femms
movq mm0, QWORD PTR [eax]
movd mm1, DWORD PTR [eax+8]
movq mm2, mm0
movq mm3, mm1
PFMUL (mm0, mm0)
PFMUL (mm1, mm1)
PFACC (mm0, mm0)
PFADD (mm1, mm0)
PFRSQRT (mm0, mm1)
punpckldq mm1, mm1
PFMUL (mm1, mm0)
PFMUL (mm2, mm0)
PFMUL (mm3, mm0)
movq QWORD PTR [eax], mm2
movd DWORD PTR [eax+8], mm3
movd radius, mm1
femms
}
#elif defined _LINUX || defined __APPLE__
long long a,c;
int b,d;
memcpy(&a,&vec[0],sizeof(a));
memcpy(&b,&vec[2],sizeof(b));
memcpy(&c,&vec[0],sizeof(c));
memcpy(&d,&vec[2],sizeof(d));
__asm __volatile__( "femms" );
__asm __volatile__
(
"pfmul %y3, %y3\n\t"
"pfmul %y0, %y0 \n\t"
"pfacc %y3, %y3 \n\t"
"pfadd %y3, %y0 \n\t"
"pfrsqrt %y0, %y3 \n\t"
"punpckldq %y0, %y0 \n\t"
"pfmul %y3, %y0 \n\t"
"pfmul %y3, %y2 \n\t"
"pfmul %y3, %y1 \n\t"
: "=y" (radius), "=y" (c), "=y" (d)
: "y" (a), "0" (b), "1" (c), "2" (d)
);
memcpy(&vec[0],&c,sizeof(c));
memcpy(&vec[2],&d,sizeof(d));
__asm __volatile__( "femms" );
#else
#error
#endif
}
return radius;
}
void FASTCALL _3DNow_VectorNormalizeFast (Vector& vec)
{
_3DNow_VectorNormalize( vec );
}
// JAY: This complains with the latest processor pack
#ifdef _MSC_VER
#pragma warning(disable: 4730)
#endif
float _3DNow_InvRSquared(const float* v)
{
Assert( s_bMathlibInitialized );
float r2 = 1.f;
#ifdef _WIN32
_asm { // AMD 3DNow only routine
mov eax, v
femms
movq mm0, QWORD PTR [eax]
movd mm1, DWORD PTR [eax+8]
movd mm2, [r2]
PFMUL (mm0, mm0)
PFMUL (mm1, mm1)
PFACC (mm0, mm0)
PFADD (mm1, mm0)
PFMAX (mm1, mm2)
PFRCP (mm0, mm1)
movd [r2], mm0
femms
}
#elif defined _LINUX || defined __APPLE__
long long a,c;
int b;
memcpy(&a,&v[0],sizeof(a));
memcpy(&b,&v[2],sizeof(b));
memcpy(&c,&v[0],sizeof(c));
__asm __volatile__( "femms" );
__asm __volatile__
(
"PFMUL %y2, %y2 \n\t"
"PFMUL %y3, %y3 \n\t"
"PFACC %y2, %y2 \n\t"
"PFADD %y2, %y3 \n\t"
"PFMAX %y3, %y4 \n\t"
"PFRCP %y3, %y2 \n\t"
"movq %y2, %y0 \n\t"
: "=y" (r2)
: "0" (r2), "y" (a), "y" (b), "y" (c)
);
__asm __volatile__( "femms" );
#else
#error
#endif
return r2;
}

16
mathlib/3dnow.h
View File

@ -1,16 +0,0 @@
//========= Copyright © 1996-2006, Valve Corporation, All rights reserved. ============//
//
// Purpose:
//
//=====================================================================================//
#ifndef _3DNOW_H
#define _3DNOW_H
float _3DNow_Sqrt(float x);
float _3DNow_RSqrt(float x);
float FASTCALL _3DNow_VectorNormalize (Vector& vec);
void FASTCALL _3DNow_VectorNormalizeFast (Vector& vec);
float _3DNow_InvRSquared(const float* v);
#endif // _3DNOW_H

8
mathlib/mathlib-2005.vcproj
View File

@ -210,10 +210,6 @@
Name="Source Files"
Filter="cpp;c;cxx;rc;def;r;odl;idl;hpj;bat"
>
<File
RelativePath=".\3dnow.cpp"
>
</File>
<File
RelativePath=".\anorms.cpp"
>
@ -388,10 +384,6 @@
Name="Header Files"
Filter="h;hpp;hxx;hm;inl"
>
<File
RelativePath=".\3dnow.h"
>
</File>
<File
RelativePath=".\noisedata.h"
>

28
mathlib/mathlib_base.cpp
View File

@ -24,8 +24,6 @@
#include "mathlib/mathlib.h"
#include "mathlib/vector.h"
#if !defined( _X360 )
#include "mathlib/amd3dx.h"
#include "3dnow.h"
#include "sse.h"
#endif
@ -3198,7 +3196,6 @@ bool CalcLineToLineIntersectionSegment(
#pragma optimize( "", on )
#endif
static bool s_b3DNowEnabled = false;
static bool s_bMMXEnabled = false;
static bool s_bSSEEnabled = false;
static bool s_bSSE2Enabled = false;
@ -3235,25 +3232,6 @@ void MathLib_Init( float gamma, float texGamma, float brightness, int overbright
s_bMMXEnabled = false;
}
// SSE Generally performs better than 3DNow when present, so this is placed
// first to allow SSE to override these settings.
if ( bAllow3DNow && pi.m_b3DNow )
{
s_b3DNowEnabled = true;
// Select the 3DNow specific routines if available;
pfVectorNormalize = _3DNow_VectorNormalize;
pfVectorNormalizeFast = _3DNow_VectorNormalizeFast;
pfInvRSquared = _3DNow_InvRSquared;
pfSqrt = _3DNow_Sqrt;
pfRSqrt = _3DNow_RSqrt;
pfRSqrtFast = _3DNow_RSqrt;
}
else
{
s_b3DNowEnabled = false;
}
if ( bAllowSSE && pi.m_bSSE )
{
s_bSSEEnabled = true;
@ -3295,12 +3273,6 @@ void MathLib_Init( float gamma, float texGamma, float brightness, int overbright
BuildGammaTable( gamma, texGamma, brightness, overbright );
}
bool MathLib_3DNowEnabled( void )
{
Assert( s_bMathlibInitialized );
return s_b3DNowEnabled;
}
bool MathLib_MMXEnabled( void )
{
Assert( s_bMathlibInitialized );

1187
public/mathlib/amd3dx.h
View File

File diff suppressed because it is too large Load Diff

1
public/mathlib/mathlib.h
View File

@ -1708,7 +1708,6 @@ float CalcDistanceSqrToLineSegment2D( Vector2D const &P, Vector2D const &vLineA,
// Init the mathlib
void MathLib_Init( float gamma = 2.2f, float texGamma = 2.2f, float brightness = 0.0f, int overbright = 2.0f, bool bAllow3DNow = true, bool bAllowSSE = true, bool bAllowSSE2 = true, bool bAllowMMX = true );
bool MathLib_3DNowEnabled( void );
bool MathLib_MMXEnabled( void );
bool MathLib_SSEEnabled( void );
bool MathLib_SSE2Enabled( void );