Inicio Explorar Ayuda
Iniciar sesión
id-Software
/
DOOM-3
espejo de https://github.com/id-Software/DOOM-3
Seguir 2
Destacar 0
Fork 0
Archivos Incidencias 0 Wiki
Rama: master
Ramas Etiquetas
DOOM-3
/
neo
/
idlib
/
math
/
Simd.h

Simd.h 12 KB
Permalink Histórico Raw

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205 
  1. /*
    2. 

  2. ===========================================================================
    3. 

  3. 

  4. Doom 3 GPL Source Code
    5. 

  5. Copyright (C) 1999-2011 id Software LLC, a ZeniMax Media company. 
    6. 

  6. 

  7. This file is part of the Doom 3 GPL Source Code (?Doom 3 Source Code?).  
    8. 

  8. 

  9. Doom 3 Source Code is free software: you can redistribute it and/or modify
    10. 

  10. it under the terms of the GNU General Public License as published by
    11. 

  11. the Free Software Foundation, either version 3 of the License, or
    12. 

  12. (at your option) any later version.
    13. 

  13. 

  14. Doom 3 Source Code is distributed in the hope that it will be useful,
    15. 

  15. but WITHOUT ANY WARRANTY; without even the implied warranty of
    16. 

  16. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    17. 

  17. GNU General Public License for more details.
    18. 

  18. 

  19. You should have received a copy of the GNU General Public License
    20. 

  20. along with Doom 3 Source Code.  If not, see <http://www.gnu.org/licenses/>.
    21. 

  21. 

  22. In addition, the Doom 3 Source Code is also subject to certain additional terms. You should have received a copy of these additional terms immediately following the terms and conditions of the GNU General Public License which accompanied the Doom 3 Source Code.  If not, please request a copy in writing from id Software at the address below.
    23. 

  23. 

  24. If you have questions concerning this license or the applicable additional terms, you may contact in writing id Software LLC, c/o ZeniMax Media Inc., Suite 120, Rockville, Maryland 20850 USA.
    25. 

  25. 

  26. ===========================================================================
    27. 

  27. */
    28. 

  28. 

  29. #ifndef __MATH_SIMD_H__
    30. 

  30. #define __MATH_SIMD_H__
    31. 

  31. 

  32. /*
    33. 

  33. ===============================================================================
    34. 

  34. 

  35. 	Single Instruction Multiple Data (SIMD)
    36. 

  36. 

  37. 	For optimal use data should be aligned on a 16 byte boundary.
    38. 

  38. 	All idSIMDProcessor routines are thread safe.
    39. 

  39. 

  40. ===============================================================================
    41. 

  41. */
    42. 

  42. 

  43. class idSIMD {
    44. 

  44. public:
    45. 

  45. 	static void			Init( void );
    46. 

  46. 	static void			InitProcessor( const char *module, bool forceGeneric );
    47. 

  47. 	static void			Shutdown( void );
    48. 

  48. 	static void			Test_f( const class idCmdArgs &args );
    49. 

  49. };
    50. 

  50. 

  51. 

  52. /*
    53. 

  53. ===============================================================================
    54. 

  54. 

  55. 	virtual base class for different SIMD processors
    56. 

  56. 

  57. ===============================================================================
    58. 

  58. */
    59. 

  59. 

  60. #ifdef _WIN32
    61. 

  61. #define VPCALL __fastcall
    62. 

  62. #else
    63. 

  63. #define VPCALL
    64. 

  64. #endif
    65. 

  65. 

  66. class idVec2;
    67. 

  67. class idVec3;
    68. 

  68. class idVec4;
    69. 

  69. class idVec5;
    70. 

  70. class idVec6;
    71. 

  71. class idVecX;
    72. 

  72. class idMat2;
    73. 

  73. class idMat3;
    74. 

  74. class idMat4;
    75. 

  75. class idMat5;
    76. 

  76. class idMat6;
    77. 

  77. class idMatX;
    78. 

  78. class idPlane;
    79. 

  79. class idDrawVert;
    80. 

  80. class idJointQuat;
    81. 

  81. class idJointMat;
    82. 

  82. struct dominantTri_s;
    83. 

  83. 

  84. const int MIXBUFFER_SAMPLES = 4096;
    85. 

  85. 

  86. typedef enum {
    87. 

  87. 	SPEAKER_LEFT = 0,
    88. 

  88. 	SPEAKER_RIGHT,
    89. 

  89. 	SPEAKER_CENTER,
    90. 

  90. 	SPEAKER_LFE,
    91. 

  91. 	SPEAKER_BACKLEFT,
    92. 

  92. 	SPEAKER_BACKRIGHT
    93. 

  93. } speakerLabel;
    94. 

  94. 

  95. 

  96. class idSIMDProcessor {
    97. 

  97. public:
    98. 

  98. 									idSIMDProcessor( void ) { cpuid = CPUID_NONE; }
    99. 

  99. 

  100. 	cpuid_t							cpuid;
    101. 

  101. 

  102. 	virtual const char * VPCALL		GetName( void ) const = 0;
    103. 

  103. 

  104. 	virtual void VPCALL Add( float *dst,			const float constant,	const float *src,		const int count ) = 0;
    105. 

  105. 	virtual void VPCALL Add( float *dst,			const float *src0,		const float *src1,		const int count ) = 0;
    106. 

  106. 	virtual void VPCALL Sub( float *dst,			const float constant,	const float *src,		const int count ) = 0;
    107. 

  107. 	virtual void VPCALL Sub( float *dst,			const float *src0,		const float *src1,		const int count ) = 0;
    108. 

  108. 	virtual void VPCALL Mul( float *dst,			const float constant,	const float *src,		const int count ) = 0;
    109. 

  109. 	virtual void VPCALL Mul( float *dst,			const float *src0,		const float *src1,		const int count ) = 0;
    110. 

  110. 	virtual void VPCALL Div( float *dst,			const float constant,	const float *src,		const int count ) = 0;
    111. 

  111. 	virtual void VPCALL Div( float *dst,			const float *src0,		const float *src1,		const int count ) = 0;
    112. 

  112. 	virtual void VPCALL MulAdd( float *dst,			const float constant,	const float *src,		const int count ) = 0;
    113. 

  113. 	virtual void VPCALL MulAdd( float *dst,			const float *src0,		const float *src1,		const int count ) = 0;
    114. 

  114. 	virtual void VPCALL MulSub( float *dst,			const float constant,	const float *src,		const int count ) = 0;
    115. 

  115. 	virtual void VPCALL MulSub( float *dst,			const float *src0,		const float *src1,		const int count ) = 0;
    116. 

  116. 

  117. 	virtual	void VPCALL Dot( float *dst,			const idVec3 &constant,	const idVec3 *src,		const int count ) = 0;
    118. 

  118. 	virtual	void VPCALL Dot( float *dst,			const idVec3 &constant,	const idPlane *src,		const int count ) = 0;
    119. 

  119. 	virtual void VPCALL Dot( float *dst,			const idVec3 &constant,	const idDrawVert *src,	const int count ) = 0;
    120. 

  120. 	virtual	void VPCALL Dot( float *dst,			const idPlane &constant,const idVec3 *src,		const int count ) = 0;
    121. 

  121. 	virtual	void VPCALL Dot( float *dst,			const idPlane &constant,const idPlane *src,		const int count ) = 0;
    122. 

  122. 	virtual void VPCALL Dot( float *dst,			const idPlane &constant,const idDrawVert *src,	const int count ) = 0;
    123. 

  123. 	virtual	void VPCALL Dot( float *dst,			const idVec3 *src0,		const idVec3 *src1,		const int count ) = 0;
    124. 

  124. 	virtual void VPCALL Dot( float &dot,			const float *src1,		const float *src2,		const int count ) = 0;
    125. 

  125. 

  126. 	virtual	void VPCALL CmpGT( byte *dst,			const float *src0,		const float constant,	const int count ) = 0;
    127. 

  127. 	virtual	void VPCALL CmpGT( byte *dst,			const byte bitNum,		const float *src0,		const float constant,	const int count ) = 0;
    128. 

  128. 	virtual	void VPCALL CmpGE( byte *dst,			const float *src0,		const float constant,	const int count ) = 0;
    129. 

  129. 	virtual	void VPCALL CmpGE( byte *dst,			const byte bitNum,		const float *src0,		const float constant,	const int count ) = 0;
    130. 

  130. 	virtual	void VPCALL CmpLT( byte *dst,			const float *src0,		const float constant,	const int count ) = 0;
    131. 

  131. 	virtual	void VPCALL CmpLT( byte *dst,			const byte bitNum,		const float *src0,		const float constant,	const int count ) = 0;
    132. 

  132. 	virtual	void VPCALL CmpLE( byte *dst,			const float *src0,		const float constant,	const int count ) = 0;
    133. 

  133. 	virtual	void VPCALL CmpLE( byte *dst,			const byte bitNum,		const float *src0,		const float constant,	const int count ) = 0;
    134. 

  134. 

  135. 	virtual	void VPCALL MinMax( float &min,			float &max,				const float *src,		const int count ) = 0;
    136. 

  136. 	virtual	void VPCALL MinMax( idVec2 &min,		idVec2 &max,			const idVec2 *src,		const int count ) = 0;
    137. 

  137. 	virtual	void VPCALL MinMax( idVec3 &min,		idVec3 &max,			const idVec3 *src,		const int count ) = 0;
    138. 

  138. 	virtual	void VPCALL MinMax( idVec3 &min,		idVec3 &max,			const idDrawVert *src,	const int count ) = 0;
    139. 

  139. 	virtual	void VPCALL MinMax( idVec3 &min,		idVec3 &max,			const idDrawVert *src,	const int *indexes,		const int count ) = 0;
    140. 

  140. 

  141. 	virtual	void VPCALL Clamp( float *dst,			const float *src,		const float min,		const float max,		const int count ) = 0;
    142. 

  142. 	virtual	void VPCALL ClampMin( float *dst,		const float *src,		const float min,		const int count ) = 0;
    143. 

  143. 	virtual	void VPCALL ClampMax( float *dst,		const float *src,		const float max,		const int count ) = 0;
    144. 

  144. 

  145. 	virtual void VPCALL Memcpy( void *dst,			const void *src,		const int count ) = 0;
    146. 

  146. 	virtual void VPCALL Memset( void *dst,			const int val,			const int count ) = 0;
    147. 

  147. 

  148. 	// these assume 16 byte aligned and 16 byte padded memory
    149. 

  149. 	virtual void VPCALL Zero16( float *dst,			const int count ) = 0;
    150. 

  150. 	virtual void VPCALL Negate16( float *dst,		const int count ) = 0;
    151. 

  151. 	virtual void VPCALL Copy16( float *dst,			const float *src,		const int count ) = 0;
    152. 

  152. 	virtual void VPCALL Add16( float *dst,			const float *src1,		const float *src2,		const int count ) = 0;
    153. 

  153. 	virtual void VPCALL Sub16( float *dst,			const float *src1,		const float *src2,		const int count ) = 0;
    154. 

  154. 	virtual void VPCALL Mul16( float *dst,			const float *src1,		const float constant,	const int count ) = 0;
    155. 

  155. 	virtual void VPCALL AddAssign16( float *dst,	const float *src,		const int count ) = 0;
    156. 

  156. 	virtual void VPCALL SubAssign16( float *dst,	const float *src,		const int count ) = 0;
    157. 

  157. 	virtual void VPCALL MulAssign16( float *dst,	const float constant,	const int count ) = 0;
    158. 

  158. 

  159. 	// idMatX operations
    160. 

  160. 	virtual void VPCALL MatX_MultiplyVecX( idVecX &dst, const idMatX &mat, const idVecX &vec ) = 0;
    161. 

  161. 	virtual void VPCALL MatX_MultiplyAddVecX( idVecX &dst, const idMatX &mat, const idVecX &vec ) = 0;
    162. 

  162. 	virtual void VPCALL MatX_MultiplySubVecX( idVecX &dst, const idMatX &mat, const idVecX &vec ) = 0;
    163. 

  163. 	virtual void VPCALL MatX_TransposeMultiplyVecX( idVecX &dst, const idMatX &mat, const idVecX &vec ) = 0;
    164. 

  164. 	virtual void VPCALL MatX_TransposeMultiplyAddVecX( idVecX &dst, const idMatX &mat, const idVecX &vec ) = 0;
    165. 

  165. 	virtual void VPCALL MatX_TransposeMultiplySubVecX( idVecX &dst, const idMatX &mat, const idVecX &vec ) = 0;
    166. 

  166. 	virtual void VPCALL MatX_MultiplyMatX( idMatX &dst, const idMatX &m1, const idMatX &m2 ) = 0;
    167. 

  167. 	virtual void VPCALL MatX_TransposeMultiplyMatX( idMatX &dst, const idMatX &m1, const idMatX &m2 ) = 0;
    168. 

  168. 	virtual void VPCALL MatX_LowerTriangularSolve( const idMatX &L, float *x, const float *b, const int n, int skip = 0 ) = 0;
    169. 

  169. 	virtual void VPCALL MatX_LowerTriangularSolveTranspose( const idMatX &L, float *x, const float *b, const int n ) = 0;
    170. 

  170. 	virtual bool VPCALL MatX_LDLTFactor( idMatX &mat, idVecX &invDiag, const int n ) = 0;
    171. 

  171. 

  172. 	// rendering
    173. 

  173. 	virtual void VPCALL BlendJoints( idJointQuat *joints, const idJointQuat *blendJoints, const float lerp, const int *index, const int numJoints ) = 0;
    174. 

  174. 	virtual void VPCALL ConvertJointQuatsToJointMats( idJointMat *jointMats, const idJointQuat *jointQuats, const int numJoints ) = 0;
    175. 

  175. 	virtual void VPCALL ConvertJointMatsToJointQuats( idJointQuat *jointQuats, const idJointMat *jointMats, const int numJoints ) = 0;
    176. 

  176. 	virtual void VPCALL TransformJoints( idJointMat *jointMats, const int *parents, const int firstJoint, const int lastJoint ) = 0;
    177. 

  177. 	virtual void VPCALL UntransformJoints( idJointMat *jointMats, const int *parents, const int firstJoint, const int lastJoint ) = 0;
    178. 

  178. 	virtual void VPCALL TransformVerts( idDrawVert *verts, const int numVerts, const idJointMat *joints, const idVec4 *weights, const int *index, const int numWeights ) = 0;
    179. 

  179. 	virtual void VPCALL TracePointCull( byte *cullBits, byte &totalOr, const float radius, const idPlane *planes, const idDrawVert *verts, const int numVerts ) = 0;
    180. 

  180. 	virtual void VPCALL DecalPointCull( byte *cullBits, const idPlane *planes, const idDrawVert *verts, const int numVerts ) = 0;
    181. 

  181. 	virtual void VPCALL OverlayPointCull( byte *cullBits, idVec2 *texCoords, const idPlane *planes, const idDrawVert *verts, const int numVerts ) = 0;
    182. 

  182. 	virtual void VPCALL DeriveTriPlanes( idPlane *planes, const idDrawVert *verts, const int numVerts, const int *indexes, const int numIndexes ) = 0;
    183. 

  183. 	virtual void VPCALL DeriveTangents( idPlane *planes, idDrawVert *verts, const int numVerts, const int *indexes, const int numIndexes ) = 0;
    184. 

  184. 	virtual void VPCALL DeriveUnsmoothedTangents( idDrawVert *verts, const dominantTri_s *dominantTris, const int numVerts ) = 0;
    185. 

  185. 	virtual void VPCALL NormalizeTangents( idDrawVert *verts, const int numVerts ) = 0;
    186. 

  186. 	virtual void VPCALL CreateTextureSpaceLightVectors( idVec3 *lightVectors, const idVec3 &lightOrigin, const idDrawVert *verts, const int numVerts, const int *indexes, const int numIndexes ) = 0;
    187. 

  187. 	virtual void VPCALL CreateSpecularTextureCoords( idVec4 *texCoords, const idVec3 &lightOrigin, const idVec3 &viewOrigin, const idDrawVert *verts, const int numVerts, const int *indexes, const int numIndexes ) = 0;
    188. 

  188. 	virtual int  VPCALL CreateShadowCache( idVec4 *vertexCache, int *vertRemap, const idVec3 &lightOrigin, const idDrawVert *verts, const int numVerts ) = 0;
    189. 

  189. 	virtual int  VPCALL CreateVertexProgramShadowCache( idVec4 *vertexCache, const idDrawVert *verts, const int numVerts ) = 0;
    190. 

  190. 

  191. 	// sound mixing
    192. 

  192. 	virtual void VPCALL UpSamplePCMTo44kHz( float *dest, const short *pcm, const int numSamples, const int kHz, const int numChannels ) = 0;
    193. 

  193. 	virtual void VPCALL UpSampleOGGTo44kHz( float *dest, const float * const *ogg, const int numSamples, const int kHz, const int numChannels ) = 0;
    194. 

  194. 	virtual void VPCALL MixSoundTwoSpeakerMono( float *mixBuffer, const float *samples, const int numSamples, const float lastV[2], const float currentV[2] ) = 0;
    195. 

  195. 	virtual void VPCALL MixSoundTwoSpeakerStereo( float *mixBuffer, const float *samples, const int numSamples, const float lastV[2], const float currentV[2] ) = 0;
    196. 

  196. 	virtual void VPCALL MixSoundSixSpeakerMono( float *mixBuffer, const float *samples, const int numSamples, const float lastV[6], const float currentV[6] ) = 0;
    197. 

  197. 	virtual void VPCALL MixSoundSixSpeakerStereo( float *mixBuffer, const float *samples, const int numSamples, const float lastV[6], const float currentV[6] ) = 0;
    198. 

  198. 	virtual void VPCALL MixedSoundToSamples( short *samples, const float *mixBuffer, const int numSamples ) = 0;
    199. 

  199. };
    200. 

  200. 

  201. // pointer to SIMD processor
    202. 

  202. extern idSIMDProcessor *SIMDProcessor;
    203. 

  203. 

  204. #endif /* !__MATH_SIMD_H__ */
    205. 

  205.