142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul/* 342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * Mesa 3-D graphics library 442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * Version: 6.3 542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * 642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * Copyright (C) 1999-2003 Brian Paul All Rights Reserved. 742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * 842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * Permission is hereby granted, free of charge, to any person obtaining a 942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * copy of this software and associated documentation files (the "Software"), 1042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * to deal in the Software without restriction, including without limitation 1142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * the rights to use, copy, modify, merge, publish, distribute, sublicense, 1242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * and/or sell copies of the Software, and to permit persons to whom the 1342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * Software is furnished to do so, subject to the following conditions: 1442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * 1542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * The above copyright notice and this permission notice shall be included 1642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * in all copies or substantial portions of the Software. 1742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * 1842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 1942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 2042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 2142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN 2242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 2342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 2442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul */ 2542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 2642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul/* 2742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * x86-64 optimizations shamelessy converted from x86/sse/3dnow assembly by 2842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul * Mikko Tiihonen 2942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul */ 3042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 3142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#ifdef USE_X86_64_ASM 3242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 3364dc397d8b6a0b1b87fde0a4f20a57fab2664ef2José Fonseca#include "main/glheader.h" 3464dc397d8b6a0b1b87fde0a4f20a57fab2664ef2José Fonseca#include "main/context.h" 3542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#include "math/m_xform.h" 3642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#include "tnl/t_context.h" 3742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#include "x86-64.h" 3881e366c65949292ab9a70e518ab5487e639db255Brian Paul#include "../x86/x86_xform.h" 3942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 4042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#ifdef DEBUG 4142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#include "math/m_debug.h" 4242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#endif 4342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 442b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanickextern void _mesa_x86_64_cpuid(unsigned int *regs); 452b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick 4642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian PaulDECLARE_XFORM_GROUP( x86_64, 4 ) 472b8d8989fb6f9c36baf166fc715182a1407ebadbIan RomanickDECLARE_XFORM_GROUP( 3dnow, 4 ) 4842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 49af52ff0c99faa88f2047c62bb5873ae9a427104dBrian Paul#else 50af52ff0c99faa88f2047c62bb5873ae9a427104dBrian Paul/* just to silence warning below */ 51af52ff0c99faa88f2047c62bb5873ae9a427104dBrian Paul#include "x86-64.h" 5242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#endif 5342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 5442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul/* 5542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_general( XFORM_ARGS ); 5642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_identity( XFORM_ARGS ); 5742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_perspective( XFORM_ARGS ); 5842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_3d( XFORM_ARGS ); 5942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_3d_no_rot( XFORM_ARGS ); 6042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_2d_no_rot( XFORM_ARGS ); 6142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulextern void _mesa_x86_64_transform_points4_2d( XFORM_ARGS ); 6242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul*/ 6342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 6442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#ifdef USE_X86_64_ASM 6542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulstatic void message( const char *msg ) 6642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul{ 67430562a980578410127aac534e11c30ac025e749Kenneth Graunke if (_mesa_getenv("MESA_DEBUG")) { 68b3aefd1cfb6aacd1695c52911dd39da50d893eceBrian Paul _mesa_debug( NULL, "%s", msg ); 6942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul } 7042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul} 7142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#endif 7242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 7342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 7442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paulvoid _mesa_init_all_x86_64_transform_asm(void) 7542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul{ 7642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#ifdef USE_X86_64_ASM 772b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick unsigned int regs[4]; 7842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 7942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul if ( _mesa_getenv( "MESA_NO_ASM" ) ) { 8042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul return; 8142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul } 8242fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 8342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul message("Initializing x86-64 optimizations\n"); 8442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 8542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 8642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _mesa_transform_tab[4][MATRIX_GENERAL] = 8742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _mesa_x86_64_transform_points4_general; 8842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _mesa_transform_tab[4][MATRIX_IDENTITY] = 8942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _mesa_x86_64_transform_points4_identity; 9042fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _mesa_transform_tab[4][MATRIX_3D] = 9142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _mesa_x86_64_transform_points4_3d; 922b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick 932b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick regs[0] = 0x80000001; 942b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick regs[1] = 0x00000000; 952b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick regs[2] = 0x00000000; 962b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick regs[3] = 0x00000000; 972b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_x86_64_cpuid(regs); 982b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick if (regs[3] & (1U << 31)) { 992b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick message("3Dnow! detected\n"); 1002b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_transform_tab[4][MATRIX_3D_NO_ROT] = 1012b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_3dnow_transform_points4_3d_no_rot; 1022b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_transform_tab[4][MATRIX_PERSPECTIVE] = 1032b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_3dnow_transform_points4_perspective; 1042b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_transform_tab[4][MATRIX_2D_NO_ROT] = 1052b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_3dnow_transform_points4_2d_no_rot; 1062b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_transform_tab[4][MATRIX_2D] = 1072b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick _mesa_3dnow_transform_points4_2d; 1082b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick 1092b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick } 1102b8d8989fb6f9c36baf166fc715182a1407ebadbIan Romanick 11142fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 1128c5ae809163c49cf3785edaee9af0f5b6cfe5f4bIan Romanick#ifdef DEBUG_MATH 11342fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _math_test_all_transform_functions("x86_64"); 11442fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _math_test_all_cliptest_functions("x86_64"); 11542fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul _math_test_all_normal_transform_functions("x86_64"); 11642fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#endif 11742fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul 11842fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul#endif 11942fa81275c67d7d1ad8d255120af0ffeeb46b963Brian Paul} 120