src/mesa/x86/common_x86.c

   1 /* $Id: common_x86.c,v 1.10 2001/03/03 20:33:30 brianp Exp $ */
   2
   3 /*
   4  * Mesa 3-D graphics library
   5  * Version:  3.5
   6  *
   7  * Copyright (C) 1999-2000  Brian Paul   All Rights Reserved.
   8  *
   9  * Permission is hereby granted, free of charge, to any person obtaining a
  10  * copy of this software and associated documentation files (the "Software"),
  11  * to deal in the Software without restriction, including without limitation
  12  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  13  * and/or sell copies of the Software, and to permit persons to whom the
  14  * Software is furnished to do so, subject to the following conditions:
  15  *
  16  * The above copyright notice and this permission notice shall be included
  17  * in all copies or substantial portions of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  21  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  22  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  23  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  24  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  */
  26
  27
  28 /*
  29  * Check CPU capabilities & initialize optimized funtions for this particular
  30  * processor.
  31  *
  32  * Written by Holger Waechtler <holger@akaflieg.extern.tu-berlin.de>
  33  * Changed by Andre Werthmann <wertmann@cs.uni-potsdam.de> for using the
  34  * new Katmai functions.
  35  */
  36
  37 #include <stdlib.h>
  38 #include <stdio.h>
  39 #if defined(USE_KATMAI_ASM) && defined(__linux__) && defined(_POSIX_SOURCE)
  40 #include <signal.h>
  41 #endif
  42
  43 #include "context.h"
  44 #include "common_x86_asm.h"
  45
  46
  47 int gl_x86_cpu_features = 0;
  48
  49 /* No reason for this to be public.
  50  */
  51 extern int gl_identify_x86_cpu_features( void );
  52
  53
  54 static void message( const char *msg )
  55 {
  56    GLboolean debug;
  57 #ifdef DEBUG
  58    debug = GL_TRUE;
  59 #else
  60    if ( getenv( "MESA_DEBUG" ) ) {
  61       debug = GL_TRUE;
  62    } else {
  63       debug = GL_FALSE;
  64    }
  65 #endif
  66    if ( debug ) {
  67       fprintf( stderr, "%s", msg );
  68    }
  69 }
  70
  71 #if defined(USE_KATMAI_ASM)
  72 /*
  73  * We must verify that the Streaming SIMD Extensions are truly supported
  74  * on this processor before we go ahead and hook out the optimized code.
  75  * Unfortunately, the CPUID bit isn't enough, as the OS must set the
  76  * OSFXSR bit in CR4 if it supports the extended FPU save and restore
  77  * required to use SSE.  Unfortunately, we can't just go ahead and read
  78  * this register, as only the kernel can do that.  Similarly, we must
  79  * verify that the OSXMMEXCPT bit in CR4 has been set by the OS,
  80  * signifying that it supports unmasked SIMD FPU exceptions.  If we take
  81  * an unmasked exception and the OS doesn't correctly support them, the
  82  * best we'll get is a SIGILL and the worst we'll get is an infinite
  83  * loop in the signal delivery from the kernel as we can't interact with
  84  * the SIMD FPU state to clear the exception bits.  Either way, this is
  85  * not good.
  86  */
  87
  88 extern void gl_test_os_katmai_support( void );
  89 extern void gl_test_os_katmai_exception_support( void );
  90
  91 #if defined(__linux__) && defined(_POSIX_SOURCE)
  92 static void sigill_handler( int signal, struct sigcontext sc )
  93 {
  94    message( "SIGILL, " );
  95
  96    /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
  97     * instructions are 3 bytes long.  We must increment the instruction
  98     * pointer manually to avoid repeated execution of the offending
  99     * instruction.
 100     *
 101     * If the SIGILL is caused by a divide-by-zero when unmasked
 102     * exceptions aren't supported, the SIMD FPU status and control
 103     * word will be restored at the end of the test, so we don't need
 104     * to worry about doing it here.  Besides, we may not be able to...
 105     */
 106    sc.eip += 3;
 107
 108    gl_x86_cpu_features &= ~(X86_FEATURE_XMM);
 109 }
 110
 111 static void sigfpe_handler( int signal, struct sigcontext sc )
 112 {
 113    message( "SIGFPE, " );
 114
 115    if ( sc.fpstate->magic != 0xffff ) {
 116       /* Our signal context has the extended FPU state, so reset the
 117        * divide-by-zero exception mask and clear the divide-by-zero
 118        * exception bit.
 119        */
 120       sc.fpstate->mxcsr |= 0x00000200;
 121       sc.fpstate->mxcsr &= 0xfffffffb;
 122    } else {
 123       /* If we ever get here, we're completely hosed.
 124        */
 125       message( "\n\n" );
 126       _mesa_problem( NULL, "SSE enabling test failed badly!" );
 127    }
 128 }
 129 #endif /* __linux__ && _POSIX_SOURCE */
 130
 131 /* If we're running on a processor that can do SSE, let's see if we
 132  * are allowed to or not.  This will catch 2.4.0 or later kernels that
 133  * haven't been configured for a Pentium III but are running on one,
 134  * and RedHat patched 2.2 kernels that have broken exception handling
 135  * support for user space apps that do SSE.
 136  *
 137  * GH: Isn't this just awful?
 138  */
 139 static void check_os_katmai_support( void )
 140 {
 141 #if defined(__linux__)
 142 #if defined(_POSIX_SOURCE)
 143    struct sigaction saved_sigill;
 144    struct sigaction saved_sigfpe;
 145
 146    /* Save the original signal handlers.
 147     */
 148    sigaction( SIGILL, NULL, &saved_sigill );
 149    sigaction( SIGFPE, NULL, &saved_sigfpe );
 150
 151    signal( SIGILL, (void (*)(int))sigill_handler );
 152    signal( SIGFPE, (void (*)(int))sigfpe_handler );
 153
 154    /* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
 155     * supports the extended FPU save and restore required for SSE.  If
 156     * we execute an SSE instruction on a PIII and get a SIGILL, the OS
 157     * doesn't support Streaming SIMD Exceptions, even if the processor
 158     * does.
 159     */
 160    if ( cpu_has_xmm ) {
 161       message( "Testing OS support for SSE... " );
 162
 163       gl_test_os_katmai_support();
 164
 165       if ( cpu_has_xmm ) {
 166          message( "yes.\n" );
 167       } else {
 168          message( "no!\n" );
 169       }
 170    }
 171
 172    /* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
 173     * it supports unmasked SIMD FPU exceptions.  If we unmask the
 174     * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
 175     * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
 176     * as expected, we're okay but we need to clean up after it.
 177     *
 178     * Are we being too stringent in our requirement that the OS support
 179     * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
 180     * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
 181     * doesn't even support them.  We at least know the user-space SSE
 182     * support is good in kernels that do support unmasked exceptions,
 183     * and therefore to be safe I'm going to leave this test in here.
 184     */
 185    if ( cpu_has_xmm ) {
 186       message( "Testing OS support for SSE unmasked exceptions... " );
 187
 188       gl_test_os_katmai_exception_support();
 189
 190       if ( cpu_has_xmm ) {
 191          message( "yes.\n" );
 192       } else {
 193          message( "no!\n" );
 194       }
 195    }
 196
 197    /* Restore the original signal handlers.
 198     */
 199    sigaction( SIGILL, &saved_sigill, NULL );
 200    sigaction( SIGFPE, &saved_sigfpe, NULL );
 201
 202    /* If we've gotten to here and the XMM CPUID bit is still set, we're
 203     * safe to go ahead and hook out the SSE code throughout Mesa.
 204     */
 205    if ( cpu_has_xmm ) {
 206       message( "Tests of OS support for SSE passed.\n" );
 207    } else {
 208       message( "Tests of OS support for SSE failed!\n" );
 209    }
 210 #else
 211    /* We can't use POSIX signal handling to test the availability of
 212     * SSE, so we disable it by default.
 213     */
 214    message( "Cannot test OS support for SSE, disabling to be safe.\n" );
 215    gl_x86_cpu_features &= ~(X86_FEATURE_XMM);
 216 #endif /* _POSIX_SOURCE */
 217 #else
 218    /* Do nothing on non-Linux platforms for now.
 219     */
 220    message( "Not testing OS support for SSE, leaving enabled.\n" );
 221 #endif /* __linux__ */
 222 }
 223
 224 #endif /* USE_KATMAI_ASM */
 225
 226
 227 void _mesa_init_all_x86_transform_asm( void )
 228 {
 229 #ifdef USE_X86_ASM
 230    gl_x86_cpu_features = gl_identify_x86_cpu_features();
 231
 232    if ( getenv( "MESA_NO_ASM" ) ) {
 233       gl_x86_cpu_features = 0;
 234    }
 235
 236    if ( gl_x86_cpu_features ) {
 237       _mesa_init_x86_transform_asm();
 238    }
 239
 240 #ifdef USE_MMX_ASM
 241    if ( cpu_has_mmx ) {
 242       if ( getenv( "MESA_NO_MMX" ) == 0 ) {
 243          message( "MMX cpu detected.\n" );
 244       } else {
 245          gl_x86_cpu_features &= ~(X86_FEATURE_MMX);
 246       }
 247    }
 248 #endif
 249
 250 #ifdef USE_3DNOW_ASM
 251    if ( cpu_has_3dnow ) {
 252       if ( getenv( "MESA_NO_3DNOW" ) == 0 ) {
 253          message( "3DNow! cpu detected.\n" );
 254          _mesa_init_3dnow_transform_asm();
 255       } else {
 256          gl_x86_cpu_features &= ~(X86_FEATURE_3DNOW);
 257       }
 258    }
 259 #endif
 260
 261 #ifdef USE_KATMAI_ASM
 262    if ( cpu_has_xmm && getenv( "MESA_FORCE_KATMAI" ) == 0 ) {
 263       check_os_katmai_support();
 264    }
 265    if ( cpu_has_xmm ) {
 266       if ( getenv( "MESA_NO_KATMAI" ) == 0 ) {
 267          message( "Katmai cpu detected.\n" );
 268          _mesa_init_katmai_transform_asm();
 269       } else {
 270          gl_x86_cpu_features &= ~(X86_FEATURE_XMM);
 271       }
 272    }
 273 #endif
 274 #endif
 275 }
 276
 277 /* Note: the above function must be called before this one, so that
 278  * gl_x86_cpu_features gets correctly initialized.
 279  */
 280 void _mesa_init_all_x86_vertex_asm( void )
 281 {
 282 #ifdef USE_X86_ASM
 283    if ( gl_x86_cpu_features ) {
 284       _mesa_init_x86_vertex_asm();
 285    }
 286
 287 #ifdef USE_3DNOW_ASM
 288    if ( cpu_has_3dnow && getenv( "MESA_NO_3DNOW" ) == 0 ) {
 289       _mesa_init_3dnow_vertex_asm();
 290    }
 291 #endif
 292
 293 #ifdef USE_KATMAI_ASM
 294    if ( cpu_has_xmm && getenv( "MESA_NO_KATMAI" ) == 0 ) {
 295       _mesa_init_katmai_vertex_asm();
 296    }
 297 #endif
 298 #endif
 299 }