src/mesa/x86/common_x86.c

   1 /* $Id: common_x86.c,v 1.20 2002/11/13 15:03:31 brianp Exp $ */
   2
   3 /*
   4  * Mesa 3-D graphics library
   5  * Version:  5.0
   6  *
   7  * Copyright (C) 1999-2002  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  * Check CPU capabilities & initialize optimized funtions for this particular
  29  * processor.
  30  *
  31  * Written by Holger Waechtler <holger@akaflieg.extern.tu-berlin.de>
  32  * Changed by Andre Werthmann <wertmann@cs.uni-potsdam.de> for using the
  33  * new Katmai functions.
  34  */
  35
  36 #include <stdlib.h>
  37 #include <stdio.h>
  38 #if defined(USE_SSE_ASM) && defined(__linux__)
  39 #include <signal.h>
  40 #endif
  41 #if defined(USE_SSE_ASM) && defined(__FreeBSD__)
  42 #include <sys/types.h>
  43 #include <sys/sysctl.h>
  44 #endif
  45
  46 #include "context.h"
  47 #include "common_x86_asm.h"
  48 #include "imports.h"
  49
  50
  51 int _mesa_x86_cpu_features = 0;
  52
  53 /* No reason for this to be public.
  54  */
  55 extern int _mesa_identify_x86_cpu_features( void );
  56
  57
  58 static void message( const char *msg )
  59 {
  60    GLboolean debug;
  61 #ifdef DEBUG
  62    debug = GL_TRUE;
  63 #else
  64    if ( getenv( "MESA_DEBUG" ) ) {
  65       debug = GL_TRUE;
  66    } else {
  67       debug = GL_FALSE;
  68    }
  69 #endif
  70    if ( debug ) {
  71       fprintf( stderr, "%s", msg );
  72    }
  73 }
  74
  75 #if defined(USE_SSE_ASM)
  76 /*
  77  * We must verify that the Streaming SIMD Extensions are truly supported
  78  * on this processor before we go ahead and hook out the optimized code.
  79  * Unfortunately, the CPUID bit isn't enough, as the OS must set the
  80  * OSFXSR bit in CR4 if it supports the extended FPU save and restore
  81  * required to use SSE.  Unfortunately, we can't just go ahead and read
  82  * this register, as only the kernel can do that.  Similarly, we must
  83  * verify that the OSXMMEXCPT bit in CR4 has been set by the OS,
  84  * signifying that it supports unmasked SIMD FPU exceptions.  If we take
  85  * an unmasked exception and the OS doesn't correctly support them, the
  86  * best we'll get is a SIGILL and the worst we'll get is an infinite
  87  * loop in the signal delivery from the kernel as we can't interact with
  88  * the SIMD FPU state to clear the exception bits.  Either way, this is
  89  * not good.
  90  */
  91
  92 extern void _mesa_test_os_sse_support( void );
  93 extern void _mesa_test_os_sse_exception_support( void );
  94
  95 #if defined(__linux__) && defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
  96 static void sigill_handler( int signal, struct sigcontext sc )
  97 {
  98    message( "SIGILL, " );
  99
 100    /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
 101     * instructions are 3 bytes long.  We must increment the instruction
 102     * pointer manually to avoid repeated execution of the offending
 103     * instruction.
 104     *
 105     * If the SIGILL is caused by a divide-by-zero when unmasked
 106     * exceptions aren't supported, the SIMD FPU status and control
 107     * word will be restored at the end of the test, so we don't need
 108     * to worry about doing it here.  Besides, we may not be able to...
 109     */
 110    sc.eip += 3;
 111
 112    _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM);
 113 }
 114
 115 static void sigfpe_handler( int signal, struct sigcontext sc )
 116 {
 117    message( "SIGFPE, " );
 118
 119    if ( sc.fpstate->magic != 0xffff ) {
 120       /* Our signal context has the extended FPU state, so reset the
 121        * divide-by-zero exception mask and clear the divide-by-zero
 122        * exception bit.
 123        */
 124       sc.fpstate->mxcsr |= 0x00000200;
 125       sc.fpstate->mxcsr &= 0xfffffffb;
 126    } else {
 127       /* If we ever get here, we're completely hosed.
 128        */
 129       message( "\n\n" );
 130       _mesa_problem( NULL, "SSE enabling test failed badly!" );
 131    }
 132 }
 133 #endif /* __linux__ && _POSIX_SOURCE && X86_FXSR_MAGIC */
 134
 135 /* If we're running on a processor that can do SSE, let's see if we
 136  * are allowed to or not.  This will catch 2.4.0 or later kernels that
 137  * haven't been configured for a Pentium III but are running on one,
 138  * and RedHat patched 2.2 kernels that have broken exception handling
 139  * support for user space apps that do SSE.
 140  *
 141  * GH: Isn't this just awful?
 142  */
 143 static void check_os_sse_support( void )
 144 {
 145 #if defined(__linux__)
 146 #if defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
 147    struct sigaction saved_sigill;
 148    struct sigaction saved_sigfpe;
 149
 150    /* Save the original signal handlers.
 151     */
 152    sigaction( SIGILL, NULL, &saved_sigill );
 153    sigaction( SIGFPE, NULL, &saved_sigfpe );
 154
 155    signal( SIGILL, (void (*)(int))sigill_handler );
 156    signal( SIGFPE, (void (*)(int))sigfpe_handler );
 157
 158    /* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
 159     * supports the extended FPU save and restore required for SSE.  If
 160     * we execute an SSE instruction on a PIII and get a SIGILL, the OS
 161     * doesn't support Streaming SIMD Exceptions, even if the processor
 162     * does.
 163     */
 164    if ( cpu_has_xmm ) {
 165       message( "Testing OS support for SSE... " );
 166
 167       _mesa_test_os_sse_support();
 168
 169       if ( cpu_has_xmm ) {
 170          message( "yes.\n" );
 171       } else {
 172          message( "no!\n" );
 173       }
 174    }
 175
 176    /* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
 177     * it supports unmasked SIMD FPU exceptions.  If we unmask the
 178     * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
 179     * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
 180     * as expected, we're okay but we need to clean up after it.
 181     *
 182     * Are we being too stringent in our requirement that the OS support
 183     * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
 184     * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
 185     * doesn't even support them.  We at least know the user-space SSE
 186     * support is good in kernels that do support unmasked exceptions,
 187     * and therefore to be safe I'm going to leave this test in here.
 188     */
 189    if ( cpu_has_xmm ) {
 190       message( "Testing OS support for SSE unmasked exceptions... " );
 191
 192       _mesa_test_os_sse_exception_support();
 193
 194       if ( cpu_has_xmm ) {
 195          message( "yes.\n" );
 196       } else {
 197          message( "no!\n" );
 198       }
 199    }
 200
 201    /* Restore the original signal handlers.
 202     */
 203    sigaction( SIGILL, &saved_sigill, NULL );
 204    sigaction( SIGFPE, &saved_sigfpe, NULL );
 205
 206    /* If we've gotten to here and the XMM CPUID bit is still set, we're
 207     * safe to go ahead and hook out the SSE code throughout Mesa.
 208     */
 209    if ( cpu_has_xmm ) {
 210       message( "Tests of OS support for SSE passed.\n" );
 211    } else {
 212       message( "Tests of OS support for SSE failed!\n" );
 213    }
 214 #else
 215    /* We can't use POSIX signal handling to test the availability of
 216     * SSE, so we disable it by default.
 217     */
 218    message( "Cannot test OS support for SSE, disabling to be safe.\n" );
 219    _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM);
 220 #endif /* _POSIX_SOURCE && X86_FXSR_MAGIC */
 221 #elif defined(__FreeBSD__)
 222    {
 223       int ret, len, enabled;
 224       len = sizeof(enabled);
 225       ret = sysctlbyname("hw.instruction_sse", &enabled, &len, NULL, 0);
 226       if (ret || !enabled)
 227          _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM);
 228    }
 229 #else
 230    /* Do nothing on other platforms for now.
 231     */
 232    message( "Not testing OS support for SSE, leaving enabled.\n" );
 233 #endif /* __linux__ */
 234 }
 235
 236 #endif /* USE_SSE_ASM */
 237
 238
 239 void _mesa_init_all_x86_transform_asm( void )
 240 {
 241    (void) message; /* silence warning */
 242 #ifdef USE_X86_ASM
 243    _mesa_x86_cpu_features = _mesa_identify_x86_cpu_features();
 244
 245    if ( getenv( "MESA_NO_ASM" ) ) {
 246       _mesa_x86_cpu_features = 0;
 247    }
 248
 249    if ( _mesa_x86_cpu_features ) {
 250       _mesa_init_x86_transform_asm();
 251    }
 252
 253 #ifdef USE_MMX_ASM
 254    if ( cpu_has_mmx ) {
 255       if ( getenv( "MESA_NO_MMX" ) == 0 ) {
 256          message( "MMX cpu detected.\n" );
 257       } else {
 258          _mesa_x86_cpu_features &= ~(X86_FEATURE_MMX);
 259       }
 260    }
 261 #endif
 262
 263 #ifdef USE_3DNOW_ASM
 264    if ( cpu_has_3dnow ) {
 265       if ( getenv( "MESA_NO_3DNOW" ) == 0 ) {
 266          message( "3DNow! cpu detected.\n" );
 267          _mesa_init_3dnow_transform_asm();
 268       } else {
 269          _mesa_x86_cpu_features &= ~(X86_FEATURE_3DNOW);
 270       }
 271    }
 272 #endif
 273
 274 #ifdef USE_SSE_ASM
 275    if ( cpu_has_xmm && getenv( "MESA_FORCE_SSE" ) == 0 ) {
 276       check_os_sse_support();
 277    }
 278    if ( cpu_has_xmm ) {
 279       if ( getenv( "MESA_NO_SSE" ) == 0 ) {
 280          message( "SSE cpu detected.\n" );
 281          _mesa_init_sse_transform_asm();
 282       } else {
 283          _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM);
 284       }
 285    }
 286 #endif
 287 #endif
 288 }
 289