michael@0: /*
michael@0:  * Copyright © 2000 SuSE, Inc.
michael@0:  * Copyright © 2007 Red Hat, Inc.
michael@0:  *
michael@0:  * Permission to use, copy, modify, distribute, and sell this software and its
michael@0:  * documentation for any purpose is hereby granted without fee, provided that
michael@0:  * the above copyright notice appear in all copies and that both that
michael@0:  * copyright notice and this permission notice appear in supporting
michael@0:  * documentation, and that the name of SuSE not be used in advertising or
michael@0:  * publicity pertaining to distribution of the software without specific,
michael@0:  * written prior permission.  SuSE makes no representations about the
michael@0:  * suitability of this software for any purpose.  It is provided "as is"
michael@0:  * without express or implied warranty.
michael@0:  *
michael@0:  * SuSE DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL
michael@0:  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL SuSE
michael@0:  * BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
michael@0:  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
michael@0:  * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
michael@0:  * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
michael@0:  */
michael@0: #ifdef HAVE_CONFIG_H
michael@0: #include <config.h>
michael@0: #endif
michael@0: 
michael@0: #include "pixman-private.h"
michael@0: 
michael@0: #if defined(USE_X86_MMX) || defined (USE_SSE2)
michael@0: 
michael@0: /* The CPU detection code needs to be in a file not compiled with
michael@0:  * "-mmmx -msse", as gcc would generate CMOV instructions otherwise
michael@0:  * that would lead to SIGILL instructions on old CPUs that don't have
michael@0:  * it.
michael@0:  */
michael@0: 
michael@0: typedef enum
michael@0: {
michael@0:     X86_MMX			= (1 << 0),
michael@0:     X86_MMX_EXTENSIONS		= (1 << 1),
michael@0:     X86_SSE			= (1 << 2) | X86_MMX_EXTENSIONS,
michael@0:     X86_SSE2			= (1 << 3),
michael@0:     X86_CMOV			= (1 << 4)
michael@0: } cpu_features_t;
michael@0: 
michael@0: #ifdef HAVE_GETISAX
michael@0: 
michael@0: #include <sys/auxv.h>
michael@0: 
michael@0: static cpu_features_t
michael@0: detect_cpu_features (void)
michael@0: {
michael@0:     cpu_features_t features = 0;
michael@0:     unsigned int result = 0;
michael@0: 
michael@0:     if (getisax (&result, 1))
michael@0:     {
michael@0: 	if (result & AV_386_CMOV)
michael@0: 	    features |= X86_CMOV;
michael@0: 	if (result & AV_386_MMX)
michael@0: 	    features |= X86_MMX;
michael@0: 	if (result & AV_386_AMD_MMX)
michael@0: 	    features |= X86_MMX_EXTENSIONS;
michael@0: 	if (result & AV_386_SSE)
michael@0: 	    features |= X86_SSE;
michael@0: 	if (result & AV_386_SSE2)
michael@0: 	    features |= X86_SSE2;
michael@0:     }
michael@0: 
michael@0:     return features;
michael@0: }
michael@0: 
michael@0: #else
michael@0: 
michael@0: #define _PIXMAN_X86_64							\
michael@0:     (defined(__amd64__) || defined(__x86_64__) || defined(_M_AMD64))
michael@0: 
michael@0: static pixman_bool_t
michael@0: have_cpuid (void)
michael@0: {
michael@0: #if _PIXMAN_X86_64 || defined (_MSC_VER)
michael@0: 
michael@0:     return TRUE;
michael@0: 
michael@0: #elif defined (__GNUC__)
michael@0:     uint32_t result;
michael@0: 
michael@0:     __asm__ volatile (
michael@0:         "pushf"				"\n\t"
michael@0:         "pop %%eax"			"\n\t"
michael@0:         "mov %%eax, %%ecx"		"\n\t"
michael@0:         "xor $0x00200000, %%eax"	"\n\t"
michael@0:         "push %%eax"			"\n\t"
michael@0:         "popf"				"\n\t"
michael@0:         "pushf"				"\n\t"
michael@0:         "pop %%eax"			"\n\t"
michael@0:         "xor %%ecx, %%eax"		"\n\t"
michael@0: 	"mov %%eax, %0"			"\n\t"
michael@0: 	: "=r" (result)
michael@0: 	:
michael@0: 	: "%eax", "%ecx");
michael@0: 
michael@0:     return !!result;
michael@0: 
michael@0: #else
michael@0: #error "Unknown compiler"
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static void
michael@0: pixman_cpuid (uint32_t feature,
michael@0: 	      uint32_t *a, uint32_t *b, uint32_t *c, uint32_t *d)
michael@0: {
michael@0: #if defined (__GNUC__)
michael@0: 
michael@0: #if _PIXMAN_X86_64
michael@0:     __asm__ volatile (
michael@0:         "cpuid"				"\n\t"
michael@0: 	: "=a" (*a), "=b" (*b), "=c" (*c), "=d" (*d)
michael@0: 	: "a" (feature));
michael@0: #else
michael@0:     /* On x86-32 we need to be careful about the handling of %ebx
michael@0:      * and %esp. We can't declare either one as clobbered
michael@0:      * since they are special registers (%ebx is the "PIC
michael@0:      * register" holding an offset to global data, %esp the
michael@0:      * stack pointer), so we need to make sure that %ebx is
michael@0:      * preserved, and that %esp has its original value when
michael@0:      * accessing the output operands.
michael@0:      */
michael@0:     __asm__ volatile (
michael@0: 	"xchg %%ebx, %1"		"\n\t"
michael@0: 	"cpuid"				"\n\t"
michael@0: 	"xchg %%ebx, %1"		"\n\t"
michael@0: 	: "=a" (*a), "=r" (*b), "=c" (*c), "=d" (*d)
michael@0: 	: "a" (feature));
michael@0: #endif
michael@0: 
michael@0: #elif defined (_MSC_VER)
michael@0:     int info[4];
michael@0: 
michael@0:     __cpuid (info, feature);
michael@0: 
michael@0:     *a = info[0];
michael@0:     *b = info[1];
michael@0:     *c = info[2];
michael@0:     *d = info[3];
michael@0: #else
michael@0: #error Unknown compiler
michael@0: #endif
michael@0: }
michael@0: 
michael@0: static cpu_features_t
michael@0: detect_cpu_features (void)
michael@0: {
michael@0:     uint32_t a, b, c, d;
michael@0:     cpu_features_t features = 0;
michael@0: 
michael@0:     if (!have_cpuid())
michael@0: 	return features;
michael@0: 
michael@0:     /* Get feature bits */
michael@0:     pixman_cpuid (0x01, &a, &b, &c, &d);
michael@0:     if (d & (1 << 15))
michael@0: 	features |= X86_CMOV;
michael@0:     if (d & (1 << 23))
michael@0: 	features |= X86_MMX;
michael@0:     if (d & (1 << 25))
michael@0: 	features |= X86_SSE;
michael@0:     if (d & (1 << 26))
michael@0: 	features |= X86_SSE2;
michael@0: 
michael@0:     /* Check for AMD specific features */
michael@0:     if ((features & X86_MMX) && !(features & X86_SSE))
michael@0:     {
michael@0: 	char vendor[13];
michael@0: 
michael@0: 	/* Get vendor string */
michael@0: 	memset (vendor, 0, sizeof vendor);
michael@0: 
michael@0: 	pixman_cpuid (0x00, &a, &b, &c, &d);
michael@0: 	memcpy (vendor + 0, &b, 4);
michael@0: 	memcpy (vendor + 4, &d, 4);
michael@0: 	memcpy (vendor + 8, &c, 4);
michael@0: 
michael@0: 	if (strcmp (vendor, "AuthenticAMD") == 0 ||
michael@0: 	    strcmp (vendor, "Geode by NSC") == 0)
michael@0: 	{
michael@0: 	    pixman_cpuid (0x80000000, &a, &b, &c, &d);
michael@0: 	    if (a >= 0x80000001)
michael@0: 	    {
michael@0: 		pixman_cpuid (0x80000001, &a, &b, &c, &d);
michael@0: 
michael@0: 		if (d & (1 << 22))
michael@0: 		    features |= X86_MMX_EXTENSIONS;
michael@0: 	    }
michael@0: 	}
michael@0:     }
michael@0: 
michael@0:     return features;
michael@0: }
michael@0: 
michael@0: #endif
michael@0: 
michael@0: static pixman_bool_t
michael@0: have_feature (cpu_features_t feature)
michael@0: {
michael@0:     static pixman_bool_t initialized;
michael@0:     static cpu_features_t features;
michael@0: 
michael@0:     if (!initialized)
michael@0:     {
michael@0: 	features = detect_cpu_features();
michael@0: 	initialized = TRUE;
michael@0:     }
michael@0: 
michael@0:     return (features & feature) == feature;
michael@0: }
michael@0: 
michael@0: #endif
michael@0: 
michael@0: pixman_implementation_t *
michael@0: _pixman_x86_get_implementations (pixman_implementation_t *imp)
michael@0: {
michael@0: #define MMX_BITS  (X86_MMX | X86_MMX_EXTENSIONS)
michael@0: #define SSE2_BITS (X86_MMX | X86_MMX_EXTENSIONS | X86_SSE | X86_SSE2)
michael@0: 
michael@0: #ifdef USE_X86_MMX
michael@0:     if (!_pixman_disabled ("mmx") && have_feature (MMX_BITS))
michael@0: 	imp = _pixman_implementation_create_mmx (imp);
michael@0: #endif
michael@0: 
michael@0: #ifdef USE_SSE2
michael@0:     if (!_pixman_disabled ("sse2") && have_feature (SSE2_BITS))
michael@0: 	imp = _pixman_implementation_create_sse2 (imp);
michael@0: #endif
michael@0: 
michael@0:     return imp;
michael@0: }