cpu.c 5.66 KB
Newer Older
1 2 3 4 5
/*
 * CPU detection code, extracted from mmx.h
 * (c)1997-99 by H. Dietz and R. Fisher
 * Converted to C and improved by Fabrice Bellard.
 *
6
 * This file is part of Libav.
7
 *
8
 * Libav is free software; you can redistribute it and/or
9 10
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * Libav is distributed in the hope that it will be useful,
14 15 16 17 18
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with Libav; if not, write to the Free Software
20 21
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */
Fabrice Bellard's avatar
Fabrice Bellard committed
22 23

#include <stdlib.h>
24
#include <string.h>
25
#include "libavutil/x86_cpu.h"
26
#include "libavutil/cpu.h"
27

28 29
/* ebx saving is necessary for PIC. gcc seems unable to see it alone */
#define cpuid(index,eax,ebx,ecx,edx)\
30
    __asm__ volatile\
31
        ("mov %%"REG_b", %%"REG_S"\n\t"\
32
         "cpuid\n\t"\
33
         "xchg %%"REG_b", %%"REG_S\
34 35
         : "=a" (eax), "=S" (ebx),\
           "=c" (ecx), "=d" (edx)\
Nick Kurshev's avatar
Nick Kurshev committed
36
         : "0" (index));
Fabrice Bellard's avatar
Fabrice Bellard committed
37

Mans Rullgard's avatar
Mans Rullgard committed
38
#define xgetbv(index,eax,edx)                                   \
39
    __asm__ (".byte 0x0f, 0x01, 0xd0" : "=a"(eax), "=d"(edx) : "c" (index))
Mans Rullgard's avatar
Mans Rullgard committed
40

Fabrice Bellard's avatar
Fabrice Bellard committed
41
/* Function to test if multimedia instructions are supported...  */
Måns Rullgård's avatar
Måns Rullgård committed
42
int ff_get_cpu_flags_x86(void)
Fabrice Bellard's avatar
Fabrice Bellard committed
43
{
44
    int rval = 0;
Fabrice Bellard's avatar
Fabrice Bellard committed
45
    int eax, ebx, ecx, edx;
46
    int max_std_level, max_ext_level, std_caps=0, ext_caps=0;
47 48
    int family=0, model=0;
    union { int i[3]; char c[12]; } vendor;
49

50 51
#if ARCH_X86_32
    x86_reg a, c;
52
    __asm__ volatile (
53 54
        /* See if CPUID instruction is supported ... */
        /* ... Get copies of EFLAGS into eax and ecx */
55
        "pushfl\n\t"
56 57
        "pop %0\n\t"
        "mov %0, %1\n\t"
58

59 60 61 62
        /* ... Toggle the ID bit in one copy and store */
        /*     to the EFLAGS reg */
        "xor $0x200000, %0\n\t"
        "push %0\n\t"
63
        "popfl\n\t"
64

65
        /* ... Get the (hopefully modified) EFLAGS */
66
        "pushfl\n\t"
67 68 69 70 71
        "pop %0\n\t"
        : "=a" (a), "=c" (c)
        :
        : "cc"
        );
72

73
    if (a == c)
Fabrice Bellard's avatar
Fabrice Bellard committed
74
        return 0; /* CPUID not supported */
75
#endif
Fabrice Bellard's avatar
Fabrice Bellard committed
76

77
    cpuid(0, max_std_level, vendor.i[0], vendor.i[2], vendor.i[1]);
78 79 80

    if(max_std_level >= 1){
        cpuid(1, eax, ebx, ecx, std_caps);
81 82
        family = ((eax>>8)&0xf) + ((eax>>20)&0xff);
        model  = ((eax>>4)&0xf) + ((eax>>12)&0xf0);
83
        if (std_caps & (1<<23))
84
            rval |= AV_CPU_FLAG_MMX;
85
        if (std_caps & (1<<25))
86
            rval |= AV_CPU_FLAG_MMX2
87
#if HAVE_SSE
88
                  | AV_CPU_FLAG_SSE;
89
        if (std_caps & (1<<26))
90
            rval |= AV_CPU_FLAG_SSE2;
91
        if (ecx & 1)
92
            rval |= AV_CPU_FLAG_SSE3;
93
        if (ecx & 0x00000200 )
94
            rval |= AV_CPU_FLAG_SSSE3;
David Conrad's avatar
David Conrad committed
95
        if (ecx & 0x00080000 )
96
            rval |= AV_CPU_FLAG_SSE4;
David Conrad's avatar
David Conrad committed
97
        if (ecx & 0x00100000 )
98
            rval |= AV_CPU_FLAG_SSE42;
Mans Rullgard's avatar
Mans Rullgard committed
99 100 101 102 103 104 105 106 107
#if HAVE_AVX
        /* Check OXSAVE and AVX bits */
        if ((ecx & 0x18000000) == 0x18000000) {
            /* Check for OS support */
            xgetbv(0, eax, edx);
            if ((eax & 0x6) == 0x6)
                rval |= AV_CPU_FLAG_AVX;
        }
#endif
108 109
#endif
                  ;
110 111 112 113 114 115 116
    }

    cpuid(0x80000000, max_ext_level, ebx, ecx, edx);

    if(max_ext_level >= 0x80000001){
        cpuid(0x80000001, eax, ebx, ecx, ext_caps);
        if (ext_caps & (1<<31))
117
            rval |= AV_CPU_FLAG_3DNOW;
118
        if (ext_caps & (1<<30))
119
            rval |= AV_CPU_FLAG_3DNOWEXT;
120
        if (ext_caps & (1<<23))
121
            rval |= AV_CPU_FLAG_MMX;
122
        if (ext_caps & (1<<22))
123
            rval |= AV_CPU_FLAG_MMX2;
124 125 126 127 128 129 130 131 132 133 134 135

        /* Allow for selectively disabling SSE2 functions on AMD processors
           with SSE2 support but not SSE4a. This includes Athlon64, some
           Opteron, and some Sempron processors. MMX, SSE, or 3DNow! are faster
           than SSE2 often enough to utilize this special-case flag.
           AV_CPU_FLAG_SSE2 and AV_CPU_FLAG_SSE2SLOW are both set in this case
           so that SSE2 is used unless explicitly disabled by checking
           AV_CPU_FLAG_SSE2SLOW. */
        if (!strncmp(vendor.c, "AuthenticAMD", 12) &&
            rval & AV_CPU_FLAG_SSE2 && !(ecx & 0x00000040)) {
            rval |= AV_CPU_FLAG_SSE2SLOW;
        }
Fabrice Bellard's avatar
Fabrice Bellard committed
136
    }
137

138 139
    if (!strncmp(vendor.c, "GenuineIntel", 12)) {
        if (family == 6 && (model == 9 || model == 13 || model == 14)) {
Justin Ruggles's avatar
Justin Ruggles committed
140 141 142 143 144 145 146 147
            /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and 6/14 (core1 "yonah")
            * theoretically support sse2, but it's usually slower than mmx,
            * so let's just pretend they don't. AV_CPU_FLAG_SSE2 is disabled and
            * AV_CPU_FLAG_SSE2SLOW is enabled so that SSE2 is not used unless
            * explicitly enabled by checking AV_CPU_FLAG_SSE2SLOW. The same
            * situation applies for AV_CPU_FLAG_SSE3 and AV_CPU_FLAG_SSE3SLOW. */
            if (rval & AV_CPU_FLAG_SSE2) rval ^= AV_CPU_FLAG_SSE2SLOW|AV_CPU_FLAG_SSE2;
            if (rval & AV_CPU_FLAG_SSE3) rval ^= AV_CPU_FLAG_SSE3SLOW|AV_CPU_FLAG_SSE3;
148 149 150 151 152 153 154 155
        }
        /* The Atom processor has SSSE3 support, which is useful in many cases,
         * but sometimes the SSSE3 version is slower than the SSE2 equivalent
         * on the Atom, but is generally faster on other processors supporting
         * SSSE3. This flag allows for selectively disabling certain SSSE3
         * functions on the Atom. */
        if (family == 6 && model == 28)
            rval |= AV_CPU_FLAG_ATOM;
156 157
    }

158
    return rval;
Fabrice Bellard's avatar
Fabrice Bellard committed
159
}