cpu.c 6.84 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 7 8
 * This file is part of FFmpeg.
 *
 * FFmpeg 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
 * FFmpeg 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 FFmpeg; 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

26
#include "libavutil/x86/asm.h"
27
#include "libavutil/x86/cpu.h"
28
#include "libavutil/cpu.h"
29

30 31 32 33 34 35 36 37 38 39
#if HAVE_YASM

#define cpuid(index, eax, ebx, ecx, edx)        \
    ff_cpu_cpuid(index, &eax, &ebx, &ecx, &edx)

#define xgetbv(index, eax, edx)                 \
    ff_cpu_xgetbv(index, &eax, &edx)

#elif HAVE_INLINE_ASM

40
/* ebx saving is necessary for PIC. gcc seems unable to see it alone */
41 42 43 44 45 46 47 48 49
#define cpuid(index, eax, ebx, ecx, edx)                        \
    __asm__ volatile (                                          \
        "mov    %%"REG_b", %%"REG_S" \n\t"                      \
        "cpuid                       \n\t"                      \
        "xchg   %%"REG_b", %%"REG_S                             \
        : "=a" (eax), "=S" (ebx), "=c" (ecx), "=d" (edx)        \
        : "0" (index))

#define xgetbv(index, eax, edx)                                 \
50
    __asm__ (".byte 0x0f, 0x01, 0xd0" : "=a"(eax), "=d"(edx) : "c" (index))
51

52 53 54 55 56 57 58 59 60 61
#define get_eflags(x)                           \
    __asm__ volatile ("pushfl     \n"           \
                      "pop    %0  \n"           \
                      : "=r"(x))

#define set_eflags(x)                           \
    __asm__ volatile ("push    %0 \n"           \
                      "popfl      \n"           \
                      :: "r"(x))

62 63
#endif /* HAVE_INLINE_ASM */

64 65 66
#if ARCH_X86_64

#define cpuid_test() 1
67

68 69 70 71
#elif HAVE_YASM

#define cpuid_test ff_cpu_cpuid_test

72
#elif HAVE_INLINE_ASM
73 74 75

static int cpuid_test(void)
{
76
    x86_reg a, c;
77

78 79 80 81 82
    /* Check if CPUID is supported by attempting to toggle the ID bit in
     * the EFLAGS register. */
    get_eflags(a);
    set_eflags(a ^ 0x200000);
    get_eflags(c);
83

84 85
    return a != c;
}
86
#endif
Fabrice Bellard's avatar
Fabrice Bellard committed
87

88 89 90 91
/* Function to test if multimedia instructions are supported...  */
int ff_get_cpu_flags_x86(void)
{
    int rval = 0;
92 93 94

#ifdef cpuid

95 96 97 98 99 100 101 102
    int eax, ebx, ecx, edx;
    int max_std_level, max_ext_level, std_caps = 0, ext_caps = 0;
    int family = 0, model = 0;
    union { int i[3]; char c[12]; } vendor;

    if (!cpuid_test())
        return 0; /* CPUID not supported */

103
    cpuid(0, max_std_level, vendor.i[0], vendor.i[2], vendor.i[1]);
104

105
    if (max_std_level >= 1) {
106
        cpuid(1, eax, ebx, ecx, std_caps);
107 108
        family = ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff);
        model  = ((eax >> 4) & 0xf) + ((eax >> 12) & 0xf0);
109
        if (std_caps & (1 << 15))
110
            rval |= AV_CPU_FLAG_CMOV;
111
        if (std_caps & (1 << 23))
112
            rval |= AV_CPU_FLAG_MMX;
113
        if (std_caps & (1 << 25))
114
            rval |= AV_CPU_FLAG_MMXEXT;
115
#if HAVE_SSE
116 117 118
        if (std_caps & (1 << 25))
            rval |= AV_CPU_FLAG_SSE;
        if (std_caps & (1 << 26))
119
            rval |= AV_CPU_FLAG_SSE2;
120
        if (ecx & 1)
121
            rval |= AV_CPU_FLAG_SSE3;
122
        if (ecx & 0x00000200 )
123
            rval |= AV_CPU_FLAG_SSSE3;
124
        if (ecx & 0x00080000 )
125
            rval |= AV_CPU_FLAG_SSE4;
126
        if (ecx & 0x00100000 )
127
            rval |= AV_CPU_FLAG_SSE42;
128 129 130 131 132 133 134 135
#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;
        }
136 137
#endif /* HAVE_AVX */
#endif /* HAVE_SSE */
138 139 140 141
    }

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

142
    if (max_ext_level >= 0x80000001) {
143
        cpuid(0x80000001, eax, ebx, ecx, ext_caps);
144
        if (ext_caps & (1U << 31))
145
            rval |= AV_CPU_FLAG_3DNOW;
146
        if (ext_caps & (1 << 30))
147
            rval |= AV_CPU_FLAG_3DNOWEXT;
148
        if (ext_caps & (1 << 23))
149
            rval |= AV_CPU_FLAG_MMX;
150
        if (ext_caps & (1 << 22))
151
            rval |= AV_CPU_FLAG_MMXEXT;
152 153 154 155 156 157 158 159 160 161 162 163

        /* 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;
        }
164 165 166 167 168 169 170 171 172

        /* XOP and FMA4 use the AVX instruction coding scheme, so they can't be
         * used unless the OS has AVX support. */
        if (rval & AV_CPU_FLAG_AVX) {
            if (ecx & 0x00000800)
                rval |= AV_CPU_FLAG_XOP;
            if (ecx & 0x00010000)
                rval |= AV_CPU_FLAG_FMA4;
        }
Fabrice Bellard's avatar
Fabrice Bellard committed
173
    }
174

175 176
    if (!strncmp(vendor.c, "GenuineIntel", 12)) {
        if (family == 6 && (model == 9 || model == 13 || model == 14)) {
177 178 179 180 181 182 183 184 185 186 187
            /* 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;
188 189 190 191 192 193 194 195
        }
        /* 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;
196 197
    }

198 199
#endif /* cpuid */

200
    return rval;
Fabrice Bellard's avatar
Fabrice Bellard committed
201
}