Debian Codegraph @

src/libmpg123/getcpuflags.h

raw
1 /*
2 getcpucpuflags: get cpuflags for ia32
3
4 copyright ?-2007 by the mpg123 project - free software under the terms of the LGPL 2.1
5 see COPYING and AUTHORS files in distribution or http:#mpg123.org
6 initially written by KIMURA Takuhiro (for 3DNow!)
7 extended for general use by Thomas Orgis
8 */
9
10 #ifndef MPG123_H_GETCPUFLAGS
11 #define MPG123_H_GETCPUFLAGS
12
13 #include "config.h"
14
15 /* standard level flags part 1 (ECX)*/
16 #define FLAG_SSE3 0x00000001
17 #define FLAG_SSSE3 0x00000200
18 #define FLAG_AVX 0x1C000000
19 /* standard level flags part 2 (EDX) */
20 #define FLAG2_MMX 0x00800000
21 #define FLAG2_SSE 0x02000000
22 #define FLAG2_SSE2 0x04000000
23 #define FLAG2_FPU 0x00000001
24 /* cpuid extended level 1 (AMD) */
25 #define XFLAG_MMX 0x00800000
26 #define XFLAG_3DNOW 0x80000000
27 #define XFLAG_3DNOWEXT 0x40000000
28 /* eXtended Control Register 0 */
29 #define XCR0FLAG_AVX 0x00000006
30
31
32 struct cpuflags
33 {
34 #if defined(OPT_ARM) || defined(OPT_NEON) || defined(OPT_NEON64)
35 unsigned int has_neon;
36 #else
37 unsigned int id;
38 unsigned int std;
39 unsigned int std2;
40 unsigned int ext;
41 unsigned int xcr0_lo;
42 #endif
43 };
44
45 unsigned int INT123_getcpuflags(struct cpuflags* cf);
46
47 #if ((defined OPT_X86) || (defined OPT_X86_64) || (defined OPT_NEON) || (defined OPT_NEON64)) && (defined OPT_MULTI)
48
49 #ifdef WANT_GETCPUFLAGS
50 #include <string.h>
51 // Wrapper needed for ignorant clang memory sanitizer that chokes
52 // because it does not know the assembly code intialized the vlaues.
53 static unsigned int wrap_getcpuflags(struct cpuflags* cf)
54 {
55 memset(cf, 0, sizeof(*cf));
56 return INT123_getcpuflags(cf);
57 }
58 #endif
59
60 // We really evaluate the CPU flags.
61 #define OPT_CPU_FLAGS
62
63 /* checks the family */
64 #define cpu_i586(s) ( ((s.id & 0xf00)>>8) == 0 || ((s.id & 0xf00)>>8) > 4 )
65 /* checking some flags... */
66 #define cpu_fpu(s) (FLAG2_FPU & s.std2)
67 #define cpu_mmx(s) (FLAG2_MMX & s.std2 || XFLAG_MMX & s.ext)
68 #define cpu_3dnow(s) (XFLAG_3DNOW & s.ext)
69 #define cpu_3dnowext(s) (XFLAG_3DNOWEXT & s.ext)
70 #define cpu_sse(s) (FLAG2_SSE & s.std2)
71 #define cpu_sse2(s) (FLAG2_SSE2 & s.std2)
72 #define cpu_sse3(s) (FLAG_SSE3 & s.std)
73 #define cpu_avx(s) ((FLAG_AVX & s.std) == FLAG_AVX && (XCR0FLAG_AVX & s.xcr0_lo) == XCR0FLAG_AVX)
74 #define cpu_fast_sse(s) ((((s.id & 0xf00)>>8) == 6 && FLAG_SSSE3 & s.std) /* for Intel/VIA; family 6 CPUs with SSSE3 */ || \
75 (((s.id & 0xf00)>>8) == 0xf && (((s.id & 0x0ff00000)>>20) > 0 && ((s.id & 0x0ff00000)>>20) != 5))) /* for AMD; family > 0xF CPUs except Bobcat */
76 #define cpu_neon(s) (s.has_neon)
77
78 #else
79
80 /* Faking stuff for non-multi builds. The same code for synth function choice is used.
81 Just no runtime dependency of result... */
82 #define cpu_flags nothing
83 #define cpu_i586(s) 1
84 #define cpu_fpu(s) 1
85 #define cpu_mmx(s) 1
86 #define cpu_3dnow(s) 1
87 #define cpu_3dnowext(s) 1
88 #define cpu_sse(s) 1
89 #define cpu_sse2(s) 1
90 #define cpu_sse3(s) 1
91 #define cpu_avx(s) 1
92 #define cpu_neon(s) 1
93
94 #endif
95
96
97 #endif