111
|
1 /* Check if system supports SIMD */
|
|
2 #include <signal.h>
|
|
3
|
|
4 #if defined(__i386__) || defined(__x86_64__)
|
|
5 # include "cpuid.h"
|
|
6 #endif
|
|
7
|
|
8 #ifdef __cplusplus
|
|
9 extern "C" {
|
|
10 #endif
|
|
11 extern void abort (void);
|
|
12 extern void exit (int);
|
|
13 #ifdef __cplusplus
|
|
14 }
|
|
15 #endif
|
|
16
|
|
17 static void
|
|
18 sig_ill_handler (int sig)
|
|
19 {
|
|
20 exit(0);
|
|
21 }
|
|
22
|
|
23 static void __attribute__((noinline))
|
|
24 check_vect (void)
|
|
25 {
|
|
26 signal(SIGILL, sig_ill_handler);
|
|
27 #if defined(__PAIRED__)
|
|
28 /* 750CL paired-single instruction, 'ps_mul %v0,%v0,%v0'. */
|
|
29 asm volatile (".long 0x10000032");
|
|
30 #elif defined(__ppc__) || defined(__ppc64__) || defined(__powerpc__) || defined(powerpc)
|
|
31 /* Altivec instruction, 'vor %v0,%v0,%v0'. */
|
|
32 asm volatile (".long 0x10000484");
|
|
33 #elif defined(__i386__) || defined(__x86_64__)
|
|
34 {
|
|
35 unsigned int a, b, c, d,
|
|
36 want_level, want_b = 0, want_c = 0, want_d = 0;
|
|
37
|
|
38 /* Determine what instruction set we've been compiled for, and detect
|
|
39 that we're running with it. This allows us to at least do a compile
|
|
40 check for, e.g. SSE4.1 when the machine only supports SSE2. */
|
|
41 # if defined(__AVX2__)
|
|
42 want_level = 7, want_b = bit_AVX2;
|
|
43 # elif defined(__AVX__)
|
|
44 want_level = 1, want_c = bit_AVX;
|
|
45 # elif defined(__SSE4_2__)
|
|
46 want_level = 1, want_c = bit_SSE4_2;
|
|
47 # elif defined(__SSE4_1__)
|
|
48 want_level = 1, want_c = bit_SSE4_1;
|
|
49 # elif defined(__SSSE3__)
|
|
50 want_level = 1, want_c = bit_SSSE3;
|
|
51 # else
|
|
52 want_level = 1, want_d = bit_SSE2;
|
|
53 # endif
|
|
54
|
|
55 if (!__get_cpuid (want_level, &a, &b, &c, &d)
|
|
56 || ((b & want_b) | (c & want_c) | (d & want_d)) == 0)
|
|
57 exit (0);
|
|
58 }
|
|
59 #elif defined(__sparc__)
|
|
60 asm volatile (".word\t0x81b007c0");
|
|
61 #elif defined(__arm__)
|
|
62 {
|
|
63 /* On some processors without NEON support, this instruction may
|
|
64 be a no-op, on others it may trap, so check that it executes
|
|
65 correctly. */
|
|
66 long long a = 0, b = 1;
|
|
67 asm ("vorr %P0, %P1, %P2"
|
|
68 : "=w" (a)
|
|
69 : "0" (a), "w" (b));
|
|
70 if (a != 1)
|
|
71 exit (0);
|
|
72 }
|
|
73 #elif defined(__mips_msa)
|
|
74 asm volatile ("or.v $w0,$w0,$w0");
|
|
75 #endif
|
|
76 signal (SIGILL, SIG_DFL);
|
|
77 }
|
131
|
78
|
|
79 #if defined (__ARM_FEATURE_SVE)
|
|
80 # if __ARM_FEATURE_SVE_BITS == 0
|
|
81 # define VECTOR_BITS 1024
|
|
82 # else
|
|
83 # define VECTOR_BITS __ARM_FEATURE_SVE_BITS
|
|
84 # endif
|
|
85 #else
|
|
86 # define VECTOR_BITS 128
|
|
87 #endif
|