TOOLS/fastmemcpybench.c

   1 /*
   2  * benchmark tool for fast_memcpy code from libvo
   3  *
   4  * NOTE: This code can not be used on Pentium MMX / II because they contain
   5  * a bug in rdtsc. For Intel processors since P6(PII) rdpmc should be used
   6  * instead. For PIII it's disputable and it seems the bug was fixed but this
   7  * was not confirmed through testing.
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write to the Free Software Foundation, Inc.,
  21  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  22  */
  23
  24 #include <stdio.h>
  25 #include <stdlib.h>
  26 #include <string.h>
  27 #include <sys/ioctl.h>
  28 #include <unistd.h>
  29 #include <fcntl.h>
  30 #include <sys/mman.h>
  31 #include <sys/time.h>
  32 #include <inttypes.h>
  33
  34 #include "config.h"
  35 #include "cpudetect.h"
  36
  37 #define BLOCK_SIZE 4096
  38 #define CONFUSION_FACTOR 0
  39
  40 #if HAVE_MMX
  41 #define COMPILE_MMX
  42 #endif
  43
  44 #if HAVE_MMX2
  45 #define COMPILE_MMX2
  46 #endif
  47
  48 #if HAVE_AMD3DNOW
  49 #define COMPILE_AMD3DNOW
  50 #endif
  51
  52 #if HAVE_SSE
  53 #define COMPILE_SSE
  54 #endif
  55
  56 #ifdef COMPILE_MMX
  57 #undef RENAME
  58 #undef HAVE_MMX
  59 #undef HAVE_MMX2
  60 #undef HAVE_AMD3DNOW
  61 #undef HAVE_SSE
  62 #undef HAVE_SSE2
  63 #define HAVE_MMX 1
  64 #define HAVE_MMX2 0
  65 #define HAVE_AMD3DNOW 0
  66 #define HAVE_SSE 0
  67 #define HAVE_SSE2 0
  68 #define RENAME(a) a ## _MMX
  69 #include "libvo/aclib_template.c"
  70 #endif
  71
  72 #ifdef COMPILE_MMX2
  73 #undef RENAME
  74 #undef HAVE_MMX
  75 #undef HAVE_MMX2
  76 #undef HAVE_AMD3DNOW
  77 #undef HAVE_SSE
  78 #undef HAVE_SSE2
  79 #define HAVE_MMX 1
  80 #define HAVE_MMX2 1
  81 #define HAVE_AMD3DNOW 0
  82 #define HAVE_SSE 0
  83 #define HAVE_SSE2 0
  84 #define RENAME(a) a ## _MMX2
  85 #include "libvo/aclib_template.c"
  86 #endif
  87
  88 #ifdef COMPILE_AMD3DNOW
  89 #undef RENAME
  90 #undef HAVE_MMX
  91 #undef HAVE_MMX2
  92 #undef HAVE_AMD3DNOW
  93 #undef HAVE_SSE
  94 #undef HAVE_SSE2
  95 #define HAVE_MMX 1
  96 #define HAVE_MMX2 0
  97 #define HAVE_AMD3DNOW 1
  98 #define HAVE_SSE 0
  99 #define HAVE_SSE2 0
 100 #define RENAME(a) a ## _3DNow
 101 #include "libvo/aclib_template.c"
 102 #endif
 103
 104 #ifdef COMPILE_SSE
 105 #undef RENAME
 106 #undef HAVE_MMX
 107 #undef HAVE_MMX2
 108 #undef HAVE_AMD3DNOW
 109 #undef HAVE_SSE
 110 #undef HAVE_SSE2
 111 #define HAVE_MMX 1
 112 #define HAVE_MMX2 1
 113 #define HAVE_AMD3DNOW 0
 114 #define HAVE_SSE 1
 115 #define HAVE_SSE2 1
 116 #define RENAME(a) a ## _SSE
 117 #include "libvo/aclib_template.c"
 118 #endif
 119
 120 //#define ARR_SIZE 100000
 121 #define ARR_SIZE (1024*768*2)
 122
 123 #ifdef CONFIG_MGA
 124
 125 #include "drivers/mga_vid.h"
 126
 127 static mga_vid_config_t mga_vid_config;
 128 static unsigned char* frame = NULL;
 129 static int f;
 130
 131 static int mga_init(void)
 132 {
 133     f = open("/dev/mga_vid", O_RDWR);
 134     if (f == -1) {
 135         fprintf(stderr, "Couldn't open /dev/mga_vid.\n");
 136         return -1;
 137     }
 138
 139     mga_vid_config.num_frames = 1;
 140     mga_vid_config.frame_size = ARR_SIZE;
 141     mga_vid_config.format     = MGA_VID_FORMAT_YUY2;
 142
 143     mga_vid_config.colkey_on   =   0;
 144     mga_vid_config.src_width   = 640;
 145     mga_vid_config.src_height  = 480;
 146     mga_vid_config.dest_width  = 320;
 147     mga_vid_config.dest_height = 200;
 148     mga_vid_config.x_org       =   0;
 149     mga_vid_config.y_org       =   0;
 150
 151     mga_vid_config.version     = MGA_VID_VERSION;
 152     if (ioctl(f, MGA_VID_CONFIG, &mga_vid_config)) {
 153         perror("Error in mga_vid_config ioctl()");
 154         printf("Your mga_vid driver version is incompatible with this MPlayer version!\n");
 155         exit(1);
 156     }
 157     ioctl(f, MGA_VID_ON, 0);
 158
 159     frame = (char*)mmap(0, mga_vid_config.frame_size*mga_vid_config.num_frames,
 160                         PROT_WRITE,MAP_SHARED, f, 0);
 161     if (!frame) {
 162         printf("Can't mmap MGA frame.\n");
 163         exit(1);
 164     }
 165
 166     //clear the buffer
 167     //memset(frames[0], 0x80, mga_vid_config.frame_size*mga_vid_config.num_frames);
 168
 169     return 0;
 170 }
 171
 172 #endif
 173
 174 // Returns current time in microseconds
 175 static unsigned int GetTimer(void)
 176 {
 177     struct timeval tv;
 178     struct timezone tz;
 179     //float s;
 180     gettimeofday(&tv, &tz);
 181     //s = tv.tv_usec; s *= 0.000001; s += tv.tv_sec;
 182     return tv.tv_sec * 1000000 + tv.tv_usec;
 183 }
 184
 185 static inline unsigned long long int read_tsc(void)
 186 {
 187     unsigned long long int retval;
 188     __asm__ volatile ("rdtsc":"=A" (retval)::"memory");
 189     return retval;
 190 }
 191
 192 unsigned char __attribute__((aligned(4096)))arr1[ARR_SIZE], arr2[ARR_SIZE];
 193
 194 int main(void)
 195 {
 196     unsigned long long int v1, v2;
 197     unsigned char *marr1, *marr2;
 198     int i;
 199     unsigned int t;
 200 #ifdef CONFIG_MGA
 201     mga_init();
 202     marr1 = &frame[3];
 203 #else
 204     marr1 = &arr1[0];
 205 #endif
 206     marr2 = &arr2[0];
 207
 208     for (i = 0; i < ARR_SIZE - 16; i++)
 209         marr1[i] = marr2[i] = i;
 210
 211 #define testblock(func, name)                                                \
 212     t  = GetTimer();                                                         \
 213     v1 = read_tsc();                                                         \
 214     for (i = 0; i < 100; i++)                                                \
 215         func(marr1, marr2, ARR_SIZE - 16);                                   \
 216     v2 = read_tsc();                                                         \
 217     t  = GetTimer() - t;                                                     \
 218     /* ARR_SIZE*100 / (1024*1024) / (t/1000000) = ARR_SIZE*95.36743 / t */   \
 219     printf(name "CPU clocks=%llu = %dus  (%5.3ffps)  %5.1fMB/s\n", v2-v1, t, \
 220            100000000.0f / (float)t, (float)ARR_SIZE*95.36743f / (float)t);
 221
 222     testblock(memcpy, "libc:   ");
 223
 224 #if HAVE_MMX
 225     testblock(fast_memcpy_MMX, "MMX:    ");
 226 #endif
 227
 228 #if HAVE_AMD3DNOW
 229     testblock(fast_memcpy_3DNow, "3DNow!: ");
 230 #endif
 231
 232 #if HAVE_MMX2
 233     testblock(fast_memcpy_MMX2, "MMX2:   ");
 234 #endif
 235
 236 #if HAVE_SSE
 237     testblock(fast_memcpy_SSE, "SSE:    ");
 238 #endif
 239
 240     return 0;
 241 }