Mercurial > mplayer.hg
annotate TOOLS/fastmemcpybench.c @ 27754:08d18fe9da52
Change all occurrences of asm and __asm to __asm__, same as was done for FFmpeg.
Neither variant is valid C99 syntax, but __asm__ is the most portable variant.
author | diego |
---|---|
date | Thu, 16 Oct 2008 18:59:27 +0000 |
parents | 13c920fdf7f5 |
children | b5a46071062a |
rev | line source |
---|---|
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
1 /* |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
2 fastmemcpybench.c used to benchmark fastmemcpy.h code from libvo. |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
3 |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
4 Note: this code can not be used on PentMMX-PII because they contain |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
5 a bug in rdtsc. For Intel processors since P6(PII) rdpmc should be used |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
6 instead. For PIII it's disputable and seems bug was fixed but I don't |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
7 tested it. |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
8 */ |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
9 |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
10 #include <stdio.h> |
572 | 11 #include <stdlib.h> |
12 #include <string.h> | |
13 #include <sys/ioctl.h> | |
14 #include <unistd.h> | |
15 #include <fcntl.h> | |
16 #include <sys/mman.h> | |
17 #include <sys/time.h> | |
13839 | 18 #include <inttypes.h> |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
19 |
572 | 20 //#define ARR_SIZE 100000 |
21 #define ARR_SIZE (1024*768*2) | |
22 | |
27365
13c920fdf7f5
Change a bunch of video-output-specific preprocessor directives from a HAVE_
diego
parents:
26759
diff
changeset
|
23 #ifdef CONFIG_MGA |
572 | 24 |
17017 | 25 #include "drivers/mga_vid.h" |
572 | 26 |
27 static mga_vid_config_t mga_vid_config; | |
28 static unsigned char* frame=NULL; | |
29 static int f; | |
30 | |
31 static int mga_init(){ | |
32 | |
33 f = open("/dev/mga_vid",O_RDWR); | |
34 if(f == -1) | |
35 { | |
36 fprintf(stderr,"Couldn't open /dev/mga_vid\n"); | |
26759
8eff880f638c
cosmetics: Remove useless parentheses from return statements.
diego
parents:
26575
diff
changeset
|
37 return -1; |
572 | 38 } |
39 | |
40 mga_vid_config.num_frames=1; | |
41 mga_vid_config.frame_size=ARR_SIZE; | |
42 mga_vid_config.format=MGA_VID_FORMAT_YUY2; | |
43 | |
44 mga_vid_config.colkey_on=0; | |
45 mga_vid_config.src_width = 640; | |
46 mga_vid_config.src_height= 480; | |
47 mga_vid_config.dest_width = 320; | |
48 mga_vid_config.dest_height= 200; | |
49 mga_vid_config.x_org= 0; | |
50 mga_vid_config.y_org= 0; | |
51 | |
52 mga_vid_config.version=MGA_VID_VERSION; | |
53 if (ioctl(f,MGA_VID_CONFIG,&mga_vid_config)) | |
54 { | |
55 perror("Error in mga_vid_config ioctl()"); | |
56 printf("Your mga_vid driver version is incompatible with this MPlayer version!\n"); | |
57 exit(1); | |
58 } | |
59 ioctl(f,MGA_VID_ON,0); | |
60 | |
61 frame = (char*)mmap(0,mga_vid_config.frame_size*mga_vid_config.num_frames,PROT_WRITE,MAP_SHARED,f,0); | |
62 if(!frame){ | |
63 printf("Can't mmap mga frame\n"); | |
64 exit(1); | |
65 } | |
66 | |
67 //clear the buffer | |
68 //memset(frames[0],0x80,mga_vid_config.frame_size*mga_vid_config.num_frames); | |
69 | |
70 return 0; | |
71 | |
72 } | |
73 | |
74 #endif | |
75 | |
76 // Returns current time in microseconds | |
26575
1ca484e74f18
Mark all functions that are only used within the file as static.
diego
parents:
24211
diff
changeset
|
77 static unsigned int GetTimer(){ |
572 | 78 struct timeval tv; |
79 struct timezone tz; | |
80 // float s; | |
81 gettimeofday(&tv,&tz); | |
82 // s=tv.tv_usec;s*=0.000001;s+=tv.tv_sec; | |
26759
8eff880f638c
cosmetics: Remove useless parentheses from return statements.
diego
parents:
26575
diff
changeset
|
83 return tv.tv_sec * 1000000 + tv.tv_usec; |
572 | 84 } |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
85 |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
86 static inline unsigned long long int read_tsc( void ) |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
87 { |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
88 unsigned long long int retval; |
27754
08d18fe9da52
Change all occurrences of asm and __asm to __asm__, same as was done for FFmpeg.
diego
parents:
27365
diff
changeset
|
89 __asm__ __volatile ("rdtsc":"=A"(retval)::"memory"); |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
90 return retval; |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
91 } |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
92 |
3076
bfc881c0e591
newly allocated memory seems to point to only 1 zero filled 4k page or something so there is a near 100% cache hit rate ... not very realistic, writeing something in the source array fixes that so the benchmark scores are meaningfull now
michael
parents:
687
diff
changeset
|
93 unsigned char __attribute__((aligned(4096)))arr1[ARR_SIZE],arr2[ARR_SIZE]; |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
94 |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
95 int main( void ) |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
96 { |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
97 unsigned long long int v1,v2; |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
98 unsigned char * marr1,*marr2; |
572 | 99 int i; |
100 unsigned int t; | |
27365
13c920fdf7f5
Change a bunch of video-output-specific preprocessor directives from a HAVE_
diego
parents:
26759
diff
changeset
|
101 #ifdef CONFIG_MGA |
572 | 102 mga_init(); |
103 marr1 = &frame[3]; | |
104 #else | |
3081 | 105 marr1 = &arr1[3]; |
572 | 106 #endif |
3081 | 107 marr2 = &arr2[9]; |
3076
bfc881c0e591
newly allocated memory seems to point to only 1 zero filled 4k page or something so there is a near 100% cache hit rate ... not very realistic, writeing something in the source array fixes that so the benchmark scores are meaningfull now
michael
parents:
687
diff
changeset
|
108 |
3082 | 109 for(i=0; i<ARR_SIZE-16; i++) marr1[i] = marr2[i] = i; |
3076
bfc881c0e591
newly allocated memory seems to point to only 1 zero filled 4k page or something so there is a near 100% cache hit rate ... not very realistic, writeing something in the source array fixes that so the benchmark scores are meaningfull now
michael
parents:
687
diff
changeset
|
110 |
572 | 111 t=GetTimer(); |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
112 v1 = read_tsc(); |
572 | 113 for(i=0;i<100;i++) memcpy(marr1,marr2,ARR_SIZE-16); |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
114 v2 = read_tsc(); |
572 | 115 t=GetTimer()-t; |
116 // ARR_SIZE*100/(1024*1024)/(t/1000000) = ARR_SIZE*95.36743/t | |
687 | 117 printf(NAME": cpu clocks=%llu = %dus (%5.3ffps) %5.1fMB/s\n",v2-v1,t,100000000.0f/(float)t,(float)ARR_SIZE*95.36743f/(float)t); |
562
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
118 return 0; |
312fee2a6816
Initial release, used to benchmark fastmemcpy.h code from libvo.
atmosfear
parents:
diff
changeset
|
119 } |