#1 2009-11-17 09:59:55
編譯cpubench, 測試效能(CPU and memory benchmark)
參考站長所寫的效能測試 (CPU and memory benchmark), 取得原始碼, 經由gcc編譯後, 產生執行檔cpubench, 測試NAS的CPU and memory benchmark.
型號: DS209+II
韌體: DSM 2.2-0942
1.) 安裝gcc
> ipkg-opt install gcc
威聯通TS-109 Pro要使用cs05q3armel feed, 才有gcc.
2.) openwrt_cpu_bench_v06.c
來源
#include
#include
#include
#include
#include
#define VERSION_STR "v0.6"
// This is the overhead of calling gettimeofday() 2 times
double overhead = -1;
double begin_secs = 0;
double real_get_seconds() {
struct timeval tv;
double total;
gettimeofday(&tv, NULL);
total = (double)tv.tv_sec;
total += ((double)tv.tv_usec)/(double)1000000.0;
//printf("Secs: %d / Usecs: %d / FSecs: %f\n", tv.tv_sec, tv.tv_usec, total);
return total;
}
double get_seconds() {
if (begin_secs == 0) {
// First call
begin_secs = real_get_seconds();
}
return (real_get_seconds() - begin_secs);
}
double run_float_bench() {
double secs;
double begin = get_seconds();
int i;
double nb=0;
for (i=0;i<300000;i++) {
nb *= i;
int n = i >> 10;
nb /= n;
}
double end = get_seconds();
secs = end-begin;
printf("Time to run float bench: %.2f[secs]\n", secs);
return secs;
}
double do_run_memory_bench() {
double begin, end;
int i;
int* buf;
int len, index;
double secs;
begin = get_seconds();
len = (1<<21)/sizeof(int); // 2Mb
buf = malloc(len*sizeof(int));
// Write to memory - sequential
for (i=0;i<len;i++) {
buf[i] = i;
}
// Read memory - sequential
for (i=0;i<len;i++) {
int a = buf[i];
}
// Read memory - random
for (i=0;i<len;i++) {
index = (i*23)%(len/2);
int a = buf[index];
}
// Write memory - random
for (i=0;i<len;i++) {
index = (i*23)%(len/2);
buf[index] = i;
}
free(buf);
end = get_seconds();
secs = (end-begin);
return secs;
}
double run_memory_bench() {
double secs = 0;
int i;
for (i=0;i<10;i++) {
secs += do_run_memory_bench();
}
printf("Time to run memory bench: %.2f[secs]\n", secs);
return secs;
}
#define NBD 9009
double run_compute_e() {
double begin, end;
double secs;
begin = get_seconds();
int N=NBD, n=N, a[NBD],x;
while(--n) {
a[n]=1+1/n;
}
for(;N>9;) {
for(n=N--;--n; a[n]=x%n, x=10*a[n-1]+x/n) {
}
}
end = get_seconds();
secs = (end-begin);
printf("Time to run computation of e (%d digits): %.2f[secs]\n", NBD, secs);
return secs;
}
double run_compute_pi() {
double begin, end;
double secs;
int i;
begin = get_seconds();
//printf("Begin: %f\n", begin);
for (i=0;i<10;i++) {
int a=10000,b=0,c=8400,d=0,e=0,f[8401],g=0;
//printf("i:%d\n", i);
for(;b-c;) {
f[b++]=a/5;
}
for(;d=0,g=c*2;c-=14,e=d%a) {
for(b=c;d+=f[b]*a,f[b]=d%--g,d/=g--,--b;d*=b) {
}
}
//printf("Mid: %lld\n", (get_seconds()-end));
}
end = get_seconds();
//printf("End: %f\n", end);
secs = (end-begin);
printf("Time to run computation of pi (2400 digits, 10 times): %.2f[secs]\n", secs);
return secs;
}
int main() {
printf("This is CPU and memory benchmark for OpenWRT "VERSION_STR". This will then take some time... (typically 30-60 seconds on a 200MHz computer)\n");
double begin = get_seconds();
double end = get_seconds();
overhead = (end-begin)*1000000;
printf("Overhead for getting time: %.0fus\n", overhead);
// Nb 1
double sec_mem = run_memory_bench();
// Nb 2
double sec_pi = run_compute_pi();
// Nb 3
double sec_e = run_compute_e();
// Nb 4
double sec_float = run_float_bench();
printf("Total time: %.1fs\n", (sec_mem+sec_e+sec_pi+sec_float));
time_t t = time(0);
struct tm ti;
localtime_r(&t, &ti);
printf("\nYou can copy/paste the following line in the wiki table at: http://wiki.openwrt.org/HardwarePerformance\n");
printf("|| %04d-%02d-%02d || ''Author'' || %.1fs || %.1fs || %.1fs || %.1fs || " VERSION_STR " || ''OS'' || ''DeviceModel'' || ''CPU model'' || ''CPU Frequency'' || ''LinkToHwPage'' ||\n", (ti.tm_year+1900), (ti.tm_mon+1), ti.tm_mday, sec_mem, sec_pi, sec_e, sec_float);
return 0;
}
將上面的原始碼, 複製到nano或vi編輯器, 注意斷行問題, 存檔為openwrt_cpu_bench_v06.c
3.) 編譯, 產生執行檔cpubench
> gcc -o cpubench openwrt_cpu_bench_v06.c
編譯過程若產生錯誤, 依錯誤訊息修正原始碼.
4.) 執行
> ./cpubench
離線
相關討論主題
主題 | 回覆 | 點閱 | 最後發表 |
---|---|---|---|
|
45 | 223135 | 2017-08-01 10:59:05 作者 ken0618102000 |
置頂 |
43 | 178067 | 2014-08-16 21:19:12 作者 dg4838 |