|
5
|
1 /*
|
|
|
2 speedtest.c
|
|
|
3 David Rowe
|
|
|
4 Created 27 Feb 2007
|
|
|
5
|
|
|
6 Measures execution speed of oslec in user mode.
|
|
|
7 */
|
|
|
8
|
|
|
9 /*
|
|
|
10 Copyright (C) 2007 David Rowe
|
|
|
11
|
|
|
12 All rights reserved.
|
|
|
13
|
|
|
14 This program is free software; you can redistribute it and/or modify
|
|
|
15 it under the terms of the GNU General Public License version 2, as
|
|
|
16 published by the Free Software Foundation.
|
|
|
17
|
|
|
18 This program is distributed in the hope that it will be useful,
|
|
|
19 but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
20 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
21 GNU General Public License for more details.
|
|
|
22
|
|
|
23 You should have received a copy of the GNU General Public License
|
|
|
24 along with this program; if not, write to the Free Software
|
|
|
25 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
|
26 */
|
|
|
27
|
|
|
28 #include <assert.h>
|
|
|
29 #include <stdlib.h>
|
|
|
30 #include <stdio.h>
|
|
|
31 #include <sys/time.h>
|
|
|
32 #include <time.h>
|
|
|
33 #include <string.h>
|
|
|
34 #include <unistd.h>
|
|
|
35 #include <stdint.h>
|
|
|
36 #include <echo.h>
|
|
|
37
|
|
|
38 #define TAPS 128
|
|
|
39 #define N 8000 /* sample rate */
|
|
|
40 #define AMP 1000
|
|
|
41 #define SECS 10 /* number of simulated seconds to process */
|
|
|
42
|
|
|
43 /* constant for isr cycle averaging */
|
|
|
44
|
|
|
45 #define LTC 5
|
|
|
46
|
|
|
47 /* number of cycles we are using per call */
|
|
|
48
|
|
|
49 long long cycles_last = 0;
|
|
|
50 long long cycles_worst = 0;
|
|
|
51 long long cycles_average = 0;
|
|
|
52
|
|
|
53 #ifdef __BLACKFIN__
|
|
|
54 /* sample cycles register of Blackfin */
|
|
|
55
|
|
|
56 static inline unsigned int cycles(void) {
|
|
|
57 int ret;
|
|
|
58
|
|
|
59 __asm__ __volatile__
|
|
|
60 (
|
|
|
61 "%0 = CYCLES;\n\t"
|
|
|
62 : "=&d" (ret)
|
|
|
63 :
|
|
|
64 : "R1"
|
|
|
65 );
|
|
|
66
|
|
|
67 return ret;
|
|
|
68 }
|
|
|
69 #elif defined(__X86__) || defined (__i386) || defined (__x86_64__)
|
|
|
70 static __inline__ uint64_t cycles() {
|
|
|
71 uint64_t x;
|
|
|
72 __asm__ volatile ("rdtsc\n\t" : "=A" (x));
|
|
|
73 return x;
|
|
|
74 }
|
|
|
75 #else
|
|
|
76 static inline volatile unsigned int cycles(void) {
|
|
|
77 /* A dummy implementation for other architectures */
|
|
|
78 static unsigned int dummy_cycles = 1;
|
|
|
79 return dummy_cycles++;
|
|
|
80 }
|
|
|
81 #endif
|
|
|
82
|
|
|
83 int main(int argc, char **argv) {
|
|
|
84 int i,j;
|
|
|
85 echo_can_state_t *ec;
|
|
|
86 short tx[N],rx[N],clean;
|
|
|
87 struct timeval tv_before;
|
|
|
88 struct timeval tv_after;
|
|
|
89 unsigned long long t_before_ms, t_after_ms;
|
|
|
90 unsigned long long before_clocks, after_clocks;
|
|
|
91 unsigned long long t_ms;
|
|
|
92 unsigned long long start_cycles;
|
|
|
93 float mips_cpu, mips_per_ec;
|
|
|
94 FILE *f;
|
|
|
95
|
|
|
96 printf("\nTesting OSLEC with %d taps (%d ms tail)\n", TAPS, (TAPS*1000)/N);
|
|
|
97 for(i=0; i<N; i++) {
|
|
|
98 tx[i] = (short)AMP*(float)rand()/RAND_MAX;
|
|
|
99 rx[i] = tx[i]/4;
|
|
|
100 }
|
|
|
101
|
|
|
102 /* note NLP not switched on to make output more interesting for bit exact
|
|
|
103 testing */
|
|
|
104
|
|
|
105 ec = echo_can_create(TAPS, ECHO_CAN_USE_ADAPTION);
|
|
|
106
|
|
|
107 /* dump output of first run for bit exact testing when optimising */
|
|
|
108
|
|
|
109 f = fopen("out.txt","wt");
|
|
|
110 assert(f != NULL);
|
|
|
111 for(i=0; i<N; i++) {
|
|
|
112 clean = echo_can_update(ec, tx[i], rx[i]);
|
|
|
113 fprintf(f,"%d\n", clean);
|
|
|
114 }
|
|
|
115 fclose(f);
|
|
|
116
|
|
|
117 gettimeofday(&tv_before, NULL);
|
|
|
118 before_clocks = cycles();
|
|
|
119 for(j=0; j<SECS; j++) {
|
|
|
120
|
|
|
121 for(i=0; i<N; i++) {
|
|
|
122 start_cycles = cycles();
|
|
|
123 clean = echo_can_update(ec, tx[i], rx[i]);
|
|
|
124 cycles_last = cycles() - start_cycles;
|
|
|
125 cycles_average += (cycles_last - cycles_average +(1<<(LTC-1))) >> LTC;
|
|
|
126
|
|
|
127 if (cycles_last > cycles_worst)
|
|
|
128 cycles_worst = cycles_last;
|
|
|
129 }
|
|
|
130
|
|
|
131 }
|
|
|
132 after_clocks = cycles();
|
|
|
133 gettimeofday(&tv_after, NULL);
|
|
|
134 t_before_ms = 1000*tv_before.tv_sec + tv_before.tv_usec/1000;
|
|
|
135 t_after_ms = 1000*tv_after.tv_sec + tv_after.tv_usec/1000;
|
|
|
136 t_ms = t_after_ms - t_before_ms;
|
|
|
137
|
|
|
138 mips_cpu = (after_clocks - before_clocks)/(1E3*t_ms);
|
|
|
139 printf("CPU executes %5.2f MIPS\n-------------------------\n\n", mips_cpu);
|
|
|
140
|
|
|
141 printf("Method 1: gettimeofday() at start and end\n");
|
|
|
142 printf(" %llu ms for %ds of speech\n", t_ms, SECS);
|
|
|
143 mips_per_ec = mips_cpu/((float)SECS*1E3/(float)t_ms);
|
|
|
144 printf(" %5.2f MIPS\n", mips_per_ec);
|
|
|
145 printf(" %5.2f instances possible at 100%% CPU load\n",
|
|
|
146 (float)SECS*1E3/(float)t_ms);
|
|
|
147
|
|
|
148 printf("Method 2: samples clock cycles at start and end\n");
|
|
|
149 printf(" %5.2f MIPS\n", (after_clocks - before_clocks)/(1E6*SECS));
|
|
|
150 printf(" %5.2f instances possible at 100%% CPU load\n",
|
|
|
151 mips_cpu/((after_clocks - before_clocks)/(1E6*SECS)));
|
|
|
152
|
|
|
153 printf("Method 3: samples clock cycles for each call, IIR average\n");
|
|
|
154 mips_per_ec = 8*(float)cycles_average/1000;
|
|
|
155 printf(" cycles_worst %lld cycles_last %lld cycles_av: %lld"
|
|
|
156 "\n %5.2f MIPS\n",
|
|
|
157 cycles_worst, cycles_last, cycles_average, mips_per_ec);
|
|
|
158 printf(" %5.2f instances possible at 100%% CPU load\n", mips_cpu/mips_per_ec);
|
|
|
159
|
|
|
160 return 0;
|
|
|
161 }
|
|
|
162
|
