/*----------------------------------------------------------------------------*/ /* * Performance-Monitoring Counters Library, for Intel/AMD Processors and Linux * Author: Don Heller, dheller@scl.ameslab.gov * Last revised: 19 January 2001 */ /*----------------------------------------------------------------------------*/ #ifndef TESTNUMBER #define TESTNUMBER 0 #endif // up to 29 /* * select a test case, print anything with at least 1 event in every trial * usage: rabbit8 [options] [trials] * see also rabbit8.sh */ /*----------------------------------------------------------------------------*/ #include #include #include /*----------------------------------------------------------------------------*/ pmc_data_t t0, t1, t2; pmc_counter_t c; int main(int argc, char * argv[]) { pmc_control_t Ctl = pmc_control_null; int e, i,j,k, out, trials = 1000; pmc_counter_t * a; pmc_data_t t3; #if (19 <= TESTNUMBER) && (TESTNUMBER <= 20) int i19; double a19[10000], b19[10000]; #endif /* initialize internal data structures, read command-line arguments */ Ctl.clean = 10; Ctl.stats = 1; /* we need the min, mean and max */ if (pmc_getargs(stderr, argv[0], &argc, &argv, &Ctl) == FALSE) { exit(RABBIT_FAILURE); } if (argc > 0) { trials = atoi(argv[0]); } printf("test number = %d, trials = %d\n", TESTNUMBER, trials); // pmc_counter_init(&c, &Ctl); c = Ctl.counters[0]; if (pmc_open(0) == FALSE) /* open /dev/pmc */ { exit(RABBIT_FAILURE); } /* for (out = 0; out < Ctl.num_counters; out++) { ... } */ for (i = 0, out = 0; i < Ctl.event_pairs; i++) { for (j = 0; j < Ctl.replication; j++, out++) { a = &Ctl.counters[out]; pmc_select(a); pmc_read(&t0); pmc_read(&t1); for (k = 0; k < trials; k++) { pmc_read(&t0); //........................................................................... #if TESTNUMBER == 0 /* nothing */ #endif #if TESTNUMBER == 1 __asm__ __volatile__ ( "xorl %%eax,%%eax\n\t" /* eax = 0 */ "cpuid" /* registers affected */ : /* output */ : /* input */ : "eax", "ebx", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 2 __asm__ __volatile__ ( "movl $0,%%eax\n\t" /* eax = 0 */ "cpuid" /* registers affected */ : /* output */ : /* input */ : "eax", "ebx", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 3 __asm__ __volatile__ ( "movl $1,%%eax\n\t" /* eax = 1 */ "cpuid" /* registers affected */ : /* output */ : /* input */ : "eax", "ebx", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 4 __asm__ __volatile__ ( "movl $2,%%eax\n\t" /* eax = 2 */ "cpuid" /* registers affected */ : /* output */ : /* input */ : "eax", "ebx", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 5 __asm__ __volatile__ ( "movl $3,%%eax\n\t" /* eax = 3 */ "cpuid" /* registers affected */ : /* output */ : /* input */ : "eax", "ebx", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 6 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 7 __asm__ __volatile__ ( "xorl %%ecx,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 8 __asm__ __volatile__ ( "movl $0,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 9 __asm__ __volatile__ ( "movl $1,%%ecx\n\t" /* ecx = 1 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,16(%%ebx)\n\t" "movl %%edx,20(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 10 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)\n\t" "xorl %%ecx,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 11 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)\n\t" "movl $0,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 12 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)\n\t" "xorl %%ecx,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)\n\t" "movl $1,%%ecx\n\t" /* ecx = 1 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,16(%%ebx)\n\t" "movl %%edx,20(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 13 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)\n\t" "movl $0,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)\n\t" "movl $1,%%ecx\n\t" /* ecx = 1 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,16(%%ebx)\n\t" "movl %%edx,20(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 14 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)\n\t" "xorl %%ecx,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)\n\t" "incl %%ecx\n\t" /* ecx = 1 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,16(%%ebx)\n\t" "movl %%edx,20(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 15 __asm__ __volatile__ ( "rdtsc\n\t" "movl %%eax,0(%%ebx)\n\t" "movl %%edx,4(%%ebx)\n\t" "movl $0,%%ecx\n\t" /* ecx = 0 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,8(%%ebx)\n\t" "movl %%edx,12(%%ebx)\n\t" "incl %%ecx\n\t" /* ecx = 1 */ "rdpmc\n\t" #if defined(PMC_P6) "andl $255,%%edx\n\t" /* clean the upper 24 bits */ #endif "movl %%eax,16(%%ebx)\n\t" "movl %%edx,20(%%ebx)" : /* output */ : "ebx" (&t2) /* input */ : "eax", "ecx", "edx" /* clobbered */ ); #endif #if TESTNUMBER == 16 pmc_read(&t1); #endif #if TESTNUMBER == 17 pmc_read(&t2); #endif #if TESTNUMBER == 18 pmc_read(&t3); #endif #if TESTNUMBER == 19 for (i19 = 0; i19 < 10000; i19++) { a19[i19] = i19; } #endif #if TESTNUMBER == 20 for (i19 = 0; i19 < 10000; i19++) { a19[i19] = i19; } for (i19 = 0; i19 < 10000; i19++) { b19[i19] = a19[i19] * 10.0; } #endif /* 20 groups of 5 */ #define INSTR(instr) \ __asm__ __volatile__ ( \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" \ instr "\n\t" instr "\n\t" instr "\n\t" instr "\n\t" instr \ : : : "ecx"); #if TESTNUMBER == 21 INSTR("addl $0,%%ecx") #endif #if TESTNUMBER == 22 INSTR("addl $4,%%ecx") #endif #if TESTNUMBER == 23 INSTR("movl $0,%%ecx") #endif #if TESTNUMBER == 24 INSTR("movl $4,%%ecx") #endif #if TESTNUMBER == 25 INSTR("incl %%ecx") #endif #if TESTNUMBER == 26 INSTR("addl $1,%%ecx") #endif #if TESTNUMBER == 27 INSTR("leal (%%ecx,%%ecx,8),%%ecx") #endif #if TESTNUMBER == 28 INSTR("leal (%%ecx,%%ecx,8),%%ecx" "\n\t" "addl $1,%%ecx") #endif #if TESTNUMBER == 29 { register int i; for (i = 0; i < 512*1024; i++) { } } #endif //........................................................................... pmc_read(&t1); pmc_accumulate(a,&t1,&t0); pmc_accumulate_clock(&c,&t1,NULL); } } } pmc_close(); /* close /dev/pmc */ printf("number of trials = %10d " "min mean max std.dev\n", trials); printf("%-49s %5d %8.2f %5d %8.2f\n", "cycles", (int)pmc_min_cycles(&c), pmc_mean_cycles(&c), (int)pmc_max_cycles(&c), sqrt(pmc_variance_cycles(&c)) ); for (i = 0, out = 0; i < Ctl.event_pairs; i++) { for (j = 0; j < Ctl.replication; j++, out++) { a = &Ctl.counters[out]; for (k = 0; k < pmc_event_counters; k++) { e = Ctl.events[i][k]; if (pmc_min_events(a,k) > 0) { printf("0x%02x %3d %-40s %5d %8.2f %5d %8.2f\n", e, e, pmc_event_name(e,k), (int)pmc_min_events(a,k), pmc_mean_events(a,k), (int)pmc_max_events(a,k), sqrt(pmc_variance_events(a,k)) ); } } } } printf("\n"); exit(RABBIT_SUCCESS); } /*----------------------------------------------------------------------------*/