bool cpu_sse41_available( ) {
cpuid_init( );
if (force_no_simd) {
return false;
} else if (cpuid_ecx & 0x00080000) {
return true;
} else {
return false;
}
}
/* Entry point in to the kernel proper */
_noreturn _asmlinkage void setup(multiboot_info_t *mbi)
{
serio_init();
vga_preinit();
/* Need this mmediately to catch exceptions */
idt_init();
/* print a pretty message */
printk("ScaraOS v0.0.7 for IA-32");
if ( mbi->flags & MBF_CMDLINE ) {
cmdline = __va(mbi->cmdline);
printk(": %s", cmdline);
}
printk("\n");
/* Prints out CPU MHz */
calibrate_delay_loop();
/* Identify CPU features, also needs to be early because
* mm init may need to know which paging features are
* available
*/
cpuid_init();
/* Fire up the kernel memory allocators */
BUG_ON((mbi->flags & MBF_MMAP) == 0);
ia32_mm_init(__va(mbi->mmap), mbi->mmap_length);
/* setup the idle task */
sched_init();
ia32_gdt_finalize();
/* enable hardware interrupts */
pic_init();
/* start jiffie counter */
pit_start_timer1();
/* initialise fdt logic ready for use */
_fdt_init();
/* Setup the init task */
kernel_thread("[init]", init_task, NULL);
/* Finally, enable interupts and let it rip */
sti();
sched();
idle_task_func();
}
int main (int argc, char** argv)
{
int domainId = -1;
int c;
bstring argString;
while ((c = getopt (argc, argv, "+c:hv")) != -1)
{
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'c':
if (! (argString = bSecureInput(10,optarg)))
{
fprintf(stderr,"Failed to read argument string!\n");
exit(EXIT_FAILURE);
}
domainId = str2int((char*) argString->data);
break;
case '?':
if (isprint (optopt))
{
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
}
else
{
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
}
return EXIT_FAILURE;
default:
HELP_MSG;
exit (EXIT_SUCCESS);
}
}
if (cpuid_init() == EXIT_FAILURE)
{
ERROR_PLAIN_PRINT(Unsupported processor!);
}
void
smp_init_bsp()
{
floating_pointer_struct_t *fp;
/* gets the details about the cpu, the type, the brand
* whether it is a multi-core package etc.
*/
cpuid_init();
memset(&AP, 0, sizeof AP);
fp = scan_for_floating_ptr_struct(0x0, 0x400);
if (fp == NULL) {
fp = scan_for_floating_ptr_struct(639*0x400, 0x400);
}
if (fp == NULL) {
fp = scan_for_floating_ptr_struct(0xf0000, 0x10000);
}
if (fp == NULL) {
/*
* If it is an SMP machine we should know now, unless the
* configuration is in an EISA/MCA bus machine with an
* extended bios data area.
*
* there is a real-mode segmented pointer pointing to the
* 4K EBDA area at 0x40E, calculate and scan it here.
*/
unsigned int address = *(unsigned short *)0x40E;
address <<= 4;
if (address) {
fp = scan_for_floating_ptr_struct(address, 0x400);
}
}
if (fp != NULL && fp->phys_addr != 0) {
if (!read_mp_config_table(fp->phys_addr)) {
//cprint(LINE_STATUS+1,0, "SMP: Error while parsing MP config table");
}
}
/*
if (fp == NULL) {
cprint(LINE_STATUS+1,0,"SMP: No floating pointer structure found");
}
*/
}
int main()
{
//Start video driver (must always be before loading message)
mm_init();
pg_init();
real_init();
video_init();
video_setdriver(video_vgatext_getdriver(),0);
//Put loading message
cli_puts("ArcaneOS Loading...\n");
//Setup kernel
gdt_init();
idt_init();
isr_init();
irq_init();
timer_init();
kb_init();
ui_init();
cpuid_init();
cmos_init();
rtc_init();
acpi_init();
power_init();
mt_init();
syscall_init();
floppy_init();
__asm__ __volatile__ ("sti");
//Enable ACPI
acpi_enable();
//Create thread for ui
mt_create_thread(mt_kernel_process,test,2);
//Endless loop to prevent bugs when all threads are sleeping
for(;;)
__asm__ __volatile__ ("hlt");
}
int main(int argc, char** argv)
{
int iter = 100;
uint32_t i;
uint32_t j;
int globalNumberOfThreads = 0;
int optPrintDomains = 0;
int c;
ThreadUserData myData;
bstring testcase = bfromcstr("none");
uint32_t numberOfWorkgroups = 0;
int tmp = 0;
double time;
const TestCase* test = NULL;
Workgroup* currentWorkgroup = NULL;
Workgroup* groups = NULL;
if (cpuid_init() == EXIT_FAILURE)
{
ERROR_PLAIN_PRINT(Unsupported processor!);
}
void likwid_markerInit(void)
{
int cpuId = likwid_getProcessorId();
char* modeStr = getenv("LIKWID_MODE");
char* maskStr = getenv("LIKWID_MASK");
if ((modeStr != NULL) && (maskStr != NULL))
{
likwid_init = 1;
}
else
{
return;
}
if (!lock_check())
{
fprintf(stderr,"Access to performance counters is locked.\n");
exit(EXIT_FAILURE);
}
cpuid_init();
numa_init();
affinity_init();
timer_init();
hashTable_init();
for(int i=0; i<MAX_NUM_THREADS; i++) thread_socketFD[i] = -1;
for(int i=0; i<MAX_NUM_NODES; i++) socket_lock[i] = LOCK_INIT;
accessClient_mode = atoi(modeStr);
str2BitMask(maskStr, &counterMask);
if (accessClient_mode != DAEMON_AM_DIRECT)
{
accessClient_init(&thread_socketFD[cpuId]);
}
msr_init(thread_socketFD[cpuId]);
thermal_init(cpuId);
switch ( cpuid_info.family )
{
case P6_FAMILY:
switch ( cpuid_info.model )
{
case PENTIUM_M_BANIAS:
case PENTIUM_M_DOTHAN:
perfmon_counter_map = pm_counter_map;
perfmon_numCounters = NUM_COUNTERS_PM;
perfmon_numCountersCore = NUM_COUNTERS_CORE_PM;
break;
case ATOM_45:
case ATOM_32:
case ATOM_22:
case ATOM:
perfmon_counter_map = core2_counter_map;
perfmon_numCounters = NUM_COUNTERS_CORE2;
perfmon_numCountersCore = NUM_COUNTERS_CORE_CORE2;
break;
case CORE_DUO:
ERROR_PLAIN_PRINT(Unsupported Processor);
break;
case XEON_MP:
case CORE2_65:
case CORE2_45:
perfmon_counter_map = core2_counter_map;
perfmon_numCounters = NUM_COUNTERS_CORE2;
perfmon_numCountersCore = NUM_COUNTERS_CORE_CORE2;
break;
case NEHALEM_EX:
case WESTMERE_EX:
perfmon_counter_map = westmereEX_counter_map;
perfmon_numCounters = NUM_COUNTERS_WESTMEREEX;
perfmon_numCountersCore = NUM_COUNTERS_CORE_WESTMEREEX;
perfmon_numCountersUncore = NUM_COUNTERS_UNCORE_WESTMEREEX;
break;
case NEHALEM_BLOOMFIELD:
case NEHALEM_LYNNFIELD:
case NEHALEM_WESTMERE_M:
case NEHALEM_WESTMERE:
perfmon_counter_map = nehalem_counter_map;
perfmon_numCounters = NUM_COUNTERS_NEHALEM;
perfmon_numCountersCore = NUM_COUNTERS_CORE_NEHALEM;
perfmon_numCountersUncore = NUM_COUNTERS_UNCORE_NEHALEM;
break;
case IVYBRIDGE:
case IVYBRIDGE_EP:
{
int socket_fd = thread_socketFD[cpuId];
hasPCICounters = 1;
power_init(0); /* FIXME Static coreId is dangerous */
pci_init(socket_fd);
perfmon_counter_map = ivybridge_counter_map;
perfmon_numCounters = NUM_COUNTERS_IVYBRIDGE;
perfmon_numCountersCore = NUM_COUNTERS_CORE_IVYBRIDGE;
perfmon_numCountersUncore = NUM_COUNTERS_UNCORE_IVYBRIDGE;
}
break;
case HASWELL:
case HASWELL_EX:
case HASWELL_M1:
case HASWELL_M2:
power_init(0); /* FIXME Static coreId is dangerous */
perfmon_counter_map = haswell_counter_map;
perfmon_numCounters = NUM_COUNTERS_HASWELL;
perfmon_numCountersCore = NUM_COUNTERS_CORE_HASWELL;
break;
case SANDYBRIDGE:
case SANDYBRIDGE_EP:
{
int socket_fd = thread_socketFD[cpuId];
hasPCICounters = 1;
power_init(0); /* FIXME Static coreId is dangerous */
pci_init(socket_fd);
perfmon_counter_map = sandybridge_counter_map;
perfmon_numCounters = NUM_COUNTERS_SANDYBRIDGE;
perfmon_numCountersCore = NUM_COUNTERS_CORE_SANDYBRIDGE;
perfmon_numCountersUncore = NUM_COUNTERS_UNCORE_SANDYBRIDGE;
}
break;
default:
ERROR_PLAIN_PRINT(Unsupported Processor);
break;
}
break;
case MIC_FAMILY:
switch ( cpuid_info.model )
{
case XEON_PHI:
perfmon_counter_map = phi_counter_map;
perfmon_numCounters = NUM_COUNTERS_PHI;
perfmon_numCountersCore = NUM_COUNTERS_CORE_PHI;
break;
default:
ERROR_PLAIN_PRINT(Unsupported Processor);
break;
}
break;
case K8_FAMILY:
perfmon_counter_map = k10_counter_map;
perfmon_numCounters = NUM_COUNTERS_K10;
perfmon_numCountersCore = NUM_COUNTERS_CORE_K10;
break;
case K10_FAMILY:
perfmon_counter_map = k10_counter_map;
perfmon_numCounters = NUM_COUNTERS_K10;
perfmon_numCountersCore = NUM_COUNTERS_CORE_K10;
break;
case K15_FAMILY:
perfmon_counter_map = interlagos_counter_map;
perfmon_numCounters = NUM_COUNTERS_INTERLAGOS;
perfmon_numCountersCore = NUM_COUNTERS_CORE_INTERLAGOS;
break;
case K16_FAMILY:
perfmon_counter_map = kabini_counter_map;
perfmon_numCounters = NUM_COUNTERS_KABINI;
perfmon_numCountersCore = NUM_COUNTERS_CORE_KABINI;
break;
default:
ERROR_PLAIN_PRINT(Unsupported Processor);
break;
}
}
int main(int argc, char** argv)
{
int iter = 100;
uint32_t i;
uint32_t j;
int globalNumberOfThreads = 0;
int optPrintDomains = 0;
int c;
ThreadUserData myData;
bstring testcase = bfromcstr("none");
uint32_t numberOfWorkgroups = 0;
int tmp = 0;
double time;
const TestCase* test = NULL;
Workgroup* currentWorkgroup = NULL;
Workgroup* groups = NULL;
cpuid_init();
numa_init();
affinity_init();
/* Handling of command line options */
if (argc == 1) { HELP_MSG; }
while ((c = getopt (argc, argv, "g:w:t:i:l:aphv")) != -1) {
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'a':
printf(TESTS"\n");
exit (EXIT_SUCCESS);
case 'w':
tmp--;
if (tmp == -1)
{
fprintf (stderr, "More workgroups configured than allocated!\n");
return EXIT_FAILURE;
}
if (!test)
{
fprintf (stderr, "You need to specify a test case first!\n");
return EXIT_FAILURE;
}
testcase = bfromcstr(optarg);
currentWorkgroup = groups+tmp; /*FIXME*/
bstr_to_workgroup(currentWorkgroup, testcase, test->type, test->streams);
bdestroy(testcase);
for (i=0; i< test->streams; i++)
{
if (currentWorkgroup->streams[i].offset%test->stride)
{
fprintf (stderr, "Stream %d: offset is not a multiple of stride!\n",i);
return EXIT_FAILURE;
}
allocator_allocateVector(&(currentWorkgroup->streams[i].ptr),
PAGE_ALIGNMENT,
currentWorkgroup->size,
currentWorkgroup->streams[i].offset,
test->type,
currentWorkgroup->streams[i].domain);
}
break;
case 'i':
iter = atoi(optarg);
break;
case 'l':
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (biseqcstr(testcase,"none"))
{
fprintf (stderr, "Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
else
{
printf("Name: %s\n",test->name);
printf("Number of streams: %d\n",test->streams);
printf("Loop stride: %d\n",test->stride);
printf("Flops: %d\n",test->flops);
printf("Bytes: %d\n",test->bytes);
switch (test->type)
{
case SINGLE:
printf("Data Type: Single precision float\n");
break;
case DOUBLE:
printf("Data Type: Double precision float\n");
break;
}
}
bdestroy(testcase);
exit (EXIT_SUCCESS);
break;
case 'p':
optPrintDomains = 1;
break;
case 'g':
numberOfWorkgroups = atoi(optarg);
allocator_init(numberOfWorkgroups * MAX_STREAMS);
tmp = numberOfWorkgroups;
groups = (Workgroup*) malloc(numberOfWorkgroups*sizeof(Workgroup));
break;
case 't':
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (biseqcstr(testcase,"none"))
{
fprintf (stderr, "Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
bdestroy(testcase);
break;
case '?':
if (isprint (optopt))
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
else
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
return EXIT_FAILURE;
default:
HELP_MSG;
}
}
if (optPrintDomains)
{
affinity_printDomains();
exit (EXIT_SUCCESS);
}
timer_init();
/* :WARNING:05/04/2010 08:58:05 AM:jt: At the moment the thread
* module only allows equally sized thread groups*/
for (i=0; i<numberOfWorkgroups; i++)
{
globalNumberOfThreads += groups[i].numberOfThreads;
}
threads_init(globalNumberOfThreads);
threads_createGroups(numberOfWorkgroups);
/* we configure global barriers only */
barrier_init(1);
barrier_registerGroup(globalNumberOfThreads);
#ifdef PERFMON
printf("Using likwid\n");
likwid_markerInit();
#endif
/* initialize data structures for threads */
for (i=0; i<numberOfWorkgroups; i++)
{
myData.iter = iter;
myData.size = groups[i].size;
myData.test = test;
myData.numberOfThreads = groups[i].numberOfThreads;
myData.processors = (int*) malloc(myData.numberOfThreads * sizeof(int));
myData.streams = (void**) malloc(test->streams * sizeof(void*));
for (j=0; j<groups[i].numberOfThreads; j++)
{
myData.processors[j] = groups[i].processorIds[j];
}
for (j=0; j< test->streams; j++)
{
myData.streams[j] = groups[i].streams[j].ptr;
}
threads_registerDataGroup(i, &myData, copyThreadData);
free(myData.processors);
free(myData.streams);
}
printf(HLINE);
printf("LIKWID MICRO BENCHMARK\n");
printf("Test: %s\n",test->name);
printf(HLINE);
printf("Using %d work groups\n",numberOfWorkgroups);
printf("Using %d threads\n",globalNumberOfThreads);
printf(HLINE);
threads_create(runTest);
threads_destroy();
allocator_finalize();
time = (double) threads_data[0].cycles / (double) timer_getCpuClock();
printf("Cycles: %llu \n", LLU_CAST threads_data[0].cycles);
printf("Iterations: %llu \n", LLU_CAST iter);
printf("Size: %d \n", currentWorkgroup->size );
printf("Vectorlength: %d \n", threads_data[0].data.size);
printf("Time: %e sec\n", time);
printf("MFlops/s:\t%.2f\n",
1.0E-06 * ((double) numberOfWorkgroups * iter * currentWorkgroup->size * test->flops/ time));
printf("MByte/s:\t%.2f\n",
1.0E-06 * ( (double) numberOfWorkgroups * iter * currentWorkgroup->size * test->bytes/ time));
printf("Cycles per update:\t%f\n",
((double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
switch ( test->type )
{
case SINGLE:
printf("Cycles per cacheline:\t%f\n",
(16.0 * (double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
break;
case DOUBLE:
printf("Cycles per cacheline:\t%f\n",
(8.0 * (double) threads_data[0].cycles / (double) (iter * threads_data[0].data.size)));
break;
}
printf(HLINE);
#ifdef PERFMON
likwid_markerClose();
#endif
return EXIT_SUCCESS;
}
int main (int argc, char** argv)
{
int socket_fd = -1;
int optInfo = 0;
int optClock = 0;
int optStethoscope = 0;
int optSockets = 0;
double runtime;
int hasDRAM = 0;
int c;
bstring argString;
bstring eventString = bfromcstr("CLOCK");
int numSockets=1;
int numThreads=0;
int threadsSockets[MAX_NUM_NODES*2];
int threads[MAX_NUM_THREADS];
threadsSockets[0] = 0;
if (argc == 1)
{
HELP_MSG;
exit (EXIT_SUCCESS);
}
while ((c = getopt (argc, argv, "+c:hiM:ps:v")) != -1)
{
switch (c)
{
case 'c':
CHECK_OPTION_STRING;
numSockets = bstr_to_cpuset_physical((uint32_t*) threadsSockets, argString);
bdestroy(argString);
optSockets = 1;
break;
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'i':
optInfo = 1;
break;
case 'M': /* Set MSR Access mode */
CHECK_OPTION_STRING;
accessClient_setaccessmode(str2int((char*) argString->data));
bdestroy(argString);
break;
case 'p':
optClock = 1;
break;
case 's':
CHECK_OPTION_STRING;
optStethoscope = str2int((char*) argString->data);
bdestroy(argString);
break;
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case '?':
if (optopt == 's' || optopt == 'M' || optopt == 'c')
{
HELP_MSG;
}
else if (isprint (optopt))
{
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
}
else
{
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
}
exit( EXIT_FAILURE);
default:
HELP_MSG;
exit (EXIT_SUCCESS);
}
}
if (!lock_check())
{
fprintf(stderr,"Access to performance counters is locked.\n");
exit(EXIT_FAILURE);
}
if (optClock && optind == argc)
{
fprintf(stderr,"Commandline option -p requires an executable.\n");
exit(EXIT_FAILURE);
}
if (optSockets && !optStethoscope && optind == argc)
{
fprintf(stderr,"Commandline option -c requires an executable if not used in combination with -s.\n");
exit(EXIT_FAILURE);
}
if (cpuid_init() == EXIT_FAILURE)
{
fprintf(stderr, "CPU not supported\n");
exit(EXIT_FAILURE);
}
if (numSockets > cpuid_topology.numSockets)
{
fprintf(stderr, "System has only %d sockets but %d are given on commandline\n",
cpuid_topology.numSockets, numSockets);
exit(EXIT_FAILURE);
}
numa_init(); /* consider NUMA node as power unit for the moment */
accessClient_init(&socket_fd);
msr_init(socket_fd);
timer_init();
/* check for supported processors */
if ((cpuid_info.model == SANDYBRIDGE_EP) ||
(cpuid_info.model == SANDYBRIDGE) ||
(cpuid_info.model == IVYBRIDGE) ||
(cpuid_info.model == IVYBRIDGE_EP) ||
(cpuid_info.model == HASWELL) ||
(cpuid_info.model == NEHALEM_BLOOMFIELD) ||
(cpuid_info.model == NEHALEM_LYNNFIELD) ||
(cpuid_info.model == NEHALEM_WESTMERE))
{
power_init(numa_info.nodes[0].processors[0]);
}
else
{
fprintf (stderr, "Query Turbo Mode only supported on Intel Nehalem/Westmere/SandyBridge/IvyBridge/Haswell processors!\n");
exit(EXIT_FAILURE);
}
double clock = (double) timer_getCpuClock();
printf(HLINE);
printf("CPU name:\t%s \n",cpuid_info.name);
printf("CPU clock:\t%3.2f GHz \n", (float) clock * 1.E-09);
printf(HLINE);
if (optInfo)
{
if (power_info.turbo.numSteps != 0)
{
printf("Base clock:\t%.2f MHz \n", power_info.baseFrequency );
printf("Minimal clock:\t%.2f MHz \n", power_info.minFrequency );
printf("Turbo Boost Steps:\n");
for (int i=0; i < power_info.turbo.numSteps; i++ )
{
printf("C%d %.2f MHz \n",i+1, power_info.turbo.steps[i] );
}
}
printf(HLINE);
}
if (cpuid_info.model == SANDYBRIDGE_EP)
{
hasDRAM = 1;
}
else if ((cpuid_info.model != SANDYBRIDGE) &&
(cpuid_info.model != SANDYBRIDGE_EP) &&
(cpuid_info.model != IVYBRIDGE) &&
(cpuid_info.model != IVYBRIDGE_EP) &&
(cpuid_info.model != HASWELL))
{
fprintf (stderr, "RAPL not supported on this processor!\n");
exit(EXIT_FAILURE);
}
if (optInfo)
{
printf("Thermal Spec Power: %g Watts \n", power_info.tdp );
printf("Minimum Power: %g Watts \n", power_info.minPower);
printf("Maximum Power: %g Watts \n", power_info.maxPower);
printf("Maximum Time Window: %g micro sec \n", power_info.maxTimeWindow);
printf(HLINE);
exit(EXIT_SUCCESS);
}
if (optClock)
{
affinity_init();
argString = bformat("S%u:0-%u", threadsSockets[0], cpuid_topology.numCoresPerSocket-1);
for (int i=1; i<numSockets; i++)
{
bstring tExpr = bformat("@S%u:0-%u", threadsSockets[i], cpuid_topology.numCoresPerSocket-1);
bconcat(argString, tExpr);
}
numThreads = bstr_to_cpuset(threads, argString);
bdestroy(argString);
perfmon_init(numThreads, threads, stdout);
perfmon_setupEventSet(eventString, NULL);
}
{
PowerData pDataPkg[MAX_NUM_NODES*2];
PowerData pDataDram[MAX_NUM_NODES*2];
printf("Measure on sockets: %d", threadsSockets[0]);
for (int i=1; i<numSockets; i++)
{
printf(", %d", threadsSockets[i]);
}
printf("\n");
if (optStethoscope)
{
if (optClock)
{
perfmon_startCounters();
}
else
{
for (int i=0; i<numSockets; i++)
{
int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
if (hasDRAM) power_start(pDataDram+i, cpuId, DRAM);
power_start(pDataPkg+i, cpuId, PKG);
}
}
sleep(optStethoscope);
if (optClock)
{
perfmon_stopCounters();
perfmon_printCounterResults();
perfmon_finalize();
}
else
{
for (int i=0; i<numSockets; i++)
{
int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
power_stop(pDataPkg+i, cpuId, PKG);
if (hasDRAM) power_stop(pDataDram+i, cpuId, DRAM);
}
}
runtime = (double) optStethoscope;
}
else
{
TimerData time;
argv += optind;
bstring exeString = bfromcstr(argv[0]);
for (int i=1; i<(argc-optind); i++)
{
bconchar(exeString, ' ');
bcatcstr(exeString, argv[i]);
}
printf("%s\n",bdata(exeString));
if (optClock)
{
perfmon_startCounters();
}
else
{
for (int i=0; i<numSockets; i++)
{
int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
if (hasDRAM) power_start(pDataDram+i, cpuId, DRAM);
power_start(pDataPkg+i, cpuId, PKG);
}
timer_start(&time);
}
if (system(bdata(exeString)) == EOF)
{
fprintf(stderr, "Failed to execute %s!\n", bdata(exeString));
exit(EXIT_FAILURE);
}
if (optClock)
{
perfmon_stopCounters();
perfmon_printCounterResults();
perfmon_finalize();
}
else
{
timer_stop(&time);
for (int i=0; i<numSockets; i++)
{
int cpuId = numa_info.nodes[threadsSockets[i]].processors[0];
power_stop(pDataPkg+i, cpuId, PKG);
if (hasDRAM) power_stop(pDataDram+i, cpuId, DRAM);
}
runtime = timer_print(&time);
}
}
if (!optClock)
{
printf("Runtime: %g second \n",runtime);
printf(HLINE);
for (int i=0; i<numSockets; i++)
{
printf("Socket %d\n",threadsSockets[i]);
printf("Domain: PKG \n");
printf("Energy consumed: %g Joules \n", power_printEnergy(pDataPkg+i));
printf("Power consumed: %g Watts \n", power_printEnergy(pDataPkg+i) / runtime );
if (hasDRAM)
{
printf("Domain: DRAM \n");
printf("Energy consumed: %g Joules \n", power_printEnergy(pDataDram+i));
printf("Power consumed: %g Watts \n", power_printEnergy(pDataDram+i) / runtime );
}
printf("\n");
}
}
}
#if 0
if ( cpuid_hasFeature(TM2) )
{
thermal_init(0);
printf("Current core temperatures:\n");
for (uint32_t i = 0; i < cpuid_topology.numCoresPerSocket; i++ )
{
printf("Core %d: %u C\n",
numa_info.nodes[socketId].processors[i],
thermal_read(numa_info.nodes[socketId].processors[i]));
}
}
#endif
msr_finalize();
return EXIT_SUCCESS;
}
int main (int argc, char** argv)
{
int socket_fd = -1;
int optSetFeature = 0;
int cpuId = 0;
int c;
bstring argString;
CpuFeature feature = HW_PREFETCHER ;
while ((c = getopt (argc, argv, "c:s:u:hv")) != -1)
{
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'u':
optSetFeature = 2;
case 's':
if (! (argString = bSecureInput(20,optarg)))
{
fprintf(stderr,"Failed to read argument string!\n");
exit(EXIT_FAILURE);
}
if (biseqcstr(argString,"HW_PREFETCHER"))
{
feature = HW_PREFETCHER;
}
else if (biseqcstr(argString,"CL_PREFETCHER"))
{
feature = CL_PREFETCHER;
}
else if (biseqcstr(argString,"DCU_PREFETCHER"))
{
feature = DCU_PREFETCHER;
}
else if (biseqcstr(argString,"IP_PREFETCHER"))
{
feature = IP_PREFETCHER;
}
else
{
fprintf(stderr,"Feature not supported!\n");
exit(EXIT_FAILURE);
}
if (!optSetFeature)
{
optSetFeature = 1;
}
break;
case 'c':
if (! (argString = bSecureInput(10,optarg)))
{
fprintf(stderr,"Failed to read argument string!\n");
exit(EXIT_FAILURE);
}
cpuId = str2int((char*) argString->data);
break;
case '?':
if (isprint (optopt))
{
fprintf (stderr, "Unknown option `-%c'.\n", optopt);
}
else
{
fprintf (stderr,
"Unknown option character `\\x%x'.\n",
optopt);
}
return EXIT_FAILURE;
default:
HELP_MSG;
exit (EXIT_SUCCESS);
}
}
if (cpuid_init() == EXIT_FAILURE)
{
ERROR_PLAIN_PRINT(Unsupported processor!);
}
