void kernel_init(multiboot_info_t *mboot_info)
{
extern char __start_bss[], __stop_bss[];
memset(__start_bss, 0, __stop_bss - __start_bss);
/* mboot_info is a physical address. while some arches currently have the
* lower memory mapped, everyone should have it mapped at kernbase by now.
* also, it might be in 'free' memory, so once we start dynamically using
* memory, we may clobber it. */
multiboot_kaddr = (struct multiboot_info*)((physaddr_t)mboot_info
+ KERNBASE);
extract_multiboot_cmdline(multiboot_kaddr);
cons_init();
print_cpuinfo();
printk("Boot Command Line: '%s'\n", boot_cmdline);
exception_table_init();
cache_init(); // Determine systems's cache properties
pmem_init(multiboot_kaddr);
kmem_cache_init(); // Sets up slab allocator
kmalloc_init();
hashtable_init();
radix_init();
cache_color_alloc_init(); // Inits data structs
colored_page_alloc_init(); // Allocates colors for agnostic processes
acpiinit();
topology_init();
kthread_init(); /* might need to tweak when this happens */
vmr_init();
file_init();
page_check();
idt_init();
kernel_msg_init();
timer_init();
vfs_init();
devfs_init();
train_timing();
kb_buf_init(&cons_buf);
arch_init();
block_init();
enable_irq();
run_linker_funcs();
/* reset/init devtab after linker funcs 3 and 4. these run NIC and medium
* pre-inits, which need to happen before devether. */
devtabreset();
devtabinit();
#ifdef CONFIG_EXT2FS
mount_fs(&ext2_fs_type, "/dev/ramdisk", "/mnt", 0);
#endif /* CONFIG_EXT2FS */
#ifdef CONFIG_ETH_AUDIO
eth_audio_init();
#endif /* CONFIG_ETH_AUDIO */
get_coreboot_info(&sysinfo);
booting = 0;
#ifdef CONFIG_RUN_INIT_SCRIPT
if (run_init_script()) {
printk("Configured to run init script, but no script specified!\n");
manager();
}
#else
manager();
#endif
}
static int lua_likwid_getAffinityInfo(lua_State* L)
{
int i,j;
if (topology_isInitialized == 0)
{
topology_init();
topology_isInitialized = 1;
cpuinfo = get_cpuInfo();
cputopo = get_cpuTopology();
}
if ((topology_isInitialized) && (cpuinfo == NULL))
{
cpuinfo = get_cpuInfo();
}
if ((topology_isInitialized) && (cputopo == NULL))
{
cputopo = get_cpuTopology();
}
if (numa_isInitialized == 0)
{
if (numa_init() == 0)
{
numa_isInitialized = 1;
numainfo = get_numaTopology();
}
}
if ((numa_isInitialized) && (numainfo == NULL))
{
numainfo = get_numaTopology();
}
if (affinity_isInitialized == 0)
{
affinity_init();
affinity_isInitialized = 1;
affinity = get_affinityDomains();
}
if ((affinity_isInitialized) && (affinity == NULL))
{
affinity = get_affinityDomains();
}
if (!affinity)
{
lua_pushstring(L,"Cannot initialize affinity groups");
lua_error(L);
}
lua_newtable(L);
lua_pushstring(L,"numberOfAffinityDomains");
lua_pushunsigned(L,affinity->numberOfAffinityDomains);
lua_settable(L,-3);
lua_pushstring(L,"numberOfSocketDomains");
lua_pushunsigned(L,affinity->numberOfSocketDomains);
lua_settable(L,-3);
lua_pushstring(L,"numberOfNumaDomains");
lua_pushunsigned(L,affinity->numberOfNumaDomains);
lua_settable(L,-3);
lua_pushstring(L,"numberOfProcessorsPerSocket");
lua_pushunsigned(L,affinity->numberOfProcessorsPerSocket);
lua_settable(L,-3);
lua_pushstring(L,"numberOfCacheDomains");
lua_pushunsigned(L,affinity->numberOfCacheDomains);
lua_settable(L,-3);
lua_pushstring(L,"numberOfCoresPerCache");
lua_pushunsigned(L,affinity->numberOfCoresPerCache);
lua_settable(L,-3);
lua_pushstring(L,"numberOfProcessorsPerCache");
lua_pushunsigned(L,affinity->numberOfProcessorsPerCache);
lua_settable(L,-3);
lua_pushstring(L,"domains");
lua_newtable(L);
for(i=0;i<affinity->numberOfAffinityDomains;i++)
{
lua_pushunsigned(L, i+1);
lua_newtable(L);
lua_pushstring(L,"tag");
lua_pushstring(L,bdata(affinity->domains[i].tag));
lua_settable(L,-3);
lua_pushstring(L,"numberOfProcessors");
lua_pushunsigned(L,affinity->domains[i].numberOfProcessors);
lua_settable(L,-3);
lua_pushstring(L,"numberOfCores");
lua_pushunsigned(L,affinity->domains[i].numberOfCores);
lua_settable(L,-3);
lua_pushstring(L,"processorList");
lua_newtable(L);
for(j=0;j<affinity->domains[i].numberOfProcessors;j++)
{
lua_pushunsigned(L,j+1);
lua_pushunsigned(L,affinity->domains[i].processorList[j]);
lua_settable(L,-3);
}
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
return 1;
}
int main(int argc, char** argv)
{
uint64_t iter = 100;
uint32_t i;
uint32_t j;
int globalNumberOfThreads = 0;
int optPrintDomains = 0;
int c;
ThreadUserData myData;
bstring testcase = bfromcstr("none");
uint64_t numberOfWorkgroups = 0;
int tmp = 0;
double time;
double cycPerUp = 0.0;
const TestCase* test = NULL;
uint64_t realSize = 0;
uint64_t realIter = 0;
uint64_t maxCycles = 0;
uint64_t minCycles = UINT64_MAX;
uint64_t cyclesClock = 0;
uint64_t demandIter = 0;
TimerData itertime;
Workgroup* currentWorkgroup = NULL;
Workgroup* groups = NULL;
uint32_t min_runtime = 1; /* 1s */
bstring HLINE = bfromcstr("");
binsertch(HLINE, 0, 80, '-');
binsertch(HLINE, 80, 1, '\n');
int (*ownprintf)(const char *format, ...);
ownprintf = &printf;
/* Handling of command line options */
if (argc == 1)
{
HELP_MSG;
exit(EXIT_SUCCESS);
}
while ((c = getopt (argc, argv, "w:t:s:l:aphvi:")) != -1) {
switch (c)
{
case 'h':
HELP_MSG;
exit (EXIT_SUCCESS);
case 'v':
VERSION_MSG;
exit (EXIT_SUCCESS);
case 'a':
ownprintf(TESTS"\n");
exit (EXIT_SUCCESS);
case 'w':
numberOfWorkgroups++;
break;
case 's':
min_runtime = atoi(optarg);
break;
case 'i':
demandIter = strtoul(optarg, NULL, 10);
if (demandIter <= 0)
{
fprintf (stderr, "Error: Iterations must be greater than 0\n");
return EXIT_FAILURE;
}
break;
case 'l':
bdestroy(testcase);
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (test == NULL)
{
fprintf (stderr, "Error: Unknown test case %s\n",optarg);
return EXIT_FAILURE;
}
else
{
ownprintf("Name: %s\n",test->name);
ownprintf("Number of streams: %d\n",test->streams);
ownprintf("Loop stride: %d\n",test->stride);
ownprintf("Flops: %d\n",test->flops);
ownprintf("Bytes: %d\n",test->bytes);
switch (test->type)
{
case INT:
ownprintf("Data Type: Integer\n");
break;
case SINGLE:
ownprintf("Data Type: Single precision float\n");
break;
case DOUBLE:
ownprintf("Data Type: Double precision float\n");
break;
}
if (test->loads >= 0)
{
ownprintf("Load Ops: %d\n",test->loads);
}
if (test->stores >= 0)
{
ownprintf("Store Ops: %d\n",test->stores);
}
if (test->branches >= 0)
{
ownprintf("Branches: %d\n",test->branches);
}
if (test->instr_const >= 0)
{
ownprintf("Constant instructions: %d\n",test->instr_const);
}
if (test->instr_loop >= 0)
{
ownprintf("Loop instructions: %d\n",test->instr_loop);
}
}
bdestroy(testcase);
exit (EXIT_SUCCESS);
break;
case 'p':
optPrintDomains = 1;
break;
case 'g':
numberOfWorkgroups = LLU_CAST atol(optarg);
tmp = numberOfWorkgroups;
break;
case 't':
bdestroy(testcase);
testcase = bfromcstr(optarg);
for (i=0; i<NUMKERNELS; i++)
{
if (biseqcstr(testcase, kernels[i].name))
{
test = kernels+i;
break;
}
}
if (test == NULL)
{
fprintf (stderr, "Error: 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 ((numberOfWorkgroups == 0) && (!optPrintDomains))
{
fprintf(stderr, "Error: At least one workgroup (-w) must be set on commandline\n");
exit (EXIT_FAILURE);
}
if (topology_init() != EXIT_SUCCESS)
{
fprintf(stderr, "Error: Unsupported processor!\n");
exit(EXIT_FAILURE);
}
if ((test == NULL) && (!optPrintDomains))
{
fprintf(stderr, "Unknown test case. Please check likwid-bench -a for available tests\n");
fprintf(stderr, "and select one using the -t commandline option\n");
exit(EXIT_FAILURE);
}
numa_init();
affinity_init();
timer_init();
if (optPrintDomains)
{
bdestroy(testcase);
AffinityDomains_t affinity = get_affinityDomains();
ownprintf("Number of Domains %d\n",affinity->numberOfAffinityDomains);
for (i=0; i < affinity->numberOfAffinityDomains; i++ )
{
ownprintf("Domain %d:\n",i);
ownprintf("\tTag %s:",bdata(affinity->domains[i].tag));
for ( uint32_t j=0; j < affinity->domains[i].numberOfProcessors; j++ )
{
ownprintf(" %d",affinity->domains[i].processorList[j]);
}
ownprintf("\n");
}
exit (EXIT_SUCCESS);
}
allocator_init(numberOfWorkgroups * MAX_STREAMS);
groups = (Workgroup*) malloc(numberOfWorkgroups*sizeof(Workgroup));
tmp = 0;
optind = 0;
while ((c = getopt (argc, argv, "w:t:s:l:i:aphv")) != -1)
{
switch (c)
{
case 'w':
currentWorkgroup = groups+tmp;
bstring groupstr = bfromcstr(optarg);
i = bstr_to_workgroup(currentWorkgroup, groupstr, test->type, test->streams);
bdestroy(groupstr);
if (i == 0)
{
for (i=0; i< test->streams; i++)
{
if (currentWorkgroup->streams[i].offset%test->stride)
{
fprintf (stderr, "Error: 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);
}
tmp++;
}
else
{
exit(EXIT_FAILURE);
}
break;
default:
continue;
break;
}
}
/* :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;
}
ownprintf(bdata(HLINE));
ownprintf("LIKWID MICRO BENCHMARK\n");
ownprintf("Test: %s\n",test->name);
ownprintf(bdata(HLINE));
ownprintf("Using %" PRIu64 " work groups\n",numberOfWorkgroups);
ownprintf("Using %d threads\n",globalNumberOfThreads);
ownprintf(bdata(HLINE));
threads_init(globalNumberOfThreads);
threads_createGroups(numberOfWorkgroups);
/* we configure global barriers only */
barrier_init(1);
barrier_registerGroup(globalNumberOfThreads);
cyclesClock = timer_getCycleClock();
#ifdef LIKWID_PERFMON
if (getenv("LIKWID_FILEPATH") != NULL)
{
ownprintf("Using Likwid Marker API\n");
}
LIKWID_MARKER_INIT;
ownprintf(bdata(HLINE));
#endif
/* initialize data structures for threads */
for (i=0; i<numberOfWorkgroups; i++)
{
myData.iter = iter;
if (demandIter > 0)
{
myData.iter = demandIter;
}
myData.min_runtime = min_runtime;
myData.size = groups[i].size;
myData.test = test;
myData.cycles = 0;
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);
}
if (demandIter == 0)
{
getIterSingle((void*) &threads_data[0]);
for (i=0; i<numberOfWorkgroups; i++)
{
iter = threads_updateIterations(i, demandIter);
}
}
#ifdef DEBUG_LIKWID
else
{
ownprintf("Using manually selected iterations per thread\n");
}
#endif
threads_create(runTest);
threads_join();
for (int i=0; i<globalNumberOfThreads; i++)
{
realSize += threads_data[i].data.size;
realIter += threads_data[i].data.iter;
if (threads_data[i].cycles > maxCycles)
{
maxCycles = threads_data[i].cycles;
}
if (threads_data[i].cycles < minCycles)
{
minCycles = threads_data[i].cycles;
}
}
time = (double) maxCycles / (double) cyclesClock;
ownprintf(bdata(HLINE));
ownprintf("Cycles:\t\t\t%" PRIu64 "\n", maxCycles);
ownprintf("CPU Clock:\t\t%" PRIu64 "\n", timer_getCpuClock());
ownprintf("Cycle Clock:\t\t%" PRIu64 "\n", cyclesClock);
ownprintf("Time:\t\t\t%e sec\n", time);
ownprintf("Iterations:\t\t%" PRIu64 "\n", realIter);
ownprintf("Iterations per thread:\t%" PRIu64 "\n",threads_data[0].data.iter);
ownprintf("Inner loop executions:\t%.0f\n", ((double)realSize)/((double)test->stride));
ownprintf("Size:\t\t\t%" PRIu64 "\n", realSize*test->bytes );
ownprintf("Size per thread:\t%" PRIu64 "\n", threads_data[0].data.size*test->bytes);
ownprintf("Number of Flops:\t%" PRIu64 "\n", (threads_data[0].data.iter * realSize * test->flops));
ownprintf("MFlops/s:\t\t%.2f\n",
1.0E-06 * ((double) threads_data[0].data.iter * realSize * test->flops/ time));
ownprintf("Data volume (Byte):\t%llu\n", LLU_CAST (threads_data[0].data.iter * realSize * test->bytes));
ownprintf("MByte/s:\t\t%.2f\n",
1.0E-06 * ( (double) threads_data[0].data.iter * realSize * test->bytes/ time));
cycPerUp = ((double) maxCycles / (double) (threads_data[0].data.iter * realSize));
ownprintf("Cycles per update:\t%f\n", cycPerUp);
switch ( test->type )
{
case INT:
case SINGLE:
ownprintf("Cycles per cacheline:\t%f\n", (16.0 * cycPerUp));
break;
case DOUBLE:
ownprintf("Cycles per cacheline:\t%f\n", (8.0 * cycPerUp));
break;
}
ownprintf("Loads per update:\t%ld\n", test->loads );
ownprintf("Stores per update:\t%ld\n", test->stores );
if ((test->loads > 0) && (test->stores > 0))
{
ownprintf("Load/store ratio:\t%.2f\n", ((double)test->loads)/((double)test->stores) );
}
if ((test->instr_loop > 0) && (test->instr_const > 0))
{
ownprintf("Instructions:\t\t%" PRIu64 "\n", LLU_CAST ((double)realSize/test->stride)*test->instr_loop*threads_data[0].data.iter + test->instr_const );
}
if (test->uops > 0)
{
ownprintf("UOPs:\t\t\t%" PRIu64 "\n", LLU_CAST ((double)realSize/test->stride)*test->uops*threads_data[0].data.iter);
}
ownprintf(bdata(HLINE));
threads_destroy(numberOfWorkgroups, test->streams);
allocator_finalize();
workgroups_destroy(&groups, numberOfWorkgroups, test->streams);
#ifdef LIKWID_PERFMON
if (getenv("LIKWID_FILEPATH") != NULL)
{
ownprintf("Writing Likwid Marker API results to file %s\n", getenv("LIKWID_FILEPATH"));
}
LIKWID_MARKER_CLOSE;
#endif
bdestroy(HLINE);
return EXIT_SUCCESS;
}
static int lua_likwid_getEventsAndCounters(lua_State* L)
{
int i;
char optString[1024];
int optStringIndex = 0;
if (topology_isInitialized == 0)
{
topology_init();
topology_isInitialized = 1;
cpuinfo = get_cpuInfo();
}
if ((topology_isInitialized) && (cpuinfo == NULL))
{
cpuinfo = get_cpuInfo();
}
perfmon_init_maps();
lua_newtable(L);
lua_pushstring(L,"Counters");
lua_newtable(L);
for(i=1;i<=perfmon_numCounters;i++)
{
optStringIndex = 0;
optString[0] = '\0';
lua_pushunsigned(L,i);
lua_newtable(L);
lua_pushstring(L,"Name");
lua_pushstring(L,counter_map[i-1].key);
lua_settable(L,-3);
lua_pushstring(L,"Options");
for(int j=1; j<NUM_EVENT_OPTIONS; j++)
{
if (counter_map[i-1].optionMask & REG_TYPE_MASK(j))
{
optStringIndex += sprintf(&(optString[optStringIndex]), "%s|", eventOptionTypeName[j]);
}
}
optString[optStringIndex-1] = '\0';
lua_pushstring(L,optString);
lua_settable(L,-3);
lua_pushstring(L,"Type");
lua_pushunsigned(L, counter_map[i-1].type);
lua_settable(L,-3);
lua_pushstring(L,"TypeName");
lua_pushstring(L, RegisterTypeNames[counter_map[i-1].type]);
lua_settable(L,-3);
lua_pushstring(L,"Index");
lua_pushunsigned(L,counter_map[i-1].index);
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
lua_pushstring(L,"Events");
lua_newtable(L);
for(i=1;i<=perfmon_numArchEvents;i++)
{
optStringIndex = 0;
optString[0] = '\0';
lua_pushunsigned(L,i);
lua_newtable(L);
lua_pushstring(L,"Name");
lua_pushstring(L,eventHash[i-1].name);
lua_settable(L,-3);
lua_pushstring(L,"ID");
lua_pushunsigned(L,eventHash[i-1].eventId);
lua_settable(L,-3);
lua_pushstring(L,"UMask");
lua_pushunsigned(L,eventHash[i-1].umask);
lua_settable(L,-3);
lua_pushstring(L,"Limit");
lua_pushstring(L,eventHash[i-1].limit);
lua_settable(L,-3);
lua_pushstring(L,"Options");
for(int j=1; j<NUM_EVENT_OPTIONS; j++)
{
if (eventHash[i-1].optionMask & REG_TYPE_MASK(j))
{
optStringIndex += sprintf(&(optString[optStringIndex]), "%s|", eventOptionTypeName[j]);
}
}
optString[optStringIndex-1] = '\0';
lua_pushstring(L,optString);
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
return 1;
}
static int lua_likwid_getNumaInfo(lua_State* L)
{
uint32_t i,j;
if (topology_isInitialized == 0)
{
topology_init();
topology_isInitialized = 1;
cpuinfo = get_cpuInfo();
cputopo = get_cpuTopology();
}
if ((topology_isInitialized) && (cpuinfo == NULL))
{
cpuinfo = get_cpuInfo();
}
if ((topology_isInitialized) && (cputopo == NULL))
{
cputopo = get_cpuTopology();
}
if (numa_isInitialized == 0)
{
if (numa_init() == 0)
{
numa_isInitialized = 1;
numainfo = get_numaTopology();
}
else
{
lua_newtable(L);
lua_pushstring(L,"numberOfNodes");
lua_pushunsigned(L,0);
lua_settable(L,-3);
lua_pushstring(L,"nodes");
lua_newtable(L);
lua_settable(L,-3);
return 1;
}
}
if ((numa_isInitialized) && (numainfo == NULL))
{
numainfo = get_numaTopology();
}
if (affinity_isInitialized == 0)
{
affinity_init();
affinity_isInitialized = 1;
affinity = get_affinityDomains();
}
if ((affinity_isInitialized) && (affinity == NULL))
{
affinity = get_affinityDomains();
}
lua_newtable(L);
lua_pushstring(L,"numberOfNodes");
lua_pushunsigned(L,numainfo->numberOfNodes);
lua_settable(L,-3);
lua_pushstring(L,"nodes");
lua_newtable(L);
for(i=0;i<numainfo->numberOfNodes;i++)
{
lua_pushinteger(L, i+1);
lua_newtable(L);
lua_pushstring(L,"id");
lua_pushunsigned(L,numainfo->nodes[i].id);
lua_settable(L,-3);
lua_pushstring(L,"totalMemory");
lua_pushunsigned(L,numainfo->nodes[i].totalMemory);
lua_settable(L,-3);
lua_pushstring(L,"freeMemory");
lua_pushunsigned(L,numainfo->nodes[i].freeMemory);
lua_settable(L,-3);
lua_pushstring(L,"numberOfProcessors");
lua_pushunsigned(L,numainfo->nodes[i].numberOfProcessors);
lua_settable(L,-3);
lua_pushstring(L,"numberOfDistances");
lua_pushunsigned(L,numainfo->nodes[i].numberOfDistances);
lua_settable(L,-3);
lua_pushstring(L,"processors");
lua_newtable(L);
for(j=0;j<numainfo->nodes[i].numberOfProcessors;j++)
{
lua_pushunsigned(L,j+1);
lua_pushunsigned(L,numainfo->nodes[i].processors[j]);
lua_settable(L,-3);
}
lua_settable(L,-3);
/*lua_pushstring(L,"processorsCompact");
lua_newtable(L);
for(j=0;j<numa->nodes[i].numberOfProcessors;j++)
{
lua_pushunsigned(L,j);
lua_pushunsigned(L,numa->nodes[i].processorsCompact[j]);
lua_settable(L,-3);
}
lua_settable(L,-3);*/
lua_pushstring(L,"distances");
lua_newtable(L);
for(j=0;j<numainfo->nodes[i].numberOfDistances;j++)
{
lua_pushinteger(L,j+1);
lua_newtable(L);
lua_pushinteger(L,j);
lua_pushunsigned(L,numainfo->nodes[i].distances[j]);
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
return 1;
}
static int lua_likwid_getPowerInfo(lua_State* L)
{
int i;
if (topology_isInitialized == 0)
{
topology_init();
topology_isInitialized = 1;
cpuinfo = get_cpuInfo();
cputopo = get_cpuTopology();
}
if ((topology_isInitialized) && (cpuinfo == NULL))
{
cpuinfo = get_cpuInfo();
}
if ((topology_isInitialized) && (cputopo == NULL))
{
cputopo = get_cpuTopology();
}
if (power_isInitialized == 0)
{
power_hasRAPL = power_init(0);
if (power_hasRAPL)
{
power_isInitialized = 1;
power = get_powerInfo();
}
else
{
return 0;
}
}
lua_newtable(L);
lua_pushstring(L,"hasRAPL");
lua_pushboolean(L,power_hasRAPL);
lua_settable(L,-3);
lua_pushstring(L,"baseFrequency");
lua_pushnumber(L,power->baseFrequency);
lua_settable(L,-3);
lua_pushstring(L,"minFrequency");
lua_pushnumber(L,power->minFrequency);
lua_settable(L,-3);
lua_pushstring(L,"powerUnit");
lua_pushnumber(L,power->powerUnit);
lua_settable(L,-3);
lua_pushstring(L,"timeUnit");
lua_pushnumber(L,power->timeUnit);
lua_settable(L,-3);
lua_pushstring(L,"turbo");
lua_newtable(L);
lua_pushstring(L,"numSteps");
lua_pushunsigned(L,power->turbo.numSteps);
lua_settable(L,-3);
lua_pushstring(L,"steps");
lua_newtable(L);
for(i=0;i<power->turbo.numSteps;i++)
{
lua_pushunsigned(L,i+1);
lua_pushnumber(L,power->turbo.steps[i]);
lua_settable(L,-3);
}
lua_settable(L,-3);
lua_settable(L,-3);
lua_pushstring(L,"domains");
lua_newtable(L);
for(i=0;i<NUM_POWER_DOMAINS;i++)
{
lua_pushstring(L,power_names[i]);
lua_newtable(L);
lua_pushstring(L, "ID");
lua_pushnumber(L, power->domains[i].type);
lua_settable(L,-3);
lua_pushstring(L, "energyUnit");
lua_pushnumber(L, power->domains[i].energyUnit);
lua_settable(L,-3);
lua_pushstring(L,"supportStatus");
if (power->domains[i].supportFlags & POWER_DOMAIN_SUPPORT_STATUS)
{
lua_pushboolean(L, 1);
}
else
{
lua_pushboolean(L, 0);
}
lua_settable(L,-3);
lua_pushstring(L,"supportPerf");
if (power->domains[i].supportFlags & POWER_DOMAIN_SUPPORT_PERF)
{
lua_pushboolean(L, 1);
}
else
{
lua_pushboolean(L, 0);
}
lua_settable(L,-3);
lua_pushstring(L,"supportPolicy");
if (power->domains[i].supportFlags & POWER_DOMAIN_SUPPORT_POLICY)
{
lua_pushboolean(L, 1);
}
else
{
lua_pushboolean(L, 0);
}
lua_settable(L,-3);
lua_pushstring(L,"supportLimit");
if (power->domains[i].supportFlags & POWER_DOMAIN_SUPPORT_LIMIT)
{
lua_pushboolean(L, 1);
}
else
{
lua_pushboolean(L, 0);
}
lua_settable(L,-3);
if (power->domains[i].supportFlags & POWER_DOMAIN_SUPPORT_INFO)
{
lua_pushstring(L,"supportInfo");
lua_pushboolean(L, 1);
lua_settable(L,-3);
lua_pushstring(L,"tdp");
lua_pushnumber(L, power->domains[i].tdp);
lua_settable(L,-3);
lua_pushstring(L,"minPower");
lua_pushnumber(L, power->domains[i].minPower);
lua_settable(L,-3);
lua_pushstring(L,"maxPower");
lua_pushnumber(L, power->domains[i].maxPower);
lua_settable(L,-3);
lua_pushstring(L,"maxTimeWindow");
lua_pushnumber(L, power->domains[i].maxTimeWindow);
lua_settable(L,-3);
}
else
{
lua_pushstring(L,"supportInfo");
lua_pushboolean(L, 0);
lua_settable(L,-3);
}
lua_settable(L,-3);
}
lua_settable(L,-3);
return 1;
}
static int lua_likwid_getCpuTopology(lua_State* L)
{
int i;
TreeNode* socketNode;
int socketCount = 0;
TreeNode* coreNode;
int coreCount = 0;
TreeNode* threadNode;
int threadCount = 0;
if (topology_isInitialized == 0)
{
topology_init();
topology_isInitialized = 1;
cputopo = get_cpuTopology();
}
if ((topology_isInitialized) && (cputopo == NULL))
{
cputopo = get_cpuTopology();
}
if (numa_isInitialized == 0)
{
if (numa_init() == 0)
{
numa_isInitialized = 1;
numainfo = get_numaTopology();
}
}
if ((numa_isInitialized) && (numainfo == NULL))
{
numainfo = get_numaTopology();
}
lua_newtable(L);
lua_pushstring(L,"numHWThreads");
lua_pushunsigned(L,cputopo->numHWThreads);
lua_settable(L,-3);
lua_pushstring(L,"activeHWThreads");
lua_pushunsigned(L,cputopo->activeHWThreads);
lua_settable(L,-3);
lua_pushstring(L,"numSockets");
lua_pushunsigned(L,cputopo->numSockets);
lua_settable(L,-3);
lua_pushstring(L,"numCoresPerSocket");
lua_pushunsigned(L,cputopo->numCoresPerSocket);
lua_settable(L,-3);
lua_pushstring(L,"numThreadsPerCore");
lua_pushunsigned(L,cputopo->numThreadsPerCore);
lua_settable(L,-3);
lua_pushstring(L,"numCacheLevels");
lua_pushinteger(L,cputopo->numCacheLevels);
lua_settable(L,-3);
lua_pushstring(L,"threadPool");
lua_newtable(L);
for(i=0;i<cputopo->numHWThreads;i++)
{
lua_pushnumber(L,i);
lua_newtable(L);
lua_pushstring(L,"threadId");
lua_pushunsigned(L,cputopo->threadPool[i].threadId);
lua_settable(L,-3);
lua_pushstring(L,"coreId");
lua_pushunsigned(L,cputopo->threadPool[i].coreId);
lua_settable(L,-3);
lua_pushstring(L,"packageId");
lua_pushunsigned(L,cputopo->threadPool[i].packageId);
lua_settable(L,-3);
lua_pushstring(L,"apicId");
lua_pushunsigned(L,cputopo->threadPool[i].apicId);
lua_settable(L,-3);
lua_pushstring(L,"inCpuSet");
lua_pushunsigned(L,cputopo->threadPool[i].inCpuSet);
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
lua_pushstring(L,"cacheLevels");
lua_newtable(L);
for(i=0;i<cputopo->numCacheLevels;i++)
{
lua_pushnumber(L,i+1);
lua_newtable(L);
lua_pushstring(L,"level");
lua_pushunsigned(L,cputopo->cacheLevels[i].level);
lua_settable(L,-3);
lua_pushstring(L,"associativity");
lua_pushunsigned(L,cputopo->cacheLevels[i].associativity);
lua_settable(L,-3);
lua_pushstring(L,"sets");
lua_pushunsigned(L,cputopo->cacheLevels[i].sets);
lua_settable(L,-3);
lua_pushstring(L,"lineSize");
lua_pushunsigned(L,cputopo->cacheLevels[i].lineSize);
lua_settable(L,-3);
lua_pushstring(L,"size");
lua_pushunsigned(L,cputopo->cacheLevels[i].size);
lua_settable(L,-3);
lua_pushstring(L,"threads");
lua_pushunsigned(L,cputopo->cacheLevels[i].threads);
lua_settable(L,-3);
lua_pushstring(L,"inclusive");
lua_pushunsigned(L,cputopo->cacheLevels[i].inclusive);
lua_settable(L,-3);
lua_pushstring(L,"type");
switch (cputopo->cacheLevels[i].type)
{
case DATACACHE:
lua_pushstring(L,"DATACACHE");
break;
case INSTRUCTIONCACHE:
lua_pushstring(L,"INSTRUCTIONCACHE");
break;
case UNIFIEDCACHE:
lua_pushstring(L,"UNIFIEDCACHE");
break;
case ITLB:
lua_pushstring(L,"ITLB");
break;
case DTLB:
lua_pushstring(L,"DTLB");
break;
case NOCACHE:
default:
lua_pushstring(L,"NOCACHE");
break;
}
lua_settable(L,-3);
lua_settable(L,-3);
}
lua_settable(L,-3);
lua_pushstring(L,"topologyTree");
lua_newtable(L);
socketNode = tree_getChildNode(cputopo->topologyTree);
while (socketNode != NULL)
{
lua_pushinteger(L, socketCount);
lua_newtable(L);
lua_pushstring(L, "ID");
lua_pushunsigned(L,socketNode->id);
lua_settable(L, -3);
lua_pushstring(L, "Childs");
lua_newtable(L);
coreCount = 0;
coreNode = tree_getChildNode(socketNode);
while (coreNode != NULL)
{
lua_pushinteger(L, coreCount);
lua_newtable(L);
lua_pushstring(L, "ID");
lua_pushunsigned(L,coreNode->id);
lua_settable(L,-3);
lua_pushstring(L, "Childs");
lua_newtable(L);
threadNode = tree_getChildNode(coreNode);
threadCount = 0;
while (threadNode != NULL)
{
lua_pushunsigned(L,threadCount);
lua_pushunsigned(L,threadNode->id);
lua_settable(L,-3);
threadNode = tree_getNextNode(threadNode);
threadCount++;
}
lua_settable(L,-3);
coreNode = tree_getNextNode(coreNode);
coreCount++;
lua_settable(L,-3);
}
lua_settable(L,-3);
socketNode = tree_getNextNode(socketNode);
socketCount++;
lua_settable(L,-3);
}
lua_settable(L,-3);
return 1;
}
int main(int argc, char* argv[])
{
int i, j;
int err;
int* cpus;
int gid;
double result = 0.0;
char estr[] = "INSTR_RETIRED_ANY:FIXC0,CPU_CLK_UNHALTED_CORE:FIXC1,CPU_CLK_UNHALTED_REF:FIXC2,TEMP_CORE:TMP0";
// Load the topology module and print some values.
err = topology_init();
if (err < 0)
{
printf("Failed to initialize LIKWID's topology module\n");
return 1;
}
// CpuInfo_t contains global information like name, CPU family, ...
CpuInfo_t info = get_cpuInfo();
// CpuTopology_t contains information about the topology of the CPUs.
CpuTopology_t topo = get_cpuTopology();
// Create affinity domains. Commonly only needed when reading Uncore counters
//affinity_init();
printf("Likwid example on a %s with %d CPUs\n", info->name, topo->numHWThreads);
cpus = (int*)malloc(topo->numHWThreads * sizeof(int));
if (!cpus)
return 1;
for (i=0;i<topo->numHWThreads;i++)
{
cpus[i] = topo->threadPool[i].apicId;
}
// Must be called before perfmon_init() but only if you want to use another
// access mode as the pre-configured one. For direct access (0) you have to
// be root.
//accessClient_setaccessmode(0);
// Initialize the perfmon module.
err = perfmon_init(topo->numHWThreads, cpus);
if (err < 0)
{
printf("Failed to initialize LIKWID's performance monitoring module\n");
topology_finalize();
return 1;
}
// Add eventset string to the perfmon module.
gid = perfmon_addEventSet(estr);
if (gid < 0)
{
printf("Failed to add event string %s to LIKWID's performance monitoring module\n", estr);
perfmon_finalize();
topology_finalize();
return 1;
}
// Setup the eventset identified by group ID (gid).
err = perfmon_setupCounters(gid);
if (err < 0)
{
printf("Failed to setup group %d in LIKWID's performance monitoring module\n", gid);
perfmon_finalize();
topology_finalize();
return 1;
}
// Start all counters in the previously set up event set.
err = perfmon_startCounters();
if (err < 0)
{
printf("Failed to start counters for group %d for thread %d\n",gid, (-1*err)-1);
perfmon_finalize();
topology_finalize();
return 1;
}
// Perform something
sleep(2);
// Stop all counters in the previously started event set.
err = perfmon_stopCounters();
if (err < 0)
{
printf("Failed to stop counters for group %d for thread %d\n",gid, (-1*err)-1);
perfmon_finalize();
topology_finalize();
return 1;
}
// Print the result of every thread/CPU for all events in estr.
char* ptr = strtok(estr,",");
j = 0;
while (ptr != NULL)
{
for (i = 0;i < topo->numHWThreads; i++)
{
result = perfmon_getResult(gid, j, cpus[i]);
printf("Measurement result for event set %s at CPU %d: %f\n", ptr, cpus[i], result);
}
ptr = strtok(NULL,",");
j++;
}
// Uninitialize the perfmon module.
perfmon_finalize();
// Uninitialize the topology module.
topology_finalize();
return 0;
}
int cpustr_to_cpulist(char* cpustring, int* cpulist, int length)
{
int insert = 0;
int len = 0;
int ret = 0;
bstring bcpustr = bfromcstr(cpustring);
struct bstrList* strlist = bstrListCreate();
bstring scattercheck = bformat("scatter");
topology_init();
CpuTopology_t cpuid_topology = get_cpuTopology();
strlist = bsplit(bcpustr, '@');
int* tmpList = (int*)malloc(length * sizeof(int));
if (tmpList == NULL)
{
bstrListDestroy(strlist);
bdestroy(scattercheck);
bdestroy(bcpustr);
return -ENOMEM;
}
for (int i=0; i< strlist->qty; i++)
{
if (binstr(strlist->entry[i], 0, scattercheck) != BSTR_ERR)
{
ret = cpustr_to_cpulist_scatter(strlist->entry[i], tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
}
else if (bstrchrp(strlist->entry[i], 'E', 0) == 0)
{
ret = cpustr_to_cpulist_expression(strlist->entry[i], tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
}
else if (bstrchrp(strlist->entry[i], 'L', 0) == 0)
{
ret = cpustr_to_cpulist_logical(strlist->entry[i], tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
}
else if (cpuid_topology->activeHWThreads < cpuid_topology->numHWThreads)
{
fprintf(stdout, "INFO: You are running LIKWID in a cpuset with %d CPUs, only logical numbering allowed", cpuid_topology->activeHWThreads);
if (((bstrchrp(strlist->entry[i], 'N', 0) == 0) ||
(bstrchrp(strlist->entry[i], 'S', 0) == 0) ||
(bstrchrp(strlist->entry[i], 'C', 0) == 0) ||
(bstrchrp(strlist->entry[i], 'M', 0) == 0)) &&
(bstrchrp(strlist->entry[i], ':', 0) != BSTR_ERR))
{
bstring newstr = bformat("L:");
bconcat(newstr, strlist->entry[i]);
ret = cpustr_to_cpulist_logical(newstr, tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
bdestroy(newstr);
}
else
{
bstring newstr = bformat("L:N:");
bconcat(newstr, strlist->entry[i]);
ret = cpustr_to_cpulist_logical(newstr, tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
bdestroy(newstr);
}
}
if (((bstrchrp(strlist->entry[i], 'N', 0) == 0) ||
(bstrchrp(strlist->entry[i], 'S', 0) == 0) ||
(bstrchrp(strlist->entry[i], 'C', 0) == 0) ||
(bstrchrp(strlist->entry[i], 'M', 0) == 0)) &&
(bstrchrp(strlist->entry[i], ':', 0) != BSTR_ERR))
{
bstring newstr = bformat("L:");
bconcat(newstr, strlist->entry[i]);
ret = cpustr_to_cpulist_logical(newstr, tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
bdestroy(newstr);
}
else
{
ret = cpustr_to_cpulist_physical(strlist->entry[i], tmpList, length);
insert += cpulist_concat(cpulist, insert, tmpList, ret);
}
}
free(tmpList);
bstrListDestroy(strlist);
return insert;
}
static int cpustr_to_cpulist_physical(bstring bcpustr, int* cpulist, int length)
{
topology_init();
CpuTopology_t cpuid_topology = get_cpuTopology();
affinity_init();
AffinityDomains_t affinity = get_affinityDomains();
bstring bdomain;
bstring blist;
int domainidx = -1;
if (bstrchrp(bcpustr, ':', 0) != BSTR_ERR)
{
struct bstrList* strlist = bstrListCreate();
strlist = bsplit(bcpustr, ':');
bdomain = bstrcpy(strlist->entry[0]);
blist = bstrcpy(strlist->entry[1]);
bstrListDestroy(strlist);
}
else
{
bdomain = bformat("N");
blist = bstrcpy(bcpustr);
}
for (int i=0; i<affinity->numberOfAffinityDomains; i++)
{
if (bstrcmp(bdomain, affinity->domains[i].tag) == 0)
{
domainidx = i;
break;
}
}
if (domainidx < 0)
{
fprintf(stderr, "Cannot find domain %s\n", bdata(bdomain));
bdestroy(bdomain);
bdestroy(blist);
return 0;
}
struct bstrList* strlist = bstrListCreate();
strlist = bsplit(blist, ',');
int insert = 0;
for (int i=0;i< strlist->qty; i++)
{
if (bstrchrp(strlist->entry[i], '-', 0) != BSTR_ERR)
{
struct bstrList* indexlist = bstrListCreate();
indexlist = bsplit(strlist->entry[i], '-');
for (int j=atoi(bdata(indexlist->entry[0])); j<=atoi(bdata(indexlist->entry[1]));j++)
{
if (cpu_in_domain(domainidx, j))
{
cpulist[insert] = j;
insert++;
if (insert == length)
{
bstrListDestroy(indexlist);
goto physical_done;
}
}
else
{
fprintf(stderr, "CPU %d not in domain %s\n", j, bdata(affinity->domains[domainidx].tag));
}
}
bstrListDestroy(indexlist);
}
else
{
int cpu = atoi(bdata(strlist->entry[i]));
if (cpu_in_domain(domainidx, cpu))
{
cpulist[insert] = cpu;
insert++;
if (insert == length)
{
goto physical_done;
}
}
else
{
fprintf(stderr, "CPU %d not in domain %s\n", cpu, bdata(affinity->domains[domainidx].tag));
}
}
}
physical_done:
bstrListDestroy(strlist);
bdestroy(bdomain);
bdestroy(blist);
return insert;
}
static int cpustr_to_cpulist_logical(bstring bcpustr, int* cpulist, int length)
{
topology_init();
CpuTopology_t cpuid_topology = get_cpuTopology();
affinity_init();
AffinityDomains_t affinity = get_affinityDomains();
int domainidx = -1;
bstring bdomain;
bstring blist;
if (bstrchrp(bcpustr, 'L', 0) != 0)
{
fprintf(stderr, "Not a valid CPU expression\n");
return 0;
}
struct bstrList* strlist = bstrListCreate();
strlist = bsplit(bcpustr, ':');
if (strlist->qty != 3)
{
fprintf(stderr, "ERROR: Invalid expression, should look like L:<domain>:<indexlist> or be in a cpuset\n");
bstrListDestroy(strlist);
return 0;
}
bdomain = bstrcpy(strlist->entry[1]);
blist = bstrcpy(strlist->entry[2]);
bstrListDestroy(strlist);
for (int i=0; i<affinity->numberOfAffinityDomains; i++)
{
if (bstrcmp(bdomain, affinity->domains[i].tag) == 0)
{
domainidx = i;
break;
}
}
if (domainidx < 0)
{
printf("Cannot find domain %s\n", bdata(bdomain));
return 0;
}
int *inlist = malloc(affinity->domains[domainidx].numberOfProcessors * sizeof(int));
if (inlist == NULL)
{
return -ENOMEM;
}
int ret = cpulist_sort(affinity->domains[domainidx].processorList, inlist, affinity->domains[domainidx].numberOfProcessors);
strlist = bstrListCreate();
strlist = bsplit(blist, ',');
int insert = 0;
for (int i=0; i< strlist->qty; i++)
{
if (bstrchrp(strlist->entry[i], '-', 0) != BSTR_ERR)
{
struct bstrList* indexlist = bstrListCreate();
indexlist = bsplit(strlist->entry[i], '-');
for (int j=atoi(bdata(indexlist->entry[0])); j<=atoi(bdata(indexlist->entry[1]));j++)
{
cpulist[insert] = inlist[j];
insert++;
if (insert == length)
{
bstrListDestroy(indexlist);
goto logical_done;
}
}
bstrListDestroy(indexlist);
}
else
{
cpulist[insert] = inlist[atoi(bdata(strlist->entry[i]))];
insert++;
if (insert == length)
{
goto logical_done;
}
}
}
logical_done:
free(inlist);
bstrListDestroy(strlist);
return insert;
}
static int cpustr_to_cpulist_expression(bstring bcpustr, int* cpulist, int length)
{
topology_init();
CpuTopology_t cpuid_topology = get_cpuTopology();
affinity_init();
AffinityDomains_t affinity = get_affinityDomains();
bstring bdomain;
int domainidx = -1;
int count = 0;
int stride = 0;
int chunk = 0;
if (bstrchrp(bcpustr, 'E', 0) != 0)
{
fprintf(stderr, "Not a valid CPU expression\n");
return 0;
}
struct bstrList* strlist = bstrListCreate();
strlist = bsplit(bcpustr, ':');
if (strlist->qty == 3)
{
bdomain = bstrcpy(strlist->entry[1]);
count = atoi(bdata(strlist->entry[2]));
stride = 1;
chunk = 1;
}
else if (strlist->qty == 5)
{
bdomain = bstrcpy(strlist->entry[1]);
count = atoi(bdata(strlist->entry[2]));
chunk = atoi(bdata(strlist->entry[3]));
stride = atoi(bdata(strlist->entry[4]));
}
for (int i=0; i<affinity->numberOfAffinityDomains; i++)
{
if (bstrcmp(bdomain, affinity->domains[i].tag) == 0)
{
domainidx = i;
break;
}
}
if (domainidx < 0)
{
fprintf(stderr, "Cannot find domain %s\n", bdata(bdomain));
bstrListDestroy(strlist);
return 0;
}
int offset = 0;
int insert = 0;
for (int i=0;i<count;i++)
{
for (int j=0;j<chunk && offset+j<affinity->domains[domainidx].numberOfProcessors;j++)
{
cpulist[insert] = affinity->domains[domainidx].processorList[offset + j];
insert++;
if (insert == length)
goto expression_done;
}
offset += stride;
if (offset >= affinity->domains[domainidx].numberOfProcessors)
{
offset = 0;
}
if (insert >= count)
goto expression_done;
}
bstrListDestroy(strlist);
return 0;
expression_done:
bstrListDestroy(strlist);
return insert;
}
