bstring parse_workgroup(Workgroup* group, const_bstring str, DataType type)
{
CpuTopology_t topo;
struct bstrList* tokens;
bstring cpustr;
int numThreads = 0;
bstring domain;
tokens = bsplit(str,':');
if (tokens->qty == 2)
{
topo = get_cpuTopology();
numThreads = topo->activeHWThreads;
cpustr = bformat("E:%s:%d", bdata(tokens->entry[0]), numThreads );
}
else if (tokens->qty == 3)
{
cpustr = bformat("E:%s:%s", bdata(tokens->entry[0]), bdata(tokens->entry[2]));
numThreads = str2int(bdata(tokens->entry[2]));
if (numThreads < 0)
{
fprintf(stderr, "Cannot convert %s to integer\n", bdata(tokens->entry[2]));
bstrListDestroy(tokens);
return NULL;
}
}
else if (tokens->qty == 5)
{
cpustr = bformat("E:%s:%s:%s:%s", bdata(tokens->entry[0]),
bdata(tokens->entry[2]),
bdata(tokens->entry[3]),
bdata(tokens->entry[4]));
numThreads = str2int(bdata(tokens->entry[2]));
if (numThreads < 0)
{
fprintf(stderr, "Cannot convert %s to integer\n", bdata(tokens->entry[2]));
bstrListDestroy(tokens);
return NULL;
}
}
else
{
fprintf(stderr, "Misformated workgroup string\n");
bstrListDestroy(tokens);
return NULL;
}
group->size = bstr_to_doubleSize(tokens->entry[1], type);
if (group->size == 0)
{
fprintf(stderr, "Stream size cannot be read, should look like <domain>:<size>\n");
bstrListDestroy(tokens);
return NULL;
}
group->processorIds = (int*) malloc(numThreads * sizeof(int));
if (group->processorIds == NULL)
{
fprintf(stderr, "No more memory to allocate list of processors\n");
bstrListDestroy(tokens);
return NULL;
}
group->numberOfThreads = numThreads;
if (cpustr_to_cpulist(bdata(cpustr),group->processorIds, numThreads) < 0 )
{
free(group->processorIds);
bstrListDestroy(tokens);
return NULL;
}
domain = bstrcpy(tokens->entry[0]);
bdestroy(cpustr);
bstrListDestroy(tokens);
return domain;
}
void bstr_to_workgroup(Workgroup* group, const_bstring str, DataType type, int numberOfStreams)
{
uint32_t i;
int parseStreams = 0;
bstring threadInfo;
bstring streams= bformat("0");
struct bstrList* tokens;
struct bstrList* subtokens;
AffinityDomains_t domains;
AffinityDomain* domain = NULL;
/* split the workgroup into the thread and the streams part */
tokens = bsplit(str,'-');
if (tokens->qty == 2)
{
threadInfo = bstrcpy(tokens->entry[0]);
streams = bstrcpy(tokens->entry[1]);
parseStreams = 1;
}
else if (tokens->qty == 1)
{
threadInfo = bstrcpy(tokens->entry[0]);
}
else
{
fprintf(stderr, "Error in parsing workgroup string\n");
}
bstrListDestroy (tokens);
tokens = bsplit(threadInfo,':');
if (tokens->qty == 5)
{
uint32_t maxNumThreads;
int chunksize;
int stride;
int counter;
int currentId = 0;
int startId = 0;
domains = get_affinityDomains();
for (i = 0; i < domains->numberOfAffinityDomains; i++)
{
if (bstrcmp(domains->domains[i].tag, tokens->entry[0]) == BSTR_OK)
{
domain = &(domains->domains[i]);
break;
}
}
if (domain == NULL)
{
fprintf(stderr, "Error: Domain %s not available on current machine.\nTry likwid-bench -p for supported domains.",
bdata(tokens->entry[0]));
exit(EXIT_FAILURE);
}
group->size = bstr_to_doubleSize(tokens->entry[1], type);
group->numberOfThreads = str2int(bdata(tokens->entry[2]));
chunksize = str2int(bdata(tokens->entry[3]));
stride = str2int(bdata(tokens->entry[4]));
maxNumThreads = ceil((double)domain->numberOfProcessors / stride) * chunksize;
if (group->numberOfThreads > maxNumThreads)
{
fprintf(stderr, "Warning: More threads (%d) requested than CPUs in domain %s fulfilling expression (%d).\n",
group->numberOfThreads, bdata(tokens->entry[0]), maxNumThreads);
}
group->processorIds = (int*) malloc(group->numberOfThreads * sizeof(int));
counter = chunksize;
counter = 0;
for (int j=0; j<group->numberOfThreads; j+=chunksize)
{
for(i=0;i<chunksize && j+i<group->numberOfThreads ;i++)
{
group->processorIds[startId++] = domain->processorList[counter+i];
}
counter += stride;
if (counter >= domain->numberOfProcessors)
{
counter = counter-domain->numberOfProcessors;
}
}
}
else if (tokens->qty == 3)
{
domains = get_affinityDomains();
for (i = 0; i < domains->numberOfAffinityDomains; i++)
{
if (bstrcmp(domains->domains[i].tag, tokens->entry[0]) == BSTR_OK)
{
domain = &(domains->domains[i]);
break;
}
}
if (domain == NULL)
{
fprintf(stderr, "Error: Domain %s not available on current machine.\nTry likwid-bench -p for supported domains.",
bdata(tokens->entry[0]));
exit(EXIT_FAILURE);
}
group->size = bstr_to_doubleSize(tokens->entry[1], type);
group->numberOfThreads = str2int(bdata(tokens->entry[2]));
if (group->numberOfThreads > domain->numberOfProcessors)
{
fprintf(stderr, "Warning: More threads (%d) requested than CPUs in domain %s (%d).\n",
group->numberOfThreads, bdata(tokens->entry[0]), domain->numberOfProcessors);
}
group->processorIds = (int*) malloc(group->numberOfThreads * sizeof(int));
for (i=0; i<group->numberOfThreads; i++)
{
group->processorIds[i] = domain->processorList[i % domain->numberOfProcessors];
}
}
else if (tokens->qty == 2)
{
domains = get_affinityDomains();
for (i = 0; i < domains->numberOfAffinityDomains; i++)
{
if (bstrcmp(domains->domains[i].tag, tokens->entry[0]) == BSTR_OK)
{
domain = &(domains->domains[i]);
break;
}
}
if (domain == NULL)
{
fprintf(stderr, "Error: Domain %s not available on current machine.\nTry likwid-bench -p for supported domains.",
bdata(tokens->entry[0]));
exit(EXIT_FAILURE);
}
group->size = bstr_to_doubleSize(tokens->entry[1], type);
group->numberOfThreads = domain->numberOfProcessors;
group->processorIds = (int*) malloc(group->numberOfThreads * sizeof(int));
for (i=0; i<group->numberOfThreads; i++)
{
group->processorIds[i] = domain->processorList[i];
}
}
else
{
fprintf(stderr, "Error in parsing workgroup string\n");
}
bstrListDestroy(tokens);
/* parse stream list */
if (parseStreams)
{
tokens = bsplit(streams,',');
if (tokens->qty < numberOfStreams)
{
fprintf(stderr, "Error: Testcase requires at least %d streams\n", numberOfStreams);
}
group->streams = (Stream*) malloc(numberOfStreams * sizeof(Stream));
for (i=0;i<(uint32_t) tokens->qty;i++)
{
subtokens = bsplit(tokens->entry[i],':');
if ( subtokens->qty == 3 )
{
int index = str2int(bdata(subtokens->entry[0]));
if (index >= numberOfStreams)
{
fprintf(stderr, "Error: Stream index %d out of range\n",index);
}
group->streams[index].domain = bstrcpy(subtokens->entry[1]);
group->streams[index].offset = str2int(bdata(subtokens->entry[2]));
}
else if ( subtokens->qty == 2 )
{
int index = str2int(bdata(subtokens->entry[0]));
if (index >= numberOfStreams)
{
fprintf(stderr, "Error: Stream index %d out of range\n",index);
}
group->streams[index].domain = bstrcpy(subtokens->entry[1]);
group->streams[index].offset = 0;
}
else
{
fprintf(stderr, "Error: Cannot parse stream placement defintition in %s\n", bdata(str));
}
bstrListDestroy(subtokens);
}
bstrListDestroy(tokens);
}
else
{
group->streams = (Stream*) malloc(numberOfStreams * sizeof(Stream));
for (i=0; i< (uint32_t)numberOfStreams; i++)
{
group->streams[i].domain = domain->tag;
group->streams[i].offset = 0;
}
}
group->size /= numberOfStreams;
return;
}