int32_t
perf_test(void)
{
struct rte_lpm *lpm = NULL;
uint64_t begin, total_time, lpm_used_entries = 0;
unsigned i, j;
uint8_t next_hop_add = 0xAA, next_hop_return = 0;
int status = 0;
uint64_t cache_line_counter = 0;
int64_t count = 0;
rte_srand(rte_rdtsc());
printf("No. routes = %u\n", (unsigned) NUM_ROUTE_ENTRIES);
print_route_distribution(large_route_table, (uint32_t) NUM_ROUTE_ENTRIES);
lpm = rte_lpm_create(__func__, SOCKET_ID_ANY, 1000000, 0);
TEST_LPM_ASSERT(lpm != NULL);
/* Measue add. */
begin = rte_rdtsc();
for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
if (rte_lpm_add(lpm, large_route_table[i].ip,
large_route_table[i].depth, next_hop_add) == 0)
status++;
}
/* End Timer. */
total_time = rte_rdtsc() - begin;
printf("Unique added entries = %d\n", status);
/* Obtain add statistics. */
for (i = 0; i < RTE_LPM_TBL24_NUM_ENTRIES; i++) {
if (lpm->tbl24[i].valid)
lpm_used_entries++;
if (i % 32 == 0){
if ((uint64_t)count < lpm_used_entries) {
cache_line_counter++;
count = lpm_used_entries;
}
}
}
printf("Used table 24 entries = %u (%g%%)\n",
(unsigned) lpm_used_entries,
(lpm_used_entries * 100.0) / RTE_LPM_TBL24_NUM_ENTRIES);
printf("64 byte Cache entries used = %u (%u bytes)\n",
(unsigned) cache_line_counter, (unsigned) cache_line_counter * 64);
printf("Average LPM Add: %g cycles\n", (double)total_time / NUM_ROUTE_ENTRIES);
/* Measure single Lookup */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i ++) {
static uint32_t ip_batch[BATCH_SIZE];
for (j = 0; j < BATCH_SIZE; j ++)
ip_batch[j] = rte_rand();
/* Lookup per batch */
begin = rte_rdtsc();
for (j = 0; j < BATCH_SIZE; j ++) {
if (rte_lpm_lookup(lpm, ip_batch[j], &next_hop_return) != 0)
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("Average LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Measure bulk Lookup */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i ++) {
static uint32_t ip_batch[BATCH_SIZE];
uint16_t next_hops[BULK_SIZE];
/* Create array of random IP addresses */
for (j = 0; j < BATCH_SIZE; j ++)
ip_batch[j] = rte_rand();
/* Lookup per batch */
begin = rte_rdtsc();
for (j = 0; j < BATCH_SIZE; j += BULK_SIZE) {
unsigned k;
rte_lpm_lookup_bulk(lpm, &ip_batch[j], next_hops, BULK_SIZE);
for (k = 0; k < BULK_SIZE; k++)
if (unlikely(!(next_hops[k] & RTE_LPM_LOOKUP_SUCCESS)))
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("BULK LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Measure LookupX4 */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i++) {
static uint32_t ip_batch[BATCH_SIZE];
uint16_t next_hops[4];
/* Create array of random IP addresses */
for (j = 0; j < BATCH_SIZE; j++)
ip_batch[j] = rte_rand();
/* Lookup per batch */
begin = rte_rdtsc();
for (j = 0; j < BATCH_SIZE; j += RTE_DIM(next_hops)) {
unsigned k;
__m128i ipx4;
ipx4 = _mm_loadu_si128((__m128i *)(ip_batch + j));
ipx4 = *(__m128i *)(ip_batch + j);
rte_lpm_lookupx4(lpm, ipx4, next_hops, UINT16_MAX);
for (k = 0; k < RTE_DIM(next_hops); k++)
if (unlikely(next_hops[k] == UINT16_MAX))
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("LPM LookupX4: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Delete */
status = 0;
begin = rte_rdtsc();
for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
/* rte_lpm_delete(lpm, ip, depth) */
status += rte_lpm_delete(lpm, large_route_table[i].ip,
large_route_table[i].depth);
}
total_time += rte_rdtsc() - begin;
printf("Average LPM Delete: %g cycles\n",
(double)total_time / NUM_ROUTE_ENTRIES);
rte_lpm_delete_all(lpm);
rte_lpm_free(lpm);
return PASS;
}
static int
test_memzone_reserve_max_aligned(void)
{
const struct rte_memzone *mz;
const struct rte_config *config;
const struct rte_memseg *ms;
int memseg_idx = 0;
int memzone_idx = 0;
uintptr_t addr_offset;
size_t len = 0;
void* last_addr;
size_t maxlen = 0;
/* random alignment */
rte_srand((unsigned)rte_rdtsc());
const unsigned align = 1 << ((rte_rand() % 8) + 5); /* from 128 up to 4k alignment */
/* get pointer to global configuration */
config = rte_eal_get_configuration();
ms = rte_eal_get_physmem_layout();
addr_offset = 0;
for (memseg_idx = 0; memseg_idx < RTE_MAX_MEMSEG; memseg_idx++){
/* ignore smaller memsegs as they can only get smaller */
if (ms[memseg_idx].len < maxlen)
continue;
/* align everything */
last_addr = RTE_PTR_ALIGN_CEIL(ms[memseg_idx].addr, RTE_CACHE_LINE_SIZE);
len = ms[memseg_idx].len - RTE_PTR_DIFF(last_addr, ms[memseg_idx].addr);
len &= ~((size_t) RTE_CACHE_LINE_MASK);
/* cycle through all memzones */
for (memzone_idx = 0; memzone_idx < RTE_MAX_MEMZONE; memzone_idx++) {
/* stop when reaching last allocated memzone */
if (config->mem_config->memzone[memzone_idx].addr == NULL)
break;
/* check if the memzone is in our memseg and subtract length */
if ((config->mem_config->memzone[memzone_idx].addr >=
ms[memseg_idx].addr) &&
(config->mem_config->memzone[memzone_idx].addr <
(RTE_PTR_ADD(ms[memseg_idx].addr, ms[memseg_idx].len)))) {
/* since the zones can now be aligned and occasionally skip
* some space, we should calculate the length based on
* reported length and start addresses difference.
*/
len -= (uintptr_t) RTE_PTR_SUB(
config->mem_config->memzone[memzone_idx].addr,
(uintptr_t) last_addr);
len -= config->mem_config->memzone[memzone_idx].len;
last_addr =
RTE_PTR_ADD(config->mem_config->memzone[memzone_idx].addr,
(size_t) config->mem_config->memzone[memzone_idx].len);
}
}
/* make sure we get the alignment offset */
if (len > maxlen) {
addr_offset = RTE_PTR_ALIGN_CEIL((uintptr_t) last_addr, align) - (uintptr_t) last_addr;
maxlen = len;
}
}
if (maxlen == 0 || maxlen == addr_offset) {
printf("There is no space left for biggest %u-aligned memzone!\n", align);
return 0;
}
maxlen -= addr_offset;
mz = rte_memzone_reserve_aligned("max_zone_aligned", 0,
SOCKET_ID_ANY, 0, align);
if (mz == NULL){
printf("Failed to reserve a big chunk of memory\n");
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
if (mz->len != maxlen) {
printf("Memzone reserve with 0 size and alignment %u did not return"
" bigest block\n", align);
printf("Expected size = %zu, actual size = %zu\n",
maxlen, mz->len);
rte_dump_physmem_layout(stdout);
rte_memzone_dump(stdout);
return -1;
}
return 0;
}
static int
test_reciprocal(void)
{
int result = 0;
uint32_t divisor_u32 = 0;
uint32_t dividend_u32;
uint32_t nresult_u32;
uint32_t rresult_u32;
uint64_t i, j;
uint64_t divisor_u64 = 0;
uint64_t dividend_u64;
uint64_t nresult_u64;
uint64_t rresult_u64;
struct rte_reciprocal reci_u32 = {0};
struct rte_reciprocal_u64 reci_u64 = {0};
rte_srand(rte_rdtsc());
printf("Validating unsigned 32bit division.\n");
for (i = 0; i < MAX_ITERATIONS; i++) {
/* Change divisor every DIVIDE_ITER iterations. */
if (i % DIVIDE_ITER == 0) {
divisor_u32 = rte_rand();
reci_u32 = rte_reciprocal_value(divisor_u32);
}
dividend_u32 = rte_rand();
nresult_u32 = dividend_u32 / divisor_u32;
rresult_u32 = rte_reciprocal_divide(dividend_u32,
reci_u32);
if (nresult_u32 != rresult_u32) {
printf("Division failed, %"PRIu32"/%"PRIu32" = "
"expected %"PRIu32" result %"PRIu32"\n",
dividend_u32, divisor_u32,
nresult_u32, rresult_u32);
result = 1;
break;
}
}
printf("Validating unsigned 64bit division.\n");
for (i = 0; i < MAX_ITERATIONS; i++) {
/* Change divisor every DIVIDE_ITER iterations. */
if (i % DIVIDE_ITER == 0) {
divisor_u64 = rte_rand();
reci_u64 = rte_reciprocal_value_u64(divisor_u64);
}
dividend_u64 = rte_rand();
nresult_u64 = dividend_u64 / divisor_u64;
rresult_u64 = rte_reciprocal_divide_u64(dividend_u64,
&reci_u64);
if (nresult_u64 != rresult_u64) {
printf("Division failed, %"PRIu64"/%"PRIu64" = "
"expected %"PRIu64" result %"PRIu64"\n",
dividend_u64, divisor_u64,
nresult_u64, rresult_u64);
result = 1;
break;
}
}
printf("Validating unsigned 64bit division with 32bit divisor.\n");
for (i = 0; i < MAX_ITERATIONS; i++) {
/* Change divisor every DIVIDE_ITER iterations. */
if (i % DIVIDE_ITER == 0) {
divisor_u64 = rte_rand() >> 32;
reci_u64 = rte_reciprocal_value_u64(divisor_u64);
}
dividend_u64 = rte_rand();
nresult_u64 = dividend_u64 / divisor_u64;
rresult_u64 = rte_reciprocal_divide_u64(dividend_u64,
&reci_u64);
if (nresult_u64 != rresult_u64) {
printf("Division failed, %"PRIu64"/%"PRIu64" = "
"expected %"PRIu64" result %"PRIu64"\n",
dividend_u64, divisor_u64,
nresult_u64, rresult_u64);
result = 1;
break;
}
}
int32_t
perf_test(void)
{
struct rte_lpm6 *lpm = NULL;
struct rte_lpm6_config config;
uint64_t begin, total_time;
unsigned i, j;
uint8_t next_hop_add = 0xAA, next_hop_return = 0;
int status = 0;
int64_t count = 0;
config.max_rules = 1000000;
config.number_tbl8s = NUMBER_TBL8S;
config.flags = 0;
rte_srand(rte_rdtsc());
printf("No. routes = %u\n", (unsigned) NUM_ROUTE_ENTRIES);
print_route_distribution(large_route_table, (uint32_t) NUM_ROUTE_ENTRIES);
lpm = rte_lpm6_create(__func__, SOCKET_ID_ANY, &config);
TEST_LPM_ASSERT(lpm != NULL);
/* Measure add. */
begin = rte_rdtsc();
for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
if (rte_lpm6_add(lpm, large_route_table[i].ip,
large_route_table[i].depth, next_hop_add) == 0)
status++;
}
/* End Timer. */
total_time = rte_rdtsc() - begin;
printf("Unique added entries = %d\n", status);
printf("Average LPM Add: %g cycles\n",
(double)total_time / NUM_ROUTE_ENTRIES);
/* Measure single Lookup */
total_time = 0;
count = 0;
for (i = 0; i < ITERATIONS; i ++) {
begin = rte_rdtsc();
for (j = 0; j < NUM_IPS_ENTRIES; j ++) {
if (rte_lpm6_lookup(lpm, large_ips_table[j].ip,
&next_hop_return) != 0)
count++;
}
total_time += rte_rdtsc() - begin;
}
printf("Average LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Measure bulk Lookup */
total_time = 0;
count = 0;
uint8_t ip_batch[NUM_IPS_ENTRIES][16];
int16_t next_hops[NUM_IPS_ENTRIES];
for (i = 0; i < NUM_IPS_ENTRIES; i++)
memcpy(ip_batch[i], large_ips_table[i].ip, 16);
for (i = 0; i < ITERATIONS; i ++) {
/* Lookup per batch */
begin = rte_rdtsc();
rte_lpm6_lookup_bulk_func(lpm, ip_batch, next_hops, NUM_IPS_ENTRIES);
total_time += rte_rdtsc() - begin;
for (j = 0; j < NUM_IPS_ENTRIES; j++)
if (next_hops[j] < 0)
count++;
}
printf("BULK LPM Lookup: %.1f cycles (fails = %.1f%%)\n",
(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
/* Delete */
status = 0;
begin = rte_rdtsc();
for (i = 0; i < NUM_ROUTE_ENTRIES; i++) {
/* rte_lpm_delete(lpm, ip, depth) */
status += rte_lpm6_delete(lpm, large_route_table[i].ip,
large_route_table[i].depth);
}
total_time += rte_rdtsc() - begin;
printf("Average LPM Delete: %g cycles\n",
(double)total_time / NUM_ROUTE_ENTRIES);
rte_lpm6_delete_all(lpm);
rte_lpm6_free(lpm);
return PASS;
}
