app/test: isolate lpm performance cases
authorThomas Monjalon <thomas.monjalon@6wind.com>
Thu, 5 May 2016 07:08:13 +0000 (09:08 +0200)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Tue, 24 May 2016 15:00:55 +0000 (17:00 +0200)
The LPM performance tests should not be mixed with
functional tests: they are long and show some statistics.

LPM6 functional tests are added to parallel tests.

Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
app/test/Makefile
app/test/autotest_data.py
app/test/test_lpm.c
app/test/test_lpm6.c
app/test/test_lpm6_perf.c [new file with mode: 0644]
app/test/test_lpm_perf.c [new file with mode: 0644]

index a4907d5..f269fe0 100644 (file)
@@ -89,7 +89,9 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_functions.c
 SRCS-$(CONFIG_RTE_LIBRTE_HASH) += test_hash_scaling.c
 
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm.c
+SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm_perf.c
 SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm6.c
+SRCS-$(CONFIG_RTE_LIBRTE_LPM) += test_lpm6_perf.c
 
 SRCS-y += test_debug.c
 SRCS-y += test_errno.c
index dde4511..8a92bb4 100644 (file)
@@ -173,6 +173,12 @@ parallel_test_group_list = [
                 "Func" :       default_autotest,
                 "Report" :     None,
                },
+               {
+                       "Name" :    "LPM6 autotest",
+                       "Command" : "lpm6_autotest",
+                       "Func" :    default_autotest,
+                       "Report" :  None,
+               },
                {
                 "Name" :       "IVSHMEM autotest",
                 "Command" :    "ivshmem_autotest",
@@ -449,19 +455,6 @@ non_parallel_test_group_list = [
                },
        ]
 },
-{
-       "Prefix" :      "lpm6",
-       "Memory" :      "512",
-       "Tests" :
-       [
-               {
-                 "Name" :       "LPM6 autotest",
-                 "Command" :    "lpm6_autotest",
-                 "Func" :       default_autotest,
-                 "Report" :     None,
-                },
-       ]
-},
 {
        "Prefix":       "timer_perf",
        "Memory" :      per_sockets(512),
index 40fbbc6..f6930fb 100644 (file)
 #include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
-#include <errno.h>
-#include <sys/queue.h>
 
-#include <rte_common.h>
-#include <rte_cycles.h>
-#include <rte_memory.h>
-#include <rte_random.h>
-#include <rte_branch_prediction.h>
-#include <rte_ip.h>
-#include <time.h>
+#include <rte_lpm.h>
 
 #include "test.h"
-
-#include "rte_lpm.h"
 #include "test_lpm_routes.h"
 #include "test_xmmt_ops.h"
 
@@ -78,7 +68,6 @@ static int32_t test14(void);
 static int32_t test15(void);
 static int32_t test16(void);
 static int32_t test17(void);
-static int32_t perf_test(void);
 
 rte_lpm_test tests[] = {
 /* Test Cases */
@@ -100,7 +89,6 @@ rte_lpm_test tests[] = {
        test15,
        test16,
        test17,
-       perf_test,
 };
 
 #define NUM_LPM_TESTS (sizeof(tests)/sizeof(tests[0]))
@@ -1230,202 +1218,7 @@ test17(void)
 }
 
 /*
- * Lookup performance test
- */
-
-#define ITERATIONS (1 << 10)
-#define BATCH_SIZE (1 << 12)
-#define BULK_SIZE 32
-
-static void
-print_route_distribution(const struct route_rule *table, uint32_t n)
-{
-       unsigned i, j;
-
-       printf("Route distribution per prefix width: \n");
-       printf("DEPTH    QUANTITY (PERCENT)\n");
-       printf("--------------------------- \n");
-
-       /* Count depths. */
-       for (i = 1; i <= 32; i++) {
-               unsigned depth_counter = 0;
-               double percent_hits;
-
-               for (j = 0; j < n; j++)
-                       if (table[j].depth == (uint8_t) i)
-                               depth_counter++;
-
-               percent_hits = ((double)depth_counter)/((double)n) * 100;
-               printf("%.2u%15u (%.2f)\n", i, depth_counter, percent_hits);
-       }
-       printf("\n");
-}
-
-int32_t
-perf_test(void)
-{
-       struct rte_lpm *lpm = NULL;
-       struct rte_lpm_config config;
-
-       config.max_rules = 1000000;
-       config.number_tbl8s = NUMBER_TBL8S;
-       config.flags = 0;
-       uint64_t begin, total_time, lpm_used_entries = 0;
-       unsigned i, j;
-       uint32_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, &config);
-       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];
-               uint32_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];
-               uint32_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;
-                       xmm_t ipx4;
-
-                       ipx4 = vect_loadu_sil128((xmm_t *)(ip_batch + j));
-                       ipx4 = *(xmm_t *)(ip_batch + j);
-                       rte_lpm_lookupx4(lpm, ipx4, next_hops, UINT32_MAX);
-                       for (k = 0; k < RTE_DIM(next_hops); k++)
-                               if (unlikely(next_hops[k] == UINT32_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;
-}
-
-/*
- * Do all unit and performance tests.
+ * Do all unit tests.
  */
 
 static int
index b464342..9163cd7 100644 (file)
  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
+
 #include <stdio.h>
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
-#include <errno.h>
-#include <sys/queue.h>
-
-#include <time.h>
-
-#include "test.h"
 
-#include <rte_common.h>
-#include <rte_cycles.h>
 #include <rte_memory.h>
-#include <rte_random.h>
-#include <rte_branch_prediction.h>
-#include <rte_ip.h>
+#include <rte_lpm6.h>
 
-#include "rte_lpm6.h"
+#include "test.h"
 #include "test_lpm6_routes.h"
 
 #define TEST_LPM_ASSERT(cond) do {                                            \
@@ -88,7 +79,6 @@ static int32_t test24(void);
 static int32_t test25(void);
 static int32_t test26(void);
 static int32_t test27(void);
-static int32_t perf_test(void);
 
 rte_lpm6_test tests6[] = {
 /* Test Cases */
@@ -120,7 +110,6 @@ rte_lpm6_test tests6[] = {
        test25,
        test26,
        test27,
-       perf_test,
 };
 
 #define NUM_LPM6_TESTS                (sizeof(tests6)/sizeof(tests6[0]))
@@ -1762,143 +1751,7 @@ test27(void)
 }
 
 /*
- * Lookup performance test
- */
-
-#define ITERATIONS (1 << 10)
-#define BATCH_SIZE 100000
-
-static void
-print_route_distribution(const struct rules_tbl_entry *table, uint32_t n)
-{
-       unsigned i, j;
-
-       printf("Route distribution per prefix width: \n");
-       printf("DEPTH    QUANTITY (PERCENT)\n");
-       printf("--------------------------- \n");
-
-       /* Count depths. */
-       for(i = 1; i <= 128; i++) {
-               unsigned depth_counter = 0;
-               double percent_hits;
-
-               for (j = 0; j < n; j++)
-                       if (table[j].depth == (uint8_t) i)
-                               depth_counter++;
-
-               percent_hits = ((double)depth_counter)/((double)n) * 100;
-               printf("%.2u%15u (%.2f)\n", i, depth_counter, percent_hits);
-       }
-       printf("\n");
-}
-
-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;
-}
-
-/*
- * Do all unit and performance tests.
+ * Do all unit tests.
  */
 static int
 test_lpm6(void)
diff --git a/app/test/test_lpm6_perf.c b/app/test/test_lpm6_perf.c
new file mode 100644 (file)
index 0000000..b7d4631
--- /dev/null
@@ -0,0 +1,191 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <rte_cycles.h>
+#include <rte_random.h>
+#include <rte_memory.h>
+#include <rte_lpm6.h>
+
+#include "test.h"
+#include "test_lpm6_routes.h"
+
+#define TEST_LPM_ASSERT(cond) do {                                            \
+       if (!(cond)) {                                                        \
+               printf("Error at line %d: \n", __LINE__);                     \
+               return -1;                                                    \
+       }                                                                     \
+} while(0)
+
+#define ITERATIONS (1 << 10)
+#define BATCH_SIZE 100000
+#define NUMBER_TBL8S                                           (1 << 16)
+
+static void
+print_route_distribution(const struct rules_tbl_entry *table, uint32_t n)
+{
+       unsigned i, j;
+
+       printf("Route distribution per prefix width: \n");
+       printf("DEPTH    QUANTITY (PERCENT)\n");
+       printf("--------------------------- \n");
+
+       /* Count depths. */
+       for(i = 1; i <= 128; i++) {
+               unsigned depth_counter = 0;
+               double percent_hits;
+
+               for (j = 0; j < n; j++)
+                       if (table[j].depth == (uint8_t) i)
+                               depth_counter++;
+
+               percent_hits = ((double)depth_counter)/((double)n) * 100;
+               printf("%.2u%15u (%.2f)\n", i, depth_counter, percent_hits);
+       }
+       printf("\n");
+}
+
+static int
+test_lpm6_perf(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 0;
+}
+
+static struct test_command lpm6_perf_cmd = {
+       .command = "lpm6_perf_autotest",
+       .callback = test_lpm6_perf,
+};
+REGISTER_TEST_COMMAND(lpm6_perf_cmd);
diff --git a/app/test/test_lpm_perf.c b/app/test/test_lpm_perf.c
new file mode 100644 (file)
index 0000000..41da811
--- /dev/null
@@ -0,0 +1,249 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+
+#include <rte_cycles.h>
+#include <rte_random.h>
+#include <rte_branch_prediction.h>
+#include <rte_lpm.h>
+
+#include "test.h"
+#include "test_lpm_routes.h"
+#include "test_xmmt_ops.h"
+
+#define TEST_LPM_ASSERT(cond) do {                                            \
+       if (!(cond)) {                                                        \
+               printf("Error at line %d: \n", __LINE__);                     \
+               return -1;                                                    \
+       }                                                                     \
+} while(0)
+
+#define ITERATIONS (1 << 10)
+#define BATCH_SIZE (1 << 12)
+#define BULK_SIZE 32
+
+static void
+print_route_distribution(const struct route_rule *table, uint32_t n)
+{
+       unsigned i, j;
+
+       printf("Route distribution per prefix width: \n");
+       printf("DEPTH    QUANTITY (PERCENT)\n");
+       printf("--------------------------- \n");
+
+       /* Count depths. */
+       for (i = 1; i <= 32; i++) {
+               unsigned depth_counter = 0;
+               double percent_hits;
+
+               for (j = 0; j < n; j++)
+                       if (table[j].depth == (uint8_t) i)
+                               depth_counter++;
+
+               percent_hits = ((double)depth_counter)/((double)n) * 100;
+               printf("%.2u%15u (%.2f)\n", i, depth_counter, percent_hits);
+       }
+       printf("\n");
+}
+
+static int
+test_lpm_perf(void)
+{
+       struct rte_lpm *lpm = NULL;
+       struct rte_lpm_config config;
+
+       config.max_rules = 1000000;
+       config.number_tbl8s = 256;
+       config.flags = 0;
+       uint64_t begin, total_time, lpm_used_entries = 0;
+       unsigned i, j;
+       uint32_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, &config);
+       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];
+               uint32_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];
+               uint32_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;
+                       xmm_t ipx4;
+
+                       ipx4 = vect_loadu_sil128((xmm_t *)(ip_batch + j));
+                       ipx4 = *(xmm_t *)(ip_batch + j);
+                       rte_lpm_lookupx4(lpm, ipx4, next_hops, UINT32_MAX);
+                       for (k = 0; k < RTE_DIM(next_hops); k++)
+                               if (unlikely(next_hops[k] == UINT32_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 0;
+}
+
+static struct test_command lpm_perf_cmd = {
+       .command = "lpm_perf_autotest",
+       .callback = test_lpm_perf,
+};
+REGISTER_TEST_COMMAND(lpm_perf_cmd);