test/hash: use compiler atomics for sync
authorDharmik Thakkar <dharmik.thakkar@arm.com>
Wed, 22 Sep 2021 21:52:05 +0000 (16:52 -0500)
committerDavid Marchand <david.marchand@redhat.com>
Tue, 19 Oct 2021 15:15:10 +0000 (17:15 +0200)
Convert rte_atomic usages to compiler atomic built-ins
for stats sync.

Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Joyce Kong <joyce.kong@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
Tested-by: David Christensen <drc@linux.vnet.ibm.com>
app/test/test_hash_multiwriter.c
app/test/test_hash_readwrite.c

index afa3c7b..0c5a8ca 100644 (file)
@@ -43,8 +43,8 @@ const uint32_t nb_entries = 5*1024*1024;
 const uint32_t nb_total_tsx_insertion = 4.5*1024*1024;
 uint32_t rounded_nb_total_tsx_insertion;
 
-static rte_atomic64_t gcycles;
-static rte_atomic64_t ginsertions;
+static uint64_t gcycles;
+static uint64_t ginsertions;
 
 static int use_htm;
 
@@ -84,8 +84,8 @@ test_hash_multiwriter_worker(void *arg)
        }
 
        cycles = rte_rdtsc_precise() - begin;
-       rte_atomic64_add(&gcycles, cycles);
-       rte_atomic64_add(&ginsertions, i - offset);
+       __atomic_fetch_add(&gcycles, cycles, __ATOMIC_RELAXED);
+       __atomic_fetch_add(&ginsertions, i - offset, __ATOMIC_RELAXED);
 
        for (; i < offset + tbl_multiwriter_test_params.nb_tsx_insertion; i++)
                tbl_multiwriter_test_params.keys[i]
@@ -168,11 +168,8 @@ test_hash_multiwriter(void)
 
        tbl_multiwriter_test_params.found = found;
 
-       rte_atomic64_init(&gcycles);
-       rte_atomic64_clear(&gcycles);
-
-       rte_atomic64_init(&ginsertions);
-       rte_atomic64_clear(&ginsertions);
+       __atomic_store_n(&gcycles, 0, __ATOMIC_RELAXED);
+       __atomic_store_n(&ginsertions, 0, __ATOMIC_RELAXED);
 
        /* Get list of enabled cores */
        i = 0;
@@ -238,8 +235,8 @@ test_hash_multiwriter(void)
        printf("No key corrupted during multiwriter insertion.\n");
 
        unsigned long long int cycles_per_insertion =
-               rte_atomic64_read(&gcycles)/
-               rte_atomic64_read(&ginsertions);
+               __atomic_load_n(&gcycles, __ATOMIC_RELAXED)/
+               __atomic_load_n(&ginsertions, __ATOMIC_RELAXED);
 
        printf(" cycles per insertion: %llu\n", cycles_per_insertion);
 
index 4860768..9b192f2 100644 (file)
@@ -45,14 +45,14 @@ struct {
        struct rte_hash *h;
 } tbl_rw_test_param;
 
-static rte_atomic64_t gcycles;
-static rte_atomic64_t ginsertions;
+static uint64_t gcycles;
+static uint64_t ginsertions;
 
-static rte_atomic64_t gread_cycles;
-static rte_atomic64_t gwrite_cycles;
+static uint64_t gread_cycles;
+static uint64_t gwrite_cycles;
 
-static rte_atomic64_t greads;
-static rte_atomic64_t gwrites;
+static uint64_t greads;
+static uint64_t gwrites;
 
 static int
 test_hash_readwrite_worker(__rte_unused void *arg)
@@ -110,8 +110,8 @@ test_hash_readwrite_worker(__rte_unused void *arg)
        }
 
        cycles = rte_rdtsc_precise() - begin;
-       rte_atomic64_add(&gcycles, cycles);
-       rte_atomic64_add(&ginsertions, i - offset);
+       __atomic_fetch_add(&gcycles, cycles, __ATOMIC_RELAXED);
+       __atomic_fetch_add(&ginsertions, i - offset, __ATOMIC_RELAXED);
 
        for (; i < offset + tbl_rw_test_param.num_insert; i++)
                tbl_rw_test_param.keys[i] = RTE_RWTEST_FAIL;
@@ -209,11 +209,8 @@ test_hash_readwrite_functional(int use_htm, int use_rw_lf, int use_ext)
        int worker_cnt = rte_lcore_count() - 1;
        uint32_t tot_insert = 0;
 
-       rte_atomic64_init(&gcycles);
-       rte_atomic64_clear(&gcycles);
-
-       rte_atomic64_init(&ginsertions);
-       rte_atomic64_clear(&ginsertions);
+       __atomic_store_n(&gcycles, 0, __ATOMIC_RELAXED);
+       __atomic_store_n(&ginsertions, 0, __ATOMIC_RELAXED);
 
        if (init_params(use_ext, use_htm, use_rw_lf, use_jhash) != 0)
                goto err;
@@ -272,8 +269,8 @@ test_hash_readwrite_functional(int use_htm, int use_rw_lf, int use_ext)
        printf("No key corrupted during read-write test.\n");
 
        unsigned long long int cycles_per_insertion =
-               rte_atomic64_read(&gcycles) /
-               rte_atomic64_read(&ginsertions);
+               __atomic_load_n(&gcycles, __ATOMIC_RELAXED) /
+               __atomic_load_n(&ginsertions, __ATOMIC_RELAXED);
 
        printf("cycles per insertion and lookup: %llu\n", cycles_per_insertion);
 
@@ -313,8 +310,8 @@ test_rw_reader(void *arg)
        }
 
        cycles = rte_rdtsc_precise() - begin;
-       rte_atomic64_add(&gread_cycles, cycles);
-       rte_atomic64_add(&greads, i);
+       __atomic_fetch_add(&gread_cycles, cycles, __ATOMIC_RELAXED);
+       __atomic_fetch_add(&greads, i, __ATOMIC_RELAXED);
        return 0;
 }
 
@@ -347,8 +344,9 @@ test_rw_writer(void *arg)
        }
 
        cycles = rte_rdtsc_precise() - begin;
-       rte_atomic64_add(&gwrite_cycles, cycles);
-       rte_atomic64_add(&gwrites, tbl_rw_test_param.num_insert);
+       __atomic_fetch_add(&gwrite_cycles, cycles, __ATOMIC_RELAXED);
+       __atomic_fetch_add(&gwrites, tbl_rw_test_param.num_insert,
+                                                       __ATOMIC_RELAXED);
        return 0;
 }
 
@@ -371,15 +369,11 @@ test_hash_readwrite_perf(struct perf *perf_results, int use_htm,
 
        uint64_t start = 0, end = 0;
 
-       rte_atomic64_init(&greads);
-       rte_atomic64_init(&gwrites);
-       rte_atomic64_clear(&gwrites);
-       rte_atomic64_clear(&greads);
+       __atomic_store_n(&gwrites, 0, __ATOMIC_RELAXED);
+       __atomic_store_n(&greads, 0, __ATOMIC_RELAXED);
 
-       rte_atomic64_init(&gread_cycles);
-       rte_atomic64_clear(&gread_cycles);
-       rte_atomic64_init(&gwrite_cycles);
-       rte_atomic64_clear(&gwrite_cycles);
+       __atomic_store_n(&gread_cycles, 0, __ATOMIC_RELAXED);
+       __atomic_store_n(&gwrite_cycles, 0, __ATOMIC_RELAXED);
 
        if (init_params(0, use_htm, 0, use_jhash) != 0)
                goto err;
@@ -436,10 +430,10 @@ test_hash_readwrite_perf(struct perf *perf_results, int use_htm,
                if (tot_worker_lcore < core_cnt[n] * 2)
                        goto finish;
 
-               rte_atomic64_clear(&greads);
-               rte_atomic64_clear(&gread_cycles);
-               rte_atomic64_clear(&gwrites);
-               rte_atomic64_clear(&gwrite_cycles);
+               __atomic_store_n(&greads, 0, __ATOMIC_RELAXED);
+               __atomic_store_n(&gread_cycles, 0, __ATOMIC_RELAXED);
+               __atomic_store_n(&gwrites, 0, __ATOMIC_RELAXED);
+               __atomic_store_n(&gwrite_cycles, 0, __ATOMIC_RELAXED);
 
                rte_hash_reset(tbl_rw_test_param.h);
 
@@ -481,8 +475,8 @@ test_hash_readwrite_perf(struct perf *perf_results, int use_htm,
 
                if (reader_faster) {
                        unsigned long long int cycles_per_insertion =
-                               rte_atomic64_read(&gread_cycles) /
-                               rte_atomic64_read(&greads);
+                               __atomic_load_n(&gread_cycles, __ATOMIC_RELAXED) /
+                               __atomic_load_n(&greads, __ATOMIC_RELAXED);
                        perf_results->read_only[n] = cycles_per_insertion;
                        printf("Reader only: cycles per lookup: %llu\n",
                                                        cycles_per_insertion);
@@ -490,17 +484,17 @@ test_hash_readwrite_perf(struct perf *perf_results, int use_htm,
 
                else {
                        unsigned long long int cycles_per_insertion =
-                               rte_atomic64_read(&gwrite_cycles) /
-                               rte_atomic64_read(&gwrites);
+                               __atomic_load_n(&gwrite_cycles, __ATOMIC_RELAXED) /
+                               __atomic_load_n(&gwrites, __ATOMIC_RELAXED);
                        perf_results->write_only[n] = cycles_per_insertion;
                        printf("Writer only: cycles per writes: %llu\n",
                                                        cycles_per_insertion);
                }
 
-               rte_atomic64_clear(&greads);
-               rte_atomic64_clear(&gread_cycles);
-               rte_atomic64_clear(&gwrites);
-               rte_atomic64_clear(&gwrite_cycles);
+               __atomic_store_n(&greads, 0, __ATOMIC_RELAXED);
+               __atomic_store_n(&gread_cycles, 0, __ATOMIC_RELAXED);
+               __atomic_store_n(&gwrites, 0, __ATOMIC_RELAXED);
+               __atomic_store_n(&gwrite_cycles, 0, __ATOMIC_RELAXED);
 
                rte_hash_reset(tbl_rw_test_param.h);
 
@@ -575,8 +569,8 @@ test_hash_readwrite_perf(struct perf *perf_results, int use_htm,
 
                if (reader_faster) {
                        unsigned long long int cycles_per_insertion =
-                               rte_atomic64_read(&gread_cycles) /
-                               rte_atomic64_read(&greads);
+                               __atomic_load_n(&gread_cycles, __ATOMIC_RELAXED) /
+                               __atomic_load_n(&greads, __ATOMIC_RELAXED);
                        perf_results->read_write_r[n] = cycles_per_insertion;
                        printf("Read-write cycles per lookup: %llu\n",
                                                        cycles_per_insertion);
@@ -584,8 +578,8 @@ test_hash_readwrite_perf(struct perf *perf_results, int use_htm,
 
                else {
                        unsigned long long int cycles_per_insertion =
-                               rte_atomic64_read(&gwrite_cycles) /
-                               rte_atomic64_read(&gwrites);
+                               __atomic_load_n(&gwrite_cycles, __ATOMIC_RELAXED) /
+                               __atomic_load_n(&gwrites, __ATOMIC_RELAXED);
                        perf_results->read_write_w[n] = cycles_per_insertion;
                        printf("Read-write cycles per writes: %llu\n",
                                                        cycles_per_insertion);