Merge pull request #8824 from ivankochin/topic/tools/am-memcpy

TOOLS/PERF: Add user-side memcpy option for AM benchmarks
openucx · Jan 23, 2023 · 52a9394 · 52a9394
2 parents b41dee7 + 3ee1786
commit 52a9394
Show file tree

Hide file tree

Showing 5 changed files with 21 additions and 3 deletions.
diff --git a/src/tools/perf/api/libperf.h b/src/tools/perf/api/libperf.h
@@ -88,7 +88,8 @@ enum ucx_perf_test_flags {
     UCX_PERF_TEST_FLAG_WAKEUP           = UCS_BIT(10), /* Create context with wakeup feature enabled */
     UCX_PERF_TEST_FLAG_ERR_HANDLING     = UCS_BIT(11), /* Create UCP eps with error handling support */
     UCX_PERF_TEST_FLAG_LOOPBACK         = UCS_BIT(12), /* Use loopback connection */
-    UCX_PERF_TEST_FLAG_PREREG           = UCS_BIT(13) /* Pass pre-registered memory handle */
+    UCX_PERF_TEST_FLAG_PREREG           = UCS_BIT(13), /* Pass pre-registered memory handle */
+    UCX_PERF_TEST_FLAG_AM_RECV_COPY     = UCS_BIT(14)  /* Do additional memcopy during AM receive */
 };
 
 

diff --git a/src/tools/perf/lib/ucp_tests.cc b/src/tools/perf/lib/ucp_tests.cc
@@ -315,7 +315,13 @@ class ucp_perf_test_runner {
             return test->am_rndv_recv(data, length, param);
         }
 
-        /* TODO: Add option to do memcopy here */
+        if (test->m_perf.params.flags & UCX_PERF_TEST_FLAG_AM_RECV_COPY) {
+            ucs_assertv(length == test->m_am_rx_length,
+                        "wrong buffer length %ld != %ld",
+                        length, test->m_am_rx_length);
+            memcpy(test->m_am_rx_buffer, data, test->m_am_rx_length);
+        }
+
         test->recv_completed();
         return UCS_OK;
     }

diff --git a/src/tools/perf/perftest.h b/src/tools/perf/perftest.h
@@ -18,7 +18,7 @@
 #define MAX_BATCH_FILES         32
 #define MAX_CPUS                1024
 #define TL_RESOURCE_NAME_NONE   "<none>"
-#define TEST_PARAMS_ARGS        "t:n:s:W:O:w:D:i:H:oSCIqM:r:E:T:d:x:A:BUem:R:lz"
+#define TEST_PARAMS_ARGS        "t:n:s:W:O:w:D:i:H:oSCIqM:r:E:T:d:x:A:BUem:R:lyz"
 #define TEST_ID_UNDEFINED       -1
 
 enum {

diff --git a/src/tools/perf/perftest_params.c b/src/tools/perf/perftest_params.c
@@ -134,6 +134,7 @@ static void usage(const struct perftest_context *ctx, const char *program)
     printf("                        sleep      : go to sleep after posting requests\n");
     printf("     -H <size>      active message header size (%zu), not included in message size\n",
                                 ctx->params.super.ucp.am_hdr_size);
+    printf("     -y             do additional memcopy to the user memory in active message receive handler\n");
     printf("     -z             pass pre-registered memory handle\n");
     printf("\n");
     printf("   NOTE: When running UCP tests, transport and device should be specified by\n");
@@ -414,6 +415,9 @@ ucs_status_t parse_test_params(perftest_params_t *params, char opt,
             return UCS_ERR_INVALID_PARAM;
         }
         return UCS_OK;
+    case 'y':
+        params->super.flags |= UCX_PERF_TEST_FLAG_AM_RECV_COPY;
+        return UCS_OK;
     case 'z':
         params->super.flags |= UCX_PERF_TEST_FLAG_PREREG;
         return UCS_OK;

diff --git a/test/gtest/ucp/test_ucp_perf.cc b/test/gtest/ucp/test_ucp_perf.cc
@@ -273,6 +273,13 @@ const test_perf::test_spec test_ucp_perf::tests[] =
     ucs_offsetof(ucx_perf_result_t, latency.total_average), 1e6, 0.001, 60.0,
     0 },
 
+  { "am_lat_memcpy", "usec",
+    UCX_PERF_API_UCP, UCX_PERF_CMD_AM, UCX_PERF_TEST_TYPE_PINGPONG,
+    UCX_PERF_WAIT_MODE_POLL,
+    UCP_PERF_DATATYPE_CONTIG, 0, 1, { 8 }, 1, 100000lu,
+    ucs_offsetof(ucx_perf_result_t, latency.total_average), 1e6, 0.001, 60.0,
+    UCX_PERF_TEST_FLAG_AM_RECV_COPY },
+
   { "am_lat_b", "usec",
     UCX_PERF_API_UCP, UCX_PERF_CMD_AM, UCX_PERF_TEST_TYPE_PINGPONG,
     UCX_PERF_WAIT_MODE_SLEEP,