Profiling of round 2 processing

Enclave/route.cpp

@@ -504,10 +504,18 @@ void ecall_routing_proceed(void *cbpointer)
 
             // Obliviously tally the number of messages we received in
             // round1 destined for each storage node
+#ifdef PROFILE_ROUTING
+            uint32_t inserted = round1.inserted;
+            unsigned long start_round2 = printf_with_rtclock("begin round2 processing (%u,%u)\n", inserted, round1.bufsize);
+            unsigned long start_tally = printf_with_rtclock("begin tally (%u)\n", inserted);
+#endif
             uint32_t msg_size = g_teems_config.msg_size;
             nodenum_t num_storage_nodes = g_teems_config.num_storage_nodes;
             std::vector<uint32_t> tally = obliv_tally_stg(
                 round1.buf, msg_size, round1.inserted, num_storage_nodes);
+#ifdef PROFILE_ROUTING
+            printf_with_rtclock_diff(start_tally, "end tally (%u)\n", inserted);
+#endif
 
             // Note: tally contains private values!  It's OK to
             // non-obliviously check for an error condition, though.
@@ -529,14 +537,27 @@ void ecall_routing_proceed(void *cbpointer)
 
             // Obliviously add padding for each storage node according
             // to the (private) padding tally.
+#ifdef PROFILE_ROUTING
+            unsigned long start_pad = printf_with_rtclock("begin pad (%u)\n", tot_padding);
+#endif
             obliv_pad_stg(round1.buf + round1.inserted * msg_size,
                 msg_size, tally, tot_padding);
+#ifdef PROFILE_ROUTING
+            printf_with_rtclock_diff(start_pad, "end pad (%u)\n", tot_padding);
+#endif
 
             round1.inserted += tot_padding;
 
             // Obliviously shuffle the messages
+#ifdef PROFILE_ROUTING
+            unsigned long start_shuffle = printf_with_rtclock("begin shuffle (%u,%u)\n", round1.inserted, round1.bufsize);
+#endif
             uint32_t num_shuffled = shuffle_mtobliv(g_teems_config.nthreads,
                 round1.buf, msg_size, round1.inserted, round1.bufsize);
+#ifdef PROFILE_ROUTING
+            printf_with_rtclock_diff(start_pad, "end shuffle (%u,%u)\n", round1.inserted, round1.bufsize);
+            printf_with_rtclock_diff(start_round2, "end round2 processing (%u,%u)\n", inserted, round1.bufsize);
+#endif
 
             // Now we can handle the messages non-obliviously, since we
             // know there will be exactly msgs_per_stg messages to each