2
0

SWDEV-295522: Fixing Performance Issue

Removing DEBUG_TRACES and the unnecessary use of roctracer_op_string, made the MS app reporting 78 to 81 stable samples per second, depending on the type of the trace, while the main app without rocprof reports 100 to 106. More detailed numbers will be posted in the ticket.

Change-Id: Ida25d3bfc72047afaa27326d697be76d97564334


[ROCm/roctracer commit: e7327aaac7]
Este cometimento está contido em:
Ammar ELWazir
2022-04-07 00:07:24 +00:00
ascendente f1de3ae0e5
cometimento 35ea06abd0
2 ficheiros modificados com 1 adições e 46 eliminações
-19
Ver ficheiro
@@ -294,8 +294,6 @@ static inline void CorrelationIdRegistr(const activity_correlation_id_t& correla
std::lock_guard<correlation_id_mutex_t> lck(correlation_id_mutex);
const auto ret = correlation_id_map.insert({correlation_id, correlation_id_tls});
if (ret.second == false) EXC_ABORT(ROCTRACER_STATUS_ERROR, "HCC activity id is not unique(" << correlation_id << ")");
DEBUG_TRACE("CorrelationIdRegistr id(%lu) id_tls(%lu)\n", correlation_id, correlation_id_tls);
}
static inline activity_correlation_id_t CorrelationIdLookup(const activity_correlation_id_t& correlation_id) {
@@ -304,9 +302,6 @@ static inline activity_correlation_id_t CorrelationIdLookup(const activity_corre
if (it == correlation_id_map.end()) EXC_ABORT(ROCTRACER_STATUS_ERROR, "HCC activity id lookup failed(" << correlation_id << ")");
const activity_correlation_id_t ret_val = it->second;
correlation_id_map.erase(it);
DEBUG_TRACE("CorrelationIdLookup id(%lu) ret(%lu)\n", correlation_id, ret_val);
return ret_val;
}
@@ -394,11 +389,6 @@ void* HIP_SyncApiDataCallback(
// Clearing correlatin ID
correlation_id_tls = 0;
}
const char * name = roctracer_op_string(ACTIVITY_DOMAIN_HIP_API, op_id, 0);
DEBUG_TRACE("HIP_SyncApiDataCallback(\"%s\") phase(%d): op(%u) record(%p) data(%p) pool(%p) depth(%d) correlation_id(%lu) time_ns(%lu)\n",
name, phase, op_id, record, data, pool, (int)(record_pair_stack->size()), (data_ptr) ? data_ptr->correlation_id : 0, timer.timestamp_ns());
return ret;
}
@@ -488,11 +478,6 @@ void* HIP_SyncActivityCallback(
// Clearing correlatin ID
correlation_id_tls = 0;
}
const char * name = roctracer_op_string(ACTIVITY_DOMAIN_HIP_API, op_id, 0);
DEBUG_TRACE("HIP_SyncActivityCallback(\"%s\") phase(%d): op(%u) record(%p) data(%p) pool(%p) depth(%d) correlation_id(%lu) beg_ns(%lu) end_ns(%lu)\n",
name, phase, op_id, record, data, pool, (int)(record_pair_stack->size()), (data_ptr) ? data_ptr->correlation_id : 0, timestamp_ns);
return ret;
}
@@ -507,10 +492,6 @@ void HCC_AsyncActivityCallback(uint32_t op_id, void* record, void* arg) {
record_ptr->correlation_id = CorrelationIdLookup(record_ptr->correlation_id);
if (record_ptr->correlation_id == 0) return;
pool->Write(*record_ptr);
const char * name = roctracer_op_string(ACTIVITY_DOMAIN_HCC_OPS, record_ptr->op, record_ptr->kind);
DEBUG_TRACE("HCC_AsyncActivityCallback(\"%s\"): op(%u) kind(%u) record(%p) pool(%p) correlation_id(%d) beg_ns(%lu) end_ns(%lu)\n",
name, record_ptr->op, record_ptr->kind, record, pool, record_ptr->correlation_id, record_ptr->begin_ns, record_ptr->end_ns);
}
// Open output file