AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

LINT: Add cpplint, clang-format and pre-commit support

Browse Source 
Change-Id: I3cbb787ef27d90486b212dfb1a8c77c460acc2ac
Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>


[ROCm/rdc commit: 434e40305d]
This commit is contained in:
Galantsev, Dmitrii
2023-12-04 15:24:34 -06:00
parent 61cf14d7cc
commit ea624cbb7c
137 changed files with 9122 additions and 10469 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rdc/example/job_stats_example.cc
+130 -142
View File
@@ -21,162 +21,150 @@ THE SOFTWARE.
*/
#include <unistd.h>
#include <iostream>
#include "rdc/rdc.h"
int main(int, char **) {
rdc_status_t result;
rdc_handle_t rdc_handle;
bool standalone = false;
char hostIpAddress[] = {"127.0.0.1:50051"};
char group_name[] = {"group1"};
char job_id[] = {"123"};
int main(int, char**) {
rdc_status_t result;
rdc_handle_t rdc_handle;
bool standalone = false;
char hostIpAddress[] = {"127.0.0.1:50051"};
char group_name[] = {"group1"};
char job_id[] = {"123"};
// Select the embedded mode and standalone mode dynamically.
std::cout << "Start rdci in: \n";
std::cout << "0 - Embedded mode \n";
std::cout << "1 - Standalone mode \n";
while (!(std::cin >> standalone)) {
std::cout << "Invalid input.\n";
std::cin.clear();
std::cin.ignore();
}
std::cout << std::endl;
std::cout << (standalone ? "Standalone mode selected.\n" : "Embedded mode selected.\n");
// Select the embedded mode and standalone mode dynamically.
std::cout << "Start rdci in: \n";
std::cout << "0 - Embedded mode \n";
std::cout << "1 - Standalone mode \n";
while (!(std::cin >> standalone)) {
std::cout << "Invalid input.\n";
std::cin.clear();
std::cin.ignore();
}
std::cout << std::endl;
std::cout << (standalone?
"Standalone mode selected.\n":"Embedded mode selected.\n");
// Init the rdc
result = rdc_init(0);
// Init the rdc
result = rdc_init(0);
if (result != RDC_ST_OK) {
std::cout << "Error initializing RDC. Return: " << rdc_status_string(result) << std::endl;
goto cleanup;
} else {
std::cout << "RDC Initialized.\n";
}
if (standalone) { // standalone
result = rdc_connect(hostIpAddress, &rdc_handle, nullptr, nullptr, nullptr);
if (result != RDC_ST_OK) {
std::cout << "Error initializing RDC. Return: " <<
rdc_status_string(result) << std::endl;
goto cleanup;
} else {
std::cout << "RDC Initialized.\n";
std::cout << "Error connecting to remote rdcd. Return: " << rdc_status_string(result)
<< std::endl;
goto cleanup;
}
if (standalone) { // standalone
result = rdc_connect(hostIpAddress, &rdc_handle,
nullptr, nullptr, nullptr);
if ( result != RDC_ST_OK ) {
std::cout << "Error connecting to remote rdcd. Return: "
<< rdc_status_string(result) << std::endl;
goto cleanup;
}
} else { // embedded
result = rdc_start_embedded(RDC_OPERATION_MODE_MANUAL, &rdc_handle);
if (result != RDC_ST_OK) {
std::cout << "Error starting embedded RDC engine. Return: "
<< rdc_status_string(result) << std::endl;
goto cleanup;
}
}
// Now we can use the same API for both standalone and embedded
// (1) create group and add GPUs
rdc_gpu_group_t group_id;
result = rdc_group_gpu_create(rdc_handle, RDC_GROUP_EMPTY,
group_name, &group_id);
} else { // embedded
result = rdc_start_embedded(RDC_OPERATION_MODE_MANUAL, &rdc_handle);
if (result != RDC_ST_OK) {
std::cout << "Error creating group. Return: "
<< rdc_status_string(result);
std::cout << "Error starting embedded RDC engine. Return: " << rdc_status_string(result)
<< std::endl;
goto cleanup;
}
}
// Now we can use the same API for both standalone and embedded
// (1) create group and add GPUs
rdc_gpu_group_t group_id;
result = rdc_group_gpu_create(rdc_handle, RDC_GROUP_EMPTY, group_name, &group_id);
if (result != RDC_ST_OK) {
std::cout << "Error creating group. Return: " << rdc_status_string(result);
goto cleanup;
}
result = rdc_group_gpu_add(rdc_handle, group_id, 0); // Add GPU 0
if (result != RDC_ST_OK) {
std::cout << "Error adding group. Return: " << rdc_status_string(result);
goto cleanup;
}
// (2) start the recording. Set the sample frequency to once per second.
result = rdc_job_start_stats(rdc_handle, group_id, job_id, 1000000);
if (result != RDC_ST_OK) {
std::cout << "Error start job stats. Return: " << rdc_status_string(result);
goto cleanup;
}
// For standalone mode, the daemon will update and cache the samples
// In manual mode, we must call rdc_field_update_all periodically to
// take samples.
if (!standalone) { // embedded manual mode
for (int i = 5; i > 0; i--) { // As an example, we will take 5 samples
result = rdc_field_update_all(rdc_handle, 0);
if (result != RDC_ST_OK) {
std::cout << "Error update all fields. Return: " << rdc_status_string(result);
goto cleanup;
}
usleep(1000000);
}
} else { // standalone mode, do nothing
usleep(5000000); // sleep 5 seconds before fetch the stats
}
result = rdc_group_gpu_add(rdc_handle, group_id, 0); // Add GPU 0
if (result != RDC_ST_OK) {
std::cout << "Error adding group. Return: "
<< rdc_status_string(result);
goto cleanup;
}
// (3) stop the Slurm job, which will stop the watch
// We do not have to stop the job to get stats. The rdc_job_get_stats can be
// called at any time before stop
result = rdc_job_stop_stats(rdc_handle, job_id);
if (result != RDC_ST_OK) {
std::cout << "Error stop job stats. Return: " << rdc_status_string(result);
goto cleanup;
}
// (2) start the recording. Set the sample frequency to once per second.
result = rdc_job_start_stats(rdc_handle, group_id,
job_id, 1000000);
if (result != RDC_ST_OK) {
std::cout << "Error start job stats. Return: "
<< rdc_status_string(result);
goto cleanup;
}
// (4) Get the stats
rdc_job_info_t job_info;
result = rdc_job_get_stats(rdc_handle, job_id, &job_info);
// For standalone mode, the daemon will update and cache the samples
// In manual mode, we must call rdc_field_update_all periodically to
// take samples.
if (!standalone) { // embedded manual mode
for (int i=5; i > 0 ; i--) { // As an example, we will take 5 samples
result = rdc_field_update_all(rdc_handle, 0);
if (result != RDC_ST_OK) {
std::cout << "Error update all fields. Return: "
<< rdc_status_string(result);
goto cleanup;
}
usleep(1000000);
}
} else { // standalone mode, do nothing
usleep(5000000); // sleep 5 seconds before fetch the stats
}
if (result == RDC_ST_OK) {
std::cout << "|------- Execution Stats ----------+"
<< "------------------------------------\n";
std::cout << "| Start Time * | " << job_info.summary.start_time << "\n";
std::cout << "| End Time * | " << job_info.summary.end_time << "\n";
std::cout << "| Total Execution Time (sec) * | "
<< (job_info.summary.end_time - job_info.summary.start_time) << "\n";
std::cout << "+------- Performance Stats --------+"
<< "------------------------------------\n";
std::cout << "| Energy Consumed (Joules) | " << job_info.summary.energy_consumed
<< "\n";
std::cout << "| Power Usage (Watts) | "
<< "Max: " << job_info.summary.power_usage.max_value
<< " Min: " << job_info.summary.power_usage.min_value
<< " Avg: " << job_info.summary.power_usage.average << "\n";
std::cout << "| GPU Clock (MHz) | "
<< "Max: " << job_info.summary.gpu_clock.max_value
<< " Min: " << job_info.summary.gpu_clock.min_value
<< " Avg: " << job_info.summary.gpu_clock.average << "\n";
std::cout << "| GPU Utilization (%) | "
<< "Max: " << job_info.summary.gpu_utilization.max_value
<< " Min: " << job_info.summary.gpu_utilization.min_value
<< " Avg: " << job_info.summary.gpu_utilization.average << "\n";
std::cout << "| Max GPU Memory Used (bytes) * | " << job_info.summary.max_gpu_memory_used
<< "\n";
std::cout << "| Memory Utilization (%) | "
<< "Max: " << job_info.summary.memory_utilization.max_value
<< " Min: " << job_info.summary.memory_utilization.min_value
<< " Avg: " << job_info.summary.memory_utilization.average << "\n";
std::cout << "+----------------------------------+"
<< "------------------------------------\n";
} else {
std::cout << "No data for job stats found." << std::endl;
}
// (3) stop the Slurm job, which will stop the watch
// We do not have to stop the job to get stats. The rdc_job_get_stats can be
// called at any time before stop
result = rdc_job_stop_stats(rdc_handle, job_id);
if (result != RDC_ST_OK) {
std::cout << "Error stop job stats. Return: "
<< rdc_status_string(result);
goto cleanup;
}
// (4) Get the stats
rdc_job_info_t job_info;
result = rdc_job_get_stats(rdc_handle, job_id, &job_info);
if (result == RDC_ST_OK) {
std::cout << "|------- Execution Stats ----------+"
<<"------------------------------------\n";
std::cout << "| Start Time * | "
<< job_info.summary.start_time<< "\n";
std::cout << "| End Time * | "
<< job_info.summary.end_time <<"\n";
std::cout << "| Total Execution Time (sec) * | "
<< (job_info.summary.end_time-job_info.summary.start_time)
<< "\n";
std::cout << "+------- Performance Stats --------+"
<< "------------------------------------\n";
std::cout << "| Energy Consumed (Joules) | "
<< job_info.summary.energy_consumed << "\n";
std::cout << "| Power Usage (Watts) | "
<< "Max: " << job_info.summary.power_usage.max_value
<< " Min: "<< job_info.summary.power_usage.min_value
<< " Avg: "<< job_info.summary.power_usage.average << "\n";
std::cout << "| GPU Clock (MHz) | "
<< "Max: " <<job_info.summary.gpu_clock.max_value
<<" Min: " << job_info.summary.gpu_clock.min_value
<<" Avg: "<< job_info.summary.gpu_clock.average << "\n";
std::cout << "| GPU Utilization (%) | "
<< "Max: " <<job_info.summary.gpu_utilization.max_value
<<" Min: "<< job_info.summary.gpu_utilization.min_value
<<" Avg: "<< job_info.summary.gpu_utilization.average << "\n";
std::cout << "| Max GPU Memory Used (bytes) * | "
<< job_info.summary.max_gpu_memory_used << "\n";
std::cout << "| Memory Utilization (%) | "
<< "Max: " << job_info.summary.memory_utilization.max_value
<<" Min: "<< job_info.summary.memory_utilization.min_value
<<" Avg: "<< job_info.summary.memory_utilization.average << "\n";
std::cout << "+----------------------------------+"
<< "------------------------------------\n";
} else {
std::cout << "No data for job stats found." << std::endl;
}
// Cleanup consists of shutting down RDC.
cleanup:
std::cout << "Cleaning up.\n";
if (standalone)
rdc_disconnect(rdc_handle);
else
rdc_stop_embedded(rdc_handle);
rdc_shutdown();
return result;
// Cleanup consists of shutting down RDC.
cleanup:
std::cout << "Cleaning up.\n";
if (standalone)
rdc_disconnect(rdc_handle);
else
rdc_stop_embedded(rdc_handle);
rdc_shutdown();
return result;
}