AI/rocm-systems
2
0
انشعاب 0
کد مسائل تقاضاهای واکشی Actions Packages پروژه‌ها انتشارها دانشنامه فعالیت

Merge "Added slowdown temperature metrics" into amd-dev

فهرست منبع
This commit is contained in:
Maisam Arif
2023-09-19 16:47:22 -04:00
کامیت شده توسط Gerrit Code Review
والد d756f5ee72 93c54f52cd
کامیت 72cd9ee6bd
2فایلهای تغییر یافته به همراه76 افزوده شده و 25 حذف شده
دانلود پرونده وصله دانلود فایل تغییرات diff Expand all files Collapse all files
amdsmi_cli/amdsmi_cli.py
+5 -4
مشاهده پرونده
@@ -76,10 +76,11 @@ if __name__ == "__main__":
amd_smi_commands.logger.destination = args.file amd_smi_commands.logger.destination = args.file
if args.loglevel: if args.loglevel:
logging_dict = {'DEBUG' : logging.DEBUG, logging_dict = {'DEBUG' : logging.DEBUG,
'INFO' : logging.INFO, 'INFO' : logging.INFO,
'WARNING': logging.WARNING, 'WARNING': logging.WARNING,
'ERROR': logging.ERROR, 'ERROR': logging.ERROR,
'CRITICAL': logging.CRITICAL} 'CRITICAL': logging.CRITICAL}
# Enable debug logs on amdsmi library ie. RSMI_LOGGING = 1 in environment or otherwise
logging.basicConfig(format='%(levelname)s: %(message)s', level=logging_dict[args.loglevel]) logging.basicConfig(format='%(levelname)s: %(message)s', level=logging_dict[args.loglevel])
# Execute subcommands # Execute subcommands
amdsmi_cli/amdsmi_commands.py
+71 -21
مشاهده پرونده
@@ -301,37 +301,74 @@ class AMDSMICommands():
if not self.all_arguments: if not self.all_arguments:
raise e raise e
# Edge temperature limits
try: try:
temp_edge_limit_error = False slowdown_temp_edge_limit_error = False
temp_edge_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu, slowdown_temp_edge_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu,
amdsmi_interface.AmdSmiTemperatureType.EDGE, amdsmi_interface.AmdSmiTemperatureMetric.CRITICAL) amdsmi_interface.AmdSmiTemperatureType.EDGE, amdsmi_interface.AmdSmiTemperatureMetric.CRITICAL)
except amdsmi_exception.AmdSmiLibraryException as e: except amdsmi_exception.AmdSmiLibraryException as e:
temp_edge_limit_error = True slowdown_temp_edge_limit_error = True
temp_edge_limit = e.get_error_info() slowdown_temp_edge_limit = e.get_error_info()
if not self.all_arguments: if not self.all_arguments:
raise e raise e
if temp_edge_limit == 0: if slowdown_temp_edge_limit == 0:
temp_edge_limit_error = True slowdown_temp_edge_limit_error = True
temp_edge_limit = 'N/A' slowdown_temp_edge_limit = 'N/A'
try: try:
temp_hotspot_limit_error = False shutdown_temp_edge_limit_error = False
temp_hotspot_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu, shutdown_temp_edge_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu,
amdsmi_interface.AmdSmiTemperatureType.EDGE, amdsmi_interface.AmdSmiTemperatureMetric.EMERGENCY)
except amdsmi_exception.AmdSmiLibraryException as e:
shutdown_temp_edge_limit_error = True
shutdown_temp_edge_limit = e.get_error_info()
if not self.all_arguments:
raise e
if shutdown_temp_edge_limit == 0:
shutdown_temp_edge_limit_error = True
shutdown_temp_edge_limit = 'N/A'
# Hotspot/Junction temperature limits
try:
slowdown_temp_hotspot_limit_error = False
slowdown_temp_hotspot_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu,
amdsmi_interface.AmdSmiTemperatureType.HOTSPOT, amdsmi_interface.AmdSmiTemperatureMetric.CRITICAL) amdsmi_interface.AmdSmiTemperatureType.HOTSPOT, amdsmi_interface.AmdSmiTemperatureMetric.CRITICAL)
except amdsmi_exception.AmdSmiLibraryException as e: except amdsmi_exception.AmdSmiLibraryException as e:
temp_hotspot_limit_error = True slowdown_temp_hotspot_limit_error = True
temp_hotspot_limit = e.get_error_info() slowdown_temp_hotspot_limit = e.get_error_info()
if not self.all_arguments: if not self.all_arguments:
raise e raise e
try: try:
temp_vram_limit_error = False shutdown_temp_hotspot_limit_error = False
temp_vram_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu, shutdown_temp_hotspot_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu,
amdsmi_interface.AmdSmiTemperatureType.HOTSPOT, amdsmi_interface.AmdSmiTemperatureMetric.EMERGENCY)
except amdsmi_exception.AmdSmiLibraryException as e:
shutdown_temp_hotspot_limit_error = True
shutdown_temp_hotspot_limit = e.get_error_info()
if not self.all_arguments:
raise e
# VRAM temperature limits
try:
slowdown_temp_vram_limit_error = False
slowdown_temp_vram_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu,
amdsmi_interface.AmdSmiTemperatureType.VRAM, amdsmi_interface.AmdSmiTemperatureMetric.CRITICAL) amdsmi_interface.AmdSmiTemperatureType.VRAM, amdsmi_interface.AmdSmiTemperatureMetric.CRITICAL)
except amdsmi_exception.AmdSmiLibraryException as e: except amdsmi_exception.AmdSmiLibraryException as e:
temp_vram_limit_error = True slowdown_temp_vram_limit_error = True
temp_vram_limit = e.get_error_info() slowdown_temp_vram_limit = e.get_error_info()
if not self.all_arguments:
raise e
try:
shutdown_temp_vram_limit_error = False
shutdown_temp_vram_limit = amdsmi_interface.amdsmi_get_temp_metric(args.gpu,
amdsmi_interface.AmdSmiTemperatureType.VRAM, amdsmi_interface.AmdSmiTemperatureMetric.EMERGENCY)
except amdsmi_exception.AmdSmiLibraryException as e:
shutdown_temp_vram_limit_error = True
shutdown_temp_vram_limit = e.get_error_info()
if not self.all_arguments: if not self.all_arguments:
raise e raise e
@@ -342,18 +379,31 @@ class AMDSMICommands():
current_power_limit = f"{current_power_limit} {unit}" current_power_limit = f"{current_power_limit} {unit}"
unit = '\N{DEGREE SIGN}C' unit = '\N{DEGREE SIGN}C'
if not temp_edge_limit_error: if not slowdown_temp_edge_limit_error:
temp_edge_limit = f"{temp_edge_limit} {unit}" slowdown_temp_edge_limit = f"{slowdown_temp_edge_limit} {unit}"
if not temp_hotspot_limit_error: if not slowdown_temp_hotspot_limit_error:
temp_hotspot_limit = f"{temp_hotspot_limit} {unit}" slowdown_temp_hotspot_limit = f"{slowdown_temp_hotspot_limit} {unit}"
if not temp_vram_limit_error: if not slowdown_temp_vram_limit_error:
temp_vram_limit = f"{temp_vram_limit} {unit}" slowdown_temp_vram_limit = f"{slowdown_temp_vram_limit} {unit}"
if not shutdown_temp_edge_limit_error:
shutdown_temp_edge_limit = f"{shutdown_temp_edge_limit} {unit}"
if not shutdown_temp_hotspot_limit_error:
shutdown_temp_hotspot_limit = f"{shutdown_temp_hotspot_limit} {unit}"
if not shutdown_temp_vram_limit_error:
shutdown_temp_vram_limit = f"{shutdown_temp_vram_limit} {unit}"
limit_info = {} limit_info = {}
# Power limits # Power limits
limit_info['max_power'] = max_power_limit limit_info['max_power'] = max_power_limit
limit_info['current_power'] = current_power_limit limit_info['current_power'] = current_power_limit
# Shutdown limits
limit_info['slowdown_edge_temperature'] = slowdown_temp_edge_limit
limit_info['slowdown_hotspot_temperature'] = slowdown_temp_hotspot_limit
limit_info['slowdown_vram_temperature'] = slowdown_temp_vram_limit
limit_info['shutdown_edge_temperature'] = shutdown_temp_edge_limit
limit_info['shutdown_hotspot_temperature'] = shutdown_temp_hotspot_limit
limit_info['shutdown_vram_temperature'] = shutdown_temp_vram_limit
static_dict['limit'] = limit_info static_dict['limit'] = limit_info
if args.driver: if args.driver:
try: try: