AI/rocm-systems
2
0 Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

[SWDEV-488303] Fixed process list information source

Telusuri Sumber 
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
Change-Id: Iec3416cb5ca1bdd806c3225b514bbf3dbf8c0d2e
This commit is contained in:
Maisam Arif authored and Arif, Maisam committed 2025-05-30 20:48:29 -05:00
1 orang tua cc4dfd834f
melakukan cebb0799cb
9 files changed
+108 -122
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

amdsmi_cli/amdsmi_commands.py
+16 -8
Melihat File
@@ -2165,14 +2165,14 @@ class AMDSMICommands():
clk_value = int(clk.get("value", 0))
else:
if isinstance(clk, str):
clk_value = int(str(clk).split()[0])
clk_value = int(str(clk).split()[0])
else:
clk_value = int(clk)
if isinstance(min_clk, dict):
min_clk_value = int(min_clk.get("value", 0))
else:
if isinstance(min_clk, str):
min_clk_value = int(str(min_clk).split()[0])
min_clk_value = int(str(min_clk).split()[0])
else:
min_clk_value = int(min_clk)
# If the clk value is less than the min_clk value, then deep sleep is enabled
@@ -5755,15 +5755,22 @@ class AMDSMICommands():
# Clean processes dictionary
filtered_process_values = []
for process_info in process_list:
process_info.pop('mem') # Remove 'mem' value
process_info.pop('engine_usage') # Remove 'engine_usage' value
process_info['mem_usage'] = process_info.pop('mem')
process_info['cu_occupancy'] = process_info.pop('cu_occupancy')
memory_usage_unit = "B"
if self.logger.is_human_readable_format():
process_info['mem_usage'] = self.helpers.convert_bytes_to_readable(process_info['mem_usage'])
for usage_metric in process_info['memory_usage']:
process_info["memory_usage"][usage_metric] = self.helpers.convert_bytes_to_readable(process_info["memory_usage"][usage_metric])
memory_usage_unit = ""
process_info['mem_usage'] = self.helpers.unit_format(self.logger,
process_info['mem_usage'],
memory_usage_unit)
for usage_metric in process_info['memory_usage']:
process_info['memory_usage'][usage_metric] = self.helpers.unit_format(self.logger,
process_info['memory_usage'][usage_metric],
@@ -5796,8 +5803,8 @@ class AMDSMICommands():
# Build the process table's title and header
self.logger.secondary_table_title = "PROCESS INFO"
self.logger.secondary_table_header = 'GPU'.rjust(3) + "NAME".rjust(22) + "PID".rjust(9) + "GTT_MEM".rjust(10) + \
"CPU_MEM".rjust(10) + "VRAM_MEM".rjust(10) + "CU%".rjust(9)
self.logger.secondary_table_header = 'GPU'.rjust(3) + "NAME".rjust(19) + "PID".rjust(9) + "GTT_MEM".rjust(10) + \
"CPU_MEM".rjust(10) + "VRAM_MEM".rjust(10) + "MEM_USG".rjust(10) + "CU%".rjust(9)
if watching_output:
self.logger.secondary_table_header = 'TIMESTAMP'.rjust(10) + ' ' + self.logger.secondary_table_header
@@ -6446,7 +6453,7 @@ class AMDSMICommands():
else:
with self.logger.destination.open('a', encoding="utf-8") as output_file:
output_file.write(legend_output + '\n')
def ras(self, args, multiple_devices=False, gpu=None, cper=None, afid=None,
severity=None, folder=None, file_limit=None, cper_file=None, follow=None):
@@ -6494,7 +6501,7 @@ class AMDSMICommands():
if not args.gpu:
return
if not isinstance(args.gpu, list):
args.gpu = [args.gpu]
@@ -6506,6 +6513,7 @@ class AMDSMICommands():
break
time.sleep(1)
def default(self, args):
"""Display the default amdsmi view when no args are given."""
@@ -6560,7 +6568,7 @@ class AMDSMICommands():
else:
partition_mode = f"{current_comp}/{current_mem}"
gpu_info_dict.update({"partition_mode": partition_mode})
# GPU name market name and OAM ID
try:
asic_info = amdsmi_interface.amdsmi_get_gpu_asic_info(processor)
amdsmi_cli/amdsmi_helpers.py
-1
Melihat File
@@ -1102,7 +1102,6 @@ class AMDSMIHelpers():
# Loop through all entries in the dictionary.
for entry_index, entry in enumerate(entries.values()):
# Assume 'entry' is a dictionary with keys: "error_severity" and "notify_type".
error_severity = entry.get("error_severity", "Unknown")
notify_type = entry.get("notify_type", "Unknown")
amdsmi_cli/amdsmi_logger.py
+7 -4
Melihat File
@@ -213,23 +213,26 @@ class AMDSMILogger():
for process_dict in value:
if process_dict['process_info'] == "No running processes detected":
# Add N/A for empty process_info
table_values += "N/A".rjust(20) + "N/A".rjust(9) + "N/A".rjust(10) + \
"N/A".rjust(10) + "N/A".rjust(10) + "N/A".rjust(9) + '\n'
table_values += "N/A".rjust(17) + "N/A".rjust(9) + "N/A".rjust(10) + \
"N/A".rjust(10) + "N/A".rjust(10) + "N/A".rjust(10) + \
"N/A".rjust(9) + '\n'
else:
#Fix this herre
for process_key, process_value in process_dict['process_info'].items():
string_process_value = str(process_value)
if process_key == "name":
# Truncate name if too long
process_name = string_process_value[:20]
process_name = string_process_value[:17]
if process_name == "":
process_name = "N/A"
table_values += process_name.rjust(20)
table_values += process_name.rjust(17)
elif process_key == "pid":
table_values += string_process_value.rjust(9)
elif process_key == "memory_usage":
for memory_key, memory_value in process_value.items():
table_values += str(memory_value).rjust(10)
elif process_key == "mem_usage":
table_values += string_process_value.rjust(10)
elif process_key == "cu_occupancy":
table_values += string_process_value.rjust(9)
# Add the stored gpu and stored timestamp to the next line