Cleanup of system profiling scripts

script/app-mlperf-inference/customize.py

@@ -465,14 +465,16 @@ def postprocess(i):
             power_end_time = pd.Timestamp(datetime.strptime(match_end, '%m-%d-%Y %H:%M:%S.%f')).replace(tzinfo=timezone.utc)
         #converts timestamp key value to datetime objects
         sys_utilisation_log['timestamp'] = pd.to_datetime(sys_utilisation_log['timestamp'])
+        '''
         for i in range(len(sys_utilisation_log['timestamp'])):
             print(f"{sys_utilisation_log['timestamp'][i]} {power_begin_time}")
             print(sys_utilisation_log['timestamp'][i]>=power_begin_time)
-        print(f"{sys_utilisation_log['timestamp'][0]} {power_begin_time}")
-        print(sys_utilisation_log['timestamp'][0]>=power_begin_time)
+        '''
+        #print(f"{sys_utilisation_log['timestamp'][0]} {power_begin_time}")
+        #print(sys_utilisation_log['timestamp'][0]>=power_begin_time)
         filtered_log = sys_utilisation_log[(sys_utilisation_log['timestamp'] >= power_begin_time) & 
                                (sys_utilisation_log['timestamp'] <= power_end_time)]
-        print(filtered_log)
+        #print(filtered_log)
         # Calculate average of cpu_utilisation and used_memory_gb
         system_utilisation_info_dump["avg_cpu_utilisation"] = filtered_log['cpu_utilisation'].mean()
         system_utilisation_info_dump["avg_used_memory_gb"] = filtered_log['used_memory_gb'].mean()

script/benchmark-program/customize.py

@@ -56,19 +56,20 @@ def preprocess(i):
         elif env.get('CM_MLPERF_DEVICE', '') == "cpu":
             env['CM_SYS_UTILISATION_SCRIPT_TAGS'] += ',_cpu'
         # this section is for supplying the input arguments/tags
-        env['CM_SYS_UTILISATION_SCRIPT_TAGS'] += ' --log_dir=\"' + logs_dir + '\"'   # specify the logs directory
+        env['CM_SYS_UTILISATION_SCRIPT_TAGS'] += ' --log_dir=\'' + logs_dir + '\''   # specify the logs directory
         if env.get('CM_SYSTEM_INFO_MEASUREMENT_INTERVAL', '') != '':        # specifying the interval in which the system information should be measured
             env['CM_SYS_UTILISATION_SCRIPT_TAGS'] += ' --interval=\"' + env['CM_SYSTEM_INFO_MEASUREMENT_INTERVAL'] + '\"'
 
     # generate the pre run cmd - recording runtime system infos 
     pre_run_cmd = ""
     if env.get('CM_PROFILE_NVIDIA_POWER', '') == "on":
-        pass
+        #pass
         # Note: To be fixed
         # running the script as a process in background
-        # pre_run_cmd = pre_run_cmd + '\"' + 'cm run script --tags=runtime,system,utilisation' + env['CM_SYS_UTILISATION_SCRIPT_TAGS'] + ' &\" '
+         pre_run_cmd = pre_run_cmd +  'cm run script --tags=runtime,system,utilisation' + env['CM_SYS_UTILISATION_SCRIPT_TAGS'] + ' --quiet  & '
         # obtain the command if of the background process
-        # pre_run_cmd += 'cmd_pid=$!' + ' && ' + 'echo $cmd_pid' + ' && '
+         pre_run_cmd += ' cmd_pid=\$!' + ' && ' + 'echo CMD_PID=\$cmd_pid'
+        #pre_run_cmd += ' && '
         # # check whether the command is running or not
         # pre_run_cmd += "ps -p $cmd_pid || { echo 'Process $cmd_pid is not running, exiting'; exit 1; }"
     env['CM_PRE_RUN_CMD'] = pre_run_cmd
@@ -77,8 +78,8 @@ def preprocess(i):
     # generate the post run cmd - for killing the process that records runtime system infos
     post_run_cmd = ""
     if env.get('CM_PROFILE_NVIDIA_POWER', '') == "on":
-        pass
-        # post_run_cmd += "echo 'killing process ${cmd_pid}' && kill -TERM ${cmd_pid} && test $? -eq 0 || exit $? "
+        #pass
+        post_run_cmd += "echo 'killing process \${cmd_pid}' && kill -TERM \${cmd_pid}"
     env['CM_POST_RUN_CMD'] = post_run_cmd
     print(f"Post run command for killing the process that measures the runtime system information: {post_run_cmd}")
 

script/benchmark-program/run.sh

@@ -59,7 +59,9 @@ else
   test $exitstatus -eq 0 || $exitstatus
 fi
 
+eval ${CM_POST_RUN_CMD}
 test $? -eq 0 || exit $? 
 
+test $exitstatus -eq 0 || $exitstatus
+
 
-eval ${CM_POST_RUN_CMD}

script/detect-cpu/run.sh

@@ -1,5 +1,16 @@
 #!/bin/bash
 
+# Function to extract a field from /proc/cpuinfo
+extract_field() {
+  local key="$1"
+  local default="$2"
+  # Use awk to find the first occurrence and extract the value
+  local value=$(awk -F: -v key="$key" '$1 ~ key {print $2; exit}' /proc/cpuinfo | sed 's/^[[:space:]]*//;s/[[:space:]]*$//')
+
+  # Check if value is empty and assign default if needed
+  echo "${value:-$default}"
+}
+
 if [[ ${CM_HOST_OS_FLAVOR} == "macos" ]]; then
     sysctl -a | grep hw > tmp-lscpu.out
 else
@@ -8,15 +19,30 @@ else
     echo "CM_HOST_MEMORY_CAPACITY=$memory_capacity">>tmp-run-env.out
     disk_capacity=`df -h --total -l |grep total |tr -s ' '|cut -d' ' -f2`
     echo "CM_HOST_DISK_CAPACITY=$disk_capacity">>tmp-run-env.out
+
     # extract cpu information which are not there in lscpu
-    cpuinfo=$(cat /proc/cpuinfo)
-    echo "CM_HOST_CPU_WRITE_PROTECT_SUPPORT=$(echo "$cpuinfo" | grep -m 1 "wp" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CPU_MICROCODE=$(echo "$cpuinfo" | grep -m 1 "microcode" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CPU_FPU_SUPPORT=$(echo "$cpuinfo" | grep -m 1 "fpu" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CPU_FPU_EXCEPTION_SUPPORT=$(echo "$cpuinfo" | grep -m 1 "fpu_exception" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CPU_BUGS=$(echo "$cpuinfo" | grep -m 1 "bugs" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CPU_TLB_SIZE=$(echo "$cpuinfo" | grep -m 1 "TLB size" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CPU_CFLUSH_SIZE=$(echo "$cpuinfo" | grep -m 1 "clush size" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_CACHE_ALLIGNMENT_SIZE=$(echo "$cpuinfo" | grep -m 1 "cache_alignment" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out
-    echo "CM_HOST_POWER_MANAGEMENT=$(echo "$cpuinfo" | grep -m 1 "power management" | cut -d ':' -f 2 | xargs)">>tmp-run-env.out    
+    CM_HOST_CPU_WRITE_PROTECT_SUPPORT=$(extract_field "wp" "Not Found")
+    CM_HOST_CPU_MICROCODE=$(extract_field "microcode" "Not Found")
+    CM_HOST_CPU_FPU_SUPPORT=$(extract_field "fpu" "Not Found")
+    CM_HOST_CPU_FPU_EXCEPTION_SUPPORT=$(extract_field "fpu_exception" "Not Found")
+    CM_HOST_CPU_BUGS=$(extract_field "bugs" "Not Found")
+    CM_HOST_CPU_TLB_SIZE=$(extract_field "TLB size" "Not Found")
+    CM_HOST_CPU_CFLUSH_SIZE=$(extract_field "clflush size" "Not Found")
+    CM_HOST_CACHE_ALIGNMENT_SIZE=$(extract_field "cache_alignment" "Not Found")
+    CM_HOST_POWER_MANAGEMENT=$(extract_field "power management" "Not Found")
+
+    # Write results to a file
+    {
+      echo "CM_HOST_CPU_WRITE_PROTECT_SUPPORT=$CM_HOST_CPU_WRITE_PROTECT_SUPPORT"
+      echo "CM_HOST_CPU_MICROCODE=$CM_HOST_CPU_MICROCODE"
+      echo "CM_HOST_CPU_FPU_SUPPORT=$CM_HOST_CPU_FPU_SUPPORT"
+      echo "CM_HOST_CPU_FPU_EXCEPTION_SUPPORT=$CM_HOST_CPU_FPU_EXCEPTION_SUPPORT"
+      echo "CM_HOST_CPU_BUGS=$CM_HOST_CPU_BUGS"
+      echo "CM_HOST_CPU_TLB_SIZE=$CM_HOST_CPU_TLB_SIZE"
+      echo "CM_HOST_CPU_CFLUSH_SIZE=$CM_HOST_CPU_CFLUSH_SIZE"
+      echo "CM_HOST_CACHE_ALIGNMENT_SIZE=$CM_HOST_CACHE_ALIGNMENT_SIZE"
+      echo "CM_HOST_POWER_MANAGEMENT=$CM_HOST_POWER_MANAGEMENT"
+    } >> tmp-run-env.out
 fi
+
+