[rocm_smi.py] error feedback improvement

Cleaning overally verbose error reporting system.

Signed-off-by: Elena Sakhnovitch
Signed-off-by: Sreekant Somasekharan
Change-Id: Icc96086810b8dcfc426848b8c349a2572026c3bd


[ROCm/rocm_smi_lib commit: 4dd2398f3d]

projects/rocm-smi-lib/python_smi_tools/rocm_smi.py

@@ -509,7 +509,6 @@ def printLog(device, metricName, value):
         logstr = 'GPU[%s]\t\t: %s' % (device, metricName)
     if device is None:
         logstr = logstr[13:]
-    logging.debug(logstr)
     # Force thread safe printing
     print(logstr + '\n', end='')
 
@@ -2026,7 +2025,7 @@ def showProfile(deviceList):
     status = rsmi_power_profile_status_t()
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_power_profile_presets_get(device, 0, byref(status))
-        if rsmi_ret_ok(ret, device, 'profiles'):
+        if rsmi_ret_ok(ret, device, 'profiles', silent=False):
             binaryMaskString = str(format(status.available_profiles, '07b'))[::-1]
             bitMaskPosition = 0
             profileNumber = 0
@@ -2058,7 +2057,7 @@ def showRange(deviceList, rangeType):
     odvf = rsmi_od_volt_freq_data_t()
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_od_volt_info_get(device, byref(odvf))
-        if rsmi_ret_ok(ret, device, 'od volt'):
+        if rsmi_ret_ok(ret, device, 'od volt', silent=False):
             if rangeType == 'sclk':
                 printLog(device, 'Valid sclk range: %sMhz - %sMhz' % (
                 int(odvf.curr_sclk_range.lower_bound / 1000000), int(odvf.curr_sclk_range.upper_bound / 1000000)), None)
@@ -2080,8 +2079,6 @@ def showRange(deviceList, rangeType):
                                  None)
                 else:
                     printLog(device, 'Unable to display %s range' % (rangeType), None)
-        else:
-            printLog(device, 'Unable to display %s range' % (rangeType), None)
     printLogSpacer()
 
 
@@ -2294,13 +2291,11 @@ def showVoltageCurve(deviceList):
     odvf = rsmi_od_volt_freq_data_t()
     for device in deviceList:
         ret = rocmsmi.rsmi_dev_od_volt_info_get(device, byref(odvf))
-        if rsmi_ret_ok(ret, device, 'od volt'):
+        if rsmi_ret_ok(ret, device, 'od volt', silent=False):
             for position in range(3):
                 printLog(device, 'Voltage point %d: %sMhz %smV' % (
                 position, int(list(odvf.curve.vc_points)[position].frequency / 1000000),
                 int(list(odvf.curve.vc_points)[position].voltage)), None)
-        else:
-            printLog(device, 'Voltage Curve is not supported', None)
     printLogSpacer()
 
 
@@ -2790,10 +2785,10 @@ def rsmi_ret_ok(my_ret, device=None, metric=None, silent=False):
             returnString += ' %s: ' % (metric)
         returnString += '%s\t' % (err_str.value.decode())
         if not PRINT_JSON:
-            if silent:
-                logging.info('%s', returnString)
-            else:
-                logging.error('%s', returnString)
+            logging.debug('%s', returnString)
+            if not silent:
+                if my_ret in rsmi_status_verbose_err_out:
+                    printLog(device, rsmi_status_verbose_err_out[my_ret], None)
         RETCODE = my_ret
         return False
     return True

projects/rocm-smi-lib/python_smi_tools/rsmiBindings.py

@@ -69,6 +69,27 @@ class rsmi_status_t(c_int):
     RSMI_STATUS_UNKNOWN_ERROR = 0xFFFFFFFF
 
 
+#Dictionary of rsmi ret codes and it's verbose output
+rsmi_status_verbose_err_out = {
+    rsmi_status_t.RSMI_STATUS_SUCCESS: 'Operation was successful',
+    rsmi_status_t.RSMI_STATUS_INVALID_ARGS: 'Invalid arguments provided',
+    rsmi_status_t.RSMI_STATUS_NOT_SUPPORTED: 'Not supported on the given system',
+    rsmi_status_t.RSMI_STATUS_FILE_ERROR: 'Problem accessing a file',
+    rsmi_status_t.RSMI_STATUS_PERMISSION: 'Permission denied',
+    rsmi_status_t.RSMI_STATUS_OUT_OF_RESOURCES: 'Unable to acquire memory or other resource',
+    rsmi_status_t.RSMI_STATUS_INTERNAL_EXCEPTION: 'An internal exception was caught',
+    rsmi_status_t.RSMI_STATUS_INPUT_OUT_OF_BOUNDS: 'Provided input is out of allowable or safe range',
+    rsmi_status_t.RSMI_INITIALIZATION_ERROR: 'Error occured during rsmi initialization',
+    rsmi_status_t.RSMI_STATUS_NOT_YET_IMPLEMENTED: 'Requested function is not implemented on this setup',
+    rsmi_status_t.RSMI_STATUS_NOT_FOUND: 'Item searched for but not found',
+    rsmi_status_t.RSMI_STATUS_INSUFFICIENT_SIZE: 'Insufficient resources available',
+    rsmi_status_t.RSMI_STATUS_INTERRUPT: 'Interrupt occured during execution',
+    rsmi_status_t.RSMI_STATUS_UNEXPECTED_SIZE: 'Unexpected amount of data read',
+    rsmi_status_t.RSMI_STATUS_NO_DATA: 'No data found for the given input',
+    rsmi_status_t.RSMI_STATUS_UNKNOWN_ERROR: 'Unknown error occured'
+}
+
+
 class rsmi_init_flags_t(c_int):
     RSMI_INIT_FLAG_ALL_GPUS = 0x1