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

Merge remote-tracking branch 'rocmsmi/amd-staging' into HEAD

Browse Source 
Change-Id: I65ed7f3a0d1b6e58bc8377932d7c39db21d1b422
This commit is contained in:
Galantsev, Dmitrii
2023-09-21 19:10:41 -05:00
parent def17accbe e0483f2ee2
commit 5c41319c83
38 changed files with 1475 additions and 1055 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocm_smi/python_smi_tools/rocm_smi.py
+163 -100
View File
@@ -173,10 +173,12 @@ def formatMatrixToJSON(deviceList, matrix, metricName):
printSysLog(metricName.format(deviceList[row_indx], deviceList[col_ind]), valueStr)
def getBus(device):
def getBus(device, silent=False):
""" Return the bus identifier of a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
bdfid = c_uint64(0)
ret = rocmsmi.rsmi_dev_pci_id_get(device, byref(bdfid))
@@ -188,16 +190,18 @@ def getBus(device):
function = bdfid.value & 0x7
pic_id = '{:04X}:{:02X}:{:02X}.{:0X}'.format(domain, bus, device, function)
if rsmi_ret_ok(ret, device, 'get_pci_id'):
if rsmi_ret_ok(ret, device, 'get_pci_id', silent):
return pic_id
def getFanSpeed(device):
def getFanSpeed(device, silent=True):
""" Return a tuple with the fan speed (value,%) for a specified device,
or (None,None) if either current fan speed or max fan speed cannot be
obtained
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is on.
"""
fanLevel = c_int64()
fanMax = c_int64()
@@ -209,7 +213,7 @@ def getFanSpeed(device):
/sys/class/drm/cardX/device/hwmon/hwmonX/pwmX
"""
ret = rocmsmi.rsmi_dev_fan_speed_get(device, sensor_ind, byref(fanLevel))
if rsmi_ret_ok(ret, device, 'get_fan_speed', True):
if rsmi_ret_ok(ret, device, 'get_fan_speed', silent):
fl = fanLevel.value
last_ret = ret
@@ -217,7 +221,7 @@ def getFanSpeed(device):
/sys/class/drm/cardX/device/hwmon/hwmonX/pwmX
"""
ret = rocmsmi.rsmi_dev_fan_speed_max_get(device, sensor_ind, byref(fanMax))
if rsmi_ret_ok(ret, device, 'get_fan_max_speed', True):
if rsmi_ret_ok(ret, device, 'get_fan_max_speed', silent):
fm = fanMax.value
""" In case we had an error before, we don't overwrite it with a
@@ -232,59 +236,67 @@ def getFanSpeed(device):
return (last_ret, fl, round((float(fl) / float(fm)) * 100, 2))
def getGpuUse(device):
def getGpuUse(device, silent=False):
""" Return the current GPU usage as a percentage
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
percent = c_uint32()
ret = rocmsmi.rsmi_dev_busy_percent_get(device, byref(percent))
if rsmi_ret_ok(ret, device, 'GPU Utilization '):
if rsmi_ret_ok(ret, device, 'GPU Utilization ', silent):
return percent.value
return -1
def getId(device):
def getId(device, silent=False):
""" Return the hexadecimal value of a device's ID
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
dv_id = c_short()
ret = rocmsmi.rsmi_dev_id_get(device, byref(dv_id))
if rsmi_ret_ok(ret, device, 'get_device_id'):
if rsmi_ret_ok(ret, device, 'get_device_id', silent):
return hex(dv_id.value)
def getRev(device):
def getRev(device, silent=False):
""" Return the hexadecimal value of a device's Revision
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
dv_rev = c_short()
ret = rocmsmi.rsmi_dev_revision_get(device, byref(dv_rev))
if rsmi_ret_ok(ret, device, 'get_device_rev'):
if rsmi_ret_ok(ret, device, 'get_device_rev', silent):
return hex(dv_rev.value)
def getMaxPower(device):
def getMaxPower(device, silent=False):
""" Return the maximum power cap of a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
power_cap = c_uint64()
ret = rocmsmi.rsmi_dev_power_cap_get(device, 0, byref(power_cap))
if rsmi_ret_ok(ret, device, 'get_power_cap'):
if rsmi_ret_ok(ret, device, 'get_power_cap', silent):
return power_cap.value / 1000000
return -1
def getMemInfo(device, memType, quiet=False):
def getMemInfo(device, memType, silent=False):
""" Returns a tuple of (memory_used, memory_total) of
the requested memory type usage for the device specified
@param device: DRM device identifier
@param type: [vram|vis_vram|gtt] Memory type to return
@param quiet=Turn on to silience error output
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off,
which exposes any issue accessing the different
memory types.
@@ -300,11 +312,11 @@ def getMemInfo(device, memType, quiet=False):
memTotal = None
ret = rocmsmi.rsmi_dev_memory_usage_get(device, memory_type_l.index(memType), byref(memoryUse))
if rsmi_ret_ok(ret, device, 'get_memory_usage_' + str(memType), quiet):
if rsmi_ret_ok(ret, device, 'get_memory_usage_' + str(memType), silent):
memUsed = memoryUse.value
ret = rocmsmi.rsmi_dev_memory_total_get(device, memory_type_l.index(memType), byref(memoryTot))
if rsmi_ret_ok(ret, device, 'get_memory_total_' + str(memType), quiet):
if rsmi_ret_ok(ret, device, 'get_memory_total_' + str(memType), silent):
memTotal = memoryTot.value
return (memUsed, memTotal)
@@ -334,14 +346,16 @@ def getProcessName(pid):
return pName
def getPerfLevel(device):
def getPerfLevel(device, silent=False):
""" Return the current performance level of a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
perf = rsmi_dev_perf_level_t()
ret = rocmsmi.rsmi_dev_perf_level_get(device, byref(perf))
if rsmi_ret_ok(ret, device, 'get_perf_level'):
if rsmi_ret_ok(ret, device, 'get_perf_level', silent):
return perf_level_string(perf.value)
return 'N/A'
@@ -369,42 +383,48 @@ def getPidList():
return
def getPower(device):
def getPower(device, silent=False):
""" Return the current power level of a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
power = c_uint32()
ret = rocmsmi.rsmi_dev_power_ave_get(device, 0, byref(power))
if rsmi_ret_ok(ret, device, 'get_power_avg'):
if rsmi_ret_ok(ret, device, 'get_power_avg', silent):
return power.value / 1000000
return 'N/A'
def getRasEnablement(device, block):
def getRasEnablement(device, block, silent=True):
""" Return RAS enablement state for a given device
@param device: DRM device identifier
@param block: RAS block identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is on.
"""
state = rsmi_ras_err_state_t()
ret = rocmsmi.rsmi_dev_ecc_status_get(device, rsmi_gpu_block_d[block], byref(state))
if rsmi_ret_ok(ret, device, 'get_ecc_status_' + str(block), True):
if rsmi_ret_ok(ret, device, 'get_ecc_status_' + str(block), silent):
return rsmi_ras_err_stale_machine[state.value].upper()
return 'N/A'
def getTemp(device, sensor):
def getTemp(device, sensor, silent=True):
""" Display the current temperature from a given device's sensor
@param device: DRM device identifier
@param sensor: Temperature sensor identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is on.
"""
temp = c_int64(0)
metric = rsmi_temperature_metric_t.RSMI_TEMP_CURRENT
ret = rocmsmi.rsmi_dev_temp_metric_get(c_uint32(device), temp_type_lst.index(sensor), metric, byref(temp))
if rsmi_ret_ok(ret, device, 'get_temp_metric' + str(sensor), True):
if rsmi_ret_ok(ret, device, 'get_temp_metric' + str(sensor), silent):
return temp.value / 1000
return 'N/A'
@@ -428,52 +448,60 @@ def findFirstAvailableTemp(device):
continue
return (ret_temp_type, ret_temp)
def getVbiosVersion(device):
def getVbiosVersion(device, silent=False):
""" Returns the VBIOS version for a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
vbios = create_string_buffer(256)
ret = rocmsmi.rsmi_dev_vbios_version_get(device, vbios, 256)
if ret == rsmi_status_t.RSMI_STATUS_NOT_SUPPORTED:
return "Unsupported"
elif rsmi_ret_ok(ret, device):
elif rsmi_ret_ok(ret, device, silent=silent):
return vbios.value.decode()
def getVersion(deviceList, component):
def getVersion(deviceList, component, silent=False):
""" Return the software version for the specified component
@param deviceList: List of DRM devices (can be a single-item list)
@param component: Component (currently only driver)
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
ver_str = create_string_buffer(256)
ret = rocmsmi.rsmi_version_str_get(component, ver_str, 256)
if rsmi_ret_ok(ret, None, 'get_version_str_' + str(component)):
if rsmi_ret_ok(ret, None, 'get_version_str_' + str(component), silent):
return ver_str.value.decode()
return None
def getComputePartition(device):
def getComputePartition(device, silent=True):
""" Return the current compute partition of a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is on.
"""
currentComputePartition = create_string_buffer(256)
ret = rocmsmi.rsmi_dev_compute_partition_get(device, currentComputePartition, 256)
if rsmi_ret_ok(ret, device, 'get_compute_partition', silent=True) and currentComputePartition.value.decode():
if rsmi_ret_ok(ret, device, 'get_compute_partition', silent) and currentComputePartition.value.decode():
return str(currentComputePartition.value.decode())
return "N/A"
def getMemoryPartition(device):
def getMemoryPartition(device, silent=True):
""" Return the current memory partition of a given device
@param device: DRM device identifier
@param silent=Turn on to silence error output
(you plan to handle manually). Default is on.
"""
currentNPSMode = create_string_buffer(256)
ret = rocmsmi.rsmi_dev_nps_mode_get(device, currentNPSMode, 256)
if rsmi_ret_ok(ret, device, 'get_NPS_mode', silent=True) and currentNPSMode.value.decode():
if rsmi_ret_ok(ret, device, 'get_NPS_mode', silent) and currentNPSMode.value.decode():
return str(currentNPSMode.value.decode())
return "N/A"
@@ -610,10 +638,21 @@ def printLog(device, metricName, value=None, extraSpace=False, useItalics=False)
lock.acquire()
if useItalics:
logstr = italics + logstr + end
if extraSpace:
print('\n' + logstr + '\n', end='', flush=True)
else:
print(logstr + '\n', end='', flush=True)
try:
if extraSpace:
print('\n', end='')
print(logstr + '\n', end='')
sys.stdout.flush()
# when piped into programs like 'head' - print throws an error.
# silently ignore instead
except(BrokenPipeError, IOError):
# https://docs.python.org/3/library/signal.html#note-on-sigpipe
# Python flushes standard streams on exit; redirect remaining output
# to devnull to avoid another BrokenPipeError at shutdown
devnull = os.open(os.devnull, os.O_WRONLY)
os.dup2(devnull, sys.stdout.fileno())
sys.exit(1) # Python exits with error code 1 on EPIPE
lock.release()
@@ -785,12 +824,10 @@ def resetFans(deviceList):
for device in deviceList:
sensor_ind = c_uint32(0)
ret = rocmsmi.rsmi_dev_fan_reset(device, sensor_ind)
if (ret == rsmi_status_t.RSMI_STATUS_NOT_SUPPORTED) or (ret == rsmi_status_t.RSMI_STATUS_PERMISSION):
if not rsmi_ret_ok(rsmi_status_t.RSMI_STATUS_NOT_SUPPORTED, device, 'reset_fan'):
continue
if rsmi_ret_ok(ret, device, silent=True):
printLog(device, 'Successfully reset fan speed to driver control', None)
else:
if rsmi_ret_ok(ret, device, 'reset_fan'):
printLog(device, 'Successfully reset fan speed to driver control', None)
printLog(device, 'Not supported on the given system', None)
printLogSpacer()
@@ -1311,8 +1348,10 @@ def setFanSpeed(deviceList, fan):
else:
fanLevel = int(str(fan))
ret = rocmsmi.rsmi_dev_fan_speed_set(device, 0, int(fanLevel))
if rsmi_ret_ok(ret, device, 'set_fan_speed'):
if rsmi_ret_ok(ret, device, silent=True):
printLog(device, 'Successfully set fan speed to level %s' % (str(int(fanLevel))), None)
else:
printLog(device, 'Not supported on the given system', None)
printLogSpacer()
@@ -1595,10 +1634,13 @@ def showAllConcise(deviceList):
MAX_ALL_CONCISE_WIDTH = 100
appWidth_temp = appWidth
appWidth = MAX_ALL_CONCISE_WIDTH
silent = True
printLogSpacer(' Concise Info ')
deviceList.sort()
(temp_type, _) = findFirstAvailableTemp(deviceList[0])
temp_type = '(' + temp_type_lst[0] + ')'
if len(deviceList) >= 1:
(temp_type, _) = findFirstAvailableTemp(deviceList[0])
available_temp_type = temp_type.lower()
available_temp_type = available_temp_type.replace('(', '')
available_temp_type = available_temp_type.replace(')', '')
@@ -1620,9 +1662,9 @@ def showAllConcise(deviceList):
values = {}
degree_sign = u'\N{DEGREE SIGN}'
for device in deviceList:
gpu_dev_product_info = getDevProductInfo(device)
gpu_dev_product_info = getDevProductInfo(device, silent)
gpu_dev_product_info_names = list(gpu_dev_product_info[device])
temp_val = str(getTemp(device, available_temp_type))
temp_val = str(getTemp(device, available_temp_type, silent))
if temp_val != 'N/A':
temp_val += degree_sign + 'C'
avgPwr = str(getPower(device))
@@ -1630,26 +1672,25 @@ def showAllConcise(deviceList):
avgPwr += 'W'
else:
avgPwr = 'N/A'
combined_partition = (getMemoryPartition(device) + ", "
+ getComputePartition(device))
concise = True
sclk = showCurrentClocks([device], 'sclk', concise)
mclk = showCurrentClocks([device], 'mclk', concise)
(retCode, fanLevel, fanSpeed) = getFanSpeed(device)
combined_partition = (getMemoryPartition(device, silent) + ", "
+ getComputePartition(device, silent))
sclk = showCurrentClocks([device], 'sclk', concise=silent)
mclk = showCurrentClocks([device], 'mclk', concise=silent)
(retCode, fanLevel, fanSpeed) = getFanSpeed(device, silent)
fan = str(fanSpeed) + '%'
if getPerfLevel(device) != -1:
perf = getPerfLevel(device)
if getPerfLevel(device, silent) != -1:
perf = getPerfLevel(device, silent)
else:
perf = 'Unsupported'
if getMaxPower(device) != -1:
pwrCap = str(getMaxPower(device)) + 'W'
if getMaxPower(device, silent) != -1:
pwrCap = str(getMaxPower(device, silent)) + 'W'
else:
pwrCap = 'Unsupported'
if getGpuUse(device) != -1:
gpu_busy = str(getGpuUse(device)) + '%'
if getGpuUse(device, silent) != -1:
gpu_busy = str(getGpuUse(device, silent)) + '%'
else:
gpu_busy = 'Unsupported'
vram_used, vram_total = getMemInfo(device, 'vram', True)
vram_used, vram_total = getMemInfo(device, 'vram', silent)
mem_use_pct = 0
if vram_used is None:
mem_use_pct='Unsupported'
@@ -1683,7 +1724,7 @@ def showAllConcise(deviceList):
for device in deviceList:
printLog(None, "".join(str(word).ljust(max_widths[col]) for col, word in
zip(range(len(max_widths)), values['card%s' % (str(device))])), None)
gpu_dev_product_info = getDevProductInfo(device)
gpu_dev_product_info = getDevProductInfo(device, silent)
gpu_dev_product_info_names = list(gpu_dev_product_info[device])
if (len(gpu_dev_product_info_names) > 1):
printLog(None, "".join(str(word).ljust(max_widths[col]) for col, word in
@@ -1707,19 +1748,20 @@ def showAllConciseHw(deviceList):
header = ['GPU', 'DID', 'DREV', 'GFX RAS', 'SDMA RAS', 'UMC RAS', 'VBIOS', 'BUS']
head_widths = [len(head) + 2 for head in header]
values = {}
silent = True
for device in deviceList:
gpuid = getId(device)
gpuid = getId(device, silent)
if str(gpuid).startswith('0x'):
gpuid = str(gpuid)[2:]
gpurev = getRev(device)
gpurev = getRev(device, silent)
if str(gpurev).startswith('0x'):
gpurev = str(gpurev)[2:]
gfxRas = getRasEnablement(device, 'GFX')
sdmaRas = getRasEnablement(device, 'SDMA')
umcRas = getRasEnablement(device, 'UMC')
vbios = getVbiosVersion(device)
bus = getBus(device)
gfxRas = getRasEnablement(device, 'GFX', silent)
sdmaRas = getRasEnablement(device, 'SDMA', silent)
umcRas = getRasEnablement(device, 'UMC', silent)
vbios = getVbiosVersion(device, silent)
bus = getBus(device, silent)
values['card%s' % (str(device))] = [device, gpuid, gpurev, gfxRas, sdmaRas, umcRas, vbios, bus]
val_widths = {}
for device in deviceList:
@@ -1760,15 +1802,19 @@ def showClocks(deviceList):
for clk_type in sorted(rsmi_clk_names_dict):
if rocmsmi.rsmi_dev_gpu_clk_freq_get(device, rsmi_clk_names_dict[clk_type], None) == 1:
ret = rocmsmi.rsmi_dev_gpu_clk_freq_get(device, rsmi_clk_names_dict[clk_type], byref(freq))
if rsmi_ret_ok(ret, device, 'get_clk_freq_' + clk_type, True):
printLog(device, 'Supported %s frequencies on GPU%s' % (clk_type, str(device)), None)
for x in range(freq.num_supported):
fr = '{:>.0f}Mhz'.format(freq.frequency[x] / 1000000)
if x == freq.current:
printLog(device, str(x), str(fr) + ' *')
else:
printLog(device, str(x), str(fr))
printLog(device, '', None)
if ret == rsmi_status_t.RSMI_STATUS_UNEXPECTED_DATA:
printLog(device, 'Clock [%s] on device [%s] exists but EMPTY! Likely driver error!' % (clk_type, str(device)))
continue
if not rsmi_ret_ok(ret, device, 'get_clk_freq_' + clk_type, True):
continue
printLog(device, 'Supported %s frequencies on GPU%s' % (clk_type, str(device)), None)
for x in range(freq.num_supported):
fr = '{:>.0f}Mhz'.format(freq.frequency[x] / 1000000)
if x == freq.current:
printLog(device, str(x), str(fr) + ' *')
else:
printLog(device, str(x), str(fr))
printLog(device, '', None)
else:
logging.debug('{} frequency is unsupported on device[{}]'.format(clk_type, device))
printLog(device, '', None)
@@ -1814,8 +1860,8 @@ def showCurrentClocks(deviceList, clk_defined=None, concise=False):
if concise: # in case function is used for concise output, no need to print.
return '{:.0f}Mhz'.format(fr)
printLog(device, '{} clock level'.format(clk_defined), '{} ({:.0f}Mhz)'.format(levl, fr))
else:
printErrLog(device, '%s clock is unsupported' % (clk_defined))
elif not concise:
logging.debug('{} clock is unsupported on device[{}]'.format(clk_defined, device))
else: # if clk is not defined, will display all current clk
for clk_type in sorted(rsmi_clk_names_dict):
@@ -1832,7 +1878,7 @@ def showCurrentClocks(deviceList, clk_defined=None, concise=False):
printLog(device, '%s clock level:' % (clk_type), levl)
else:
printLog(device, '%s clock level: %s' % (clk_type, levl), '(%sMhz)' % (str(fr)[:-2]))
else:
elif not concise:
logging.debug('{} clock is unsupported on device[{}]'.format(clk_type, device))
# pcie clocks
if rocmsmi.rsmi_dev_pci_bandwidth_get(device, None) == 1:
@@ -1845,9 +1891,10 @@ def showCurrentClocks(deviceList, clk_defined=None, concise=False):
fr = '{:.1f}GT/s x{}'.format(bw.transfer_rate.frequency[current_f] / 1000000000,
bw.lanes[current_f])
printLog(device, 'pcie clock level', '{} ({})'.format(current_f, fr))
else:
logging.debug('PCIe clock is unsupported on device[{}]'.format(device))
printLogSpacer()
elif not concise:
logging.debug('{} clock is unsupported on device[{}]'.format('PCIe', device))
if not concise:
printLogSpacer()
def showCurrentFans(deviceList):
@@ -2113,6 +2160,7 @@ def showMemUse(deviceList):
@param deviceList: List of DRM devices (can be a single-item list)
"""
memoryUse = c_uint64()
avgMemBandwidth = c_uint16()
printLogSpacer(' Current Memory Use ')
for device in deviceList:
ret = rocmsmi.rsmi_dev_memory_busy_percent_get(device, byref(memoryUse))
@@ -2124,6 +2172,12 @@ def showMemUse(deviceList):
printLog(device, utilization_counter_name[ut_counter.type], ut_counter.val)
else:
printLog(device, 'Memory Activity', 'N/A')
ret = rocmsmi.rsmi_dev_activity_avg_mm_get(device, byref(avgMemBandwidth))
if rsmi_ret_ok(ret, device, silent=True):
printLog(device, 'Avg. Memory Bandwidth', avgMemBandwidth.value)
else:
printLog(device, 'Not supported on the given system', None)
printLogSpacer()
@@ -2404,47 +2458,51 @@ def showProductName(deviceList):
printLogSpacer()
def getDevProductInfo(device):
def getDevProductInfo(device, silent=False):
""" Show the requested product name for the device requested
@param device: Device we want to get the info for
@param silent=Turn on to silence error output
(you plan to handle manually). Default is off.
"""
# Retrieve card vendor
MAX_BUFF_SIZE = 256
MAX_DESC_SIZE = 20
device_info = "N/A"
device_series = "N/A"
device_model = "N/A"
gpu_revision = "N/A"
device_list = {}
vendor = create_string_buffer(MAX_BUFF_SIZE)
ret = rocmsmi.rsmi_dev_vendor_name_get(device, vendor, MAX_BUFF_SIZE)
# Only continue if GPU vendor is AMD
if rsmi_ret_ok(ret, device, 'get_vendor_name') and isAmdDevice(device):
if rsmi_ret_ok(ret, device, 'get_vendor_name', silent) and isAmdDevice(device):
# Retrieve the device series
series = create_string_buffer(MAX_BUFF_SIZE)
ret = rocmsmi.rsmi_dev_name_get(device, series, MAX_BUFF_SIZE)
if rsmi_ret_ok(ret, device, 'get_name'):
if rsmi_ret_ok(ret, device, 'get_name', silent):
try:
device_series = series.value.decode()
except UnicodeDecodeError:
device_series = "N/A"
printErrLog(device, "Unable to read card series")
if not silent:
printErrLog(device, "Unable to read card series")
# Retrieve the device model
model = create_string_buffer(MAX_BUFF_SIZE)
ret = rocmsmi.rsmi_dev_subsystem_name_get(device, model, MAX_BUFF_SIZE)
if rsmi_ret_ok(ret, device, 'get_subsystem_name'):
if rsmi_ret_ok(ret, device, 'get_subsystem_name', silent):
try:
device_model = model.value.decode()
device_model = padHexValue(device_model, 4)
except UnicodeDecodeError:
device_model = "N/A"
printErrLog(device, "Unable to read device model")
if not silent:
printErrLog(device, "Unable to read device model")
try:
gpu_revision = padHexValue(getRev(device), 2)
except Exception as exc:
gpu_revision = "N/A"
printErrLog(device, "Unable to read card revision %s" % (exc))
if not silent:
printErrLog(device, "Unable to read card revision %s" % (exc))
device_series_str = str(device_series[:MAX_DESC_SIZE])
device_series_str = device_series_str.ljust(MAX_DESC_SIZE, ' ')
@@ -2790,7 +2848,9 @@ def getGraphColor(percentage):
def showTempGraph(deviceList):
deviceList.sort()
(temp_type, temp_value) = findFirstAvailableTemp(deviceList[0])
temp_type = '(' + temp_type_lst[0] + ')'
if len(deviceList) >= 1:
(temp_type, _) = findFirstAvailableTemp(deviceList[0])
printLogSpacer(' Temperature Graph ' + temp_type + ' ')
temp_type = temp_type.lower()
temp_type = temp_type.replace('(', '')
@@ -3381,7 +3441,7 @@ def rsmi_ret_ok(my_ret, device=None, metric=None, silent=False):
@param my_ret: Return of RSMI call (rocm_smi_lib API)
@param metric: Parameter of GPU currently being analyzed
@param silent: Echo verbose error reponse.
True siliences err output, False does not silience err output (default).
True silences err output, False does not silence err output (default).
"""
global RETCODE
global PRINT_JSON
@@ -3398,8 +3458,8 @@ def rsmi_ret_ok(my_ret, device=None, metric=None, silent=False):
if err_str.value is not None:
returnString += '%s\t' % (err_str.value.decode())
if not PRINT_JSON:
logging.debug('%s', returnString)
if not silent:
logging.debug('%s', returnString)
if my_ret in rsmi_status_verbose_err_out:
printLog(device, metric + ", " + rsmi_status_verbose_err_out[my_ret], None)
RETCODE = my_ret
@@ -3465,8 +3525,7 @@ def save(deviceList, savefilepath):
# The code below is for when this script is run as an executable instead of when imported as a module
if __name__ == '__main__':
parser = argparse.ArgumentParser(
description='AMD ROCm System Management Interface | ROCM-SMI version: %s | Kernel version: %s' % (
__version__, getVersion(None, rsmi_sw_component_t.RSMI_SW_COMP_DRIVER)),
description=f'AMD ROCm System Management Interface | ROCM-SMI version: {__version__}',
formatter_class=lambda prog: argparse.HelpFormatter(prog, max_help_position=90, width=120))
groupDev = parser.add_argument_group()
groupDisplayOpt = parser.add_argument_group('Display Options')
@@ -3626,6 +3685,11 @@ if __name__ == '__main__':
args = parser.parse_args()
# Must set PRINT_JSON early so the prints can be silenced
if args.json or args.csv:
PRINT_JSON = True
# Initialize rsmiBindings
rocmsmi = initRsmiBindings(silent=PRINT_JSON)
# Initialize the rocm SMI library
initializeRsmi()
@@ -3661,8 +3725,7 @@ if __name__ == '__main__':
sys.exit(1)
# If we want JSON/CSV output, initialize the keys (devices)
if args.json or args.csv:
PRINT_JSON = True
if PRINT_JSON:
for device in deviceList:
JSON_DATA['card' + str(device)] = {}
rocm_smi/python_smi_tools/rsmiBindings.py
+33 -26
View File
@@ -1,5 +1,6 @@
#!/usr/bin/env python3
"""ROCm_SMI_LIB CLI Tool Python Bindings"""
# NOTE: You MUST call rsmiBindings.initRsmiBindings() when using this library!
# TODO: Get most (or all) of these from rocm_smi.h to avoid mismatches and redundancy
from __future__ import print_function
@@ -14,36 +15,42 @@ import os
# relative path changed accordingly.
# if ROCM_SMI_LIB_PATH is set, we can load 'librocm_smi64.so' from that location
#
# Library load is wrapped in a function so prints can be hidden for PRINT_JSON mode.
path_librocm = str()
rocm_smi_lib_path = os.getenv('ROCM_SMI_LIB_PATH')
if (rocm_smi_lib_path != None):
path_librocm = rocm_smi_lib_path
else:
path_librocm = os.path.dirname(os.path.realpath(__file__)) + '/../../@CMAKE_INSTALL_LIBDIR@/librocm_smi64.so.@VERSION_MAJOR@'
def initRsmiBindings(silent=False):
def print_silent(*args):
if not silent:
print(args)
if not os.path.isfile(path_librocm):
print('Unable to find %s . Trying /opt/rocm*' % path_librocm)
for root, dirs, files in os.walk('/opt', followlinks=True):
if 'librocm_smi64.so.@VERSION_MAJOR@' in files:
path_librocm = os.path.join(os.path.realpath(root), 'librocm_smi64.so.@VERSION_MAJOR@')
if os.path.isfile(path_librocm):
print('Using lib from %s' % path_librocm)
rocm_smi_lib_path = os.getenv('ROCM_SMI_LIB_PATH')
if (rocm_smi_lib_path != None):
path_librocm = rocm_smi_lib_path
else:
print('Unable to find librocm_smi64.so.@VERSION_MAJOR@')
else:
print('Library loaded from: %s ' % path_librocm)
path_librocm = os.path.dirname(os.path.realpath(__file__)) + '/../../@CMAKE_INSTALL_LIBDIR@/librocm_smi64.so.@VERSION_MAJOR@'
# ----------> TODO: Support static libs as well as SO
try:
cdll.LoadLibrary(path_librocm)
rocmsmi = CDLL(path_librocm)
except OSError:
print('Unable to load the rocm_smi library.\n'\
'Set LD_LIBRARY_PATH to the folder containing librocm_smi64.so.@VERSION_MAJOR@\n'\
'{0}Please refer to https://github.com/'\
'RadeonOpenCompute/rocm_smi_lib for the installation guide.{1}'\
.format('\33[33m', '\033[0m'))
exit()
if not os.path.isfile(path_librocm):
print_silent('Unable to find %s . Trying /opt/rocm*' % path_librocm)
for root, dirs, files in os.walk('/opt', followlinks=True):
if 'librocm_smi64.so.@VERSION_MAJOR@' in files:
path_librocm = os.path.join(os.path.realpath(root), 'librocm_smi64.so.@VERSION_MAJOR@')
if os.path.isfile(path_librocm):
print_silent('Using lib from %s' % path_librocm)
else:
print('Unable to find librocm_smi64.so.@VERSION_MAJOR@')
else:
print_silent('Library loaded from: %s ' % path_librocm)
# ----------> TODO: Support static libs as well as SO
try:
cdll.LoadLibrary(path_librocm)
return CDLL(path_librocm)
except OSError:
print('Unable to load the rocm_smi library.\n'\
'Set LD_LIBRARY_PATH to the folder containing librocm_smi64.so.@VERSION_MAJOR@\n'\
'{0}Please refer to https://github.com/'\
'RadeonOpenCompute/rocm_smi_lib for the installation guide.{1}'\
.format('\33[33m', '\033[0m'))
exit()
# Device ID
dv_id = c_uint64()