AI/rocm-systems
2
0
Форкнуть 0
Код Задачи Запросы на слияние Действия Пакеты Проекты Релизы Вики Активность

Merge amd-staging into amd-master

Просмотр исходного кода 
Signed-off-by: Hao Zhou <Hao.Zhou@amd.com>
Change-Id: Ic324e60cd33d0db539537a978710d9c87c1dbd2e
Этот коммит содержится в:
Hao Zhou
2021-12-09 10:04:29 +08:00
родитель 98baeca615 1aeb27c4c9
Коммит 3ef213258b
5 изменённых файлов: 94 добавлений и 30 удалений
Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы
python_smi_tools/rocm_smi.py
+52 -7
Просмотреть файл
@@ -483,7 +483,7 @@ def printLog(device, metricName, value):
logstr = logstr[13:]
logging.debug(logstr)
# Force thread safe printing
print(logstr + '\n\r', end='')
print(logstr + '\n', end='')
def printListLog(metricName, valuesList):
@@ -1623,7 +1623,7 @@ def getCoarseGrainUtil(device, typeName=None):
utilization_counters[i].type = c_int(i)
ret = rocmsmi.rsmi_utilization_count_get(device, utilization_counters, length, byref(timestamp))
if rsmi_ret_ok(ret, device):
if rsmi_ret_ok(ret, device, typeName, True):
return utilization_counters
return -1
@@ -1643,6 +1643,9 @@ def showGpuUse(deviceList):
if util_counters != -1:
for ut_counter in util_counters:
printLog(device, utilization_counter_name[ut_counter.type], ut_counter.val)
else:
printLog(device, 'GFX Activity', 'N/A')
printLogSpacer()
@@ -1732,6 +1735,8 @@ def showMemUse(deviceList):
if util_counters != -1:
for ut_counter in util_counters:
printLog(device, utilization_counter_name[ut_counter.type], ut_counter.val)
else:
printLog(device, 'Memory Activity', 'N/A')
printLogSpacer()
@@ -2314,7 +2319,7 @@ def showAccessibleTopology(deviceList):
if rsmi_ret_ok(ret):
gpu_links_type[srcdevice][destdevice] = accessible.value
else:
printErrLog(srcdevice, 'Cannot read link accessibility: Unsupported on this machine', None)
printErrLog(srcdevice, 'Cannot read link accessibility: Unsupported on this machine')
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_type, "(Topology) Link accessibility between DRM devices {} and {}")
return
@@ -2353,7 +2358,7 @@ def showWeightTopology(deviceList):
if rsmi_ret_ok(ret):
gpu_links_weight[srcdevice][destdevice] = weight
else:
printErrLog(srcdevice, 'Cannot read Link Weight: Not supported on this machine', None)
printErrLog(srcdevice, 'Cannot read Link Weight: Not supported on this machine')
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_weight, "(Topology) Weight between DRM devices {} and {}")
@@ -2396,7 +2401,7 @@ def showHopsTopology(deviceList):
if rsmi_ret_ok(ret):
gpu_links_hops[srcdevice][destdevice] = hops
else:
printErrLog(srcdevice, 'Cannot read Link Hops: Not supported on this machine', None)
printErrLog(srcdevice, 'Cannot read Link Hops: Not supported on this machine')
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_hops, "(Topology) Hops between DRM devices {} and {}")
@@ -2444,7 +2449,7 @@ def showTypeTopology(deviceList):
else:
gpu_links_type[srcdevice][destdevice] = "XXXX"
else:
printErrLog(srcdevice, 'Cannot read Link Type: Not supported on this machine', None)
printErrLog(srcdevice, 'Cannot read Link Type: Not supported on this machine')
if PRINT_JSON:
formatMatrixToJSON(deviceList, gpu_links_type, "(Topology) Link type between DRM devices {} and {}")
return
@@ -2485,7 +2490,7 @@ def showNumaTopology(deviceList):
if rsmi_ret_ok(ret):
printLog(device, "(Topology) Numa Affinity", numa_numbers.value)
else:
printErrLog(device, 'Cannot read Numa Affinity', None)
printErrLog(device, 'Cannot read Numa Affinity')
def showHwTopology(deviceList):
@@ -2504,6 +2509,43 @@ def showHwTopology(deviceList):
showNumaTopology(deviceList)
def showNodesBw(deviceList):
""" Display max and min bandwidth between nodes.
Currently supports XGMI only.
This reads the HW Topology file and displays the matrix for the nodes
@param deviceList: List of DRM devices (can be a single-item list)
"""
devices_ind = range(len(deviceList))
minBW = c_uint32()
maxBW = c_uint32()
gpu_links_type = [[0 for x in devices_ind] for y in devices_ind]
printLogSpacer(' Bandwidth ')
for srcdevice in deviceList:
for destdevice in deviceList:
if srcdevice != destdevice:
ret = rocmsmi.rsmi_minmax_bandwidth_get(srcdevice, destdevice, byref(minBW), byref(maxBW))
if rsmi_ret_ok(ret, " {} to {}".format(srcdevice, destdevice),None ):
gpu_links_type[srcdevice][destdevice] = "{}-{}".format(minBW.value, maxBW.value)
else:
gpu_links_type[srcdevice][destdevice] = "N/A"
if PRINT_JSON:
formatMatrixToJSON(deviceList, "{}-{}".format(minBW.value, maxBW.value), " min-max bandwidth between DRM devices {} and {}".format(srcdevice, destdevice))
return
printTableRow(None, ' ')
for row in deviceList:
tmp = 'GPU%d' % row
printTableRow('%-12s', tmp)
printEmptyLine()
for gpu1 in deviceList:
tmp = 'GPU%d' % gpu1
printTableRow('%-6s', tmp)
for gpu2 in deviceList:
printTableRow('%-12s', gpu_links_type[gpu1][gpu2])
printEmptyLine()
printLog(None,"Format: min-max; Units: mps", None)
printLog(None,'"0-0" min-max bandwidth indicates devices are not connected dirrectly', None)
def checkAmdGpus(deviceList):
""" Check if there are any AMD GPUs being queried,
return False if there are none
@@ -2847,6 +2889,7 @@ if __name__ == '__main__':
groupDisplay.add_argument('--showtoponuma', help='Shows the numa nodes ', action='store_true')
groupDisplay.add_argument('--showenergycounter', help='Energy accumulator that stores amount of energy consumed',
action='store_true')
groupDisplay.add_argument('--shownodesbw', help='Shows the numa nodes ', action='store_true')
groupActionReset.add_argument('-r', '--resetclocks', help='Reset clocks and OverDrive to default',
action='store_true')
@@ -3084,6 +3127,8 @@ if __name__ == '__main__':
showProductName(deviceList)
if args.showxgmierr:
showXgmiErr(deviceList)
if args.shownodesbw:
showNodesBw(deviceList)
if args.showtopo:
showHwTopology(deviceList)
if args.showtopoaccess:
src/rocm_smi.cc
+32 -20
Просмотреть файл
@@ -2081,6 +2081,7 @@ rsmi_dev_temp_metric_get(uint32_t dv_ind, uint32_t sensor_type,
rsmi_status_t ret;
amd::smi::MonitorTypes mon_type;
uint16_t val_ui16;
switch (metric) {
case RSMI_TEMP_CURRENT:
@@ -2129,6 +2130,10 @@ rsmi_dev_temp_metric_get(uint32_t dv_ind, uint32_t sensor_type,
mon_type = amd::smi::kMonInvalid;
}
if (temperature == nullptr) {
return RSMI_STATUS_INVALID_ARGS;
}
// The HBM temperature is retreived from the gpu_metrics
if (sensor_type == RSMI_TEMP_TYPE_HBM_0
|| sensor_type == RSMI_TEMP_TYPE_HBM_1
@@ -2144,25 +2149,26 @@ rsmi_dev_temp_metric_get(uint32_t dv_ind, uint32_t sensor_type,
return ret;
}
if (temperature == nullptr) {
return RSMI_STATUS_INVALID_ARGS;
}
if (sensor_type == RSMI_TEMP_TYPE_HBM_0) {
*temperature = gpu_metrics.temperature_hbm[0] *
CENTRIGRADE_TO_MILLI_CENTIGRADE;
} else if (sensor_type == RSMI_TEMP_TYPE_HBM_1) {
*temperature = gpu_metrics.temperature_hbm[1] *
CENTRIGRADE_TO_MILLI_CENTIGRADE;
} else if (sensor_type == RSMI_TEMP_TYPE_HBM_2) {
*temperature = gpu_metrics.temperature_hbm[2] *
CENTRIGRADE_TO_MILLI_CENTIGRADE;
} else if (sensor_type == RSMI_TEMP_TYPE_HBM_3) {
*temperature = gpu_metrics.temperature_hbm[3] *
CENTRIGRADE_TO_MILLI_CENTIGRADE;
} else {
return RSMI_STATUS_NOT_SUPPORTED;
switch (sensor_type) {
case RSMI_TEMP_TYPE_HBM_0:
val_ui16 = gpu_metrics.temperature_hbm[0];
break;
case RSMI_TEMP_TYPE_HBM_1:
val_ui16 = gpu_metrics.temperature_hbm[1];
break;
case RSMI_TEMP_TYPE_HBM_2:
val_ui16 = gpu_metrics.temperature_hbm[2];
break;
case RSMI_TEMP_TYPE_HBM_3:
val_ui16 = gpu_metrics.temperature_hbm[3];
break;
default:
return RSMI_STATUS_INVALID_ARGS;
}
if (val_ui16 == UINT16_MAX)
return RSMI_STATUS_NOT_SUPPORTED;
else
*temperature = val_ui16 * CENTRIGRADE_TO_MILLI_CENTIGRADE;
return RSMI_STATUS_SUCCESS;
} // end HBM temperature
@@ -2797,6 +2803,8 @@ rsmi_utilization_count_get(uint32_t dv_ind,
rsmi_status_t ret;
rsmi_gpu_metrics_t gpu_metrics;
uint32_t val_ui32;
ret = rsmi_dev_gpu_metrics_info_get(dv_ind, &gpu_metrics);
if (ret != RSMI_STATUS_SUCCESS) {
return ret;
@@ -2810,14 +2818,18 @@ rsmi_utilization_count_get(uint32_t dv_ind,
for (uint32_t index = 0 ; index < count; index++) {
switch (utilization_counters[index].type) {
case RSMI_COARSE_GRAIN_GFX_ACTIVITY:
utilization_counters[index].value = gpu_metrics.gfx_activity_acc;
val_ui32 = gpu_metrics.gfx_activity_acc;
break;
case RSMI_COARSE_GRAIN_MEM_ACTIVITY:
utilization_counters[index].value = gpu_metrics.mem_actvity_acc;
val_ui32 = gpu_metrics.mem_actvity_acc;
break;
default:
return RSMI_STATUS_INVALID_ARGS;
}
if (val_ui32 == UINT32_MAX)
return RSMI_STATUS_NOT_SUPPORTED;
else
utilization_counters[index].value = val_ui32;
}
*timestamp = gpu_metrics.system_clock_counter;
tests/rocm_smi_test/functional/frequencies_read_write.cc
+8 -1
Просмотреть файл
@@ -141,8 +141,15 @@ void TestFrequenciesReadWrite::Run(void) {
std::endl;
}
ret = rsmi_dev_gpu_clk_freq_set(dv_ind, rsmi_clk, freq_bitmask);
//Certain ASICs does not allow to set particular clocks. If set function for a clock returns
//permission error despite root access, manually set ret value to success and return
if (ret == RSMI_STATUS_PERMISSION && geteuid() == 0) {
std::cout << "\t**Set " << FreqEnumToStr(rsmi_clk) <<
": Not supported on this machine. Skipping..." << std::endl;
ret = RSMI_STATUS_SUCCESS;
return;
}
CHK_ERR_ASRT(ret)
ret = rsmi_dev_gpu_clk_freq_get(dv_ind, rsmi_clk, &f);
if (ret != RSMI_STATUS_SUCCESS) {
return;
tests/rocm_smi_test/functional/hw_topology_read.cc
+1 -1
Просмотреть файл
@@ -136,7 +136,7 @@ void TestHWTopologyRead::Run(void) {
gpu_links[dv_ind_src][dv_ind_dst].type = "X";
gpu_links[dv_ind_src][dv_ind_dst].hops = 0;
gpu_links[dv_ind_src][dv_ind_dst].weight = 0;
gpu_links[dv_ind_src][dv_ind_dst].accessible = false;
gpu_links[dv_ind_src][dv_ind_dst].accessible = true;
} else {
RSMI_IO_LINK_TYPE type;
err = rsmi_topo_get_link_type(dv_ind_src, dv_ind_dst,
tests/rocm_smi_test/functional/temp_read.cc
+1 -1
Просмотреть файл
@@ -124,7 +124,7 @@ void TestTempRead::Run(void) {
// Verify api support checking functionality is working
err = rsmi_dev_temp_metric_get(i, type, met, nullptr);
ASSERT_EQ(err, RSMI_STATUS_NOT_SUPPORTED);
ASSERT_EQ(err, RSMI_STATUS_INVALID_ARGS);
return;
} else {
CHK_ERR_ASRT(err)