mirror of
https://github.com/intel/compute-runtime.git
synced 2026-01-09 06:23:01 +08:00
feature(sysman): supports product helper for temperature module
Related-To: NEO-8720 Signed-off-by: Kulkarni, Ashwin Kumar <ashwin.kumar.kulkarni@intel.com>
This commit is contained in:
committed by
Compute-Runtime-Automation
parent
985186242e
commit
d7aea3e745
@@ -9,17 +9,12 @@
|
||||
|
||||
#include "shared/source/debug_settings/debug_settings_manager.h"
|
||||
|
||||
#include "level_zero/sysman/source/shared/linux/product_helper/sysman_product_helper.h"
|
||||
#include "level_zero/sysman/source/shared/linux/zes_os_sysman_imp.h"
|
||||
|
||||
namespace L0 {
|
||||
namespace Sysman {
|
||||
|
||||
constexpr uint32_t numSocTemperatureEntries = 7; // entries would be PCH or GT_TEMP, DRAM, SA, PSF, DE, PCIE, TYPEC
|
||||
constexpr uint32_t numCoreTemperatureEntries = 4; // entries would be CORE0, CORE1, CORE2, CORE3
|
||||
constexpr uint32_t numComputeTemperatureEntries = 3; // entries would be IA, GT and LLC
|
||||
constexpr uint32_t invalidMaxTemperature = 125;
|
||||
constexpr uint32_t invalidMinTemperature = 10;
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getProperties(zes_temp_properties_t *pProperties) {
|
||||
pProperties->type = type;
|
||||
pProperties->onSubdevice = 0;
|
||||
@@ -31,157 +26,16 @@ ze_result_t LinuxTemperatureImp::getProperties(zes_temp_properties_t *pPropertie
|
||||
return ZE_RESULT_SUCCESS;
|
||||
}
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getGlobalMaxTemperatureNoSubDevice(double *pTemperature) {
|
||||
auto isValidTemperature = [](auto temperature) {
|
||||
if ((temperature > invalidMaxTemperature) || (temperature < invalidMinTemperature)) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): temperature:%f is not in valid limits \n", __FUNCTION__, temperature);
|
||||
return false;
|
||||
}
|
||||
return true;
|
||||
};
|
||||
|
||||
auto getMaxTemperature = [&](auto temperature, auto numTemperatureEntries) {
|
||||
uint32_t maxTemperature = 0;
|
||||
for (uint32_t count = 0; count < numTemperatureEntries; count++) {
|
||||
uint32_t localTemperatureVal = (temperature >> (8 * count)) & 0xff;
|
||||
if (isValidTemperature(localTemperatureVal)) {
|
||||
if (localTemperatureVal > maxTemperature) {
|
||||
maxTemperature = localTemperatureVal;
|
||||
}
|
||||
}
|
||||
}
|
||||
return maxTemperature;
|
||||
};
|
||||
|
||||
ze_result_t result = ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
|
||||
uint32_t maxComputeTemperature = 0;
|
||||
uint32_t maxCoreTemperature = 0;
|
||||
std::string key;
|
||||
if (productFamily == IGFX_DG1) {
|
||||
uint32_t computeTemperature = 0;
|
||||
key = "COMPUTE_TEMPERATURES";
|
||||
result = pPmt->readValue(key, computeTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for COMPUTE_TEMPERATURES is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
// Check max temperature among IA, GT and LLC sensors across COMPUTE_TEMPERATURES
|
||||
maxComputeTemperature = getMaxTemperature(computeTemperature, numComputeTemperatureEntries);
|
||||
|
||||
uint32_t coreTemperature = 0;
|
||||
key = "CORE_TEMPERATURES";
|
||||
result = pPmt->readValue(key, coreTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for CORE_TEMPERATURES is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
// Check max temperature among CORE0, CORE1, CORE2, CORE3 sensors across CORE_TEMPERATURES
|
||||
maxCoreTemperature = getMaxTemperature(coreTemperature, numCoreTemperatureEntries);
|
||||
}
|
||||
|
||||
// SOC_TEMPERATURES is present in all product families
|
||||
uint64_t socTemperature = 0;
|
||||
key = "SOC_TEMPERATURES";
|
||||
result = pPmt->readValue(key, socTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for SOC_TEMPERATURES is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
// Check max temperature among possible sensors like PCH or GT_TEMP, DRAM, SA, PSF, DE, PCIE, TYPEC across SOC_TEMPERATURES
|
||||
uint32_t maxSocTemperature = getMaxTemperature(socTemperature, numSocTemperatureEntries);
|
||||
|
||||
*pTemperature = static_cast<double>(std::max({maxComputeTemperature, maxCoreTemperature, maxSocTemperature}));
|
||||
|
||||
return result;
|
||||
}
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getGlobalMaxTemperature(double *pTemperature) {
|
||||
// For XE_HP_SDV and PVC single tile devices, telemetry info is retrieved from
|
||||
// tile's telem node rather from root device telem node.
|
||||
if ((!isSubdevice) && (!(productFamily == IGFX_PVC))) {
|
||||
return getGlobalMaxTemperatureNoSubDevice(pTemperature);
|
||||
}
|
||||
uint32_t globalMaxTemperature = 0;
|
||||
std::string key("TileMaxTemperature");
|
||||
ze_result_t result = pPmt->readValue(key, globalMaxTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for TileMaxTemperature is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
*pTemperature = static_cast<double>(globalMaxTemperature);
|
||||
return result;
|
||||
}
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getGpuMaxTemperatureNoSubDevice(double *pTemperature) {
|
||||
double gpuMaxTemperature = 0;
|
||||
uint64_t socTemperature = 0;
|
||||
// Gpu temperature is obtained from GT_TEMP in SOC_TEMPERATURE's bit 0 to 7.
|
||||
std::string key = "SOC_TEMPERATURES";
|
||||
auto result = pPmt->readValue(key, socTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for SOC_TEMPERATURES is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
gpuMaxTemperature = static_cast<double>(socTemperature & 0xff);
|
||||
|
||||
if (productFamily == IGFX_DG1) {
|
||||
// In DG1 platform, Gpu Max Temperature is obtained from COMPUTE_TEMPERATURE only
|
||||
uint32_t computeTemperature = 0;
|
||||
std::string key("COMPUTE_TEMPERATURES");
|
||||
ze_result_t result = pPmt->readValue(key, computeTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for COMPUTE_TEMPERATURES is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
|
||||
// GT temperature could be read via 8th to 15th bit in the value read in temperature
|
||||
computeTemperature = (computeTemperature >> 8) & 0xff;
|
||||
gpuMaxTemperature = static_cast<double>(computeTemperature);
|
||||
}
|
||||
*pTemperature = gpuMaxTemperature;
|
||||
return ZE_RESULT_SUCCESS;
|
||||
return pSysmanProductHelper->getGlobalMaxTemperature(pPmt, pTemperature);
|
||||
}
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getGpuMaxTemperature(double *pTemperature) {
|
||||
if ((!isSubdevice) && (!(productFamily == IGFX_PVC))) {
|
||||
return getGpuMaxTemperatureNoSubDevice(pTemperature);
|
||||
}
|
||||
uint32_t gpuMaxTemperature = 0;
|
||||
std::string key("GTMaxTemperature");
|
||||
ze_result_t result = pPmt->readValue(key, gpuMaxTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for GTMaxTemperature is returning error:0x%x \n", __FUNCTION__, result);
|
||||
return result;
|
||||
}
|
||||
*pTemperature = static_cast<double>(gpuMaxTemperature);
|
||||
return result;
|
||||
return pSysmanProductHelper->getGpuMaxTemperature(pPmt, pTemperature);
|
||||
}
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getMemoryMaxTemperature(double *pTemperature) {
|
||||
ze_result_t result = ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
|
||||
uint32_t numHbmModules = 0u;
|
||||
if (productFamily == IGFX_PVC) {
|
||||
numHbmModules = 4u;
|
||||
} else {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s() returning UNSUPPORTED_FEATURE \n", __FUNCTION__);
|
||||
return ZE_RESULT_ERROR_UNSUPPORTED_FEATURE;
|
||||
}
|
||||
|
||||
std::vector<uint32_t> maxDeviceTemperatureList;
|
||||
for (auto hbmModuleIndex = 0u; hbmModuleIndex < numHbmModules; hbmModuleIndex++) {
|
||||
uint32_t maxDeviceTemperature = 0;
|
||||
// To read HBM 0's max device temperature key would be HBM0MaxDeviceTemperature
|
||||
std::string key = "HBM" + std::to_string(hbmModuleIndex) + "MaxDeviceTemperature";
|
||||
result = pPmt->readValue(key, maxDeviceTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
NEO::printDebugString(NEO::debugManager.flags.PrintDebugMessages.get(), stderr, "Error@ %s(): Pmt->readvalue() for %s is returning error:0x%x \n", __FUNCTION__, key.c_str(), result);
|
||||
return result;
|
||||
}
|
||||
maxDeviceTemperatureList.push_back(maxDeviceTemperature);
|
||||
}
|
||||
|
||||
*pTemperature = static_cast<double>(*std::max_element(maxDeviceTemperatureList.begin(), maxDeviceTemperatureList.end()));
|
||||
return result;
|
||||
return pSysmanProductHelper->getMemoryMaxTemperature(pPmt, pTemperature);
|
||||
}
|
||||
|
||||
ze_result_t LinuxTemperatureImp::getSensorTemperature(double *pTemperature) {
|
||||
@@ -189,21 +43,12 @@ ze_result_t LinuxTemperatureImp::getSensorTemperature(double *pTemperature) {
|
||||
switch (type) {
|
||||
case ZES_TEMP_SENSORS_GLOBAL:
|
||||
result = getGlobalMaxTemperature(pTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
return result;
|
||||
}
|
||||
break;
|
||||
case ZES_TEMP_SENSORS_GPU:
|
||||
result = getGpuMaxTemperature(pTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
return result;
|
||||
}
|
||||
break;
|
||||
case ZES_TEMP_SENSORS_MEMORY:
|
||||
result = getMemoryMaxTemperature(pTemperature);
|
||||
if (result != ZE_RESULT_SUCCESS) {
|
||||
return result;
|
||||
}
|
||||
break;
|
||||
default:
|
||||
*pTemperature = 0;
|
||||
@@ -215,13 +60,19 @@ ze_result_t LinuxTemperatureImp::getSensorTemperature(double *pTemperature) {
|
||||
}
|
||||
|
||||
bool LinuxTemperatureImp::isTempModuleSupported() {
|
||||
if ((!isSubdevice) && (!(productFamily == IGFX_PVC))) {
|
||||
if (type == ZES_TEMP_SENSORS_MEMORY) {
|
||||
return false;
|
||||
}
|
||||
bool result = (pPmt != nullptr);
|
||||
switch (type) {
|
||||
case ZES_TEMP_SENSORS_GLOBAL:
|
||||
case ZES_TEMP_SENSORS_GPU:
|
||||
break;
|
||||
case ZES_TEMP_SENSORS_MEMORY:
|
||||
result &= pSysmanProductHelper->isMemoryMaxTemperatureSupported();
|
||||
break;
|
||||
default:
|
||||
result = false;
|
||||
break;
|
||||
}
|
||||
|
||||
return (pPmt != nullptr);
|
||||
return result;
|
||||
}
|
||||
|
||||
void LinuxTemperatureImp::setSensorType(zes_temp_sensors_t sensorType) {
|
||||
@@ -232,7 +83,7 @@ LinuxTemperatureImp::LinuxTemperatureImp(OsSysman *pOsSysman, ze_bool_t onSubdev
|
||||
uint32_t subdeviceId) : subdeviceId(subdeviceId), isSubdevice(onSubdevice) {
|
||||
LinuxSysmanImp *pLinuxSysmanImp = static_cast<LinuxSysmanImp *>(pOsSysman);
|
||||
pPmt = pLinuxSysmanImp->getPlatformMonitoringTechAccess(subdeviceId);
|
||||
productFamily = pLinuxSysmanImp->getProductFamily();
|
||||
pSysmanProductHelper = pLinuxSysmanImp->getSysmanProductHelper();
|
||||
}
|
||||
|
||||
std::unique_ptr<OsTemperature> OsTemperature::create(OsSysman *pOsSysman, ze_bool_t onSubdevice, uint32_t subdeviceId, zes_temp_sensors_t sensorType) {
|
||||
|
||||
@@ -20,6 +20,7 @@ namespace Sysman {
|
||||
|
||||
class SysfsAccess;
|
||||
class PlatformMonitoringTech;
|
||||
class SysmanProductHelper;
|
||||
class LinuxTemperatureImp : public OsTemperature, NEO::NonCopyableOrMovableClass {
|
||||
public:
|
||||
ze_result_t getProperties(zes_temp_properties_t *pProperties) override;
|
||||
@@ -36,15 +37,11 @@ class LinuxTemperatureImp : public OsTemperature, NEO::NonCopyableOrMovableClass
|
||||
|
||||
private:
|
||||
ze_result_t getGlobalMaxTemperature(double *pTemperature);
|
||||
ze_result_t getGlobalMinTemperature(double *pTemperature);
|
||||
ze_result_t getGpuMaxTemperature(double *pTemperature);
|
||||
ze_result_t getGpuMinTemperature(double *pTemperature);
|
||||
ze_result_t getMemoryMaxTemperature(double *pTemperature);
|
||||
ze_result_t getGlobalMaxTemperatureNoSubDevice(double *pTemperature);
|
||||
ze_result_t getGpuMaxTemperatureNoSubDevice(double *pTemperature);
|
||||
uint32_t subdeviceId = 0;
|
||||
ze_bool_t isSubdevice = 0;
|
||||
PRODUCT_FAMILY productFamily = IGFX_UNKNOWN;
|
||||
SysmanProductHelper *pSysmanProductHelper = nullptr;
|
||||
};
|
||||
|
||||
} // namespace Sysman
|
||||
|
||||
Reference in New Issue
Block a user