tuxclocker/nvidia.h

89 lines
2.2 KiB
C++

#ifndef NVIDIA_H
#define NVIDIA_H
#include <QObject>
#include <QDebug>
#include <QtX11Extras/QX11Info>
#include <QProcess>
#include "/opt/cuda/include/nvml.h"
class nvidia : public QObject
{
Q_OBJECT
public:
explicit nvidia(QObject *parent = nullptr);
struct GPU
{
bool status;
int index;
char *name;
char *uuid;
char *utils;
bool overVoltAvailable = false;
bool overClockAvailable = false;
bool memOverClockAvailable = false;
bool voltageReadable = false;
bool coreClkReadable = false;
bool memClkReadable = false;
bool manualFanCtrl = false;
int maxVoltageOffset;
int minVoltageOffset;
int minCoreClkOffset;
int maxCoreClkOffset;
int minMemClkOffset;
int maxMemClkOffset;
int coreFreq;
int memFreq;
int temp;
int voltage;
int fanSpeed;
uint powerDraw;
uint coreUtil;
uint memUtil;
uint minPowerLim;
uint maxPowerLim;
uint powerLim;
int totalVRAM;
int usedVRAM;
};
QVector <GPU> GPUList;
int gpuCount = 0;
private:
Display *dpy;
nvmlDevice_t *device;
signals:
public slots:
bool setupXNVCtrl();
bool setupNVML(int GPUIndex);
void queryGPUCount();
void queryGPUNames();
void queryGPUUIDs();
void queryGPUFeatures();
void queryGPUVoltage(int GPUIndex);
void queryGPUTemp(int GPUIndex);
void queryGPUFrequencies(int GPUIndex);
void queryGPUFanSpeed(int GPUIndex);
void queryGPUUsedVRAM(int GPUIndex);
void queryGPUUtils(int GPUIndex);
void queryGPUPowerDraw(int GPUIndex);
void queryGPUPowerLimit(int GPUIndex);
void queryGPUPowerLimitLimits(int GPUIndex);
bool assignGPUFanSpeed(int GPUIndex, int targetValue);
bool assignGPUFanCtlMode(int GPUIndex, int targetValue);
bool assignGPUFreqOffset(int GPUIndex, int targetValue);
bool assignGPUMemClockOffset(int GPUIndex, int targetValue);
bool assignGPUVoltageOffset(int GPUIndex, int targetValue);
// NVML functions know the GPU index already based on the dev object passed in setupNVML()
bool assignGPUPowerLimit(uint targetValue);
private slots:
};
#endif // NVIDIA_H