| version 1.3 | version 1.4 |
|---|
| |
| #ifndef DEVICECUDA_H | #ifndef DEVICECUDA_H |
| #define DEVICECUDA_H | #define DEVICECUDA_H |
| | |
| | #ifdef NAMD_CUDA |
| | #include <cuda_runtime.h> |
| | |
| #define CUDA_EVENT_ID_POLL_REMOTE 98 | #define CUDA_EVENT_ID_POLL_REMOTE 98 |
| #define CUDA_TRACE_POLL_REMOTE \ | #define CUDA_TRACE_POLL_REMOTE \ |
| traceUserEvent(CUDA_EVENT_ID_POLL_REMOTE) | traceUserEvent(CUDA_EVENT_ID_POLL_REMOTE) |
| |
| do { int dev; cudaGetDevice(&dev); traceUserBracketEvent( \ | do { int dev; cudaGetDevice(&dev); traceUserBracketEvent( \ |
| CUDA_EVENT_ID_BASE + 2 * dev + 1, START, END); } while (0) | CUDA_EVENT_ID_BASE + 2 * dev + 1, START, END); } while (0) |
| | |
| #ifdef WIN32 | |
| #define __thread __declspec(thread) | |
| #endif | |
| | |
| // | // |
| // Class that handles PE <=> CUDA device mapping | // Class that handles PE <=> CUDA device mapping |
| // | // |
| |
| int nomergegrids; | int nomergegrids; |
| int nostreaming; | int nostreaming; |
| | |
| | // Number of devices on this node |
| | int deviceCount; |
| | |
| | // Number of devices on this node that are used for computation |
| | int ndevices; |
| | |
| | // List of device IDs that are used for computation |
| | int *devices; |
| | |
| // True when GPU is shared between PEs | // True when GPU is shared between PEs |
| bool sharedGpu; | bool sharedGpu; |
| // Index of next GPU sharing this GPU | // Index of next GPU sharing this GPU |
| |
| int *pesSharingDevice; | int *pesSharingDevice; |
| // True when what??? | // True when what??? |
| int gpuIsMine; | int gpuIsMine; |
| | // Value of __CUDA_ARCH__ |
| | int cuda_arch; |
| | |
| | // Device ID for this Pe |
| | int deviceID; |
| | |
| | // Device properties for all devices on this node |
| | cudaDeviceProp* deviceProps; |
| | |
| void register_user_events(); | void register_user_events(); |
| | |
| |
| | |
| void initialize(); | void initialize(); |
| | |
| | int getDeviceCount() {return deviceCount;} |
| | int getNumDevice() {return ndevices;} |
| | |
| bool device_shared_with_pe(int pe); | bool device_shared_with_pe(int pe); |
| bool one_device_per_node(); | bool one_device_per_node(); |
| | |
| |
| | |
| int getGpuIsMine() {return gpuIsMine;} | int getGpuIsMine() {return gpuIsMine;} |
| void setGpuIsMine(const int val) {gpuIsMine = val;} | void setGpuIsMine(const int val) {gpuIsMine = val;} |
| | |
| | int get_cuda_arch() {return cuda_arch;} |
| | |
| | int getDeviceID() {return deviceID;} |
| | int getDeviceIDbyRank(int rank) {return devices[rank];} |
| | int getDeviceIDforPe(int pe); |
| | int getMasterPeForDeviceID(int deviceID); |
| | |
| | int getMaxNumThreads(); |
| | int getMaxNumBlocks(); |
| }; | }; |
| | #endif //NAMD_CUDA |
| | |
| #endif // DEVICECUDA_H | #endif // DEVICECUDA_H |