gpu_details.h

// -*-c++-*-
#ifndef STORMM_HPC_STATUS
#define STORMM_HPC_STATUS
 
#include <vector>
#include <string>
#ifdef STORMM_USE_HPC
#  ifdef STORMM_USE_CUDA
#    include <cuda_runtime.h>
#  endif
#endif
#include "copyright.h"
#include "Constants/scaling.h"
 
namespace stormm {
namespace card {

constexpr long long int significant_gpu_memory = constants::mega;
  
class GpuDetails {
public:

  GpuDetails();
#ifdef STORMM_USE_HPC
#  ifdef STORMM_USE_CUDA
  GpuDetails(const cudaDeviceProp &devprop, int dev_index);
#  endif
#endif

  bool getAvailability() const;

  bool getGpuSupported() const;

  int getArchMajor() const;

  int getArchMinor() const;

  int getSMPCount() const;

  int getCardRam() const;

  int getMaxThreadsPerBlock() const;
  
  int getMaxThreadsPerSMP() const;

  int getMaxBlocksPerSMP() const;

  int getMaxSharedPerBlock() const;
  
  int getMaxSharedPerSMP() const;

  int getGlobalCacheSize() const;

  int getRegistersPerBlock() const;
  
  int getRegistersPerSMP() const;
  
  std::string getCardName() const;

  void setSMPCount(int smp_count_in);

  bool operator==(const GpuDetails &right) const;
  bool operator!=(const GpuDetails &right) const;
  
private:
  bool available;            
  bool supported;            
  int arch_major;            
  int arch_minor;            
  int smp_count;             
  int card_ram;              
  int max_threads_per_block; 
  int max_threads_per_smp;   
  int max_blocks_per_smp;    
  int max_shared_per_block;  
  int max_shared_per_smp;    
  int global_cache_size;     
  int registers_per_block;   
  int registers_per_smp;     
  std::string card_name;     
};
  
} // namespace card
} // namespace stormm

extern stormm::card::GpuDetails null_gpu;
 
#endif