stormm::card::GpuDetails Class Reference

Pertinent aspects of one particular GPU. Condensing the data for each GPU in this manner helps to ensure that one cache line will obtain all statistics for a single GPU. More...

#include <gpu_details.h>

Public Member Functions
bool	getAvailability () const
	Get the availability of the GPU.
bool	getGpuSupported () const
	Get whether the architecture of the GPU is supported.
int	getArchMajor () const
	Get the major architectural number of the GPU.
int	getArchMinor () const
	Get the minor architectural number of the GPU.
int	getSMPCount () const
	Get the number of streaming multiprocessors on the GPU.
int	getCardRam () const
	Get the amount of RAM on the GPU, in megabytes (assuming that all of the device RAM is available for any given process)
int	getMaxThreadsPerBlock () const
	Get the maximum number of threads per block supported by this GPU.
int	getMaxThreadsPerSMP () const
	Get the maximum number of threads per streaming multiprocessor on this GPU.
int	getMaxBlocksPerSMP () const
	Get the maximum number of blocks per streaming multiprocessor on this GPU.
int	getMaxSharedPerBlock () const
	Get the maximum amount of L1 shared memory per block on this GPU.
int	getMaxSharedPerSMP () const
	Get the available L1 shared memory per streaming multiprocessor on this GPU.
int	getGlobalCacheSize () const
	Get the total amount of global cache (L2) on the card, in bytes.
int	getRegistersPerBlock () const
	Get the maximum number of registers available per block on this GPU.
int	getRegistersPerSMP () const
	Get the maximum number of registers available per streaming multiprocessor on this GPU.
std::string	getCardName () const
	Get the name of the GPU.
void	setSMPCount (int smp_count_in)
	Set the number of streaming multiprocessors. This is useful for experimentation in mock settings, or perhaps running STORMM on GPUs when MiG is engaged.
	GpuDetails ()
	Constructors include a blank constructor (which automatically labels the GPU as unavailable) and constructors based on cudaDeviceProp or hipDeviceProp.
bool	operator== (const GpuDetails &right) const
	Overload the == and != operators to compare GpuDetails objects.
bool	operator!= (const GpuDetails &right) const

Detailed Description

Pertinent aspects of one particular GPU. Condensing the data for each GPU in this manner helps to ensure that one cache line will obtain all statistics for a single GPU.

Constructor & Destructor Documentation

GpuDetails()

stormm::card::GpuDetails::GpuDetails

(

)

Constructors include a blank constructor (which automatically labels the GPU as unavailable) and constructors based on cudaDeviceProp or hipDeviceProp.

Parameters

devprop	A CUDA device properties object reported by cudaGetDeviceProperties()
dev_index	Index of the GPU in a longer list produced by the CUDA runtime library

Member Function Documentation

operator==()

bool stormm::card::GpuDetails::operator==

(

const GpuDetails &

right

)

const

Overload the == and != operators to compare GpuDetails objects.

Parameters

right

The other GPU to compare against

setSMPCount()

void stormm::card::GpuDetails::setSMPCount

(

int

smp_count_in

)

Set the number of streaming multiprocessors. This is useful for experimentation in mock settings, or perhaps running STORMM on GPUs when MiG is engaged.

Parameters

smp_count_in

The number of streaming multiprocessors

---

The documentation for this class was generated from the following files:

• src/Accelerator/gpu_details.h
• src/Accelerator/gpu_details.cpp