cuda相关函数

内存相关

函数	形参	返回值	意义
__host__ ​ __device__ ​cudaError_t cudaMalloc ( void** devPtr, size_t size )	devPtr:指向分配内存 size:需要分配的大小	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorMemoryAllocation = 2:内存分配失败	在设备上分配内存
__host__ ​ __device__ ​cudaError_t cudaFree ( void* devPtr )	devPtr:需要释放的设备内存指针	cudaSuccess = 0:成功 cudaErrorInvalidDevicePointer = 17:不是有效的设备指针	释放设备内存
__host__ ​cudaError_t cudaMemcpy ( void* dst, const void* src, size_t count, cudaMemcpyKind kind )	dst:目的地址 src:源地址 count:字节数 kind:方向，包括： cudaMemcpyHostToHost, cudaMemcpyHostToDevice, cudaMemcpyDeviceToHost, cudaMemcpyDeviceToDevice, cudaMemcpyDefault	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidMemcpyDirection = 21:kind参数错误	从src拷贝count个字节到dst

设备属性相关

函数	形参	返回值	意义
__host__ ​ __device__ ​cudaError_t cudaGetDevice ( int* device )	device:返回正在运行的设备	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对	获取正则运行的设备
__host__ ​ __device__ ​cudaError_t cudaGetDeviceCount ( int* count )	count:返回可用设备的个数	cudaSuccess = 0:成功	获取设备个数
__host__ ​cudaError_t cudaGetDeviceProperties ( cudaDeviceProp* prop, int device )	prop:设备属性 device:设备号	cudaSuccess = 0:成功 cudaErrorInvalidDevice = 10:不是有效设备号	获取设备的性质
__host__ ​cudaError_t cudaChooseDevice ( int* device, const cudaDeviceProp* prop )	device:返回最符合的设备号 prop:设备属性	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对	选择最符号的设备号
__host__ ​cudaError_t cudaMemcpyToSymbol ( const void* symbol, const void* src, size_t count, size_t offset = 0, cudaMemcpyKind kind = cudaMemcpyHostToDevice )	symbol:目的常量地址 src:源地址 count:拷贝字节数 offset:symbol起始地址的偏移 kind:传输方向 stream:流的方向	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidSymbol = 13:不是有效识别符 cudaErrorInvalidMemcpyDirection = 21:拷贝方向不对 cudaErrorNoKernelImageForDevice = 48:没有可用的设备映像	拷贝数据到常量内存

event相关

函数	形参	返回值	意义
__host__ ​cudaError_t cudaEventCreate ( cudaEvent_t* event )	event:返回创建的event obj	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorLaunchFailure = 4:执行kernel过程中发生故障，常见的有无效指针、访问月结等 cudaErrorMemoryAllocation = 2:没有足够多的内存可分配	创建event
__host__ ​ device ​cudaError_t cudaEventDestroy ( cudaEvent_t event )	event:销毁的event	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorLaunchFailure = 4:执行kernel过程中发生故障	销毁event
__host__ ​cudaError_t cudaEventElapsedTime ( float* ms, cudaEvent_t start, cudaEvent_t end )	ms:返回的时间差 start开始事件: stop:停止事件	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidResourceHandle = 33：传给API的资源句柄不对，常见类型有 cudaStream_t and cudaEvent_t cudaErrorLaunchFailure = 4:执行kernel过程中发生故障	计算两个事件的时间差
__host__ ​\ _device_ ​cudaError_t cudaEventRecord ( cudaEvent_t event, cudaStream_t stream = 0 )	event:需要记录的时间 stream:需要记录事件的流	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidResourceHandle = 33：传给API的资源句柄不对 cudaErrorLaunchFailure = 4:执行kernel过程中发生故障	记录某个事件
__host__ ​cudaError_t cudaEventSynchronize ( cudaEvent_t event )	event:等待完成的时间	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidResourceHandle = 33：传给API的资源句柄不对 cudaErrorLaunchFailure = 4:执行kernel过程中发生故障	等待时间完成

texture

函数	形参	返回值	意义
template < class T, int dim, enum cudaTextureReadMode readMode > __host__ ​cudaError_t cudaBindTexture ( size_t* offset, const texture < T, dim, readMode > & tex, const void* devPtr, size_t size = UINT_MAX )	offset:偏移,一般为NULL tex:绑定的texture名 devPtr:被绑定的内存地址 size:绑定的大小	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidTexture = 18:绑定的texture不对	将devPtr绑定为texture内存
template < class T, int dim, enum cudaTextureReadMode readMode > __host__ ​cudaError_t cudaBindTexture2D ( size_t* offset, const texture < T, dim, readMode > & tex, const void* devPtr, size_t width, size_t height, size_t pitch )	offset:偏移 tex:绑定的texture引用 devPtr:需要绑定的地址	cudaSuccess = 0:成功 cudaErrorInvalidValue = 11:形参不对 cudaErrorInvalidTexture = 18:绑定的texture不对	将devPtr绑定为texture2D内存
template < class T, int dim, enum cudaTextureReadMode readMode > __host__ ​cudaError_t cudaBindTexture2D ( size_t* offset, const texture < T, dim, readMode > & tex, const void* devPtr, const cudaChannelFormatDesc& desc, size_t width, size_t height, size_t pitch )
template < class T, int dim, enum cudaTextureReadMode readMode > __host__ ​cudaError_t cudaBindTexture ( size_t* offset, const texture < T, dim, readMode > & tex, const void* devPtr, const cudaChannelFormatDesc& desc, size_t size = UINT_MAX )
template < class T, int dim, enum cudaTextureReadMode readMode > __host__ ​cudaError_t cudaUnbindTexture ( const texture < T, dim, readMode > & tex )			取消绑定
template T tex1Dfetch(cudaTextureObject_t texObj, int x)	x:使用非标准化坐标		取数据
template T tex1D(cudaTextureObject_t texObj, float x)	使用texture坐标
template T tex2D(cudaTextureObject_t texObj, float x, float y)	使用texture坐标

参考链接:https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__MEMORY.html#group__CUDART__MEMORY_1g37d37965bfb4803b6d4e59ff26856356