darknet_gpu.hpp File Reference

Some GPU-specific includes and definitions. More...

#include <cuda.h>
#include <cuda_runtime.h>
#include <cuda_runtime_api.h>
#include <curand.h>
#include <cublas_v2.h>
#include <cudnn.h>
#include <hip/hip_runtime.h>
#include <rocrand/rocrand.h>
#include <hiprand/hiprand.h>
#include <hipblas/hipblas.h>
#include <miopen/miopen.h>

Include dependency graph for darknet_gpu.hpp:

This graph shows which files directly or indirectly include this file:

Macros
#define	CUBLAS_OP_N   HIPBLAS_OP_N
#define	CUBLAS_OP_T   HIPBLAS_OP_T
#define	CUBLAS_STATUS_SUCCESS   HIPBLAS_STATUS_SUCCESS
#define	cublasCreate   hipblasCreate
#define	cublasHandle_t   hipblasHandle_t
#define	cublasSetStream   hipblasSetStream
#define	cublasSgemm   hipblasSgemm
#define	cublasStatus_t   hipblasStatus_t
#define	CUcontext   hipCtx_t
#define	cuCtxGetCurrent   hipCtxGetCurrent
#define	CUDA_SUCCESS   hipSuccess
#define	cudaDeviceProp   hipDeviceProp_t
#define	cudaDeviceScheduleBlockingSync   hipDeviceScheduleBlockingSync
#define	cudaDeviceSynchronize   hipDeviceSynchronize
#define	cudaDriverGetVersion   hipDriverGetVersion
#define	cudaError   hipError_t
#define	cudaError_t   hipError_t
#define	cudaErrorInsufficientDriver   hipErrorInsufficientDriver
#define	cudaErrorNoDevice   hipErrorNoDevice
#define	cudaEvent_t   hipEvent_t
#define	cudaEventCreate   hipEventCreate
#define	cudaEventCreateWithFlags   hipEventCreateWithFlags
#define	cudaEventDestroy   hipEventDestroy
#define	cudaEventDisableTiming   hipEventDisableTiming
#define	cudaEventElapsedTime   hipEventElapsedTime
#define	cudaEventRecord   hipEventRecord
#define	cudaEventSynchronize   hipEventSynchronize
#define	cudaFree   hipFree
#define	cudaFreeHost   hipHostFree
#define	cudaGetDevice   hipGetDevice
#define	cudaGetDeviceCount   hipGetDeviceCount
#define	cudaGetDeviceProperties   hipGetDeviceProperties
#define	cudaGetErrorName   hipGetErrorName
#define	cudaGetErrorString   hipGetErrorString
#define	cudaGetLastError   hipGetLastError
#define	cudaGraph_t   hipGraph_t
#define	cudaGraphExec_t   hipGraphExec_t
#define	cudaGraphInstantiate   hipGraphInstantiate
#define	cudaGraphLaunch   hipGraphLaunch
#define	cudaHostAlloc   hipHostMalloc
#define	cudaHostAlloc   hipHostMalloc
#define	cudaHostAllocDefault   hipHostMallocDefault
#define	cudaHostRegisterMapped   hipHostRegisterMapped
#define	cudaMalloc   hipMalloc
#define	cudaMemcpy   hipMemcpy
#define	cudaMemcpyAsync   hipMemcpyAsync
#define	cudaMemcpyDefault   hipMemcpyDefault
#define	cudaMemcpyDeviceToHost   hipMemcpyDeviceToHost
#define	cudaMemcpyHostToDevice   hipMemcpyHostToDevice
#define	cudaMemGetInfo   hipMemGetInfo
#define	cudaMemset   hipMemset
#define	cudaPeekAtLastError   hipPeekAtLastError
#define	cudaReadModeElementType   hipReadModeElementType
#define	cudaRuntimeGetVersion   hipRuntimeGetVersion
#define	cudaSetDevice   hipSetDevice
#define	cudaSetDeviceFlags   hipSetDeviceFlags
#define	cudaStream_t   hipStream_t
#define	cudaStreamBeginCapture   hipStreamBeginCapture
#define	cudaStreamCaptureModeGlobal   hipStreamCaptureModeGlobal
#define	cudaStreamCreate   hipStreamCreate
#define	cudaStreamCreateWithFlags   hipStreamCreateWithFlags
#define	cudaStreamEndCapture   hipStreamEndCapture
#define	cudaStreamNonBlocking   hipStreamNonBlocking
#define	cudaStreamSynchronize   hipStreamSynchronize
#define	cudaStreamWaitEvent   hipStreamWaitEvent
#define	cudaSuccess   hipSuccess
#define	CUDNN_BATCHNORM_SPATIAL   miopenBatchNormMode_t::miopenBNSpatial
#define	CUDNN_CROSS_CORRELATION   miopenConvolutionMode_t::miopenConvolution
#define	CUDNN_DATA_FLOAT   miopenDataType_t::miopenFloat
#define	CUDNN_DATA_HALF   miopenDataType_t::miopenHalf
#define	CUDNN_STATUS_SUCCESS   miopenStatus_t::miopenStatusSuccess
#define	CUDNN_TENSOR_NCHW   0
#define	cudnnBatchNormalizationBackward   miopenBatchNormalizationBackward
#define	cudnnBatchNormalizationForwardTraining   miopenBatchNormalizationForwardTraining
#define	cudnnConvolutionBackwardData(h, alpha, wd, w, dyd, dy, cd, algo, ws, s, beta, dxd, dx)   miopenConvolutionBackwardData(h, alpha, dyd, dy, wd, w, cd, algo, beta, dxd, dx, wd, s)
#define	cudnnConvolutionBackwardFilter(h, alpha, xd, x, dyd, dy, cd, algo, ws, s, beta, dwd, dw)   miopenConvolutionBackwardWeights(h, alpha, dyd, dy, xd, x, cd, algo, beta, dwd, dw, ws, s)
#define	cudnnConvolutionBwdDataAlgo_t   miopenConvBwdDataAlgorithm_t
#define	cudnnConvolutionBwdFilterAlgo_t   miopenConvBwdWeightsAlgorithm_t
#define	cudnnConvolutionDescriptor_t   miopenConvolutionDescriptor_t
#define	cudnnConvolutionForward(h, alpha, xd, x, wd, w, cd, algo, ws, s, beta, yd, y)   miopenConvolutionForward(h, alpha, xd, x, wd, w, cd, algo, beta, yd, y, ws, s)
#define	cudnnConvolutionFwdAlgo_t   miopenConvFwdAlgorithm_t
#define	cudnnConvolutionMode_t   miopenConvolutionMode_t
#define	cudnnCreate   miopenCreate
#define	cudnnCreateConvolutionDescriptor   miopenCreateConvolutionDescriptor
#define	cudnnCreateFilterDescriptor   miopenCreateTensorDescriptor /* TODO: check this, does tensor replace filter? */
#define	cudnnCreateTensorDescriptor   miopenCreateTensorDescriptor
#define	cudnnDataType_t   miopenDataType_t
#define	cudnnDataType_t   miopenDataType_t
#define	cudnnDestroyTensorDescriptor   miopenDestroyTensorDescriptor
#define	cudnnFilterDescriptor_t   miopenTensorDescriptor_t
#define	cudnnGetConvolutionBackwardDataWorkspaceSize(h, wd, dyd, cd, dxd, algo, s)   miopenConvolutionBackwardDataGetWorkSpaceSize(h, dyd, wd, cd, dxd, s)
#define	cudnnGetConvolutionBackwardFilterWorkspaceSize(h, xd, dyd, cd, gd, algo, s)   miopenConvolutionBackwardWeightsGetWorkSpaceSize(h, dyd, xd, cd, dwd, s)
#define	cudnnGetConvolutionForwardWorkspaceSize(h, xd, wd, cd, yd, algo, s)   miopenConvolutionForwardGetWorkSpaceSize(h, wd, xd, cd, yd, s)
#define	cudnnHandle_t   miopenHandle_t
#define	cudnnPoolingDescriptor_t   miopenPoolingDescriptor_t
#define	cudnnSetConvolution2dDescriptor(d, h, w, u, v, x, y, mode)   miopenInitConvolutionDescriptor(d, mode, h, w, u, v, x, y)
#define	cudnnSetFilter4dDescriptor(d, t, i, n, c, h, w)   miopenSet4dTensorDescriptor(d, t, n, c, h, w)
#define	cudnnSetStream   miopenSetStream
#define	cudnnSetTensor4dDescriptor(d, i, t, b, c, h, w)   miopenSet4dTensorDescriptor(d, t, b, c, h, w)
#define	cudnnStatus_t   miopenStatus_t
#define	cudnnTensorDescriptor_t   miopenTensorDescriptor_t
#define	CURAND_RNG_PSEUDO_DEFAULT   HIPRAND_RNG_PSEUDO_DEFAULT
#define	curandCreateGenerator   hiprandCreateGenerator
#define	curandGenerateUniform   hiprandGenerateUniform
#define	curandGenerator_t   hiprandGenerator_t
#define	curandSetPseudoRandomGeneratorSeed   hiprandSetPseudoRandomGeneratorSeed
#define	CUresult   hipError_t
#define	SHFL_DOWN(val, offset)   shfl_xor(val, offset)

Detailed Description

Some GPU-specific includes and definitions.

Handles differences between NVIDIA CUDA and AMD ROCm.

Macro Definition Documentation

CUBLAS_OP_N

#define CUBLAS_OP_N   HIPBLAS_OP_N

CUBLAS_OP_T

#define CUBLAS_OP_T   HIPBLAS_OP_T

CUBLAS_STATUS_SUCCESS

#define CUBLAS_STATUS_SUCCESS   HIPBLAS_STATUS_SUCCESS

cublasCreate

#define cublasCreate   hipblasCreate

cublasHandle_t

#define cublasHandle_t   hipblasHandle_t

cublasSetStream

#define cublasSetStream   hipblasSetStream

cublasSgemm

#define cublasSgemm   hipblasSgemm

cublasStatus_t

#define cublasStatus_t   hipblasStatus_t

CUcontext

#define CUcontext   hipCtx_t

cuCtxGetCurrent

#define cuCtxGetCurrent   hipCtxGetCurrent

CUDA_SUCCESS

#define CUDA_SUCCESS   hipSuccess

cudaDeviceProp

#define cudaDeviceProp   hipDeviceProp_t

cudaDeviceScheduleBlockingSync

#define cudaDeviceScheduleBlockingSync   hipDeviceScheduleBlockingSync

cudaDeviceSynchronize

#define cudaDeviceSynchronize   hipDeviceSynchronize

cudaDriverGetVersion

#define cudaDriverGetVersion   hipDriverGetVersion

cudaError

#define cudaError   hipError_t

cudaError_t

#define cudaError_t   hipError_t

cudaErrorInsufficientDriver

#define cudaErrorInsufficientDriver   hipErrorInsufficientDriver

cudaErrorNoDevice

#define cudaErrorNoDevice   hipErrorNoDevice

cudaEvent_t

#define cudaEvent_t   hipEvent_t

cudaEventCreate

#define cudaEventCreate   hipEventCreate

cudaEventCreateWithFlags

#define cudaEventCreateWithFlags   hipEventCreateWithFlags

cudaEventDestroy

#define cudaEventDestroy   hipEventDestroy

cudaEventDisableTiming

#define cudaEventDisableTiming   hipEventDisableTiming

cudaEventElapsedTime

#define cudaEventElapsedTime   hipEventElapsedTime

cudaEventRecord

#define cudaEventRecord   hipEventRecord

cudaEventSynchronize

#define cudaEventSynchronize   hipEventSynchronize

cudaFree

#define cudaFree   hipFree

cudaFreeHost

#define cudaFreeHost   hipHostFree

cudaGetDevice

#define cudaGetDevice   hipGetDevice

cudaGetDeviceCount

#define cudaGetDeviceCount   hipGetDeviceCount

cudaGetDeviceProperties

#define cudaGetDeviceProperties   hipGetDeviceProperties

cudaGetErrorName

#define cudaGetErrorName   hipGetErrorName

cudaGetErrorString

#define cudaGetErrorString   hipGetErrorString

cudaGetLastError

#define cudaGetLastError   hipGetLastError

cudaGraph_t

#define cudaGraph_t   hipGraph_t

cudaGraphExec_t

#define cudaGraphExec_t   hipGraphExec_t

cudaGraphInstantiate

#define cudaGraphInstantiate   hipGraphInstantiate

cudaGraphLaunch

#define cudaGraphLaunch   hipGraphLaunch

cudaHostAlloc [1/2]

#define cudaHostAlloc   hipHostMalloc

cudaHostAlloc [2/2]

#define cudaHostAlloc   hipHostMalloc

cudaHostAllocDefault

#define cudaHostAllocDefault   hipHostMallocDefault

cudaHostRegisterMapped

#define cudaHostRegisterMapped   hipHostRegisterMapped

cudaMalloc

#define cudaMalloc   hipMalloc

cudaMemcpy

#define cudaMemcpy   hipMemcpy

cudaMemcpyAsync

#define cudaMemcpyAsync   hipMemcpyAsync

cudaMemcpyDefault

#define cudaMemcpyDefault   hipMemcpyDefault

cudaMemcpyDeviceToHost

#define cudaMemcpyDeviceToHost   hipMemcpyDeviceToHost

cudaMemcpyHostToDevice

#define cudaMemcpyHostToDevice   hipMemcpyHostToDevice

cudaMemGetInfo

#define cudaMemGetInfo   hipMemGetInfo

cudaMemset

#define cudaMemset   hipMemset

cudaPeekAtLastError

#define cudaPeekAtLastError   hipPeekAtLastError

cudaReadModeElementType

#define cudaReadModeElementType   hipReadModeElementType

cudaRuntimeGetVersion

#define cudaRuntimeGetVersion   hipRuntimeGetVersion

cudaSetDevice

#define cudaSetDevice   hipSetDevice

cudaSetDeviceFlags

#define cudaSetDeviceFlags   hipSetDeviceFlags

cudaStream_t

#define cudaStream_t   hipStream_t

cudaStreamBeginCapture

#define cudaStreamBeginCapture   hipStreamBeginCapture

cudaStreamCaptureModeGlobal

#define cudaStreamCaptureModeGlobal   hipStreamCaptureModeGlobal

cudaStreamCreate

#define cudaStreamCreate   hipStreamCreate

cudaStreamCreateWithFlags

#define cudaStreamCreateWithFlags   hipStreamCreateWithFlags

cudaStreamEndCapture

#define cudaStreamEndCapture   hipStreamEndCapture

cudaStreamNonBlocking

#define cudaStreamNonBlocking   hipStreamNonBlocking

cudaStreamSynchronize

#define cudaStreamSynchronize   hipStreamSynchronize

cudaStreamWaitEvent

#define cudaStreamWaitEvent   hipStreamWaitEvent

cudaSuccess

#define cudaSuccess   hipSuccess

CUDNN_BATCHNORM_SPATIAL

#define CUDNN_BATCHNORM_SPATIAL   miopenBatchNormMode_t::miopenBNSpatial

CUDNN_CROSS_CORRELATION

#define CUDNN_CROSS_CORRELATION   miopenConvolutionMode_t::miopenConvolution

CUDNN_DATA_FLOAT

#define CUDNN_DATA_FLOAT   miopenDataType_t::miopenFloat

CUDNN_DATA_HALF

#define CUDNN_DATA_HALF   miopenDataType_t::miopenHalf

CUDNN_STATUS_SUCCESS

#define CUDNN_STATUS_SUCCESS   miopenStatus_t::miopenStatusSuccess

CUDNN_TENSOR_NCHW

#define CUDNN_TENSOR_NCHW   0

cudnnBatchNormalizationBackward

#define cudnnBatchNormalizationBackward   miopenBatchNormalizationBackward

cudnnBatchNormalizationForwardTraining

#define cudnnBatchNormalizationForwardTraining   miopenBatchNormalizationForwardTraining

cudnnConvolutionBackwardData

#define cudnnConvolutionBackwardData	(		h,
			alpha,
			wd,
			w,
			dyd,
			dy,
			cd,
			algo,
			ws,
			s,
			beta,
			dxd,
			dx
	)		miopenConvolutionBackwardData(h, alpha, dyd, dy, wd, w, cd, algo, beta, dxd, dx, wd, s)

cudnnConvolutionBackwardFilter

#define cudnnConvolutionBackwardFilter	(		h,
			alpha,
			xd,
			x,
			dyd,
			dy,
			cd,
			algo,
			ws,
			s,
			beta,
			dwd,
			dw
	)		miopenConvolutionBackwardWeights(h, alpha, dyd, dy, xd, x, cd, algo, beta, dwd, dw, ws, s)

cudnnConvolutionBwdDataAlgo_t

#define cudnnConvolutionBwdDataAlgo_t   miopenConvBwdDataAlgorithm_t

cudnnConvolutionBwdFilterAlgo_t

#define cudnnConvolutionBwdFilterAlgo_t   miopenConvBwdWeightsAlgorithm_t

cudnnConvolutionDescriptor_t

#define cudnnConvolutionDescriptor_t   miopenConvolutionDescriptor_t

cudnnConvolutionForward

#define cudnnConvolutionForward	(		h,
			alpha,
			xd,
			x,
			wd,
			w,
			cd,
			algo,
			ws,
			s,
			beta,
			yd,
			y
	)		miopenConvolutionForward(h, alpha, xd, x, wd, w, cd, algo, beta, yd, y, ws, s)

cudnnConvolutionFwdAlgo_t

#define cudnnConvolutionFwdAlgo_t   miopenConvFwdAlgorithm_t

cudnnConvolutionMode_t

#define cudnnConvolutionMode_t   miopenConvolutionMode_t

cudnnCreate

#define cudnnCreate   miopenCreate

cudnnCreateConvolutionDescriptor

#define cudnnCreateConvolutionDescriptor   miopenCreateConvolutionDescriptor

cudnnCreateFilterDescriptor

#define cudnnCreateFilterDescriptor   miopenCreateTensorDescriptor /* TODO: check this, does tensor replace filter? */

cudnnCreateTensorDescriptor

#define cudnnCreateTensorDescriptor   miopenCreateTensorDescriptor

cudnnDataType_t [1/2]

#define cudnnDataType_t   miopenDataType_t

cudnnDataType_t [2/2]

#define cudnnDataType_t   miopenDataType_t

cudnnDestroyTensorDescriptor

#define cudnnDestroyTensorDescriptor   miopenDestroyTensorDescriptor

cudnnFilterDescriptor_t

#define cudnnFilterDescriptor_t   miopenTensorDescriptor_t

cudnnGetConvolutionBackwardDataWorkspaceSize

#define cudnnGetConvolutionBackwardDataWorkspaceSize	(		h,
			wd,
			dyd,
			cd,
			dxd,
			algo,
			s
	)		miopenConvolutionBackwardDataGetWorkSpaceSize(h, dyd, wd, cd, dxd, s)

cudnnGetConvolutionBackwardFilterWorkspaceSize

#define cudnnGetConvolutionBackwardFilterWorkspaceSize	(		h,
			xd,
			dyd,
			cd,
			gd,
			algo,
			s
	)		miopenConvolutionBackwardWeightsGetWorkSpaceSize(h, dyd, xd, cd, dwd, s)

cudnnGetConvolutionForwardWorkspaceSize

#define cudnnGetConvolutionForwardWorkspaceSize	(		h,
			xd,
			wd,
			cd,
			yd,
			algo,
			s
	)		miopenConvolutionForwardGetWorkSpaceSize(h, wd, xd, cd, yd, s)

cudnnHandle_t

#define cudnnHandle_t   miopenHandle_t

cudnnPoolingDescriptor_t

#define cudnnPoolingDescriptor_t   miopenPoolingDescriptor_t

cudnnSetConvolution2dDescriptor

#define cudnnSetConvolution2dDescriptor	(		d,
			h,
			w,
			u,
			v,
			x,
			y,
			mode
	)		miopenInitConvolutionDescriptor(d, mode, h, w, u, v, x, y)

cudnnSetFilter4dDescriptor

#define cudnnSetFilter4dDescriptor	(		d,
			t,
			i,
			n,
			c,
			h,
			w
	)		miopenSet4dTensorDescriptor(d, t, n, c, h, w)

cudnnSetStream

#define cudnnSetStream   miopenSetStream

cudnnSetTensor4dDescriptor

#define cudnnSetTensor4dDescriptor	(		d,
			i,
			t,
			b,
			c,
			h,
			w
	)		miopenSet4dTensorDescriptor(d, t, b, c, h, w)

cudnnStatus_t

#define cudnnStatus_t   miopenStatus_t

cudnnTensorDescriptor_t

#define cudnnTensorDescriptor_t   miopenTensorDescriptor_t

CURAND_RNG_PSEUDO_DEFAULT

#define CURAND_RNG_PSEUDO_DEFAULT   HIPRAND_RNG_PSEUDO_DEFAULT

curandCreateGenerator

#define curandCreateGenerator   hiprandCreateGenerator

curandGenerateUniform

#define curandGenerateUniform   hiprandGenerateUniform

curandGenerator_t

#define curandGenerator_t   hiprandGenerator_t

curandSetPseudoRandomGeneratorSeed

#define curandSetPseudoRandomGeneratorSeed   hiprandSetPseudoRandomGeneratorSeed

CUresult

#define CUresult   hipError_t

SHFL_DOWN

#define SHFL_DOWN	(		val,
			offset
	)		shfl_xor(val, offset)