dark_cuda.cpp File Reference

Include dependency graph for dark_cuda.cpp:

Macros
#define	cudaEventWaitDefault   0x00

Functions
cublasHandle_t	blas_handle ()
void	check_cuda_error (cudaError_t status, const char *const filename, const char *const funcname, const int line)
void	check_cuda_error_extended (cudaError_t status, const char *const filename, const char *const funcname, const int line)
void	cublas_check_error (cublasStatus_t status)
void	cublas_check_error_extended (cublasStatus_t status, const char *const filename, const char *const function, const int line)
float	cuda_compare (float *x_gpu, float *x, size_t n, char *s)
void	cuda_free (float *x_gpu)
void	cuda_free_host (float *x_cpu)
void *	cuda_get_context ()
int	cuda_get_device ()
dim3	cuda_gridsize (size_t n)
float *	cuda_make_array (float *x, size_t n)
	Allocate memory on the GPU.
float *	cuda_make_array_pinned (float *x, size_t n)
float *	cuda_make_array_pinned_preallocated (float *x, size_t n)
void **	cuda_make_array_pointers (void **x, size_t n)
int *	cuda_make_int_array (size_t n)
int *	cuda_make_int_array_new_api (int *x, size_t n)
void	cuda_pull_array (float *x_gpu, float *x, size_t n)
void	cuda_pull_array_async (float *x_gpu, float *x, size_t n)
void	cuda_push_array (float *x_gpu, float *x, size_t n)
void	cuda_random (float *x_gpu, size_t n)
void	cuda_set_device (int n)
	This is part of the original C API. This function does nothing when Darknet was built to run on the CPU.
void	cudnn_check_error (cudnnStatus_t status, const char *const filename, const char *const function, const int line)
void	cudnn_check_error_extended (cudnnStatus_t status, const char *const filename, const char *const function, const int line)
cudnnHandle_t	cudnn_handle ()
void	free_pinned_memory ()
cudaStream_t	get_cuda_stream ()
int	get_gpu_compute_capability (int i, char *device_name)
int	get_number_of_blocks (int array_size, int block_size)
void	pre_allocate_pinned_memory (const size_t size)
void	reset_wait_stream_events ()
void	show_cuda_cudnn_info ()
cudaStream_t	switch_stream (int i)
void	wait_stream (int i)

Variables
static cublasHandle_t	blasHandle [16]
static int	blasInit [16] = { 0 }
int	cuda_debug_sync = 0
static cudnnHandle_t	cudnnHandle [16]
static int	cudnnInit [16] = { 0 }
static volatile int	event_counter = 0
static const int	max_events = 1024
static int	streamInit [16] = { 0 }
static cudaStream_t	streamsArray [16]
static cudnnHandle_t	switchCudnnHandle [16]
static int	switchCudnnInit [16]
static cudaEvent_t	switchEventsArray [1024]
static int	switchStreamInit [16] = { 0 }
static cudaStream_t	switchStreamsArray [16]

Macro Definition Documentation

cudaEventWaitDefault

#define cudaEventWaitDefault   0x00

Function Documentation

blas_handle()

cublasHandle_t blas_handle

(

)

Here is the call graph for this function:

Here is the caller graph for this function:

check_cuda_error()

void check_cuda_error	(	cudaError_t	status,
		const char *const	filename,
		const char *const	funcname,
		const int	line
	)

Here is the call graph for this function:

Here is the caller graph for this function:

check_cuda_error_extended()

void check_cuda_error_extended	(	cudaError_t	status,
		const char *const	filename,
		const char *const	funcname,
		const int	line
	)

Here is the call graph for this function:

cublas_check_error()

void cublas_check_error

(

cublasStatus_t

status

)

Here is the caller graph for this function:

cublas_check_error_extended()

void cublas_check_error_extended	(	cublasStatus_t	status,
		const char *const	filename,
		const char *const	function,
		const int	line
	)

Here is the call graph for this function:

cuda_compare()

float cuda_compare	(	float *	x_gpu,
		float *	x,
		size_t	n,
		char *	s
	)

Here is the call graph for this function:

cuda_free()

void cuda_free

(

float *

x_gpu

)

cuda_free_host()

void cuda_free_host

(

float *

x_cpu

)

Here is the caller graph for this function:

cuda_get_context()

void * cuda_get_context

(

)

cuda_get_device()

int cuda_get_device

(

)

Here is the caller graph for this function:

cuda_gridsize()

dim3 cuda_gridsize

(

size_t

n

)

cuda_make_array()

float * cuda_make_array	(	float *	x,
		size_t	n
	)

Allocate memory on the GPU.

If x is not null, then copy the given floats from the host pointer.

Returns

a pointer to the CUDA memory allocation.

Warning

The copy is asynchronous and may not have finished when this function returns!

Here is the call graph for this function:

cuda_make_array_pinned()

float * cuda_make_array_pinned	(	float *	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_make_array_pinned_preallocated()

float * cuda_make_array_pinned_preallocated	(	float *	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_make_array_pointers()

void ** cuda_make_array_pointers	(	void **	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_make_int_array()

int * cuda_make_int_array

(

size_t

n

)

Here is the caller graph for this function:

cuda_make_int_array_new_api()

int * cuda_make_int_array_new_api	(	int *	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_pull_array()

void cuda_pull_array	(	float *	x_gpu,
		float *	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_pull_array_async()

void cuda_pull_array_async	(	float *	x_gpu,
		float *	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_push_array()

void cuda_push_array	(	float *	x_gpu,
		float *	x,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_random()

void cuda_random	(	float *	x_gpu,
		size_t	n
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cuda_set_device()

void cuda_set_device

(

int

n

)

This is part of the original C API. This function does nothing when Darknet was built to run on the CPU.

Here is the caller graph for this function:

cudnn_check_error()

void cudnn_check_error	(	cudnnStatus_t	status,
		const char *const	filename,
		const char *const	function,
		const int	line
	)

Here is the call graph for this function:

Here is the caller graph for this function:

cudnn_check_error_extended()

void cudnn_check_error_extended	(	cudnnStatus_t	status,
		const char *const	filename,
		const char *const	function,
		const int	line
	)

Here is the call graph for this function:

cudnn_handle()

cudnnHandle_t cudnn_handle

(

)

Here is the call graph for this function:

Here is the caller graph for this function:

free_pinned_memory()

void free_pinned_memory

(

)

Here is the call graph for this function:

Here is the caller graph for this function:

get_cuda_stream()

cudaStream_t get_cuda_stream

(

)

Here is the call graph for this function:

get_gpu_compute_capability()

int get_gpu_compute_capability	(	int	i,
		char *	device_name
	)

Here is the caller graph for this function:

get_number_of_blocks()

int get_number_of_blocks	(	int	array_size,
		int	block_size
	)

pre_allocate_pinned_memory()

void pre_allocate_pinned_memory

(

const size_t

size

)

Here is the call graph for this function:

Here is the caller graph for this function:

reset_wait_stream_events()

void reset_wait_stream_events

(

)

Here is the caller graph for this function:

show_cuda_cudnn_info()

void show_cuda_cudnn_info

(

)

Here is the call graph for this function:

Here is the caller graph for this function:

switch_stream()

cudaStream_t switch_stream

(

int

i

)

Here is the call graph for this function:

Here is the caller graph for this function:

wait_stream()

void wait_stream

(

int

i

)

Here is the call graph for this function:

Here is the caller graph for this function:

Variable Documentation

blasHandle

cublasHandle_t blasHandle[16]

static

blasInit

int blasInit[16] = { 0 }

static

cuda_debug_sync

int cuda_debug_sync = 0

Todo:

V3 is cuda_debug_sync still necessary?

cudnnHandle

cudnnHandle_t cudnnHandle[16]

static

cudnnInit

int cudnnInit[16] = { 0 }

static

event_counter

volatile int event_counter = 0

static

max_events

const int max_events = 1024

static

streamInit

int streamInit[16] = { 0 }

static

streamsArray

cudaStream_t streamsArray[16]

static

switchCudnnHandle

cudnnHandle_t switchCudnnHandle[16]

static

switchCudnnInit

int switchCudnnInit[16]

static

switchEventsArray

cudaEvent_t switchEventsArray[1024]

static

switchStreamInit

int switchStreamInit[16] = { 0 }

static

switchStreamsArray

cudaStream_t switchStreamsArray[16]

static