acc_aux.c File Reference

#include "acc_aux.h"
#include <stdint.h>

Include dependency graph for acc_aux.c:

Macros
#define	CLAMP(x, low, high)   (((x) > (high)) ? (high) : (((x) < (low)) ? (low) : (x)))
#define	MAX_DIM   12800
#define	MAX_IN_DIM   256
#define	LOG_2_MAX_IN_DIM   8
#define	LOG_2_MAX_DIM   7

Functions
static double	load_imb (double *A, int nthreads)
double	get_acc_async_cost ()
static unsigned int	next_power_2 (unsigned int v)
static unsigned int	previous_power_2 (unsigned int v)
static uint32_t	my_log2 (const uint32_t x)
static unsigned int	nearst_2_100 (unsigned int v)
static unsigned int	nearst_k (unsigned int v)
double	estimate_acc_time (int m, int n, int k)
double	estimate_acc_gemm_time (int m, int n, int k)
double	estimate_acc_scatter_time (int m, int n, int k)
double	estimate_cpu_time (int m, int n, int k)
double	acc_data_send_time (size_t sz)
void	LookUpTableInit (int my_rank)
double	estimate_acc_scatter_time_strat1 (Ublock_info_t *Ublock_info, int_t nub, Remain_info_t *Lblock_info, int_t nlb)
int_t	fixed_cpu_acc_partition (Ublock_info_t *Ublock_info_Phi, int_t num_u_blks_Phi, int_t Rnbrow, int_t ldu_Phi)
int_t	tuned_partition (int_t num_u_blks_Phi, Ublock_info_t *Ublock_info_Phi, Remain_info_t *Remain_info, int_t RemainBlk, double cpu_time_0, int_t Rnbrow, int_t ldu_Phi)

Variables
double	cpu_bandwidth
int	communication_overlap
double	acc_async_cost
int_t	fixed_partition
double	frac
double	CpuDgemmLookUp [8][8][9]
double	PhiDgemmLookUp [8][8][9]
double	PhiBWLookUp [8]
double	MicPciBandwidth [18]
double	MicScatterBW [24][24]
double	l_count [24]
double	u_count [24]

Macro Definition Documentation

CLAMP

#define CLAMP	(		x,
			low,
			high
	)		(((x) > (high)) ? (high) : (((x) < (low)) ? (low) : (x)))

LOG_2_MAX_DIM

#define LOG_2_MAX_DIM   7

LOG_2_MAX_IN_DIM

#define LOG_2_MAX_IN_DIM   8

MAX_DIM

#define MAX_DIM   12800

MAX_IN_DIM

#define MAX_IN_DIM   256

Function Documentation

acc_data_send_time()

double acc_data_send_time

(

size_t

sz

)

Here is the call graph for this function:

estimate_acc_gemm_time()

double estimate_acc_gemm_time	(	int	m,
		int	n,
		int	k
	)

Here is the call graph for this function:

Here is the caller graph for this function:

estimate_acc_scatter_time()

double estimate_acc_scatter_time	(	int	m,
		int	n,
		int	k
	)

Here is the call graph for this function:

estimate_acc_scatter_time_strat1()

double estimate_acc_scatter_time_strat1	(	Ublock_info_t *	Ublock_info,
		int_t	nub,
		Remain_info_t *	Lblock_info,
		int_t	nlb
	)

Here is the caller graph for this function:

estimate_acc_time()

double estimate_acc_time	(	int	m,
		int	n,
		int	k
	)

Here is the call graph for this function:

Here is the caller graph for this function:

estimate_cpu_time()

double estimate_cpu_time	(	int	m,
		int	n,
		int	k
	)

Here is the call graph for this function:

Here is the caller graph for this function:

fixed_cpu_acc_partition()

int_t fixed_cpu_acc_partition	(	Ublock_info_t *	Ublock_info_Phi,
		int_t	num_u_blks_Phi,
		int_t	Rnbrow,
		int_t	ldu_Phi
	)

Here is the call graph for this function:

get_acc_async_cost()

double get_acc_async_cost

(

)

Here is the caller graph for this function:

load_imb()

static double load_imb ( double *  A, int  nthreads  )

inlinestatic

LookUpTableInit()

void LookUpTableInit

(

int

my_rank

)

Here is the call graph for this function:

Here is the caller graph for this function:

my_log2()

static uint32_t my_log2 ( const uint32_t  x )

inlinestatic

Here is the caller graph for this function:

nearst_2_100()

static unsigned int nearst_2_100 ( unsigned int  v )

inlinestatic

Here is the call graph for this function:

Here is the caller graph for this function:

nearst_k()

static unsigned int nearst_k ( unsigned int  v )

inlinestatic

Here is the call graph for this function:

Here is the caller graph for this function:

next_power_2()

static unsigned int next_power_2 ( unsigned int  v )

inlinestatic

previous_power_2()

static unsigned int previous_power_2 ( unsigned int  v )

inlinestatic

tuned_partition()

int_t tuned_partition	(	int_t	num_u_blks_Phi,
		Ublock_info_t *	Ublock_info_Phi,
		Remain_info_t *	Remain_info,
		int_t	RemainBlk,
		double	cpu_time_0,
		int_t	Rnbrow,
		int_t	ldu_Phi
	)

Here is the call graph for this function:

Here is the caller graph for this function:

Variable Documentation

acc_async_cost

double acc_async_cost

communication_overlap

int communication_overlap

cpu_bandwidth

double cpu_bandwidth

CpuDgemmLookUp

double CpuDgemmLookUp[8][8][9]

fixed_partition

int_t fixed_partition

frac

double frac

l_count

double l_count[24]

MicPciBandwidth

double MicPciBandwidth[18]

MicScatterBW

double MicScatterBW[24][24]

PhiBWLookUp

double PhiBWLookUp[8]

PhiDgemmLookUp

double PhiDgemmLookUp[8][8][9]

u_count

double u_count[24]