#include <clm4rm.h>
#include <stdlib.h>
#include <qdebug.h>

Functions
void	print_event_info (cl_event event)

void	clm4rm_mul_block (clmatrix_t C, clmatrix_t A, clmatrix_t B, int r0, int r1, cl_command_queue queue, clm4rm_conditions cond)

void	clm4rm_mul (clmatrix_t C, clmatrix_t A, clmatrix_t B, cl_command_queue queue, clm4rm_conditions cond)
	Boolean matrix multiplication on the GPU using the method of the Four Russians. C := A * B. More...

void	clcubic_mul_enqeue (clmatrix_t C, clmatrix_t A, clmatrix_t B, int tile_n, int tile_m, size_t work_offset[2], size_t work_size[2], int uptri, cl_command_queue queue, clm4rm_conditions cond)

void	clcubic_mul (clmatrix_t C, clmatrix_t A, clmatrix_t B, size2_t max_tile, cl_command_queue queue, clm4rm_conditions cond)
	Boolean matrix multiplication on the GPU using nested loops. C := A*B. More...

void	clutri_mul (clmatrix_t C, clmatrix_t A, clmatrix_t B, size2_t max_tile, cl_command_queue queue, clm4rm_conditions cond)
	Boolean matrix multiplication on the GPU using nested loops. C := A*B Assumes matrixes to be upper triangular. More...

void	printb (uint32_t x, int k)

void	print3 (uint32_t x, int k)

void	create_index_tables (uint32_t k)

Variables
cl_kernel	clm4rm_mul_kernel
	OpenCL kernel for Four-Russians matrix multiplication. More...

cl_kernel	clcubic_mul_kernel [MAX_TILE_M+1]
	OpenCL kernels for cubic matrix multiplication. Each kernel for a tile size. Actual tile sizes are injected as macros. More...

cl_kernel	clutri_mul_kernel [MAX_TILE_M+1]
	OpenCL kernels for cubic upper-triangle matrix multiplication. Each kernel for a tile size. Actual tile sizes are injected as macros. More...

Function Documentation

◆ clcubic_mul()

void clcubic_mul	(	clmatrix_t *	C,
		clmatrix_t *	A,
		clmatrix_t *	B,
		size2_t	max_tile,
		cl_command_queue	queue,
		clm4rm_conditions *	cond
	)

Boolean matrix multiplication on the GPU using nested loops. C := A*B.

Matrix is partitioned into three parts:

       +----------------------------------+----+
       |                                  |    |
       |      |      |      |      |      |    |
       |                                  |    |
       | - - -+ - - -+ - - -+ - - -+ - - -|    |
       |                                  |    |
       |      |      |      |      |      |    |
       |                                  |    |
       | - - -+ - - -+ - - -+ - - -+ - - -|    |
       |                                  |REST|
       |      |      |      |      |      |RIGHT
       |                                  |    |
       | - - -+ - - -+ - - -+ - - -+ - - -|    |
       |                                  |    |
       |      |      |      |      |      |    |
       |                                  |    |
       | - - -+ - - -+ - - -+ - - -+ - - -|    |
       |                                  |    |
       |      |      |      |      |      |    |
       |                                  |    |
       |------+------+------+------+------+----|
       |   REST BOTTOM                         |
       | - - -+ - - -+ - - -+ - - -+ - - -+ - -+

Definition at line 132 of file clm4rm_multiplication.cpp.

◆ clcubic_mul_enqeue()

void clcubic_mul_enqeue	(	clmatrix_t *	C,
		clmatrix_t *	A,
		clmatrix_t *	B,
		int	tile_n,
		int	tile_m,
		size_t	work_offset[2],
		size_t	work_size[2],
		int	uptri,
		cl_command_queue	queue,
		clm4rm_conditions *	cond
	)

Definition at line 241 of file clm4rm_multiplication.cpp.

◆ clm4rm_mul()

void clm4rm_mul	(	clmatrix_t *	C,
		clmatrix_t *	A,
		clmatrix_t *	B,
		cl_command_queue	queue,
		clm4rm_conditions *	cond
	)

Boolean matrix multiplication on the GPU using the method of the Four Russians. C := A * B.

The function returns immediately. The operation is scheduled for asynchronous execution of the GPU. Use post-condition events to wait for the execution of the operation.

Parameters

C	a matrix structure; receives the resutl
A	a matrix structure
B	a matrix structure
queue	OpenCL command queue
cond	keeps track of pre-conditions and newly created post-conditions

Definition at line 30 of file clm4rm_multiplication.cpp.

◆ clm4rm_mul_block()

void clm4rm_mul_block	(	clmatrix_t *	C,
		clmatrix_t *	A,
		clmatrix_t *	B,
		int	r0,
		int	r1,
		cl_command_queue	queue,
		clm4rm_conditions *	cond
	)

Definition at line 44 of file clm4rm_multiplication.cpp.

◆ clutri_mul()

void clutri_mul	(	clmatrix_t *	C,
		clmatrix_t *	A,
		clmatrix_t *	B,
		size2_t	max_tile,
		cl_command_queue	queue,
		clm4rm_conditions *	cond
	)

Boolean matrix multiplication on the GPU using nested loops. C := A*B Assumes matrixes to be upper triangular.

Matrix is partitioned into three parts

       +----------------------------------+----+
       |                                  |    |
       |      |      |      |      |      |    |
       |                                  |    |
       | - - -+ - - -+ - - -+ - - -+ - - -|    |
       |                                  |    |
       |      |      |      |      |      |    |
       |                                  |    |
       |      + - - -+ - - -+ - - -+ - - -|    |
       |                                  |REST|
       |             |      |      |      |RIGHT
       |                                  |    |
       |             + - - -+ - - -+ - - -|    |
       |                                  |    |
       |                    |      |      |    |
       |   (empty)                        |    |
       |                    + - - -+ - - -|    |
       |                                  |    |
       |                           |      |    |
       |                                  |    |
       | - - - - - - + - - - - - - + - - -+    |
       |   (empty)                        |    |
       | - - -+ - - -+ - - -+ - - -+ - - -+ - -+

Definition at line 339 of file clm4rm_multiplication.cpp.

◆ create_index_tables()

void create_index_tables ( uint32_t k )

Definition at line 453 of file clm4rm_multiplication.cpp.

◆ print3()

void print3	(	uint32_t	x,
		int	k
	)

Definition at line 439 of file clm4rm_multiplication.cpp.

◆ print_event_info()

void print_event_info ( cl_event event )

Definition at line 228 of file clm4rm_multiplication.cpp.

◆ printb()

void printb	(	uint32_t	x,
		int	k
	)

Definition at line 433 of file clm4rm_multiplication.cpp.

Variable Documentation

◆ clcubic_mul_kernel

cl_kernel clcubic_mul_kernel[MAX_TILE_M+1]

OpenCL kernels for cubic matrix multiplication. Each kernel for a tile size. Actual tile sizes are injected as macros.

Definition at line 71 of file clm4rm.cpp.

◆ clm4rm_mul_kernel

cl_kernel clm4rm_mul_kernel

OpenCL kernel for Four-Russians matrix multiplication.

Definition at line 68 of file clm4rm.cpp.

◆ clutri_mul_kernel

cl_kernel clutri_mul_kernel[MAX_TILE_M+1]

OpenCL kernels for cubic upper-triangle matrix multiplication. Each kernel for a tile size. Actual tile sizes are injected as macros.

Definition at line 72 of file clm4rm.cpp.

Functions

Variables