densemat_iter_macros.h File Reference

Macros for iterating through densemats. More...

#include "ghost/omp.h"

Include dependency graph for densemat_iter_macros.h:

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

Go to the source code of this file.

Macros
#define	DENSEMAT_DT   char
#define	DENSEMAT_ELSIZE(vec)   vec->elSize
#define	DENSEMAT_COMPACT(vec)   (!(vec->traits.flags & GHOST_DENSEMAT_SCATTERED))
#define	DENSEMAT_COMPACT_ROWS(vec)   (!(vec->traits.flags & GHOST_DENSEMAT_SCATTERED_ROWS))
#define	DENSEMAT_SCATTERED_COLS(vec)   ((vec->traits.flags & GHOST_DENSEMAT_SCATTERED_COLS))
#define	DENSEMAT_SINGLECOL_STRIDE1(vec)   (vec->traits.ncols == 1 && ((vec->traits.storage == GHOST_DENSEMAT_COLMAJOR) || ((vec->traits.storage == GHOST_DENSEMAT_ROWMAJOR) && (vec->stride == 1))))
#define	DENSEMAT_ITER(vec, call)
	Iterate over a densemats and execute a statement for each entry. More...
#define	DENSEMAT_ITER_INIT(vec, call)
	Iterate over a densemats and execute a statement for each entry. More...
#define	DENSEMAT_ITER2(vec1, vec2, call)   DENSEMAT_ITER2_OFFS(vec1,vec2,0,0,call)
#define	DENSEMAT_ITER2_OFFS(vec1, vec2, vec2roffs, vec2coffs, call)
	Iterate over two densemats synchronously and execute a statement for each entry. An offset may be given for the second input densemat. More...
#define	DENSEMAT_ITER2_COMPACT_OFFS_TRANSPOSED(vec1, vec2, vec2roffs, vec2coffs, call)
	Iterate over two densemats synchronously and execute a statement for each entry. The second densemat is stored transposed (compared to the first). An offset may be given for the second input densemat. More...
#define	DENSEMAT_ITER_BEGIN_COMPACT_SINGLECOL(vec, valptr, row, col, memrow, memcol)
#define	DENSEMAT_ITER_COMPACT_SINGLECOL_PAD(vec, valptr, row, col, memrow, memcol)
#define	DENSEMAT_ITER_BEGIN_COMPACT(vec, valptr, row, col, memrow, memcol)
#define	DENSEMAT_ITER_COMPACT_PAD(vec, valptr, row, col, memrow, memcol)
#define	DENSEMAT_ITER_END()
#define	DENSEMAT_ITER2_BEGIN_COMPACT(vec1, vec2, row, col, memrow1, memrow2, memcol1, memcol2)   DENSEMAT_ITER2_BEGIN_COMPACT_OFFS(vec1,vec2,row,col,memrow1,memrow2,memcol1,memcol2,0,0)
#define	DENSEMAT_ITER2_BEGIN_COMPACT_OFFS(vec1, vec2, valptr1, valptr2, row, col, memrow1, memrow2, memcol1, memcol2, vec2roffs, vec2coffs)
#define	DENSEMAT_ITER2_BEGIN_COMPACT_OFFS_SINGLECOL(vec1, vec2, valptr1, valptr2, row, col, memrow1, memrow2, memcol1, memcol2, vec2roffs, vec2coffs)
#define	DENSEMAT_ITER2_BEGIN_COMPACT_OFFS_TRANSPOSED(vec1, vec2, row, col, memrow1, memrow2, memcol1, memcol2, vec2roffs, vec2coffs)
#define	DENSEMAT_ITER2_BEGIN_SCATTEREDCOLS_COMPACTROWS1_OFFS(vec1, vec2, row, col, memrow1, memrow2, memcol1, memcol2, vec2roffs, vec2coffs)

Detailed Description

Macros for iterating through densemats.

Author

Moritz Kreutzer morit.nosp@m.z.kr.nosp@m.eutze.nosp@m.r@fa.nosp@m.u.de

Macro Definition Documentation

#define DENSEMAT_COMPACT

(

vec

)

(!(vec->traits.flags & GHOST_DENSEMAT_SCATTERED))

#define DENSEMAT_COMPACT_ROWS

(

vec

)

(!(vec->traits.flags & GHOST_DENSEMAT_SCATTERED_ROWS))

#define DENSEMAT_DT   char

#define DENSEMAT_ELSIZE

(

vec

)

vec->elSize

#define DENSEMAT_ITER	(		vec,
			call
	)

Iterate over a densemats and execute a statement for each entry.

This macro sets the following variables: row,col,memrow,memcol,valptr

Parameters

vec	The densemat.
call	The statement to call for each entry.

Returns

#define DENSEMAT_ITER2	(		vec1,
			vec2,
			call
	)		DENSEMAT_ITER2_OFFS(vec1,vec2,0,0,call)

See Also

DENSEMAT_ITER2_OFFS with offsets set to (0,0).

#define DENSEMAT_ITER2_BEGIN_COMPACT	(		vec1,
			vec2,
			row,
			col,
			memrow1,
			memrow2,
			memcol1,
			memcol2
	)		DENSEMAT_ITER2_BEGIN_COMPACT_OFFS(vec1,vec2,row,col,memrow1,memrow2,memcol1,memcol2,0,0)

#define DENSEMAT_ITER2_BEGIN_COMPACT_OFFS	(		vec1,
			vec2,
			valptr1,
			valptr2,
			row,
			col,
			memrow1,
			memrow2,
			memcol1,
			memcol2,
			vec2roffs,
			vec2coffs
	)

Value:

_Pragma("omp for schedule(runtime) private(col,memcol1,memcol2,memrow1,memrow2,valptr1,valptr2,cuvalptr1,cuvalptr2)")\
    for (row=0; row<DM_NROWS(vec1); row++) {\
        memrow1 = row;\
        memrow2 = row;\
        for (col = 0; col<vec1->traits.ncols; col++) {\
            memcol1 = col;\
            memcol2 = col;

#define DENSEMAT_ITER2_BEGIN_COMPACT_OFFS_SINGLECOL	(		vec1,
			vec2,
			valptr1,
			valptr2,
			row,
			col,
			memrow1,
			memrow2,
			memcol1,
			memcol2,
			vec2roffs,
			vec2coffs
	)

Value:

memcol1 = 0;\
    memcol2 = 0;\
    col = 0;\
    _Pragma("omp for schedule(runtime) private(memrow1,memrow2,valptr1,valptr2,cuvalptr1,cuvalptr2)")\
    for (row=0; row<DM_NROWS(vec1); row++) {\
        memrow1 = row;\
        memrow2 = row;\
        {\

#define DENSEMAT_ITER2_BEGIN_COMPACT_OFFS_TRANSPOSED	(		vec1,
			vec2,
			row,
			col,
			memrow1,
			memrow2,
			memcol1,
			memcol2,
			vec2roffs,
			vec2coffs
	)

Value:

_Pragma("omp for schedule(runtime) private(col,memcol1,memcol2,memrow1,memrow2,valptr1,valptr2,cuvalptr1,cuvalptr2)")\
    for (row=0; row<DM_NROWS(vec1); row++) {\
        memrow1 = row;\
        memrow2 = row;\
        for (col = 0; col<vec1->traits.ncols; col++) {\
            memcol1 = col;\
            memcol2 = col;

#define DENSEMAT_ITER2_BEGIN_SCATTEREDCOLS_COMPACTROWS1_OFFS	(		vec1,
			vec2,
			row,
			col,
			memrow1,
			memrow2,
			memcol1,
			memcol2,
			vec2roffs,
			vec2coffs
	)

Value:

memcol1 = -1;\
    for (col = 0; col<vec1->traits.ncols; col++) {\
        memcol1 = ghost_bitmap_next(vec1->colmask,memcol1);\
        memcol2 = col;\
        _Pragma("omp for schedule(runtime) private(memrow1,memrow2,valptr1,valptr2,cuvalptr1,cuvalptr2)")\
        for (row=0; row<DM_NROWS(vec1); row++) {\
            memrow1 = row;\
            memrow2 = row;\

#define DENSEMAT_ITER2_COMPACT_OFFS_TRANSPOSED	(		vec1,
			vec2,
			vec2roffs,
			vec2coffs,
			call
	)

Value:

ghost_lidx row=0,col=0,memrow1=0,memcol1=0,memrow2=0,memcol2=0;\
    DENSEMAT_DT *valptr1 = NULL, *valptr2 = NULL, *cuvalptr1 = NULL, *cuvalptr2 = NULL;\
    if (ghost_omp_in_parallel()) {\
        DENSEMAT_ITER2_BEGIN_COMPACT_OFFS_TRANSPOSED(vec1,vec2,row,col,memrow1,memrow2,memcol1,memcol2,vec2roffs,vec2coffs);\
        valptr1 = DENSEMAT_VALPTR(vec1,row,col);\
        valptr2 = DENSEMAT_VALPTR_TRANSPOSED(vec2,row+vec2roffs,col+vec2coffs);\
        cuvalptr1 = DENSEMAT_CUVALPTR(vec1,row,col);\
        cuvalptr2 = DENSEMAT_CUVALPTR_TRANSPOSED(vec2,row+vec2roffs,col+vec2coffs);\
        call;\
        DENSEMAT_ITER_END();\
    } else {\
        _Pragma("omp parallel")\
        {\
            DENSEMAT_ITER2_BEGIN_COMPACT_OFFS_TRANSPOSED(vec1,vec2,row,col,memrow1,memrow2,memcol1,memcol2,vec2roffs,vec2coffs);\
            valptr1 = DENSEMAT_VALPTR(vec1,row,col);\
            valptr2 = DENSEMAT_VALPTR_TRANSPOSED(vec2,row+vec2roffs,col+vec2coffs);\
            cuvalptr1 = DENSEMAT_CUVALPTR(vec1,row,col);\
            cuvalptr2 = DENSEMAT_CUVALPTR_TRANSPOSED(vec2,row+vec2roffs,col+vec2coffs);\
            call;\
            DENSEMAT_ITER_END();\
        }\
    }\
    /* Trick the compiler to not produce warnings about unused variables */\
    if ((row+col+memrow1+memcol1+memrow2+memcol2 < 0) || \
            (valptr1 == (DENSEMAT_DT *)0xbeef) || (valptr2 == (DENSEMAT_DT *)0xbeef) || \
            (cuvalptr1 == (DENSEMAT_DT *)0xbeef) || (cuvalptr2 == (DENSEMAT_DT *)0xbeef)) \
            {printf("Never happens\n");}

Iterate over two densemats synchronously and execute a statement for each entry. The second densemat is stored transposed (compared to the first). An offset may be given for the second input densemat.

This macro sets the following variables: row,col,memrow1,memcol1,memrow2,memcol2

Parameters

vec1	The first densemat
vec2	The second densemat
vec2roffs	The row offset to the second densemat.
vec2coffs	The col offset to the second densemat.
call	The statement to call for each entry.

Returns

#define DENSEMAT_ITER2_OFFS	(		vec1,
			vec2,
			vec2roffs,
			vec2coffs,
			call
	)

Iterate over two densemats synchronously and execute a statement for each entry. An offset may be given for the second input densemat.

This macro sets the following variables: row,col,memrow1,memcol1,memrow2,memcol2,valptr1,valptr2

Parameters

vec1	The first densemat
vec2	The second densemat
vec2roffs	The row offset to the second densemat.
vec2coffs	The col offset to the second densemat.
call	The statement to call for each entry.

Returns

#define DENSEMAT_ITER_BEGIN_COMPACT	(		vec,
			valptr,
			row,
			col,
			memrow,
			memcol
	)

Value:

_Pragma("omp for schedule(runtime) private(col,memrow,memcol,valptr,cuvalptr)")\
    for (row = 0; row<DM_NROWSPAD(vec); row++) {\
        memrow = row;\
        col = 0;\
        if( row<DM_NROWS(vec) )\
        for (; col<vec->traits.ncols; col++) {\
            memcol = col;\

#define DENSEMAT_ITER_BEGIN_COMPACT_SINGLECOL	(		vec,
			valptr,
			row,
			col,
			memrow,
			memcol
	)

Value:

col = 0;\
    memcol = 0;\
    _Pragma("omp for schedule(runtime) private(memrow,valptr,cuvalptr)")\
    for (row = 0; row<DM_NROWSPAD(vec); row++) {\
        memrow = row;\
        if( row < DM_NROWS(vec) ) {\

#define DENSEMAT_ITER_COMPACT_PAD	(		vec,
			valptr,
			row,
			col,
			memrow,
			memcol
	)

Value:

}\
        for (; col<vec->traits.ncolspadded; col++) {\
            valptr = DENSEMAT_VALPTR(vec,row,col);\
            memset(valptr,0,DENSEMAT_ELSIZE(vec)*sizeof(DENSEMAT_DT));\

#define DENSEMAT_ITER_COMPACT_SINGLECOL_PAD	(		vec,
			valptr,
			row,
			col,
			memrow,
			memcol
	)

Value:

} else {\
            valptr = DENSEMAT_VALPTR_SINGLECOL_STRIDE1(vec,row,col);\
            memset(valptr,0,DENSEMAT_ELSIZE(vec)*sizeof(DENSEMAT_DT));\

#define DENSEMAT_ITER_END

(

)

Value:

}\
    }

#define DENSEMAT_ITER_INIT	(		vec,
			call
	)

Iterate over a densemats and execute a statement for each entry.

This macro sets the following variables: row,col,memrow,memcol,valptr In addition to DENSEMAT_ITER, this macro intializes padding to zero

Parameters

vec	The densemat.
call	The statement to call for each entry.

Returns

#define DENSEMAT_SCATTERED_COLS

(

vec

)

((vec->traits.flags & GHOST_DENSEMAT_SCATTERED_COLS))

#define DENSEMAT_SINGLECOL_STRIDE1

(

vec

)

(vec->traits.ncols == 1 && ((vec->traits.storage == GHOST_DENSEMAT_COLMAJOR) || ((vec->traits.storage == GHOST_DENSEMAT_ROWMAJOR) && (vec->stride == 1))))