Mentions légales du service

Skip to content
Snippets Groups Projects
Commit 684dc885 authored by Mathieu Faverge's avatar Mathieu Faverge
Browse files

Merge branch 'starpu/codelet_refactor' into 'master' 

StarPU:  refactor GPU codelets

See merge request !267
parents b647c9c2 fba5d132
No related branches found
No related tags found
1 merge request!267StarPU: refactor GPU codelets
Hide whitespace changes
Inline Side-by-side
runtime/starpu/codelets/codelet_ztrsm.c
+ 25
− 22
View file @ 684dc885
......@@ -19,6 +19,7 @@
* @author Cedric Castagnede
* @author Lucas Barros de Assis
* @author Florent Pruvost
* @author Gwenole Lucas
* @date 2021-03-16
* @precisions normal z -> c d s
*
......@@ -42,37 +43,34 @@ struct cl_ztrsm_args_s {
static void
cl_ztrsm_cpu_func(void *descr[], void *cl_arg)
{
struct cl_ztrsm_args_s clargs;
struct cl_ztrsm_args_s *clargs = (struct cl_ztrsm_args_s*)cl_arg;
CHAM_tile_t *tileA;
CHAM_tile_t *tileB;
tileA = cti_interface_get(descr[0]);
tileB = cti_interface_get(descr[1]);
starpu_codelet_unpack_args( cl_arg, &clargs );
TCORE_ztrsm( clargs.side, clargs.uplo, clargs.transA, clargs.diag,
clargs.m, clargs.n, clargs.alpha, tileA, tileB );
TCORE_ztrsm( clargs->side, clargs->uplo, clargs->transA, clargs->diag,
clargs->m, clargs->n, clargs->alpha, tileA, tileB );
}
#ifdef CHAMELEON_USE_CUDA
static void
cl_ztrsm_cuda_func(void *descr[], void *cl_arg)
{
struct cl_ztrsm_args_s clargs;
struct cl_ztrsm_args_s *clargs = (struct cl_ztrsm_args_s*)cl_arg;
CHAM_tile_t *tileA;
CHAM_tile_t *tileB;
tileA = cti_interface_get(descr[0]);
tileB = cti_interface_get(descr[1]);
starpu_codelet_unpack_args( cl_arg, &clargs );
RUNTIME_getStream(stream);
CUDA_ztrsm(
clargs.side, clargs.uplo, clargs.transA, clargs.diag,
clargs.m, clargs.n,
(cuDoubleComplex*)&(clargs.alpha),
clargs->side, clargs->uplo, clargs->transA, clargs->diag,
clargs->m, clargs->n,
(cuDoubleComplex*)&(clargs->alpha),
tileA->mat, tileA->ld,
tileB->mat, tileB->ld,
stream );
......@@ -97,29 +95,34 @@ void INSERT_TASK_ztrsm( const RUNTIME_option_t *options,
CHAMELEON_Complex64_t alpha, const CHAM_desc_t *A, int Am, int An,
const CHAM_desc_t *B, int Bm, int Bn )
{
struct cl_ztrsm_args_s clargs = {
.side = side,
.uplo = uplo,
.transA = transA,
.diag = diag,
.m = m,
.n = n,
.alpha = alpha,
.tileA = A->get_blktile( A, Am, An ),
.tileB = B->get_blktile( B, Bm, Bn ),
};
struct cl_ztrsm_args_s *clargs = NULL;
void (*callback)(void*);
RUNTIME_request_t *request = options->request;
starpu_option_request_t *schedopt = (starpu_option_request_t *)(request->schedopt);
int workerid;
int exec = 0;
char *cl_name = "ztrsm";
/* Handle cache */
CHAMELEON_BEGIN_ACCESS_DECLARATION;
CHAMELEON_ACCESS_R(A, Am, An);
CHAMELEON_ACCESS_RW(B, Bm, Bn);
exec = __chameleon_need_exec;
CHAMELEON_END_ACCESS_DECLARATION;
if ( exec ) {
clargs = malloc( sizeof( struct cl_ztrsm_args_s ) );
clargs->side = side;
clargs->uplo = uplo;
clargs->transA = transA;
clargs->diag = diag;
clargs->m = m;
clargs->n = n;
clargs->alpha = alpha;
clargs->tileA = A->get_blktile( A, Am, An );
clargs->tileB = B->get_blktile( B, Bm, Bn );
}
/* Callback fro profiling information */
callback = options->profiling ? cl_ztrsm_callback : NULL;
......@@ -130,7 +133,7 @@ void INSERT_TASK_ztrsm( const RUNTIME_option_t *options,
rt_starpu_insert_task(
&cl_ztrsm,
/* Task codelet arguments */
STARPU_VALUE, &clargs, sizeof(struct cl_ztrsm_args_s),
STARPU_CL_ARGS, clargs, sizeof(struct cl_ztrsm_args_s),
STARPU_R, RTBLKADDR(A, CHAMELEON_Complex64_t, Am, An),
STARPU_RW, RTBLKADDR(B, CHAMELEON_Complex64_t, Bm, Bn),
......
runtime/starpu/codelets/codelet_ztrtri.c
+ 19
− 15
View file @ 684dc885
......@@ -41,17 +41,16 @@ struct cl_ztrtri_args_s {
static void
cl_ztrtri_cpu_func(void *descr[], void *cl_arg)
{
struct cl_ztrtri_args_s clargs;
struct cl_ztrtri_args_s *clargs = (struct cl_ztrtri_args_s *)cl_arg;
CHAM_tile_t *tileA;
int info = 0;
tileA = cti_interface_get(descr[0]);
starpu_codelet_unpack_args( cl_arg, &clargs );
TCORE_ztrtri( clargs.uplo, clargs.diag, clargs.n, tileA, &info );
TCORE_ztrtri( clargs->uplo, clargs->diag, clargs->n, tileA, &info );
if ( (clargs.sequence->status == CHAMELEON_SUCCESS) && (info != 0) ) {
RUNTIME_sequence_flush( NULL, clargs.sequence, clargs.request, clargs.iinfo+info );
if ( (clargs->sequence->status == CHAMELEON_SUCCESS) && (info != 0) ) {
RUNTIME_sequence_flush( NULL, clargs->sequence, clargs->request, clargs->iinfo+info );
}
}
#endif /* !defined(CHAMELEON_SIMULATION) */
......@@ -66,26 +65,31 @@ void INSERT_TASK_ztrtri( const RUNTIME_option_t *options,
const CHAM_desc_t *A, int Am, int An,
int iinfo )
{
struct cl_ztrtri_args_s clargs = {
.uplo = uplo,
.diag = diag,
.n = n,
.tileA = A->get_blktile( A, Am, An ),
.iinfo = iinfo,
.sequence = options->sequence,
.request = options->request,
};
struct cl_ztrtri_args_s *clargs = NULL;
void (*callback)(void*);
RUNTIME_request_t *request = options->request;
starpu_option_request_t *schedopt = (starpu_option_request_t *)(request->schedopt);
int workerid;
int exec = 0;
char *cl_name = "ztrtri";
/* Handle cache */
CHAMELEON_BEGIN_ACCESS_DECLARATION;
CHAMELEON_ACCESS_RW(A, Am, An);
exec = __chameleon_need_exec;
CHAMELEON_END_ACCESS_DECLARATION;
if ( exec ) {
clargs = malloc( sizeof( struct cl_ztrtri_args_s ) );
clargs->uplo = uplo;
clargs->diag = diag;
clargs->n = n;
clargs->tileA = A->get_blktile( A, Am, An );
clargs->iinfo = iinfo;
clargs->sequence = options->sequence;
clargs->request = options->request;
}
/* Callback fro profiling information */
callback = options->profiling ? cl_ztrtri_callback : NULL;
......@@ -96,7 +100,7 @@ void INSERT_TASK_ztrtri( const RUNTIME_option_t *options,
rt_starpu_insert_task(
&cl_ztrtri,
/* Task codelet arguments */
STARPU_VALUE, &clargs, sizeof(struct cl_ztrtri_args_s),
STARPU_CL_ARGS, clargs, sizeof(struct cl_ztrtri_args_s),
STARPU_RW, RTBLKADDR(A, CHAMELEON_Complex64_t, Am, An),
/* Common task arguments */
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment