| Message ID | 286d95c2-dbf3-c4d8-3a79-f1dbb7e650b2@mentor.com |
|---|---|
| State | New |
| Headers | show |
| Series | [libgomp,openacc] Use GOMP_ASYNC_SYNC in GOACC_declare | expand |
On 11/17/2017 09:45 AM, Tom de Vries wrote: > Hi, > > GOACC_enter_exit_data has this prototype: > ... > void > GOACC_enter_exit_data (int device, size_t mapnum, > void **hostaddrs, size_t *sizes, > unsigned short *kinds, > int async, int num_waits, ...) > ... > > And GOACC_declare calls GOACC_enter_exit_data with async arg zero: > ... > case GOMP_MAP_DELETE: > GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], > &kinds[i], 0, 0); > ... > > Async arg zero means some async queue (see openacc 2.0a, 2.14.1 "async > clause" for more details). > > The declare directive has no async clause, so the arg should be > GOMP_ASYNC_SYNC. > > Tested libgomp testsuite on x86_64 with nvptx accelerator. > > OK for trunk? Assuming no objections, committed to trunk as attached. Thanks, - Tom [openacc, libgomp] Use GOMP_ASYNC_SYNC in GOACC_declare 2018-05-09 Tom de Vries <tom@codesourcery.com> PR libgomp/82901 * oacc-parallel.c (GOACC_declare): Use GOMP_ASYNC_SYNC as async argument to GOACC_enter_exit_data. --- libgomp/oacc-parallel.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/libgomp/oacc-parallel.c b/libgomp/oacc-parallel.c index a71b399..f270491 100644 --- a/libgomp/oacc-parallel.c +++ b/libgomp/oacc-parallel.c @@ -524,7 +524,7 @@ GOACC_declare (int device, size_t mapnum, case GOMP_MAP_POINTER: case GOMP_MAP_DELETE: GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_FORCE_DEVICEPTR: @@ -533,19 +533,19 @@ GOACC_declare (int device, size_t mapnum, case GOMP_MAP_ALLOC: if (!acc_is_present (hostaddrs[i], sizes[i])) GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_TO: GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_FROM: kinds[i] = GOMP_MAP_FORCE_FROM; GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_FORCE_PRESENT:
Use GOMP_ASYNC_SYNC in GOACC_declare Fix for PR82901 - "GOACC_declare calls GOACC_enter_exit_data with async arg 0" --- libgomp/oacc-parallel.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/libgomp/oacc-parallel.c b/libgomp/oacc-parallel.c index a8cff9e..d82d500 100644 --- a/libgomp/oacc-parallel.c +++ b/libgomp/oacc-parallel.c @@ -524,7 +524,7 @@ GOACC_declare (int device, size_t mapnum, case GOMP_MAP_POINTER: case GOMP_MAP_DELETE: GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_FORCE_DEVICEPTR: @@ -533,19 +533,19 @@ GOACC_declare (int device, size_t mapnum, case GOMP_MAP_ALLOC: if (!acc_is_present (hostaddrs[i], sizes[i])) GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_TO: GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_FROM: kinds[i] = GOMP_MAP_FORCE_FROM; GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], - &kinds[i], 0, 0); + &kinds[i], GOMP_ASYNC_SYNC, 0); break; case GOMP_MAP_FORCE_PRESENT:
Hi, GOACC_enter_exit_data has this prototype: ... void GOACC_enter_exit_data (int device, size_t mapnum, void **hostaddrs, size_t *sizes, unsigned short *kinds, int async, int num_waits, ...) ... And GOACC_declare calls GOACC_enter_exit_data with async arg zero: ... case GOMP_MAP_DELETE: GOACC_enter_exit_data (device, 1, &hostaddrs[i], &sizes[i], &kinds[i], 0, 0); ... Async arg zero means some async queue (see openacc 2.0a, 2.14.1 "async clause" for more details). The declare directive has no async clause, so the arg should be GOMP_ASYNC_SYNC. Tested libgomp testsuite on x86_64 with nvptx accelerator. OK for trunk? Thanks, - Tom