ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Add some debugging output and fix some places where the output id and

Просмотр исходного кода 
verbosity level were swapped

This commit was SVN r24740.
Этот коммит содержится в:
Brian Barrett 2011-06-01 17:20:18 +00:00
родитель 8f401a0563
Коммит b778d785fb
4 изменённых файлов: 61 добавлений и 14 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

19
ompi/mca/mtl/portals4/mtl_portals4.c
Просмотреть файл

@ -21,6 +21,8 @@
#include <portals4.h> #include <portals4.h>
#include "orte/util/name_fns.h"
#include "ompi/proc/proc.h"
#include "ompi/mca/mtl/mtl.h" #include "ompi/mca/mtl/mtl.h"
#include "opal/class/opal_list.h" #include "opal/class/opal_list.h"
#include "ompi/runtime/ompi_module_exchange.h" #include "ompi/runtime/ompi_module_exchange.h"
@ -67,6 +69,16 @@ ompi_mtl_portals4_add_procs(struct mca_mtl_base_module_t *mtl,
ptl_process_t *id; ptl_process_t *id;
size_t size; size_t size;
if (procs[i]->proc_arch != ompi_proc_local()->proc_arch) {
opal_output(ompi_mtl_base_output,
"Portals 4 MTL does not support heterogeneous operations.");
opal_output(ompi_mtl_base_output,
"Proc %s architecture %x, mine %x.",
ORTE_NAME_PRINT(&procs[i]->proc_name),
procs[i]->proc_arch, ompi_proc_local()->proc_arch);
return OMPI_ERROR;
}
mtl_peer_data[i] = malloc(sizeof(struct mca_mtl_base_endpoint_t)); mtl_peer_data[i] = malloc(sizeof(struct mca_mtl_base_endpoint_t));
if (NULL == mtl_peer_data[i]) { if (NULL == mtl_peer_data[i]) {
PtlMEUnlink(ompi_mtl_portals4.long_overflow_me_h); PtlMEUnlink(ompi_mtl_portals4.long_overflow_me_h);
@ -97,6 +109,11 @@ ompi_mtl_portals4_add_procs(struct mca_mtl_base_module_t *mtl,
mtl_peer_data[i]->ptl_proc = *id; mtl_peer_data[i]->ptl_proc = *id;
mtl_peer_data[i]->send_count = 0; mtl_peer_data[i]->send_count = 0;
mtl_peer_data[i]->recv_count = 0; mtl_peer_data[i]->recv_count = 0;
opal_output_verbose(25, ompi_mtl_base_output,
"Peer %d: %x,%x", (int) i,
(int) mtl_peer_data[i]->ptl_proc.phys.nid,
(int) mtl_peer_data[i]->ptl_proc.phys.pid);
} }
return OMPI_SUCCESS; return OMPI_SUCCESS;
@ -159,7 +176,7 @@ ompi_mtl_portals4_progress(void)
while (true) { while (true) {
ret = PtlEQGet(ompi_mtl_portals4.eq_h, &ev); ret = PtlEQGet(ompi_mtl_portals4.eq_h, &ev);
if (PTL_OK == ret) { if (PTL_OK == ret) {
OPAL_OUTPUT_VERBOSE((ompi_mtl_base_output, 50, OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output,
"Found event of type %d\n", ev.type)); "Found event of type %d\n", ev.type));
switch (ev.type) { switch (ev.type) {
case PTL_EVENT_GET: case PTL_EVENT_GET:

12
ompi/mca/mtl/portals4/mtl_portals4_component.c
Просмотреть файл

@ -123,6 +123,18 @@ ompi_mtl_portals4_component_open(void)
return OMPI_ERR_NOT_SUPPORTED; return OMPI_ERR_NOT_SUPPORTED;
} }
opal_output_verbose(1, ompi_mtl_base_output,
"Eager limit: %d", (int) ompi_mtl_portals4.eager_limit);
opal_output_verbose(1, ompi_mtl_base_output,
"Short receive blocks: %d", ompi_mtl_portals4.recv_short_num);
opal_output_verbose(1, ompi_mtl_base_output,
"Queue size: %d", ompi_mtl_portals4.queue_size);
opal_output_verbose(1, ompi_mtl_base_output,
"Long protocol: %s",
(ompi_mtl_portals4.protocol == eager) ? "Eager" :
(ompi_mtl_portals4.protocol == rndv) ? "Rendezvous" :
(ompi_mtl_portals4.protocol == triggered) ? "Triggered" : "Other");
ompi_mtl_portals4.ni_h = PTL_INVALID_HANDLE; ompi_mtl_portals4.ni_h = PTL_INVALID_HANDLE;
ompi_mtl_portals4.eq_h = PTL_INVALID_HANDLE; ompi_mtl_portals4.eq_h = PTL_INVALID_HANDLE;
ompi_mtl_portals4.zero_md_h = PTL_INVALID_HANDLE; ompi_mtl_portals4.zero_md_h = PTL_INVALID_HANDLE;

11
ompi/mca/mtl/portals4/mtl_portals4_recv.c
Просмотреть файл

@ -129,6 +129,7 @@ ompi_mtl_portals4_recv_progress(ptl_event_t *ev,
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR; ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR;
} }
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output, "recv completed"));
ptl_request->super.completion_callback(&ptl_request->super); ptl_request->super.completion_callback(&ptl_request->super);
break; break;
@ -162,6 +163,7 @@ ompi_mtl_portals4_recv_progress(ptl_event_t *ev,
if (ompi_mtl_portals4.protocol == triggered) { if (ompi_mtl_portals4.protocol == triggered) {
PtlCTFree(ptl_request->ct_h); PtlCTFree(ptl_request->ct_h);
} }
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output, "recv completed"));
ptl_request->super.completion_callback(&ptl_request->super); ptl_request->super.completion_callback(&ptl_request->super);
break; break;
@ -230,6 +232,7 @@ ompi_mtl_portals4_recv_progress(ptl_event_t *ev,
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR; ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR;
} }
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output, "recv completed"));
ptl_request->super.completion_callback(&ptl_request->super); ptl_request->super.completion_callback(&ptl_request->super);
} else { } else {
@ -338,7 +341,7 @@ ompi_mtl_portals4_irecv(struct mca_mtl_base_module_t* mtl,
remote_proc.phys.nid = PTL_NID_ANY; remote_proc.phys.nid = PTL_NID_ANY;
remote_proc.phys.pid = PTL_PID_ANY; remote_proc.phys.pid = PTL_PID_ANY;
if (ompi_mtl_portals4.protocol == triggered) { if (ompi_mtl_portals4.protocol == triggered) {
printf("Brian broke any_source\n"); abort(); printf("Brian broke any_source with triggered rndv\n"); abort();
} }
} else { } else {
ompi_proc_t* ompi_proc = ompi_comm_peer_lookup( comm, src ); ompi_proc_t* ompi_proc = ompi_comm_peer_lookup( comm, src );
@ -365,6 +368,12 @@ ompi_mtl_portals4_irecv(struct mca_mtl_base_module_t* mtl,
ptl_request->delivery_len = length; ptl_request->delivery_len = length;
ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_SUCCESS; ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_SUCCESS;
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output,
"Recv %d from %x,%x of length %d\n",
endpoint->recv_count,
endpoint->ptl_proc.phys.nid, endpoint->ptl_proc.phys.pid,
(int)length));
if (ompi_mtl_portals4.protocol == triggered && length > ompi_mtl_portals4.eager_limit) { if (ompi_mtl_portals4.protocol == triggered && length > ompi_mtl_portals4.eager_limit) {
ptl_md_t md; ptl_md_t md;

31
ompi/mca/mtl/portals4/mtl_portals4_send.c
Просмотреть файл

@ -37,7 +37,7 @@ ompi_mtl_portals4_short_callback(ptl_event_t *ev, ompi_mtl_portals4_request_t *p
assert(NULL != ptl_request->super.ompi_req); assert(NULL != ptl_request->super.ompi_req);
if (ev->ni_fail_type != PTL_NI_OK) { if (ev->ni_fail_type != PTL_NI_OK) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: short send callback ni_fail_type: %d", "%s:%d: short send callback ni_fail_type: %d",
__FILE__, __LINE__, ev->ni_fail_type); __FILE__, __LINE__, ev->ni_fail_type);
ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR; ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR;
@ -46,6 +46,7 @@ ompi_mtl_portals4_short_callback(ptl_event_t *ev, ompi_mtl_portals4_request_t *p
free(ptl_request->buffer_ptr); free(ptl_request->buffer_ptr);
} }
PtlMDRelease(ptl_request->md_h); PtlMDRelease(ptl_request->md_h);
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output, "send completed"));
ptl_request->super.completion_callback(&ptl_request->super); ptl_request->super.completion_callback(&ptl_request->super);
return OMPI_SUCCESS; return OMPI_SUCCESS;
@ -60,7 +61,7 @@ ompi_mtl_portals4_long_callback(ptl_event_t *ev, struct ompi_mtl_portals4_reques
assert(NULL != ptl_request->super.ompi_req); assert(NULL != ptl_request->super.ompi_req);
if (ev->ni_fail_type != PTL_NI_OK) { if (ev->ni_fail_type != PTL_NI_OK) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: long send callback ni_fail_type: %d", "%s:%d: long send callback ni_fail_type: %d",
__FILE__, __LINE__, ev->ni_fail_type); __FILE__, __LINE__, ev->ni_fail_type);
ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR; ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR;
@ -74,6 +75,7 @@ ompi_mtl_portals4_long_callback(ptl_event_t *ev, struct ompi_mtl_portals4_reques
free(ptl_request->buffer_ptr); free(ptl_request->buffer_ptr);
} }
PtlMDRelease(ptl_request->md_h); PtlMDRelease(ptl_request->md_h);
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output, "send completed"));
ptl_request->super.completion_callback(&ptl_request->super); ptl_request->super.completion_callback(&ptl_request->super);
} }
@ -94,7 +96,7 @@ ompi_mtl_portals4_sync_callback(ptl_event_t *ev, struct ompi_mtl_portals4_reques
assert(NULL != ptl_request->super.ompi_req); assert(NULL != ptl_request->super.ompi_req);
if (ev->ni_fail_type != PTL_NI_OK) { if (ev->ni_fail_type != PTL_NI_OK) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: sync send callback ni_fail_type: %d", "%s:%d: sync send callback ni_fail_type: %d",
__FILE__, __LINE__, ev->ni_fail_type); __FILE__, __LINE__, ev->ni_fail_type);
ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR; ptl_request->super.ompi_req->req_status.MPI_ERROR = OMPI_ERROR;
@ -108,6 +110,7 @@ ompi_mtl_portals4_sync_callback(ptl_event_t *ev, struct ompi_mtl_portals4_reques
free(ptl_request->buffer_ptr); free(ptl_request->buffer_ptr);
} }
PtlMDRelease(ptl_request->md_h); PtlMDRelease(ptl_request->md_h);
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output, "send completed"));
ptl_request->super.completion_callback(&ptl_request->super); ptl_request->super.completion_callback(&ptl_request->super);
} }
@ -144,7 +147,7 @@ ompi_mtl_portals4_short_isend(mca_pml_base_send_mode_t mode, void *start, int le
&md, &md,
&ptl_request->md_h); &ptl_request->md_h);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlMDBind failed: %d", "%s:%d: PtlMDBind failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
return ompi_mtl_portals4_get_error(ret); return ompi_mtl_portals4_get_error(ret);
@ -161,7 +164,7 @@ ompi_mtl_portals4_short_isend(mca_pml_base_send_mode_t mode, void *start, int le
ptl_request, ptl_request,
0); 0);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlPut failed: %d", "%s:%d: PtlPut failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
PtlMDRelease(ptl_request->md_h); PtlMDRelease(ptl_request->md_h);
@ -196,7 +199,7 @@ ompi_mtl_portals4_long_isend(void *start, int length, int contextid, int localra
&md, &md,
&ptl_request->md_h); &ptl_request->md_h);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlMDBind failed: %d", "%s:%d: PtlMDBind failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
return ompi_mtl_portals4_get_error(ret); return ompi_mtl_portals4_get_error(ret);
@ -223,7 +226,7 @@ ompi_mtl_portals4_long_isend(void *start, int length, int contextid, int localra
ptl_request, ptl_request,
&ptl_request->me_h); &ptl_request->me_h);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlMEAppend failed: %d", "%s:%d: PtlMEAppend failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
PtlMDRelease(ptl_request->md_h); PtlMDRelease(ptl_request->md_h);
@ -265,7 +268,7 @@ ompi_mtl_portals4_long_isend(void *start, int length, int contextid, int localra
me.match_bits); me.match_bits);
} }
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlPut failed: %d", "%s:%d: PtlPut failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
PtlMEUnlink(ptl_request->me_h); PtlMEUnlink(ptl_request->me_h);
@ -300,7 +303,7 @@ ompi_mtl_portals4_sync_isend(void *start, int length, int contextid, int localra
&md, &md,
&ptl_request->md_h); &ptl_request->md_h);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlMDBind failed: %d", "%s:%d: PtlMDBind failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
return ompi_mtl_portals4_get_error(ret); return ompi_mtl_portals4_get_error(ret);
@ -323,7 +326,7 @@ ompi_mtl_portals4_sync_isend(void *start, int length, int contextid, int localra
ptl_request, ptl_request,
&ptl_request->me_h); &ptl_request->me_h);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlMEAppend failed: %d", "%s:%d: PtlMEAppend failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
PtlMDRelease(ptl_request->md_h); PtlMDRelease(ptl_request->md_h);
@ -341,7 +344,7 @@ ompi_mtl_portals4_sync_isend(void *start, int length, int contextid, int localra
ptl_request, ptl_request,
(ptl_hdr_data_t)(uintptr_t)ptl_request); (ptl_hdr_data_t)(uintptr_t)ptl_request);
if (PTL_OK != ret) { if (PTL_OK != ret) {
opal_output_verbose(ompi_mtl_base_output, 1, opal_output_verbose(1, ompi_mtl_base_output,
"%s:%d: PtlPut failed: %d", "%s:%d: PtlPut failed: %d",
__FILE__, __LINE__, ret); __FILE__, __LINE__, ret);
PtlMEUnlink(ptl_request->me_h); PtlMEUnlink(ptl_request->me_h);
@ -381,6 +384,12 @@ ompi_mtl_portals4_isend(struct mca_mtl_base_module_t* mtl,
endpoint->send_count++; endpoint->send_count++;
OPAL_OUTPUT_VERBOSE((50, ompi_mtl_base_output,
"Send %d to %x,%x of length %d\n",
endpoint->send_count,
endpoint->ptl_proc.phys.nid, endpoint->ptl_proc.phys.pid,
(int)length));
switch (mode) { switch (mode) {
case MCA_PML_BASE_SEND_STANDARD: case MCA_PML_BASE_SEND_STANDARD:
case MCA_PML_BASE_SEND_READY: case MCA_PML_BASE_SEND_READY: