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

clean up and make consistent the reporting out from the udapl btl; report out readeable event string instead of just a number

Просмотр исходного кода 
This commit was SVN r16954.
Этот коммит содержится в:
Donald Kerr 2007-12-13 15:32:26 +00:00
родитель f7812baf5b
Коммит d05d3afaed
6 изменённых файлов: 242 добавлений и 64 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

44
ompi/mca/btl/udapl/btl_udapl.c
Просмотреть файл

@ -100,8 +100,8 @@ static int udapl_reg_mr(void *reg_data, void *base, size_t size,
&dat_size, &dat_addr);
if(rc != DAT_SUCCESS) {
opal_show_help("help-mpi-btl-udapl.txt",
"dat_lmr_create DAT_INSUFFICIENT_RESOURCES", true);
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"dat_lmr_create DAT_INSUFFICIENT_RESOURCES", true));
return OMPI_ERR_OUT_OF_RESOURCE;
}
@ -152,8 +152,8 @@ mca_btl_udapl_init(DAT_NAME_PTR ia_name, mca_btl_udapl_module_t* btl)
dat_strerror(rc, (const char**)&major,
(const char**)&minor);
opal_show_help("help-mpi-btl-udapl.txt",
"dat_ia_open fail", true, ia_name, major, minor);
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"dat_ia_open fail", true, ia_name, major, minor));
return OMPI_ERROR;
}
@ -190,21 +190,21 @@ mca_btl_udapl_init(DAT_NAME_PTR ia_name, mca_btl_udapl_module_t* btl)
/* check evd qlen against adapter max */
if (btl->udapl_dto_evd_qlen > (btl->udapl_ia_attr).max_evd_qlen) {
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"evd_qlen adapter max",
true,
"btl_udapl_dto_evd_qlen",
btl->udapl_dto_evd_qlen,
(btl->udapl_ia_attr).max_evd_qlen);
(btl->udapl_ia_attr).max_evd_qlen));
btl->udapl_dto_evd_qlen = btl->udapl_ia_attr.max_evd_qlen;
}
if (btl->udapl_conn_evd_qlen > (btl->udapl_ia_attr).max_evd_qlen) {
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"evd_qlen adapter max",
true,
"btl_udapl_conn_evd_qlen",
btl->udapl_conn_evd_qlen,
(btl->udapl_ia_attr).max_evd_qlen);
(btl->udapl_ia_attr).max_evd_qlen));
btl->udapl_conn_evd_qlen = btl->udapl_ia_attr.max_evd_qlen;
}
@ -257,9 +257,9 @@ mca_btl_udapl_init(DAT_NAME_PTR ia_name, mca_btl_udapl_module_t* btl)
/* by not erroring out here we can try to continue with
* the default endpoint parameter values
*/
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"use default endpoint params",
true);
true));
}
/* Save the port with the address information */
@ -292,12 +292,12 @@ mca_btl_udapl_init(DAT_NAME_PTR ia_name, mca_btl_udapl_module_t* btl)
if (mca_btl_udapl_component.udapl_buffer_alignment !=
DAT_OPTIMAL_ALIGNMENT) {
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"optimal buffer alignment mismatch",
true,
DAT_OPTIMAL_ALIGNMENT,
mca_btl_udapl_component.udapl_buffer_alignment,
DAT_OPTIMAL_ALIGNMENT);
DAT_OPTIMAL_ALIGNMENT));
}
/* initialize free lists */
@ -470,13 +470,14 @@ static int mca_btl_udapl_set_peer_parameters(
udapl_btl->udapl_dto_evd_qlen) {
/* user modified so warn */
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP,
("help-mpi-btl-udapl.txt",
"evd_qlen too low",
true,
"btl_udapl_dto_evd_qlen",
udapl_btl->udapl_dto_evd_qlen,
"btl_udapl_dto_evd_qlen",
potential_dto_evd_qlen);
potential_dto_evd_qlen));
} else {
udapl_btl->udapl_dto_evd_qlen = potential_dto_evd_qlen;
}
@ -540,13 +541,14 @@ static int mca_btl_udapl_set_peer_parameters(
udapl_btl->udapl_conn_evd_qlen) {
/* user modified so warn */
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP,
("help-mpi-btl-udapl.txt",
"evd_qlen too low",
true,
"btl_udapl_conn_evd_qlen",
udapl_btl->udapl_conn_evd_qlen,
"btl_udapl_conn_evd_qlen",
potential_conn_evd_qlen);
potential_conn_evd_qlen));
} else {
udapl_btl->udapl_conn_evd_qlen = potential_conn_evd_qlen;
}
@ -609,13 +611,14 @@ static int mca_btl_udapl_set_peer_parameters(
mca_btl_udapl_component.udapl_timeout) {
/* user modified so warn */
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP,
("help-mpi-btl-udapl.txt",
"connection timeout low",
true,
"btl_udapl_timeout",
mca_btl_udapl_component.udapl_timeout,
"btl_udapl_timeout",
potential_udapl_timeout);
potential_udapl_timeout));
} else {
mca_btl_udapl_component.udapl_timeout =
potential_udapl_timeout;
@ -804,7 +807,8 @@ int mca_btl_udapl_free(
} else if(frag->size == mca_btl_udapl_component.udapl_max_frag_size) {
MCA_BTL_UDAPL_FRAG_RETURN_MAX(btl, frag);
} else {
OPAL_OUTPUT((0, "[%s:%d] mca_btl_udapl_free: invalid descriptor\n", __FILE__,__LINE__));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DIAGNOSE,
("mca_btl_udapl_free: invalid descriptor\n"));
return OMPI_ERR_BAD_PARAM;
}
return OMPI_SUCCESS;
@ -1108,7 +1112,7 @@ int mca_btl_udapl_get(
mca_btl_base_endpoint_t* endpoint,
mca_btl_base_descriptor_t* des)
{
OPAL_OUTPUT((0, "udapl_get\n"));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DEVELOPER, ("udapl_get\n"));
return OMPI_ERR_NOT_IMPLEMENTED;
}

81
ompi/mca/btl/udapl/btl_udapl.h
Просмотреть файл

@ -54,6 +54,8 @@ extern "C" {
struct mca_btl_udapl_component_t {
mca_btl_base_component_1_0_1_t super; /**< base BTL component */
int32_t udapl_verbosity; /**< report out level, see
"Report Out from uDAPL BTL" below for details. */
size_t udapl_num_btls; /**< number of hcas available to the uDAPL component */
size_t udapl_max_btls; /**< maximum number of supported hcas */
struct mca_btl_udapl_module_t **udapl_btls; /**< array of available BTL modules */
@ -161,8 +163,85 @@ struct mca_btl_udapl_reg_t {
typedef struct mca_btl_udapl_reg_t mca_btl_udapl_reg_t;
/**
* Report a uDAPL error - for debugging
* Report Out from uDAPL BTL
*
* - BTL_ERROR() : Use to report out errors from uDAPL BTL. These are
* critical errors which will most likely cause the program to fail so
* this message should always be reported to the user. Defined in
* btl/base/btl_base_error.h.
* Example:
* dat_strerror(rc, (const char**)&major, (const char**)&minor);
* BTL_ERROR(("ERROR: %s %s %s\n", "dat_cr_accept", major, minor));
*
* - BTL_UDAPL_VERBOSE_OUTPUT() : Use to output different levels
* of verbosity to the user. See Note below.
* Example:
* BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DIAGNOSE,
* ("WARNING: don't %s", "jump"));
*
* - BTL_UDAPL_VERBOSE_HELP() : Use output information as defined in
* uDAPL BTL help file (help-mpi-btl-udapl.txt). See Note below.
* Example:
* BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP,
* ("help-mpi-btl-udapl.txt",
* "invalid num rdma segments", true, 22));
*
* Note : - Verbose levels are defined below. These levels are
* controlled by the mca parameter "btl_udapl_verbose".
* The verbose level is set to 10 by default so that critical
* error and useful help information will appear. Which ever value
* this param is set to, those messages as well as any lower level
* verbose messages will be reported.
* - Setting "btl_udapl_verbose" to "-1" will turn off all
* messages reported by the use of BTL_UDAPL_VERBOSE_*().
* - These macros should not be used in a critical path as they
* are always included in the compiled code.
* - These macros rely on the use of paranthesis around the "args"
* value.
*
* Values used with BTL_UDAPL_VERBOSE_*():
*
* - 0: critical user information; should always be reported;
* on by default
* - 10: useful help messages that would be reported from
* "help-mpi-btl-udapl.txt"; accessed from
* BTL_UDAPL_VERBOSE_HELP(); on by default
* - 20: general execution diagnostic information;
* may be useful to user or btl developer
* - 30: basic debugging/diagnostic information
* - 90: useful only to developers
* - 100: other components do not appear to go beyond 100 for verbose
* levels so noting here as the max for future reference
*/
#define VERBOSE_CRITICAL 0
#define VERBOSE_SHOW_HELP 10
#define VERBOSE_INFORM 20
#define VERBOSE_DIAGNOSE 30
#define VERBOSE_DEVELOPER 90
#define BTL_UDAPL_VERBOSE_OUTPUT(verbose_level, args) \
do { \
if (verbose_level <= mca_btl_udapl_component.udapl_verbosity) { \
mca_btl_base_out("[%s]%s[%s:%d:%s] ", \
orte_system_info.nodename, \
ORTE_NAME_PRINT(orte_process_info.my_name), \
__FILE__, __LINE__, __func__); \
mca_btl_base_out args; \
mca_btl_base_out("\n"); \
} \
} while(0);
#define BTL_UDAPL_VERBOSE_HELP(verbose_level, args) \
do { \
if (verbose_level <= mca_btl_udapl_component.udapl_verbosity) { \
opal_show_help args; \
} \
} while(0);
/*
* Report a uDAPL error - for debugging
*/
#if OMPI_ENABLE_DEBUG
extern void mca_btl_udapl_error(DAT_RETURN ret, char* str);

117
ompi/mca/btl/udapl/btl_udapl_component.c
Просмотреть файл

@ -59,6 +59,8 @@ void mca_btl_udapl_frag_progress_pending(mca_btl_udapl_module_t* udapl_btl,
const int connection);
static int mca_btl_udapl_modify_ia_list(DAT_COUNT *num_info_entries,
DAT_PROVIDER_INFO* datinfo);
static const char*
mca_btl_udapl_dat_event_to_string(DAT_EVENT_NUMBER event_number);
mca_btl_udapl_component_t mca_btl_udapl_component = {
@ -91,6 +93,77 @@ mca_btl_udapl_component_t mca_btl_udapl_component = {
};
/*
* Predefined and fixed size structure containing DAT_EVENT values
* and associated string as defined in: "uDAPL:User Direct Access
* Programming Library v1.2 Sept 15, 2004", DAT Collaborative Organization.
*/
static struct mca_btl_udapl_dat_events {
DAT_EVENT_NUMBER value;
const char* name;
} mca_btl_udapl_dat_events[] = {
{ DAT_DTO_COMPLETION_EVENT,
"DAT_DTO_COMPLETION_EVENT" },
{ DAT_RMR_BIND_COMPLETION_EVENT,
"DAT_RMR_BIND_COMPLETION_EVENT" },
{ DAT_CONNECTION_REQUEST_EVENT,
"DAT_CONNECTION_REQUEST_EVENT" },
{ DAT_CONNECTION_EVENT_ESTABLISHED,
"DAT_CONNECTION_EVENT_ESTABLISHED" },
{ DAT_CONNECTION_EVENT_PEER_REJECTED,
"DAT_CONNECTION_EVENT_PEER_REJECTED" },
{ DAT_CONNECTION_EVENT_NON_PEER_REJECTED,
"DAT_CONNECTION_EVENT_NON_PEER_REJECTED" },
{ DAT_CONNECTION_EVENT_ACCEPT_COMPLETION_ERROR,
"DAT_CONNECTION_EVENT_ACCEPT_COMPLETION_ERROR" },
{ DAT_CONNECTION_EVENT_DISCONNECTED,
"DAT_CONNECTION_EVENT_DISCONNECTED" },
{ DAT_CONNECTION_EVENT_BROKEN,
"DAT_CONNECTION_EVENT_BROKEN" },
{ DAT_CONNECTION_EVENT_TIMED_OUT,
"DAT_CONNECTION_EVENT_TIMED_OUT" },
{ DAT_CONNECTION_EVENT_UNREACHABLE,
"DAT_CONNECTION_EVENT_UNREACHABLE" },
{ DAT_ASYNC_ERROR_EVD_OVERFLOW,
"DAT_ASYNC_ERROR_EVD_OVERFLOW" },
{ DAT_ASYNC_ERROR_IA_CATASTROPHIC,
"DAT_ASYNC_ERROR_IA_CATASTROPHIC" },
{ DAT_ASYNC_ERROR_EP_BROKEN,
"DAT_ASYNC_ERROR_EP_BROKEN" },
{ DAT_ASYNC_ERROR_TIMED_OUT,
"DAT_ASYNC_ERROR_TIMED_OUT" },
{ DAT_ASYNC_ERROR_PROVIDER_INTERNAL_ERROR,
"DAT_ASYNC_ERROR_PROVIDER_INTERNAL_ERROR" },
{ DAT_SOFTWARE_EVENT,
"DAT_SOFTWARE_EVENT" }
};
/*
* Function to convert DAT_EVENT_NUMBER into a readable string.
*
* @param event_number (IN) DAT_EVENT_NUMBER value
*
* @return event string or a string indicating
* event number is invalid
*/
static const char *
mca_btl_udapl_dat_event_to_string(DAT_EVENT_NUMBER event_number)
{
int i;
int num_events = (sizeof(mca_btl_udapl_dat_events) /
sizeof(mca_btl_udapl_dat_events[0]));
for (i = 0; i < num_events; i++) {
if (mca_btl_udapl_dat_events[i].value == event_number) {
return (mca_btl_udapl_dat_events[i].name);
}
}
return ("Unknown DAT Event Number");
}
/**
* Report a uDAPL error - for debugging
*/
@ -348,10 +421,11 @@ static int mca_btl_udapl_modify_ia_list(DAT_COUNT *num_info_entries,
/* if if_list not NULL, either not found or user error */
if (opal_argv_count(mca_btl_udapl_component.if_list)) {
char *str = opal_argv_join(mca_btl_udapl_component.if_list, ',');
opal_show_help("help-mpi-btl-udapl.txt", "nonexistent entry",
true, orte_system_info.nodename,
((NULL != mca_btl_udapl_component.if_include) ?
"in" : "ex"), str);
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP,
("help-mpi-btl-udapl.txt", "nonexistent entry",
true, orte_system_info.nodename,
((NULL != mca_btl_udapl_component.if_include) ?
"in" : "ex"), str));
free(str);
}
@ -383,11 +457,10 @@ mca_btl_udapl_component_init (int *num_btl_modules,
mca_btl_udapl_component.if_list = NULL;
if (NULL != mca_btl_udapl_component.if_include &&
NULL != mca_btl_udapl_component.if_exclude) {
opal_show_help("help-mpi-btl-udapl.txt",
"specified include and exclude", true,
mca_btl_udapl_component.if_include,
mca_btl_udapl_component.if_exclude);
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"specified include and exclude", true,
mca_btl_udapl_component.if_include,
mca_btl_udapl_component.if_exclude));
mca_btl_udapl_component.udapl_num_btls = 0;
mca_btl_udapl_modex_send();
return NULL;
@ -898,14 +971,17 @@ int mca_btl_udapl_component_progress()
(mca_btl_base_descriptor_t*)frag);
break;
default:
OPAL_OUTPUT((0, "WARNING unknown frag type: %d\n",
frag->type));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DIAGNOSE,
("WARNING: unknown frag type: %d\n",
frag->type));
}
count++;
break;
default:
OPAL_OUTPUT((0, "WARNING unknown dto event: %d\n",
event.event_number));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DIAGNOSE,
("WARNING: DTO event: %s (%d)\n",
mca_btl_udapl_dat_event_to_string(event.event_number),
event.event_number));
}
}
@ -947,11 +1023,14 @@ int mca_btl_udapl_component_progress()
case DAT_CONNECTION_EVENT_UNREACHABLE:
/* Need to set the BTL endpoint to MCA_BTL_UDAPL_FAILED
See dat_ep_connect documentation pdf pg 198 */
BTL_OUTPUT(("WARNING : Connection event not handled : %d\n",
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_CRITICAL,
("WARNING: connection event not handled : %s (%d)\n",
mca_btl_udapl_dat_event_to_string(event.event_number),
event.event_number));
break;
default:
BTL_ERROR(("ERROR: unknown connection event : %d",
BTL_ERROR(("ERROR: connection event : %s (%d)",
mca_btl_udapl_dat_event_to_string(event.event_number),
event.event_number));
}
}
@ -969,11 +1048,15 @@ int mca_btl_udapl_component_progress()
case DAT_ASYNC_ERROR_EP_BROKEN:
case DAT_ASYNC_ERROR_TIMED_OUT:
case DAT_ASYNC_ERROR_PROVIDER_INTERNAL_ERROR:
BTL_OUTPUT(("WARNING: async event ignored : %d",
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_CRITICAL,
("WARNING: async event ignored : %s (%d)",
mca_btl_udapl_dat_event_to_string(event.event_number),
event.event_number));
break;
default:
BTL_OUTPUT(("WARNING unknown async event: %d\n",
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_CRITICAL,
("WARNING: %s (%d)\n",
mca_btl_udapl_dat_event_to_string(event.event_number),
event.event_number));
}
}

47
ompi/mca/btl/udapl/btl_udapl_endpoint.c
Просмотреть файл

@ -82,9 +82,9 @@ int mca_btl_udapl_endpoint_write_eager(mca_btl_base_endpoint_t* endpoint,
int pad = 0;
uint8_t head = endpoint->endpoint_eager_rdma_remote.head;
size_t size_plus_align = OPAL_ALIGN(
mca_btl_udapl_component.udapl_eager_frag_size,
DAT_OPTIMAL_ALIGNMENT,
size_t);
mca_btl_udapl_component.udapl_eager_frag_size,
DAT_OPTIMAL_ALIGNMENT,
size_t);
/* now that we have the head update it */
MCA_BTL_UDAPL_RDMA_NEXT_INDEX(endpoint->endpoint_eager_rdma_remote.head);
@ -117,7 +117,7 @@ int mca_btl_udapl_endpoint_write_eager(mca_btl_base_endpoint_t* endpoint,
* and then working way back
*/
remote_buf = (char *)(endpoint->endpoint_eager_rdma_remote.base.pval) +
(head * size_plus_align) +
(head * size_plus_align) +
frag->size -
frag->triplet.segment_length;
@ -169,7 +169,7 @@ int mca_btl_udapl_endpoint_send(mca_btl_base_endpoint_t* endpoint,
/* just send it already.. */
if(frag->size ==
mca_btl_udapl_component.udapl_eager_frag_size) {
if(OPAL_THREAD_ADD32(&endpoint->endpoint_eager_rdma_remote.tokens, -1) < 0) {
/* no rdma segment available so either send or queue */
OPAL_THREAD_ADD32(&endpoint->endpoint_eager_rdma_remote.tokens, 1);
@ -348,11 +348,11 @@ int mca_btl_udapl_endpoint_get_params(mca_btl_udapl_module_t* btl,
btl->udapl_max_recv_dtos) {
/* user modified, this will fail and is not acceptable */
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"max_recv_dtos too low",
true,
btl->udapl_max_recv_dtos,
mca_btl_udapl_component.udapl_num_recvs);
mca_btl_udapl_component.udapl_num_recvs));
btl->udapl_max_recv_dtos =
mca_btl_udapl_component.udapl_num_recvs;
@ -392,10 +392,11 @@ int mca_btl_udapl_endpoint_get_params(mca_btl_udapl_module_t* btl,
mca_btl_udapl_module.udapl_max_request_dtos) {
/* user has modified */
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP,
("help-mpi-btl-udapl.txt",
"max_request_dtos too low",
true,
btl->udapl_max_request_dtos, request_dtos);
btl->udapl_max_request_dtos, request_dtos));
} else {
btl->udapl_max_request_dtos = request_dtos;
}
@ -403,11 +404,11 @@ int mca_btl_udapl_endpoint_get_params(mca_btl_udapl_module_t* btl,
if (btl->udapl_max_request_dtos > btl->udapl_ia_attr.max_dto_per_ep) {
/* do not go beyond what is allowed by the system */
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"max_request_dtos system max",
true,
btl->udapl_max_request_dtos,
btl->udapl_ia_attr.max_dto_per_ep);
btl->udapl_ia_attr.max_dto_per_ep));
btl->udapl_max_request_dtos = btl->udapl_ia_attr.max_dto_per_ep;
}
@ -675,8 +676,9 @@ int mca_btl_udapl_endpoint_finish_connect(struct mca_btl_udapl_module_t* btl,
rc = mca_btl_udapl_endpoint_finish_max(ep);
} else {
OPAL_OUTPUT((0, "btl_udapl ERROR invalid EP state %d\n",
ep->endpoint_state));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DIAGNOSE,
("ERROR: invalid EP state %d\n",
ep->endpoint_state));
return OMPI_ERROR;
}
return rc;
@ -685,7 +687,8 @@ int mca_btl_udapl_endpoint_finish_connect(struct mca_btl_udapl_module_t* btl,
}
/* If this point is reached, no matching endpoint was found */
OPAL_OUTPUT((0, "btl_udapl ERROR could not match endpoint\n"));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_DIAGNOSE,
("btl_udapl ERROR could not match endpoint\n"));
return OMPI_ERROR;
}
@ -773,7 +776,7 @@ static int mca_btl_udapl_endpoint_finish_max(mca_btl_udapl_endpoint_t* endpoint)
frag->segment.seg_len + sizeof(mca_btl_udapl_footer_t));
assert(frag->size ==
mca_btl_udapl_component.udapl_eager_frag_size);
rc = dat_ep_post_send(endpoint->endpoint_eager, 1,
&frag->triplet, cookie, DAT_COMPLETION_DEFAULT_FLAG);
if(DAT_SUCCESS != rc) {
@ -1058,7 +1061,7 @@ static int mca_btl_udapl_endpoint_send_eager_rdma(
rdma_connect->rkey =
endpoint->endpoint_eager_rdma_local.reg->rmr_context;
rdma_connect->rdma_start.pval =
(unsigned char*)frag->base.super.ptr;
(unsigned char*)frag->base.super.ptr;
/* send fragment */
rc = mca_btl_udapl_send((mca_btl_base_module_t *)udapl_btl, endpoint,
@ -1102,16 +1105,16 @@ void mca_btl_udapl_endpoint_connect_eager_rdma(
/* NOTE: Need to find a more generic way to check ranges
* for all mca parameters.
*/
opal_show_help("help-mpi-btl-udapl.txt",
BTL_UDAPL_VERBOSE_HELP(VERBOSE_SHOW_HELP, ("help-mpi-btl-udapl.txt",
"invalid num rdma segments",
true,
mca_btl_udapl_component.udapl_eager_rdma_num);
mca_btl_udapl_component.udapl_eager_rdma_num));
goto unlock_rdma_local;
}
/* create space for fragment structures */
alloc_ptr = (char*)malloc(mca_btl_udapl_component.udapl_eager_rdma_num *
sizeof(mca_btl_udapl_frag_eager_rdma_t));
sizeof(mca_btl_udapl_frag_eager_rdma_t));
if(NULL == alloc_ptr) {
goto unlock_rdma_local;
@ -1119,13 +1122,13 @@ void mca_btl_udapl_endpoint_connect_eager_rdma(
/* get size of one fragment's data region */
size_plus_align = OPAL_ALIGN(
mca_btl_udapl_component.udapl_eager_frag_size,
DAT_OPTIMAL_ALIGNMENT, size_t);
mca_btl_udapl_component.udapl_eager_frag_size,
DAT_OPTIMAL_ALIGNMENT, size_t);
/* create and register memory for all rdma segments */
buf = udapl_btl->super.btl_mpool->mpool_alloc(udapl_btl->super.btl_mpool,
(size_plus_align * mca_btl_udapl_component.udapl_eager_rdma_num),
0, 0,
0, 0,
(mca_mpool_base_registration_t**)&endpoint->endpoint_eager_rdma_local.reg);
if(!buf)

6
ompi/mca/btl/udapl/btl_udapl_mca.c
Просмотреть файл

@ -236,6 +236,12 @@ int mca_btl_udapl_register_mca_params(void)
NULL, &mca_btl_udapl_component.if_exclude,
REGSTR_EMPTY_OK), tmp_rc, rc);
CHECK_PARAM_REGISTER_RETURN_VALUE(mca_btl_udapl_reg_int("verbose",
"Verbosity level of the uDAPL BTL (-1 thru 100)",
VERBOSE_SHOW_HELP,
&(mca_btl_udapl_component.udapl_verbosity),
REGINT_NEG_ONE_OK), tmp_rc, rc);
/* register uDAPL module parameters */
CHECK_PARAM_REGISTER_RETURN_VALUE(mca_btl_udapl_reg_int("async_evd_qlen",
"The asynchronous event dispatcher queue length.",

11
ompi/mca/btl/udapl/btl_udapl_proc.c
Просмотреть файл

@ -11,6 +11,7 @@
* All rights reserved.
* Copyright (c) 2006 Sandia National Laboratories. All rights
* reserved.
* Copyright (c) 2007 Sun Microsystems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -129,15 +130,17 @@ mca_btl_udapl_proc_t* mca_btl_udapl_proc_create(ompi_proc_t* ompi_proc)
(void*)&udapl_proc->proc_addrs,
&size);
if(OMPI_SUCCESS != rc) {
opal_output(0, "[%s:%d] ompi_modex_recv failed for peer %s",
__FILE__,__LINE__,ORTE_NAME_PRINT(&ompi_proc->proc_name));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_CRITICAL,
("ompi_modex_recv failed for peer %s",
ORTE_NAME_PRINT(&ompi_proc->proc_name)));
OBJ_RELEASE(udapl_proc);
return NULL;
}
if((size % sizeof(mca_btl_udapl_addr_t)) != 0) {
opal_output(0, "[%s:%d] invalid udapl address for peer %s",
__FILE__,__LINE__,ORTE_NAME_PRINT(&ompi_proc->proc_name));
BTL_UDAPL_VERBOSE_OUTPUT(VERBOSE_CRITICAL,
("invalid udapl address for peer %s",
ORTE_NAME_PRINT(&ompi_proc->proc_name)));
OBJ_RELEASE(udapl_proc);
return NULL;
}