1
1
openmpi/opal/mca/btl/scif/btl_scif_send.c

300 строки
10 KiB
C
Исходник Обычный вид История

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2014 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "btl_scif.h"
#include "btl_scif_frag.h"
#define BUFFER_FREE(s,e,hbm) (((s) > (e) || ((s) == (e) && !hbm)) ? (s) - (e) : (mca_btl_scif_component.segment_size - (e)))
/* attempt to reserve a contiguous segment from the remote endpoint */
static inline int mca_btl_scif_send_get_buffer (mca_btl_base_endpoint_t *endpoint, size_t size, unsigned char * restrict *dst)
{
/* the high bit helps determine if the buffer is empty or full */
bool hbm = (endpoint->send_buffer.start >> 31) == (endpoint->send_buffer.end >> 31);
const unsigned int segment_size = mca_btl_scif_component.segment_size;
unsigned int start = endpoint->send_buffer.start & ~ (1 << 31);
unsigned int end = endpoint->send_buffer.end & ~ (1 << 31);
unsigned int buffer_free = BUFFER_FREE(start, end, hbm);
#if defined(SCIF_TIMING)
struct timespec ts;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
#endif
/* need space for the fragment + the header */
size += sizeof (mca_btl_scif_frag_hdr_t);
/* check if we need to free up space for this fragment */
if (OPAL_UNLIKELY(buffer_free < size)) {
BTL_VERBOSE(("not enough room for a fragment of size %u. in use buffer segment: {start: %x, end: %x, high bit matches: %d}\n",
(unsigned) size, start, end, (int) hbm));
/* read the current start pointer from the remote peer */
start = endpoint->send_buffer.start = endpoint->send_buffer.startp[0];
start &= ~ (1 << 31);
hbm = (endpoint->send_buffer.start >> 31) == (endpoint->send_buffer.end >> 31);
buffer_free = BUFFER_FREE(start, end, hbm);
opal_atomic_rmb ();
/* if this is the end of the buffer. does the fragment fit? */
if (OPAL_UNLIKELY(buffer_free > 0 && buffer_free < size && start <= end)) {
mca_btl_scif_frag_hdr_t hdr;
hdr.size = buffer_free - sizeof (mca_btl_scif_frag_hdr_t);
hdr.tag = 0xff;
#if defined(SCIF_USE_SEQ)
hdr.seq = endpoint->seq_next++;
((uint64_t *) (endpoint->send_buffer.buffer + end))[0] = *((uint64_t *) &hdr);
#else
((uint32_t *) (endpoint->send_buffer.buffer + end))[0] = *((uint32_t *) &hdr);
#endif
/* toggle the high bit */
end = 64;
endpoint->send_buffer.end = ((endpoint->send_buffer.end & (1 << 31)) ^ (1 << 31)) | end;
hbm = (endpoint->send_buffer.start >> 31) == (endpoint->send_buffer.end >> 31);
buffer_free = BUFFER_FREE(start, end, hbm);
}
if (OPAL_UNLIKELY(buffer_free < size)) {
#if defined(SCIF_TIMING)
SCIF_UPDATE_TIMER(mca_btl_scif_component.aquire_buffer_time, mca_btl_scif_component.aquire_buffer_time_max, ts);
#endif
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_ERR_OUT_OF_RESOURCE;
}
}
BTL_VERBOSE(("writing fragment of size %u to offset %u {start: %x, end: %x} of peer's buffer. free = %u",
(unsigned int) size, end, start, end, buffer_free));
*dst = endpoint->send_buffer.buffer + end;
/* align the buffer on a 64 byte boundary */
end = (end + size + 63) & ~63;
if (OPAL_UNLIKELY(segment_size == end)) {
endpoint->send_buffer.end = ((endpoint->send_buffer.end & (1 << 31)) ^ (1 << 31)) | 64;
} else {
endpoint->send_buffer.end = (endpoint->send_buffer.end & (1 << 31)) | end;
}
#if defined(SCIF_TIMING)
SCIF_UPDATE_TIMER(mca_btl_scif_component.aquire_buffer_time, mca_btl_scif_component.aquire_buffer_time_max, ts);
#endif
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_SUCCESS;
}
static void mark_buffer (struct mca_btl_base_endpoint_t *endpoint)
{
if (endpoint->port_id.node != mca_btl_scif_module.port_id.node) {
/* force the PCIe bus to flush by reading from the remote node */
volatile uint32_t start = endpoint->send_buffer.startp[0]; (void)start;
endpoint->send_buffer.endp[0] = endpoint->send_buffer.end;
endpoint->send_buffer.start = endpoint->send_buffer.startp[0];
} else {
MB();
2015-06-23 20:59:57 -07:00
endpoint->send_buffer.endp[0] = endpoint->send_buffer.end;
}
}
static int mca_btl_scif_send_frag (struct mca_btl_base_endpoint_t *endpoint,
mca_btl_scif_base_frag_t *frag)
{
size_t size = frag->hdr.size;
unsigned char * restrict dst;
BTL_VERBOSE(("btl/scif sending descriptor %p from %d -> %d. length = %" PRIu64, (void *) frag,
OPAL_PROC_MY_NAME.vpid, endpoint->peer_proc->proc_name.vpid, frag->segments[0].seg_len));
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
if (OPAL_LIKELY(OPAL_SUCCESS == mca_btl_scif_send_get_buffer (endpoint, size, &dst))) {
unsigned char * restrict data = (unsigned char * restrict) frag->segments[0].seg_addr.pval;
#if defined(SCIF_TIMING)
struct timespec ts;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
#endif
memcpy (dst + sizeof (frag->hdr), data, frag->segments[0].seg_len);
if (frag->segments[1].seg_len) {
memcpy (dst + sizeof (frag->hdr) + frag->segments[0].seg_len,
frag->segments[1].seg_addr.pval,
frag->segments[1].seg_len);
}
#if defined(SCIF_USE_SEQ)
frag->hdr.seq = endpoint->seq_next++;
/* write the tag to signal the fragment is available */
((uint64_t *) dst)[0] = *((uint64_t *) &frag->hdr);
#else
((uint32_t *) dst)[0] = *((uint32_t *) &frag->hdr);
#endif
opal_atomic_wmb ();
#if defined(SCIF_TIMING)
SCIF_UPDATE_TIMER(mca_btl_scif_component.send_time, mca_btl_scif_component.send_time_max, ts);
#endif
/* fragment is gone */
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
mca_btl_scif_frag_complete (frag, OPAL_SUCCESS);
return 1;
}
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_ERR_OUT_OF_RESOURCE;
}
int mca_btl_scif_send (struct mca_btl_base_module_t *btl,
struct mca_btl_base_endpoint_t *endpoint,
struct mca_btl_base_descriptor_t *descriptor,
mca_btl_base_tag_t tag)
{
mca_btl_scif_base_frag_t *frag = (mca_btl_scif_base_frag_t *) descriptor;
size_t size = frag->segments[0].seg_len + frag->segments[1].seg_len;
int rc;
frag->hdr.tag = tag;
frag->hdr.size = size;
if (OPAL_UNLIKELY(MCA_BTL_SCIF_EP_STATE_CONNECTED != endpoint->state)) {
rc = mca_btl_scif_ep_connect (endpoint);
if (OPAL_UNLIKELY(MCA_BTL_SCIF_EP_STATE_CONNECTED != endpoint->state)) {
/* the receiver was not ready to handle the fragment. queue up the fragment. */
descriptor->des_flags |= MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
opal_list_append (&endpoint->frag_wait_list, (opal_list_item_t *) descriptor);
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_SUCCESS;
}
}
rc = mca_btl_scif_send_frag (endpoint, frag);
if (OPAL_LIKELY(1 == rc)) {
mark_buffer (endpoint);
return 1;
}
/* the receiver was not ready to handle the fragment. queue up the fragment. */
descriptor->des_flags |= MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
opal_list_append (&endpoint->frag_wait_list, (opal_list_item_t *) descriptor);
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_SUCCESS;
}
int mca_btl_scif_sendi (struct mca_btl_base_module_t *btl,
struct mca_btl_base_endpoint_t *endpoint,
struct opal_convertor_t *convertor,
void *header, size_t header_size,
size_t payload_size, uint8_t order,
uint32_t flags, mca_btl_base_tag_t tag,
mca_btl_base_descriptor_t **descriptor)
{
size_t length = (header_size + payload_size);
unsigned char * restrict base;
mca_btl_scif_frag_hdr_t hdr;
size_t max_data;
int rc;
#if defined(SCIF_TIMING)
struct timespec ts;
#endif
assert (length < mca_btl_scif_module.super.btl_eager_limit);
assert (0 == (flags & MCA_BTL_DES_SEND_ALWAYS_CALLBACK));
if (OPAL_UNLIKELY(MCA_BTL_SCIF_EP_STATE_CONNECTED != endpoint->state)) {
rc = mca_btl_scif_ep_connect (endpoint);
if (OPAL_UNLIKELY(MCA_BTL_SCIF_EP_STATE_CONNECTED != endpoint->state)) {
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_ERR_RESOURCE_BUSY;
}
}
rc = mca_btl_scif_send_get_buffer (endpoint, length, &base);
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
if (NULL != descriptor) {
*descriptor = NULL;
}
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_ERR_OUT_OF_RESOURCE;
}
#if defined(SCIF_TIMING)
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
#endif
/* fill in the fragment header (except for the tag) */
hdr.size = length;
hdr.tag = tag;
#if defined(SCIF_USE_SEQ)
hdr.seq = endpoint->seq_next++;
#endif
/* write the match header (with MPI comm/tag/etc. info) */
memcpy (base + sizeof (hdr), header, header_size);
if (payload_size) {
uint32_t iov_count = 1;
struct iovec iov[1];
iov[0].iov_base = base + sizeof (hdr) + header_size;
iov[0].iov_len = payload_size;
/* move the data */
opal_convertor_pack (convertor, iov, &iov_count, &max_data);
assert (max_data == payload_size);
}
#if defined(SCIF_USE_SEQ)
/* signal the remote side that this fragment is available */
((uint64_t *)base)[0] = *((uint64_t *) &hdr);
#else
((uint32_t *)base)[0] = *((uint32_t *) &hdr);
#endif
opal_atomic_wmb ();
mark_buffer (endpoint);
#if defined(SCIF_TIMING)
SCIF_UPDATE_TIMER(mca_btl_scif_component.sendi_time, mca_btl_scif_component.sendi_time_max, ts);
#endif
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_SUCCESS;
}
int mca_btl_scif_progress_send_wait_list (mca_btl_base_endpoint_t *endpoint)
{
mca_btl_scif_base_frag_t *frag;
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
int rc = OPAL_SUCCESS;
while (NULL !=
(frag = (mca_btl_scif_base_frag_t *) opal_list_remove_first (&endpoint->frag_wait_list))) {
rc = mca_btl_scif_send_frag (endpoint, frag);
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
if (OPAL_UNLIKELY(OPAL_SUCCESS > rc)) {
if (OPAL_LIKELY(OPAL_ERR_OUT_OF_RESOURCE == rc)) {
opal_list_prepend (&endpoint->frag_wait_list, (opal_list_item_t *) frag);
} else {
mca_btl_scif_frag_complete (frag, rc);
}
break;
}
}
mark_buffer (endpoint);
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-) WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic. This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
return OPAL_SUCCESS;
}