1
1
openmpi/opal/mca/btl/ugni/btl_ugni_prepare.h

263 строки
9.1 KiB
C
Исходник Обычный вид История

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2011-2014 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2011 UT-Battelle, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#if !defined(MCA_BTL_UGNI_PREPARE_H)
#define MCA_BTL_UGNI_PREPARE_H
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 04:47:28 +04:00
#include "opal_config.h"
#include "btl_ugni.h"
#include "btl_ugni_frag.h"
static inline struct mca_btl_base_descriptor_t *
mca_btl_ugni_prepare_src_send_nodata (struct mca_btl_base_module_t *btl,
mca_btl_base_endpoint_t *endpoint,
uint8_t order, size_t reserve,
uint32_t flags)
{
mca_btl_ugni_base_frag_t *frag = NULL;
(void) MCA_BTL_UGNI_FRAG_ALLOC_RDMA(endpoint, frag);
if (OPAL_UNLIKELY(NULL == frag)) {
return NULL;
}
BTL_VERBOSE(("preparing src for send fragment. size = %u", (unsigned int) reserve));
frag->hdr_size = reserve + sizeof (frag->hdr.send);
frag->segments[0].base.seg_addr.pval = frag->hdr.send_ex.pml_header;
frag->segments[0].base.seg_len = reserve;
frag->segments[1].base.seg_addr.pval = NULL;
frag->segments[1].base.seg_len = 0;
frag->base.des_local = &frag->segments->base;
frag->base.des_local_count = 1;
Per RFC: Remove des_src and des_dst members from the mca_btl_base_segment_t and replace them with des_local and des_remote This change also updates the BTL version to 3.0.0. This commit does not represent the final version of BTL 3.0.0. More changes are coming. In making this change I updated all of the BTLs as well as BTL user's to use the new structure members. Please evaluate your component to ensure the changes are correct. RFC text: This is the first of several BTL interface changes I am proposing for the 1.9/2.0 release series. What: Change naming of btl descriptor members. I propose we change des_src and des_dst (and their associated counts) to be des_local and des_remote. For receive callbacks the des_local member will be used to communicate the segment information to the callback. The proposed change will include updating all of the doxygen in btl.h as well as updating all BTLs and BTL users to use the new naming scheme. Why: My btl usage makes use of both put and get operations on the same descriptor. With the current naming scheme I need to ensure that there is consistency beteen the segments described in des_src and des_dst depending on whether a put or get operation is executed. Additionally, the current naming prevents BTLs that do not require prepare/RMA matched operations (do not set MCA_BTL_FLAGS_RDMA_MATCHED) from executing multiple simultaneous put AND get operations. At the moment the descriptor can only be used with one or the other. The naming change makes it easier for BTL users to setup/modify descriptors for RMA operations as the local segment and remote segment are always in the same member field. The only issue I forsee with this change is that it will require a little more work to move BTL fixes to the 1.8 release series. This commit was SVN r32196.
2014-07-10 20:31:15 +04:00
frag->base.order = order;
frag->base.des_flags = flags;
return &frag->base;
}
static inline struct mca_btl_base_descriptor_t *
mca_btl_ugni_prepare_src_send_inplace (struct mca_btl_base_module_t *btl,
mca_btl_base_endpoint_t *endpoint,
struct opal_convertor_t *convertor,
uint8_t order, size_t reserve, size_t *size,
uint32_t flags)
{
bool use_eager_get = (*size + reserve) > mca_btl_ugni_component.smsg_max_data;
mca_btl_ugni_base_frag_t *frag = NULL;
mca_btl_ugni_reg_t *registration = NULL;
void *data_ptr;
int rc;
opal_convertor_get_current_pointer (convertor, &data_ptr);
(void) MCA_BTL_UGNI_FRAG_ALLOC_RDMA(endpoint, frag);
if (OPAL_UNLIKELY(NULL == frag)) {
return NULL;
}
BTL_VERBOSE(("preparing src for send fragment. size = %u",
(unsigned int)(*size + reserve)));
if (OPAL_UNLIKELY(true == use_eager_get)) {
rc = btl->btl_mpool->mpool_register(btl->btl_mpool, data_ptr, *size, 0,
(mca_mpool_base_registration_t **)&registration);
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 04:47:28 +04:00
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
mca_btl_ugni_frag_return (frag);
return NULL;
}
frag->flags = MCA_BTL_UGNI_FRAG_EAGER | MCA_BTL_UGNI_FRAG_IGNORE;
frag->registration = registration;
frag->segments[1].memory_handle = registration->memory_hdl;
frag->hdr_size = reserve + sizeof (frag->hdr.eager);
frag->segments[0].base.seg_addr.pval = frag->hdr.eager_ex.pml_header;
} else {
frag->hdr_size = reserve + sizeof (frag->hdr.send);
frag->segments[0].base.seg_addr.pval = frag->hdr.send_ex.pml_header;
}
frag->segments[0].base.seg_len = reserve;
frag->segments[1].base.seg_addr.pval = data_ptr;
frag->segments[1].base.seg_len = *size;
frag->base.des_local = &frag->segments->base;
frag->base.des_local_count = 2;
Per RFC: Remove des_src and des_dst members from the mca_btl_base_segment_t and replace them with des_local and des_remote This change also updates the BTL version to 3.0.0. This commit does not represent the final version of BTL 3.0.0. More changes are coming. In making this change I updated all of the BTLs as well as BTL user's to use the new structure members. Please evaluate your component to ensure the changes are correct. RFC text: This is the first of several BTL interface changes I am proposing for the 1.9/2.0 release series. What: Change naming of btl descriptor members. I propose we change des_src and des_dst (and their associated counts) to be des_local and des_remote. For receive callbacks the des_local member will be used to communicate the segment information to the callback. The proposed change will include updating all of the doxygen in btl.h as well as updating all BTLs and BTL users to use the new naming scheme. Why: My btl usage makes use of both put and get operations on the same descriptor. With the current naming scheme I need to ensure that there is consistency beteen the segments described in des_src and des_dst depending on whether a put or get operation is executed. Additionally, the current naming prevents BTLs that do not require prepare/RMA matched operations (do not set MCA_BTL_FLAGS_RDMA_MATCHED) from executing multiple simultaneous put AND get operations. At the moment the descriptor can only be used with one or the other. The naming change makes it easier for BTL users to setup/modify descriptors for RMA operations as the local segment and remote segment are always in the same member field. The only issue I forsee with this change is that it will require a little more work to move BTL fixes to the 1.8 release series. This commit was SVN r32196.
2014-07-10 20:31:15 +04:00
frag->base.order = order;
frag->base.des_flags = flags;
return &frag->base;
}
static inline struct mca_btl_base_descriptor_t *
mca_btl_ugni_prepare_src_send_buffered (struct mca_btl_base_module_t *btl,
mca_btl_base_endpoint_t *endpoint,
struct opal_convertor_t *convertor,
uint8_t order, size_t reserve, size_t *size,
uint32_t flags)
{
bool use_eager_get = (*size + reserve) > mca_btl_ugni_component.smsg_max_data;
mca_btl_ugni_reg_t *registration = NULL;
mca_btl_ugni_base_frag_t *frag = NULL;
uint32_t iov_count = 1;
struct iovec iov;
int rc;
if (OPAL_UNLIKELY(true == use_eager_get)) {
(void) MCA_BTL_UGNI_FRAG_ALLOC_EAGER_SEND(endpoint, frag);
if (OPAL_UNLIKELY(NULL == frag)) {
return NULL;
}
frag->flags = MCA_BTL_UGNI_FRAG_EAGER | MCA_BTL_UGNI_FRAG_IGNORE;
registration = (mca_btl_ugni_reg_t *) frag->base.super.registration;
frag->segments[1].memory_handle = registration->memory_hdl;
frag->hdr_size = reserve + sizeof (frag->hdr.eager);
frag->segments[0].base.seg_addr.pval = frag->hdr.eager_ex.pml_header;
} else {
(void) MCA_BTL_UGNI_FRAG_ALLOC_SMSG(endpoint, frag);
if (OPAL_UNLIKELY(NULL == frag)) {
return NULL;
}
frag->hdr_size = reserve + sizeof (frag->hdr.send);
frag->segments[0].base.seg_addr.pval = frag->hdr.send_ex.pml_header;
}
frag->flags |= MCA_BTL_UGNI_FRAG_BUFFERED;
iov.iov_len = *size;
iov.iov_base = (IOVBASE_TYPE *) frag->base.super.ptr;
rc = opal_convertor_pack (convertor, &iov, &iov_count, size);
if (OPAL_UNLIKELY(rc < 0)) {
mca_btl_ugni_frag_return (frag);
return NULL;
}
frag->segments[0].base.seg_len = reserve;
frag->segments[1].base.seg_addr.pval = frag->base.super.ptr;
frag->segments[1].base.seg_len = *size;
frag->base.des_local = &frag->segments->base;
frag->base.des_local_count = 2;
Per RFC: Remove des_src and des_dst members from the mca_btl_base_segment_t and replace them with des_local and des_remote This change also updates the BTL version to 3.0.0. This commit does not represent the final version of BTL 3.0.0. More changes are coming. In making this change I updated all of the BTLs as well as BTL user's to use the new structure members. Please evaluate your component to ensure the changes are correct. RFC text: This is the first of several BTL interface changes I am proposing for the 1.9/2.0 release series. What: Change naming of btl descriptor members. I propose we change des_src and des_dst (and their associated counts) to be des_local and des_remote. For receive callbacks the des_local member will be used to communicate the segment information to the callback. The proposed change will include updating all of the doxygen in btl.h as well as updating all BTLs and BTL users to use the new naming scheme. Why: My btl usage makes use of both put and get operations on the same descriptor. With the current naming scheme I need to ensure that there is consistency beteen the segments described in des_src and des_dst depending on whether a put or get operation is executed. Additionally, the current naming prevents BTLs that do not require prepare/RMA matched operations (do not set MCA_BTL_FLAGS_RDMA_MATCHED) from executing multiple simultaneous put AND get operations. At the moment the descriptor can only be used with one or the other. The naming change makes it easier for BTL users to setup/modify descriptors for RMA operations as the local segment and remote segment are always in the same member field. The only issue I forsee with this change is that it will require a little more work to move BTL fixes to the 1.8 release series. This commit was SVN r32196.
2014-07-10 20:31:15 +04:00
frag->base.order = order;
frag->base.des_flags = flags;
return &frag->base;
}
static inline struct mca_btl_base_descriptor_t *
mca_btl_ugni_prepare_src_send (struct mca_btl_base_module_t *btl,
mca_btl_base_endpoint_t *endpoint,
struct opal_convertor_t *convertor,
uint8_t order, size_t reserve, size_t *size,
uint32_t flags)
{
bool use_eager_get = (*size + reserve) > mca_btl_ugni_component.smsg_max_data;
bool send_in_place;
void *data_ptr;
if (!(*size)) {
return mca_btl_ugni_prepare_src_send_nodata (btl, endpoint, order, reserve, flags);
}
opal_convertor_get_current_pointer (convertor, &data_ptr);
send_in_place = !(opal_convertor_need_buffers(convertor) ||
(use_eager_get && ((uintptr_t)data_ptr & 3)));
if (send_in_place) {
return mca_btl_ugni_prepare_src_send_inplace (btl, endpoint, convertor, order,
reserve, size, flags);
} else {
return mca_btl_ugni_prepare_src_send_buffered (btl, endpoint, convertor, order,
reserve, size, flags);
}
}
static inline struct mca_btl_base_descriptor_t *
mca_btl_ugni_prepare_src_rdma (struct mca_btl_base_module_t *btl,
mca_btl_base_endpoint_t *endpoint,
mca_mpool_base_registration_t *registration,
struct opal_convertor_t *convertor,
uint8_t order, size_t *size,
uint32_t flags)
{
mca_btl_ugni_base_frag_t *frag;
void *data_ptr;
int rc;
opal_convertor_get_current_pointer (convertor, &data_ptr);
(void) MCA_BTL_UGNI_FRAG_ALLOC_RDMA(endpoint, frag);
if (OPAL_UNLIKELY(NULL == frag)) {
return NULL;
}
/*
* For medium message use FMA protocols and for large message
* use BTE protocols
*/
/* No need to register while using FMA Put (registration is
* non-null in get-- is this always true?) */
if (*size >= mca_btl_ugni_component.ugni_fma_limit || (flags & MCA_BTL_DES_FLAGS_GET)) {
if (NULL == registration) {
rc = btl->btl_mpool->mpool_register(btl->btl_mpool, data_ptr, *size, 0,
(mca_mpool_base_registration_t **) &registration);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
mca_btl_ugni_frag_return (frag);
return NULL;
}
frag->registration = (mca_btl_ugni_reg_t *) registration;
}
frag->segments[0].memory_handle = ((mca_btl_ugni_reg_t *)registration)->memory_hdl;
} else {
memset ((void *) &frag->segments[0].memory_handle, 0,
sizeof (frag->segments[0].memory_handle));
}
if ((flags & MCA_BTL_DES_FLAGS_GET) && (*size & 0x3)) {
memmove (frag->segments[0].extra_bytes, (char *) data_ptr + (*size & ~0x3),
*size & 0x3);
frag->segments[0].extra_byte_count = *size & 0x3;
} else {
frag->segments[0].extra_byte_count = 0;
}
frag->segments[0].base.seg_addr.lval = (uint64_t)(uintptr_t) data_ptr;
frag->segments[0].base.seg_len = *size;
frag->base.des_local = &frag->segments->base;
frag->base.des_local_count = 1;
frag->base.order = order;
frag->base.des_flags = flags;
return &frag->base;
}
#endif