1
1
openmpi/opal/mca/btl/ugni/btl_ugni_prepare.h
Nathan Hjelm d4afb16f5a opal: rework mpool and rcache frameworks
This commit rewrites both the mpool and rcache frameworks. Summary of
changes:

 - Before this change a significant portion of the rcache
   functionality lived in mpool components. This meant that it was
   impossible to add a new memory pool to use with rdma networks
   (ugni, openib, etc) without duplicating the functionality of an
   existing mpool component. All the registration functionality has
   been removed from the mpool and placed in the rcache framework.

 - All registration cache mpools components (udreg, grdma, gpusm,
   rgpusm) have been changed to rcache components. rcaches are
   allocated and released in the same way mpool components were.

 - It is now valid to pass NULL as the resources argument when
   creating an rcache. At this time the gpusm and rgpusm components
   support this. All other rcache components require non-NULL
   resources.

 - A new mpool component has been added: hugepage. This component
   supports huge page allocations on linux.

 - Memory pools are now allocated using "hints". Each mpool component
   is queried with the hints and returns a priority. The current hints
   supported are NULL (uses posix_memalign/malloc), page_size=x (huge
   page mpool), and mpool=x.

 - The sm mpool has been moved to common/sm. This reflects that the sm
   mpool is specialized and not meant for any general
   allocations. This mpool may be moved back into the mpool framework
   if there is any objection.

 - The opal_free_list_init arguments have been updated. The unused0
   argument is not used to pass in the registration cache module. The
   mpool registration flags are now rcache registration flags.

 - All components have been updated to make use of the new framework
   interfaces.

As this commit makes significant changes to both the mpool and rcache
frameworks both versions have been bumped to 3.0.0.

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2016-03-14 10:50:41 -06:00

202 строки
6.9 KiB
C

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2011-2015 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
#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].seg_addr.pval = frag->hdr.send_ex.pml_header;
frag->segments[0].seg_len = reserve;
frag->segments[1].seg_addr.pval = NULL;
frag->segments[1].seg_len = 0;
frag->base.des_segments = frag->segments;
frag->base.des_segment_count = 1;
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_module_t *ugni_module = (mca_btl_ugni_module_t *) btl;
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 = ugni_module->rcache->rcache_register (ugni_module->rcache, data_ptr, *size, 0,
MCA_RCACHE_ACCESS_REMOTE_READ,
(mca_rcache_base_registration_t **)&registration);
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->hdr.eager.memory_handle = registration->handle;;
frag->hdr_size = reserve + sizeof (frag->hdr.eager);
frag->segments[0].seg_addr.pval = frag->hdr.eager_ex.pml_header;
} else {
frag->hdr_size = reserve + sizeof (frag->hdr.send);
frag->segments[0].seg_addr.pval = frag->hdr.send_ex.pml_header;
}
frag->segments[0].seg_len = reserve;
frag->segments[1].seg_addr.pval = data_ptr;
frag->segments[1].seg_len = *size;
frag->base.des_segments = frag->segments;
frag->base.des_segment_count = 2;
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->hdr.eager.memory_handle = registration->handle;
frag->hdr_size = reserve + sizeof (frag->hdr.eager);
frag->segments[0].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].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].seg_len = reserve;
frag->segments[1].seg_addr.pval = frag->base.super.ptr;
frag->segments[1].seg_len = *size;
frag->base.des_segments = frag->segments;
frag->base.des_segment_count = 2;
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);
}
}
#endif