552c9ca5a0
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.
302 строки
11 KiB
C
302 строки
11 KiB
C
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
|
/*
|
|
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2009 The University of Tennessee and The University
|
|
* of Tennessee Research Foundation. All rights
|
|
* reserved.
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
|
* University of Stuttgart. All rights reserved.
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
* All rights reserved.
|
|
* Copyright (c) 2006-2007 Mellanox Technologies. All rights reserved.
|
|
* Copyright (c) 2010-2013 Cisco Systems, Inc. All rights reserved.
|
|
* Copyright (c) 2011 NVIDIA Corporation. All rights reserved.
|
|
* Copyright (c) 2012 Los Alamos National Security, LLC. All rights
|
|
* reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "opal_config.h"
|
|
|
|
#include "opal/class/ompi_free_list.h"
|
|
#include "opal/align.h"
|
|
#include "opal/util/output.h"
|
|
#include "opal/mca/mpool/mpool.h"
|
|
|
|
static void ompi_free_list_construct(ompi_free_list_t* fl);
|
|
static void ompi_free_list_destruct(ompi_free_list_t* fl);
|
|
|
|
OBJ_CLASS_INSTANCE(ompi_free_list_t, opal_atomic_lifo_t,
|
|
ompi_free_list_construct, ompi_free_list_destruct);
|
|
|
|
typedef struct ompi_free_list_item_t ompi_free_list_memory_t;
|
|
|
|
OBJ_CLASS_INSTANCE(ompi_free_list_item_t,
|
|
opal_list_item_t,
|
|
NULL, NULL);
|
|
|
|
static void ompi_free_list_construct(ompi_free_list_t* fl)
|
|
{
|
|
OBJ_CONSTRUCT(&fl->fl_lock, opal_mutex_t);
|
|
OBJ_CONSTRUCT(&fl->fl_condition, opal_condition_t);
|
|
fl->fl_max_to_alloc = 0;
|
|
fl->fl_num_allocated = 0;
|
|
fl->fl_num_per_alloc = 0;
|
|
fl->fl_num_waiting = 0;
|
|
fl->fl_frag_size = sizeof(ompi_free_list_item_t);
|
|
fl->fl_frag_alignment = 0;
|
|
fl->fl_payload_buffer_size=0;
|
|
fl->fl_payload_buffer_alignment=0;
|
|
fl->fl_frag_class = OBJ_CLASS(ompi_free_list_item_t);
|
|
fl->fl_mpool = 0;
|
|
fl->ctx = NULL;
|
|
OBJ_CONSTRUCT(&(fl->fl_allocations), opal_list_t);
|
|
}
|
|
|
|
static void ompi_free_list_destruct(ompi_free_list_t* fl)
|
|
{
|
|
opal_list_item_t *item;
|
|
ompi_free_list_item_t *fl_item;
|
|
ompi_free_list_memory_t *fl_mem;
|
|
|
|
#if 0 && OPAL_ENABLE_DEBUG
|
|
if(opal_list_get_size(&fl->super) != fl->fl_num_allocated) {
|
|
opal_output(0, "ompi_free_list: %d allocated %d returned: %s:%d\n",
|
|
fl->fl_num_allocated, opal_list_get_size(&fl->super),
|
|
fl->super.super.cls_init_file_name, fl->super.super.cls_init_lineno);
|
|
}
|
|
#endif
|
|
|
|
while(NULL != (item = opal_atomic_lifo_pop(&(fl->super)))) {
|
|
fl_item = (ompi_free_list_item_t*)item;
|
|
|
|
/* destruct the item (we constructed it), the underlying memory will be
|
|
* reclaimed when we free the slab (ompi_free_list_memory_t ptr)
|
|
* containing it */
|
|
OBJ_DESTRUCT(fl_item);
|
|
}
|
|
|
|
while(NULL != (item = opal_list_remove_first(&(fl->fl_allocations)))) {
|
|
fl_mem = (ompi_free_list_memory_t*)item;
|
|
|
|
if( NULL != fl->fl_mpool ) {
|
|
fl->fl_mpool->mpool_free(fl->fl_mpool, fl_mem->ptr,
|
|
fl_mem->registration);
|
|
} else if (fl_mem->ptr) {
|
|
free (fl_mem->ptr);
|
|
}
|
|
|
|
/* destruct the item (we constructed it), then free the memory chunk */
|
|
OBJ_DESTRUCT(item);
|
|
free(item);
|
|
}
|
|
|
|
OBJ_DESTRUCT(&fl->fl_allocations);
|
|
OBJ_DESTRUCT(&fl->fl_condition);
|
|
OBJ_DESTRUCT(&fl->fl_lock);
|
|
}
|
|
|
|
int ompi_free_list_init_ex(
|
|
ompi_free_list_t *flist,
|
|
size_t elem_size,
|
|
size_t alignment,
|
|
opal_class_t* elem_class,
|
|
int num_elements_to_alloc,
|
|
int max_elements_to_alloc,
|
|
int num_elements_per_alloc,
|
|
mca_mpool_base_module_t* mpool,
|
|
ompi_free_list_item_init_fn_t item_init,
|
|
void* ctx)
|
|
{
|
|
/* alignment must be more than zero and power of two */
|
|
if(alignment <= 1 || (alignment & (alignment - 1)))
|
|
return OPAL_ERROR;
|
|
|
|
if(elem_size > flist->fl_frag_size)
|
|
flist->fl_frag_size = elem_size;
|
|
flist->fl_frag_alignment = alignment;
|
|
if(elem_class)
|
|
flist->fl_frag_class = elem_class;
|
|
flist->fl_payload_buffer_size=flist->fl_frag_size-
|
|
flist->fl_frag_class->cls_sizeof;
|
|
flist->fl_payload_buffer_alignment=alignment;
|
|
flist->fl_max_to_alloc = max_elements_to_alloc;
|
|
flist->fl_num_allocated = 0;
|
|
flist->fl_num_per_alloc = num_elements_per_alloc;
|
|
flist->fl_mpool = mpool;
|
|
flist->item_init = item_init;
|
|
flist->ctx = ctx;
|
|
if(num_elements_to_alloc)
|
|
return ompi_free_list_grow(flist, num_elements_to_alloc);
|
|
return OPAL_SUCCESS;
|
|
}
|
|
|
|
/* this will replace ompi_free_list_init_ex */
|
|
int ompi_free_list_init_ex_new(
|
|
ompi_free_list_t *flist,
|
|
size_t frag_size,
|
|
size_t frag_alignment,
|
|
opal_class_t* frag_class,
|
|
size_t payload_buffer_size,
|
|
size_t payload_buffer_alignment,
|
|
int num_elements_to_alloc,
|
|
int max_elements_to_alloc,
|
|
int num_elements_per_alloc,
|
|
mca_mpool_base_module_t* mpool,
|
|
ompi_free_list_item_init_fn_t item_init,
|
|
void* ctx)
|
|
{
|
|
/* alignment must be more than zero and power of two */
|
|
if (frag_alignment <= 1 || (frag_alignment & (frag_alignment - 1)))
|
|
return OPAL_ERROR;
|
|
if (0 < payload_buffer_size) {
|
|
if (payload_buffer_alignment <= 1 || (payload_buffer_alignment & (payload_buffer_alignment - 1)))
|
|
return OPAL_ERROR;
|
|
}
|
|
|
|
if (frag_size > flist->fl_frag_size)
|
|
flist->fl_frag_size = frag_size;
|
|
if (frag_class)
|
|
flist->fl_frag_class = frag_class;
|
|
flist->fl_payload_buffer_size=payload_buffer_size;
|
|
flist->fl_max_to_alloc = max_elements_to_alloc;
|
|
flist->fl_num_allocated = 0;
|
|
flist->fl_num_per_alloc = num_elements_per_alloc;
|
|
flist->fl_mpool = mpool;
|
|
flist->fl_frag_alignment = frag_alignment;
|
|
flist->fl_payload_buffer_alignment = payload_buffer_alignment;
|
|
flist->item_init = item_init;
|
|
flist->ctx = ctx;
|
|
if (num_elements_to_alloc)
|
|
return ompi_free_list_grow(flist, num_elements_to_alloc);
|
|
return OPAL_SUCCESS;
|
|
}
|
|
|
|
int ompi_free_list_grow(ompi_free_list_t* flist, size_t num_elements)
|
|
{
|
|
unsigned char *ptr, *mpool_alloc_ptr = NULL, *payload_ptr = NULL;
|
|
ompi_free_list_memory_t *alloc_ptr;
|
|
size_t i, alloc_size, head_size, elem_size = 0;
|
|
mca_mpool_base_registration_t *reg = NULL;
|
|
|
|
if(flist->fl_max_to_alloc > 0)
|
|
if(flist->fl_num_allocated + num_elements > flist->fl_max_to_alloc)
|
|
num_elements = flist->fl_max_to_alloc - flist->fl_num_allocated;
|
|
|
|
if(num_elements == 0)
|
|
return OPAL_ERR_TEMP_OUT_OF_RESOURCE;
|
|
|
|
head_size = (NULL == flist->fl_mpool) ? flist->fl_frag_size:
|
|
flist->fl_frag_class->cls_sizeof;
|
|
head_size = OPAL_ALIGN(head_size, flist->fl_frag_alignment, size_t);
|
|
|
|
/* calculate head allocation size */
|
|
alloc_size = num_elements * head_size + sizeof(ompi_free_list_memory_t) +
|
|
flist->fl_frag_alignment;
|
|
|
|
alloc_ptr = (ompi_free_list_memory_t*)malloc(alloc_size);
|
|
|
|
if(NULL == alloc_ptr)
|
|
return OPAL_ERR_TEMP_OUT_OF_RESOURCE;
|
|
|
|
if (0 != flist->fl_payload_buffer_size) {
|
|
elem_size = OPAL_ALIGN(flist->fl_payload_buffer_size,
|
|
flist->fl_payload_buffer_alignment, size_t);
|
|
|
|
/* elem_size should not be 0 here */
|
|
assert (elem_size > 0);
|
|
|
|
/* allocate the rest from the mpool (or use memalign/malloc) */
|
|
if(flist->fl_mpool != NULL) {
|
|
payload_ptr = mpool_alloc_ptr =
|
|
(unsigned char *) flist->fl_mpool->mpool_alloc(flist->fl_mpool,
|
|
num_elements * elem_size,
|
|
flist->fl_payload_buffer_alignment,
|
|
MCA_MPOOL_FLAGS_CACHE_BYPASS |
|
|
MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM, ®);
|
|
} else {
|
|
#ifdef HAVE_POSIX_MEMALIGN
|
|
posix_memalign ((void **) &mpool_alloc_ptr, flist->fl_payload_buffer_alignment,
|
|
num_elements * elem_size);
|
|
payload_ptr = mpool_alloc_ptr;
|
|
#else
|
|
mpool_alloc_ptr = (unsigned char *) malloc (num_elements * elem_size +
|
|
flist->fl_payload_buffer_alignment);
|
|
payload_ptr = (unsigned char *) OPAL_ALIGN((uintptr_t)mpool_alloc_ptr,
|
|
flist->fl_payload_buffer_alignment,
|
|
uintptr_t);
|
|
#endif
|
|
}
|
|
|
|
if(NULL == mpool_alloc_ptr) {
|
|
free(alloc_ptr);
|
|
return OPAL_ERR_TEMP_OUT_OF_RESOURCE;
|
|
}
|
|
}
|
|
|
|
/* make the alloc_ptr a list item, save the chunk in the allocations list,
|
|
* and have ptr point to memory right after the list item structure */
|
|
OBJ_CONSTRUCT(alloc_ptr, ompi_free_list_item_t);
|
|
opal_list_append(&(flist->fl_allocations), (opal_list_item_t*)alloc_ptr);
|
|
|
|
alloc_ptr->registration = reg;
|
|
alloc_ptr->ptr = mpool_alloc_ptr;
|
|
|
|
ptr = (unsigned char*)alloc_ptr + sizeof(ompi_free_list_memory_t);
|
|
ptr = OPAL_ALIGN_PTR(ptr, flist->fl_frag_alignment, unsigned char*);
|
|
|
|
for(i=0; i<num_elements; i++) {
|
|
ompi_free_list_item_t* item = (ompi_free_list_item_t*)ptr;
|
|
item->registration = reg;
|
|
item->ptr = payload_ptr;
|
|
|
|
OBJ_CONSTRUCT_INTERNAL(item, flist->fl_frag_class);
|
|
|
|
/* run the initialize function if present */
|
|
if(flist->item_init) {
|
|
flist->item_init(item, flist->ctx);
|
|
}
|
|
|
|
opal_atomic_lifo_push(&(flist->super), &(item->super));
|
|
ptr += head_size;
|
|
payload_ptr += elem_size;
|
|
|
|
}
|
|
flist->fl_num_allocated += num_elements;
|
|
return OPAL_SUCCESS;
|
|
}
|
|
|
|
/**
|
|
* This function resize the free_list to contain at least the specified
|
|
* number of elements. We do not create all of them in the same memory
|
|
* segment. Instead we will several time the fl_num_per_alloc elements
|
|
* until we reach the required number of the maximum allowed by the
|
|
* initialization.
|
|
*/
|
|
int
|
|
ompi_free_list_resize_mt(ompi_free_list_t* flist, size_t size)
|
|
{
|
|
ssize_t inc_num;
|
|
int ret = OPAL_SUCCESS;
|
|
|
|
if (flist->fl_num_allocated > size) {
|
|
return OPAL_SUCCESS;
|
|
}
|
|
OPAL_THREAD_LOCK(&((flist)->fl_lock));
|
|
inc_num = (ssize_t)size - (ssize_t)flist->fl_num_allocated;
|
|
while( inc_num > 0 ) {
|
|
ret = ompi_free_list_grow(flist, flist->fl_num_per_alloc);
|
|
if( OPAL_SUCCESS != ret ) break;
|
|
inc_num = (ssize_t)size - (ssize_t)flist->fl_num_allocated;
|
|
}
|
|
OPAL_THREAD_UNLOCK(&((flist)->fl_lock));
|
|
|
|
return ret;
|
|
}
|