/*
* Initializes the mpool module.
*/
void mca_mpool_rgpusm_module_init(mca_mpool_rgpusm_module_t* mpool)
{
mpool->super.mpool_component = &mca_mpool_rgpusm_component.super;
mpool->super.mpool_base = NULL; /* no base .. */
mpool->super.mpool_alloc = NULL;
mpool->super.mpool_realloc = NULL;
mpool->super.mpool_free = mca_mpool_rgpusm_free;
mpool->super.mpool_register = mca_mpool_rgpusm_register;
mpool->super.mpool_find = mca_mpool_rgpusm_find;
mpool->super.mpool_deregister = mca_mpool_rgpusm_deregister;
mpool->super.mpool_release_memory = NULL;
mpool->super.mpool_finalize = mca_mpool_rgpusm_finalize;
mpool->super.mpool_ft_event = mca_mpool_rgpusm_ft_event;
mpool->super.rcache =
mca_rcache_base_module_create(mca_mpool_rgpusm_component.rcache_name);
mpool->super.flags = 0;
mpool->resources.reg_data = NULL;
mpool->resources.sizeof_reg = sizeof(struct mca_mpool_common_cuda_reg_t);
mpool->resources.register_mem = cuda_openmemhandle;
mpool->resources.deregister_mem = cuda_closememhandle;
OBJ_CONSTRUCT(&mpool->reg_list, ompi_free_list_t);
ompi_free_list_init_new(&mpool->reg_list, mpool->resources.sizeof_reg,
opal_cache_line_size,
OBJ_CLASS(mca_mpool_base_registration_t),
0,opal_cache_line_size,
0, -1, 32, NULL);
OBJ_CONSTRUCT(&mpool->lru_list, opal_list_t);
mpool->stat_cache_hit = mpool->stat_cache_miss = mpool->stat_evicted = 0;
mpool->stat_cache_found = mpool->stat_cache_notfound = 0;
mpool->stat_cache_valid = mpool->stat_cache_invalid = 0;
}
/*
* Initializes the mpool module.
*/
void mca_mpool_rdma_module_init(mca_mpool_rdma_module_t* mpool)
{
mpool->super.mpool_component = &mca_mpool_rdma_component.super;
mpool->super.mpool_base = NULL; /* no base .. */
mpool->super.mpool_alloc = mca_mpool_rdma_alloc;
mpool->super.mpool_realloc = mca_mpool_rdma_realloc;
mpool->super.mpool_free = mca_mpool_rdma_free;
mpool->super.mpool_register = mca_mpool_rdma_register;
mpool->super.mpool_find = mca_mpool_rdma_find;
mpool->super.mpool_deregister = mca_mpool_rdma_deregister;
mpool->super.mpool_release_memory = mca_mpool_rdma_release_memory;
if(mca_mpool_rdma_component.print_stats == true)
mpool->super.mpool_finalize = mca_mpool_rdma_finalize;
else
mpool->super.mpool_finalize = NULL;
mpool->super.rcache =
mca_rcache_base_module_create(mca_mpool_rdma_component.rcache_name);
mpool->super.flags = MCA_MPOOL_FLAGS_MPI_ALLOC_MEM;
OBJ_CONSTRUCT(&mpool->reg_list, ompi_free_list_t);
ompi_free_list_init(&mpool->reg_list, mpool->resources.sizeof_reg,
OBJ_CLASS(mca_mpool_base_registration_t), 0, -1, 32,
NULL);
OBJ_CONSTRUCT(&mpool->mru_list, opal_list_t);
mpool->stat_cache_hit = mpool->stat_cache_miss = mpool->stat_evicted = 0;
mpool->stat_cache_found = mpool->stat_cache_notfound = 0;
}
/*
* Initializes the mpool module.
*/
void mca_mpool_rdma_module_init(mca_mpool_rdma_module_t* mpool)
{
mpool->super.mpool_component = &mca_mpool_rdma_component.super;
mpool->super.mpool_base = NULL; /* no base .. */
mpool->super.mpool_alloc = mca_mpool_rdma_alloc;
mpool->super.mpool_realloc = mca_mpool_rdma_realloc;
mpool->super.mpool_free = mca_mpool_rdma_free;
mpool->super.mpool_register = mca_mpool_rdma_register;
mpool->super.mpool_find = mca_mpool_rdma_find;
mpool->super.mpool_deregister = mca_mpool_rdma_deregister;
mpool->super.mpool_release_memory = mca_mpool_rdma_release_memory;
mpool->super.mpool_finalize = mca_mpool_rdma_finalize;
mpool->super.mpool_ft_event = mca_mpool_rdma_ft_event;
mpool->super.rcache =
mca_rcache_base_module_create(mca_mpool_rdma_component.rcache_name);
mpool->super.flags = MCA_MPOOL_FLAGS_MPI_ALLOC_MEM;
OBJ_CONSTRUCT(&mpool->reg_list, ompi_free_list_t);
ompi_free_list_init_new(&mpool->reg_list, mpool->resources.sizeof_reg,
opal_cache_line_size,
OBJ_CLASS(mca_mpool_base_registration_t),
0,opal_cache_line_size,
0, -1, 32, NULL);
OBJ_CONSTRUCT(&mpool->mru_list, opal_list_t);
OBJ_CONSTRUCT(&mpool->gc_list, opal_list_t);
mpool->stat_cache_hit = mpool->stat_cache_miss = mpool->stat_evicted = 0;
mpool->stat_cache_found = mpool->stat_cache_notfound = 0;
/* Set this here (vs in component.c) because
ompi_mpi_leave_pinned* may have been set after MCA params were
read (e.g., by the openib btl) */
mca_mpool_rdma_component.leave_pinned = (int)
(1 == ompi_mpi_leave_pinned || ompi_mpi_leave_pinned_pipeline);
}
static void mca_mpool_grdma_pool_contructor (mca_mpool_grdma_pool_t *pool)
{
memset ((void *)((uintptr_t)pool + sizeof (pool->super)), 0, sizeof (*pool) - sizeof (pool->super));
OBJ_CONSTRUCT(&pool->lru_list, opal_list_t);
OBJ_CONSTRUCT(&pool->gc_list, opal_list_t);
pool->rcache = mca_rcache_base_module_create(mca_mpool_grdma_component.rcache_name);
}
static int init_vader_endpoint (struct mca_btl_base_endpoint_t *ep, struct opal_proc_t *proc, int remote_rank) {
mca_btl_vader_component_t *component = &mca_btl_vader_component;
union vader_modex_t *modex;
size_t msg_size;
int rc;
OBJ_CONSTRUCT(ep, mca_btl_vader_endpoint_t);
ep->peer_smp_rank = remote_rank;
if (remote_rank != MCA_BTL_VADER_LOCAL_RANK) {
OPAL_MODEX_RECV(rc, &component->super.btl_version,
&proc->proc_name, (void **) &modex, &msg_size);
if (OPAL_SUCCESS != rc) {
return rc;
}
/* attatch to the remote segment */
#if OPAL_BTL_VADER_HAVE_XPMEM
if (MCA_BTL_VADER_XPMEM == mca_btl_vader_component.single_copy_mechanism) {
/* always use xpmem if it is available */
ep->segment_data.xpmem.apid = xpmem_get (modex->xpmem.seg_id, XPMEM_RDWR, XPMEM_PERMIT_MODE, (void *) 0666);
ep->segment_data.xpmem.rcache = mca_rcache_base_module_create("vma");
(void) vader_get_registation (ep, modex->xpmem.segment_base, mca_btl_vader_component.segment_size,
MCA_MPOOL_FLAGS_PERSIST, (void **) &ep->segment_base);
} else {
#endif
/* store a copy of the segment information for detach */
ep->segment_data.other.seg_ds = malloc (msg_size);
if (NULL == ep->segment_data.other.seg_ds) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
memcpy (ep->segment_data.other.seg_ds, &modex->seg_ds, msg_size);
ep->segment_base = opal_shmem_segment_attach (ep->segment_data.other.seg_ds);
if (NULL == ep->segment_base) {
return OPAL_ERROR;
}
#if OPAL_BTL_VADER_HAVE_XPMEM
}
#endif
OBJ_CONSTRUCT(&ep->lock, opal_mutex_t);
free (modex);
} else {
/* set up the segment base so we can calculate a virtual to real for local pointers */
ep->segment_base = component->my_segment;
}
ep->fifo = (struct vader_fifo_t *) ep->segment_base;
return OPAL_SUCCESS;
}
static int init_vader_endpoint (struct mca_btl_base_endpoint_t *ep, struct ompi_proc_t *proc, int remote_rank) {
const int fbox_in_offset = MCA_BTL_VADER_LOCAL_RANK - (MCA_BTL_VADER_LOCAL_RANK > remote_rank);
const int fbox_out_offset = remote_rank - (MCA_BTL_VADER_LOCAL_RANK < remote_rank);
mca_btl_vader_component_t *component = &mca_btl_vader_component;
struct vader_modex_t *modex;
size_t msg_size;
int rc;
ep->peer_smp_rank = remote_rank;
if (remote_rank != MCA_BTL_VADER_LOCAL_RANK) {
if (OMPI_SUCCESS != (rc = ompi_modex_recv(&component->super.btl_version,
proc, (void *)&modex, &msg_size))) {
return rc;
}
/* attatch to the remote segment */
#if OMPI_BTL_VADER_HAVE_XPMEM
/* always use xpmem if it is available */
ep->apid = xpmem_get (modex->seg_id, XPMEM_RDWR, XPMEM_PERMIT_MODE, (void *) 0666);
ep->rcache = mca_rcache_base_module_create("vma");
(void) vader_get_registation (ep, modex->segment_base, mca_btl_vader_component.segment_size,
MCA_MPOOL_FLAGS_PERSIST, (void **) &ep->segment_base);
#else
int offset = offsetof (opal_shmem_ds_t, seg_name);
memcpy (&ep->seg_ds, modex->buffer, offset);
memcpy (&ep->seg_ds.seg_base_addr, modex->buffer + offset, sizeof (ep->seg_ds.seg_base_addr));
offset += sizeof (ep->seg_ds.seg_base_addr);
strncpy (ep->seg_ds.seg_name, modex->buffer + offset, OPAL_PATH_MAX);
ep->segment_base = opal_shmem_segment_attach (&ep->seg_ds);
if (NULL == ep->segment_base) {
return rc;
}
#endif
free (modex);
ep->next_fbox_out = 0;
ep->next_fbox_in = 0;
ep->next_sequence = 0;
ep->expected_sequence = 0;
ep->fbox_in = (struct mca_btl_vader_fbox_t * restrict) (ep->segment_base + MCA_BTL_VADER_FIFO_SIZE +
fbox_in_offset * MCA_BTL_VADER_FBOX_PEER_SIZE);
ep->fbox_out = (struct mca_btl_vader_fbox_t * restrict) (component->my_segment + MCA_BTL_VADER_FIFO_SIZE +
fbox_out_offset * MCA_BTL_VADER_FBOX_PEER_SIZE);
} else {
static int mca_btl_scif_setup_rcache (mca_btl_scif_module_t *scif_module)
{
mca_rcache_base_resources_t rcache_resources;
int rc;
/* initialize the grdma rcache */
rcache_resources.cache_name = "scif";
rcache_resources.reg_data = (void *) scif_module;
rcache_resources.sizeof_reg = sizeof (mca_btl_scif_reg_t);
rcache_resources.register_mem = scif_reg_mem;
rcache_resources.deregister_mem = scif_dereg_mem;
scif_module->rcache = mca_rcache_base_module_create ("grdma", scif_module, &rcache_resources);
if (NULL == scif_module->rcache) {
BTL_ERROR(("error creating grdma rcache"));
return OPAL_ERROR;
}
/* setup free lists for fragments. dma fragments will be used for
* rma operations and in-place sends. eager frags will be used for
* buffered sends. */
rc = opal_free_list_init (&scif_module->dma_frags,
sizeof (mca_btl_scif_dma_frag_t), 64,
OBJ_CLASS(mca_btl_scif_dma_frag_t),
128, opal_getpagesize (),
mca_btl_scif_component.scif_free_list_num,
mca_btl_scif_component.scif_free_list_max,
mca_btl_scif_component.scif_free_list_inc,
NULL, 0, NULL, NULL, NULL);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
return rc;
}
rc = opal_free_list_init (&scif_module->eager_frags,
sizeof (mca_btl_scif_eager_frag_t), 8,
OBJ_CLASS(mca_btl_scif_eager_frag_t),
128 + scif_module->super.btl_eager_limit, 64,
mca_btl_scif_component.scif_free_list_num,
mca_btl_scif_component.scif_free_list_max,
mca_btl_scif_component.scif_free_list_inc,
NULL, 0, NULL, NULL, NULL);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
BTL_ERROR(("error creating eager receive fragment free list"));
return rc;
}
return OPAL_SUCCESS;
}
static int init_vader_endpoint (struct mca_btl_base_endpoint_t *ep, struct ompi_proc_t *proc, int remote_rank) {
const int fbox_in_offset = MCA_BTL_VADER_LOCAL_RANK - (MCA_BTL_VADER_LOCAL_RANK > remote_rank);
const int fbox_out_offset = remote_rank - (MCA_BTL_VADER_LOCAL_RANK < remote_rank);
mca_btl_vader_component_t *component = &mca_btl_vader_component;
struct vader_modex_t *modex;
size_t msg_size;
int rc;
ep->peer_smp_rank = remote_rank;
if (remote_rank != MCA_BTL_VADER_LOCAL_RANK) {
if (OMPI_SUCCESS != (rc = ompi_modex_recv(&component->super.btl_version,
proc, (void *)&modex, &msg_size))) {
return rc;
}
ep->apid = xpmem_get (modex->seg_id, XPMEM_RDWR, XPMEM_PERMIT_MODE, (void *) 0666);
ep->rcache = mca_rcache_base_module_create("vma");
ep->next_fbox_out = 0;
ep->next_fbox_in = 0;
/* attatch to the remote segment */
(void) vader_get_registation (ep, modex->segment_base, mca_btl_vader_component.segment_size,
MCA_MPOOL_FLAGS_PERSIST, (void **) &ep->segment_base);
ep->fifo = (struct vader_fifo_t *) ep->segment_base;
ep->fbox_in = ep->segment_base + 4096 + fbox_in_offset * MCA_BTL_VADER_FBOX_PEER_SIZE;
ep->fbox_out = component->my_segment + 4096 + fbox_out_offset * MCA_BTL_VADER_FBOX_PEER_SIZE;
} else {
/* set up the segment base so we can calculate a virtual to real for local pointers */
ep->segment_base = component->my_segment;
ep->fifo = (struct vader_fifo_t *) ep->segment_base;
}
return OMPI_SUCCESS;
}
