Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

UCT/GDR_COPY: Optimize memory registration in fast-path #10342

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

UCT/GDR_COPY: Optimize memory registration in fast-path #10342

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
56799c5
UCS/RCACHE: Use built-in atomics for rwlock
Artemy-Mellanox Dec 1, 2024
42e80f8
UCP/MM: Optimize mem type fast-path
Artemy-Mellanox Dec 1, 2024
7fbaf35
UCT/GDR_COPY: Optimize memory registration in fast-path
Artemy-Mellanox Dec 1, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
41 changes: 39 additions & 2 deletions src/ucp/core/ucp_mm.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1187,6 +1187,44 @@ ucs_status_t ucp_mem_unmap(ucp_context_h context, ucp_mem_h memh)
return UCS_OK;
}

static ucs_status_t
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If we always use one md for mtype, maybe we can avoid ucp_memh as well, just use uct objects?

ucp_memh_uct_reg_one(ucp_mem_h memh, ucp_md_index_t md_index)
{
uct_md_mem_reg_params_t reg_params = {
.field_mask = UCT_MD_MEM_REG_FIELD_FLAGS,
.flags = UCT_MD_MEM_ACCESS_ALL,
};
ucp_context_h context = memh->context;
ucs_memory_type_t mem_type = memh->mem_type;
void *address = ucp_memh_address(memh);
size_t length = ucp_memh_length(memh);
ucs_status_t status;

if (((context->cache_md_map[mem_type] & UCS_BIT(md_index)) == 0) ||
(context->rcache == NULL)) {
status = uct_md_mem_reg_v2(context->tl_mds[md_index].md, address, length,
&reg_params, &memh->uct[md_index]);
if (status != UCS_OK) {
return status;
}

memh->parent = memh;
} else {
status = ucp_memh_get(context, address, length, mem_type,
UCS_BIT(md_index), UCT_MD_MEM_ACCESS_ALL,
"mem_type", &memh->parent);

if (status != UCS_OK) {
return status;
}

memh->uct[md_index] = memh->parent->uct[md_index];
}

memh->md_map = UCS_BIT(md_index);
return UCS_OK;
}

ucs_status_t ucp_mem_type_reg_buffers(ucp_worker_h worker, void *remote_addr,
size_t length, ucs_memory_type_t mem_type,
ucp_md_index_t md_index, ucp_mem_h memh,
Expand Down Expand Up @@ -1214,8 +1252,7 @@ ucs_status_t ucp_mem_type_reg_buffers(ucp_worker_h worker, void *remote_addr,

tl_md = &context->tl_mds[md_index];
cmpt = context->tl_cmpts[tl_md->cmpt_index].cmpt;
status = ucp_memh_init_uct_reg(context, memh, UCS_BIT(md_index),
UCT_MD_MEM_ACCESS_ALL, "mem_type");
status = ucp_memh_uct_reg_one(memh, md_index);
if (status != UCS_OK) {
goto out;
}
Expand Down
1 change: 1 addition & 0 deletions src/ucs/Makefile.am
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,7 @@ nobase_dist_libucs_la_HEADERS = \
type/param.h \
type/init_once.h \
type/spinlock.h \
type/rwlock.h \
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we move "introducing fast rwlock to ucs" to separate PR, including gtest for functionality and performance measurement (print lock overhead using UCS_TEST_MESSAGE + write in the PR desc how fast is new lock compared to glibc)

type/status.h \
type/thread_mode.h \
type/cpu_set.h \
Expand Down
49 changes: 20 additions & 29 deletions src/ucs/memory/rcache.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@
#include <ucs/sys/string.h>
#include <ucs/arch/cpu.h>
#include <ucs/type/spinlock.h>
#include <ucs/type/rwlock.h>
#include <ucs/vfs/base/vfs_obj.h>
#include <ucm/api/ucm.h>

Expand Down Expand Up @@ -424,7 +425,7 @@ void ucs_mem_region_destroy_internal(ucs_rcache_t *rcache,
UCS_STATS_UPDATE_COUNTER(rcache->stats, UCS_RCACHE_DEREGS, 1);

if (drop_lock) {
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
}

UCS_PROFILE_NAMED_CALL_VOID_ALWAYS("mem_dereg",
Expand All @@ -433,7 +434,7 @@ void ucs_mem_region_destroy_internal(ucs_rcache_t *rcache,
region);

if (drop_lock) {
pthread_rwlock_wrlock(&rcache->pgt_lock);
ucs_rwlock_write_lock(&rcache->pgt_lock);
}
}

Expand Down Expand Up @@ -493,14 +494,14 @@ static inline void ucs_rcache_region_put_internal(ucs_rcache_t *rcache,

/* Destroy region and de-register memory */
if (flags & UCS_RCACHE_REGION_PUT_FLAG_TAKE_PGLOCK) {
pthread_rwlock_wrlock(&rcache->pgt_lock);
ucs_rwlock_write_lock(&rcache->pgt_lock);
}

ucs_mem_region_destroy_internal(rcache, region,
flags & UCS_RCACHE_REGION_PUT_FLAG_TAKE_PGLOCK);

if (flags & UCS_RCACHE_REGION_PUT_FLAG_TAKE_PGLOCK) {
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
}
}

Expand Down Expand Up @@ -649,15 +650,15 @@ static void ucs_rcache_unmapped_callback(ucm_event_type_t event_type,
* no rcache operations are performed to clean it.
*/
if (!(rcache->params.flags & UCS_RCACHE_FLAG_SYNC_EVENTS) &&
!pthread_rwlock_trywrlock(&rcache->pgt_lock)) {
!ucs_rwlock_write_trylock(&rcache->pgt_lock)) {
/* coverity[double_lock] */
ucs_rcache_invalidate_range(rcache, start, end,
UCS_RCACHE_REGION_PUT_FLAG_ADD_TO_GC);
UCS_STATS_UPDATE_COUNTER(rcache->stats, UCS_RCACHE_UNMAPS, 1);
/* coverity[double_lock] */
ucs_rcache_check_inv_queue(rcache, UCS_RCACHE_REGION_PUT_FLAG_ADD_TO_GC);
/* coverity[double_unlock] */
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
return;
}

Expand Down Expand Up @@ -703,11 +704,11 @@ static void ucs_rcache_purge(ucs_rcache_t *rcache)
/* Lock must be held in write mode */
static void ucs_rcache_clean(ucs_rcache_t *rcache)
{
pthread_rwlock_wrlock(&rcache->pgt_lock);
ucs_rwlock_write_lock(&rcache->pgt_lock);
/* coverity[double_lock]*/
ucs_rcache_check_inv_queue(rcache, 0);
ucs_rcache_check_gc_list(rcache, 1);
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
}

/* Lock must be held in write mode */
Expand Down Expand Up @@ -940,7 +941,7 @@ ucs_status_t ucs_rcache_create_region(ucs_rcache_t *rcache, void *address,
ucs_trace_func("rcache=%s, address=%p, length=%zu", rcache->name, address,
length);

pthread_rwlock_wrlock(&rcache->pgt_lock);
ucs_rwlock_write_lock(&rcache->pgt_lock);

retry:
/* Align to page size */
Expand Down Expand Up @@ -1061,7 +1062,7 @@ ucs_status_t ucs_rcache_create_region(ucs_rcache_t *rcache, void *address,
*region_p = region;
out_unlock:
/* coverity[double_unlock]*/
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
return status;
}

Expand All @@ -1082,7 +1083,7 @@ ucs_status_t ucs_rcache_get(ucs_rcache_t *rcache, void *address, size_t length,
ucs_trace_func("rcache=%s, address=%p, length=%zu", rcache->name, address,
length);

pthread_rwlock_rdlock(&rcache->pgt_lock);
ucs_rwlock_read_lock(&rcache->pgt_lock);
UCS_STATS_UPDATE_COUNTER(rcache->stats, UCS_RCACHE_GETS, 1);
if (ucs_queue_is_empty(&rcache->inv_q)) {
pgt_region = UCS_PROFILE_CALL(ucs_pgtable_lookup, &rcache->pgtable,
Expand All @@ -1096,12 +1097,12 @@ ucs_status_t ucs_rcache_get(ucs_rcache_t *rcache, void *address, size_t length,
ucs_rcache_region_lru_get(rcache, region);
*region_p = region;
UCS_STATS_UPDATE_COUNTER(rcache->stats, UCS_RCACHE_HITS_FAST, 1);
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_read_unlock(&rcache->pgt_lock);
return UCS_OK;
}
}
}
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_read_unlock(&rcache->pgt_lock);

/* Fall back to slow version (with rw lock) in following cases:
* - invalidation list not empty
Expand Down Expand Up @@ -1132,7 +1133,7 @@ void ucs_rcache_region_invalidate(ucs_rcache_t *rcache,
comp = ucs_mpool_get(&rcache->mp);
ucs_spin_unlock(&rcache->lock);

pthread_rwlock_wrlock(&rcache->pgt_lock);
ucs_rwlock_write_lock(&rcache->pgt_lock);
if (comp != NULL) {
comp->func = cb;
comp->arg = arg;
Expand All @@ -1145,7 +1146,7 @@ void ucs_rcache_region_invalidate(ucs_rcache_t *rcache,
/* coverity[double_lock] */
ucs_rcache_region_invalidate_internal(rcache, region, 0);
/* coverity[double_unlock] */
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
UCS_STATS_UPDATE_COUNTER(rcache->stats, UCS_RCACHE_PUTS, 1);
}

Expand All @@ -1170,10 +1171,10 @@ static void ucs_rcache_before_fork(void)
* again on-demand.
* - Other use cases shouldn't be affected
*/
pthread_rwlock_wrlock(&rcache->pgt_lock);
ucs_rwlock_write_lock(&rcache->pgt_lock);
/* coverity[double_lock] */
ucs_rcache_invalidate_range(rcache, 0, UCS_PGT_ADDR_MAX, 0);
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_write_unlock(&rcache->pgt_lock);
}
}
pthread_mutex_unlock(&ucs_rcache_global_context.lock);
Expand Down Expand Up @@ -1272,7 +1273,6 @@ static UCS_CLASS_INIT_FUNC(ucs_rcache_t, const ucs_rcache_params_t *params,
{
ucs_status_t status;
size_t mp_obj_size, mp_align;
int ret;
ucs_mpool_params_t mp_params;

if (params->region_struct_size < sizeof(ucs_rcache_region_t)) {
Expand All @@ -1294,16 +1294,10 @@ static UCS_CLASS_INIT_FUNC(ucs_rcache_t, const ucs_rcache_params_t *params,

self->params = *params;

ret = pthread_rwlock_init(&self->pgt_lock, NULL);
if (ret) {
ucs_error("pthread_rwlock_init() failed: %m");
status = UCS_ERR_INVALID_PARAM;
goto err_destroy_stats;
}

ucs_rwlock_init(&self->pgt_lock);
status = ucs_spinlock_init(&self->lock, 0);
if (status != UCS_OK) {
goto err_destroy_rwlock;
goto err_destroy_stats;
}

status = ucs_pgtable_init(&self->pgtable, ucs_rcache_pgt_dir_alloc,
Expand Down Expand Up @@ -1376,8 +1370,6 @@ static UCS_CLASS_INIT_FUNC(ucs_rcache_t, const ucs_rcache_params_t *params,
ucs_pgtable_cleanup(&self->pgtable);
err_destroy_inv_q_lock:
ucs_spinlock_destroy(&self->lock);
err_destroy_rwlock:
pthread_rwlock_destroy(&self->pgt_lock);
err_destroy_stats:
UCS_STATS_NODE_FREE(self->stats);
err_free_name:
Expand Down Expand Up @@ -1408,7 +1400,6 @@ static UCS_CLASS_CLEANUP_FUNC(ucs_rcache_t)
ucs_mpool_cleanup(&self->mp, 1);
ucs_pgtable_cleanup(&self->pgtable);
ucs_spinlock_destroy(&self->lock);
pthread_rwlock_destroy(&self->pgt_lock);
UCS_STATS_NODE_FREE(self->stats);
ucs_free(self->name);
ucs_free(self->distribution);
Expand Down
12 changes: 12 additions & 0 deletions src/ucs/memory/rcache.inl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@
#define UCS_RCACHE_INL_

#include "rcache_int.h"
#include <ucs/profile/profile.h>
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why needed?


static UCS_F_ALWAYS_INLINE int
ucs_rcache_region_test(ucs_rcache_region_t *region, int prot, size_t alignment)
Expand Down Expand Up @@ -80,6 +81,17 @@ ucs_rcache_lookup_unsafe(ucs_rcache_t *rcache, void *address, size_t length,
return region;
}

static UCS_F_ALWAYS_INLINE ucs_rcache_region_t *
ucs_rcache_lookup(ucs_rcache_t *rcache, void *address, size_t length,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we integrate this optimizations into ucs_rcache_get?

size_t alignment, int prot)
{
ucs_rcache_region_t *region;

ucs_rwlock_read_lock(&rcache->pgt_lock);
region = ucs_rcache_lookup_unsafe(rcache, address, length, alignment, prot);
ucs_rwlock_read_unlock(&rcache->pgt_lock);
return region;
}

static UCS_F_ALWAYS_INLINE void
ucs_rcache_region_put_unsafe(ucs_rcache_t *rcache, ucs_rcache_region_t *region)
Expand Down
3 changes: 2 additions & 1 deletion src/ucs/memory/rcache_int.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
#include <ucs/stats/stats.h>
#include <ucs/sys/ptr_arith.h>
#include <ucs/type/spinlock.h>
#include <ucs/type/rwlock.h>


#define ucs_rcache_region_log_lvl(_level, _message, ...) \
Expand Down Expand Up @@ -66,7 +67,7 @@ typedef struct ucs_rcache_distribution {
struct ucs_rcache {
ucs_rcache_params_t params; /**< rcache parameters (immutable) */

pthread_rwlock_t pgt_lock; /**< Protects the page table and all
ucs_rwlock_t pgt_lock; /**< Protects the page table and all
regions whose refcount is 0 */
ucs_pgtable_t pgtable; /**< page table to hold the regions */

Expand Down
4 changes: 2 additions & 2 deletions src/ucs/memory/rcache_vfs.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,9 +54,9 @@ static void ucs_rcache_vfs_show_primitive(void *obj, ucs_string_buffer_t *strb,
{
ucs_rcache_t *rcache = obj;

pthread_rwlock_rdlock(&rcache->pgt_lock);
ucs_rwlock_read_lock(&rcache->pgt_lock);
ucs_vfs_show_primitive(obj, strb, arg_ptr, arg_u64);
pthread_rwlock_unlock(&rcache->pgt_lock);
ucs_rwlock_read_unlock(&rcache->pgt_lock);
}

static void ucs_rcache_vfs_init_regions_distribution(ucs_rcache_t *rcache)
Expand Down
Loading
Loading