ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

opal/common/ucx: Simplify Worker Pool TLS structure

Просмотр исходного кода 
Get rid of unneeded context and memory region identifiers

Signed-off-by: Artem Polyakov <artpol84@gmail.com>
Этот коммит содержится в:
Artem Polyakov 2018-11-28 13:57:37 -08:00
родитель 1e7bf7085d
Коммит 9fb9cfbe8e
3 изменённых файлов: 37 добавлений и 45 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

73
opal/mca/common/ucx/common_ucx_wpool.c
Просмотреть файл

@ -144,7 +144,6 @@ opal_common_ucx_wpool_init(opal_common_ucx_wpool_t *wpool,
return rc; return rc;
} }
wpool->cur_ctxid = wpool->cur_memid = 0;
OBJ_CONSTRUCT(&wpool->mutex, opal_mutex_t); OBJ_CONSTRUCT(&wpool->mutex, opal_mutex_t);
OBJ_CONSTRUCT(&wpool->tls_list, opal_list_t); OBJ_CONSTRUCT(&wpool->tls_list, opal_list_t);
@ -400,8 +399,7 @@ opal_common_ucx_wpctx_create(opal_common_ucx_wpool_t *wpool, int comm_size,
opal_common_ucx_ctx_t *ctx = calloc(1, sizeof(*ctx)); opal_common_ucx_ctx_t *ctx = calloc(1, sizeof(*ctx));
int ret = OPAL_SUCCESS; int ret = OPAL_SUCCESS;
ctx->ctx_id = OPAL_ATOMIC_ADD_FETCH32(&wpool->cur_ctxid, 1); WPOOL_DBG_OUT(_dbg_ctx, "ctx_create: ctx = %p\n", (void*)ctx);
WPOOL_DBG_OUT(_dbg_ctx, "ctx_create: ctx_id = %d\n", (int)ctx->ctx_id);
OBJ_CONSTRUCT(&ctx->mutex, opal_mutex_t); OBJ_CONSTRUCT(&ctx->mutex, opal_mutex_t);
OBJ_CONSTRUCT(&ctx->tls_workers, opal_list_t); OBJ_CONSTRUCT(&ctx->tls_workers, opal_list_t);
@ -553,10 +551,7 @@ int opal_common_ucx_wpmem_create(opal_common_ucx_ctx_t *ctx,
ucs_status_t status; ucs_status_t status;
int ret = OPAL_SUCCESS; int ret = OPAL_SUCCESS;
mem->mem_id = OPAL_ATOMIC_ADD_FETCH32(&ctx->wpool->cur_memid, 1); WPOOL_DBG_OUT(_dbg_mem, "for ctx = %p\n", (void *)ctx);
WPOOL_DBG_OUT(_dbg_mem, "ctx = %p, mem_id = %d\n",
(void *)ctx, (int)mem->mem_id);
mem->released = 0; mem->released = 0;
mem->refcntr = 1; /* application holding this memory handler */ mem->refcntr = 1; /* application holding this memory handler */
@ -820,7 +815,7 @@ static void _common_ucx_tls_cleanup(_tlocal_table_t *tls)
// Cleanup memory table // Cleanup memory table
size = tls->mem_tbl_size; size = tls->mem_tbl_size;
for (i = 0; i < size; i++) { for (i = 0; i < size; i++) {
if (!tls->mem_tbl[i]->mem_id){ if (NULL == tls->mem_tbl[i]->gmem){
continue; continue;
} }
_tlocal_mem_record_cleanup(tls->mem_tbl[i]); _tlocal_mem_record_cleanup(tls->mem_tbl[i]);
@ -830,7 +825,7 @@ static void _common_ucx_tls_cleanup(_tlocal_table_t *tls)
// Cleanup ctx table // Cleanup ctx table
size = tls->ctx_tbl_size; size = tls->ctx_tbl_size;
for (i = 0; i < size; i++) { for (i = 0; i < size; i++) {
if (!tls->ctx_tbl[i]->ctx_id){ if (NULL == tls->ctx_tbl[i]->gctx){
continue; continue;
} }
_tlocal_ctx_record_cleanup(tls->ctx_tbl[i]); _tlocal_ctx_record_cleanup(tls->ctx_tbl[i]);
@ -883,23 +878,22 @@ _tlocal_tls_memtbl_extend(_tlocal_table_t *tbl, size_t append)
static inline _tlocal_ctx_t * static inline _tlocal_ctx_t *
_tlocal_ctx_search(_tlocal_table_t *tls, int ctx_id) _tlocal_ctx_search(_tlocal_table_t *tls, opal_common_ucx_ctx_t *ctx)
{ {
size_t i; size_t i;
for(i=0; i<tls->ctx_tbl_size; i++) { for(i=0; i<tls->ctx_tbl_size; i++) {
if( tls->ctx_tbl[i]->ctx_id == ctx_id){ if (tls->ctx_tbl[i]->gctx == ctx){
return tls->ctx_tbl[i]; return tls->ctx_tbl[i];
} }
} }
WPOOL_DBG_OUT(_dbg_tls, "tls = %p, ctx_id = %d\n", WPOOL_DBG_OUT(_dbg_tls, "tls = %p, ctx = %p\n", (void *)tls, (void*)ctx);
(void *)tls, (int)ctx_id);
return NULL; return NULL;
} }
static int static int
_tlocal_ctx_record_cleanup(_tlocal_ctx_t *ctx_rec) _tlocal_ctx_record_cleanup(_tlocal_ctx_t *ctx_rec)
{ {
if (0 == ctx_rec->ctx_id) { if (NULL == ctx_rec->gctx) {
return OPAL_SUCCESS; return OPAL_SUCCESS;
} }
/* Remove myself from the communication context structure /* Remove myself from the communication context structure
@ -927,11 +921,13 @@ _tlocal_add_ctx(_tlocal_table_t *tls, opal_common_ucx_ctx_t *ctx)
/* Try to find available record in the TLS table /* Try to find available record in the TLS table
* In parallel perform deferred cleanups */ * In parallel perform deferred cleanups */
for (i=0; i<tls->ctx_tbl_size; i++) { for (i=0; i<tls->ctx_tbl_size; i++) {
if (tls->ctx_tbl[i]->gctx->released ) { if (NULL != tls->ctx_tbl[i]->gctx) {
/* Found dirty record, need to clean first */ if (tls->ctx_tbl[i]->gctx->released ) {
_tlocal_ctx_record_cleanup(tls->ctx_tbl[i]); /* Found dirty record, need to clean first */
_tlocal_ctx_record_cleanup(tls->ctx_tbl[i]);
}
} }
if ((0 == tls->ctx_tbl[i]->ctx_id) && (0 > free_idx)) { if ((NULL != tls->ctx_tbl[i]->gctx) && (0 > free_idx)) {
/* Found clean record */ /* Found clean record */
free_idx = i; free_idx = i;
} }
@ -947,7 +943,6 @@ _tlocal_add_ctx(_tlocal_table_t *tls, opal_common_ucx_ctx_t *ctx)
} }
} }
tls->ctx_tbl[free_idx]->ctx_id = ctx->ctx_id;
tls->ctx_tbl[free_idx]->gctx = ctx; tls->ctx_tbl[free_idx]->gctx = ctx;
tls->ctx_tbl[free_idx]->winfo = _wpool_get_idle(tls->wpool, ctx->comm_size); tls->ctx_tbl[free_idx]->winfo = _wpool_get_idle(tls->wpool, ctx->comm_size);
if (NULL == tls->ctx_tbl[free_idx]->winfo) { if (NULL == tls->ctx_tbl[free_idx]->winfo) {
@ -1009,13 +1004,13 @@ static int _tlocal_ctx_connect(_tlocal_ctx_t *ctx_rec, int target)
/* TLS memory management */ /* TLS memory management */
static inline _tlocal_mem_t * static inline _tlocal_mem_t *
_tlocal_search_mem(_tlocal_table_t *tls, int mem_id) _tlocal_search_mem(_tlocal_table_t *tls, opal_common_ucx_wpmem_t *gmem)
{ {
size_t i; size_t i;
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p mem_id = %d\n", WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p mem = %p\n",
(void *)tls, (int)mem_id); (void *)tls, (void*)gmem);
for(i=0; i<tls->mem_tbl_size; i++) { for(i=0; i<tls->mem_tbl_size; i++) {
if( tls->mem_tbl[i]->mem_id == mem_id){ if( tls->mem_tbl[i]->gmem == gmem){
return tls->mem_tbl[i]; return tls->mem_tbl[i];
} }
} }
@ -1066,12 +1061,14 @@ static _tlocal_mem_t *_tlocal_add_mem(_tlocal_table_t *tls,
/* Try to find available spot in the table */ /* Try to find available spot in the table */
for (i=0; i<tls->mem_tbl_size; i++) { for (i=0; i<tls->mem_tbl_size; i++) {
if (tls->mem_tbl[i]->gmem->released) { if (NULL == tls->mem_tbl[i]->gmem) {
/* Found a dirty record. Need to clean it first */ if (tls->mem_tbl[i]->gmem->released) {
_tlocal_mem_record_cleanup(tls->mem_tbl[i]); /* Found a dirty record. Need to clean it first */
break; _tlocal_mem_record_cleanup(tls->mem_tbl[i]);
break;
}
} }
if ((0 == tls->mem_tbl[i]->mem_id) && (0 > free_idx)) { if ((NULL == tls->mem_tbl[i]->gmem) && (0 > free_idx)) {
/* Found a clear record */ /* Found a clear record */
free_idx = i; free_idx = i;
} }
@ -1086,17 +1083,17 @@ static _tlocal_mem_t *_tlocal_add_mem(_tlocal_table_t *tls,
} }
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p\n", (void *)tls); WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p\n", (void *)tls);
} }
tls->mem_tbl[free_idx]->mem_id = mem->mem_id;
tls->mem_tbl[free_idx]->gmem = mem; tls->mem_tbl[free_idx]->gmem = mem;
tls->mem_tbl[free_idx]->mem = calloc(1, sizeof(*tls->mem_tbl[free_idx]->mem)); tls->mem_tbl[free_idx]->mem = calloc(1, sizeof(*tls->mem_tbl[free_idx]->mem));
ctx_rec = _tlocal_ctx_search(tls, mem->ctx->ctx_id); ctx_rec = _tlocal_ctx_search(tls, mem->ctx);
if (NULL == ctx_rec) { if (NULL == ctx_rec) {
// TODO: act accordingly - cleanup // TODO: act accordingly - cleanup
return NULL; return NULL;
} }
WPOOL_DBG_OUT("tls = %p, ctx_id = %d\n", WPOOL_DBG_OUT("tls = %p, ctx = %p\n",
(void *)tls, (int)mem->ctx->ctx_id); (void *)tls, (void*)mem->ctx);
tls->mem_tbl[free_idx]->mem->worker = ctx_rec->winfo; tls->mem_tbl[free_idx]->mem->worker = ctx_rec->winfo;
tls->mem_tbl[free_idx]->mem->rkeys = calloc(mem->ctx->comm_size, tls->mem_tbl[free_idx]->mem->rkeys = calloc(mem->ctx->comm_size,
@ -1162,10 +1159,10 @@ opal_common_ucx_tlocal_fetch_spath(opal_common_ucx_wpmem_t *mem, int target)
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p\n",(void*)tls); WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p\n",(void*)tls);
/* Obtain the worker structure */ /* Obtain the worker structure */
ctx_rec = _tlocal_ctx_search(tls, mem->ctx->ctx_id); ctx_rec = _tlocal_ctx_search(tls, mem->ctx);
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "ctx_id = %d, ctx_rec=%p\n", WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "ctx_id = %p, ctx_rec=%p\n",
(int)mem->ctx->ctx_id, (void *)ctx_rec); (void*)mem->ctx, (void *)ctx_rec);
if (OPAL_UNLIKELY(NULL == ctx_rec)) { if (OPAL_UNLIKELY(NULL == ctx_rec)) {
ctx_rec = _tlocal_add_ctx(tls, mem->ctx); ctx_rec = _tlocal_add_ctx(tls, mem->ctx);
if (NULL == ctx_rec) { if (NULL == ctx_rec) {
@ -1191,9 +1188,9 @@ opal_common_ucx_tlocal_fetch_spath(opal_common_ucx_wpmem_t *mem, int target)
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "ep = %p\n", (void *)ep); WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "ep = %p\n", (void *)ep);
/* Obtain the memory region info */ /* Obtain the memory region info */
mem_rec = _tlocal_search_mem(tls, mem->mem_id); mem_rec = _tlocal_search_mem(tls, mem);
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p mem_rec = %p mem_id = %d\n", WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p mem_rec = %p mem_id = %p\n",
(void *)tls, (void *)mem_rec, (int)mem->mem_id); (void *)tls, (void *)mem_rec, (void*)mem);
if (OPAL_UNLIKELY(mem_rec == NULL)) { if (OPAL_UNLIKELY(mem_rec == NULL)) {
mem_rec = _tlocal_add_mem(tls, mem); mem_rec = _tlocal_add_mem(tls, mem);
WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p mem = %p\n", WPOOL_DBG_OUT(_dbg_tls || _dbg_mem, "tls = %p mem = %p\n",

3
opal/mca/common/ucx/common_ucx_wpool.h
Просмотреть файл

@ -37,12 +37,10 @@ typedef struct {
opal_list_t idle_workers; opal_list_t idle_workers;
opal_list_t active_workers; opal_list_t active_workers;
opal_atomic_int32_t cur_ctxid, cur_memid;
opal_list_t tls_list; opal_list_t tls_list;
} opal_common_ucx_wpool_t; } opal_common_ucx_wpool_t;
typedef struct { typedef struct {
int ctx_id;
opal_mutex_t mutex; opal_mutex_t mutex;
opal_atomic_int32_t refcntr; opal_atomic_int32_t refcntr;
@ -61,7 +59,6 @@ typedef struct {
} opal_common_ucx_ctx_t; } opal_common_ucx_ctx_t;
typedef struct { typedef struct {
int mem_id;
/* reference context to which memory region belongs */ /* reference context to which memory region belongs */
opal_common_ucx_ctx_t *ctx; opal_common_ucx_ctx_t *ctx;

6
opal/mca/common/ucx/common_ucx_wpool_int.h
Просмотреть файл

@ -6,7 +6,6 @@
#include "common_ucx_wpool.h" #include "common_ucx_wpool.h"
typedef struct { typedef struct {
int ctx_id;
opal_common_ucx_ctx_t *gctx; opal_common_ucx_ctx_t *gctx;
opal_common_ucx_winfo_t *winfo; opal_common_ucx_winfo_t *winfo;
} _tlocal_ctx_t; } _tlocal_ctx_t;
@ -17,7 +16,6 @@ typedef struct {
} _mem_info_t; } _mem_info_t;
typedef struct { typedef struct {
int mem_id;
opal_common_ucx_wpmem_t *gmem; opal_common_ucx_wpmem_t *gmem;
_mem_info_t *mem; _mem_info_t *mem;
opal_common_ucx_tlocal_fast_ptrs_t *mem_tls_ptr; opal_common_ucx_tlocal_fast_ptrs_t *mem_tls_ptr;
@ -59,13 +57,13 @@ static int _tlocal_tls_memtbl_extend(_tlocal_table_t *tbl, size_t append);
static _tlocal_table_t* _common_ucx_tls_init(opal_common_ucx_wpool_t *wpool); static _tlocal_table_t* _common_ucx_tls_init(opal_common_ucx_wpool_t *wpool);
static void _common_ucx_tls_cleanup(_tlocal_table_t *tls); static void _common_ucx_tls_cleanup(_tlocal_table_t *tls);
static inline _tlocal_ctx_t *_tlocal_ctx_search(_tlocal_table_t *tls, static inline _tlocal_ctx_t *_tlocal_ctx_search(_tlocal_table_t *tls,
int ctx_id); opal_common_ucx_ctx_t *ctx);
static int _tlocal_ctx_record_cleanup(_tlocal_ctx_t *ctx_rec); static int _tlocal_ctx_record_cleanup(_tlocal_ctx_t *ctx_rec);
static _tlocal_ctx_t *_tlocal_add_ctx(_tlocal_table_t *tls, static _tlocal_ctx_t *_tlocal_add_ctx(_tlocal_table_t *tls,
opal_common_ucx_ctx_t *ctx); opal_common_ucx_ctx_t *ctx);
static int _tlocal_ctx_connect(_tlocal_ctx_t *ctx, int target); static int _tlocal_ctx_connect(_tlocal_ctx_t *ctx, int target);
static inline _tlocal_mem_t *_tlocal_search_mem(_tlocal_table_t *tls, static inline _tlocal_mem_t *_tlocal_search_mem(_tlocal_table_t *tls,
int mem_id); opal_common_ucx_wpmem_t *gmem);
static _tlocal_mem_t *_tlocal_add_mem(_tlocal_table_t *tls, static _tlocal_mem_t *_tlocal_add_mem(_tlocal_table_t *tls,
opal_common_ucx_wpmem_t *mem); opal_common_ucx_wpmem_t *mem);
static int _tlocal_mem_create_rkey(_tlocal_mem_t *mem_rec, ucp_ep_h ep, int target); static int _tlocal_mem_create_rkey(_tlocal_mem_t *mem_rec, ucp_ep_h ep, int target);