Improve pipeline (cons and prod in //)

This commit is contained in:
Thomas Preud'homme 2011-05-10 13:43:55 +02:00
parent f01db158c2
commit f05cfdcd92
19 changed files with 392 additions and 284 deletions

View File

@ -13,7 +13,7 @@ COMMDIR:=communication
# Compilation flags
# I know -finline-functions and -finline-functions-called-once are enabled by
# -O3 but I did this in case gcc behaviour change one day
CFLAGS:=-c -g -O3 -finline-functions -finline-functions-called-once -Wall -Werror
CFLAGS:=-c -g -O3 -finline-functions -finline-functions-called-once -Wall -Wextra -Werror
LDFLAGS:=-L$(LIBDIR) -lpthread -ldl
# Executables

View File

@ -12,9 +12,9 @@ struct channel
{
void * volatile buf[2 * BUF_SIZE / sizeof(void *)] __attribute__ ((aligned (CACHE_LINE_SIZE)));
int unused[20] __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile int state __attribute__ ((aligned (CACHE_LINE_SIZE)));
int sender_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
int receiver_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile unsigned int state:1 __attribute__ ((aligned (CACHE_LINE_SIZE)));
unsigned int sender_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
unsigned int receiver_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
};

View File

@ -22,8 +22,8 @@ struct lvl_2
struct channel
{
struct lvl_2 queue[SLOTS] __attribute__ ((aligned (CACHE_LINE_SIZE)));
int head __attribute__ ((aligned(CACHE_LINE_SIZE)));
int tail __attribute__ ((aligned(CACHE_LINE_SIZE)));
unsigned int head __attribute__ ((aligned(CACHE_LINE_SIZE)));
unsigned int tail __attribute__ ((aligned(CACHE_LINE_SIZE)));
};
__BEGIN_DECLS
@ -31,7 +31,7 @@ __BEGIN_DECLS
// TODO: Make it send only one data
static inline void send(struct channel *channel, void **addr)
{
static __thread int chkidx = 0;
static __thread unsigned int chkidx = 0;
// If all slots are full, spin
if (!chkidx)

View File

@ -19,8 +19,8 @@
struct channel
{
void * volatile *shared_space;
int head __attribute__ ((aligned (CACHE_LINE_SIZE)));
int tail __attribute__ ((aligned (CACHE_LINE_SIZE)));
unsigned int head __attribute__ ((aligned (CACHE_LINE_SIZE)));
unsigned int tail __attribute__ ((aligned (CACHE_LINE_SIZE)));
};
__BEGIN_DECLS
@ -29,7 +29,7 @@ extern int adjust_slip(struct channel *channel);
static inline void send(struct channel *channel, void **addr)
{
static __thread int nb_iter = 0;
static __thread unsigned int nb_iter = 0;
assert(addr != NULL);
if (nb_iter == ADJUST_FREQ)

View File

@ -14,8 +14,8 @@
struct channel
{
void * volatile *shared_space;
volatile int cons_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile int prod_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile unsigned int cons_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile unsigned int prod_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
};
__BEGIN_DECLS

View File

@ -9,22 +9,22 @@
struct control
{
volatile int read;
volatile int write;
volatile unsigned int read;
volatile unsigned int write;
};
struct cons
{
int localWrite;
int nextRead;
int rBatch;
unsigned int localWrite;
unsigned int nextRead;
unsigned int rBatch;
};
struct prod
{
int localRead;
int nextWrite;
int wBatch;
unsigned int localRead;
unsigned int nextWrite;
unsigned int wBatch;
};
@ -38,13 +38,15 @@ struct channel
__BEGIN_DECLS
extern const int batchSize;
extern const unsigned int batchSize;
static inline void send(struct channel *channel, void **addr)
{
while (1)
{
int afterNextWrite = (channel->prod.nextWrite + 1) % SHARED_SPACE_VOIDPTR;
unsigned int afterNextWrite;
afterNextWrite = (channel->prod.nextWrite + 1) % SHARED_SPACE_VOIDPTR;
if (afterNextWrite == channel->prod.localRead)
{
if (afterNextWrite == channel->ctrl.read)

View File

@ -7,7 +7,8 @@ struct channel
{
};
static inline void send(struct channel *channel, void **addr) {}
static inline void send(struct channel *channel __attribute__ ((unused)),
void **addr __attribute__ ((unused))) {}
__END_DECLS

View File

@ -15,7 +15,7 @@ __BEGIN_DECLS
static inline void send(struct channel *channel, void **addr)
{
int nb_read;
unsigned int nb_read;
void *addr_ptr;
nb_read = 0;

View File

@ -10,8 +10,8 @@
struct channel
{
void * volatile *shared_space;
volatile int cons_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile int prod_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile unsigned int cons_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
volatile unsigned int prod_idx __attribute__ ((aligned (CACHE_LINE_SIZE)));
};
@ -19,8 +19,8 @@ __BEGIN_DECLS
static inline void send(struct channel *channel, void **addr)
{
static __thread int local_cons_idx = 0;
int local_prod, next_prod;
static __thread unsigned int local_cons_idx = 0;
unsigned int local_prod, next_prod;
local_prod = channel->prod_idx;
next_prod = (local_prod + 1) % SHARED_SPACE_VOIDPTR;

View File

@ -17,7 +17,7 @@ static int lines_per_calc;
int init_calc(int lines_per_calc_param)
{
int i;
unsigned int i;
assert(!(CACHE_SIZE % (lines_per_calc_param * CACHE_LINE_SIZE)));
lines_per_calc = lines_per_calc_param;

View File

@ -59,7 +59,7 @@ void *recv_one_data(struct channel *channel)
*/
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
{
int nb_read;
size_t nb_read;
nb_read = 0;
while (channel->state && nb_read < count)
@ -76,13 +76,7 @@ ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
n = channel->receiver_idx + (BUF_SIZE / sizeof(void *));
channel->receiver_idx = n % ((2 * BUF_SIZE) / sizeof(void *));
for(; i < n; i++)
{
/*
* The behaviour of this is not documented but we know
* the values inside buf won't change during this affectation
*/
*buf++ = channel->buf[i];
}
nb_read += BUF_SIZE / sizeof(void *);
channel->state = 0;
}

View File

@ -51,16 +51,17 @@ void *recv_one_data(struct channel *channel)
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
{
int n;
size_t n;
n = 0;
// If all slots are empty, spin
while (channel->queue[channel->head].flag)
{
unsigned int i;
// Dequeue a chunk of data items
memcpy(buf, (const void *)
channel->queue[channel->head].chunk,
SUB_SLOTS * sizeof(*buf));
for(i = 0; i < SUB_SLOTS; i++)
*buf++ = channel->queue[channel->head].chunk[i];
n += SUB_SLOTS;
channel->queue[channel->head].flag = 0;
channel->head = (channel->head + 1) % SLOTS;

View File

@ -16,7 +16,7 @@ void *create_comm_channel(void)
{
if (!posix_memalign((void *) &channel->shared_space, CACHE_LINE_SIZE, SHARED_SPACE_SIZE))
{
int i;
unsigned int i;
channel->head = 0;
channel->tail = 0;
@ -39,9 +39,9 @@ int destroy_comm_channel(void *channel)
int adjust_slip(struct channel *channel)
{
int dist, dist_old, unused;
unsigned int dist, dist_old, unused;
puts("adjust_slip is called"); /* Must be removed after calibration */
//puts("adjust_slip is called"); /* Must be removed after calibration */
unused = 0;
dist = (channel->head + SHARED_SPACE_VOIDPTR - channel->tail) % SHARED_SPACE_VOIDPTR;
if (dist < DANGER)
@ -49,7 +49,7 @@ int adjust_slip(struct channel *channel)
dist_old = 0;
do
{
int i;
unsigned int i;
dist_old = dist;
@ -89,14 +89,12 @@ void *recv_one_data(struct channel *channel)
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
{
int n, next_adjust;
unsigned int n;
static __thread int nb_iter = 0;
next_adjust = ADJUST_FREQ - nb_iter;
for(n = 0; n < count; n++)
for(n = 0; n < count; n++, nb_iter++)
{
/* if ((nb_iter + n) % ADJUST_FREQ == 0) */
if (n && (n % next_adjust == ADJUST_FREQ))
if (nb_iter % ADJUST_FREQ == 0)
{
adjust_slip(channel);
nb_iter = 0;
@ -112,6 +110,5 @@ ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
channel->shared_space[channel->tail] = NULL;
channel->tail = (channel->tail + 1) % SHARED_SPACE_VOIDPTR;
}
nb_iter = (nb_iter + n) % ADJUST_FREQ;
return n;
}

View File

@ -35,7 +35,7 @@ int destroy_comm_channel(void *channel)
void *recv_one_data(struct channel *channel)
{
int cons_idx;
unsigned int cons_idx;
void *result;
cons_idx = channel->cons_idx;
@ -48,7 +48,7 @@ void *recv_one_data(struct channel *channel)
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
{
int n, cons_idx;
unsigned int n, cons_idx;
n = 0;
for(cons_idx = channel->cons_idx; cons_idx != channel->prod_idx; cons_idx = (cons_idx + 1) % SHARED_SPACE_VOIDPTR, channel->cons_idx = cons_idx)

View File

@ -8,7 +8,7 @@
#include <specific_comm.h>
const int batchSize = 50; // Check with SHARED_SPACE_SIZE
const unsigned int batchSize = 64; // Check with SHARED_SPACE_SIZE
void *create_comm_channel(void)
{
@ -68,7 +68,7 @@ void *recv_one_data(struct channel *channel)
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
{
int n;
unsigned int n;
for(n = 0; n < count; n++)
{
@ -79,10 +79,6 @@ ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
channel->cons.localWrite = channel->ctrl.write;
}
/*
* The behaviour of this is not documented but we know
* the values inside buf won't change during this affectation
*/
*buf++ = channel->shared_space[channel->cons.nextRead];
channel->cons.nextRead = (channel->cons.nextRead + 1) % SHARED_SPACE_VOIDPTR;
channel->cons.rBatch++;

View File

@ -16,7 +16,7 @@ void *create_comm_channel(void)
return (void *) &store_var;
}
int destroy_comm_channel(void *unused)
int destroy_comm_channel(void *unused __attribute__ ((unused)))
{
return 0;
}
@ -29,7 +29,7 @@ int destroy_comm_channel(void *unused)
* @warning recv_one_data should not be used in conjonction of
* recv_some_data
*/
void *recv_one_data(struct channel *channel)
void *recv_one_data(struct channel *channel __attribute__ ((unused)))
{
return NULL;
}
@ -43,7 +43,9 @@ void *recv_one_data(struct channel *channel)
* recv_one_data
* @warning count must be a multiple of BUF_SIZE
*/
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
ssize_t recv_some_data(struct channel *channel __attribute__ ((unused)),
void **buf __attribute__ ((unused)),
size_t count __attribute__ ((unused)))
{
return count;
}

View File

@ -39,7 +39,8 @@ int destroy_comm_channel(void *channel)
void *recv_one_data(struct channel *channel)
{
void *result, **res_ptr;
int n, nb_read;
int n;
unsigned int nb_read;
nb_read = 0;
res_ptr = &result;

View File

@ -36,7 +36,7 @@ int destroy_comm_channel(void *channel)
void *recv_one_data(struct channel *channel)
{
void *result;
int cons_idx, prod_idx;
unsigned int cons_idx, prod_idx;
cons_idx = channel->cons_idx;
prod_idx = channel->prod_idx;
@ -55,7 +55,7 @@ void *recv_one_data(struct channel *channel)
ssize_t recv_some_data(struct channel *channel, void **buf, size_t count)
{
int n, cons_idx, prod_idx;
unsigned int n, cons_idx, prod_idx;
n = 0;
cons_idx = channel->cons_idx;

View File

@ -18,22 +18,33 @@
#include <specific_comm.h>
//#define MAX_BLOCK_ENTRIES (page_size / sizeof(void *))
#define MAX_BLOCK_ENTRIES (4096 * CACHE_LINE_SIZE / sizeof(void *)) // Must be a multiple of BUF_SIZE
//#define MAX_BLOCK_ENTRIES ((BUF_SIZE * 4 + page_size) & ~(page_size - 1)) // Big buffer size is not a good idea
/* Must be a multiple of BUF_SIZE */
#define MAX_BLOCK_ENTRIES (2048 * CACHE_LINE_SIZE / sizeof(void *))
/*#define MAX_BLOCK_ENTRIES ((BUF_SIZE * 4 + page_size) & ~(page_size - 1)) // Big buffer size is not a good idea */
#define toString(x) doStringification(x)
#define doStringification(x) #x
#define MIN(x,y) ((x < y) ? x : y)
#define WORDS_PER_LINE (CACHE_LINE_SIZE / sizeof(uintptr_t))
#define PROD 1
#define CONS 2
#define SOURCE 0 /* Initial producer (calling do_calc()) */
#define INTERM 1
#define SINK 2 /* Final consumer (doing the check) */
typedef struct prod_cons_thread
typedef struct data_xchg
{
void *prod_comm_channel;
void *cons_comm_channel;
int flags; // PROD, CONS or both
int cpu_binding; // id of the CPU to run the thread on
} prod_cons_thread_t;
void ** volatile data_buf[2][MAX_BLOCK_ENTRIES];
volatile int filled_buf_entries[2];
int unused[20];
volatile int buf_status:1 __attribute__((aligned (CACHE_LINE_SIZE)));
} data_xchg_t;
typedef struct node_param
{
void *prev_comm_channel; /* Channel with previous node */
void *next_comm_channel; /* Channel with next mode */
data_xchg_t *c2p_xfer; /* Consumer to producer local transfer */
int type; /* SOURCE, INTERM or SINK */
int thread_idx;
} node_param_t;
typedef int inc_check_t;
@ -44,18 +55,16 @@ static int (*init_calc)(int) = NULL;
static void **(*do_calc)(void) = NULL;
static int (*end_calc)(void) = NULL;
static int shared = 0; /* We are not shared by default */
pthread_cond_t cond_cons_has_finished = PTHREAD_COND_INITIALIZER;
pthread_mutex_t mutex_cons_has_finished = PTHREAD_MUTEX_INITIALIZER;
static int init_calc_arg = 0;
static long init_calc_arg = 0;
static int block_reception = 1;
static int nb_nodes = 2; // Nb of nodes participating to the chain of pipelines
static long nb_nodes = 2; /* Nb of nodes in the chain of pipelines */
static int check_recv_match_send = 0;
static uintptr_t single_prod_check_val; // /!\ Implies only one real producer
static inc_check_t *single_prod_check_ctxt; // /!\ Implies only one real producer
static int nb_cpus = 4; // TOFIX: don't hardcode this
static uintptr_t single_prod_check_val; /* /!\ Only one real producer */
static inc_check_t *single_prod_check_ctxt; /* /!\ Only one real producer */
static int nb_cpus = 4; /* TOFIX: don't hardcode this */
static int page_size = 0;
void usage(char *argv[])
static void usage(char *argv[])
{
char format[] = "-n <num_buf> -p <num_prod> [options]";
char options[] = "Required options:\n"
@ -70,7 +79,7 @@ void usage(char *argv[])
"-l nb_nodes\t\t\tNumber of nodes in the pipeline chain\n"
"-d\t\t\t\tReceive one piece of data\n"
"-h\t\t\t\tPrint this help\n"
"-s <level>\t\t\tShare the same L<level> cache or not\n"
"-s\t\t\t\tShare the same L2 cache or not\n"
"\t\t\t\tIf level is:\n"
"\t\t\t\t\t> 0, then the same L<level> must be shared\n"
"\t\t\t\t\t< 0, then different L<level> must be used\n"
@ -80,24 +89,24 @@ void usage(char *argv[])
printf("%s\n", options);
}
int do_noinit(int unused)
static int do_noinit(int unused __attribute__ ((unused)))
{
return 0;
}
void **do_nocalc(void)
static void **do_nocalc(void)
{
static int an_int, *an_int_ptr = &an_int;
return (void **) &an_int_ptr;
}
int do_noend(void)
static int do_noend(void)
{
return 0;
}
int inc_check_init(int init_value, inc_check_t **context)
static int inc_check_init(int init_value, inc_check_t **context)
{
inc_check_t *ctxt;
@ -109,24 +118,24 @@ int inc_check_init(int init_value, inc_check_t **context)
return 0;
}
int inc_check_next(inc_check_t *context, uintptr_t *next_value)
static int inc_check_next(inc_check_t *context, uintptr_t *next_value)
{
*next_value = (*context)++;
return 0;
}
int inc_check_end(inc_check_t *context)
static int inc_check_end(inc_check_t *context)
{
free(context);
return 0;
}
int do_checkinit(int init_value)
static int do_checkinit(int init_value)
{
return inc_check_init(init_value, &single_prod_check_ctxt);
}
void **do_checkcalc(void)
static void **do_checkcalc(void)
{
int ret;
@ -137,17 +146,17 @@ void **do_checkcalc(void)
return (void **) single_prod_check_val;
}
int do_checkend(void)
static int do_checkend(void)
{
return inc_check_end(single_prod_check_ctxt);
}
int analyse_options(int argc, char *argv[])
static int analyse_options(int argc, char *argv[])
{
int opt;
opterr = 0;
while ((opt = getopt(argc, argv, ":bc:dhl:kn:s::"/*p:"*/)) != -1)
while ((opt = getopt(argc, argv, ":bc:dhl:kn:s"/*:p:"*/)) != -1)
{
switch (opt)
{
@ -261,40 +270,10 @@ int analyse_options(int argc, char *argv[])
break;
#endif
case 's' :
if ((optind != argc) && (*argv[optind] != '-'))
{
int share_level;
char *inval;
share_level = strtol(argv[optind], &inval, 10);
if ((*argv[optind] == '\0') || (*inval != '\0'))
{
fprintf(stderr, "Option '-p' needs an integer argument\n");
return -1;
}
if ((share_level == LONG_MIN) || ((share_level == LONG_MAX) && errno == ERANGE))
{
fprintf(stderr, "Shared memory level must be between %ld and %ld, both inclusive\n", LONG_MIN, LONG_MAX);
return -1;
}
/* TODO: Real management of shared memory level */
/* TODO: -x: We want level x not to be shared; 0 do as we want, only memory is guaranteed to be shared */
if (share_level <= 0)
shared = 0;
else
shared = 1;
optind++;
}
shared = 1;
break;
case '?' :
fprintf(stderr, "Option inconnue\n");
/*if (!strncmp("--check", argv[optind], strlen("--check")))
{
check_recv_match_send = 1;
optind++;
optopt = (int) *argv[optind];
fprintf(stderr, "--check required\n");
break;
}*/
return -1;
case ':' :
fprintf(stderr, "Option %s needs an argument\n", argv[optind]);
@ -345,9 +324,9 @@ int analyse_options(int argc, char *argv[])
return 0;
}
int producer(void *prod_channel)
static int initial_producer(node_param_t *node_param)
{
int i, j;
unsigned int i, j;
if (init_calc(init_calc_arg))
{
@ -355,9 +334,8 @@ int producer(void *prod_channel)
return 1;
}
for(i = 0; i < nb_bufs_sent; i++) {
//printf("[%p] Send %d new CACHE_LINE\n", (void *) pthread_self(), BUF_SIZE / CACHE_LINE_SIZE);
for(j = 0; j < WORDS_PER_LINE; j++)
send(prod_channel, do_calc());
send(node_param->next_comm_channel, do_calc());
}
if (end_calc())
{
@ -368,178 +346,348 @@ int producer(void *prod_channel)
return 0;
}
void on_message(void *val)
static int intermediate_producer(node_param_t *node_param)
{
//printf("Receive value: %p\n", (void *) val);
int i, j;
long long total_data_sent = 0, total_to_send;
i = 0;
total_to_send = nb_bufs_sent * WORDS_PER_LINE;
while (total_data_sent < total_to_send)
{
while (!node_param->c2p_xfer->buf_status);
for(j = 0; j < node_param->c2p_xfer->filled_buf_entries[i]; j++)
send(node_param->next_comm_channel, node_param->c2p_xfer->data_buf[i][j]);
total_data_sent += node_param->c2p_xfer->filled_buf_entries[i];
node_param->c2p_xfer->buf_status = 0;
i = !i;
}
printf("[%p] Producer finished !\n", (void*) pthread_self());
return 0;
}
int consumer(void *cons_channel)
static void on_message(void *val __attribute__ ((unused)))
{
/*printf("Receive value: %p\n", (void *) val);*/
}
static void *consumer_block(node_param_t *node_param)
{
int i = 0;
int delayed_error;
unsigned long long total_data_received = 0, total_to_receive;
uintptr_t cons_check_value;
inc_check_t *cons_check_context;
cons_check_context = NULL;
delayed_error = 0;
if (check_recv_match_send)
{
if (inc_check_init(init_calc_arg, &cons_check_context))
{
fprintf(stderr, "Initialization of check has failed\n");
return &page_size; /* &page_size can't be NULL */
}
}
cons_check_value = init_calc_arg;
total_to_receive = nb_bufs_sent * WORDS_PER_LINE;
while (total_data_received < total_to_receive)
{
int j;
ssize_t nb_data_received;
size_t to_receive;
to_receive = MIN(MAX_BLOCK_ENTRIES, total_to_receive - total_data_received);
nb_data_received = recv_some_data(node_param->prev_comm_channel,
(void **) node_param->c2p_xfer->data_buf[i], to_receive);
node_param->c2p_xfer->filled_buf_entries[i] = nb_data_received;
if (unlikely(!(node_param->type & SINK)))
{
/* Check with nb lines sent & recv */
while (node_param->c2p_xfer->buf_status);
node_param->c2p_xfer->buf_status = 1;
}
total_data_received += nb_data_received;
for (j = 0; j < nb_data_received; j++)
{
if (unlikely(check_recv_match_send))
{
if (inc_check_next(cons_check_context, &cons_check_value))
{
if (!delayed_error)
{
fprintf(stderr, "Error while checking received value match sent value\n");
delayed_error = 1;
}
}
if (cons_check_value != (uintptr_t) node_param->c2p_xfer->data_buf[i][j])
{
if (!delayed_error)
{
fprintf(stderr, "Mismatch between expected(%lu) and received values(%lu)\n", cons_check_value, (uintptr_t) node_param->c2p_xfer->data_buf[i][j]);
delayed_error = 1;
}
}
}
on_message(node_param->c2p_xfer->data_buf[i][j]);
}
i = !i;
/*printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), nb_data_received, nb_data_received ? "s" : "");*/
}
if (unlikely(!(node_param->type & SINK)))
printf("[%p] Consumer finished !\n", (void*) pthread_self());
if (delayed_error)
return &page_size; /* &page_size can't be NULL */
return NULL;
}
static int consumer_data(node_param_t *node_param)
{
unsigned int i, j;
int delayed_error;
uintptr_t cons_check_value;
inc_check_t *cons_check_context;
cons_check_context = NULL;
delayed_error = 0;
if (inc_check_init(init_calc_arg, &cons_check_context))
if (check_recv_match_send)
{
fprintf(stderr, "Initialization of check has failed\n");
return -1; /* &page_size can't be NULL, whatever NULL is bound to */
if (inc_check_init(init_calc_arg, &cons_check_context))
{
fprintf(stderr, "Initialization of check has failed\n");
return -1; /* &page_size can't be NULL */
}
}
cons_check_value = init_calc_arg;
if (block_reception)
{
long long total_data_received = 0;
void *data_buf[MAX_BLOCK_ENTRIES];
while (total_data_received < nb_bufs_sent * WORDS_PER_LINE)
for(i = 0; i < nb_bufs_sent; i++) {
for(j = 0; j < WORDS_PER_LINE; j++)
{
int i;
ssize_t nb_data_received;
void *data;
nb_data_received = recv_some_data(cons_channel, data_buf, MAX_BLOCK_ENTRIES);
total_data_received += nb_data_received;
for (i = 0; i < nb_data_received; i++)
data = recv_one_data(node_param->prev_comm_channel);
if (unlikely(check_recv_match_send))
{
if (check_recv_match_send)
if (inc_check_next(cons_check_context, &cons_check_value))
{
if (inc_check_next(cons_check_context, &cons_check_value))
if (!delayed_error)
{
if (!delayed_error)
{
fprintf(stderr, "Error while checking received value match sent value\n");
delayed_error = 1;
}
}
if (cons_check_value != (uintptr_t) data_buf[i])
{
if (!delayed_error)
{
fprintf(stderr, "Mismatch between expected(%lu) and received values(%lu)\n", cons_check_value, (uintptr_t) data_buf[i]);
delayed_error = 1;
}
fprintf(stderr, "Error while checking received value match sent value\n");
delayed_error = 1;
}
}
on_message(data_buf[i]);
}
//printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), nb_data_received, nb_data_received ? "s" : "");
}
}
else
{
int i, j;
for(i = 0; i < nb_bufs_sent; i++) {
for(j = 0; j < WORDS_PER_LINE; j++)
{
void *data;
data = recv_one_data(cons_channel);
if (check_recv_match_send)
if (cons_check_value != (uintptr_t) data)
{
if (inc_check_next(cons_check_context, &cons_check_value))
if (!delayed_error)
{
if (!delayed_error)
{
fprintf(stderr, "Error while checking received value match sent value\n");
delayed_error = 1;
}
}
if (cons_check_value != (uintptr_t) data)
{
if (!delayed_error)
{
fprintf(stderr, "Mismatch between expected(%lu) and received values(%lu)\n", cons_check_value, (uintptr_t) data);
delayed_error = 1;
}
fprintf(stderr, "Mismatch between expected(%lu) and received values(%lu)\n", cons_check_value, (uintptr_t) data);
delayed_error = 1;
}
}
on_message(data);
//printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), WORDS_PER_LINE, WORDS_PER_LINE ? "s" : "");
}
on_message(data);
if (likely(node_param->type == INTERM))
send(node_param->next_comm_channel, data);
/*printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), WORDS_PER_LINE, WORDS_PER_LINE ? "s" : "");*/
}
}
printf("[%p] Consumer finished !\n", (void*) pthread_self());
if (unlikely(!(node_param->type & SINK)))
printf("[%p] Consumer finished !\n", (void*) pthread_self());
if (delayed_error)
return -1;
return 0;
}
int consprod(void *cons_channel, void *prod_channel)
static int alloc_consprod_databuf(data_xchg_t **c2p_xfer_addr)
{
if (block_reception)
{
long long total_data_received = 0;
void *data_buf[MAX_BLOCK_ENTRIES];
int ret;
data_xchg_t *c2p_xfer;
while (total_data_received < nb_bufs_sent * WORDS_PER_LINE)
{
int i;
ssize_t nb_data_received;
nb_data_received = recv_some_data(cons_channel, data_buf, MAX_BLOCK_ENTRIES);
total_data_received += nb_data_received;
for (i = 0; i < nb_data_received; i++)
send(prod_channel, data_buf[i]);
//printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), nb_data_received, nb_data_received ? "s" : "");
}
}
else
{
int i, j;
for(i = 0; i < nb_bufs_sent; i++) {
for(j = 0; j < WORDS_PER_LINE; j++)
send(prod_channel, recv_one_data(cons_channel));
//printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), WORDS_PER_LINE, WORDS_PER_LINE ? "s" : "");
}
}
printf("[%p] Producer/consumer finished !\n", (void*) pthread_self());
ret = posix_memalign((void **) &c2p_xfer, page_size, sizeof(*c2p_xfer));
c2p_xfer->buf_status = 0;
if (ret)
return EXIT_FAILURE;
*c2p_xfer_addr = c2p_xfer;
return 0;
}
void *node(prod_cons_thread_t *thread_params)
static int interm_block_xfer(node_param_t *node_param)
{
int return_value;
void *pthread_return_value;
pthread_t prod_tid;
return_value = 0;
if (alloc_consprod_databuf(&node_param->c2p_xfer))
return -1;
/* Create producer to next node, we are consumer of previous node */
if (pthread_create(&prod_tid, NULL, (void *(*)(void *)) consumer_block,
node_param))
{
perror("pthread_create cons (block_reception)");
return_value = -1;
goto free_c2p_xfer_ressources;
}
return_value = intermediate_producer(node_param);
pthread_join(prod_tid, &pthread_return_value);
if (pthread_return_value != NULL)
return_value = -1;
free_c2p_xfer_ressources:
free(node_param->c2p_xfer);
return return_value;
}
static int set_cpu_binding(int thread_idx)
{
int cpu_binding;
pthread_t tid;
cpu_set_t cpuset;
/* Should work in most cases */
if (shared)
cpu_binding = thread_idx % nb_cpus;
else
cpu_binding = (2 * thread_idx) % nb_cpus;
tid = pthread_self();
CPU_ZERO(&cpuset);
CPU_SET(thread_params->cpu_binding, &cpuset);
CPU_SET(cpu_binding, &cpuset);
if (pthread_setaffinity_np(tid, sizeof(cpu_set_t), &cpuset))
{
perror("pthread_setaffinity_np");
return NULL;
return -1;
}
switch (thread_params->flags & (PROD | CONS))
return 0;
}
static void *node(node_param_t *node_param)
{
int return_value;
if (set_cpu_binding(node_param->thread_idx))
return &page_size;
switch (node_param->type)
{
case PROD:
return_value = producer(thread_params->prod_comm_channel);
case SOURCE:
return_value = initial_producer(node_param);
break;
case CONS:
return_value = consumer(thread_params->cons_comm_channel);
case INTERM:
if (block_reception)
return_value = interm_block_xfer(node_param);
else
return_value = consumer_data(node_param);
break;
case (PROD | CONS):
return_value = consprod(thread_params->cons_comm_channel,
thread_params->prod_comm_channel);
case SINK:
if (block_reception)
{
if (alloc_consprod_databuf(&node_param->c2p_xfer))
return &page_size;
if (consumer_block(node_param) == NULL) //TODO: allocate data_buf
return_value = 0;
else
return_value = -1;
free(node_param->c2p_xfer);
}
else
return_value = consumer_data(node_param);
break;
default:
return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
return &page_size; /* &page_size can't be NULL */
}
if (return_value)
return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
return &page_size; /* &page_size can't be NULL */
return NULL;
}
static int set_node_params(node_param_t *node_param,
node_param_t *prev_node_param, int thread_idx)
{
if (thread_idx == nb_nodes - 1)
node_param->type = SINK;
else
{
if (thread_idx)
node_param->type = INTERM;
else
node_param->type = SOURCE;
node_param->next_comm_channel = create_comm_channel();
if (node_param->next_comm_channel == NULL)
return -1;
}
if (prev_node_param != NULL)
{
node_param->prev_comm_channel =
prev_node_param->next_comm_channel;
}
node_param->thread_idx = thread_idx;
return 0;
}
static int create_threads(int *p_nb_nodes, pthread_t *node_tids,
node_param_t *node_params)
{
int i, nb_nodes;
nb_nodes = *p_nb_nodes;
for (i = 0; i < nb_nodes; i++)
{
node_param_t *prev_node_param;
prev_node_param = (i) ? &node_params[i - 1] : NULL;
if (set_node_params(&node_params[i], prev_node_param, i))
{
*p_nb_nodes = i - 1;
return -1;
}
if (pthread_create(&node_tids[i], NULL,
(void *(*)(void *)) node, &node_params[i]))
{
perror("pthread_create node");
destroy_comm_channel(node_params[i].next_comm_channel);
*p_nb_nodes = i - 1;
return -1;
}
}
*p_nb_nodes = i - 1;
return 0;
}
static int join_threads(int nb_threads, pthread_t *tids)
{
int i, return_value;
void *pthread_return_value;
for (i = 0, return_value = 0; i < nb_threads; i++)
{
pthread_join(tids[i], &pthread_return_value);
if (pthread_return_value != NULL)
return_value = EXIT_FAILURE;
}
return return_value;
}
static int destroy_threads(int last_allocated, node_param_t *node_params)
{
int i, return_value;
for (i = last_allocated, return_value = 0; i >= 0; i--)
{
if (node_params[i].type != SINK)
{
if (destroy_comm_channel(node_params[i].next_comm_channel))
return_value = -1;
}
}
return return_value;
}
int main(int argc, char *argv[])
{
int return_value, nb_threads;
pthread_t *tids;
int i, return_value;
void *pthread_return_value;
prod_cons_thread_t *thread_params;
node_param_t *node_params;
return_value = EXIT_SUCCESS;
if (analyse_options(argc, argv))
@ -547,59 +695,25 @@ int main(int argc, char *argv[])
page_size = sysconf(_SC_PAGE_SIZE);
if (page_size <= 0)
return EXIT_FAILURE;
thread_params = malloc(nb_nodes * sizeof(prod_cons_thread_t));
if (thread_params == NULL)
nb_threads = nb_nodes;
node_params = malloc(nb_threads * sizeof(node_param_t));
if (node_params == NULL)
return EXIT_FAILURE;
tids = malloc(nb_nodes * sizeof(pthread_t));
tids = malloc(nb_threads * sizeof(pthread_t));
if (tids == NULL)
{
return_value = EXIT_FAILURE;
goto error_alloc_tids;
}
for (i = 0; i < nb_nodes - 1; i++)
{
if (i)
thread_params[i].flags = PROD | CONS;
else
thread_params[i].flags = PROD;
// Should work in most cases
if (shared)
thread_params[i].cpu_binding = i % nb_cpus;
else
thread_params[i].cpu_binding = (2 * i) % nb_cpus;
thread_params[i].prod_comm_channel = create_comm_channel();
if (thread_params[i].prod_comm_channel == NULL)
{
return_value = EXIT_FAILURE;
goto error_create_channels;
}
if (i)
thread_params[i].cons_comm_channel =
thread_params[i - 1].prod_comm_channel;
pthread_create(&tids[i], NULL, (void *(*)(void *)) node, &thread_params[i]);
}
thread_params[i].flags = CONS;
if (shared)
thread_params[i].cpu_binding = i % nb_cpus;
else
thread_params[i].cpu_binding = (2 * i) % nb_cpus;
thread_params[i].cons_comm_channel =
thread_params[i - 1].prod_comm_channel;
pthread_create(&tids[i], NULL, (void *(*)(void *)) node, &thread_params[i]);
for (i = 0; i < nb_nodes; i++)
{
pthread_join(tids[i], &pthread_return_value);
if (pthread_return_value != NULL)
return_value = EXIT_FAILURE;
}
i--;
if (create_threads(&nb_threads, tids, node_params))
goto error_create_channels;
if (join_threads(nb_threads, tids))
return_value = EXIT_FAILURE;
error_create_channels:
for (i-- ; i >= 0; i--) {
if (destroy_comm_channel(thread_params[i].prod_comm_channel))
return_value = EXIT_FAILURE;
}
if (destroy_threads(nb_threads, node_params))
return_value = EXIT_FAILURE;
free(tids);
error_alloc_tids:
free(thread_params);
free(node_params);
return return_value;
}