void rosout_post(UrosString *strp, uros_bool_t costant, uint8_t level,
const char *fileszp, int line, const char *funcp) {
static uint32_t seq = 0;
struct msg__rosgraph_msgs__Log *msgp;
urosAssert(urosStringIsValid(strp));
msgp = urosNew(NULL, struct msg__rosgraph_msgs__Log);
urosAssert(msgp != NULL);
init_msg__rosgraph_msgs__Log(msgp);
msgp->header.frame_id = urosStringAssignZ(costant ? "1" : "0");
msgp->header.seq = seq++;
msgp->header.stamp.sec = urosGetTimestampMsec();
msgp->header.stamp.nsec = (msgp->header.stamp.sec % 1000) * 1000000;
msgp->header.stamp.sec /= 1000;
msgp->level = level;
msgp->name = urosNode.config.nodeName;
msgp->msg = *strp;
msgp->file = urosStringAssignZ(fileszp);
msgp->function = urosStringAssignZ(funcp);
msgp->line = line;
fifo_enqueue(&rosoutQueue, (void *)msgp);
}
static void process()
{
while (end_scatter_gather == 0)
{
unsigned int tmp_token;
if(fifo_dequeue_non_blocking(&scatter_to_process, &tmp_token))
{
fifo_enqueue(&process_to_gather, tmp_token);
// display critical section //
uart_lock_acquire();
uart_puts("process on core ");
uart_putd(core_id());
uart_puts(" - processing token ");
uart_putd(tmp_token);
uart_putc('\n');
uart_lock_release();
// display critical section //
}
}
// display critical section //
uart_lock_acquire();
uart_puts("process on core ");
uart_putd(core_id());
uart_puts(" - END\n");
uart_lock_release();
// display critical section //
}
static void scatter()
{
unsigned long int nb_token = 0;
while (nb_token < MAX_TOKEN)
{
// display critical section //
uart_lock_acquire();
uart_puts("scatter on core ");
uart_putd(core_id());
uart_puts(" - generating token ");
uart_putd(nb_token);
uart_putc('\n');
uart_lock_release();
// display critical section //
fifo_enqueue(&scatter_to_process, nb_token);
nb_token++;
}
// display critical section //
uart_lock_acquire();
uart_puts("scatter on core ");
uart_putd(core_id());
uart_puts(" - END\n");
uart_lock_release();
// display critical section //
}
/** output one character */
void uart_putc(uint8_t data)
{
/* store data */
fifo_enqueue((fifo_t *)&global_uart.tx, data);
/* enable interrupt */
_UCSRB_UART0 |= _BV(_UDRIE_UART0);
}
static int ilo_pkt_enqueue(struct ilo_hwinfo *hw, struct ccb *ccb,
int dir, int id, int len)
{
char *fifobar;
int entry;
if (dir == SENDQ)
fifobar = ccb->ccb_u1.send_fifobar;
else
fifobar = ccb->ccb_u3.recv_fifobar;
entry = mk_entry(id, len);
return fifo_enqueue(hw, fifobar, entry);
}
/*
* This function is called as a result of the "eth_drv_recv()" call above.
* It's job is to actually fetch data for a packet from the hardware once
* memory buffers have been allocated for the packet. Note that the buffers
* may come in pieces, using a scatter-gather list. This allows for more
* efficient processing in the upper layers of the stack.
*/
static void dp83902a_recv(struct nic_priv_data *dp, int len)
{
u8 *base = dp->base;
int i, mlen;
u8 saved_char = 0;
bool saved;
/* Read incoming packet data */
DP_OUT(base, DP_CR, DP_CR_PAGE0 | DP_CR_NODMA | DP_CR_START);
DP_OUT(base, DP_RBCL, len & 0xFF);
DP_OUT(base, DP_RBCH, len >> 8);
DP_OUT(base, DP_RSAL, 4); /* Past header */
DP_OUT(base, DP_RSAH, dp->rx_next);
DP_OUT(base, DP_ISR, DP_ISR_RDC); /* Clear end of DMA */
DP_OUT(base, DP_CR, DP_CR_RDMA | DP_CR_START);
saved = false;
for (i = 0; i < 1; i++) {
mlen = len;
while (0 < mlen) {
/* Saved byte from previous loop? */
if (saved) {
fifo_enqueue(dp->rx_rb, &saved_char, TRUE);
mlen--;
saved = false;
continue;
}
{
u8 tmp;
DP_IN_DATA(dp->data, tmp);
fifo_enqueue(dp->rx_rb, &tmp, TRUE);
mlen--;
}
}
}
}
struct mempool *mempool_ram_create(u32 entity_size,
u32 page_count,
u32 mem_flags)
{
u32 e;
virtual_addr_t va;
struct mempool *mp;
if (!entity_size ||
((VMM_PAGE_SIZE * page_count) < entity_size)) {
return NULL;
}
mp = vmm_zalloc(sizeof(struct mempool));
if (!mp) {
return NULL;
}
mp->type = MEMPOOL_TYPE_RAM;
mp->entity_size = entity_size;
mp->entity_count =
udiv64((VMM_PAGE_SIZE * page_count), entity_size);
mp->f = fifo_alloc(sizeof(virtual_addr_t), mp->entity_count);
if (!mp->f) {
vmm_free(mp);
return NULL;
}
mp->entity_base = vmm_host_alloc_pages(page_count, mem_flags);
if (!mp->entity_base) {
fifo_free(mp->f);
vmm_free(mp);
return NULL;
}
mp->d.ram.page_count = page_count;
mp->d.ram.mem_flags = mem_flags;
for (e = 0; e < mp->entity_count; e++) {
va = mp->entity_base + e * entity_size;
fifo_enqueue(mp->f, &va, FALSE);
}
return mp;
}
struct mempool *mempool_raw_create(u32 entity_size,
physical_addr_t phys,
virtual_size_t size,
u32 mem_flags)
{
u32 e;
virtual_addr_t va;
struct mempool *mp;
if (!entity_size || (size < entity_size)) {
return NULL;
}
mp = vmm_zalloc(sizeof(struct mempool));
if (!mp) {
return NULL;
}
mp->type = MEMPOOL_TYPE_RAW;
mp->entity_size = entity_size;
mp->entity_count = udiv64(size, entity_size);
mp->f = fifo_alloc(sizeof(virtual_addr_t), mp->entity_count);
if (!mp->f) {
vmm_free(mp);
return NULL;
}
mp->entity_base = vmm_host_memmap(phys, size, mem_flags);
if (!mp->entity_base) {
fifo_free(mp->f);
vmm_free(mp);
return NULL;
}
mp->d.raw.phys = phys;
mp->d.raw.size = size;
mp->d.raw.mem_flags = mem_flags;
for (e = 0; e < mp->entity_count; e++) {
va = mp->entity_base + e * entity_size;
fifo_enqueue(mp->f, &va, FALSE);
}
return mp;
}
int mempool_free(struct mempool *mp, void *entity)
{
virtual_addr_t entity_va;
if (!mp) {
return VMM_EFAIL;
}
if (!mempool_check_ptr(mp, entity)) {
return VMM_EINVALID;
}
entity_va = (virtual_addr_t)entity;
if (!fifo_enqueue(mp->f, &entity_va, FALSE)) {
return VMM_ENOSPC;
}
return VMM_OK;
}
int main( int argc, char **argv ) {
fifo_queue_t q;
fifo_init(&q);
#pragma omp parallel
#pragma omp single nowait
{
int i;
for(i=1;i<5;++i) {
#pragma omp task
{
int j;
for(j = 0; j < 1000; ++j) {
fifo_enqueue(&q, i*1000+j);
}
}
#pragma omp task
{
int d, j;
for(j = 0; j < 1000; ++j) {
d = fifo_dequeue(&q);
if (d)
printf("dequeue %d\n", d);
}
}
}
}
int d;
while (true) {
d = fifo_dequeue(&q);
if (d == -1)
break;
printf("dequeue %d\n", d);
}
assert(fifo_empty(&q));
#pragma omp taskwait
fifo_free(&q);
return 0;
}
struct mempool *mempool_heap_create(u32 entity_size,
u32 entity_count)
{
u32 e;
virtual_addr_t va;
struct mempool *mp;
if (!entity_size || !entity_count) {
return NULL;
}
mp = vmm_zalloc(sizeof(struct mempool));
if (!mp) {
return NULL;
}
mp->type = MEMPOOL_TYPE_HEAP;
mp->entity_size = entity_size;
mp->entity_count = entity_count;
mp->f = fifo_alloc(sizeof(virtual_addr_t), mp->entity_count);
if (!mp->f) {
vmm_free(mp);
return NULL;
}
mp->entity_base =
(virtual_addr_t)vmm_malloc(entity_size * entity_count);
if (!mp->entity_base) {
fifo_free(mp->f);
vmm_free(mp);
return NULL;
}
for (e = 0; e < mp->entity_count; e++) {
va = mp->entity_base + e * entity_size;
fifo_enqueue(mp->f, &va, FALSE);
}
return mp;
}
int mempool_free(struct mempool *mp, void *buf)
{
virtual_addr_t buf_va;
if (!mp) {
return VMM_EFAIL;
}
buf_va = (virtual_addr_t)buf;
if ((buf_va < mp->page_base) ||
((mp->page_base + (mp->page_count * VMM_PAGE_SIZE)) < buf_va)) {
return VMM_EINVALID;
}
if (!fifo_enqueue(mp->f, &buf_va, FALSE)) {
return VMM_EFAIL;
}
return VMM_OK;
}
void *enquer(void *arg)
{
int i, j;
inc(&b1);
while (b1!=(NENQER+NDEQER));
inc(&b2);
while (b2!=(NENQER+NDEQER));
for (j=0; j<NITER; j++) {
i=anf32(&n, 1);
fifo_enqueue(q, (void *)i);
}
dec(&c1);
while (c1!=0);
dec(&c2);
while (c2!=0);
return NULL;
}
int main(int argc, char *argv[])
{
pthread_t thrs[NENQER+NDEQER];
int i, j;
int qlen=FIFO_LEN_DEFAULT;
if (argc>1) {
qlen=atoi(argv[1]);
}
q=fifo_create(qlen); /* smaller buffer to saturate the buffer */
# if (CHECK)
chararr=(char *)calloc(NITER*NENQER, sizeof(char));
# endif
for (i=0; i<NENQER; i++)
pthread_create(thrs+i, NULL, enquer, NULL);
for (; i<NENQER+NDEQER; i++)
pthread_create(thrs+i, NULL, dequer, NULL);
for (i=0; i<NENQER; i++)
pthread_join(thrs[i], NULL);
for (j=0; j<NDEQER; j++)
fifo_enqueue(q, (void *)0);
for (; i<NENQER+NDEQER; i++)
pthread_join(thrs[i], NULL);
# if (CHECK)
for (j=0; j<NITER*NENQER; j++)
if (chararr[j]==0) {
printf("%d hasn't been seen\n", j+1);
}
# endif
return 0;
}
struct mempool *mempool_create(u32 buf_size, u32 buf_count)
{
u32 b;
virtual_addr_t va;
struct mempool *mp;
mp = vmm_zalloc(sizeof(struct mempool));
if (!mp) {
return NULL;
}
mp->f = fifo_alloc(sizeof(virtual_addr_t), buf_count);
if (!mp->f) {
vmm_free(mp);
return NULL;
}
mp->buf_count = buf_count;
mp->buf_size = buf_size;
mp->page_count = VMM_SIZE_TO_PAGE(buf_size * buf_count);
mp->page_base = vmm_host_alloc_pages(mp->page_count,
VMM_MEMORY_FLAGS_NORMAL);
if (!mp->page_base) {
fifo_free(mp->f);
vmm_free(mp);
return NULL;
}
for (b = 0; b < mp->buf_count; b++) {
va = mp->page_base + b * buf_size;
fifo_enqueue(mp->f, &va, FALSE);
}
return mp;
}
static void* worker(tdat_t *td)
{
#ifdef HAVE_SIGNAL_H
/* set this thread as cancellable */
if (set_cancellable() != 0)
{
td->status = OPNORM_THREAD;
return NULL;
}
#endif
/*
unpack arguments, mostly for readability but also saves
a few dereferences
*/
tdat_shared_t *ts = td->shared;
index_t
n = ts->n,
m = ts->m;
double
p = ts->p,
q = ts->q,
eps = ts->eps,
LCRP = ts->LCRP,
SCD = ts->SCD;
const double *M = ts->M;
fifo_t *fifo = ts->fifo;
/* working data */
double pcent[n];
cube_t cube0, cube1;
patch_t patch;
double tmax = 0.0;
patch.centres = pcent;
int fifoerr;
while (1)
{
/* thread cancellation point */
pthread_testcancel();
/* dequeue a cube */
if (pthread_mutex_lock(&(ts->fifolock)) < 0)
{
td->status = OPNORM_THREAD;
return NULL;
}
fifoerr = fifo_dequeue(fifo, &cube0);
if (pthread_mutex_unlock(&(ts->fifolock)) < 0)
{
td->status = OPNORM_THREAD;
return NULL;
}
if (fifoerr != FIFO_OK)
{
td->status = (fifoerr == FIFO_EMPTY ?
OPNORM_OK :
OPNORM_FIFO);
return NULL;
}
cube_print(&cube0, n, ">");
/* nfifo is the total number dequeue */
td->nfifo++;
/* cube subdivide */
int hwnexp = cube0.hwnexp + 1;
double halfwidth = ldexp(1, -hwnexp);
/*
if halfwidth < DBL_EPSILON then we cannot
calulate the centres of the subdivided cubes
accurately, we break out and report that the
requested accuracy could not be achieved
*/
if (halfwidth < DBL_EPSILON)
{
td->status = OPNORM_INACC;
return NULL;
}
for (size_t k = 0 ; k < (1UL << (n-1)) ; k++)
{
cube1.side = cube0.side;
cube1.hwnexp = hwnexp;
/*
we give our cube1 a temporary set of centres
while we evaluate and decide whether to jetison
or enqueue it, only if the latter do we make a
malloc and copy the temporary centres. this
saves a *lot* of malloc/free pairs
*/
double centres[n];
cube1.centres = centres;
size_t k0 = k;
for (size_t j = 0 ; j < n ; j++)
{
if (cube0.side == j)
{
cube1.centres[j] = cube0.centres[j];
continue;
}
cube1.centres[j] = cube0.centres[j] +
((k0 % 2) ? halfwidth : -halfwidth);
k0 /= 2;
}
cube_print(&cube1, n, "<");
/* get the corresponding patch */
cube_to_patch(&cube1, n, p, LCRP, SCD, &patch);
patch_print(patch, n);
double ratio = radius_to_ratio(patch.radius, p);
/*
check for patch viability - this check almost
always succeeds (on Drury K, this is false in
80/164016 cases, so 0.05% of the time) very
small beer. Yet it introduces a branch point,
so one might think it worth removing. Timing
tests indicate that there is no speedup in
doing so, so we keep it.
*/
if (ratio > 0)
{
/* evaluate M at patch centre */
double v[m];
ts->matvec(M, pcent, m, n, v);
td->neval++;
double t = pnorm(v, m, q);
/* test first with the previous value of tmax */
if (t < tmax)
{
/* test whether we can jettison this cube */
if (t < (tmax * ratio * (1 + eps))) continue;
/*
note that the global ts->tmax >= tmax so it would
be pointless (and cost a mutex access) to test
for that here
*/
}
else
{
if (pthread_mutex_lock(&(ts->maxlock)) < 0)
{
td->status = OPNORM_THREAD;
return NULL;
}
/* update local tmax from global */
tmax = ts->tmax;
/*
if we have found a new global maximum then we
update it (and the global maximising vector)
as well as the local copy
*/
if (t > tmax)
{
ts->tmax = tmax = t;
if (ts->vmax)
memcpy(ts->vmax, pcent, n*sizeof(double));
}
if (pthread_mutex_unlock(&(ts->maxlock)) < 0)
{
td->status = OPNORM_THREAD;
return NULL;
}
/*
test whether we can jettison this cube but
now with the updated value of tmax
*/
if (t < (tmax * ratio * (1 + eps))) continue;
}
}
/*
we will enqueue this cube, so we need to
allocate and copy its temporary centres set
*/
if (! (cube1.centres = malloc(n*sizeof(double))))
{
td->status = OPNORM_ALLOC;
return NULL;
}
memcpy(cube1.centres, centres, n*sizeof(double));
if (pthread_mutex_lock(&(ts->fifolock)) < 0)
{
free(cube1.centres);
td->status = OPNORM_THREAD;
return NULL;
}
fifoerr = fifo_enqueue(fifo, &cube1);
if (pthread_mutex_unlock(&(ts->fifolock)) < 0)
{
td->status = OPNORM_THREAD;
return NULL;
}
switch(fifoerr)
{
case FIFO_OK:
break;
case FIFO_USERMAX:
td->status = OPNORM_FIFOMAX;
return NULL;
default:
td->status = OPNORM_FIFO;
return NULL;
}
}
free(cube0.centres);
}
/* we should not arrive here */
td->status = OPNORM_BUG;
return NULL;
}