static void hfa_parse_gwq_status(struct cgpu_info *hashfast, struct hashfast_info *info,
struct hf_header *h)
{
struct hf_gwq_data *g = (struct hf_gwq_data *)(h + 1);
struct work *work;
applog(LOG_DEBUG, "HFA %d: OP_GWQ_STATUS, device_head %4d tail %4d my tail %4d shed %3d inflight %4d",
hashfast->device_id, g->sequence_head, g->sequence_tail, info->hash_sequence_tail,
g->shed_count, HF_SEQUENCE_DISTANCE(info->hash_sequence_head,g->sequence_tail));
mutex_lock(&info->lock);
info->hash_count += g->hash_count;
info->device_sequence_head = g->sequence_head;
info->device_sequence_tail = g->sequence_tail;
info->shed_count = g->shed_count;
/* Free any work that is no longer required */
while (info->device_sequence_tail != info->hash_sequence_tail) {
if (++info->hash_sequence_tail >= info->num_sequence)
info->hash_sequence_tail = 0;
if (unlikely(!(work = info->works[info->hash_sequence_tail]))) {
applog(LOG_ERR, "HFA %d: Bad work sequence tail",
hashfast->device_id);
hashfast->shutdown = true;
break;
}
applog(LOG_DEBUG, "HFA %d: Completing work on hash_sequence_tail %d",
hashfast->device_id, info->hash_sequence_tail);
free_work(work);
info->works[info->hash_sequence_tail] = NULL;
}
mutex_unlock(&info->lock);
}
bool do_job_prepare(struct thr_info *mythr, struct timeval *tvp_now)
{
struct cgpu_info *proc = mythr->cgpu;
struct device_drv *api = proc->drv;
struct timeval tv_worktime;
mythr->tv_morework.tv_sec = -1;
mythr->_job_transition_in_progress = true;
if (mythr->work)
timersub(tvp_now, &mythr->work->tv_work_start, &tv_worktime);
if ((!mythr->work) || abandon_work(mythr->work, &tv_worktime, proc->max_hashes))
{
mythr->work_restart = false;
request_work(mythr);
// FIXME: Allow get_work to return NULL to retry on notification
if (mythr->next_work)
free_work(mythr->next_work);
mythr->next_work = get_and_prepare_work(mythr);
if (!mythr->next_work)
return false;
mythr->starting_next_work = true;
api->job_prepare(mythr, mythr->next_work, mythr->_max_nonce);
}
else
{
mythr->starting_next_work = false;
api->job_prepare(mythr, mythr->work, mythr->_max_nonce);
}
job_prepare_complete(mythr);
return true;
}
PRIVATE void error
(
NumericType **Numeric,
WorkType *Work
)
{
free_work (Work) ;
UMFPACK_free_numeric ((void **) Numeric) ;
ASSERT (UMF_malloc_count == init_count) ;
}
static void gridseed_test_btc_nonce(struct cgpu_info *gridseed, GRIDSEED_INFO *info,
struct thr_info *thr, unsigned char *data)
{
struct work *work;
uint32_t nonce;
int workid, index, i;
bool valid = false;
bool nowork = false;
memcpy(&workid, data+8, 4);
memcpy(&nonce, data+4, 4);
nonce = htole32(nonce);
mutex_lock(&info->qlock);
nowork = (info->soft_queue_len <= 0);
for(i=0; i<info->soft_queue_len; i++) {
struct work *dupwork;
work = info->workqueue[i];
if (work->devflag == false)
continue;
if (work->id > workid)
break;
dupwork = copy_work(work);
if (dupwork == NULL)
continue;
if (test_nonce(dupwork, nonce)) {
submit_tested_work(thr, dupwork);
index = i;
valid = true;
free_work(dupwork);
break;
} else
free_work(dupwork);
}
if (valid)
__gridseed_purge_work_queue(gridseed, info, index);
mutex_unlock(&info->qlock);
if (!valid && !nowork)
inc_hw_errors(thr);
return;
}
void pme_free_all_work(struct pme_solve_work_t **work, int nthread)
{
int thread;
for (thread = 0; thread < nthread; thread++)
{
free_work(&(*work)[thread]);
}
sfree(work);
*work = NULL;
}
static void hfa_free_all_work(struct hashfast_info *info)
{
while (info->device_sequence_tail != info->hash_sequence_head) {
struct work *work;
if (++info->hash_sequence_tail >= info->num_sequence)
info->hash_sequence_tail = 0;
if (unlikely(!(work = info->works[info->hash_sequence_tail])))
break;
free_work(work);
info->works[info->hash_sequence_tail] = NULL;
}
}
static
void do_queue_flush(struct thr_info *mythr)
{
struct cgpu_info *proc = mythr->cgpu;
struct device_drv *api = proc->drv;
api->queue_flush(mythr);
if (mythr->next_work)
{
free_work(mythr->next_work);
mythr->next_work = NULL;
}
}
static void hfa_parse_gwq_status(struct cgpu_info *hashfast, struct hashfast_info *info,
struct hf_header *h)
{
struct hf_gwq_data *g = (struct hf_gwq_data *)(h + 1);
struct work *work;
applog(LOG_DEBUG, "HFA %d: OP_GWQ_STATUS, device_head %4d tail %4d my tail %4d shed %3d inflight %4d",
hashfast->device_id, g->sequence_head, g->sequence_tail, info->hash_sequence_tail,
g->shed_count, HF_SEQUENCE_DISTANCE(info->hash_sequence_head,g->sequence_tail));
/* This is a special flag that the thermal overload has been tripped */
if (unlikely(h->core_address & 0x80)) {
applog(LOG_WARNING, "HFA %d Thermal overload tripped! Resetting device",
hashfast->device_id);
hfa_send_shutdown(hashfast);
if (hfa_reset(hashfast, info)) {
applog(LOG_NOTICE, "HFA %d: Succesfully reset, continuing operation",
hashfast->device_id);
return;
}
applog(LOG_WARNING, "HFA %d Failed to reset device, killing off thread to allow re-hotplug",
hashfast->device_id);
usb_nodev(hashfast);
return;
}
mutex_lock(&info->lock);
info->hash_count += g->hash_count;
info->device_sequence_head = g->sequence_head;
info->device_sequence_tail = g->sequence_tail;
info->shed_count = g->shed_count;
/* Free any work that is no longer required */
while (info->device_sequence_tail != info->hash_sequence_tail) {
if (++info->hash_sequence_tail >= info->num_sequence)
info->hash_sequence_tail = 0;
if (unlikely(!(work = info->works[info->hash_sequence_tail]))) {
applog(LOG_ERR, "HFA %d: Bad work sequence tail",
hashfast->device_id);
hashfast->shutdown = true;
break;
}
applog(LOG_DEBUG, "HFA %d: Completing work on hash_sequence_tail %d",
hashfast->device_id, info->hash_sequence_tail);
free_work(work);
info->works[info->hash_sequence_tail] = NULL;
}
mutex_unlock(&info->lock);
}
void mt_job_transition(struct thr_info *mythr)
{
struct timeval tv_now;
timer_set_now(&tv_now);
if (mythr->starting_next_work)
{
mythr->next_work->tv_work_start = tv_now;
if (mythr->prev_work)
free_work(mythr->prev_work);
mythr->prev_work = mythr->work;
mythr->work = mythr->next_work;
mythr->next_work = NULL;
}
mythr->tv_jobstart = tv_now;
mythr->_job_transition_in_progress = false;
}
static
struct work *get_and_prepare_work(struct thr_info *thr)
{
struct cgpu_info *proc = thr->cgpu;
struct device_drv *api = proc->drv;
struct work *work;
work = get_work(thr);
if (!work)
return NULL;
if (api->prepare_work && !api->prepare_work(thr, work)) {
free_work(work);
applog(LOG_ERR, "%"PRIpreprv": Work prepare failed, disabling!", proc->proc_repr);
proc->deven = DEV_RECOVER_ERR;
run_cmd(cmd_idle);
return NULL;
}
return work;
}
GLOBAL Int UMFPACK_numeric
(
const Int Ap [ ],
const Int Ai [ ],
const double Ax [ ],
#ifdef COMPLEX
const double Az [ ],
#endif
void *SymbolicHandle,
void **NumericHandle,
const double Control [UMFPACK_CONTROL],
double User_Info [UMFPACK_INFO]
)
{
/* ---------------------------------------------------------------------- */
/* local variables */
/* ---------------------------------------------------------------------- */
double Info2 [UMFPACK_INFO], alloc_init, relpt, relpt2, droptol,
front_alloc_init, stats [2] ;
double *Info ;
WorkType WorkSpace, *Work ;
NumericType *Numeric ;
SymbolicType *Symbolic ;
Int n_row, n_col, n_inner, newsize, i, status, *inew, npiv, ulen, scale ;
Unit *mnew ;
/* ---------------------------------------------------------------------- */
/* get the amount of time used by the process so far */
/* ---------------------------------------------------------------------- */
umfpack_tic (stats) ;
/* ---------------------------------------------------------------------- */
/* initialize and check inputs */
/* ---------------------------------------------------------------------- */
#ifndef NDEBUG
UMF_dump_start ( ) ;
init_count = UMF_malloc_count ;
DEBUGm4 (("\nUMFPACK numeric: U transpose version\n")) ;
#endif
/* If front_alloc_init negative then allocate that size of front in
* UMF_start_front. If alloc_init negative, then allocate that initial
* size of Numeric->Memory. */
relpt = GET_CONTROL (UMFPACK_PIVOT_TOLERANCE,
UMFPACK_DEFAULT_PIVOT_TOLERANCE) ;
relpt2 = GET_CONTROL (UMFPACK_SYM_PIVOT_TOLERANCE,
UMFPACK_DEFAULT_SYM_PIVOT_TOLERANCE) ;
alloc_init = GET_CONTROL (UMFPACK_ALLOC_INIT, UMFPACK_DEFAULT_ALLOC_INIT) ;
front_alloc_init = GET_CONTROL (UMFPACK_FRONT_ALLOC_INIT,
UMFPACK_DEFAULT_FRONT_ALLOC_INIT) ;
scale = GET_CONTROL (UMFPACK_SCALE, UMFPACK_DEFAULT_SCALE) ;
droptol = GET_CONTROL (UMFPACK_DROPTOL, UMFPACK_DEFAULT_DROPTOL) ;
relpt = MAX (0.0, MIN (relpt, 1.0)) ;
relpt2 = MAX (0.0, MIN (relpt2, 1.0)) ;
droptol = MAX (0.0, droptol) ;
front_alloc_init = MIN (1.0, front_alloc_init) ;
if (scale != UMFPACK_SCALE_NONE && scale != UMFPACK_SCALE_MAX)
{
scale = UMFPACK_DEFAULT_SCALE ;
}
if (User_Info != (double *) NULL)
{
/* return Info in user's array */
Info = User_Info ;
/* clear the parts of Info that are set by UMFPACK_numeric */
for (i = UMFPACK_NUMERIC_SIZE ; i <= UMFPACK_MAX_FRONT_NCOLS ; i++)
{
Info [i] = EMPTY ;
}
for (i = UMFPACK_NUMERIC_DEFRAG ; i < UMFPACK_IR_TAKEN ; i++)
{
Info [i] = EMPTY ;
}
}
else
{
/* no Info array passed - use local one instead */
Info = Info2 ;
for (i = 0 ; i < UMFPACK_INFO ; i++)
{
Info [i] = EMPTY ;
}
}
Symbolic = (SymbolicType *) SymbolicHandle ;
Numeric = (NumericType *) NULL ;
if (!UMF_valid_symbolic (Symbolic))
{
Info [UMFPACK_STATUS] = UMFPACK_ERROR_invalid_Symbolic_object ;
return (UMFPACK_ERROR_invalid_Symbolic_object) ;
}
/* compute alloc_init automatically for AMD or other symmetric ordering */
if (/* Symbolic->ordering == UMFPACK_ORDERING_AMD */ alloc_init >= 0
&& Symbolic->amd_lunz > 0)
{
alloc_init = (Symbolic->nz + Symbolic->amd_lunz) / Symbolic->lunz_bound;
alloc_init = MIN (1.0, alloc_init) ;
alloc_init *= UMF_REALLOC_INCREASE ;
}
n_row = Symbolic->n_row ;
n_col = Symbolic->n_col ;
n_inner = MIN (n_row, n_col) ;
/* check for integer overflow in Numeric->Memory minimum size */
if (INT_OVERFLOW (Symbolic->dnum_mem_init_usage * sizeof (Unit)))
{
/* :: int overflow, initial Numeric->Memory size :: */
/* There's no hope to allocate a Numeric object big enough simply to
* hold the initial matrix, so return an out-of-memory condition */
DEBUGm4 (("out of memory: numeric int overflow\n")) ;
Info [UMFPACK_STATUS] = UMFPACK_ERROR_out_of_memory ;
return (UMFPACK_ERROR_out_of_memory) ;
}
Info [UMFPACK_STATUS] = UMFPACK_OK ;
Info [UMFPACK_NROW] = n_row ;
Info [UMFPACK_NCOL] = n_col ;
Info [UMFPACK_SIZE_OF_UNIT] = (double) (sizeof (Unit)) ;
if (!Ap || !Ai || !Ax || !NumericHandle)
{
Info [UMFPACK_STATUS] = UMFPACK_ERROR_argument_missing ;
return (UMFPACK_ERROR_argument_missing) ;
}
Info [UMFPACK_NZ] = Ap [n_col] ;
*NumericHandle = (void *) NULL ;
/* ---------------------------------------------------------------------- */
/* allocate the Work object */
/* ---------------------------------------------------------------------- */
/* (1) calls UMF_malloc 15 or 17 times, to obtain temporary workspace of
* size c+1 Entry's and 2*(n_row+1) + 3*(n_col+1) + (n_col+n_inner+1) +
* (nn+1) + * 3*(c+1) + 2*(r+1) + max(r,c) + (nfr+1) integers plus 2*nn
* more integers if diagonal pivoting is to be done. r is the maximum
* number of rows in any frontal matrix, c is the maximum number of columns
* in any frontal matrix, n_inner is min (n_row,n_col), nn is
* max (n_row,n_col), and nfr is the number of frontal matrices. For a
* square matrix, this is c+1 Entry's and about 8n + 3c + 2r + max(r,c) +
* nfr integers, plus 2n more for diagonal pivoting.
*/
Work = &WorkSpace ;
Work->n_row = n_row ;
Work->n_col = n_col ;
Work->nfr = Symbolic->nfr ;
Work->nb = Symbolic->nb ;
Work->n1 = Symbolic->n1 ;
if (!work_alloc (Work, Symbolic))
{
DEBUGm4 (("out of memory: numeric work\n")) ;
Info [UMFPACK_STATUS] = UMFPACK_ERROR_out_of_memory ;
error (&Numeric, Work) ;
return (UMFPACK_ERROR_out_of_memory) ;
}
ASSERT (UMF_malloc_count == init_count + 16 + 2*Symbolic->prefer_diagonal) ;
/* ---------------------------------------------------------------------- */
/* allocate Numeric object */
/* ---------------------------------------------------------------------- */
/* (2) calls UMF_malloc 10 or 11 times, for a total space of
* sizeof (NumericType) bytes, 4*(n_row+1) + 4*(n_row+1) integers, and
* (n_inner+1) Entry's, plus n_row Entry's if row scaling is to be done.
* sizeof (NumericType) is a small constant. Next, it calls UMF_malloc
* once, for the variable-sized part of the Numeric object
* (Numeric->Memory). The size of this object is the larger of
* (Control [UMFPACK_ALLOC_INIT]) * (the approximate upper bound computed
* by UMFPACK_symbolic), and the minimum required to start the numerical
* factorization. * This request is reduced if it fails.
*/
if (!numeric_alloc (&Numeric, Symbolic, alloc_init, scale))
{
DEBUGm4 (("out of memory: initial numeric\n")) ;
Info [UMFPACK_STATUS] = UMFPACK_ERROR_out_of_memory ;
error (&Numeric, Work) ;
return (UMFPACK_ERROR_out_of_memory) ;
}
DEBUG0 (("malloc: init_count "ID" UMF_malloc_count "ID"\n",
init_count, UMF_malloc_count)) ;
ASSERT (UMF_malloc_count == init_count
+ (16 + 2*Symbolic->prefer_diagonal)
+ (11 + (scale != UMFPACK_SCALE_NONE))) ;
/* set control parameters */
Numeric->relpt = relpt ;
Numeric->relpt2 = relpt2 ;
Numeric->droptol = droptol ;
Numeric->alloc_init = alloc_init ;
Numeric->front_alloc_init = front_alloc_init ;
Numeric->scale = scale ;
DEBUG0 (("umf relpt %g %g init %g %g inc %g red %g\n",
relpt, relpt2, alloc_init, front_alloc_init,
UMF_REALLOC_INCREASE, UMF_REALLOC_REDUCTION)) ;
/* ---------------------------------------------------------------------- */
/* scale and factorize */
/* ---------------------------------------------------------------------- */
/* (3) During numerical factorization (inside UMF_kernel), the variable-size
* block of memory is increased in size via a call to UMF_realloc if it is
* found to be too small. During factorization, this block holds the
* pattern and values of L and U at the top end, and the elements
* (contibution blocks) and the current frontal matrix (Work->F*) at the
* bottom end. The peak size of the variable-sized object is estimated in
* UMFPACK_*symbolic (Info [UMFPACK_VARIABLE_PEAK_ESTIMATE]), although this
* upper bound can be very loose. The size of the Symbolic object
* (which is currently allocated) is in Info [UMFPACK_SYMBOLIC_SIZE], and
* is between 2*n and 13*n integers.
*/
DEBUG0 (("Calling umf_kernel\n")) ;
status = UMF_kernel (Ap, Ai, Ax,
#ifdef COMPLEX
Az,
#endif
Numeric, Work, Symbolic) ;
Info [UMFPACK_STATUS] = status ;
if (status < UMFPACK_OK)
{
/* out of memory, or pattern has changed */
error (&Numeric, Work) ;
return (status) ;
}
Info [UMFPACK_FORCED_UPDATES] = Work->nforced ;
Info [UMFPACK_VARIABLE_INIT] = Numeric->init_usage ;
if (Symbolic->prefer_diagonal)
{
Info [UMFPACK_NOFF_DIAG] = Work->noff_diagonal ;
}
DEBUG0 (("malloc: init_count "ID" UMF_malloc_count "ID"\n",
init_count, UMF_malloc_count)) ;
npiv = Numeric->npiv ; /* = n_inner for nonsingular matrices */
ulen = Numeric->ulen ; /* = 0 for square nonsingular matrices */
/* ---------------------------------------------------------------------- */
/* free Work object */
/* ---------------------------------------------------------------------- */
/* (4) After numerical factorization all of the objects allocated in step
* (1) are freed via UMF_free, except that one object of size n_col+1 is
* kept if there are off-diagonal nonzeros in the last pivot row (can only
* occur for singular or rectangular matrices). This is Work->Upattern,
* which is transfered to Numeric->Upattern if ulen > 0.
*/
DEBUG0 (("malloc: init_count "ID" UMF_malloc_count "ID"\n",
init_count, UMF_malloc_count)) ;
free_work (Work) ;
DEBUG0 (("malloc: init_count "ID" UMF_malloc_count "ID"\n",
init_count, UMF_malloc_count)) ;
DEBUG0 (("Numeric->ulen: "ID" scale: "ID"\n", ulen, scale)) ;
ASSERT (UMF_malloc_count == init_count + (ulen > 0) +
(11 + (scale != UMFPACK_SCALE_NONE))) ;
/* ---------------------------------------------------------------------- */
/* reduce Lpos, Lilen, Lip, Upos, Uilen and Uip to size npiv+1 */
/* ---------------------------------------------------------------------- */
/* (5) Six components of the Numeric object are reduced in size if the
* matrix is singular or rectangular. The original size is 3*(n_row+1) +
* 3*(n_col+1) integers. The new size is 6*(npiv+1) integers. For
* square non-singular matrices, these two sizes are the same.
*/
if (npiv < n_row)
{
/* reduce Lpos, Uilen, and Uip from size n_row+1 to size npiv */
inew = (Int *) UMF_realloc (Numeric->Lpos, npiv+1, sizeof (Int)) ;
if (inew)
{
Numeric->Lpos = inew ;
}
inew = (Int *) UMF_realloc (Numeric->Uilen, npiv+1, sizeof (Int)) ;
if (inew)
{
Numeric->Uilen = inew ;
}
inew = (Int *) UMF_realloc (Numeric->Uip, npiv+1, sizeof (Int)) ;
if (inew)
{
Numeric->Uip = inew ;
}
}
if (npiv < n_col)
{
/* reduce Upos, Lilen, and Lip from size n_col+1 to size npiv */
inew = (Int *) UMF_realloc (Numeric->Upos, npiv+1, sizeof (Int)) ;
if (inew)
{
Numeric->Upos = inew ;
}
inew = (Int *) UMF_realloc (Numeric->Lilen, npiv+1, sizeof (Int)) ;
if (inew)
{
Numeric->Lilen = inew ;
}
inew = (Int *) UMF_realloc (Numeric->Lip, npiv+1, sizeof (Int)) ;
if (inew)
{
Numeric->Lip = inew ;
}
}
/* ---------------------------------------------------------------------- */
/* reduce Numeric->Upattern from size n_col+1 to size ulen+1 */
/* ---------------------------------------------------------------------- */
/* (6) The size of Numeric->Upattern (formerly Work->Upattern) is reduced
* from size n_col+1 to size ulen + 1. If ulen is zero, the object does
* not exist. */
DEBUG4 (("ulen: "ID" Upattern "ID"\n", ulen, (Int) Numeric->Upattern)) ;
ASSERT (IMPLIES (ulen == 0, Numeric->Upattern == (Int *) NULL)) ;
if (ulen > 0 && ulen < n_col)
{
inew = (Int *) UMF_realloc (Numeric->Upattern, ulen+1, sizeof (Int)) ;
if (inew)
{
Numeric->Upattern = inew ;
}
}
/* ---------------------------------------------------------------------- */
/* reduce Numeric->Memory to hold just the LU factors at the head */
/* ---------------------------------------------------------------------- */
/* (7) The variable-sized block (Numeric->Memory) is reduced to hold just L
* and U, via a call to UMF_realloc, since the frontal matrices are no
* longer needed.
*/
newsize = Numeric->ihead ;
if (newsize < Numeric->size)
{
mnew = (Unit *) UMF_realloc (Numeric->Memory, newsize, sizeof (Unit)) ;
if (mnew)
{
/* realloc succeeded (how can it fail since the size is reduced?) */
Numeric->Memory = mnew ;
Numeric->size = newsize ;
}
}
Numeric->ihead = Numeric->size ;
Numeric->itail = Numeric->ihead ;
Numeric->tail_usage = 0 ;
Numeric->ibig = EMPTY ;
/* UMF_mem_alloc_tail_block can no longer be called (no tail marker) */
/* ---------------------------------------------------------------------- */
/* report the results and return the Numeric object */
/* ---------------------------------------------------------------------- */
UMF_set_stats (
Info,
Symbolic,
(double) Numeric->max_usage, /* actual peak Numeric->Memory */
(double) Numeric->size, /* actual final Numeric->Memory */
Numeric->flops, /* actual "true flops" */
(double) Numeric->lnz + n_inner, /* actual nz in L */
(double) Numeric->unz + Numeric->nnzpiv, /* actual nz in U */
(double) Numeric->maxfrsize, /* actual largest front size */
(double) ulen, /* actual Numeric->Upattern size */
(double) npiv, /* actual # pivots found */
(double) Numeric->maxnrows, /* actual largest #rows in front */
(double) Numeric->maxncols, /* actual largest #cols in front */
scale != UMFPACK_SCALE_NONE,
Symbolic->prefer_diagonal,
ACTUAL) ;
Info [UMFPACK_ALLOC_INIT_USED] = Numeric->alloc_init ;
Info [UMFPACK_NUMERIC_DEFRAG] = Numeric->ngarbage ;
Info [UMFPACK_NUMERIC_REALLOC] = Numeric->nrealloc ;
Info [UMFPACK_NUMERIC_COSTLY_REALLOC] = Numeric->ncostly ;
Info [UMFPACK_COMPRESSED_PATTERN] = Numeric->isize ;
Info [UMFPACK_LU_ENTRIES] = Numeric->nLentries + Numeric->nUentries +
Numeric->npiv ;
Info [UMFPACK_UDIAG_NZ] = Numeric->nnzpiv ;
Info [UMFPACK_RSMIN] = Numeric->rsmin ;
Info [UMFPACK_RSMAX] = Numeric->rsmax ;
Info [UMFPACK_WAS_SCALED] = Numeric->scale ;
/* nz in L and U with no dropping of small entries */
Info [UMFPACK_ALL_LNZ] = Numeric->all_lnz + n_inner ;
Info [UMFPACK_ALL_UNZ] = Numeric->all_unz + Numeric->nnzpiv ;
Info [UMFPACK_NZDROPPED] =
(Numeric->all_lnz - Numeric->lnz)
+ (Numeric->all_unz - Numeric->unz) ;
/* estimate of the reciprocal of the condition number. */
if (SCALAR_IS_ZERO (Numeric->min_udiag)
|| SCALAR_IS_ZERO (Numeric->max_udiag)
|| SCALAR_IS_NAN (Numeric->min_udiag)
|| SCALAR_IS_NAN (Numeric->max_udiag))
{
/* rcond is zero if there is any zero or NaN on the diagonal */
Numeric->rcond = 0.0 ;
}
else
{
/* estimate of the recipricol of the condition number. */
/* This is NaN if diagonal is zero-free, but has one or more NaN's. */
Numeric->rcond = Numeric->min_udiag / Numeric->max_udiag ;
}
Info [UMFPACK_UMIN] = Numeric->min_udiag ;
Info [UMFPACK_UMAX] = Numeric->max_udiag ;
Info [UMFPACK_RCOND] = Numeric->rcond ;
if (Numeric->nnzpiv < n_inner
|| SCALAR_IS_ZERO (Numeric->rcond) || SCALAR_IS_NAN (Numeric->rcond))
{
/* there are zeros and/or NaN's on the diagonal of U */
DEBUG0 (("Warning, matrix is singular in umfpack_numeric\n")) ;
DEBUG0 (("nnzpiv "ID" n_inner "ID" rcond %g\n", Numeric->nnzpiv,
n_inner, Numeric->rcond)) ;
status = UMFPACK_WARNING_singular_matrix ;
Info [UMFPACK_STATUS] = status ;
}
Numeric->valid = NUMERIC_VALID ;
*NumericHandle = (void *) Numeric ;
/* Numeric has 11 to 13 objects */
ASSERT (UMF_malloc_count == init_count + 11 +
+ (ulen > 0) /* Numeric->Upattern */
+ (scale != UMFPACK_SCALE_NONE)) ; /* Numeric->Rs */
/* ---------------------------------------------------------------------- */
/* get the time used by UMFPACK_numeric */
/* ---------------------------------------------------------------------- */
umfpack_toc (stats) ;
Info [UMFPACK_NUMERIC_WALLTIME] = stats [0] ;
Info [UMFPACK_NUMERIC_TIME] = stats [1] ;
/* return UMFPACK_OK or UMFPACK_WARNING_singular_matrix */
return (status) ;
}
// Miner loop to manage a single processor (with possibly multiple threads per processor)
void minerloop_scanhash(struct thr_info *mythr)
{
struct cgpu_info *cgpu = mythr->cgpu;
struct device_drv *api = cgpu->drv;
struct timeval tv_start, tv_end;
struct timeval tv_hashes, tv_worktime;
uint32_t max_nonce = api->can_limit_work ? api->can_limit_work(mythr) : 0xffffffff;
int64_t hashes;
struct work *work;
const bool primary = (!mythr->device_thread) || mythr->primary_thread;
#ifdef HAVE_PTHREAD_CANCEL
pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL);
#endif
while (likely(!cgpu->shutdown)) {
mythr->work_restart = false;
request_work(mythr);
work = get_and_prepare_work(mythr);
if (!work)
break;
timer_set_now(&work->tv_work_start);
do {
thread_reportin(mythr);
/* Only allow the mining thread to be cancelled when
* it is not in the driver code. */
pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL);
timer_set_now(&tv_start);
hashes = api->scanhash(mythr, work, work->blk.nonce + max_nonce);
timer_set_now(&tv_end);
pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL);
pthread_testcancel();
thread_reportin(mythr);
timersub(&tv_end, &tv_start, &tv_hashes);
if (!hashes_done(mythr, hashes, &tv_hashes, api->can_limit_work ? &max_nonce : NULL))
goto disabled;
if (unlikely(mythr->work_restart)) {
/* Apart from device_thread 0, we stagger the
* starting of every next thread to try and get
* all devices busy before worrying about
* getting work for their extra threads */
if (!primary) {
struct timespec rgtp;
rgtp.tv_sec = 0;
rgtp.tv_nsec = 250 * mythr->device_thread * 1000000;
nanosleep(&rgtp, NULL);
}
break;
}
if (unlikely(mythr->pause || cgpu->deven != DEV_ENABLED))
disabled:
mt_disable(mythr);
timersub(&tv_end, &work->tv_work_start, &tv_worktime);
} while (!abandon_work(work, &tv_worktime, cgpu->max_hashes));
free_work(work);
}
}
static int64_t serial_fpga_scanwork(struct thr_info *thr)
{
struct cgpu_info *serial_fpga;
int fd;
int ret;
struct FPGA_INFO *info;
unsigned char ob_bin[44], nonce_buf[SERIAL_READ_SIZE];
char *ob_hex;
uint32_t nonce;
int64_t hash_count;
struct timeval tv_start, tv_finish, elapsed, tv_end, diff;
int curr_hw_errors, i, j;
uint32_t * ob;
ob = (uint32_t *)ob_bin;
int count;
double Hs, W, fullnonce;
int read_count;
int64_t estimate_hashes;
uint32_t values;
int64_t hash_count_range;
struct work *work;
applog(LOG_DEBUG, "serial_fpga_scanwork...");
if (thr->cgpu->deven == DEV_DISABLED)
return -1;
serial_fpga = thr->cgpu;
info = serial_fpga->device_data;
work = get_work(thr, thr->id);
if (info->device_fd == -1) {
applog(LOG_INFO, "Attemping to Reopen Serial FPGA on %s", serial_fpga->device_path);
fd = serial_open(serial_fpga->device_path, SERIAL_IO_SPEED, SERIAL_READ_TIMEOUT, false);
if (unlikely(-1 == fd)) {
applog(LOG_ERR, "Failed to open Serial FPGA on %s",
serial_fpga->device_path);
return -1;
}
else
info->device_fd = fd;
}
fd = info->device_fd;
memset(ob_bin, 0, sizeof(ob_bin));
// Currently, extra nonces are not supported
//
memset((unsigned char*)work->data + 144, 0, 12);
//
//
calc_midstate(work);
memcpy(ob_bin, work->midstate, 32); // Midstate
memcpy(ob_bin + 32, work->data + 128, 12); // Remaining Bytes From Block Header
// Send Bytes To FPGA In Reverse Order
unsigned char swap[44];
uint32_t * sw;
sw = (uint32_t *)swap;
for (j=0; j<8; j++) {
sw[j] = swab32(ob[j]);
}
memcpy(swap + 32, ob_bin + 32, 12);
for (j=0; j<44; j++) {
ob_bin[j] = swap[j];
}
//unsigned char* b = (unsigned char*)(ob_bin);
//applog(LOG_WARNING, "swap: %02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x", b[28],b[29],b[30],b[31],b[32],b[33],b[34],b[35],b[36],b[37],b[38],b[39],b[40],b[41],b[42],b[43]);
//applog(LOG_WARNING, "swap: %02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x", b[0],b[1],b[2],b[3],b[4],b[5],b[6],b[7],b[8],b[9],b[10],b[11],b[12],b[13],b[14],b[15],b[16],b[17],b[18],b[19],b[20],b[21],b[22],b[23],b[24],b[25],b[26],b[27],b[28],b[29],b[30],b[31],b[32],b[33],b[34],b[35],b[36],b[37],b[38],b[39],b[40],b[41],b[42],b[43]);
//#ifndef WIN32
// tcflush(fd, TCOFLUSH);
//#endif
// Send Data To FPGA
ret = write(fd, ob_bin, sizeof(ob_bin));
if (ret != sizeof(ob_bin)) {
applog(LOG_ERR, "%s%i: Serial Send Error (ret=%d)", serial_fpga->drv->name, serial_fpga->device_id, ret);
serial_fpga_close(thr);
dev_error(serial_fpga, REASON_DEV_COMMS_ERROR);
return 0;
}
if (opt_debug) {
ob_hex = bin2hex(ob_bin, sizeof(ob_bin));
applog(LOG_DEBUG, "Serial FPGA %d sent: %s",
serial_fpga->device_id, ob_hex);
free(ob_hex);
}
elapsed.tv_sec = 0;
elapsed.tv_usec = 0;
cgtime(&tv_start);
applog(LOG_DEBUG, "%s%i: Begin Scan For Nonces", serial_fpga->drv->name, serial_fpga->device_id);
while (thr && !thr->work_restart) {
memset(nonce_buf,0,4);
// Check Serial Port For 1/10 Sec For Nonce
ret = read(fd, nonce_buf, SERIAL_READ_SIZE);
// Calculate Elapsed Time
cgtime(&tv_end);
timersub(&tv_end, &tv_start, &elapsed);
if (ret == 0) { // No Nonce Found
if (elapsed.tv_sec > info->timeout) {
applog(LOG_DEBUG, "%s%i: End Scan For Nonces - Time = %d sec", serial_fpga->drv->name, serial_fpga->device_id, elapsed.tv_sec);
break;
}
continue;
}
else if (ret < SERIAL_READ_SIZE) {
applog(LOG_ERR, "%s%i: Serial Read Error (ret=%d)", serial_fpga->drv->name, serial_fpga->device_id, ret);
serial_fpga_close(thr);
dev_error(serial_fpga, REASON_DEV_COMMS_ERROR);
break;
}
memcpy((char *)&nonce, nonce_buf, SERIAL_READ_SIZE);
#if !defined (__BIG_ENDIAN__) && !defined(MIPSEB)
nonce = swab32(nonce);
#endif
curr_hw_errors = serial_fpga->hw_errors;
applog(LOG_INFO, "%s%i: Nonce Found - %08X (%5.1fMhz)", serial_fpga->drv->name, serial_fpga->device_id, nonce, (double)(1/(info->Hs * 1000000)));
submit_nonce(thr, work, nonce);
// Update Hashrate
if (serial_fpga->hw_errors == curr_hw_errors)
info->Hs = ((double)(elapsed.tv_sec) + ((double)(elapsed.tv_usec))/((double)1000000)) / (double)nonce;
}
// Estimate Number Of Hashes
hash_count = ((double)(elapsed.tv_sec) + ((double)(elapsed.tv_usec))/((double)1000000)) / info->Hs;
free_work(work);
return hash_count;
}
