int
alanRun(alanContext *actx) {
char me[]="alanRun", err[BIFF_STRLEN];
int tid, hack=AIR_FALSE;
alanTask task[ALAN_THREAD_MAX];
if (_alanCheck(actx)) {
sprintf(err, "%s: ", me);
biffAdd(ALAN, err); return 1;
}
if (!( actx->_nlev[0] && actx->_nlev[0] )) {
sprintf(err, "%s: _nlev[0,1] not allocated: "
"call alanUpdate + alanInit", me);
biffAdd(ALAN, err); return 1;
}
if (!airThreadCapable && 1 == actx->numThreads) {
hack = airThreadNoopWarning;
airThreadNoopWarning = AIR_FALSE;
}
actx->changeMutex = airThreadMutexNew();
actx->iterBarrier = airThreadBarrierNew(actx->numThreads);
actx->averageChange = 0;
actx->changeCount = 0;
actx->stop = alanStopNot;
for (tid=0; tid<actx->numThreads; tid++) {
task[tid].actx = actx;
task[tid].idx = tid;
task[tid].thread = airThreadNew();
airThreadStart(task[tid].thread, _alanTuringWorker,
(void *)&(task[tid]));
}
for (tid=0; tid<actx->numThreads; tid++) {
airThreadJoin(task[tid].thread, &(task[tid].me));
task[tid].thread = airThreadNix(task[tid].thread);
}
actx->iterBarrier = airThreadBarrierNix(actx->iterBarrier);
actx->changeMutex = airThreadMutexNix(actx->changeMutex);
if (!airThreadCapable && 1 == actx->numThreads) {
airThreadNoopWarning = hack;
}
/* we assume that someone set actx->stop */
return 0;
}
int
pushStart(pushContext *pctx) {
char me[]="pushStart", err[BIFF_STRLEN];
unsigned int tidx;
if (_pushContextCheck(pctx)) {
sprintf(err, "%s: trouble", me);
biffAdd(PUSH, err); return 1;
}
airSrandMT(pctx->seed);
/* the ordering of things below is important: gage and fiber contexts
have to be set up before they're copied by task setup */
if (_pushTensorFieldSetup(pctx)
|| _pushGageSetup(pctx)
|| _pushFiberSetup(pctx)
|| _pushTaskSetup(pctx)
|| _pushBinSetup(pctx)
|| _pushThingSetup(pctx)) {
sprintf(err, "%s: trouble setting up context", me);
biffAdd(PUSH, err); return 1;
}
/* HEY: this should be done by the user */
pctx->process[0] = _pushForce;
pctx->process[1] = _pushUpdate;
pctx->finished = AIR_FALSE;
if (pctx->numThread > 1) {
pctx->binMutex = airThreadMutexNew();
pctx->stageBarrierA = airThreadBarrierNew(pctx->numThread);
pctx->stageBarrierB = airThreadBarrierNew(pctx->numThread);
}
/* start threads 1 and up running; they'll all hit stageBarrierA */
for (tidx=1; tidx<pctx->numThread; tidx++) {
if (pctx->verbose > 1) {
fprintf(stderr, "%s: spawning thread %d\n", me, tidx);
}
airThreadStart(pctx->task[tidx]->thread, _pushWorker,
(void *)(pctx->task[tidx]));
}
return 0;
}
int
pullStart(pullContext *pctx) {
char me[]="pullStart", err[BIFF_STRLEN];
unsigned int tidx;
fprintf(stderr, "!%s: hello %p\n", me, pctx);
pctx->iter = 0; /* have to initialize this here because of seedOnly hack */
/* the ordering of steps below is important! e.g. gage context has
to be set up (_pullVolumeSetup) by before its copied (_pullTaskSetup) */
if (_pullContextCheck(pctx)
|| _pullVolumeSetup(pctx)
|| _pullInfoSetup(pctx)
|| _pullTaskSetup(pctx)
|| _pullBinSetup(pctx)
|| _pullPointSetup(pctx)) {
sprintf(err, "%s: trouble setting up context", me);
biffAdd(PULL, err); return 1;
}
fprintf(stderr, "!%s: setup done-ish\n", me);
if (pctx->threadNum > 1) {
pctx->binMutex = airThreadMutexNew();
pctx->iterBarrierA = airThreadBarrierNew(pctx->threadNum);
pctx->iterBarrierB = airThreadBarrierNew(pctx->threadNum);
/* start threads 1 and up running; they'll all hit iterBarrierA */
for (tidx=1; tidx<pctx->threadNum; tidx++) {
if (pctx->verbose > 1) {
fprintf(stderr, "%s: spawning thread %d\n", me, tidx);
}
airThreadStart(pctx->task[tidx]->thread, _pullWorker,
(void *)(pctx->task[tidx]));
}
} else {
pctx->binMutex = NULL;
pctx->iterBarrierA = NULL;
pctx->iterBarrierB = NULL;
}
pctx->timeIteration = 0;
pctx->timeRun = 0;
return 0;
}
/*
******** hooverRender()
**
** because of the biff usage(), only one thread can call hooverRender(),
** and no promises if the threads themselves call biff...
*/
int
hooverRender(hooverContext *ctx, int *errCodeP, int *errThreadP) {
char me[]="hooverRender", err[BIFF_STRLEN];
_hooverExtraContext *ec;
_hooverThreadArg args[HOOVER_THREAD_MAX];
_hooverThreadArg *errArg;
airThread *thread[HOOVER_THREAD_MAX];
_htpu u;
void *render;
int ret;
airArray *mop;
int threadIdx;
if (!( errCodeP && errThreadP )) {
sprintf(err, "%s: got NULL int return pointer", me);
biffAdd(HOOVER, err);
return hooverErrInit;
}
/* this calls limnCameraUpdate() */
if (hooverContextCheck(ctx)) {
sprintf(err, "%s: problem detected in given context", me);
biffAdd(HOOVER, err);
*errCodeP = 0;
*errThreadP = 0;
return hooverErrInit;
}
if (!(ec = _hooverExtraContextNew(ctx))) {
sprintf(err, "%s: problem creating thread context", me);
biffAdd(HOOVER, err);
*errCodeP = 0;
*errThreadP = 0;
return hooverErrInit;
}
mop = airMopNew();
airMopAdd(mop, ec, (airMopper)_hooverExtraContextNix, airMopAlways);
if ( (ret = (ctx->renderBegin)(&render, ctx->user)) ) {
*errCodeP = ret;
*errCodeP = 0;
*errThreadP = 0;
airMopError(mop);
return hooverErrRenderBegin;
}
for (threadIdx=0; threadIdx<ctx->numThreads; threadIdx++) {
args[threadIdx].ctx = ctx;
args[threadIdx].ec = ec;
args[threadIdx].render = render;
args[threadIdx].whichThread = threadIdx;
args[threadIdx].whichErr = hooverErrNone;
args[threadIdx].errCode = 0;
thread[threadIdx] = airThreadNew();
}
ctx->workIdx = 0;
if (1 < ctx->numThreads) {
ctx->workMutex = airThreadMutexNew();
} else {
ctx->workMutex = NULL;
}
/* (done): call airThreadStart() once per thread, passing the
address of a distinct (and appropriately intialized)
_hooverThreadArg to each. If return of airThreadStart() is
non-zero, put its return in *errCodeP, the number of the
problematic in *errThreadP, and return hooverErrThreadCreate.
Then call airThreadJoin() on all the threads, passing &errArg as
"retval". On non-zero return, set *errCodeP and *errThreadP,
and return hooverErrThreadJoin. If return of airThreadJoin() is
zero, but the errArg is non-NULL, then assume that this errArg
is actually just the passed _hooverThreadArg returned to us, and
from this copy errArg->errCode into *errCodeP, and return
errArg->whichErr */
if (1 < ctx->numThreads && !airThreadCapable) {
fprintf(stderr, "%s: WARNING: not multi-threaded; will do %d "
"\"threads\" serially !!!\n", me, ctx->numThreads);
}
for (threadIdx=0; threadIdx<ctx->numThreads; threadIdx++) {
if ((ret = airThreadStart(thread[threadIdx], _hooverThreadBody,
(void *) &args[threadIdx]))) {
*errCodeP = ret;
*errThreadP = threadIdx;
airMopError(mop);
return hooverErrThreadCreate;
}
}
for (threadIdx=0; threadIdx<ctx->numThreads; threadIdx++) {
u.h = &errArg;
if ((ret = airThreadJoin(thread[threadIdx], u.v))) {
*errCodeP = ret;
*errThreadP = threadIdx;
airMopError(mop);
return hooverErrThreadJoin;
}
if (errArg != NULL) {
*errCodeP = errArg->errCode;
*errThreadP = threadIdx;
return errArg->whichErr;
}
thread[threadIdx] = airThreadNix(thread[threadIdx]);
}
if (1 < ctx->numThreads) {
ctx->workMutex = airThreadMutexNix(ctx->workMutex);
}
if ( (ret = (ctx->renderEnd)(render, ctx->user)) ) {
*errCodeP = ret;
*errThreadP = -1;
return hooverErrRenderEnd;
}
render = NULL;
airMopOkay(mop);
*errCodeP = 0;
*errThreadP = 0;
return hooverErrNone;
}
int
coilStart(coilContext *cctx) {
char me[]="coilStart", err[BIFF_STRLEN];
int valIdx, valLen;
coil_t (*lup)(const void*, size_t), *val;
unsigned tidx, elIdx;
if (!cctx) {
sprintf(err, "%s: got NULL pointer", me);
biffAdd(COIL, err); return 1;
}
cctx->task = (coilTask **)calloc(cctx->numThreads, sizeof(coilTask *));
if (!(cctx->task)) {
sprintf(err, "%s: couldn't allocate array of tasks", me);
biffAdd(COIL, err); return 1;
}
/* we create tasks for ALL threads, including me, thread 0 */
cctx->task[0] = NULL;
for (tidx=0; tidx<cctx->numThreads; tidx++) {
cctx->task[tidx] = _coilTaskNew(cctx, tidx);
if (!(cctx->task[tidx])) {
sprintf(err, "%s: couldn't allocate task %d", me, tidx);
biffAdd(COIL, err); return 1;
}
}
cctx->finished = AIR_FALSE;
if (cctx->numThreads > 1) {
cctx->nextSliceMutex = airThreadMutexNew();
cctx->filterBarrier = airThreadBarrierNew(cctx->numThreads);
cctx->updateBarrier = airThreadBarrierNew(cctx->numThreads);
}
/* initialize the values in cctx->nvol */
val = (coil_t*)(cctx->nvol->data);
valLen = cctx->kind->valLen;
#if COIL_TYPE_FLOAT
lup = nrrdFLookup[cctx->nin->type];
#else
lup = nrrdDLookup[cctx->nin->type];
#endif
for (elIdx=0; elIdx<cctx->size[0]*cctx->size[1]*cctx->size[2]; elIdx++) {
for (valIdx=0; valIdx<valLen; valIdx++) {
val[valIdx + 0*valLen] = lup(cctx->nin->data, valIdx + valLen*elIdx);
val[valIdx + 1*valLen] = 0;
}
val += 2*valLen;
}
/* start threads 1 and up running; they'll all hit filterBarrier */
if (cctx->numThreads > 1) {
for (tidx=1; tidx<cctx->numThreads; tidx++) {
if (cctx->verbose > 1) {
fprintf(stderr, "%s: spawning thread %d\n", me, tidx);
}
airThreadStart(cctx->task[tidx]->thread, _coilWorker,
(void *)(cctx->task[tidx]));
}
}
/* set things as though we've just finished an update phase */
cctx->nextSlice = cctx->size[2];
cctx->todoFilter = AIR_TRUE;
cctx->todoUpdate = AIR_FALSE;
return 0;
}
