/*
** this is the core of the worker threads: as long as there are bins
** left to process, get the next one, and process it
*/
int
_pullProcess(pullTask *task) {
char me[]="_pullProcess", err[BIFF_STRLEN];
unsigned int binIdx;
task->stuckNum = 0;
while (task->pctx->binNextIdx < task->pctx->binNum) {
/* get the index of the next bin to process */
if (task->pctx->threadNum > 1) {
airThreadMutexLock(task->pctx->binMutex);
}
/* note that we entirely skip bins with no points */
do {
binIdx = task->pctx->binNextIdx;
if (task->pctx->binNextIdx < task->pctx->binNum) {
task->pctx->binNextIdx++;
}
} while (binIdx < task->pctx->binNum
&& 0 == task->pctx->bin[binIdx].pointNum);
if (task->pctx->threadNum > 1) {
airThreadMutexUnlock(task->pctx->binMutex);
}
if (binIdx == task->pctx->binNum) {
/* no more bins to process! */
break;
}
if (pullBinProcess(task, binIdx)) {
sprintf(err, "%s(%u): had trouble on bin %u", me,
task->threadIdx, binIdx);
biffAdd(PULL, err); return 1;
}
}
return 0;
}
/*
** this is run once per task (thread), per stage
*/
int
_pushStageRun(pushTask *task, int stageIdx) {
char me[]="_pushStageRun", err[BIFF_STRLEN];
unsigned int binIdx;
while (task->pctx->binIdx < task->pctx->numBin) {
if (task->pctx->numThread > 1) {
airThreadMutexLock(task->pctx->binMutex);
}
do {
binIdx = task->pctx->binIdx;
if (task->pctx->binIdx < task->pctx->numBin) {
task->pctx->binIdx++;
}
} while (binIdx < task->pctx->numBin
&& 0 == task->pctx->bin[binIdx].numThing
&& 0 == task->pctx->bin[binIdx].numPoint);
if (task->pctx->numThread > 1) {
airThreadMutexUnlock(task->pctx->binMutex);
}
if (binIdx == task->pctx->numBin) {
/* no more bins to process */
break;
}
if (task->pctx->process[stageIdx](task, binIdx,
task->pctx->stageParm[stageIdx])) {
sprintf(err, "%s(%d): had trouble running stage %d", me,
task->threadIdx, stageIdx);
biffAdd(PUSH, err); return 1;
}
}
return 0;
}
int
_coilThisZGet(coilTask *task, int doFilter) {
int thisZ, *thisFlag, *thatFlag;
if (doFilter) {
thisFlag = &(task->cctx->todoFilter);
thatFlag = &(task->cctx->todoUpdate);
} else {
thisFlag = &(task->cctx->todoUpdate);
thatFlag = &(task->cctx->todoFilter);
}
if (task->cctx->numThreads > 1) {
airThreadMutexLock(task->cctx->nextSliceMutex);
}
if (task->cctx->nextSlice == task->cctx->size[2]
&& *thisFlag) {
/* we're the first thread to start this phase */
task->cctx->nextSlice = 0;
*thisFlag = AIR_FALSE;
}
thisZ = task->cctx->nextSlice;
if (task->cctx->nextSlice < task->cctx->size[2]) {
task->cctx->nextSlice++;
if (task->cctx->nextSlice == task->cctx->size[2]) {
/* we just grabbed the last slice of this phase */
*thatFlag = AIR_TRUE;
}
}
if (task->cctx->numThreads > 1) {
airThreadMutexUnlock(task->cctx->nextSliceMutex);
}
return thisZ;
}
int
airThreadBarrierWait(airThreadBarrier *barrier) {
airThreadMutexLock(barrier->doneMutex);
barrier->numDone += 1;
if (barrier->numDone < barrier->numUsers) {
airThreadCondWait(barrier->doneCond, barrier->doneMutex);
} else {
barrier->numDone = 0;
airThreadCondBroadcast(barrier->doneCond);
}
airThreadMutexUnlock(barrier->doneMutex);
return 0;
}
void *
_hooverThreadBody(void *_arg) {
_hooverThreadArg *arg;
void *thread;
int ret, /* to catch return values from callbacks */
sampleI, /* which sample we're on */
inside, /* we're inside the volume */
vI, uI; /* integral coords in image */
double tmp,
mm, /* lowest position in index space, for all axes */
Mx, My, Mz, /* highest position in index space on each axis */
u, v, /* floating-point coords in image */
uvScale, /* how to scale (u,v) to go from image to
near plane, according to ortho or perspective */
lx, ly, lz, /* half edge-lengths of volume */
rayLen=0, /* length of segment formed by ray line intersecting
the near and far clipping planes */
rayT, /* current position along ray (world-space) */
rayDirW[3], /* unit-length ray direction (world-space) */
rayDirI[3], /* rayDirW transformed into index space;
not unit length, but a unit change in
world space along rayDirW translates to
this change in index space along rayDirI */
rayPosW[3], /* current ray location (world-space) */
rayPosI[3], /* current ray location (index-space) */
rayStartW[3], /* ray start on near plane (world-space) */
rayStartI[3], /* ray start on near plane (index-space) */
rayStep, /* distance between samples (world-space) */
vOff[3], uOff[3]; /* offsets in arg->ec->wU and arg->ec->wV
directions towards start of ray */
arg = (_hooverThreadArg *)_arg;
if ( (ret = (arg->ctx->threadBegin)(&thread,
arg->render,
arg->ctx->user,
arg->whichThread)) ) {
arg->errCode = ret;
arg->whichErr = hooverErrThreadBegin;
return arg;
}
lx = arg->ec->volHLen[0];
ly = arg->ec->volHLen[1];
lz = arg->ec->volHLen[2];
if (nrrdCenterNode == arg->ctx->volCentering) {
mm = 0;
Mx = arg->ctx->volSize[0]-1;
My = arg->ctx->volSize[1]-1;
Mz = arg->ctx->volSize[2]-1;
} else {
mm = -0.5;
Mx = arg->ctx->volSize[0]-0.5;
My = arg->ctx->volSize[1]-0.5;
Mz = arg->ctx->volSize[2]-0.5;
}
if (arg->ctx->cam->orthographic) {
ELL_3V_COPY(rayDirW, arg->ctx->cam->N);
rayDirI[0] = AIR_DELTA(-lx, rayDirW[0], lx, mm, Mx);
rayDirI[1] = AIR_DELTA(-ly, rayDirW[1], ly, mm, My);
rayDirI[2] = AIR_DELTA(-lz, rayDirW[2], lz, mm, Mz);
rayLen = arg->ctx->cam->vspFaar - arg->ctx->cam->vspNeer;
uvScale = 1.0;
} else {
uvScale = arg->ctx->cam->vspNeer/arg->ctx->cam->vspDist;
}
while (1) {
/* the work assignment is simply the next scanline to be rendered:
the result of all this is setting vI */
if (arg->ctx->workMutex) {
airThreadMutexLock(arg->ctx->workMutex);
}
vI = arg->ctx->workIdx;
if (arg->ctx->workIdx < arg->ctx->imgSize[1]) {
arg->ctx->workIdx += 1;
}
if (arg->ctx->workMutex) {
airThreadMutexUnlock(arg->ctx->workMutex);
}
if (vI == arg->ctx->imgSize[1]) {
/* we're done! */
break;
}
if (nrrdCenterCell == arg->ctx->imgCentering) {
v = uvScale*AIR_AFFINE(-0.5, vI, arg->ctx->imgSize[1]-0.5,
arg->ctx->cam->vRange[0],
arg->ctx->cam->vRange[1]);
} else {
v = uvScale*AIR_AFFINE(0.0, vI, arg->ctx->imgSize[1]-1.0,
arg->ctx->cam->vRange[0],
arg->ctx->cam->vRange[1]);
}
ELL_3V_SCALE(vOff, v, arg->ctx->cam->V);
for (uI=0; uI<arg->ctx->imgSize[0]; uI++) {
if (nrrdCenterCell == arg->ctx->imgCentering) {
u = uvScale*AIR_AFFINE(-0.5, uI, arg->ctx->imgSize[0]-0.5,
arg->ctx->cam->uRange[0],
arg->ctx->cam->uRange[1]);
} else {
u = uvScale*AIR_AFFINE(0.0, uI, arg->ctx->imgSize[0]-1.0,
arg->ctx->cam->uRange[0],
arg->ctx->cam->uRange[1]);
}
ELL_3V_SCALE(uOff, u, arg->ctx->cam->U);
ELL_3V_ADD3(rayStartW, uOff, vOff, arg->ec->rayZero);
rayStartI[0] = AIR_AFFINE(-lx, rayStartW[0], lx, mm, Mx);
rayStartI[1] = AIR_AFFINE(-ly, rayStartW[1], ly, mm, My);
rayStartI[2] = AIR_AFFINE(-lz, rayStartW[2], lz, mm, Mz);
if (!arg->ctx->cam->orthographic) {
ELL_3V_SUB(rayDirW, rayStartW, arg->ctx->cam->from);
ELL_3V_NORM(rayDirW, rayDirW, tmp);
rayDirI[0] = AIR_DELTA(-lx, rayDirW[0], lx, mm, Mx);
rayDirI[1] = AIR_DELTA(-ly, rayDirW[1], ly, mm, My);
rayDirI[2] = AIR_DELTA(-lz, rayDirW[2], lz, mm, Mz);
rayLen = ((arg->ctx->cam->vspFaar - arg->ctx->cam->vspNeer)/
ELL_3V_DOT(rayDirW, arg->ctx->cam->N));
}
if ( (ret = (arg->ctx->rayBegin)(thread,
arg->render,
arg->ctx->user,
uI, vI, rayLen,
rayStartW, rayStartI,
rayDirW, rayDirI)) ) {
arg->errCode = ret;
arg->whichErr = hooverErrRayBegin;
return arg;
}
sampleI = 0;
rayT = 0;
while (1) {
ELL_3V_SCALE_ADD2(rayPosW, 1.0, rayStartW, rayT, rayDirW);
ELL_3V_SCALE_ADD2(rayPosI, 1.0, rayStartI, rayT, rayDirI);
inside = (AIR_IN_CL(mm, rayPosI[0], Mx) &&
AIR_IN_CL(mm, rayPosI[1], My) &&
AIR_IN_CL(mm, rayPosI[2], Mz));
rayStep = (arg->ctx->sample)(thread,
arg->render,
arg->ctx->user,
sampleI, rayT,
inside,
rayPosW, rayPosI);
if (!AIR_EXISTS(rayStep)) {
/* sampling failed */
arg->errCode = 0;
arg->whichErr = hooverErrSample;
return arg;
}
if (!rayStep) {
/* ray decided to finish itself */
break;
}
/* else we moved to a new location along the ray */
rayT += rayStep;
if (!AIR_IN_CL(0, rayT, rayLen)) {
/* ray stepped outside near-far clipping region, its done. */
break;
}
sampleI++;
}
if ( (ret = (arg->ctx->rayEnd)(thread,
arg->render,
arg->ctx->user)) ) {
arg->errCode = ret;
arg->whichErr = hooverErrRayEnd;
return arg;
}
} /* end this scanline */
} /* end while(1) assignment of scanlines */
if ( (ret = (arg->ctx->threadEnd)(thread,
arg->render,
arg->ctx->user)) ) {
arg->errCode = ret;
arg->whichErr = hooverErrThreadEnd;
return arg;
}
/* returning NULL actually indicates that there was NOT an error */
return NULL;
}
void *
_alanTuringWorker(void *_task) {
alan_t *tendata, *ten, react,
conf, Dxx, Dxy, Dyy, /* Dxz, Dyz, */
*tpx, *tmx, *tpy, *tmy, /* *tpz, *tmz, */
*lev0, *lev1, *parm, deltaT, alpha, beta, A, B,
*v[27], lapA, lapB, corrA, corrB,
deltaA, deltaB, diffA, diffB, change;
int dim, iter, stop, startW, endW, idx,
px, mx, py, my, pz, mz,
startY, endY, startZ, endZ, sx, sy, sz, x, y, z;
alanTask *task;
task = (alanTask *)_task;
dim = task->actx->dim;
sx = task->actx->size[0];
sy = task->actx->size[1];
sz = (2 == dim ? 1 : task->actx->size[2]);
parm = (alan_t*)(task->actx->nparm->data);
diffA = AIR_CAST(alan_t, task->actx->diffA/pow(task->actx->deltaX, dim));
diffB = AIR_CAST(alan_t, task->actx->diffB/pow(task->actx->deltaX, dim));
startW = task->idx*sy/task->actx->numThreads;
endW = (task->idx+1)*sy/task->actx->numThreads;
tendata = task->actx->nten ? (alan_t *)task->actx->nten->data : NULL;
react = task->actx->react;
if (2 == dim) {
startZ = 0;
endZ = 1;
startY = startW;
endY = endW;
} else {
startZ = startW;
endZ = endW;
startY = 0;
endY = sy;
}
for (iter = 0;
(alanStopNot == task->actx->stop
&& (0 == task->actx->maxIteration
|| iter < task->actx->maxIteration));
iter++) {
if (0 == task->idx) {
task->actx->iter = iter;
task->actx->nlev = task->actx->_nlev[(iter+1) % 2];
}
lev0 = (alan_t*)(task->actx->_nlev[iter % 2]->data);
lev1 = (alan_t*)(task->actx->_nlev[(iter+1) % 2]->data);
stop = alanStopNot;
change = 0;
conf = 1; /* if you have no data; this will stay 1 */
for (z = startZ; z < endZ; z++) {
if (task->actx->wrap) {
pz = AIR_MOD(z+1, sz);
mz = AIR_MOD(z-1, sz);
} else {
pz = AIR_MIN(z+1, sz-1);
mz = AIR_MAX(z-1, 0);
}
for (y = startY; y < endY; y++) {
if (task->actx->wrap) {
py = AIR_MOD(y+1, sy);
my = AIR_MOD(y-1, sy);
} else {
py = AIR_MIN(y+1, sy-1);
my = AIR_MAX(y-1, 0);
}
for (x = 0; x < sx; x++) {
if (task->actx->wrap) {
px = AIR_MOD(x+1, sx);
mx = AIR_MOD(x-1, sx);
} else {
px = AIR_MIN(x+1, sx-1);
mx = AIR_MAX(x-1, 0);
}
idx = x + sx*(y + sy*z);
A = lev0[0 + 2*idx];
B = lev0[1 + 2*idx];
deltaT = parm[0 + 3*idx];
alpha = parm[1 + 3*idx];
beta = parm[2 + 3*idx];
lapA = lapB = corrA = corrB = 0;
if (2 == dim) {
/*
** 0 1 2 ----> X
** 3 4 5
** 6 7 8
** |
** v Y
*/
v[1] = lev0 + 2*( x + sx*(my));
v[3] = lev0 + 2*(mx + sx*( y));
v[5] = lev0 + 2*(px + sx*( y));
v[7] = lev0 + 2*( x + sx*(py));
if (tendata) {
/*
** 0 1 2 Dxy/2 Dyy -Dxy/2
** 3 4 5 Dxx -2*(Dxx + Dyy) Dxx
** 6 7 8 -Dxy/2 Dyy Dxy/2
*/
v[0] = lev0 + 2*(mx + sx*(my));
v[2] = lev0 + 2*(px + sx*(my));
v[6] = lev0 + 2*(mx + sx*(py));
v[8] = lev0 + 2*(px + sx*(py));
ten = tendata + 4*idx;
conf = AIR_CAST(alan_t, (AIR_CLAMP(0.3, ten[0], 1) - 0.3)/0.7);
if (conf) {
Dxx = ten[1];
Dxy = ten[2];
Dyy = ten[3];
lapA = (Dxy*(v[0][0] + v[8][0] - v[2][0] - v[6][0])/2
+ Dxx*(v[3][0] + v[5][0]) + Dyy*(v[1][0] + v[7][0])
- 2*(Dxx + Dyy)*A);
lapB = (Dxy*(v[0][1] + v[8][1] - v[2][1] - v[6][1])/2
+ Dxx*(v[3][1] + v[5][1]) + Dyy*(v[1][1] + v[7][1])
- 2*(Dxx + Dyy)*B);
if (!(task->actx->homogAniso)) {
tpx = tendata + 4*(px + sx*( y + sy*( z)));
tmx = tendata + 4*(mx + sx*( y + sy*( z)));
tpy = tendata + 4*( x + sx*(py + sy*( z)));
tmy = tendata + 4*( x + sx*(my + sy*( z)));
corrA = ((tpx[1]-tmx[1])*(v[5][0]-v[3][0])/4+ /* Dxx,x*A,x */
(tpx[2]-tmx[2])*(v[7][0]-v[1][0])/4+ /* Dxy,x*A,y */
(tpy[2]-tmy[2])*(v[5][0]-v[3][0])/4+ /* Dxy,y*A,x */
(tpy[3]-tmy[3])*(v[7][0]-v[1][0])); /* Dyy,y*A,y */
corrB = ((tpx[1]-tmx[1])*(v[5][1]-v[3][1])/4+ /* Dxx,x*B,x */
(tpx[2]-tmx[2])*(v[7][1]-v[1][1])/4+ /* Dxy,x*B,y */
(tpy[2]-tmy[2])*(v[5][1]-v[3][1])/4+ /* Dxy,y*B,x */
(tpy[3]-tmy[3])*(v[7][1]-v[1][1])); /* Dyy,y*B,y */
}
} else {
/* no confidence; you diffuse */
lapA = v[1][0] + v[3][0] + v[5][0] + v[7][0] - 4*A;
lapB = v[1][1] + v[3][1] + v[5][1] + v[7][1] - 4*B;
}
} else {
/* no data; you diffuse */
lapA = v[1][0] + v[3][0] + v[5][0] + v[7][0] - 4*A;
lapB = v[1][1] + v[3][1] + v[5][1] + v[7][1] - 4*B;
}
} else {
/* 3 == dim */
/*
** 0 1 2 ---- X
** 3 4 5
** 6 7 8
** /
** / 9 10 11
** Y 12 13 14
** 15 16 17
**
** 18 19 20
** 21 22 23
** 24 25 26
** |
** |
** Z
*/
v[ 4] = lev0 + 2*( x + sx*( y + sy*(mz)));
v[10] = lev0 + 2*( x + sx*(my + sy*( z)));
v[12] = lev0 + 2*(mx + sx*( y + sy*( z)));
v[14] = lev0 + 2*(px + sx*( y + sy*( z)));
v[16] = lev0 + 2*( x + sx*(py + sy*( z)));
v[22] = lev0 + 2*( x + sx*( y + sy*(pz)));
if (tendata) {
if (!(task->actx->homogAniso)) {
}
} else {
lapA = (v[ 4][0] + v[10][0] + v[12][0]
+ v[14][0] + v[16][0] + v[22][0] - 6*A);
lapB = (v[ 4][1] + v[10][1] + v[12][1]
+ v[14][1] + v[16][1] + v[22][1] - 6*B);
}
}
deltaA = deltaT*(react*conf*task->actx->K*(alpha - A*B)
+ diffA*(lapA + corrA));
if (AIR_ABS(deltaA) > task->actx->maxPixelChange) {
stop = alanStopDiverged;
}
change += AIR_ABS(deltaA);
deltaB = deltaT*(react*conf*task->actx->K*(A*B - B - beta)
+ diffB*(lapB + corrB));
if (!( AIR_EXISTS(deltaA) && AIR_EXISTS(deltaB) )) {
stop = alanStopNonExist;
}
A += deltaA;
B = AIR_MAX(0, B + deltaB);
lev1[0 + 2*idx] = A;
lev1[1 + 2*idx] = B;
}
}
}
/* add change to global sum in a threadsafe way */
airThreadMutexLock(task->actx->changeMutex);
task->actx->averageChange += change/(sx*sy*sz);
task->actx->changeCount += 1;
if (task->actx->changeCount == task->actx->numThreads) {
/* I must be the last thread to reach this point; all
others must have passed the mutex unlock, and are
sitting at the barrier */
if (alanStopNot != stop) {
/* there was some problem in going from lev0 to lev1, which
we deal with now by setting actx->stop */
task->actx->stop = stop;
} else if (task->actx->averageChange < task->actx->minAverageChange) {
/* we converged */
task->actx->stop = alanStopConverged;
} else {
/* we keep going */
_alanPerIteration(task->actx, iter);
if (task->actx->perIteration) {
task->actx->perIteration(task->actx, iter);
}
}
task->actx->averageChange = 0;
task->actx->changeCount = 0;
}
airThreadMutexUnlock(task->actx->changeMutex);
/* force all threads to line up here, once per iteration */
airThreadBarrierWait(task->actx->iterBarrier);
}
if (iter == task->actx->maxIteration) {
/* HEY: all threads will agree on this, right? */
task->actx->stop = alanStopMaxIteration;
}
/* else: the non-alanStopNot value of task->actx->stop made us stop */
return _task;
}
