/// Loads the mesh from the thread-local file stream cache
void loadCompressed(const fs::path& filePath, const int idx) {
if (EXPECT_NOT_TAKEN(idx < 0)) {
Log(EError, "Unable to unserialize mesh, "
"shape index is negative! (requested %i out of 0..%i)", idx);
}
// Get the thread local cache; create it if this is the first time
FileStreamCache* cache = m_cache.get();
if (EXPECT_NOT_TAKEN(cache == NULL)) {
cache = new FileStreamCache();
m_cache.set(cache);
mitsuba::pushSceneCleanupHandler(&SerializedMesh::flushCache);
}
boost::shared_ptr<MeshLoader> meshLoader = cache->get(filePath);
Assert(meshLoader != NULL);
TriMesh::loadCompressed(meshLoader->seekStream((size_t) idx));
}
Vector Warp::squareToCosineHemisphere(const Point2 &sample) {
Point2 p = Warp::squareToUniformDiskConcentric(sample);
Float z = math::safe_sqrt(1.0f - p.x*p.x - p.y*p.y);
/* Guard against numerical imprecisions */
if (EXPECT_NOT_TAKEN(z == 0))
z = 1e-10f;
return Vector(p.x, p.y, z);
}
bool develop(const Point2i &sourceOffset, const Vector2i &size,
const Point2i &targetOffset, Bitmap *target) const {
const Bitmap *source = m_storage->getBitmap();
const FormatConverter *cvt = FormatConverter::getInstance(
std::make_pair(Bitmap::EFloat, target->getComponentFormat())
);
size_t sourceBpp = source->getBytesPerPixel();
size_t targetBpp = target->getBytesPerPixel();
const uint8_t *sourceData = source->getUInt8Data()
+ (sourceOffset.x + sourceOffset.y * source->getWidth()) * sourceBpp;
uint8_t *targetData = target->getUInt8Data()
+ (targetOffset.x + targetOffset.y * target->getWidth()) * targetBpp;
if (EXPECT_NOT_TAKEN(m_pixelFormats.size() != 1)) {
/* Special case for general multi-channel images -- just develop the first component(s) */
for (int i=0; i<size.y; ++i) {
for (int j=0; j<size.x; ++j) {
Float weight = *((Float *) (sourceData + (j+1)*sourceBpp - sizeof(Float)));
Float invWeight = weight != 0 ? ((Float) 1 / weight) : (Float) 0;
cvt->convert(Bitmap::ESpectrum, 1.0f, sourceData + j*sourceBpp,
target->getPixelFormat(), target->getGamma(), targetData + j * targetBpp,
1, invWeight);
}
sourceData += source->getWidth() * sourceBpp;
targetData += target->getWidth() * targetBpp;
}
} else if (size.x == m_cropSize.x && target->getWidth() == m_storage->getWidth()) {
/* Develop a connected part of the underlying buffer */
cvt->convert(source->getPixelFormat(), 1.0f, sourceData,
target->getPixelFormat(), target->getGamma(), targetData,
size.x*size.y);
} else {
/* Develop a rectangular subregion */
for (int i=0; i<size.y; ++i) {
cvt->convert(source->getPixelFormat(), 1.0f, sourceData,
target->getPixelFormat(), target->getGamma(), targetData,
size.x);
sourceData += source->getWidth() * sourceBpp;
targetData += target->getWidth() * targetBpp;
}
}
return true;
}
Vector squareToHemispherePSA(const Point2 &sample) {
Float r = std::sqrt(sample.x);
Float phi = 2.0f * M_PI * sample.y;
Float dirX = r * std::cos(phi);
Float dirY = r * std::sin(phi);
Float z = std::sqrt(1 - std::min((Float) 1, dirX*dirX + dirY*dirY));
if (EXPECT_NOT_TAKEN(z == 0)) {
/* Guard against numerical imprecisions */
return normalize(Vector(
dirX, dirY, Epsilon));
}
return Vector(
dirX, dirY, z
);
}
MTS_NAMESPACE_BEGIN
void Intersection::computePartials(const RayDifferential &ray) {
Float A[2][2], Bx[2], By[2], x[2];
int axes[2];
/* Compute the texture coordinates partials wrt.
changes in the screen-space position. Based on PBRT */
if (hasUVPartials)
return;
hasUVPartials = true;
if (!ray.hasDifferentials || (dpdu.isZero() && dpdv.isZero())) {
dudx = dvdx = dudy = dvdy = 0.0f;
return;
}
/* Offset of the plane passing through the surface */
const Float d = -dot(geoFrame.n, Vector(p));
const Float txRecip = dot(geoFrame.n, ray.rxDirection),
tyRecip = dot(geoFrame.n, ray.ryDirection);
if (EXPECT_NOT_TAKEN(txRecip == 0 || tyRecip == 0)) {
dudx = dvdx = dudy = dvdy = 0.0f;
return;
}
/* Ray distances traveled */
const Float tx = -(dot(geoFrame.n, Vector(ray.rxOrigin)) + d) /
txRecip;
const Float ty = -(dot(geoFrame.n, Vector(ray.ryOrigin)) + d) /
tyRecip;
/* Calculate the U and V partials by solving two out
of a set of 3 equations in an overconstrained system */
Float absX = std::abs(geoFrame.n.x),
absY = std::abs(geoFrame.n.y),
absZ = std::abs(geoFrame.n.z);
if (absX > absY && absX > absZ) {
axes[0] = 1;
axes[1] = 2;
} else if (absY > absZ) {
axes[0] = 0;
axes[1] = 2;
} else {
axes[0] = 0;
axes[1] = 1;
}
A[0][0] = dpdu[axes[0]];
A[0][1] = dpdv[axes[0]];
A[1][0] = dpdu[axes[1]];
A[1][1] = dpdv[axes[1]];
/* Auxilary intersection point of the adjacent rays */
Point px = ray.rxOrigin + ray.rxDirection * tx,
py = ray.ryOrigin + ray.ryDirection * ty;
Bx[0] = px[axes[0]] - p[axes[0]];
Bx[1] = px[axes[1]] - p[axes[1]];
By[0] = py[axes[0]] - p[axes[0]];
By[1] = py[axes[1]] - p[axes[1]];
if (EXPECT_TAKEN(solveLinearSystem2x2(A, Bx, x))) {
dudx = x[0];
dvdx = x[1];
} else {
dudx = 1;
dvdx = 0;
}
if (EXPECT_TAKEN(solveLinearSystem2x2(A, By, x))) {
dudy = x[0];
dvdy = x[1];
} else {
dudy = 0;
dudy = 1;
}
}
Float lookupFloat(const Point &_p) const {
const Point p = m_worldToGrid.transformAffine(_p);
int x = (int) p.x, y = (int) p.y, z = (int) p.z;
if (EXPECT_NOT_TAKEN(
x < 0 || x >= m_cellCount.x ||
y < 0 || y >= m_cellCount.y ||
z < 0 || z >= m_cellCount.z))
return 0.0f;
BlockCache *cache = m_cache.get();
if (EXPECT_NOT_TAKEN(cache == NULL)) {
cache = new BlockCache(m_blocksPerCore,
boost::bind(&CachingDataSource::renderBlock, this, _1),
boost::bind(&CachingDataSource::destroyBlock, this, _1));
m_cache.set(cache);
}
#if defined(VOLCACHE_DEBUG)
if (cache->isFull()) {
/* For debugging: when the cache is full, dump locations
of all cache records into an OBJ file and exit */
std::vector<Vector3i> keys;
cache->get_keys(std::back_inserter(keys));
std::ofstream os("keys.obj");
os << "o Keys" << endl;
for (size_t i=0; i<keys.size(); i++) {
Vector3i key = keys[i];
key = key * m_blockSize + Vector3i(m_blockSize/2);
Point p(key.x * m_voxelWidth + m_aabb.min.x,
key.y * m_voxelWidth + m_aabb.min.y,
key.z * m_voxelWidth + m_aabb.min.z);
os << "v " << p.x << " " << p.y << " " << p.z << endl;
}
/// Need to generate some fake geometry so that blender will import the points
for (size_t i=3; i<=keys.size(); i++)
os << "f " << i << " " << i-1 << " " << i-2 << endl;
os.close();
_exit(-1);
}
#endif
bool hit = false;
float *blockData = cache->get(Vector3i(
(x & m_blockMask) >> m_blockShift,
(y & m_blockMask) >> m_blockShift,
(z & m_blockMask) >> m_blockShift), hit);
statsHitRate.incrementBase();
if (hit)
++statsHitRate;
if (blockData == NULL)
return 0.0f;
const int x1 = x & m_voxelMask, y1 = y & m_voxelMask, z1 = z & m_voxelMask,
x2 = x1 + 1, y2 = y1 + 1, z2 = z1 + 1;
const Float fx = p.x - x, fy = p.y - y, fz = p.z - z,
_fx = 1.0f - fx, _fy = 1.0f - fy, _fz = 1.0f - fz;
const float
&d000 = blockData[(z1*m_blockRes + y1)*m_blockRes + x1],
&d001 = blockData[(z1*m_blockRes + y1)*m_blockRes + x2],
&d010 = blockData[(z1*m_blockRes + y2)*m_blockRes + x1],
&d011 = blockData[(z1*m_blockRes + y2)*m_blockRes + x2],
&d100 = blockData[(z2*m_blockRes + y1)*m_blockRes + x1],
&d101 = blockData[(z2*m_blockRes + y1)*m_blockRes + x2],
&d110 = blockData[(z2*m_blockRes + y2)*m_blockRes + x1],
&d111 = blockData[(z2*m_blockRes + y2)*m_blockRes + x2];
float result = ((d000*_fx + d001*fx)*_fy +
(d010*_fx + d011*fx)*fy)*_fz +
((d100*_fx + d101*fx)*_fy +
(d110*_fx + d111*fx)*fy)*fz;
return result;
}
void PreviewWorker::processCoherent(const WorkUnit *workUnit, WorkResult *workResult,
const bool &stop) {
#if defined(MTS_HAS_COHERENT_RT)
const RectangularWorkUnit *rect = static_cast<const RectangularWorkUnit *>(workUnit);
ImageBlock *block = static_cast<ImageBlock *>(workResult);
block->setOffset(rect->getOffset());
block->setSize(rect->getSize());
/* Some constants */
const int sx = rect->getOffset().x, sy = block->getOffset().y;
const int ex = sx + rect->getSize().x, ey = sy + rect->getSize().y;
const int width = rect->getSize().x;
const SSEVector MM_ALIGN16 xOffset(0.0f, 1.0f, 0.0f, 1.0f);
const SSEVector MM_ALIGN16 yOffset(0.0f, 0.0f, 1.0f, 1.0f);
const int pixelOffset[] = {0, 1, width, width+1};
const __m128 clamping = _mm_set1_ps(1/(m_minDist*m_minDist));
uint8_t temp[MTS_KD_INTERSECTION_TEMP*4];
const __m128 camTL[3] = {
_mm_set1_ps(m_cameraTL.x),
_mm_set1_ps(m_cameraTL.y),
_mm_set1_ps(m_cameraTL.z)
};
const __m128 camDx[3] = {
_mm_set1_ps(m_cameraDx.x),
_mm_set1_ps(m_cameraDx.y),
_mm_set1_ps(m_cameraDx.z)
};
const __m128 camDy[3] = {
_mm_set1_ps(m_cameraDy.x),
_mm_set1_ps(m_cameraDy.y),
_mm_set1_ps(m_cameraDy.z)
};
const __m128 lumPos[3] = {
_mm_set1_ps(m_vpl.its.p.x),
_mm_set1_ps(m_vpl.its.p.y),
_mm_set1_ps(m_vpl.its.p.z)
};
const __m128 lumDir[3] = {
_mm_set1_ps(m_vpl.its.shFrame.n.x),
_mm_set1_ps(m_vpl.its.shFrame.n.y),
_mm_set1_ps(m_vpl.its.shFrame.n.z)
};
/* Some local variables */
int pos = 0;
int numRays = 0;
RayPacket4 MM_ALIGN16 primRay4, secRay4;
Intersection4 MM_ALIGN16 its4, secIts4;
RayInterval4 MM_ALIGN16 itv4, secItv4;
SSEVector MM_ALIGN16 nSecD[3], cosThetaLight, invLengthSquared;
Spectrum emitted[4], direct[4];
Intersection its;
Vector wo, wi;
its.hasUVPartials = false;
bool diffuseVPL = false, vplOnSurface = false;
Spectrum vplWeight;
if (m_vpl.type == ESurfaceVPL && (m_diffuseSources || m_vpl.its.shape->getBSDF()->getType() == BSDF::EDiffuseReflection)) {
diffuseVPL = true;
vplOnSurface = true;
vplWeight = m_vpl.its.shape->getBSDF()->getDiffuseReflectance(m_vpl.its) * m_vpl.P / M_PI;
} else if (m_vpl.type == ELuminaireVPL) {
vplOnSurface = m_vpl.luminaire->getType() & Luminaire::EOnSurface;
diffuseVPL = m_vpl.luminaire->getType() & Luminaire::EDiffuseDirection;
EmissionRecord eRec(m_vpl.luminaire,
ShapeSamplingRecord(m_vpl.its.p, m_vpl.its.shFrame.n), m_vpl.its.shFrame.n);
vplWeight = m_vpl.P * m_vpl.luminaire->evalDirection(eRec);
}
primRay4.o[0].ps = _mm_set1_ps(m_cameraO.x);
primRay4.o[1].ps = _mm_set1_ps(m_cameraO.y);
primRay4.o[2].ps = _mm_set1_ps(m_cameraO.z);
secItv4.mint.ps = _mm_set1_ps(ShadowEpsilon);
/* Work on 2x2 sub-blocks */
for (int y=sy; y<ey; y += 2, pos += width) {
for (int x=sx; x<ex; x += 2, pos += 2) {
/* Generate camera rays without normalization */
const __m128
xPixel = _mm_add_ps(xOffset.ps, _mm_set1_ps((float) x)),
yPixel = _mm_add_ps(yOffset.ps, _mm_set1_ps((float) y));
primRay4.d[0].ps = _mm_add_ps(camTL[0], _mm_add_ps(
_mm_mul_ps(xPixel, camDx[0]), _mm_mul_ps(yPixel, camDy[0])));
primRay4.d[1].ps = _mm_add_ps(camTL[1], _mm_add_ps(
_mm_mul_ps(xPixel, camDx[1]), _mm_mul_ps(yPixel, camDy[1])));
primRay4.d[2].ps = _mm_add_ps(camTL[2], _mm_add_ps(
_mm_mul_ps(xPixel, camDx[2]), _mm_mul_ps(yPixel, camDy[2])));
primRay4.dRcp[0].ps = _mm_div_ps(SSEConstants::one.ps, primRay4.d[0].ps);
primRay4.dRcp[1].ps = _mm_div_ps(SSEConstants::one.ps, primRay4.d[1].ps);
primRay4.dRcp[2].ps = _mm_div_ps(SSEConstants::one.ps, primRay4.d[2].ps);
/* Ray coherence test */
const int primSignsX = _mm_movemask_ps(primRay4.d[0].ps);
const int primSignsY = _mm_movemask_ps(primRay4.d[1].ps);
const int primSignsZ = _mm_movemask_ps(primRay4.d[2].ps);
const bool primCoherent =
(primSignsX == 0 || primSignsX == 0xF)
&& (primSignsY == 0 || primSignsY == 0xF)
&& (primSignsZ == 0 || primSignsZ == 0xF);
/* Trace the primary rays */
its4.t = SSEConstants::p_inf;
if (EXPECT_TAKEN(primCoherent)) {
primRay4.signs[0][0] = primSignsX ? 1 : 0;
primRay4.signs[1][0] = primSignsY ? 1 : 0;
primRay4.signs[2][0] = primSignsZ ? 1 : 0;
m_kdtree->rayIntersectPacket(primRay4, itv4, its4, temp);
} else {
m_kdtree->rayIntersectPacketIncoherent(primRay4, itv4, its4, temp);
}
numRays += 4;
/* Generate secondary rays */
secRay4.o[0].ps = _mm_add_ps(primRay4.o[0].ps, _mm_mul_ps(its4.t.ps, primRay4.d[0].ps));
secRay4.o[1].ps = _mm_add_ps(primRay4.o[1].ps, _mm_mul_ps(its4.t.ps, primRay4.d[1].ps));
secRay4.o[2].ps = _mm_add_ps(primRay4.o[2].ps, _mm_mul_ps(its4.t.ps, primRay4.d[2].ps));
secRay4.d[0].ps = _mm_sub_ps(lumPos[0], secRay4.o[0].ps);
secRay4.d[1].ps = _mm_sub_ps(lumPos[1], secRay4.o[1].ps);
secRay4.d[2].ps = _mm_sub_ps(lumPos[2], secRay4.o[2].ps);
/* Normalization */
const __m128
lengthSquared = _mm_add_ps(_mm_add_ps(
_mm_mul_ps(secRay4.d[0].ps, secRay4.d[0].ps),
_mm_mul_ps(secRay4.d[1].ps, secRay4.d[1].ps)),
_mm_mul_ps(secRay4.d[2].ps, secRay4.d[2].ps)),
invLength = _mm_rsqrt_ps(lengthSquared);
invLengthSquared.ps = _mm_min_ps(_mm_rcp_ps(lengthSquared), clamping);
nSecD[0].ps = _mm_mul_ps(secRay4.d[0].ps, invLength);
nSecD[1].ps = _mm_mul_ps(secRay4.d[1].ps, invLength);
nSecD[2].ps = _mm_mul_ps(secRay4.d[2].ps, invLength);
secRay4.dRcp[0].ps = _mm_div_ps(SSEConstants::one.ps, secRay4.d[0].ps);
secRay4.dRcp[1].ps = _mm_div_ps(SSEConstants::one.ps, secRay4.d[1].ps);
secRay4.dRcp[2].ps = _mm_div_ps(SSEConstants::one.ps, secRay4.d[2].ps);
cosThetaLight.ps = _mm_sub_ps(_mm_setzero_ps(),
_mm_add_ps(_mm_add_ps(
_mm_mul_ps(nSecD[0].ps, lumDir[0]),
_mm_mul_ps(nSecD[1].ps, lumDir[1])),
_mm_mul_ps(nSecD[2].ps, lumDir[2])));
secItv4.maxt.ps = _mm_set1_ps(1-ShadowEpsilon);
/* Shading (scalar) --- this is way too much work and should be
rewritten to be smarter in special cases */
for (int idx=0; idx<4; ++idx) {
if (EXPECT_NOT_TAKEN(its4.t.f[idx] == std::numeric_limits<float>::infinity())) {
/* Don't trace a secondary ray */
secItv4.maxt.f[idx] = 0;
emitted[idx] = m_scene->LeBackground(Ray(
Point(primRay4.o[0].f[idx], primRay4.o[1].f[idx], primRay4.o[2].f[idx]),
Vector(primRay4.d[0].f[idx], primRay4.d[1].f[idx], primRay4.d[2].f[idx]),
0.0f
)) * m_backgroundScale;
memset(&direct[idx], 0, sizeof(Spectrum));
continue;
}
const unsigned int primIndex = its4.primIndex.i[idx];
const Shape *shape = (*m_shapes)[its4.shapeIndex.i[idx]];
const BSDF *bsdf = shape->getBSDF();
if (EXPECT_NOT_TAKEN(!bsdf)) {
memset(&emitted[idx], 0, sizeof(Spectrum));
memset(&direct[idx], 0, sizeof(Spectrum));
continue;
}
if (EXPECT_TAKEN(primIndex != KNoTriangleFlag)) {
const TriMesh *mesh = static_cast<const TriMesh *>(shape);
const Triangle &t = mesh->getTriangles()[primIndex];
const Normal *normals = mesh->getVertexNormals();
const Point2 *texcoords = mesh->getVertexTexcoords();
const Spectrum *colors = mesh->getVertexColors();
const TangentSpace * tangents = mesh->getVertexTangents();
const Float beta = its4.u.f[idx],
gamma = its4.v.f[idx],
alpha = 1.0f - beta - gamma;
const uint32_t idx0 = t.idx[0], idx1 = t.idx[1], idx2 = t.idx[2];
if (EXPECT_TAKEN(normals)) {
const Normal &n0 = normals[idx0],
&n1 = normals[idx1],
&n2 = normals[idx2];
its.shFrame.n = normalize(n0 * alpha + n1 * beta + n2 * gamma);
} else {
const Point *positions = mesh->getVertexPositions();
const Point &p0 = positions[idx0],
&p1 = positions[idx1],
&p2 = positions[idx2];
Vector sideA = p1 - p0, sideB = p2 - p0;
Vector n = cross(sideA, sideB);
Float nLengthSqr = n.lengthSquared();
if (nLengthSqr != 0)
n /= std::sqrt(nLengthSqr);
its.shFrame.n = Normal(n);
}
if (EXPECT_TAKEN(texcoords)) {
const Point2 &t0 = texcoords[idx0],
&t1 = texcoords[idx1],
&t2 = texcoords[idx2];
its.uv = t0 * alpha + t1 * beta + t2 * gamma;
} else {
its.uv = Point2(0.0f);
}
if (EXPECT_NOT_TAKEN(colors)) {
const Spectrum &c0 = colors[idx0],
&c1 = colors[idx1],
&c2 = colors[idx2];
its.color = c0 * alpha + c1 * beta + c2 * gamma;
}
if (EXPECT_NOT_TAKEN(tangents)) {
const TangentSpace &t0 = tangents[idx0],
&t1 = tangents[idx1],
&t2 = tangents[idx2];
its.dpdu = t0.dpdu * alpha + t1.dpdu * beta + t2.dpdu * gamma;
its.dpdv = t0.dpdv * alpha + t1.dpdv * beta + t2.dpdv * gamma;
}
} else {
Ray ray(
Point(primRay4.o[0].f[idx], primRay4.o[1].f[idx], primRay4.o[2].f[idx]),
Vector(primRay4.d[0].f[idx], primRay4.d[1].f[idx], primRay4.d[2].f[idx]),
0.0f
);
its.t = its4.t.f[idx];
shape->fillIntersectionRecord(ray, temp + idx * MTS_KD_INTERSECTION_TEMP + 8, its);
bsdf = its.shape->getBSDF();
}
wo.x = nSecD[0].f[idx]; wo.y = nSecD[1].f[idx]; wo.z = nSecD[2].f[idx];
if (EXPECT_TAKEN(!shape->isLuminaire())) {
memset(&emitted[idx], 0, sizeof(Spectrum));
} else {
Vector d(-primRay4.d[0].f[idx], -primRay4.d[1].f[idx], -primRay4.d[2].f[idx]);
emitted[idx] = shape->getLuminaire()->Le(ShapeSamplingRecord(its.p, its.shFrame.n), d);
}
if (EXPECT_TAKEN(bsdf->getType() == BSDF::EDiffuseReflection && diffuseVPL)) {
/* Fast path */
direct[idx] = (bsdf->getDiffuseReflectance(its) * vplWeight)
* (std::max((Float) 0.0f, dot(wo, its.shFrame.n))
* (vplOnSurface ? (std::max(cosThetaLight.f[idx], (Float) 0.0f) * INV_PI) : INV_PI)
* invLengthSquared.f[idx]);
} else {
wi.x = -primRay4.d[0].f[idx];
wi.y = -primRay4.d[1].f[idx];
wi.z = -primRay4.d[2].f[idx];
its.p.x = secRay4.o[0].f[idx];
its.p.y = secRay4.o[1].f[idx];
its.p.z = secRay4.o[2].f[idx];
if (EXPECT_NOT_TAKEN(bsdf->getType() & BSDF::EAnisotropic)) {
its.shFrame.s = normalize(its.dpdu - its.shFrame.n
* dot(its.shFrame.n, its.dpdu));
its.shFrame.t = cross(its.shFrame.n, its.shFrame.s);
} else {
coordinateSystem(its.shFrame.n, its.shFrame.s, its.shFrame.t);
}
const Float ctLight = cosThetaLight.f[idx];
wi = normalize(wi);
its.wi = its.toLocal(wi);
wo = its.toLocal(wo);
if (!diffuseVPL) {
if (m_vpl.type == ESurfaceVPL) {
BSDFQueryRecord bRec(m_vpl.its, m_vpl.its.toLocal(wi));
bRec.quantity = EImportance;
vplWeight = m_vpl.its.shape->getBSDF()->eval(bRec) * m_vpl.P;
} else {
EmissionRecord eRec(m_vpl.luminaire,
ShapeSamplingRecord(m_vpl.its.p, m_vpl.its.shFrame.n), wi);
eRec.type = EmissionRecord::EPreview;
vplWeight = m_vpl.luminaire->evalDirection(eRec) * m_vpl.P;
}
}
if (EXPECT_TAKEN(ctLight >= 0)) {
direct[idx] = (bsdf->eval(BSDFQueryRecord(its, wo)) * vplWeight
* ((vplOnSurface ? std::max(ctLight, (Float) 0.0f) : 1.0f) * invLengthSquared.f[idx]));
} else {
memset(&direct[idx], 0, sizeof(Spectrum));
}
}
++numRays;
}
/* Shoot the secondary rays */
const int secSignsX = _mm_movemask_ps(secRay4.d[0].ps);
const int secSignsY = _mm_movemask_ps(secRay4.d[1].ps);
const int secSignsZ = _mm_movemask_ps(secRay4.d[2].ps);
const bool secCoherent =
(secSignsX == 0 || secSignsX == 0xF)
&& (secSignsY == 0 || secSignsY == 0xF)
&& (secSignsZ == 0 || secSignsZ == 0xF);
/* Shoot the secondary rays */
secIts4.t = SSEConstants::p_inf;
if (EXPECT_TAKEN(secCoherent)) {
secRay4.signs[0][0] = secSignsX ? 1 : 0;
secRay4.signs[1][0] = secSignsY ? 1 : 0;
secRay4.signs[2][0] = secSignsZ ? 1 : 0;
m_kdtree->rayIntersectPacket(secRay4, secItv4, secIts4, temp);
} else {
m_kdtree->rayIntersectPacketIncoherent(secRay4, secItv4, secIts4, temp);
}
for (int idx=0; idx<4; ++idx) {
if (EXPECT_TAKEN(secIts4.t.f[idx] == std::numeric_limits<float>::infinity()))
block->setPixel(pos+pixelOffset[idx], direct[idx]+emitted[idx]);
else
block->setPixel(pos+pixelOffset[idx], emitted[idx]);
}
}
}
block->setExtra(numRays);
#else
Log(EError, "Coherent raytracing support was not compiled into this binary!");
#endif
}
Spectrum Li(const RayDifferential &r, RadianceQueryRecord &rRec) const {
/* Some aliases and local variables */
const Scene *scene = rRec.scene;
Intersection &its = rRec.its, bsdfIts;
RayDifferential ray(r);
LuminaireSamplingRecord lRec;
Spectrum Li(0.0f);
Point2 sample;
/* Perform the first ray intersection (or ignore if the
intersection has already been provided). */
if (!rRec.rayIntersect(ray)) {
/* If no intersection could be found, possibly return
radiance from a background luminaire */
if (rRec.type & RadianceQueryRecord::EEmittedRadiance)
return scene->LeBackground(ray);
else
return Spectrum(0.0f);
}
/* Possibly include emitted radiance if requested */
if (its.isLuminaire() && (rRec.type & RadianceQueryRecord::EEmittedRadiance))
Li += its.Le(-ray.d);
/* Include radiance from a subsurface integrator if requested */
if (its.hasSubsurface() && (rRec.type & RadianceQueryRecord::ESubsurfaceRadiance))
Li += its.LoSub(scene, rRec.sampler, -ray.d, rRec.depth);
const BSDF *bsdf = its.getBSDF(ray);
if (EXPECT_NOT_TAKEN(!bsdf)) {
/* The direct illumination integrator doesn't support
surfaces without a BSDF (e.g. medium transitions)
-- give up. */
return Li;
}
/* Leave here if direct illumination was not requested */
if (!(rRec.type & RadianceQueryRecord::EDirectSurfaceRadiance))
return Li;
/* ==================================================================== */
/* Luminaire sampling */
/* ==================================================================== */
bool adaptiveQuery = (rRec.extra & RadianceQueryRecord::EAdaptiveQuery);
Point2 *sampleArray;
size_t numLuminaireSamples = m_luminaireSamples,
numBSDFSamples = m_bsdfSamples;
Float fracLum = m_fracLum, fracBSDF = m_fracBSDF,
weightLum = m_weightLum, weightBSDF = m_weightBSDF;
if (rRec.depth > 1 || adaptiveQuery) {
/* This integrator is used recursively by another integrator.
Be less accurate as this sample will not directly be observed. */
numBSDFSamples = numLuminaireSamples = 1;
fracLum = fracBSDF = .5f;
weightLum = weightBSDF = 1.0f;
}
if (numLuminaireSamples > 1) {
sampleArray = rRec.sampler->next2DArray(numLuminaireSamples);
} else {
sample = rRec.nextSample2D(); sampleArray = &sample;
}
for (size_t i=0; i<numLuminaireSamples; ++i) {
/* Estimate the direct illumination if this is requested */
if (scene->sampleLuminaire(its.p, ray.time, lRec, sampleArray[i])) {
/* Allocate a record for querying the BSDF */
BSDFQueryRecord bRec(its, its.toLocal(-lRec.d));
/* Evaluate BSDF * cos(theta) */
const Spectrum bsdfVal = bsdf->eval(bRec);
if (!bsdfVal.isZero()) {
/* Calculate prob. of having sampled that direction
using BSDF sampling */
Float bsdfPdf = (lRec.luminaire->isIntersectable()
|| lRec.luminaire->isBackgroundLuminaire()) ?
bsdf->pdf(bRec) : 0;
/* Weight using the power heuristic */
const Float weight = miWeight(lRec.pdf * fracLum,
bsdfPdf * fracBSDF) * weightLum;
Li += lRec.value * bsdfVal * weight;
}
}
}
/* ==================================================================== */
/* BSDF sampling */
/* ==================================================================== */
if (numBSDFSamples > 1) {
sampleArray = rRec.sampler->next2DArray(numBSDFSamples);
} else {
sample = rRec.nextSample2D(); sampleArray = &sample;
}
for (size_t i=0; i<numBSDFSamples; ++i) {
/* Sample BSDF * cos(theta) */
BSDFQueryRecord bRec(its, rRec.sampler, ERadiance);
Float bsdfPdf;
Spectrum bsdfVal = bsdf->sample(bRec, bsdfPdf, sampleArray[i]);
if (bsdfVal.isZero())
continue;
/* Trace a ray in this direction */
Ray bsdfRay(its.p, its.toWorld(bRec.wo), ray.time);
if (scene->rayIntersect(bsdfRay, bsdfIts)) {
/* Intersected something - check if it was a luminaire */
if (bsdfIts.isLuminaire()) {
lRec = LuminaireSamplingRecord(bsdfIts, -bsdfRay.d);
lRec.value = bsdfIts.Le(-bsdfRay.d);
} else {
continue;
}
} else {
/* No intersection found. Possibly, there is a background
luminaire such as an environment map? */
if (scene->hasBackgroundLuminaire()) {
lRec.luminaire = scene->getBackgroundLuminaire();
lRec.d = -bsdfRay.d;
lRec.value = lRec.luminaire->Le(bsdfRay);
} else {
continue;
}
}
const Float lumPdf = (!(bRec.sampledType & BSDF::EDelta)) ?
scene->pdfLuminaire(its.p, lRec) : 0;
const Float weight = miWeight(bsdfPdf * fracBSDF,
lumPdf * fracLum) * weightBSDF;
Li += lRec.value * bsdfVal * weight;
}
return Li;
}
Spectrum Li(const RayDifferential &r, RadianceQueryRecord &rRec) const {
/* Some aliases and local variables */
const Scene *scene = rRec.scene;
Intersection &its = rRec.its;
RayDifferential ray(r);
Spectrum Li(0.0f);
/* Perform the first ray intersection (or ignore if the
intersection has already been provided). */
rRec.rayIntersect(ray);
ray.mint = Epsilon;
Spectrum pathThroughput(1.0f);
while (rRec.depth <= m_maxDepth || m_maxDepth < 0) {
if (!its.isValid()) {
/* If no intersection could be found, potentially return
radiance from a background luminaire if it exists */
if (rRec.type & RadianceQueryRecord::EEmittedRadiance)
Li += pathThroughput * scene->LeBackground(ray);
break;
}
const BSDF *bsdf = its.getBSDF(ray);
if (EXPECT_NOT_TAKEN(bsdf == NULL)) {
/* The MI path tracer doesn't support
surfaces without a BSDF (e.g. medium transitions)
-- give up. */
break;
}
/* Possibly include emitted radiance if requested */
if (its.isLuminaire() && (rRec.type & RadianceQueryRecord::EEmittedRadiance))
Li += pathThroughput * its.Le(-ray.d);
/* Include radiance from a subsurface integrator if requested */
if (its.hasSubsurface() && (rRec.type & RadianceQueryRecord::ESubsurfaceRadiance))
Li += pathThroughput * its.LoSub(scene, rRec.sampler, -ray.d, rRec.depth);
if (m_maxDepth > 0 && rRec.depth >= m_maxDepth)
break;
/* ==================================================================== */
/* Luminaire sampling */
/* ==================================================================== */
/* Prevent light leaks due to the use of shading normals */
Float wiDotGeoN = -dot(its.geoFrame.n, ray.d),
wiDotShN = Frame::cosTheta(its.wi);
if (wiDotGeoN * wiDotShN < 0 && m_strictNormals)
break;
/* Estimate the direct illumination if this is requested */
LuminaireSamplingRecord lRec;
if (rRec.type & RadianceQueryRecord::EDirectSurfaceRadiance &&
scene->sampleLuminaire(its.p, ray.time, lRec, rRec.nextSample2D())) {
/* Allocate a record for querying the BSDF */
const Vector wo = -lRec.d;
const BSDFQueryRecord bRec(its, its.toLocal(wo));
/* Evaluate BSDF * cos(theta) */
const Spectrum bsdfVal = bsdf->fCos(bRec);
Float woDotGeoN = dot(its.geoFrame.n, wo);
/* Prevent light leaks due to the use of shading normals */
if (!bsdfVal.isZero() && (!m_strictNormals
|| woDotGeoN * Frame::cosTheta(bRec.wo) > 0)) {
/* Calculate prob. of having sampled that direction
using BSDF sampling */
Float bsdfPdf = (lRec.luminaire->isIntersectable()
|| lRec.luminaire->isBackgroundLuminaire()) ?
bsdf->pdf(bRec) : 0;
/* Weight using the power heuristic */
const Float weight = miWeight(lRec.pdf, bsdfPdf);
Li += pathThroughput * lRec.value * bsdfVal * weight;
}
}
/* ==================================================================== */
/* BSDF sampling */
/* ==================================================================== */
/* Sample BSDF * cos(theta) */
BSDFQueryRecord bRec(its);
Float bsdfPdf;
Spectrum bsdfVal = bsdf->sampleCos(bRec, bsdfPdf, rRec.nextSample2D());
if (bsdfVal.isZero())
break;
bsdfVal /= bsdfPdf;
/* Prevent light leaks due to the use of shading normals */
const Vector wo = its.toWorld(bRec.wo);
Float woDotGeoN = dot(its.geoFrame.n, wo);
if (woDotGeoN * Frame::cosTheta(bRec.wo) <= 0 && m_strictNormals)
break;
/* Trace a ray in this direction */
ray = Ray(its.p, wo, ray.time);
bool hitLuminaire = false;
if (scene->rayIntersect(ray, its)) {
/* Intersected something - check if it was a luminaire */
if (its.isLuminaire()) {
lRec = LuminaireSamplingRecord(its, -ray.d);
lRec.value = its.Le(-ray.d);
hitLuminaire = true;
}
} else {
/* No intersection found. Possibly, there is a background
luminaire such as an environment map? */
if (scene->hasBackgroundLuminaire()) {
lRec.luminaire = scene->getBackgroundLuminaire();
lRec.value = lRec.luminaire->Le(ray);
lRec.d = -ray.d;
hitLuminaire = true;
} else {
rRec.depth++;
break;
}
}
/* If a luminaire was hit, estimate the local illumination and
weight using the power heuristic */
if (hitLuminaire &&
(rRec.type & RadianceQueryRecord::EDirectSurfaceRadiance)) {
/* Prob. of having generated this sample using luminaire sampling */
const Float lumPdf = (!(bRec.sampledType & BSDF::EDelta)) ?
scene->pdfLuminaire(ray.o, lRec) : 0;
const Float weight = miWeight(bsdfPdf, lumPdf);
Li += pathThroughput * lRec.value * bsdfVal * weight;
}
/* ==================================================================== */
/* Indirect illumination */
/* ==================================================================== */
/* Set the recursive query type */
/* Stop if no surface was hit by the BSDF sample or if indirect illumination
was not requested */
if (!its.isValid() || !(rRec.type & RadianceQueryRecord::EIndirectSurfaceRadiance))
break;
rRec.type = RadianceQueryRecord::ERadianceNoEmission;
/* Russian roulette - Possibly stop the recursion. Don't do this when
dealing with a transmission component, since solid angle compression
factors cause problems with the heuristic below */
if (rRec.depth >= m_rrDepth && !(bRec.sampledType & BSDF::ETransmission)) {
/* Assuming that BSDF importance sampling is perfect,
'bsdfVal.max()' should equal the maximum albedo
over all spectral samples */
Float approxAlbedo = std::min((Float) 0.9f, bsdfVal.max());
if (rRec.nextSample1D() > approxAlbedo)
break;
else
pathThroughput /= approxAlbedo;
}
pathThroughput *= bsdfVal;
rRec.depth++;
}
/* Store statistics */
avgPathLength.incrementBase();
avgPathLength += rRec.depth;
return Li;
}
