/* Calculates the fade time from the changes in gain and listener to source
* angle between updates. The result is a the time, in seconds, for the
* transition to complete.
*/
static ALfloat CalcFadeTime(ALfloat oldGain, ALfloat newGain, const aluVector *olddir, const aluVector *newdir)
{
ALfloat gainChange, angleChange, change;
/* Calculate the normalized dB gain change. */
newGain = maxf(newGain, 0.0001f);
oldGain = maxf(oldGain, 0.0001f);
gainChange = fabsf(log10f(newGain / oldGain) / log10f(0.0001f));
/* Calculate the angle change only when there is enough gain to notice it. */
angleChange = 0.0f;
if(gainChange > 0.0001f || newGain > 0.0001f)
{
/* No angle change when the directions are equal or degenerate (when
* both have zero length).
*/
if(newdir->v[0] != olddir->v[0] || newdir->v[1] != olddir->v[1] || newdir->v[2] != olddir->v[2])
{
ALfloat dotp = aluDotproduct(olddir, newdir);
angleChange = acosf(clampf(dotp, -1.0f, 1.0f)) / F_PI;
}
}
/* Use the largest of the two changes, and apply a significance shaping
* function to it. The result is then scaled to cover a 15ms transition
* range.
*/
change = maxf(angleChange * 25.0f, gainChange) * 2.0f;
return minf(change, 1.0f) * 0.015f;
}
std::pair<int, int> find_min_max(int a[], int length)
{
int min(0), max(0), i(0);
if (length % 2)
{
min = a[0];
max = a[0];
i = 1;
}
else
{
min = minf(a[0], a[1]);
max = maxf(a[0], a[1]);
i = 2;
}
for (; i < length; i += 2)
{
int min2(0), max2(0);
if (a[i] < a[i+1])
{
min2 = a[i];
max2 = a[i+1];
}
else
{
max2 = a[i];
min2 = a[i+1];
}
min = minf(min, min2);
max = maxf(max, max2);
}
return std::make_pair(min, max);
}
static void morpho_dilate(CompBuf *cbuf)
{
int x, y;
float *p, *rectf = cbuf->rect;
for (y = 0; y < cbuf->y; y++) {
for (x = 0; x < cbuf->x - 1; x++) {
p = rectf + cbuf->x * y + x;
*p = maxf(*p, *(p + 1));
}
}
for (y = 0; y < cbuf->y; y++) {
for (x = cbuf->x - 1; x >= 1; x--) {
p = rectf + cbuf->x * y + x;
*p = maxf(*p, *(p - 1));
}
}
for (x = 0; x < cbuf->x; x++) {
for (y = 0; y < cbuf->y - 1; y++) {
p = rectf + cbuf->x * y + x;
*p = maxf(*p, *(p + cbuf->x));
}
}
for (x = 0; x < cbuf->x; x++) {
for (y = cbuf->y - 1; y >= 1; y--) {
p = rectf + cbuf->x * y + x;
*p = maxf(*p, *(p - cbuf->x));
}
}
}
Vec3<float> Scene::calculate_shadowing_optimal_point(Vec3<float> near_values[4],Vec3<float> far_values[4], float &radius) {
float max_top=near_values[0].y,max_bot=near_values[0].y,max_right=near_values[0].x,max_left=near_values[0].x;
float depth=0;
for(int i=0;i<4;i++) {
float cur_max_right = maxf(near_values[i].x,far_values[i].x);
float cur_max_left = minf(near_values[i].x,far_values[i].x);
float cur_max_top = maxf(near_values[i].y,far_values[i].y);
float cur_max_bot = minf(near_values[i].y,far_values[i].y);
max_top=maxf(max_top,cur_max_top);
max_bot=minf(max_bot,cur_max_bot);
max_right=maxf(max_right,cur_max_right);
max_left=minf(max_left,cur_max_left);
depth+=(near_values[0].z+far_values[0].z);
}
float max_height=max_top-max_bot;
float max_width=max_right-max_left;
radius=maxf(max_height,max_width);
return Vec3<float>((max_right+max_left)/2,(max_top+max_bot)/2,depth/8);
}
void sreScissors::UpdateWithProjectedPoint(float x, float y, double z) {
if (z >= - 1.001d) {
// Beyond the near plane.
z = maxd(- 1.0d, z);
double depth = 0.5d * z + 0.5d;
near = mind(depth, near);
far = maxd(depth, far);
left = minf(x, left);
right = maxf(x, right);
bottom = minf(y, bottom);
top = maxf(y, top);
return;
}
sreMessage(SRE_MESSAGE_WARNING,
"Unexpected vertex in front of the near plane in UpdateWorldSpaceBoundingHull "
"z = %lf", z);
// In front of the near plane.
near = 0;
far = 1.0;
// We know that the light volume intersects the frustum, so it must extend to
// both sides of the near plane.
// Assume it fills the whole viewport (not optimal).
left = - 1.0f;
right = 1.0f;
bottom = - 1.0f;
top = 1.0f;
}
/* Calculates the normalized HRTF transition factor (delta) from the changes
* in gain and listener to source angle between updates. The result is a
* normalized delta factor that can be used to calculate moving HRIR stepping
* values.
*/
ALfloat CalcHrtfDelta(ALfloat oldGain, ALfloat newGain, const ALfloat olddir[3], const ALfloat newdir[3])
{
ALfloat gainChange, angleChange, change;
// Calculate the normalized dB gain change.
newGain = maxf(newGain, 0.0001f);
oldGain = maxf(oldGain, 0.0001f);
gainChange = fabsf(log10f(newGain / oldGain) / log10f(0.0001f));
// Calculate the normalized listener to source angle change when there is
// enough gain to notice it.
angleChange = 0.0f;
if(gainChange > 0.0001f || newGain > 0.0001f)
{
// No angle change when the directions are equal or degenerate (when
// both have zero length).
if(newdir[0]-olddir[0] || newdir[1]-olddir[1] || newdir[2]-olddir[2])
angleChange = acosf(olddir[0]*newdir[0] +
olddir[1]*newdir[1] +
olddir[2]*newdir[2]) / F_PI;
}
// Use the largest of the two changes for the delta factor, and apply a
// significance shaping function to it.
change = maxf(angleChange * 25.0f, gainChange) * 2.0f;
return minf(change, 1.0f);
}
void cmp_model_debug(struct cmp_obj* obj, void* data, cmphandle_t cur_hdl, float dt,
const struct gfx_view_params* params)
{
struct cmp_model* m = (struct cmp_model*)data;
if (m->model_hdl == INVALID_HANDLE)
return;
struct gfx_model* gmodel = rs_get_model(m->model_hdl);
if (gmodel == NULL)
return;
gfx_canvas_setztest(TRUE);
gfx_canvas_setwireframe(TRUE, TRUE);
uint tri_cnt = 0;
uint vert_cnt = 0;
for (uint i = 0; i < gmodel->node_cnt; i++) {
struct cmp_xform* xf = (struct cmp_xform*)cmp_getinstancedata(m->xforms[i]);
struct vec4f node_pos;
mat3_get_trans(&node_pos, &xf->ws_mat);
gfx_canvas_text3d(gmodel->nodes[i].name, &node_pos, ¶ms->viewproj);
gfx_canvas_setztest(FALSE);
gfx_canvas_coords(&xf->ws_mat, ¶ms->cam_pos, 0.2f);
if (gmodel->nodes[i].mesh_id != INVALID_INDEX) {
struct gfx_model_mesh* gmesh = &gmodel->meshes[gmodel->nodes[i].mesh_id];
struct gfx_model_geo* geo = &gmodel->geos[gmesh->geo_id];
/* either draw skeleton or wireframe mesh */
if (geo->skeleton != NULL) {
struct gfx_model_instance* inst = m->model_inst;
gfx_canvas_setztest(FALSE);
float scale = maxf(aabb_getwidth(&gmodel->bb),
maxf(aabb_getheight(&gmodel->bb), aabb_getdepth(&gmodel->bb)));
scale *= 0.05f;
cmp_model_drawpose(geo, inst->poses[gmesh->geo_id], params, &xf->ws_mat, scale);
gfx_canvas_setztest(TRUE);
} else {
gfx_canvas_setfillcolor_solid(&g_color_yellow);
gfx_canvas_geo(geo, &xf->ws_mat);
}
tri_cnt += geo->tri_cnt;
vert_cnt += geo->vert_cnt;
}
}
/* model info */
char info[64];
struct vec4f farpt;
gfx_canvas_settextcolor(&g_color_grey);
cmp_bounds_getfarcorner(&farpt, obj->bounds_cmp, ¶ms->cam_pos);
sprintf(info, "node_cnt: %d\ntri_cnt: %d\nvertex_cnt: %d", gmodel->node_cnt,
tri_cnt, vert_cnt);
gfx_canvas_text3dmultiline(info, &farpt, ¶ms->viewproj);
gfx_canvas_setfillcolor_solid(&g_color_white);
gfx_canvas_settextcolor(&g_color_white);
gfx_canvas_setwireframe(FALSE, TRUE);
}
static float
cube(float *pos, float *size)
{
float d[3];
float dst;
abs3f(d, pos);
sub3f(d, d, size);
dst = minf(maxf(d[0], maxf(d[1],d[2])),0.0);
max3f(d, d,(float[]){0,0,0});
struct aabb* aabb_merge(struct aabb* rb, const struct aabb* b1, const struct aabb* b2)
{
struct vec4f minpt;
struct vec4f maxpt;
vec3_setf(&minpt, minf(b1->minpt.x, b2->minpt.x), minf(b1->minpt.y, b2->minpt.y),
minf(b1->minpt.z, b2->minpt.z));
vec3_setf(&maxpt, maxf(b1->maxpt.x, b2->maxpt.x), maxf(b1->maxpt.y, b2->maxpt.y),
maxf(b1->maxpt.z, b2->maxpt.z));
return aabb_setv(rb, &minpt, &maxpt);
}
int main(void) {
int a = 10, b = 20;
std::cout << "MAXM1 = " << MAXM(a, b) << std::endl;
std::cout << "MAXM2 = " << MAXM(a, b + 0.2) << std::endl;
std::cout << "MAXM3 = " << MAXM(a, b++) << std::endl;
std::cout << "maxf1 = " << maxf(a, b) << std::endl;
// what's the problem with the following line?
//std::cout << "maxf = " << maxf((double)a,b+0.2) << std::endl;
std::cout << "maxf2 = " << maxf(a + 0.1, b + 0.2) << std::endl;
std::cout << "maxf3 = " << maxf(a, b++) << std::endl;
std::cout << "a = " << a << ", b = " << b << std::endl;
}
struct sphere* sphere_xform(struct sphere* rs, const struct sphere* s, const struct mat3f* m)
{
/* we have to transform radius by scale value of transform matrix
* to determine scale value, we find the highest scale component of the matrix */
float xlen = m->m11*m->m11 + m->m12*m->m12 + m->m13*m->m13;
float ylen = m->m21*m->m21 + m->m22*m->m22 + m->m23*m->m23;
float zlen = m->m31*m->m31 + m->m32*m->m32 + m->m33*m->m33;
float isqr_scale = maxf(xlen, ylen);
isqr_scale = maxf(isqr_scale, zlen);
/* transform center */
struct vec4f c;
vec3_setf(&c, s->x, s->y, s->z);
vec3_transformsrt(&c, &c, m);
return sphere_setf(rs, c.x, c.y, c.z, s->r*sqrtf(isqr_scale));
}
double diff_seq(cpx **seq, cpx **ref, const int n)
{
double diff = 0.0;
for (int i = 0; i < n; ++i)
diff = maxf(diff, diff_seq(seq[i], ref[i], 1.f, n));
return diff;
}
// input : simultaneous masking threshold *frqthr,
// previous masking threshold *tmpthr,
// Integrations *a (short-time) and *b (long-time)
// output: tracked Integrations *a and *b, time constant *tau
static void
CalcTemporalThreshold ( float* a, float* b, float* tau, float* frqthr, float* tmpthr )
{
int n;
float tmp;
for ( n = 0; n < PART_LONG; n++ ) {
// following calculations relative to threshold in quiet
frqthr[n] *= invLtq[n];
tmpthr[n] *= invLtq[n];
// new post-masking 'tmp' via time constant tau, if old post-masking > Ltq (=1)
tmp = tmpthr[n] > 1.f ? POW ( tmpthr[n], tau[n] ) : 1.f;
// calculate time constant for post-masking in next frame,
// if new time constant has to be calculated (new tmpMask < frqMask)
a[n] += 0.5f * (frqthr[n] - a[n]); // short time integrator
b[n] += 0.15f * (frqthr[n] - b[n]); // long time integrator
if (tmp < frqthr[n])
tau[n] = a[n] <= b[n] ? 0.8f : 0.2f + b[n] / a[n] * 0.6f;
// use post-masking of (Re-Normalization)
tmpthr[n] = maxf (frqthr[n], tmp) * partLtq[n];
}
return;
}
// NoiseShaper for a subband
void
QuantizeSubbandWithNoiseShaping ( unsigned int* qu_output, const float* input, const int res, float* errors, const float* FIR )
{
float signal;
float mult = A [res];
float invmult = C [res];
int offset = D [res];
int n, quant;
memset(errors, 0, 6 * sizeof *errors); // arghh, it produces pops on each frame boundary!
for ( n = 0; n < 36; n++) {
signal = input[n] * NoiseInjectionCompensation1D [res] - (FIR[5]*errors[n+0] + FIR[4]*errors[n+1] + FIR[3]*errors[n+2] + FIR[2]*errors[n+3] + FIR[1]*errors[n+4] + FIR[0]*errors[n+5]);
quant = mpc_lrintf(signal * mult);
// calculate the current error and save it for error refeeding
errors [n + 6] = invmult * quant - signal * NoiseInjectionCompensation1D [res];
// limitation to +/-D
quant = minf ( quant, +offset );
quant = maxf ( quant, -offset );
qu_output[n] = (unsigned int)(quant + offset);
}
}
static void heat_set_H(LaplacianSystem *sys, int vertex)
{
float dist, mindist, h;
int j, numclosest = 0;
mindist= 1e10;
/* compute minimum distance */
for(j=0; j<sys->heat.numsource; j++) {
dist= heat_source_distance(sys, vertex, j);
if(dist < mindist)
mindist= dist;
}
sys->heat.mindist[vertex]= mindist;
/* count number of sources with approximately this minimum distance */
for(j=0; j<sys->heat.numsource; j++)
if(heat_source_closest(sys, vertex, j))
numclosest++;
sys->heat.p[vertex]= (numclosest > 0)? 1.0f/numclosest: 0.0f;
/* compute H entry */
if(numclosest > 0) {
mindist= maxf(mindist, 1e-4f);
h= numclosest*C_WEIGHT/(mindist*mindist);
}
else
h= 0.0f;
sys->heat.H[vertex]= h;
}
float
ISNR_Schaetzer_Trans ( const float* input, const float SNRcomp, const int res )
{
int k;
float fac = A [res];
float invfac = C [res];
float signal, error, ret, err, sig;
// Summation of the absolute power and the quadratic error
k = 0;
signal = error = 1.e-30f;
for ( ; k < 12; k++ ) {
sig = input[k] * NoiseInjectionCompensation1D [res];
err = mpc_nearbyintf(sig * fac) * invfac - sig;
error += err * err;
signal += sig * sig;
}
err = signal > error ? error / (SNRcomp * signal) : error / signal;
ret = err;
signal = error = 1.e-30f;
for ( ; k < 24; k++ ) {
sig = input[k] * NoiseInjectionCompensation1D [res];
err = mpc_nearbyintf(sig * fac) * invfac - sig;
error += err * err;
signal += sig * sig;
}
err = signal > error ? error / (SNRcomp * signal) : error / signal;
ret = maxf(ret, err);
signal = error = 1.e-30f;
for ( ; k < 36; k++ ) {
sig = input[k] * NoiseInjectionCompensation1D [res];
err = mpc_nearbyintf(sig * fac) * invfac - sig;
error += err * err;
signal += sig * sig;
}
err = signal > error ? error / (SNRcomp * signal) : error / signal;
ret = maxf(ret, err);
return ret;
}
double diff_forward_sinus(cpx *seq, const int n)
{
if (seq == NULL)
return 1;
if (n < 4) {
return 1.0;
}
const int n2 = (n >> 1);
cpx *neg = seq + 1;
cpx *pos = seq + n - 1;
double diff = cpx_abs(seq[0]);
diff = maxf(diff, neg->x);
diff = maxf(diff, pos->x);
for (int i = 2; i < n - 3; ++i) {
diff = maxf(diff, cpx_abs(seq[i]));
}
diff = maxf(diff, abs(abs(neg->y) - n2));
diff = maxf(diff, abs(abs(pos->y) - n2));
return diff / n2;
}
inline
FLOAT F_Get_Translate_Handle_Radius( const Vec3D& eyePos, const Vec3D& objPos )
{
FLOAT eyeDist = (eyePos - objPos).LengthSqr();
eyeDist = maxf(eyeDist,0.01f);
eyeDist = mxSqrt(eyeDist);
static FLOAT PICK_DIST = 0.01f;
HOT_FLOAT(PICK_DIST);
return PICK_DIST * eyeDist;
}
void collect_statistics(CallStatP call_stats)
{
int i=0;
CallStat min=call_stats[0], max=call_stats[0], mean=call_stats[0];
CallStat stddev;
memset(&stddev, 0, sizeof(CallStat));
for (i=1; i<g_num_repeats; ++i) {
min.real_time=minf(min.real_time, call_stats[i].real_time);
min.user_time=minf(min.user_time, call_stats[i].user_time);
min.sys_time=minf(min.sys_time, call_stats[i].sys_time);
max.real_time=maxf(max.real_time, call_stats[i].real_time);
max.user_time=maxf(max.user_time, call_stats[i].user_time);
max.sys_time=maxf(max.sys_time, call_stats[i].sys_time);
mean.real_time+=call_stats[i].real_time;
mean.user_time+=call_stats[i].user_time;
mean.sys_time+=call_stats[i].sys_time;
}
mean.real_time/=g_num_repeats;
mean.user_time/=g_num_repeats;
mean.sys_time/=g_num_repeats;
for (i=0; i<g_num_repeats; ++i) {
float d1=call_stats[i].real_time-mean.real_time;
stddev.real_time+=d1*d1;
float d2=call_stats[i].user_time-mean.user_time;
stddev.real_time+=d2*d2;
float d3=call_stats[i].sys_time-mean.sys_time;
stddev.sys_time+=d3*d3;
}
stddev.real_time=sqrt(stddev.real_time/g_num_repeats);
stddev.user_time=sqrt(stddev.user_time/g_num_repeats);
stddev.sys_time=sqrt(stddev.sys_time/g_num_repeats);
callstat_print(g_output_stream, &mean, "mean", g_output_format);
callstat_print(g_output_stream, &min, "min", g_output_format);
callstat_print(g_output_stream, &max, "max", g_output_format);
callstat_print(g_output_stream, &stddev, "stddev", g_output_format);
callstat_print_sep(g_output_stream, g_output_format);
}
void drawEmancipator(emancipator_s* e)
{
if(!e || !e->used)return;
e->modelInstance.brightness=maxf(1.0f-(e->counter*1.0f)/BLACKENINGTIME,0.0f);
e->modelInstance.alpha=(e->counter<BLACKENINGTIME)?(1.0f):(maxf(1.0f-((e->counter-BLACKENINGTIME)*1.0f)/FADINGTIME,0.0f));
gsPushMatrix();
gsSwitchRenderMode(md2GsMode);
gsTranslate(e->position.x, e->position.y, e->position.z);
// TODO : transpose ?
gsMultMatrix3(e->transformationMatrix);
gsRotateX(e->axis.x*e->angle);
gsRotateY(e->axis.y*e->angle);
gsRotateZ(e->axis.z*e->angle);
md2InstanceDraw(&e->modelInstance);
gsPopMatrix();
}
double diff_seq(cpx *seq, cpx *ref, float scalar, const int n)
{
if (seq == NULL)
return 1;
double mDiff = 0.0;
double mVal = -1;
cpx rScale{scalar, 0};
for (int i = 0; i < n; ++i) {
cpx norm = cpx_mul(seq + i, &rScale);
if (invalidCpx(norm))
return 1.0;
#if defined(_NVIDIA)
mVal = maxf(mVal, maxf(cuCabsf(norm), cuCabsf(ref[i])));
double tmp = cuCabsf(cuCsubf(norm, ref[i]));
#else
mVal = maxf(mVal, maxf(cpx_abs(norm), cpx_abs(ref[i])));
double tmp = cpx_abs(cpx_sub(&norm, ref + i));
#endif
mDiff = tmp > mDiff ? tmp : mDiff;
}
return (mDiff / mVal);
}
ALfloat lpCoeffCalc(ALfloat g, ALfloat cw)
{
ALfloat a = 0.0f;
if(g < 0.9999f) /* 1-epsilon */
{
/* Be careful with gains < 0.001, as that causes the coefficient head
* towards 1, which will flatten the signal */
g = maxf(g, 0.001f);
a = (1 - g*cw - sqrtf(2*g*(1-cw) - g*g*(1 - cw*cw))) /
(1 - g);
}
return a;
}
/**
* @brief
* Calculate the Norm Error of Numerical Solution at Spicific Time.
*/
void NormErr(structMesh *mesh, physics *phys,
double time, double **c,
double *L2, double *Linf) {
double **temp = Matrix_create(mesh->ndim1, mesh->ndim2);
ExactSol(mesh, phys, time, temp);
double err = 0;
double maxerr = 0.0;
int dim1, dim2;
for (dim1=0; dim1<mesh->ndim1; dim1++) {
for (dim2=0; dim2<mesh->ndim2; dim2++) {
double dc = c[dim1][dim2]-temp[dim1][dim2];
maxerr = maxf(fabs(dc), maxerr);
err += dc*dc;
}
}
*L2 = sqrt(err/mesh->ndim1/mesh->ndim2);
*Linf = maxerr;
Matrix_free(temp);
return;
}
/* Calculates the moving HRIR target coefficients, target delays, and
* stepping values for the given polar elevation and azimuth in radians.
* Linear interpolation is used to increase the apparent resolution of the
* HRIR data set. The coefficients are also normalized and attenuated by the
* specified gain. Stepping resolution and count is determined using the
* given delta factor between 0.0 and 1.0.
*/
ALuint GetMovingHrtfCoeffs(const struct Hrtf *Hrtf, ALfloat elevation, ALfloat azimuth, ALfloat gain, ALfloat delta, ALint counter, ALfloat (*coeffs)[2], ALuint *delays, ALfloat (*coeffStep)[2], ALint *delayStep)
{
ALuint evidx[2], azidx[2];
ALuint lidx[4], ridx[4];
ALfloat mu[3], blend[4];
ALfloat left, right;
ALfloat step;
ALuint i;
// Claculate elevation indices and interpolation factor.
CalcEvIndices(Hrtf, elevation, evidx, &mu[2]);
// Calculate azimuth indices and interpolation factor for the first
// elevation.
CalcAzIndices(Hrtf, evidx[0], azimuth, azidx, &mu[0]);
// Calculate the first set of linear HRIR indices for left and right
// channels.
lidx[0] = Hrtf->evOffset[evidx[0]] + azidx[0];
lidx[1] = Hrtf->evOffset[evidx[0]] + azidx[1];
ridx[0] = Hrtf->evOffset[evidx[0]] + ((Hrtf->azCount[evidx[0]]-azidx[0]) % Hrtf->azCount[evidx[0]]);
ridx[1] = Hrtf->evOffset[evidx[0]] + ((Hrtf->azCount[evidx[0]]-azidx[1]) % Hrtf->azCount[evidx[0]]);
// Calculate azimuth indices and interpolation factor for the second
// elevation.
CalcAzIndices(Hrtf, evidx[1], azimuth, azidx, &mu[1]);
// Calculate the second set of linear HRIR indices for left and right
// channels.
lidx[2] = Hrtf->evOffset[evidx[1]] + azidx[0];
lidx[3] = Hrtf->evOffset[evidx[1]] + azidx[1];
ridx[2] = Hrtf->evOffset[evidx[1]] + ((Hrtf->azCount[evidx[1]]-azidx[0]) % Hrtf->azCount[evidx[1]]);
ridx[3] = Hrtf->evOffset[evidx[1]] + ((Hrtf->azCount[evidx[1]]-azidx[1]) % Hrtf->azCount[evidx[1]]);
// Calculate the stepping parameters.
delta = maxf(floorf(delta*(Hrtf->sampleRate*0.015f) + 0.5f), 1.0f);
step = 1.0f / delta;
/* Calculate 4 blending weights for 2D bilinear interpolation. */
blend[0] = (1.0f-mu[0]) * (1.0f-mu[2]);
blend[1] = ( mu[0]) * (1.0f-mu[2]);
blend[2] = (1.0f-mu[1]) * ( mu[2]);
blend[3] = ( mu[1]) * ( mu[2]);
/* Calculate the HRIR delays using linear interpolation. Then calculate
* the delay stepping values using the target and previous running
* delays.
*/
left = (ALfloat)(delays[0] - (delayStep[0] * counter));
right = (ALfloat)(delays[1] - (delayStep[1] * counter));
delays[0] = fastf2u(Hrtf->delays[lidx[0]]*blend[0] + Hrtf->delays[lidx[1]]*blend[1] +
Hrtf->delays[lidx[2]]*blend[2] + Hrtf->delays[lidx[3]]*blend[3] +
0.5f) << HRTFDELAY_BITS;
delays[1] = fastf2u(Hrtf->delays[ridx[0]]*blend[0] + Hrtf->delays[ridx[1]]*blend[1] +
Hrtf->delays[ridx[2]]*blend[2] + Hrtf->delays[ridx[3]]*blend[3] +
0.5f) << HRTFDELAY_BITS;
delayStep[0] = fastf2i(step * (delays[0] - left));
delayStep[1] = fastf2i(step * (delays[1] - right));
/* Calculate the sample offsets for the HRIR indices. */
lidx[0] *= Hrtf->irSize;
lidx[1] *= Hrtf->irSize;
lidx[2] *= Hrtf->irSize;
lidx[3] *= Hrtf->irSize;
ridx[0] *= Hrtf->irSize;
ridx[1] *= Hrtf->irSize;
ridx[2] *= Hrtf->irSize;
ridx[3] *= Hrtf->irSize;
/* Calculate the normalized and attenuated target HRIR coefficients using
* linear interpolation when there is enough gain to warrant it. Zero
* the target coefficients if gain is too low. Then calculate the
* coefficient stepping values using the target and previous running
* coefficients.
*/
if(gain > 0.0001f)
{
gain *= 1.0f/32767.0f;
for(i = 0;i < HRIR_LENGTH;i++)
{
left = coeffs[i][0] - (coeffStep[i][0] * counter);
right = coeffs[i][1] - (coeffStep[i][1] * counter);
coeffs[i][0] = (Hrtf->coeffs[lidx[0]+i]*blend[0] +
Hrtf->coeffs[lidx[1]+i]*blend[1] +
Hrtf->coeffs[lidx[2]+i]*blend[2] +
Hrtf->coeffs[lidx[3]+i]*blend[3]) * gain;
coeffs[i][1] = (Hrtf->coeffs[ridx[0]+i]*blend[0] +
Hrtf->coeffs[ridx[1]+i]*blend[1] +
Hrtf->coeffs[ridx[2]+i]*blend[2] +
Hrtf->coeffs[ridx[3]+i]*blend[3]) * gain;
coeffStep[i][0] = step * (coeffs[i][0] - left);
coeffStep[i][1] = step * (coeffs[i][1] - right);
}
}
else
{
for(i = 0;i < HRIR_LENGTH;i++)
{
left = coeffs[i][0] - (coeffStep[i][0] * counter);
right = coeffs[i][1] - (coeffStep[i][1] * counter);
coeffs[i][0] = 0.0f;
coeffs[i][1] = 0.0f;
coeffStep[i][0] = step * -left;
coeffStep[i][1] = step * -right;
}
}
/* The stepping count is the number of samples necessary for the HRIR to
* complete its transition. The mixer will only apply stepping for this
* many samples.
*/
return fastf2u(delta);
}
static ALvoid ALautowahState_process(ALautowahState *state, ALuint SamplesToDo, const ALfloat *SamplesIn, ALfloat (*SamplesOut)[BUFFERSIZE], ALuint NumChannels)
{
ALuint it, kt;
ALuint base;
for(base = 0;base < SamplesToDo;)
{
ALfloat temps[256];
ALuint td = minu(256, SamplesToDo-base);
ALfloat gain = state->GainCtrl;
for(it = 0;it < td;it++)
{
ALfloat smp = SamplesIn[it+base];
ALfloat a[3], b[3];
ALfloat alpha, w0;
ALfloat amplitude;
ALfloat cutoff;
/* Similar to compressor, we get the current amplitude of the
* incoming signal, and attack or release to reach it. */
amplitude = fabsf(smp);
if(amplitude > gain)
gain = minf(gain*state->AttackRate, amplitude);
else if(amplitude < gain)
gain = maxf(gain*state->ReleaseRate, amplitude);
gain = maxf(gain, GAIN_SILENCE_THRESHOLD);
/* FIXME: What range does the filter cover? */
cutoff = lerp(20.0f, 20000.0f, minf(gain/state->PeakGain, 1.0f));
/* The code below is like calling ALfilterState_setParams with
* ALfilterType_LowPass. However, instead of passing a bandwidth,
* we use the resonance property for Q. This also inlines the call.
*/
w0 = F_TAU * cutoff / state->Frequency;
/* FIXME: Resonance controls the resonant peak, or Q. How? Not sure
* that Q = resonance*0.1. */
alpha = sinf(w0) / (2.0f * state->Resonance*0.1f);
b[0] = (1.0f - cosf(w0)) / 2.0f;
b[1] = 1.0f - cosf(w0);
b[2] = (1.0f - cosf(w0)) / 2.0f;
a[0] = 1.0f + alpha;
a[1] = -2.0f * cosf(w0);
a[2] = 1.0f - alpha;
state->LowPass.a1 = a[1] / a[0];
state->LowPass.a2 = a[2] / a[0];
state->LowPass.b1 = b[1] / a[0];
state->LowPass.b2 = b[2] / a[0];
state->LowPass.input_gain = b[0] / a[0];
temps[it] = ALfilterState_processSingle(&state->LowPass, smp);
}
state->GainCtrl = gain;
for(kt = 0;kt < NumChannels;kt++)
{
ALfloat gain = state->Gain[kt];
if(!(fabsf(gain) > GAIN_SILENCE_THRESHOLD))
continue;
for(it = 0;it < td;it++)
SamplesOut[kt][base+it] += gain * temps[it];
}
base += td;
}
}
static ALvoid ALcompressorState_process(ALcompressorState *state, ALuint SamplesToDo, const ALfloat *SamplesIn, ALfloat (*SamplesOut)[BUFFERSIZE])
{
ALuint it, kt;
ALuint base;
for(base = 0;base < SamplesToDo;)
{
ALfloat temps[64];
ALuint td = minu(SamplesToDo-base, 64);
if(state->Enabled)
{
ALfloat output, smp, amplitude;
ALfloat gain = state->GainCtrl;
for(it = 0;it < td;it++)
{
smp = SamplesIn[it+base];
amplitude = fabsf(smp);
if(amplitude > gain)
gain = minf(gain+state->AttackRate, amplitude);
else if(amplitude < gain)
gain = maxf(gain-state->ReleaseRate, amplitude);
output = 1.0f / clampf(gain, 0.5f, 2.0f);
temps[it] = smp * output;
}
state->GainCtrl = gain;
}
else
{
ALfloat output, smp, amplitude;
ALfloat gain = state->GainCtrl;
for(it = 0;it < td;it++)
{
smp = SamplesIn[it+base];
amplitude = 1.0f;
if(amplitude > gain)
gain = minf(gain+state->AttackRate, amplitude);
else if(amplitude < gain)
gain = maxf(gain-state->ReleaseRate, amplitude);
output = 1.0f / clampf(gain, 0.5f, 2.0f);
temps[it] = smp * output;
}
state->GainCtrl = gain;
}
for(kt = 0;kt < MAX_OUTPUT_CHANNELS;kt++)
{
ALfloat gain = state->Gain[kt];
if(!(gain > GAIN_SILENCE_THRESHOLD))
continue;
for(it = 0;it < td;it++)
SamplesOut[kt][base+it] += gain * temps[it];
}
base += td;
}
}
void gfx_csm_render(gfx_cmdqueue cmdqueue, gfx_rendertarget rt,
const struct gfx_view_params* params, struct gfx_batch_item* batch_items, uint batch_cnt,
void* userdata, OUT struct gfx_rpath_result* result)
{
ASSERT(batch_cnt != 0);
PRF_OPENSAMPLE("rpath-csm");
int supports_shared_cbuff = gfx_check_feature(GFX_FEATURE_RANGED_CBUFFERS);
if (supports_shared_cbuff)
csm_submit_batchdata(cmdqueue, batch_items, batch_cnt, g_csm->sharedbuff);
gfx_cmdqueue_resetsrvs(cmdqueue);
gfx_output_setrendertarget(cmdqueue, g_csm->shadow_rt);
gfx_output_setviewport(cmdqueue, 0, 0, CSM_SHADOW_SIZE, CSM_SHADOW_SIZE);
gfx_output_setrasterstate(cmdqueue, g_csm->rs_bias);
gfx_output_setdepthstencilstate(cmdqueue, g_csm->ds_depth, 0);
gfx_output_clearrendertarget(cmdqueue, g_csm->shadow_rt, NULL, 1.0f, 0, GFX_CLEAR_DEPTH);
struct gfx_cblock* cb_frame = g_csm->cb_frame;
struct gfx_cblock* cb_frame_gs = g_csm->cb_frame_gs;
struct mat3f* views[CSM_CASCADE_CNT];
float fovfactors[4];
float texelsz[4] = {1.0f / (float)CSM_SHADOW_SIZE, 0, 0, 0};
for (uint i = 0; i < CSM_CASCADE_CNT && i < 4; i++) {
views[i] = &g_csm->cascades[i].view;
fovfactors[i] = maxf(g_csm->cascades[i].proj.m11, g_csm->cascades[i].proj.m22);
}
gfx_cb_set4f(cb_frame, SHADER_NAME(c_texelsz), texelsz);
gfx_cb_set4f(cb_frame, SHADER_NAME(c_fovfactors), fovfactors);
gfx_cb_set4f(cb_frame, SHADER_NAME(c_lightdir), g_csm->light_dir.f);
gfx_cb_set3mvp(cb_frame, SHADER_NAME(c_views), (const struct mat3f**)views, CSM_CASCADE_CNT);
gfx_cb_set4mv(cb_frame, SHADER_NAME(c_cascade_mats), g_csm->cascade_vps, CSM_CASCADE_CNT);
gfx_shader_updatecblock(cmdqueue, cb_frame);
gfx_cb_set4fv(cb_frame_gs, SHADER_NAME(c_cascade_planes), g_csm->cascade_planes,
4*CSM_CASCADE_CNT);
gfx_shader_updatecblock(cmdqueue, cb_frame_gs);
for (uint i = 0; i < batch_cnt; i++) {
struct gfx_batch_item* bitem = &batch_items[i];
struct gfx_shader* shader = gfx_shader_get(bitem->shader_id);
ASSERT(shader);
gfx_shader_bind(cmdqueue, shader);
/* do not send cb_xforms to shader if we are using shared buffer (bind later before draw) */
struct gfx_cblock* cbs[3];
uint xforms_shared_idx;
if (supports_shared_cbuff) {
cbs[0] = cb_frame;
cbs[1] = cb_frame_gs;
xforms_shared_idx = 2;
} else {
cbs[0] = cb_frame;
cbs[1] = cb_frame_gs;
cbs[2] = g_csm->cb_xforms;
xforms_shared_idx = 3;
}
gfx_shader_bindcblocks(cmdqueue, shader, (const struct gfx_cblock**)cbs, xforms_shared_idx);
/* batch draw */
for (int k = 0; k < bitem->nodes.item_cnt; k++) {
struct gfx_batch_node* bnode_first = &((struct gfx_batch_node*)bitem->nodes.buffer)[k];
csm_preparebatchnode(cmdqueue, bnode_first, shader);
if (bnode_first->poses[0] != NULL) {
gfx_shader_bindcblock_tbuffer(cmdqueue, shader, SHADER_NAME(tb_skins),
g_csm->tb_skins);
}
struct linked_list* node = bnode_first->bll;
while (node != NULL) {
struct gfx_batch_node* bnode = (struct gfx_batch_node*)node->data;
csm_drawbatchnode(cmdqueue, bnode, shader, xforms_shared_idx);
node = node->next;
}
}
}
/* switch back */
gfx_output_setrasterstate(cmdqueue, NULL);
gfx_output_setdepthstencilstate(cmdqueue, NULL, 0);
if (g_csm->debug_csm)
csm_renderpreview(cmdqueue, params);
PRF_CLOSESAMPLE(); /* csm */
}
/* read command line, initialize, and create threads */
int main(int argc, char *argv[]) {
int i, j;
long l; /* use long in case of a 64-bit system */
point glob_min, glob_max;
int glob_sum;
worker_input work_params[MAXWORKERS];
pthread_attr_t attr;
pthread_t workerid[MAXWORKERS];
/* set global thread attributes */
pthread_attr_init(&attr);
pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM);
// Set the threads to be joinable
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
/* initialize mutex and condition variable */
pthread_mutex_init(&barrier, NULL);
pthread_cond_init(&go, NULL);
/* read command line args if any */
size = (argc > 1)? atoi(argv[1]) : MAXSIZE;
numWorkers = (argc > 2)? atoi(argv[2]) : MAXWORKERS;
if (size > MAXSIZE) size = MAXSIZE;
if (numWorkers > MAXWORKERS) numWorkers = MAXWORKERS;
stripSize = size/numWorkers;
/* initialize the matrix */
for (i = 0; i < size; i++) {
for (j = 0; j < size; j++) {
matrix[i][j] = 1;//rand()%999;
}
}
/* print the matrix */
#ifdef DEBUG
for (i = 0; i < size; i++) {
printf("[ ");
for (j = 0; j < size; j++) {
printf(" %d", matrix[i][j]);
}
printf(" ]\n");
}
#endif
/* do the parallel work: create the workers */
// here we send in a an argument which we will get back loaded with min, max and sum values for that thread
start_time = read_timer();
for (l = 0; l < numWorkers; ++l){
worker_input wi = {.id = l, .sum = 0};
work_params[l] = wi; // saved in an array for perfomance issues (removing this meant a slowdown of a factor 4)
pthread_create(&workerid[l], &attr, Worker, (void *) &work_params[l]);
}
worker_input wi;
//initialize end variables to first value of the matrix
glob_min.x = glob_max.x = 0;
glob_min.y = glob_max.y = 0;
glob_sum = 0;
// A variable for the pointer so we can handle the return value
void* status;
for(l = 0; l < numWorkers; ++l){
// we join the the different threads into the main thread and sums up the total results
pthread_join(workerid[l], &status);
wi = *((worker_input *) status);
glob_sum += wi.sum;
glob_min = *(minf(&glob_min, &wi.min));
glob_max = *(maxf(&glob_max, &wi.max));
}
// get end time
end_time = read_timer();
// print results
printf("The total is %d\n", glob_sum);
printf("The smallest value is %d at x:%d, y:%d\n", matrix[glob_min.x][glob_min.y], glob_min.x, glob_min.y );
printf("The largest value is %d at x:%d, y:%d\n", matrix[glob_max.x][glob_max.y], glob_max.x, glob_max.y );
printf("The execution time is %g sec\n", end_time - start_time);
// printf("Total values are: Sum %d, min %d, max %d\n", glob_sum, glob_min, glob_max);
pthread_exit(NULL);
}
/* Each worker sums the values in one strip of the matrix.
After a barrier, worker(0) computes and prints the total
This is pretty much the same as in matrixSum_min_max.c except that
the barrier function and calculation of final min and max values has
moved to the main thread.
*/
void *Worker(void *arg) {
worker_input* wi = (worker_input*) arg;
point current;
int i, j, first, last;
#ifdef DEBUG
printf("worker %ld (pthread id %d) has started\n", (*wi).id, pthread_self());
#endif
/* determine first and last rows of my strip */
first = (*wi).id*stripSize;
last = ((*wi).id == numWorkers - 1) ? (size - 1) : (first + stripSize - 1);
(*wi).min.x = (*wi).max.x = first;
(*wi).min.y = (*wi).max.y = 0;
/* sum values in my strip */
for (i = first; i <= last; i++){
current.x =i;
for (j = 0; j < size; j++){
current.y = j;
(*wi).sum += matrix[i][j];
(*wi).min = *(minf( &((*wi).min), ¤t ));
(*wi).max = *(maxf( &((*wi).max), ¤t ));
}
}
pthread_exit(wi);
}
void sreFrustum::CalculateLightScissors(sreLight *light) {
if (light->type & (SRE_LIGHT_SPOT | SRE_LIGHT_BEAM)) {
// Approximate the bounding volume of the light by the bounding box of the bounding cylinder.
Vector3D up;
if (fabsf(light->cylinder.axis.x) < 0.01f && fabsf(light->cylinder.axis.z) < 0.01f) {
if (light->cylinder.axis.y > 0)
up = Vector3D(0, 0, - 1.0f);
else
up = Vector3D(0, 0, 1.0f);
}
else
up = Vector3D(0, 1.0f, 0);
// Calculate tangent planes.
Vector3D N2 = Cross(up, light->cylinder.axis);
N2.Normalize();
Vector3D N3 = Cross(light->cylinder.axis, N2);
Point3DPadded B[8];
Point3D E1 = light->cylinder.center - 0.5f * light->cylinder.length * light->cylinder.axis;
Point3D E2 = E1 + light->cylinder.length * light->cylinder.axis;
B[0] = E2 + light->cylinder.radius * N2 + light->cylinder.radius * N3;
B[1] = E1 + light->cylinder.radius * N2 + light->cylinder.radius * N3;
B[2] = E1 - light->cylinder.radius * N2 + light->cylinder.radius * N3;
B[3] = E2 - light->cylinder.radius * N2 + light->cylinder.radius * N3;
B[4] = E2 + light->cylinder.radius * N2 - light->cylinder.radius * N3;
B[5] = E1 + light->cylinder.radius * N2 - light->cylinder.radius * N3;
B[6] = E1 - light->cylinder.radius * N2 - light->cylinder.radius * N3;
B[7] = E2 - light->cylinder.radius * N2 - light->cylinder.radius * N3;
scissors.SetEmptyRegion();
scissors.UpdateWithWorldSpaceBoundingBox(B, 8, *this);
scissors.ClampEmptyRegion();
scissors.ClampRegionAndDepthBounds();
#if 0
printf("Light %d (spot): ");
sreMessage(SRE_MESSAGE_INFO,
"Scissors = (%f, %f, %f, %f)", scissors.left, scissors.right, scissors.bottom, scissors.top);
#endif
return;
}
// Point source light.
scissors.SetFullRegionAndDepthBounds();
// Transform the light position from world space to eye space.
Point3D L = (sre_internal_view_matrix * light->vector).GetPoint3D();
// Calculate the depth range.
// The near and far tangent planes parallel to the z-axis can be represented by 4D vectors
// T= <0, 0, 1.0, Lz + r> and T = <0, 0, 1.0, L.z - r> respectively.
// Multiply the z coordinates by the projection matrix to arrive at projected depths.
// Divide by w coordinate. Note that we assume an infinite view frustum.
float Lz_near = L.z + light->sphere.radius;
if (Lz_near <= - nearD) {
Vector4D V = sre_internal_projection_matrix * Vector4D(0, 0, Lz_near, 1.0);
scissors.near = 0.5 * (double)V.z / (double)V.w + 0.5;
}
float Lz_far = L.z - light->sphere.radius;
if (Lz_far <= - nearD) {
Vector4D V = sre_internal_projection_matrix * Vector4D(0, 0, Lz_far, 1.0);
scissors.far = 0.5 * (double)V.z / (double)V.w + 0.5;
}
else
scissors.far = 0;
// Calculate the determinant D
float D = 4 * (sqrf(light->sphere.radius) * sqrf(L.x) - (sqrf(L.x) + sqrf(L.z)) *
(sqrf(light->sphere.radius) - sqrf(L.z)));
if (D <= 0) {
// The light source's bounding sphere fills the entire viewport.
return;
}
// Calculate the tangent planes <Nx, 0, Nz, 0> of the light volume.
float Nx1 = (light->sphere.radius * L.x + sqrtf(D / 4)) /
(sqrf(L.x) + sqrf(L.z));
float Nx2 = (light->sphere.radius * L.x - sqrtf(D / 4)) /
(sqrf(L.x) + sqrf(L.z));
float Nz1 = (light->sphere.radius - Nx1 * L.x) / L.z;
float Nz2 = (light->sphere.radius - Nx2 * L.x) / L.z;
// The point P at which the plane T is tangent to the bounding sphere is
// given by <Lx - rNx, 0, Lz - rNz, 1>
float Pz1 = L.z - light->sphere.radius * Nz1;
float Pz2 = L.z - light->sphere.radius * Nz2;
if (Pz1 < 0) {
// Plane 1 may shrink the scissors rectangle.
float x = Nz1 * e / Nx1;
float Px1 = L.x - light->sphere.radius * Nx1;
#if 0
int xvp = 0 + (x + 1) * sre_internal_window_width / 2;
if (Px1 < L.x) {
// Left-side boundary
scissors.left = max(xvp, 0);
}
else {
// Right-side boundary
scissors.right = min(xvp, sre_internal_window_width);
}
#else
if (Px1 < L.x) {
// Left-side boundary
scissors.left = maxf(x, - 1.0);
}
else {
// Right-side boundary
scissors.right = minf(x, 1.0);
}
#endif
}
if (Pz2 < 0) {
// Plane 2 may shrink the scissors rectangle.
float x = Nz2 * e / Nx2;
float Px2 = L.x - light->sphere.radius * Nx2;
#if 0
int xvp = 0 + (x + 1) * sre_internal_window_width / 2;
if (Px2 < L.x) {
// Left-side boundary
scissors.left = max(xvp, 0);
}
else {
// Right-side boundary
scissors.right = min(xvp, sre_internal_window_width);
}
#else
if (Px2 < L.x) {
// Left-side boundary
scissors.left = maxf(x, - 1.0);
}
else {
// Right-side boundary
scissors.right = minf(x, 1.0);
}
#endif
}
// Calculate the tangent planes <0, Ny, Nz, 0> of the light volume.
float term1 = sqrf(light->sphere.radius) * sqrf(L.y) -
(sqrf(L.y) + sqrf(L.z)) *
(sqrf(light->sphere.radius) - sqrf(L.z));
float Ny1 = (light->sphere.radius * L.y + sqrtf(term1)) /
(sqrf(L.y) + sqrf(L.z));
float Ny2 = (light->sphere.radius * L.y - sqrtf(term1)) /
(sqrf(L.y) + sqrf(L.z));
Nz1 = (light->sphere.radius - Ny1 * L.y) / L.z;
Nz2 = (light->sphere.radius - Ny2 * L.y) / L.z;
Pz1 = L.z - light->sphere.radius * Nz1;
Pz2 = L.z - light->sphere.radius * Nz2;
if (Pz1 < 0) {
// Plane 3 may shrink the scissors rectangle.
float y = Nz1 * e * ratio / Ny1;
float Py1 = L.y - light->sphere.radius * Ny1;
#if 0
int yvp = 0 + (y + 1) * sre_internal_window_height / 2;
if (Py1 < L.y) {
// Bottom boundary.
scissors.bottom = max(yvp, 0);
}
else {
// Top boundary.
scissors.top = min(yvp, sre_internal_window_height);
}
#else
if (Py1 < L.y) {
// Bottom boundary.
scissors.bottom = maxf(y, - 1.0);
}
else {
// Top boundary.
scissors.top = minf(y, 1.0);
}
#endif
}
if (Pz2 < 0) {
// Plane 4 may shrink the scissors rectangle.
float y = Nz2 * e * ratio / Ny2;
float Py2 = L.y - light->sphere.radius * Ny2;
#if 0
int yvp = 0 + (y + 1) * sre_internal_window_height / 2;
if (Py2 < L.y) {
// Bottom boundary.
scissors.bottom = max(yvp, 0);
}
else {
// Top boundary.
scissors.top = min(yvp, sre_internal_window_height);
}
#else
if (Py2 < L.y) {
// Bottom boundary.
scissors.bottom = maxf(y, - 1.0);
}
else {
// Top boundary.
scissors.top = minf(y, 1.0);
}
#endif
}
// scissors.ClampRegionToScreen();
// printf("Scissors = (%f, %f, %f, %f)\n", scissors.left, scissors.right, scissors.bottom, scissors.top);
}
void sreFrustum::CalculateShadowCasterVolume(const Vector4D& lightpos, int nu_frustum_planes) {
nu_frustum_planes = maxf(nu_frustum_planes, SRE_NU_FRUSTUM_PLANES);
// Note: for beam lights, this might be inaccurate.
if (lightpos.w == 1.0f && Intersects(lightpos.GetPoint3D(), frustum_world)) {
// If the point light position is inside the view frustum, we only need to consider the
// set of visible objects.
shadow_caster_volume.nu_planes = nu_frustum_planes;
for (int i = 0; i < nu_frustum_planes; i++)
shadow_caster_volume.plane[i] = frustum_world.plane[i];
goto end;
}
// Calculate the convex hull enclosing the view frustum and the light source.
shadow_caster_volume.nu_planes = 0;
// Calculate the dot products between the frustum planes and the light source.
float dot[6] DST_ALIGNED(16);
// The SIMD-accelerated version check for 16-bytes aligned arrays, which is hard to
// guarantee in this case; otherwise, no SIMD will be used.
#if 1
dstCalculateDotProductsNx1(nu_frustum_planes, &frustum_world.plane[0],
lightpos, &dot[0]);
#else
for (int i = 0; i < nu_frustum_planes; i++)
dot[i] = Dot(frustum_world.plane[i], lightpos);
#endif
// For each frustum plane, add it if it is part of the convex hull.
for (int i = 0; i < nu_frustum_planes; i++)
if (dot[i] > 0) {
shadow_caster_volume.plane[shadow_caster_volume.nu_planes] = frustum_world.plane[i];
shadow_caster_volume.nu_planes++;
}
// printf("Shadow caster volume planes: from frustum: %d\n", shadow_caster_volume.nu_planes);
if (shadow_caster_volume.nu_planes == 0 && lightpos.w == 1.0f) {
// Special case: the point light source is behind the camera and there is no far plane.
// As a result, there is no plane from the frustum with a positive dot product. In
// this case, construct a volume consisting of the lightsource and four planes parallel
// to the frustum side planes but starting at the lightsource.
for (int i = 0; i < 4; i++) {
// Copy the normal.
shadow_caster_volume.plane[i] = frustum_world.plane[i];
// Calculate the distance such that the lightsource is in the plane.
shadow_caster_volume.plane[i].w = - Dot(frustum_world.plane[i].GetVector3D(),
lightpos.GetPoint3D());
}
shadow_caster_volume.nu_planes = 4;
goto end;
}
// For each pair of adjacent frustum planes, if one has a positive dot product and the other
// does not, calculate a new plane defined by the edge between those two frustum planes and
// the position of the light source, making sure the plane's normal direction faces inward.
// For directional lights, the plane runs parallel to the direction of the light.
nu_shadow_caster_edges = 0;
int n;
if (nu_frustum_planes == 5)
n = 8;
else
n = 12;
for (int i = 0; i < n; i++)
if ((dot[adjacent_plane[i].plane0] > 0 && dot[adjacent_plane[i].plane1] <= 0) ||
(dot[adjacent_plane[i].plane0] <= 0 && dot[adjacent_plane[i].plane1] > 0)) {
// printf("Setting plane from adjacent planes %d and %d using frustum vertices %d and %d\n",
// adjacent_plane[i].plane0, adjacent_plane[i].plane1, adjacent_plane[i].vertex0,
// adjacent_plane[i].vertex1);
if (lightpos.w == 1.0f)
shadow_caster_volume.plane[shadow_caster_volume.nu_planes] = dstPlaneFromPoints(
frustum_world.hull.vertex[adjacent_plane[i].vertex0],
frustum_world.hull.vertex[adjacent_plane[i].vertex1],
lightpos.GetPoint3D()
);
else {
shadow_caster_volume.plane[shadow_caster_volume.nu_planes] = dstPlaneFromPoints(
frustum_world.hull.vertex[adjacent_plane[i].vertex0],
frustum_world.hull.vertex[adjacent_plane[i].vertex1],
frustum_world.hull.vertex[adjacent_plane[i].vertex0] + lightpos.GetVector3D()
);
shadow_caster_edge[nu_shadow_caster_edges][0] = adjacent_plane[i].vertex0;
shadow_caster_edge[nu_shadow_caster_edges][1] = adjacent_plane[i].vertex1;
nu_shadow_caster_edges++;
}
// Make sure the normal is pointed inward, check by taking the dot product with the frustum
// "centroid".
shadow_caster_volume.plane[shadow_caster_volume.nu_planes].OrientPlaneTowardsPoint(
frustum_world.sphere.center);
shadow_caster_volume.nu_planes++;
}
end: ;
#if 0
printf("Light (%lf, %lf, %lf, %lf), %d planes in shadow caster volume.\n",
lightpos.x, lightpos.y, lightpos.z, lightpos.w, shadow_caster_volume.nu_planes);
for (int i = 0; i < shadow_caster_volume.nu_planes; i++)
printf("Plane %d: (%lf, %lf, %lf, %lf) ", i,
shadow_caster_volume.plane[i].x, shadow_caster_volume.plane[i].y,
shadow_caster_volume.plane[i].y, shadow_caster_volume.plane[i].w);
printf("\n");
#endif
}