static void _BezierPatchLowQuality(u8 *&dest, u16 *&indices, int &count, int tess_u, int tess_v, const BezierPatch &patch, u32 origVertType) {
const float third = 1.0f / 3.0f;
// Fast and easy way - just draw the control points, generate some very basic normal vector subsitutes.
// Very inaccurate though but okay for Loco Roco. Maybe should keep it as an option.
float u_base = patch.u_index / 3.0f;
float v_base = patch.v_index / 3.0f;
GEPatchPrimType prim_type = patch.primType;
for (int tile_v = 0; tile_v < 3; tile_v++) {
for (int tile_u = 0; tile_u < 3; tile_u++) {
int point_index = tile_u + tile_v * 4;
SimpleVertex v0 = *patch.points[point_index];
SimpleVertex v1 = *patch.points[point_index + 1];
SimpleVertex v2 = *patch.points[point_index + 4];
SimpleVertex v3 = *patch.points[point_index + 5];
// Generate UV. TODO: Do this even if UV specified in control points?
if ((origVertType & GE_VTYPE_TC_MASK) == 0) {
float u = u_base + tile_u * third;
float v = v_base + tile_v * third;
v0.uv[0] = u;
v0.uv[1] = v;
v1.uv[0] = u + third;
v1.uv[1] = v;
v2.uv[0] = u;
v2.uv[1] = v + third;
v3.uv[0] = u + third;
v3.uv[1] = v + third;
}
// Generate normal if lighting is enabled (otherwise there's no point).
// This is a really poor quality algorithm, we get facet normals.
if (patch.computeNormals) {
Vec3Packedf norm = Cross(v1.pos - v0.pos, v2.pos - v0.pos);
norm.Normalize();
if (patch.patchFacing)
norm *= -1.0f;
v0.nrm = norm;
v1.nrm = norm;
v2.nrm = norm;
v3.nrm = norm;
}
int total = patch.index * 3 * 3 * 4; // A patch has 3x3 tiles, and each tiles have 4 vertices.
int tile_index = tile_u + tile_v * 3;
int idx0 = total + tile_index * 4 + 0;
int idx1 = total + tile_index * 4 + 1;
int idx2 = total + tile_index * 4 + 2;
int idx3 = total + tile_index * 4 + 3;
CopyQuad(dest, &v0, &v1, &v2, &v3);
CopyQuadIndex(indices, prim_type, idx0, idx1, idx2, idx3);
count += 6;
}
}
}
// Prepare mesh of one patch for "Instanced Tessellation".
static void TessellateBezierPatchHardware(u8 *&dest, u16 *indices, int &count, int tess_u, int tess_v, GEPatchPrimType primType) {
SimpleVertex *&vertices = (SimpleVertex*&)dest;
float inv_u = 1.0f / (float)tess_u;
float inv_v = 1.0f / (float)tess_v;
// Generating simple input vertices for the bezier-computing vertex shader.
for (int tile_v = 0; tile_v < tess_v + 1; ++tile_v) {
for (int tile_u = 0; tile_u < tess_u + 1; ++tile_u) {
SimpleVertex &vert = vertices[tile_v * (tess_u + 1) + tile_u];
vert.pos.x = (float)tile_u * inv_u;
vert.pos.y = (float)tile_v * inv_v;
}
}
// Combine the vertices into triangles.
for (int tile_v = 0; tile_v < tess_v; ++tile_v) {
for (int tile_u = 0; tile_u < tess_u; ++tile_u) {
int idx0 = tile_v * (tess_u + 1) + tile_u;
int idx1 = tile_v * (tess_u + 1) + tile_u + 1;
int idx2 = (tile_v + 1) * (tess_u + 1) + tile_u;
int idx3 = (tile_v + 1) * (tess_u + 1) + tile_u + 1;
CopyQuadIndex(indices, primType, idx0, idx1, idx2, idx3);
count += 6;
}
}
}
// Prepare mesh of one patch for "Instanced Tessellation".
static void TessellateSplinePatchHardware(u8 *&dest, u16 *indices, int &count, const SplinePatchLocal &spatch) {
SimpleVertex *&vertices = (SimpleVertex*&)dest;
float inv_u = 1.0f / (float)spatch.tess_u;
float inv_v = 1.0f / (float)spatch.tess_v;
// Generating simple input vertices for the spline-computing vertex shader.
for (int tile_v = 0; tile_v < spatch.tess_v + 1; ++tile_v) {
for (int tile_u = 0; tile_u < spatch.tess_u + 1; ++tile_u) {
SimpleVertex &vert = vertices[tile_v * (spatch.tess_u + 1) + tile_u];
vert.pos.x = (float)tile_u * inv_u;
vert.pos.y = (float)tile_v * inv_v;
// TODO: Move to shader uniform and unify this method spline and bezier if necessary.
// For compute normal
vert.nrm.x = inv_u;
vert.nrm.y = inv_v;
}
}
// Combine the vertices into triangles.
for (int tile_v = 0; tile_v < spatch.tess_v; ++tile_v) {
for (int tile_u = 0; tile_u < spatch.tess_u; ++tile_u) {
int idx0 = tile_v * (spatch.tess_u + 1) + tile_u;
int idx1 = tile_v * (spatch.tess_u + 1) + tile_u + 1;
int idx2 = (tile_v + 1) * (spatch.tess_u + 1) + tile_u;
int idx3 = (tile_v + 1) * (spatch.tess_u + 1) + tile_u + 1;
CopyQuadIndex(indices, spatch.primType, idx0, idx1, idx2, idx3);
count += 6;
}
}
}
static void _BezierPatchHighQuality(u8 *&dest, u16 *&indices, int &count, int tess_u, int tess_v, const BezierPatch &patch, u32 origVertType, int maxVertices) {
const float third = 1.0f / 3.0f;
// Downsample until it fits, in case crazy tesselation factors are sent.
while ((tess_u + 1) * (tess_v + 1) > maxVertices) {
tess_u /= 2;
tess_v /= 2;
}
// First compute all the vertices and put them in an array
SimpleVertex *&vertices = (SimpleVertex*&)dest;
Vec3Packedf *horiz = new Vec3Packedf[(tess_u + 1) * 4];
Vec3Packedf *horiz2 = horiz + (tess_u + 1) * 1;
Vec3Packedf *horiz3 = horiz + (tess_u + 1) * 2;
Vec3Packedf *horiz4 = horiz + (tess_u + 1) * 3;
Vec3Packedf *derivU1 = new Vec3Packedf[(tess_u + 1) * 4];
Vec3Packedf *derivU2 = derivU1 + (tess_u + 1) * 1;
Vec3Packedf *derivU3 = derivU1 + (tess_u + 1) * 2;
Vec3Packedf *derivU4 = derivU1 + (tess_u + 1) * 3;
bool computeNormals = patch.computeNormals;
// Precompute the horizontal curves to we only have to evaluate the vertical ones.
for (int i = 0; i < tess_u + 1; i++) {
float u = ((float)i / (float)tess_u);
horiz[i] = Bernstein3D(patch.points[0]->pos, patch.points[1]->pos, patch.points[2]->pos, patch.points[3]->pos, u);
horiz2[i] = Bernstein3D(patch.points[4]->pos, patch.points[5]->pos, patch.points[6]->pos, patch.points[7]->pos, u);
horiz3[i] = Bernstein3D(patch.points[8]->pos, patch.points[9]->pos, patch.points[10]->pos, patch.points[11]->pos, u);
horiz4[i] = Bernstein3D(patch.points[12]->pos, patch.points[13]->pos, patch.points[14]->pos, patch.points[15]->pos, u);
if (computeNormals) {
derivU1[i] = Bernstein3DDerivative(patch.points[0]->pos, patch.points[1]->pos, patch.points[2]->pos, patch.points[3]->pos, u);
derivU2[i] = Bernstein3DDerivative(patch.points[4]->pos, patch.points[5]->pos, patch.points[6]->pos, patch.points[7]->pos, u);
derivU3[i] = Bernstein3DDerivative(patch.points[8]->pos, patch.points[9]->pos, patch.points[10]->pos, patch.points[11]->pos, u);
derivU4[i] = Bernstein3DDerivative(patch.points[12]->pos, patch.points[13]->pos, patch.points[14]->pos, patch.points[15]->pos, u);
}
}
for (int tile_v = 0; tile_v < tess_v + 1; ++tile_v) {
for (int tile_u = 0; tile_u < tess_u + 1; ++tile_u) {
float u = ((float)tile_u / (float)tess_u);
float v = ((float)tile_v / (float)tess_v);
float bu = u;
float bv = v;
// TODO: Should be able to precompute the four curves per U, then just Bernstein per V. Will benefit large tesselation factors.
const Vec3Packedf &pos1 = horiz[tile_u];
const Vec3Packedf &pos2 = horiz2[tile_u];
const Vec3Packedf &pos3 = horiz3[tile_u];
const Vec3Packedf &pos4 = horiz4[tile_u];
SimpleVertex &vert = vertices[tile_v * (tess_u + 1) + tile_u];
if (computeNormals) {
const Vec3Packedf &derivU1_ = derivU1[tile_u];
const Vec3Packedf &derivU2_ = derivU2[tile_u];
const Vec3Packedf &derivU3_ = derivU3[tile_u];
const Vec3Packedf &derivU4_ = derivU4[tile_u];
Vec3Packedf derivU = Bernstein3D(derivU1_, derivU2_, derivU3_, derivU4_, bv);
Vec3Packedf derivV = Bernstein3DDerivative(pos1, pos2, pos3, pos4, bv);
vert.nrm = Cross(derivU, derivV).Normalized();
if (patch.patchFacing)
vert.nrm *= -1.0f;
}
else {
vert.nrm.SetZero();
}
vert.pos = Bernstein3D(pos1, pos2, pos3, pos4, bv);
if ((origVertType & GE_VTYPE_TC_MASK) == 0) {
// Generate texcoord
vert.uv[0] = u + patch.u_index * third;
vert.uv[1] = v + patch.v_index * third;
} else {
// Sample UV from control points
patch.sampleTexUV(u, v, vert.uv[0], vert.uv[1]);
}
if (origVertType & GE_VTYPE_COL_MASK) {
patch.sampleColor(u, v, vert.color);
} else {
memcpy(vert.color, patch.points[0]->color, 4);
}
}
}
delete[] derivU1;
delete[] horiz;
GEPatchPrimType prim_type = patch.primType;
// Combine the vertices into triangles.
for (int tile_v = 0; tile_v < tess_v; ++tile_v) {
for (int tile_u = 0; tile_u < tess_u; ++tile_u) {
int total = patch.index * (tess_u + 1) * (tess_v + 1);
int idx0 = total + tile_v * (tess_u + 1) + tile_u;
int idx1 = total + tile_v * (tess_u + 1) + tile_u + 1;
int idx2 = total + (tile_v + 1) * (tess_u + 1) + tile_u;
int idx3 = total + (tile_v + 1) * (tess_u + 1) + tile_u + 1;
CopyQuadIndex(indices, prim_type, idx0, idx1, idx2, idx3);
count += 6;
}
}
dest += (tess_u + 1) * (tess_v + 1) * sizeof(SimpleVertex);
}
static void SplinePatchFullQuality(u8 *&dest, u16 *indices, int &count, const SplinePatchLocal &spatch, u32 origVertType, int quality, int maxVertices) {
// Full (mostly) correct tessellation of spline patches.
// Not very fast.
float *knot_u = new float[spatch.count_u + 4];
float *knot_v = new float[spatch.count_v + 4];
spline_knot(spatch.count_u - 1, spatch.type_u, knot_u);
spline_knot(spatch.count_v - 1, spatch.type_v, knot_v);
// Increase tesselation based on the size. Should be approximately right?
int patch_div_s = (spatch.count_u - 3) * spatch.tess_u;
int patch_div_t = (spatch.count_v - 3) * spatch.tess_v;
if (quality > 1) {
patch_div_s /= quality;
patch_div_t /= quality;
}
// Downsample until it fits, in case crazy tesselation factors are sent.
while ((patch_div_s + 1) * (patch_div_t + 1) > maxVertices) {
patch_div_s /= 2;
patch_div_t /= 2;
}
if (patch_div_s < 2) patch_div_s = 2;
if (patch_div_t < 2) patch_div_t = 2;
// First compute all the vertices and put them in an array
SimpleVertex *&vertices = (SimpleVertex*&)dest;
float tu_width = (float)spatch.count_u - 3.0f;
float tv_height = (float)spatch.count_v - 3.0f;
// int max_idx = spatch.count_u * spatch.count_v;
bool computeNormals = spatch.computeNormals;
float one_over_patch_div_s = 1.0f / (float)(patch_div_s);
float one_over_patch_div_t = 1.0f / (float)(patch_div_t);
for (int tile_v = 0; tile_v < patch_div_t + 1; tile_v++) {
float v = (float)tile_v * (float)(spatch.count_v - 3) * one_over_patch_div_t;
if (v < 0.0f)
v = 0.0f;
for (int tile_u = 0; tile_u < patch_div_s + 1; tile_u++) {
float u = (float)tile_u * (float)(spatch.count_u - 3) * one_over_patch_div_s;
if (u < 0.0f)
u = 0.0f;
SimpleVertex *vert = &vertices[tile_v * (patch_div_s + 1) + tile_u];
Vec4f vert_color(0, 0, 0, 0);
Vec3f vert_pos;
vert_pos.SetZero();
Vec3f vert_nrm;
if (origNrm) {
vert_nrm.SetZero();
}
if (origCol) {
vert_color.SetZero();
} else {
memcpy(vert->color, spatch.points[0]->color, 4);
}
if (origTc) {
vert->uv[0] = 0.0f;
vert->uv[1] = 0.0f;
} else {
vert->uv[0] = tu_width * ((float)tile_u * one_over_patch_div_s);
vert->uv[1] = tv_height * ((float)tile_v * one_over_patch_div_t);
}
// Collect influences from surrounding control points.
float u_weights[4];
float v_weights[4];
int iu = (int)u;
int iv = (int)v;
// TODO: Would really like to fix the surrounding logic somehow to get rid of these but I can't quite get it right..
// Without the previous epsilons and with large count_u, we will end up doing an out of bounds access later without these.
if (iu >= spatch.count_u - 3) iu = spatch.count_u - 4;
if (iv >= spatch.count_v - 3) iv = spatch.count_v - 4;
spline_n_4(iu, u, knot_u, u_weights);
spline_n_4(iv, v, knot_v, v_weights);
// Handle degenerate patches. without this, spatch.points[] may read outside the number of initialized points.
int patch_w = std::min(spatch.count_u - iu, 4);
int patch_h = std::min(spatch.count_v - iv, 4);
for (int ii = 0; ii < patch_w; ++ii) {
for (int jj = 0; jj < patch_h; ++jj) {
float u_spline = u_weights[ii];
float v_spline = v_weights[jj];
float f = u_spline * v_spline;
if (f > 0.0f) {
#ifdef _M_SSE
Vec4f fv(_mm_set_ps1(f));
#else
Vec4f fv = Vec4f::AssignToAll(f);
#endif
int idx = spatch.count_u * (iv + jj) + (iu + ii);
/*
if (idx >= max_idx) {
char temp[512];
snprintf(temp, sizeof(temp), "count_u: %d count_v: %d patch_w: %d patch_h: %d ii: %d jj: %d iu: %d iv: %d patch_div_s: %d patch_div_t: %d\n", spatch.count_u, spatch.count_v, patch_w, patch_h, ii, jj, iu, iv, patch_div_s, patch_div_t);
OutputDebugStringA(temp);
DebugBreak();
}*/
SimpleVertex *a = spatch.points[idx];
AccumulateWeighted(vert_pos, a->pos, fv);
if (origTc) {
vert->uv[0] += a->uv[0] * f;
vert->uv[1] += a->uv[1] * f;
}
if (origCol) {
Vec4f a_color = Vec4f::FromRGBA(a->color_32);
AccumulateWeighted(vert_color, a_color, fv);
}
if (origNrm) {
AccumulateWeighted(vert_nrm, a->nrm, fv);
}
}
}
}
vert->pos = vert_pos;
if (origNrm) {
#ifdef _M_SSE
const __m128 normalize = SSENormalizeMultiplier(useSSE4, vert_nrm.vec);
vert_nrm.vec = _mm_mul_ps(vert_nrm.vec, normalize);
#else
vert_nrm.Normalize();
#endif
vert->nrm = vert_nrm;
} else {
vert->nrm.SetZero();
vert->nrm.z = 1.0f;
}
if (origCol) {
vert->color_32 = vert_color.ToRGBA();
}
}
}
delete[] knot_u;
delete[] knot_v;
// Hacky normal generation through central difference.
if (spatch.computeNormals && !origNrm) {
#ifdef _M_SSE
const __m128 facing = spatch.patchFacing ? _mm_set_ps1(-1.0f) : _mm_set_ps1(1.0f);
#endif
for (int v = 0; v < patch_div_t + 1; v++) {
Vec3f vl_pos = vertices[v * (patch_div_s + 1)].pos;
Vec3f vc_pos = vertices[v * (patch_div_s + 1)].pos;
for (int u = 0; u < patch_div_s + 1; u++) {
const int l = std::max(0, u - 1);
const int t = std::max(0, v - 1);
const int r = std::min(patch_div_s, u + 1);
const int b = std::min(patch_div_t, v + 1);
const Vec3f vr_pos = vertices[v * (patch_div_s + 1) + r].pos;
#ifdef _M_SSE
const __m128 right = _mm_sub_ps(vr_pos.vec, vl_pos.vec);
const Vec3f vb_pos = vertices[b * (patch_div_s + 1) + u].pos;
const Vec3f vt_pos = vertices[t * (patch_div_s + 1) + u].pos;
const __m128 down = _mm_sub_ps(vb_pos.vec, vt_pos.vec);
const __m128 crossed = SSECrossProduct(right, down);
const __m128 normalize = SSENormalizeMultiplier(useSSE4, crossed);
Vec3f finalNrm = _mm_mul_ps(normalize, _mm_mul_ps(crossed, facing));
vertices[v * (patch_div_s + 1) + u].nrm = finalNrm;
#else
const Vec3Packedf &right = vr_pos - vl_pos;
const Vec3Packedf &down = vertices[b * (patch_div_s + 1) + u].pos - vertices[t * (patch_div_s + 1) + u].pos;
vertices[v * (patch_div_s + 1) + u].nrm = Cross(right, down).Normalized();
if (spatch.patchFacing) {
vertices[v * (patch_div_s + 1) + u].nrm *= -1.0f;
}
#endif
// Rotate for the next one to the right.
vl_pos = vc_pos;
vc_pos = vr_pos;
}
}
}
GEPatchPrimType prim_type = spatch.primType;
// Tesselate.
for (int tile_v = 0; tile_v < patch_div_t; ++tile_v) {
for (int tile_u = 0; tile_u < patch_div_s; ++tile_u) {
int idx0 = tile_v * (patch_div_s + 1) + tile_u;
int idx1 = tile_v * (patch_div_s + 1) + tile_u + 1;
int idx2 = (tile_v + 1) * (patch_div_s + 1) + tile_u;
int idx3 = (tile_v + 1) * (patch_div_s + 1) + tile_u + 1;
CopyQuadIndex(indices, prim_type, idx0, idx1, idx2, idx3);
count += 6;
}
}
}
static void _SplinePatchLowQuality(u8 *&dest, u16 *indices, int &count, const SplinePatchLocal &spatch, u32 origVertType) {
// Fast and easy way - just draw the control points, generate some very basic normal vector substitutes.
// Very inaccurate but okay for Loco Roco. Maybe should keep it as an option because it's fast.
const int tile_min_u = (spatch.type_u & START_OPEN) ? 0 : 1;
const int tile_min_v = (spatch.type_v & START_OPEN) ? 0 : 1;
const int tile_max_u = (spatch.type_u & END_OPEN) ? spatch.count_u - 1 : spatch.count_u - 2;
const int tile_max_v = (spatch.type_v & END_OPEN) ? spatch.count_v - 1 : spatch.count_v - 2;
float tu_width = (float)spatch.count_u - 3.0f;
float tv_height = (float)spatch.count_v - 3.0f;
tu_width /= (float)(tile_max_u - tile_min_u);
tv_height /= (float)(tile_max_v - tile_min_v);
GEPatchPrimType prim_type = spatch.primType;
bool computeNormals = spatch.computeNormals;
bool patchFacing = spatch.patchFacing;
int i = 0;
for (int tile_v = tile_min_v; tile_v < tile_max_v; ++tile_v) {
for (int tile_u = tile_min_u; tile_u < tile_max_u; ++tile_u) {
int point_index = tile_u + tile_v * spatch.count_u;
SimpleVertex v0 = *spatch.points[point_index];
SimpleVertex v1 = *spatch.points[point_index + 1];
SimpleVertex v2 = *spatch.points[point_index + spatch.count_u];
SimpleVertex v3 = *spatch.points[point_index + spatch.count_u + 1];
// Generate UV. TODO: Do this even if UV specified in control points?
if ((origVertType & GE_VTYPE_TC_MASK) == 0) {
float u = (tile_u - tile_min_u) * tu_width;
float v = (tile_v - tile_min_v) * tv_height;
v0.uv[0] = u;
v0.uv[1] = v;
v1.uv[0] = u + tu_width;
v1.uv[1] = v;
v2.uv[0] = u;
v2.uv[1] = v + tv_height;
v3.uv[0] = u + tu_width;
v3.uv[1] = v + tv_height;
}
// Generate normal if lighting is enabled (otherwise there's no point).
// This is a really poor quality algorithm, we get facet normals.
if (computeNormals) {
Vec3Packedf norm = Cross(v1.pos - v0.pos, v2.pos - v0.pos);
norm.Normalize();
if (patchFacing)
norm *= -1.0f;
v0.nrm = norm;
v1.nrm = norm;
v2.nrm = norm;
v3.nrm = norm;
}
int idx0 = i * 4 + 0;
int idx1 = i * 4 + 1;
int idx2 = i * 4 + 2;
int idx3 = i * 4 + 3;
i++;
CopyQuad(dest, &v0, &v1, &v2, &v3);
CopyQuadIndex(indices, prim_type, idx0, idx1, idx2, idx3);
count += 6;
}
}
}
static void _BezierPatchHighQuality(u8 *&dest, u16 *&indices, int &count, int tess_u, int tess_v, const BezierPatch &patch, u32 origVertType) {
const float third = 1.0f / 3.0f;
// First compute all the vertices and put them in an array
SimpleVertex *&vertices = (SimpleVertex*&)dest;
PrecomputedCurves<Vec3f> prepos(tess_u + 1);
PrecomputedCurves<Vec4f> precol(tess_u + 1);
PrecomputedCurves<Math3D::Vec2f> pretex(tess_u + 1);
PrecomputedCurves<Vec3f> prederivU(tess_u + 1);
const bool computeNormals = patch.computeNormals;
const bool sampleColors = (origVertType & GE_VTYPE_COL_MASK) != 0;
const bool sampleTexcoords = (origVertType & GE_VTYPE_TC_MASK) != 0;
// Precompute the horizontal curves to we only have to evaluate the vertical ones.
for (int i = 0; i < tess_u + 1; i++) {
float u = ((float)i / (float)tess_u);
prepos.horiz1[i] = Bernstein3D(patch.points[0]->pos, patch.points[1]->pos, patch.points[2]->pos, patch.points[3]->pos, u);
prepos.horiz2[i] = Bernstein3D(patch.points[4]->pos, patch.points[5]->pos, patch.points[6]->pos, patch.points[7]->pos, u);
prepos.horiz3[i] = Bernstein3D(patch.points[8]->pos, patch.points[9]->pos, patch.points[10]->pos, patch.points[11]->pos, u);
prepos.horiz4[i] = Bernstein3D(patch.points[12]->pos, patch.points[13]->pos, patch.points[14]->pos, patch.points[15]->pos, u);
if (sampleColors) {
precol.horiz1[i] = Bernstein3D(patch.points[0]->color_32, patch.points[1]->color_32, patch.points[2]->color_32, patch.points[3]->color_32, u);
precol.horiz2[i] = Bernstein3D(patch.points[4]->color_32, patch.points[5]->color_32, patch.points[6]->color_32, patch.points[7]->color_32, u);
precol.horiz3[i] = Bernstein3D(patch.points[8]->color_32, patch.points[9]->color_32, patch.points[10]->color_32, patch.points[11]->color_32, u);
precol.horiz4[i] = Bernstein3D(patch.points[12]->color_32, patch.points[13]->color_32, patch.points[14]->color_32, patch.points[15]->color_32, u);
}
if (sampleTexcoords) {
pretex.horiz1[i] = Bernstein3D(Math3D::Vec2f(patch.points[0]->uv), Math3D::Vec2f(patch.points[1]->uv), Math3D::Vec2f(patch.points[2]->uv), Math3D::Vec2f(patch.points[3]->uv), u);
pretex.horiz2[i] = Bernstein3D(Math3D::Vec2f(patch.points[4]->uv), Math3D::Vec2f(patch.points[5]->uv), Math3D::Vec2f(patch.points[6]->uv), Math3D::Vec2f(patch.points[7]->uv), u);
pretex.horiz3[i] = Bernstein3D(Math3D::Vec2f(patch.points[8]->uv), Math3D::Vec2f(patch.points[9]->uv), Math3D::Vec2f(patch.points[10]->uv), Math3D::Vec2f(patch.points[11]->uv), u);
pretex.horiz4[i] = Bernstein3D(Math3D::Vec2f(patch.points[12]->uv), Math3D::Vec2f(patch.points[13]->uv), Math3D::Vec2f(patch.points[14]->uv), Math3D::Vec2f(patch.points[15]->uv), u);
}
if (computeNormals) {
prederivU.horiz1[i] = Bernstein3DDerivative(patch.points[0]->pos, patch.points[1]->pos, patch.points[2]->pos, patch.points[3]->pos, u);
prederivU.horiz2[i] = Bernstein3DDerivative(patch.points[4]->pos, patch.points[5]->pos, patch.points[6]->pos, patch.points[7]->pos, u);
prederivU.horiz3[i] = Bernstein3DDerivative(patch.points[8]->pos, patch.points[9]->pos, patch.points[10]->pos, patch.points[11]->pos, u);
prederivU.horiz4[i] = Bernstein3DDerivative(patch.points[12]->pos, patch.points[13]->pos, patch.points[14]->pos, patch.points[15]->pos, u);
}
}
for (int tile_v = 0; tile_v < tess_v + 1; ++tile_v) {
for (int tile_u = 0; tile_u < tess_u + 1; ++tile_u) {
float u = ((float)tile_u / (float)tess_u);
float v = ((float)tile_v / (float)tess_v);
float bv = v;
SimpleVertex &vert = vertices[tile_v * (tess_u + 1) + tile_u];
if (computeNormals) {
const Vec3f derivU = prederivU.Bernstein3D(tile_u, bv);
const Vec3f derivV = prepos.Bernstein3DDerivative(tile_u, bv);
vert.nrm = Cross(derivU, derivV).Normalized();
if (patch.patchFacing)
vert.nrm *= -1.0f;
} else {
vert.nrm.SetZero();
}
vert.pos = prepos.Bernstein3D(tile_u, bv);
if (!sampleTexcoords) {
// Generate texcoord
vert.uv[0] = u + patch.u_index * third;
vert.uv[1] = v + patch.v_index * third;
} else {
// Sample UV from control points
const Math3D::Vec2f res = pretex.Bernstein3D(tile_u, bv);
vert.uv[0] = res.x;
vert.uv[1] = res.y;
}
if (sampleColors) {
vert.color_32 = precol.Bernstein3D(tile_u, bv).ToRGBA();
} else {
memcpy(vert.color, patch.points[0]->color, 4);
}
}
}
GEPatchPrimType prim_type = patch.primType;
// Combine the vertices into triangles.
for (int tile_v = 0; tile_v < tess_v; ++tile_v) {
for (int tile_u = 0; tile_u < tess_u; ++tile_u) {
int total = patch.index * (tess_u + 1) * (tess_v + 1);
int idx0 = total + tile_v * (tess_u + 1) + tile_u;
int idx1 = total + tile_v * (tess_u + 1) + tile_u + 1;
int idx2 = total + (tile_v + 1) * (tess_u + 1) + tile_u;
int idx3 = total + (tile_v + 1) * (tess_u + 1) + tile_u + 1;
CopyQuadIndex(indices, prim_type, idx0, idx1, idx2, idx3);
count += 6;
}
}
dest += (tess_u + 1) * (tess_v + 1) * sizeof(SimpleVertex);
}
