De-duplicate EWA filter code between renderer and compositor
[blender-staging.git] / source / blender / compositor / intern / COM_MemoryBuffer.cpp
index c1916f4a68f88b615f65bb4a8d1ef31ed4e87c5c..c59ecced93c9e2c9233cb9f4473c507580d511f4 100644 (file)
@@ -178,162 +178,58 @@ void MemoryBuffer::addPixel(int x, int y, const float color[4])
        }
 }
 
-
-// table of (exp(ar) - exp(a)) / (1 - exp(a)) for r in range [0, 1] and a = -2
-// used instead of actual gaussian, otherwise at high texture magnifications circular artifacts are visible
-#define EWA_MAXIDX 255
-static const float EWA_WTS[EWA_MAXIDX + 1] = {
-       1.f, 0.990965f, 0.982f, 0.973105f, 0.96428f, 0.955524f, 0.946836f, 0.938216f, 0.929664f,
-       0.921178f, 0.912759f, 0.904405f, 0.896117f, 0.887893f, 0.879734f, 0.871638f, 0.863605f,
-       0.855636f, 0.847728f, 0.839883f, 0.832098f, 0.824375f, 0.816712f, 0.809108f, 0.801564f,
-       0.794079f, 0.786653f, 0.779284f, 0.771974f, 0.76472f, 0.757523f, 0.750382f, 0.743297f,
-       0.736267f, 0.729292f, 0.722372f, 0.715505f, 0.708693f, 0.701933f, 0.695227f, 0.688572f,
-       0.68197f, 0.67542f, 0.66892f, 0.662471f, 0.656073f, 0.649725f, 0.643426f, 0.637176f,
-       0.630976f, 0.624824f, 0.618719f, 0.612663f, 0.606654f, 0.600691f, 0.594776f, 0.588906f,
-       0.583083f, 0.577305f, 0.571572f, 0.565883f, 0.56024f, 0.55464f, 0.549084f, 0.543572f,
-       0.538102f, 0.532676f, 0.527291f, 0.521949f, 0.516649f, 0.511389f, 0.506171f, 0.500994f,
-       0.495857f, 0.490761f, 0.485704f, 0.480687f, 0.475709f, 0.470769f, 0.465869f, 0.461006f,
-       0.456182f, 0.451395f, 0.446646f, 0.441934f, 0.437258f, 0.432619f, 0.428017f, 0.42345f,
-       0.418919f, 0.414424f, 0.409963f, 0.405538f, 0.401147f, 0.39679f, 0.392467f, 0.388178f,
-       0.383923f, 0.379701f, 0.375511f, 0.371355f, 0.367231f, 0.363139f, 0.359079f, 0.355051f,
-       0.351055f, 0.347089f, 0.343155f, 0.339251f, 0.335378f, 0.331535f, 0.327722f, 0.323939f,
-       0.320186f, 0.316461f, 0.312766f, 0.3091f, 0.305462f, 0.301853f, 0.298272f, 0.294719f,
-       0.291194f, 0.287696f, 0.284226f, 0.280782f, 0.277366f, 0.273976f, 0.270613f, 0.267276f,
-       0.263965f, 0.26068f, 0.257421f, 0.254187f, 0.250979f, 0.247795f, 0.244636f, 0.241502f,
-       0.238393f, 0.235308f, 0.232246f, 0.229209f, 0.226196f, 0.223206f, 0.220239f, 0.217296f,
-       0.214375f, 0.211478f, 0.208603f, 0.20575f, 0.20292f, 0.200112f, 0.197326f, 0.194562f,
-       0.191819f, 0.189097f, 0.186397f, 0.183718f, 0.18106f, 0.178423f, 0.175806f, 0.17321f,
-       0.170634f, 0.168078f, 0.165542f, 0.163026f, 0.16053f, 0.158053f, 0.155595f, 0.153157f,
-       0.150738f, 0.148337f, 0.145955f, 0.143592f, 0.141248f, 0.138921f, 0.136613f, 0.134323f,
-       0.132051f, 0.129797f, 0.12756f, 0.125341f, 0.123139f, 0.120954f, 0.118786f, 0.116635f,
-       0.114501f, 0.112384f, 0.110283f, 0.108199f, 0.106131f, 0.104079f, 0.102043f, 0.100023f,
-       0.0980186f, 0.09603f, 0.094057f, 0.0920994f, 0.0901571f, 0.08823f, 0.0863179f, 0.0844208f,
-       0.0825384f, 0.0806708f, 0.0788178f, 0.0769792f, 0.0751551f, 0.0733451f, 0.0715493f, 0.0697676f,
-       0.0679997f, 0.0662457f, 0.0645054f, 0.0627786f, 0.0610654f, 0.0593655f, 0.0576789f, 0.0560055f,
-       0.0543452f, 0.0526979f, 0.0510634f, 0.0494416f, 0.0478326f, 0.0462361f, 0.0446521f, 0.0430805f,
-       0.0415211f, 0.039974f, 0.0384389f, 0.0369158f, 0.0354046f, 0.0339052f, 0.0324175f, 0.0309415f,
-       0.029477f, 0.0280239f, 0.0265822f, 0.0251517f, 0.0237324f, 0.0223242f, 0.020927f, 0.0195408f,
-       0.0181653f, 0.0168006f, 0.0154466f, 0.0141031f, 0.0127701f, 0.0114476f, 0.0101354f, 0.00883339f,
-       0.00754159f, 0.00625989f, 0.00498819f, 0.00372644f, 0.00247454f, 0.00123242f, 0.f
-};
-
-static void ellipse_bounds(float A, float B, float C, float F, float &xmax, float &ymax)
-{
-       float denom = 4.0f * A * C - B * B;
-       if (denom > 0.0f && A != 0.0f && C != 0.0f) {
-               xmax = sqrtf(F) / (2.0f * A) * (sqrtf(F * (4.0f * A - B * B / C)) + B * B * sqrtf(F / (C * denom)));
-               ymax = sqrtf(F) / (2.0f * C) * (sqrtf(F * (4.0f * C - B * B / A)) + B * B * sqrtf(F / (A * denom)));
+typedef struct ReadEWAData {
+       MemoryBuffer *buffer;
+       PixelSampler sampler;
+       float ufac, vfac;
+} ReadEWAData;
+
+static void read_ewa_pixel_sampled(void *userdata, int x, int y, float result[4])
+{
+       ReadEWAData *data = (ReadEWAData *) userdata;
+       switch (data->sampler) {
+               case COM_PS_NEAREST:
+                       data->buffer->read(result, x, y);
+                       break;
+               case COM_PS_BILINEAR:
+                       data->buffer->readBilinear(result,
+                                                  (float)x + data->ufac,
+                                                  (float)y + data->vfac);
+                       break;
+               case COM_PS_BICUBIC:
+                       /* TOOD(sergey): no readBicubic method yet */
+                       data->buffer->readBilinear(result,
+                                                  (float)x + data->ufac,
+                                                  (float)y + data->vfac);
+                       break;
+               default:
+                       zero_v4(result);
+                       break;
        }
-       else {
-               xmax = 0.0f;
-               ymax = 0.0f;
-       }
-}
-
-static void ellipse_params(float Ux, float Uy, float Vx, float Vy,
-                           float &A, float &B, float &C, float &F, float &umax, float &vmax)
-{
-       A = Vx * Vx + Vy * Vy;
-       B = -2.0f * (Ux * Vx + Uy * Vy);
-       C = Ux * Ux + Uy * Uy;
-       F = A * C - B * B * 0.25f;
-
-       float factor = (F != 0.0f ? (float)(EWA_MAXIDX + 1) / F : 0.0f);
-       A *= factor;
-       B *= factor;
-       C *= factor;
-       F = (float)(EWA_MAXIDX + 1);
-
-       ellipse_bounds(A, B, C, sqrtf(F), umax, vmax);
 }
 
-/**
- * Filtering method based on
- * "Creating raster omnimax images from multiple perspective views using the elliptical weighted average filter"
- * by Ned Greene and Paul S. Heckbert (1986)
- */
 void MemoryBuffer::readEWA(float result[4], const float uv[2], const float derivatives[2][2], PixelSampler sampler)
 {
-       zero_v4(result);
-       int width = this->getWidth(), height = this->getHeight();
-       if (width == 0 || height == 0)
-               return;
-
-       float u = uv[0], v = uv[1];
-       float Ux = derivatives[0][0], Vx = derivatives[1][0], Uy = derivatives[0][1], Vy = derivatives[1][1];
-       float A, B, C, F, ue, ve;
-       ellipse_params(Ux, Uy, Vx, Vy, A, B, C, F, ue, ve);
+       ReadEWAData data;
+       data.buffer = this;
+       data.sampler = sampler;
+       data.ufac = uv[0] - floorf(uv[0]);
+       data.vfac = uv[1] - floorf(uv[1]);
 
-       /* Note: highly eccentric ellipses can lead to large texture space areas to filter!
-        * This is limited somewhat by the EWA_WTS size in the loop, but a nicer approach
-        * could be the one found in
-        * "High Quality Elliptical Texture Filtering on GPU"
-        * by Pavlos Mavridis and Georgios Papaioannou
-        * in which the eccentricity of the ellipse is clamped.
-        */
-
-       int U0 = (int)u;
-       int V0 = (int)v;
-       /* pixel offset for interpolation */
-       float ufac = u - floorf(u), vfac = v - floorf(v);
-       /* filter size */
-       int u1 = (int)(u - ue);
-       int u2 = (int)(u + ue);
-       int v1 = (int)(v - ve);
-       int v2 = (int)(v + ve);
-
-       /* sane clamping to avoid unnecessarily huge loops */
-       /* note: if eccentricity gets clamped (see above),
-        * the ue/ve limits can also be lowered accordingly
+       int width = this->getWidth(), height = this->getHeight();
+       /* TODO(sergey): Render pipeline uses normalized coordinates and derivatives,
+        * but compositor uses pixel space. For now let's just divide the values and
+        * switch compositor to normalized space for EWA later.
         */
-       if (U0 - u1 > EWA_MAXIDX) u1 = U0 - EWA_MAXIDX;
-       if (u2 - U0 > EWA_MAXIDX) u2 = U0 + EWA_MAXIDX;
-       if (V0 - v1 > EWA_MAXIDX) v1 = V0 - EWA_MAXIDX;
-       if (v2 - V0 > EWA_MAXIDX) v2 = V0 + EWA_MAXIDX;
-
-       /* Early output check for cases the whole region is outside of the buffer. */
-       if ((u2 < m_rect.xmin || u1 >= m_rect.xmax) ||
-           (v2 < m_rect.ymin || v1 >= m_rect.ymax))
-       {
-               zero_v4(result);
-               return;
-       }
-
-       /* Clamp sampling rectagle to the buffer dimensions. */
-       u1 = max_ii(u1, m_rect.xmin);
-       u2 = min_ii(u2, m_rect.xmax);
-       v1 = max_ii(v1, m_rect.ymin);
-       v2 = min_ii(v2, m_rect.ymax);
-
-       float DDQ = 2.0f * A;
-       float U = u1 - U0;
-       float ac1 = A * (2.0f * U + 1.0f);
-       float ac2 = A * U * U;
-       float BU = B * U;
-
-       float sum = 0.0f;
-       for (int v = v1; v <= v2; ++v) {
-               float V = v - V0;
-
-               float DQ = ac1 + B * V;
-               float Q = (C * V + BU) * V + ac2;
-               for (int u = u1; u <= u2; ++u) {
-                       if (Q < F) {
-                               float tc[4];
-                               const float wt = EWA_WTS[CLAMPIS((int)Q, 0, EWA_MAXIDX)];
-                               switch (sampler) {
-                                       case COM_PS_NEAREST: read(tc, u, v); break;
-                                       case COM_PS_BILINEAR: readBilinear(tc, (float)u + ufac, (float)v + vfac); break;
-                                       case COM_PS_BICUBIC: readBilinear(tc, (float)u + ufac, (float)v + vfac); break; /* XXX no readBicubic method yet */
-                                       default: zero_v4(tc); break;
-                               }
-                               madd_v4_v4fl(result, tc, wt);
-                               sum += wt;
-                       }
-                       Q += DQ;
-                       DQ += DDQ;
-               }
-       }
-       
-       mul_v4_fl(result, (sum != 0.0f ? 1.0f / sum : 0.0f));
+       float uv_normal[2] = {uv[0] / width, uv[1] / height};
+       float du_normal[2] = {derivatives[0][0] / width, derivatives[0][1] / height};
+       float dv_normal[2] = {derivatives[1][0] / width, derivatives[1][1] / height};
+
+       BLI_ewa_filter(this->getWidth(), this->getHeight(),
+                      false,
+                      true,
+                      uv_normal, du_normal, dv_normal,
+                      read_ewa_pixel_sampled,
+                      &data,
+                      result);
 }