Cycles: Add new Sky Texture method including direct sunlight

This commit adds a new model to the Sky Texture node, which is based on a
method by Nishita et al. and works by basically simulating volumetric
scattering in the atmosphere.

By making some approximations (such as only considering single scattering),
we get a fairly simple and fast simulation code that takes into account
Rayleigh and Mie scattering as well as Ozone absorption.

This code is used to precompute a 512x128 texture which is then looked up
during render time, and is fast enough to allow real-time tweaking in the
viewport.

Due to the nature of the simulation, it exposes several parameters that
allow for lots of flexibility in choosing the look and matching real-world
conditions (such as Air/Dust/Ozone density and altitude).

Additionally, the same volumetric approach can be used to compute absorption
of the direct sunlight, so the model also supports adding direct sunlight.
This makes it significantly easier to set up Sun+Sky illumination where
the direction, intensity and color of the sun actually matches the sky.

In order to support properly sampling the direct sun component, the commit
also adds logic for sampling a specific area to the kernel light sampling
code. This is combined with portal and background map sampling using MIS.

This sampling logic works for the common case of having one Sky texture
going into the Background shader, but if a custom input to the Vector
node is used or if there are multiple Sky textures, it falls back to using
only background map sampling (while automatically setting the resolution to
4096x2048 if auto resolution is used).

More infos and preview can be found here:
https://docs.google.com/document/d/1gQta0ygFWXTrl5Pmvl_nZRgUw0mWg0FJeRuNKS36m08/view

Underlying model, implementation and documentation by Marco (@nacioss).
Improvements, cleanup and sun sampling by @lukasstockner.

Differential Revision: https://developer.blender.org/D7896

1 files changed, 4 insertions, 496 deletions

diff --git a/intern/cycles/kernel/kernel_light.h b/intern/cycles/kernel/kernel_light.h
index 04472212d0c..0448d0165b9 100644
--- a/intern/cycles/kernel/kernel_light.h
+++ b/intern/cycles/kernel/kernel_light.h
@@ -14,6 +14,8 @@
  * limitations under the License.
  */
 
+#include "kernel_light_background.h"
+
 CCL_NAMESPACE_BEGIN
 
 /* Light Sample result */
@@ -33,500 +35,6 @@ typedef struct LightSample {
   LightType type; /* type of light */
 } LightSample;
 
-/* Area light sampling */
-
-/* Uses the following paper:
- *
- * Carlos Urena et al.
- * An Area-Preserving Parametrization for Spherical Rectangles.
- *
- * https://www.solidangle.com/research/egsr2013_spherical_rectangle.pdf
- *
- * Note: light_p is modified when sample_coord is true.
- */
-ccl_device_inline float rect_light_sample(float3 P,
-                                          float3 *light_p,
-                                          float3 axisu,
-                                          float3 axisv,
-                                          float randu,
-                                          float randv,
-                                          bool sample_coord)
-{
-  /* In our name system we're using P for the center,
-   * which is o in the paper.
-   */
-
-  float3 corner = *light_p - axisu * 0.5f - axisv * 0.5f;
-  float axisu_len, axisv_len;
-  /* Compute local reference system R. */
-  float3 x = normalize_len(axisu, &axisu_len);
-  float3 y = normalize_len(axisv, &axisv_len);
-  float3 z = cross(x, y);
-  /* Compute rectangle coords in local reference system. */
-  float3 dir = corner - P;
-  float z0 = dot(dir, z);
-  /* Flip 'z' to make it point against Q. */
-  if (z0 > 0.0f) {
-    z *= -1.0f;
-    z0 *= -1.0f;
-  }
-  float x0 = dot(dir, x);
-  float y0 = dot(dir, y);
-  float x1 = x0 + axisu_len;
-  float y1 = y0 + axisv_len;
-  /* Compute internal angles (gamma_i). */
-  float4 diff = make_float4(x0, y1, x1, y0) - make_float4(x1, y0, x0, y1);
-  float4 nz = make_float4(y0, x1, y1, x0) * diff;
-  nz = nz / sqrt(z0 * z0 * diff * diff + nz * nz);
-  float g0 = safe_acosf(-nz.x * nz.y);
-  float g1 = safe_acosf(-nz.y * nz.z);
-  float g2 = safe_acosf(-nz.z * nz.w);
-  float g3 = safe_acosf(-nz.w * nz.x);
-  /* Compute predefined constants. */
-  float b0 = nz.x;
-  float b1 = nz.z;
-  float b0sq = b0 * b0;
-  float k = M_2PI_F - g2 - g3;
-  /* Compute solid angle from internal angles. */
-  float S = g0 + g1 - k;
-
-  if (sample_coord) {
-    /* Compute cu. */
-    float au = randu * S + k;
-    float fu = (cosf(au) * b0 - b1) / sinf(au);
-    float cu = 1.0f / sqrtf(fu * fu + b0sq) * (fu > 0.0f ? 1.0f : -1.0f);
-    cu = clamp(cu, -1.0f, 1.0f);
-    /* Compute xu. */
-    float xu = -(cu * z0) / max(sqrtf(1.0f - cu * cu), 1e-7f);
-    xu = clamp(xu, x0, x1);
-    /* Compute yv. */
-    float z0sq = z0 * z0;
-    float y0sq = y0 * y0;
-    float y1sq = y1 * y1;
-    float d = sqrtf(xu * xu + z0sq);
-    float h0 = y0 / sqrtf(d * d + y0sq);
-    float h1 = y1 / sqrtf(d * d + y1sq);
-    float hv = h0 + randv * (h1 - h0), hv2 = hv * hv;
-    float yv = (hv2 < 1.0f - 1e-6f) ? (hv * d) / sqrtf(1.0f - hv2) : y1;
-
-    /* Transform (xu, yv, z0) to world coords. */
-    *light_p = P + xu * x + yv * y + z0 * z;
-  }
-
-  /* return pdf */
-  if (S != 0.0f)
-    return 1.0f / S;
-  else
-    return 0.0f;
-}
-
-ccl_device_inline float3 ellipse_sample(float3 ru, float3 rv, float randu, float randv)
-{
-  to_unit_disk(&randu, &randv);
-  return ru * randu + rv * randv;
-}
-
-ccl_device float3 disk_light_sample(float3 v, float randu, float randv)
-{
-  float3 ru, rv;
-
-  make_orthonormals(v, &ru, &rv);
-
-  return ellipse_sample(ru, rv, randu, randv);
-}
-
-ccl_device float3 distant_light_sample(float3 D, float radius, float randu, float randv)
-{
-  return normalize(D + disk_light_sample(D, randu, randv) * radius);
-}
-
-ccl_device float3
-sphere_light_sample(float3 P, float3 center, float radius, float randu, float randv)
-{
-  return disk_light_sample(normalize(P - center), randu, randv) * radius;
-}
-
-ccl_device float spot_light_attenuation(float3 dir,
-                                        float spot_angle,
-                                        float spot_smooth,
-                                        LightSample *ls)
-{
-  float3 I = ls->Ng;
-
-  float attenuation = dot(dir, I);
-
-  if (attenuation <= spot_angle) {
-    attenuation = 0.0f;
-  }
-  else {
-    float t = attenuation - spot_angle;
-
-    if (t < spot_smooth && spot_smooth != 0.0f)
-      attenuation *= smoothstepf(t / spot_smooth);
-  }
-
-  return attenuation;
-}
-
-ccl_device float lamp_light_pdf(KernelGlobals *kg, const float3 Ng, const float3 I, float t)
-{
-  float cos_pi = dot(Ng, I);
-
-  if (cos_pi <= 0.0f)
-    return 0.0f;
-
-  return t * t / cos_pi;
-}
-
-/* Background Light */
-
-#ifdef __BACKGROUND_MIS__
-
-ccl_device float3 background_map_sample(KernelGlobals *kg, float randu, float randv, float *pdf)
-{
-  /* for the following, the CDF values are actually a pair of floats, with the
-   * function value as X and the actual CDF as Y.  The last entry's function
-   * value is the CDF total. */
-  int res_x = kernel_data.integrator.pdf_background_res_x;
-  int res_y = kernel_data.integrator.pdf_background_res_y;
-  int cdf_width = res_x + 1;
-
-  /* this is basically std::lower_bound as used by pbrt */
-  int first = 0;
-  int count = res_y;
-
-  while (count > 0) {
-    int step = count >> 1;
-    int middle = first + step;
-
-    if (kernel_tex_fetch(__light_background_marginal_cdf, middle).y < randv) {
-      first = middle + 1;
-      count -= step + 1;
-    }
-    else
-      count = step;
-  }
-
-  int index_v = max(0, first - 1);
-  kernel_assert(index_v >= 0 && index_v < res_y);
-
-  float2 cdf_v = kernel_tex_fetch(__light_background_marginal_cdf, index_v);
-  float2 cdf_next_v = kernel_tex_fetch(__light_background_marginal_cdf, index_v + 1);
-  float2 cdf_last_v = kernel_tex_fetch(__light_background_marginal_cdf, res_y);
-
-  /* importance-sampled V direction */
-  float dv = inverse_lerp(cdf_v.y, cdf_next_v.y, randv);
-  float v = (index_v + dv) / res_y;
-
-  /* this is basically std::lower_bound as used by pbrt */
-  first = 0;
-  count = res_x;
-  while (count > 0) {
-    int step = count >> 1;
-    int middle = first + step;
-
-    if (kernel_tex_fetch(__light_background_conditional_cdf, index_v * cdf_width + middle).y <
-        randu) {
-      first = middle + 1;
-      count -= step + 1;
-    }
-    else
-      count = step;
-  }
-
-  int index_u = max(0, first - 1);
-  kernel_assert(index_u >= 0 && index_u < res_x);
-
-  float2 cdf_u = kernel_tex_fetch(__light_background_conditional_cdf,
-                                  index_v * cdf_width + index_u);
-  float2 cdf_next_u = kernel_tex_fetch(__light_background_conditional_cdf,
-                                       index_v * cdf_width + index_u + 1);
-  float2 cdf_last_u = kernel_tex_fetch(__light_background_conditional_cdf,
-                                       index_v * cdf_width + res_x);
-
-  /* importance-sampled U direction */
-  float du = inverse_lerp(cdf_u.y, cdf_next_u.y, randu);
-  float u = (index_u + du) / res_x;
-
-  /* compute pdf */
-  float sin_theta = sinf(M_PI_F * v);
-  float denom = (M_2PI_F * M_PI_F * sin_theta) * cdf_last_u.x * cdf_last_v.x;
-
-  if (sin_theta == 0.0f || denom == 0.0f)
-    *pdf = 0.0f;
-  else
-    *pdf = (cdf_u.x * cdf_v.x) / denom;
-
-  /* compute direction */
-  return equirectangular_to_direction(u, v);
-}
-
-/* TODO(sergey): Same as above, after the release we should consider using
- * 'noinline' for all devices.
- */
-ccl_device float background_map_pdf(KernelGlobals *kg, float3 direction)
-{
-  float2 uv = direction_to_equirectangular(direction);
-  int res_x = kernel_data.integrator.pdf_background_res_x;
-  int res_y = kernel_data.integrator.pdf_background_res_y;
-  int cdf_width = res_x + 1;
-
-  float sin_theta = sinf(uv.y * M_PI_F);
-
-  if (sin_theta == 0.0f)
-    return 0.0f;
-
-  int index_u = clamp(float_to_int(uv.x * res_x), 0, res_x - 1);
-  int index_v = clamp(float_to_int(uv.y * res_y), 0, res_y - 1);
-
-  /* pdfs in V direction */
-  float2 cdf_last_u = kernel_tex_fetch(__light_background_conditional_cdf,
-                                       index_v * cdf_width + res_x);
-  float2 cdf_last_v = kernel_tex_fetch(__light_background_marginal_cdf, res_y);
-
-  float denom = (M_2PI_F * M_PI_F * sin_theta) * cdf_last_u.x * cdf_last_v.x;
-
-  if (denom == 0.0f)
-    return 0.0f;
-
-  /* pdfs in U direction */
-  float2 cdf_u = kernel_tex_fetch(__light_background_conditional_cdf,
-                                  index_v * cdf_width + index_u);
-  float2 cdf_v = kernel_tex_fetch(__light_background_marginal_cdf, index_v);
-
-  return (cdf_u.x * cdf_v.x) / denom;
-}
-
-ccl_device_inline bool background_portal_data_fetch_and_check_side(
-    KernelGlobals *kg, float3 P, int index, float3 *lightpos, float3 *dir)
-{
-  int portal = kernel_data.integrator.portal_offset + index;
-  const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, portal);
-
-  *lightpos = make_float3(klight->co[0], klight->co[1], klight->co[2]);
-  *dir = make_float3(klight->area.dir[0], klight->area.dir[1], klight->area.dir[2]);
-
-  /* Check whether portal is on the right side. */
-  if (dot(*dir, P - *lightpos) > 1e-4f)
-    return true;
-
-  return false;
-}
-
-ccl_device_inline float background_portal_pdf(
-    KernelGlobals *kg, float3 P, float3 direction, int ignore_portal, bool *is_possible)
-{
-  float portal_pdf = 0.0f;
-
-  int num_possible = 0;
-  for (int p = 0; p < kernel_data.integrator.num_portals; p++) {
-    if (p == ignore_portal)
-      continue;
-
-    float3 lightpos, dir;
-    if (!background_portal_data_fetch_and_check_side(kg, P, p, &lightpos, &dir))
-      continue;
-
-    /* There's a portal that could be sampled from this position. */
-    if (is_possible) {
-      *is_possible = true;
-    }
-    num_possible++;
-
-    int portal = kernel_data.integrator.portal_offset + p;
-    const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, portal);
-    float3 axisu = make_float3(
-        klight->area.axisu[0], klight->area.axisu[1], klight->area.axisu[2]);
-    float3 axisv = make_float3(
-        klight->area.axisv[0], klight->area.axisv[1], klight->area.axisv[2]);
-    bool is_round = (klight->area.invarea < 0.0f);
-
-    if (!ray_quad_intersect(P,
-                            direction,
-                            1e-4f,
-                            FLT_MAX,
-                            lightpos,
-                            axisu,
-                            axisv,
-                            dir,
-                            NULL,
-                            NULL,
-                            NULL,
-                            NULL,
-                            is_round))
-      continue;
-
-    if (is_round) {
-      float t;
-      float3 D = normalize_len(lightpos - P, &t);
-      portal_pdf += fabsf(klight->area.invarea) * lamp_light_pdf(kg, dir, -D, t);
-    }
-    else {
-      portal_pdf += rect_light_sample(P, &lightpos, axisu, axisv, 0.0f, 0.0f, false);
-    }
-  }
-
-  if (ignore_portal >= 0) {
-    /* We have skipped a portal that could be sampled as well. */
-    num_possible++;
-  }
-
-  return (num_possible > 0) ? portal_pdf / num_possible : 0.0f;
-}
-
-ccl_device int background_num_possible_portals(KernelGlobals *kg, float3 P)
-{
-  int num_possible_portals = 0;
-  for (int p = 0; p < kernel_data.integrator.num_portals; p++) {
-    float3 lightpos, dir;
-    if (background_portal_data_fetch_and_check_side(kg, P, p, &lightpos, &dir))
-      num_possible_portals++;
-  }
-  return num_possible_portals;
-}
-
-ccl_device float3 background_portal_sample(KernelGlobals *kg,
-                                           float3 P,
-                                           float randu,
-                                           float randv,
-                                           int num_possible,
-                                           int *sampled_portal,
-                                           float *pdf)
-{
-  /* Pick a portal, then re-normalize randv. */
-  randv *= num_possible;
-  int portal = (int)randv;
-  randv -= portal;
-
-  /* TODO(sergey): Some smarter way of finding portal to sample
-   * is welcome.
-   */
-  for (int p = 0; p < kernel_data.integrator.num_portals; p++) {
-    /* Search for the sampled portal. */
-    float3 lightpos, dir;
-    if (!background_portal_data_fetch_and_check_side(kg, P, p, &lightpos, &dir))
-      continue;
-
-    if (portal == 0) {
-      /* p is the portal to be sampled. */
-      int portal = kernel_data.integrator.portal_offset + p;
-      const ccl_global KernelLight *klight = &kernel_tex_fetch(__lights, portal);
-      float3 axisu = make_float3(
-          klight->area.axisu[0], klight->area.axisu[1], klight->area.axisu[2]);
-      float3 axisv = make_float3(
-          klight->area.axisv[0], klight->area.axisv[1], klight->area.axisv[2]);
-      bool is_round = (klight->area.invarea < 0.0f);
-
-      float3 D;
-      if (is_round) {
-        lightpos += ellipse_sample(axisu * 0.5f, axisv * 0.5f, randu, randv);
-        float t;
-        D = normalize_len(lightpos - P, &t);
-        *pdf = fabsf(klight->area.invarea) * lamp_light_pdf(kg, dir, -D, t);
-      }
-      else {
-        *pdf = rect_light_sample(P, &lightpos, axisu, axisv, randu, randv, true);
-        D = normalize(lightpos - P);
-      }
-
-      *pdf /= num_possible;
-      *sampled_portal = p;
-      return D;
-    }
-
-    portal--;
-  }
-
-  return make_float3(0.0f, 0.0f, 0.0f);
-}
-
-ccl_device_inline float3
-background_light_sample(KernelGlobals *kg, float3 P, float randu, float randv, float *pdf)
-{
-  /* Probability of sampling portals instead of the map. */
-  float portal_sampling_pdf = kernel_data.integrator.portal_pdf;
-
-  /* Check if there are portals in the scene which we can sample. */
-  if (portal_sampling_pdf > 0.0f) {
-    int num_portals = background_num_possible_portals(kg, P);
-    if (num_portals > 0) {
-      if (portal_sampling_pdf == 1.0f || randu < portal_sampling_pdf) {
-        if (portal_sampling_pdf < 1.0f) {
-          randu /= portal_sampling_pdf;
-        }
-        int portal;
-        float3 D = background_portal_sample(kg, P, randu, randv, num_portals, &portal, pdf);
-        if (num_portals > 1) {
-          /* Ignore the chosen portal, its pdf is already included. */
-          *pdf += background_portal_pdf(kg, P, D, portal, NULL);
-        }
-        /* We could also have sampled the map, so combine with MIS. */
-        if (portal_sampling_pdf < 1.0f) {
-          float cdf_pdf = background_map_pdf(kg, D);
-          *pdf = (portal_sampling_pdf * (*pdf) + (1.0f - portal_sampling_pdf) * cdf_pdf);
-        }
-        return D;
-      }
-      else {
-        /* Sample map, but with nonzero portal_sampling_pdf for MIS. */
-        randu = (randu - portal_sampling_pdf) / (1.0f - portal_sampling_pdf);
-      }
-    }
-    else {
-      /* We can't sample a portal.
-       * Check if we can sample the map instead.
-       */
-      if (portal_sampling_pdf == 1.0f) {
-        /* Use uniform as a fallback if we can't sample the map. */
-        *pdf = 1.0f / M_4PI_F;
-        return sample_uniform_sphere(randu, randv);
-      }
-      else {
-        portal_sampling_pdf = 0.0f;
-      }
-    }
-  }
-
-  float3 D = background_map_sample(kg, randu, randv, pdf);
-  /* Use MIS if portals could be sampled as well. */
-  if (portal_sampling_pdf > 0.0f) {
-    float portal_pdf = background_portal_pdf(kg, P, D, -1, NULL);
-    *pdf = (portal_sampling_pdf * portal_pdf + (1.0f - portal_sampling_pdf) * (*pdf));
-  }
-  return D;
-}
-
-ccl_device float background_light_pdf(KernelGlobals *kg, float3 P, float3 direction)
-{
-  /* Probability of sampling portals instead of the map. */
-  float portal_sampling_pdf = kernel_data.integrator.portal_pdf;
-
-  float portal_pdf = 0.0f, map_pdf = 0.0f;
-  if (portal_sampling_pdf > 0.0f) {
-    /* Evaluate PDF of sampling this direction by portal sampling. */
-    bool is_possible = false;
-    portal_pdf = background_portal_pdf(kg, P, direction, -1, &is_possible) * portal_sampling_pdf;
-    if (!is_possible) {
-      /* Portal sampling is not possible here because all portals point to the wrong side.
-       * If map sampling is possible, it would be used instead,
-       * otherwise fallback sampling is used. */
-      if (portal_sampling_pdf == 1.0f) {
-        return kernel_data.integrator.pdf_lights / M_4PI_F;
-      }
-      else {
-        /* Force map sampling. */
-        portal_sampling_pdf = 0.0f;
-      }
-    }
-  }
-  if (portal_sampling_pdf < 1.0f) {
-    /* Evaluate PDF of sampling this direction by map sampling. */
-    map_pdf = background_map_pdf(kg, direction) * (1.0f - portal_sampling_pdf);
-  }
-  return (portal_pdf + map_pdf) * kernel_data.integrator.pdf_lights;
-}
-#endif
-
 /* Regular Light */
 
 ccl_device_inline bool lamp_light_sample(
@@ -594,7 +102,7 @@ ccl_device_inline bool lamp_light_sample(
         /* spot light attenuation */
         float3 dir = make_float3(klight->spot.dir[0], klight->spot.dir[1], klight->spot.dir[2]);
         ls->eval_fac *= spot_light_attenuation(
-            dir, klight->spot.spot_angle, klight->spot.spot_smooth, ls);
+            dir, klight->spot.spot_angle, klight->spot.spot_smooth, ls->Ng);
         if (ls->eval_fac == 0.0f) {
           return false;
         }
@@ -732,7 +240,7 @@ ccl_device bool lamp_light_eval(
       /* spot light attenuation */
       float3 dir = make_float3(klight->spot.dir[0], klight->spot.dir[1], klight->spot.dir[2]);
       ls->eval_fac *= spot_light_attenuation(
-          dir, klight->spot.spot_angle, klight->spot.spot_smooth, ls);
+          dir, klight->spot.spot_angle, klight->spot.spot_smooth, ls->Ng);
 
       if (ls->eval_fac == 0.0f)
         return false;