Merge branch 'blender-v3.1-release'

3 files changed, 54 insertions, 27 deletions

diff --git a/intern/cycles/kernel/bvh/util.h b/intern/cycles/kernel/bvh/util.h
index 1fd3a3f2850..71045157372 100644
--- a/intern/cycles/kernel/bvh/util.h
+++ b/intern/cycles/kernel/bvh/util.h
@@ -5,27 +5,6 @@
 
 CCL_NAMESPACE_BEGIN
 
-/* Ray offset to avoid self intersection.
- *
- * This function should be used to compute a modified ray start position for
- * rays leaving from a surface. This is from "A Fast and Robust Method for Avoiding
- * Self-Intersection" see https://research.nvidia.com/publication/2019-03_A-Fast-and
- */
-ccl_device_inline float3 ray_offset(float3 P, float3 Ng)
-{
-  const float int_scale = 256.0f;
-  int3 of_i = make_int3((int)(int_scale * Ng.x), (int)(int_scale * Ng.y), (int)(int_scale * Ng.z));
-
-  float3 p_i = make_float3(__int_as_float(__float_as_int(P.x) + ((P.x < 0) ? -of_i.x : of_i.x)),
-                           __int_as_float(__float_as_int(P.y) + ((P.y < 0) ? -of_i.y : of_i.y)),
-                           __int_as_float(__float_as_int(P.z) + ((P.z < 0) ? -of_i.z : of_i.z)));
-  const float origin = 1.0f / 32.0f;
-  const float float_scale = 1.0f / 65536.0f;
-  return make_float3(fabsf(P.x) < origin ? P.x + float_scale * Ng.x : p_i.x,
-                     fabsf(P.y) < origin ? P.y + float_scale * Ng.y : p_i.y,
-                     fabsf(P.z) < origin ? P.z + float_scale * Ng.z : p_i.z);
-}
-
 #if defined(__KERNEL_CPU__)
 ccl_device int intersections_compare(const void *a, const void *b)
 {
diff --git a/intern/cycles/kernel/integrator/init_from_bake.h b/intern/cycles/kernel/integrator/init_from_bake.h
index e616123e9e7..b84059d6676 100644
--- a/intern/cycles/kernel/integrator/init_from_bake.h
+++ b/intern/cycles/kernel/integrator/init_from_bake.h
@@ -30,6 +30,50 @@ ccl_device_inline float bake_clamp_mirror_repeat(float u, float max)
   return ((((int)fu) & 1) ? 1.0f - u : u) * max;
 }
 
+/* Offset towards center of triangle to avoid ray-tracing precision issues. */
+ccl_device const float2 bake_offset_towards_center(KernelGlobals kg,
+                                                   const int prim,
+                                                   const float u,
+                                                   const float v)
+{
+  float3 tri_verts[3];
+  triangle_vertices(kg, prim, tri_verts);
+
+  /* Empirically determined values, by no means perfect. */
+  const float position_offset = 1e-4f;
+  const float uv_offset = 1e-5f;
+
+  /* Offset position towards center, amount relative to absolute size of position coordinates. */
+  const float3 P = u * tri_verts[0] + v * tri_verts[1] + (1.0f - u - v) * tri_verts[2];
+  const float3 center = (tri_verts[0] + tri_verts[1] + tri_verts[2]) / 3.0f;
+  const float3 to_center = center - P;
+
+  const float3 offset_P = P + normalize(to_center) *
+                                  min(len(to_center), max(max3(fabs(P)), 1.0f) * position_offset);
+
+  /* Compute barycentric coordinates at new position. */
+  const float3 v1 = tri_verts[1] - tri_verts[0];
+  const float3 v2 = tri_verts[2] - tri_verts[0];
+  const float3 vP = offset_P - tri_verts[0];
+
+  const float d11 = dot(v1, v1);
+  const float d12 = dot(v1, v2);
+  const float d22 = dot(v2, v2);
+  const float dP1 = dot(vP, v1);
+  const float dP2 = dot(vP, v2);
+
+  const float denom = d11 * d22 - d12 * d12;
+  if (denom == 0.0f) {
+    return make_float2(0.0f, 0.0f);
+  }
+
+  const float offset_v = clamp((d22 * dP1 - d12 * dP2) / denom, uv_offset, 1.0f - uv_offset);
+  const float offset_w = clamp((d11 * dP2 - d12 * dP1) / denom, uv_offset, 1.0f - uv_offset);
+  const float offset_u = clamp(1.0f - offset_v - offset_w, uv_offset, 1.0f - uv_offset);
+
+  return make_float2(offset_u, offset_v);
+}
+
 /* Return false to indicate that this pixel is finished.
  * Used by CPU implementation to not attempt to sample pixel for multiple samples once its known
  * that the pixel did converge. */
@@ -87,7 +131,7 @@ ccl_device bool integrator_init_from_bake(KernelGlobals kg,
   /* Initialize path state for path integration. */
   path_state_init_integrator(kg, state, sample, rng_hash);
 
-  /* Barycentric UV with sub-pixel offset. */
+  /* Barycentric UV. */
   float u = primitive[2];
   float v = primitive[3];
 
@@ -96,6 +140,14 @@ ccl_device bool integrator_init_from_bake(KernelGlobals kg,
   float dvdx = differential[2];
   float dvdy = differential[3];
 
+  /* Exactly at vertex? Nudge inwards to avoid self-intersection. */
+  if ((u == 0.0f || u == 1.0f) && (v == 0.0f || v == 1.0f)) {
+    const float2 uv = bake_offset_towards_center(kg, prim, u, v);
+    u = uv.x;
+    v = uv.y;
+  }
+
+  /* Sub-pixel offset. */
   if (sample > 0) {
     u = bake_clamp_mirror_repeat(u + dudx * (filter_x - 0.5f) + dudy * (filter_y - 0.5f), 1.0f);
     v = bake_clamp_mirror_repeat(v + dvdx * (filter_x - 0.5f) + dvdy * (filter_y - 0.5f),
diff --git a/intern/cycles/kernel/integrator/shade_surface.h b/intern/cycles/kernel/integrator/shade_surface.h
index f548d91031d..d2442755646 100644
--- a/intern/cycles/kernel/integrator/shade_surface.h
+++ b/intern/cycles/kernel/integrator/shade_surface.h
@@ -352,12 +352,8 @@ ccl_device_forceinline void integrate_surface_ao(KernelGlobals kg,
   float ao_pdf;
   sample_cos_hemisphere(ao_N, bsdf_u, bsdf_v, &ao_D, &ao_pdf);
 
-  if (!(dot(sd->Ng, ao_D) > 0.0f && ao_pdf != 0.0f)) {
-    return;
-  }
-
   Ray ray ccl_optional_struct_init;
-  ray.P = sd->P;
+  ray.P = shadow_ray_offset(kg, sd, ao_D);
   ray.D = ao_D;
   ray.t = kernel_data.integrator.ao_bounces_distance;
   ray.time = sd->time;