doc/python_api/sphinx_doc_gen.py

@@ -1949,7 +1949,7 @@ if html_theme == "furo":
             "sidebar/scroll-start.html",
             "sidebar/navigation.html",
             "sidebar/scroll-end.html",
-            # "sidebar/variant-selector.html",
+            "sidebar/variant-selector.html",
         ]
     }
 """)

extern/CMakeLists.txt

@@ -104,10 +104,6 @@ if(WITH_MOD_FLUID)
   add_subdirectory(mantaflow)
 endif()
 
-if(WITH_COMPOSITOR_CPU)
-  add_subdirectory(smaa_areatex)
-endif()
-
 if(WITH_VULKAN_BACKEND)
   add_subdirectory(vulkan_memory_allocator)
 endif()

extern/smaa_areatex/CMakeLists.txt

@@ -1,5 +0,0 @@
-# SPDX-FileCopyrightText: 2017 Blender Foundation
-#
-# SPDX-License-Identifier: GPL-2.0-or-later
-
-add_executable(smaa_areatex smaa_areatex.cpp)

extern/smaa_areatex/README.blender

@@ -1,5 +0,0 @@
-Project: smaa-cpp
-URL: https://github.com/iRi-E/smaa-cpp
-License: MIT
-Upstream version: 0.4.0
-Local modifications:

intern/cycles/device/metal/kernel.mm

@@ -106,7 +106,7 @@ struct ShaderCache {
 
   friend ShaderCache *get_shader_cache(id<MTLDevice> mtlDevice);
 
-  void compile_thread_func(int thread_index);
+  void compile_thread_func();
 
   using PipelineCollection = std::vector<unique_ptr<MetalKernelPipeline>>;
 
@@ -174,7 +174,7 @@ void ShaderCache::wait_for_all()
   }
 }
 
-void ShaderCache::compile_thread_func(int /*thread_index*/)
+void ShaderCache::compile_thread_func()
 {
   while (running) {
 
@@ -309,7 +309,7 @@ void ShaderCache::load_kernel(DeviceKernel device_kernel,
 
       metal_printf("Spawning %d Cycles kernel compilation threads\n", max_mtlcompiler_threads);
       for (int i = 0; i < max_mtlcompiler_threads; i++) {
-        compile_threads.push_back(std::thread([&] { compile_thread_func(i); }));
+        compile_threads.push_back(std::thread([this] { this->compile_thread_func(); }));
       }
     }
   }

intern/cycles/kernel/integrator/shade_volume.h

@@ -64,6 +64,11 @@ typedef struct VolumeShaderCoefficients {
   Spectrum emission;
 } VolumeShaderCoefficients;
 
+typedef struct EquiangularCoefficients {
+  float3 P;
+  float2 t_range;
+} EquiangularCoefficients;
+
 /* Evaluate shader to get extinction coefficient at P. */
 ccl_device_inline bool shadow_volume_shader_sample(KernelGlobals kg,
                                                    IntegratorShadowState state,
@@ -264,18 +269,18 @@ ccl_device void volume_shadow_heterogeneous(KernelGlobals kg,
 #  define VOLUME_SAMPLE_PDF_CUTOFF 1e-8f
 
 ccl_device float volume_equiangular_sample(ccl_private const Ray *ccl_restrict ray,
-                                           const float3 light_P,
+                                           ccl_private const EquiangularCoefficients &coeffs,
                                            const float xi,
                                            ccl_private float *pdf)
 {
-  const float tmin = ray->tmin;
-  const float tmax = ray->tmax;
-  const float delta = dot((light_P - ray->P), ray->D);
-  const float D = safe_sqrtf(len_squared(light_P - ray->P) - delta * delta);
+  const float delta = dot((coeffs.P - ray->P), ray->D);
+  const float D = safe_sqrtf(len_squared(coeffs.P - ray->P) - delta * delta);
   if (UNLIKELY(D == 0.0f)) {
     *pdf = 0.0f;
     return 0.0f;
   }
+  const float tmin = coeffs.t_range.x;
+  const float tmax = coeffs.t_range.y;
   const float theta_a = atan2f(tmin - delta, D);
   const float theta_b = atan2f(tmax - delta, D);
   const float t_ = D * tanf((xi * theta_b) + (1 - xi) * theta_a);
@@ -289,17 +294,17 @@ ccl_device float volume_equiangular_sample(ccl_private const Ray *ccl_restrict r
 }
 
 ccl_device float volume_equiangular_pdf(ccl_private const Ray *ccl_restrict ray,
-                                        const float3 light_P,
+                                        ccl_private const EquiangularCoefficients &coeffs,
                                         const float sample_t)
 {
-  const float delta = dot((light_P - ray->P), ray->D);
-  const float D = safe_sqrtf(len_squared(light_P - ray->P) - delta * delta);
+  const float delta = dot((coeffs.P - ray->P), ray->D);
+  const float D = safe_sqrtf(len_squared(coeffs.P - ray->P) - delta * delta);
   if (UNLIKELY(D == 0.0f)) {
     return 0.0f;
   }
 
-  const float tmin = ray->tmin;
-  const float tmax = ray->tmax;
+  const float tmin = coeffs.t_range.x;
+  const float tmax = coeffs.t_range.y;
   const float t_ = sample_t - delta;
 
   const float theta_a = atan2f(tmin - delta, D);
@@ -313,6 +318,29 @@ ccl_device float volume_equiangular_pdf(ccl_private const Ray *ccl_restrict ray,
   return pdf;
 }
 
+ccl_device_inline bool volume_equiangular_valid_ray_segment(KernelGlobals kg,
+                                                            const float3 ray_P,
+                                                            const float3 ray_D,
+                                                            ccl_private float2 *t_range,
+                                                            const ccl_private LightSample *ls)
+{
+  if (ls->type == LIGHT_SPOT) {
+    ccl_global const KernelLight *klight = &kernel_data_fetch(lights, ls->lamp);
+    return spot_light_valid_ray_segment(klight, ray_P, ray_D, t_range);
+  }
+  if (ls->type == LIGHT_AREA) {
+    ccl_global const KernelLight *klight = &kernel_data_fetch(lights, ls->lamp);
+    return area_light_valid_ray_segment(&klight->area, ray_P - klight->co, ray_D, t_range);
+  }
+  if (ls->type == LIGHT_TRIANGLE) {
+    return triangle_light_valid_ray_segment(kg, ray_P - ls->P, ray_D, t_range, ls);
+  }
+
+  /* Point light, the whole range of the ray is visible. */
+  kernel_assert(ls->type == LIGHT_POINT);
+  return true;
+}
+
 /* Distance sampling */
 
 ccl_device float volume_distance_sample(float max_t,
@@ -403,7 +431,7 @@ typedef struct VolumeIntegrateState {
 ccl_device_forceinline void volume_integrate_step_scattering(
     ccl_private const ShaderData *sd,
     ccl_private const Ray *ray,
-    const float3 equiangular_light_P,
+    ccl_private const EquiangularCoefficients &equiangular_coeffs,
     ccl_private const VolumeShaderCoefficients &ccl_restrict coeff,
     const Spectrum transmittance,
     ccl_private VolumeIntegrateState &ccl_restrict vstate,
@@ -474,7 +502,7 @@ ccl_device_forceinline void volume_integrate_step_scattering(
 
           /* Multiple importance sampling. */
           if (vstate.use_mis) {
-            const float equiangular_pdf = volume_equiangular_pdf(ray, equiangular_light_P, new_t);
+            const float equiangular_pdf = volume_equiangular_pdf(ray, equiangular_coeffs, new_t);
             const float mis_weight = power_heuristic(vstate.distance_pdf * distance_pdf,
                                                      equiangular_pdf);
             result.direct_throughput *= 2.0f * mis_weight;
@@ -509,7 +537,7 @@ ccl_device_forceinline void volume_integrate_heterogeneous(
     ccl_global float *ccl_restrict render_buffer,
     const float object_step_size,
     const VolumeSampleMethod direct_sample_method,
-    const float3 equiangular_light_P,
+    ccl_private const EquiangularCoefficients &equiangular_coeffs,
     ccl_private VolumeIntegrateResult &result)
 {
   PROFILING_INIT(kg, PROFILING_SHADE_VOLUME_INTEGRATE);
@@ -560,7 +588,7 @@ ccl_device_forceinline void volume_integrate_heterogeneous(
   /* Equiangular sampling: compute distance and PDF in advance. */
   if (vstate.direct_sample_method == VOLUME_SAMPLE_EQUIANGULAR) {
     result.direct_t = volume_equiangular_sample(
-        ray, equiangular_light_P, vstate.rscatter, &vstate.equiangular_pdf);
+        ray, equiangular_coeffs, vstate.rscatter, &vstate.equiangular_pdf);
   }
 #  ifdef __PATH_GUIDING__
   result.direct_sample_method = vstate.direct_sample_method;
@@ -614,7 +642,7 @@ ccl_device_forceinline void volume_integrate_heterogeneous(
 
           /* Scattering and absorption. */
           volume_integrate_step_scattering(
-              sd, ray, equiangular_light_P, coeff, transmittance, vstate, result);
+              sd, ray, equiangular_coeffs, coeff, transmittance, vstate, result);
         }
         else {
           /* Absorption only. */
@@ -673,7 +701,8 @@ ccl_device_forceinline bool integrate_volume_equiangular_sample_light(
     ccl_private const Ray *ccl_restrict ray,
     ccl_private const ShaderData *ccl_restrict sd,
     ccl_private const RNGState *ccl_restrict rng_state,
-    ccl_private float3 *ccl_restrict P)
+    ccl_private EquiangularCoefficients *ccl_restrict equiangular_coeffs,
+    ccl_private LightSample &ccl_restrict ls)
 {
   /* Test if there is a light or BSDF that needs direct light. */
   if (!kernel_data.integrator.use_direct_light) {
@@ -685,7 +714,6 @@ ccl_device_forceinline bool integrate_volume_equiangular_sample_light(
   const uint bounce = INTEGRATOR_STATE(state, path, bounce);
   const float3 rand_light = path_state_rng_3D(kg, rng_state, PRNG_LIGHT);
 
-  LightSample ls ccl_optional_struct_init;
   if (!light_sample_from_volume_segment(kg,
                                         rand_light,
                                         sd->time,
@@ -708,9 +736,10 @@ ccl_device_forceinline bool integrate_volume_equiangular_sample_light(
     return false;
   }
 
-  *P = ls.P;
+  equiangular_coeffs->P = ls.P;
 
-  return true;
+  return volume_equiangular_valid_ray_segment(
+      kg, ray->P, ray->D, &equiangular_coeffs->t_range, &ls);
 }
 
 /* Path tracing: sample point on light and evaluate light shader, then
@@ -725,41 +754,26 @@ ccl_device_forceinline void integrate_volume_direct_light(
 #  ifdef __PATH_GUIDING__
     ccl_private const Spectrum unlit_throughput,
 #  endif
-    ccl_private const Spectrum throughput)
+    ccl_private const Spectrum throughput,
+    ccl_private LightSample &ccl_restrict ls)
 {
   PROFILING_INIT(kg, PROFILING_SHADE_VOLUME_DIRECT_LIGHT);
 
-  if (!kernel_data.integrator.use_direct_light) {
+  if (!kernel_data.integrator.use_direct_light || ls.emitter_id == EMITTER_NONE) {
     return;
   }
 
-  /* Sample position on the same light again, now from the shading point where we scattered.
-   *
-   * Note that this means we sample the light tree twice when equiangular sampling is used.
-   * We could consider sampling the light tree just once and use the same light position again.
-   *
-   * This would make the PDFs for MIS weights more complicated due to having to account for
-   * both distance/equiangular and direct/indirect light sampling, but could be more accurate.
-   * Additionally we could end up behind the light or outside a spot light cone, which might
-   * waste a sample. Though on the other hand it would be possible to prevent that with
-   * equiangular sampling restricted to a smaller sub-segment where the light has influence. */
-  LightSample ls ccl_optional_struct_init;
+  /* Sample position on the same light again, now from the shading point where we scattered. */
   {
     const uint32_t path_flag = INTEGRATOR_STATE(state, path, flag);
     const uint bounce = INTEGRATOR_STATE(state, path, bounce);
     const float3 rand_light = path_state_rng_3D(kg, rng_state, PRNG_LIGHT);
+    const float3 N = zero_float3();
+    const int object_receiver = light_link_receiver_nee(kg, sd);
+    const int shader_flags = SD_BSDF_HAS_TRANSMISSION;
 
-    if (!light_sample_from_position(kg,
-                                    rng_state,
-                                    rand_light,
-                                    sd->time,
-                                    P,
-                                    zero_float3(),
-                                    light_link_receiver_nee(kg, sd),
-                                    SD_BSDF_HAS_TRANSMISSION,
-                                    bounce,
-                                    path_flag,
-                                    &ls))
+    if (!light_sample<false>(
+            kg, rand_light, sd->time, P, N, object_receiver, shader_flags, bounce, path_flag, &ls))
     {
       return;
     }
@@ -877,6 +891,7 @@ ccl_device_forceinline bool integrate_volume_phase_scatter(
     KernelGlobals kg,
     IntegratorState state,
     ccl_private ShaderData *sd,
+    ccl_private const Ray *ray,
     ccl_private const RNGState *rng_state,
     ccl_private const ShaderVolumePhases *phases)
 {
@@ -929,6 +944,7 @@ ccl_device_forceinline bool integrate_volume_phase_scatter(
   INTEGRATOR_STATE_WRITE(state, ray, P) = sd->P;
   INTEGRATOR_STATE_WRITE(state, ray, D) = normalize(phase_wo);
   INTEGRATOR_STATE_WRITE(state, ray, tmin) = 0.0f;
+  INTEGRATOR_STATE_WRITE(state, ray, previous_dt) = ray->tmax - ray->tmin;
   INTEGRATOR_STATE_WRITE(state, ray, tmax) = FLT_MAX;
 #  ifdef __RAY_DIFFERENTIALS__
   INTEGRATOR_STATE_WRITE(state, ray, dP) = differential_make_compact(sd->dP);
@@ -957,7 +973,8 @@ ccl_device_forceinline bool integrate_volume_phase_scatter(
 
   /* Update path state */
   INTEGRATOR_STATE_WRITE(state, path, mis_ray_pdf) = phase_pdf;
-  INTEGRATOR_STATE_WRITE(state, path, mis_origin_n) = zero_float3();
+  const float3 previous_P = ray->P + ray->D * ray->tmin;
+  INTEGRATOR_STATE_WRITE(state, path, mis_origin_n) = sd->P - previous_P;
   INTEGRATOR_STATE_WRITE(state, path, min_ray_pdf) = fminf(
       unguided_phase_pdf, INTEGRATOR_STATE(state, path, min_ray_pdf));
 
@@ -989,11 +1006,15 @@ ccl_device VolumeIntegrateEvent volume_integrate(KernelGlobals kg,
 
   /* Sample light ahead of volume stepping, for equiangular sampling. */
   /* TODO: distant lights are ignored now, but could instead use even distribution. */
+  LightSample ls ccl_optional_struct_init;
+  ls.emitter_id = EMITTER_NONE;
   const bool need_light_sample = !(INTEGRATOR_STATE(state, path, flag) & PATH_RAY_TERMINATE);
-  float3 equiangular_P = zero_float3();
-  const bool have_equiangular_sample = need_light_sample &&
-                                       integrate_volume_equiangular_sample_light(
-                                           kg, state, ray, &sd, &rng_state, &equiangular_P);
+
+  EquiangularCoefficients equiangular_coeffs = {zero_float3(), make_float2(ray->tmin, ray->tmax)};
+
+  const bool have_equiangular_sample =
+      need_light_sample && integrate_volume_equiangular_sample_light(
+                               kg, state, ray, &sd, &rng_state, &equiangular_coeffs, ls);
 
   VolumeSampleMethod direct_sample_method = (have_equiangular_sample) ?
                                                 volume_stack_sample_method(kg, state) :
@@ -1023,7 +1044,7 @@ ccl_device VolumeIntegrateEvent volume_integrate(KernelGlobals kg,
                                  render_buffer,
                                  step_size,
                                  direct_sample_method,
-                                 equiangular_P,
+                                 equiangular_coeffs,
                                  result);
 
   /* Perform path termination. The intersect_closest will have already marked this path
@@ -1091,7 +1112,8 @@ ccl_device VolumeIntegrateEvent volume_integrate(KernelGlobals kg,
 #  ifdef __PATH_GUIDING__
                                   unlit_throughput,
 #  endif
-                                  result.direct_throughput);
+                                  result.direct_throughput,
+                                  ls);
   }
 
   /* Indirect light.
@@ -1130,7 +1152,7 @@ ccl_device VolumeIntegrateEvent volume_integrate(KernelGlobals kg,
 #    endif
 #  endif
 
-    if (integrate_volume_phase_scatter(kg, state, &sd, &rng_state, &result.indirect_phases)) {
+    if (integrate_volume_phase_scatter(kg, state, &sd, ray, &rng_state, &result.indirect_phases)) {
       return VOLUME_PATH_SCATTERED;
     }
     else {

intern/cycles/kernel/integrator/state_template.h

@@ -75,6 +75,9 @@ KERNEL_STRUCT_MEMBER(ray, float, tmax, KERNEL_FEATURE_PATH_TRACING)
 KERNEL_STRUCT_MEMBER(ray, float, time, KERNEL_FEATURE_PATH_TRACING)
 KERNEL_STRUCT_MEMBER(ray, float, dP, KERNEL_FEATURE_PATH_TRACING)
 KERNEL_STRUCT_MEMBER(ray, float, dD, KERNEL_FEATURE_PATH_TRACING)
+#ifdef __LIGHT_TREE__
+KERNEL_STRUCT_MEMBER(ray, float, previous_dt, KERNEL_FEATURE_PATH_TRACING)
+#endif
 KERNEL_STRUCT_END(ray)
 
 /*************************** Intersection result ******************************/

intern/cycles/kernel/light/area.h

@@ -233,6 +233,11 @@ ccl_device bool area_light_spread_clamp_light(const float3 P,
   return true;
 }
 
+ccl_device_forceinline bool area_light_is_ellipse(const ccl_global KernelAreaLight *light)
+{
+  return light->invarea < 0.0f;
+}
+
 /* Common API. */
 /* Compute `eval_fac` and `pdf`. Also sample a new position on the light if `sample_coord`. */
 template<bool in_volume_segment>
@@ -338,7 +343,7 @@ ccl_device_inline bool area_light_sample(const ccl_global KernelLight *klight,
   const float light_v = dot(inplane, klight->area.axis_v) / klight->area.len_v;
 
   if (!in_volume_segment) {
-    const bool is_ellipse = (klight->area.invarea < 0.0f);
+    const bool is_ellipse = area_light_is_ellipse(&klight->area);
 
     /* Sampled point lies outside of the area light. */
     if (is_ellipse && (sqr(light_u) + sqr(light_v) > 0.25f)) {
@@ -380,7 +385,7 @@ ccl_device_inline bool area_light_intersect(const ccl_global KernelLight *klight
 {
   /* Area light. */
   const float invarea = fabsf(klight->area.invarea);
-  const bool is_ellipse = (klight->area.invarea < 0.0f);
+  const bool is_ellipse = area_light_is_ellipse(&klight->area);
   if (invarea == 0.0f) {
     return false;
   }
@@ -428,6 +433,55 @@ ccl_device_inline bool area_light_sample_from_intersection(
   return area_light_eval<false>(klight, ray_P, &light_P, ls, zero_float2(), false);
 }
 
+/* Returns the maximal distance between the light center and the boundary. */
+ccl_device_forceinline float area_light_max_extent(const ccl_global KernelAreaLight *light)
+{
+  return 0.5f * (area_light_is_ellipse(light) ? fmaxf(light->len_u, light->len_v) :
+                                                len(make_float2(light->len_u, light->len_v)));
+}
+
+/* Find the ray segment lit by the area light. */
+ccl_device_inline bool area_light_valid_ray_segment(const ccl_global KernelAreaLight *light,
+                                                    float3 P,
+                                                    float3 D,
+                                                    ccl_private float2 *t_range)
+{
+  bool valid;
+  const float tan_half_spread = light->tan_half_spread;
+  float3 axis = light->dir;
+
+  const bool angle_almost_zero = (tan_half_spread < 1e-5f);
+  if (angle_almost_zero) {
+    /* Map to local coordinate of the light. Do not use `itfm` in `KernelLight` as there might be
+     * additional scaling in the light size. */
+    const Transform tfm = make_transform(light->axis_u, light->axis_v, axis);
+    P = transform_point(&tfm, P);
+    D = transform_direction(&tfm, D);
+    axis = make_float3(0.0f, 0.0f, 1.0f);
+
+    const float half_len_u = 0.5f * light->len_u;
+    const float half_len_v = 0.5f * light->len_v;
+    if (area_light_is_ellipse(light)) {
+      valid = ray_infinite_cylinder_intersect(P, D, half_len_u, half_len_v, t_range);
+    }
+    else {
+      const float3 bbox_min = make_float3(-half_len_u, -half_len_v, 0.0f);
+      const float3 bbox_max = make_float3(half_len_u, half_len_v, FLT_MAX);
+      valid = ray_aabb_intersect(bbox_min, bbox_max, P, D, t_range);
+    }
+  }
+  else {
+    /* Conservative estimation with the smallest possible cone covering the whole spread. */
+    const float3 apex_to_point = P + area_light_max_extent(light) / tan_half_spread * axis;
+    const float cos_angle_sq = 1.0f / (1.0f + sqr(tan_half_spread));
+
+    valid = ray_cone_intersect(axis, apex_to_point, D, cos_angle_sq, t_range);
+  }
+
+  /* Limit the range to the positive side of the area light. */
+  return valid && ray_plane_intersect(axis, P, D, t_range);
+}
+
 template<bool in_volume_segment>
 ccl_device_forceinline bool area_light_tree_parameters(const ccl_global KernelLight *klight,
                                                        const float3 centroid,
@@ -464,9 +518,8 @@ ccl_device_forceinline bool area_light_tree_parameters(const ccl_global KernelLi
   const bool shape_above_surface = dot(N, centroid - P) + fabsf(dot(N, extentu)) +
                                        fabsf(dot(N, extentv)) >
                                    0;
-  const bool in_volume = is_zero(N);
 
-  return (front_facing && shape_above_surface) || in_volume;
+  return front_facing && shape_above_surface;
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/light/common.h

@@ -12,9 +12,9 @@ CCL_NAMESPACE_BEGIN
 
 typedef struct LightSample {
   float3 P;            /* position on light, or direction for distant light */
-  float3 Ng;           /* normal on light */
-  float3 D;            /* direction from shading point to light */
+  packed_float3 Ng;    /* normal on light */
   float t;             /* distance to light (FLT_MAX for distant light) */
+  float3 D;            /* direction from shading point to light */
   float u, v;          /* parametric coordinate on primitive */
   float pdf;           /* pdf for selecting light and point on light */
   float pdf_selection; /* pdf for selecting light */
@@ -25,6 +25,7 @@ typedef struct LightSample {
   int lamp;            /* lamp id */
   int group;           /* lightgroup */
   LightType type;      /* type of light */
+  int emitter_id;      /* index in the emitter array */
 } LightSample;
 
 /* Utilities */

intern/cycles/kernel/light/distribution.h

@@ -41,36 +41,14 @@ ccl_device int light_distribution_sample(KernelGlobals kg, const float rand)
   return index;
 }
 
-template<bool in_volume_segment>
 ccl_device_noinline bool light_distribution_sample(KernelGlobals kg,
-                                                   const float3 rand,
-                                                   const float time,
-                                                   const float3 P,
-                                                   const float3 N,
-                                                   const int object_receiver,
-                                                   const int shader_flags,
-                                                   const int bounce,
-                                                   const uint32_t path_flag,
+                                                   const float rand,
                                                    ccl_private LightSample *ls)
 {
   /* Sample light index from distribution. */
-  /* The first two dimensions of the Sobol sequence have better stratification. */
-  const int index = light_distribution_sample(kg, rand.z);
-  const float pdf_selection = kernel_data.integrator.distribution_pdf_lights;
-  const float2 rand_uv = float3_to_float2(rand);
-  return light_sample<in_volume_segment>(kg,
-                                         rand_uv,
-                                         time,
-                                         P,
-                                         N,
-                                         object_receiver,
-                                         shader_flags,
-                                         bounce,
-                                         path_flag,
-                                         index,
-                                         0,
-                                         pdf_selection,
-                                         ls);
+  ls->emitter_id = light_distribution_sample(kg, rand);
+  ls->pdf_selection = kernel_data.integrator.distribution_pdf_lights;
+  return true;
 }
 
 ccl_device_inline float light_distribution_pdf_lamp(KernelGlobals kg)

intern/cycles/kernel/light/light.h

@@ -177,7 +177,7 @@ ccl_device_inline bool light_sample(KernelGlobals kg,
 
 template<bool in_volume_segment>
 ccl_device_noinline bool light_sample(KernelGlobals kg,
-                                      const float2 rand,
+                                      const float3 rand_light,
                                       const float time,
                                       const float3 P,
                                       const float3 N,
@@ -185,33 +185,31 @@ ccl_device_noinline bool light_sample(KernelGlobals kg,
                                       const int shader_flags,
                                       const int bounce,
                                       const uint32_t path_flag,
-                                      const int emitter_index,
-                                      const int object_id,
-                                      const float pdf_selection,
                                       ccl_private LightSample *ls)
 {
+  /* The first two dimensions of the Sobol sequence have better stratification, use them to sample
+   * position on the light. */
+  const float2 rand = float3_to_float2(rand_light);
+
   int prim;
   MeshLight mesh_light;
 #ifdef __LIGHT_TREE__
   if (kernel_data.integrator.use_light_tree) {
     ccl_global const KernelLightTreeEmitter *kemitter = &kernel_data_fetch(light_tree_emitters,
-                                                                           emitter_index);
+                                                                           ls->emitter_id);
     prim = kemitter->light.id;
     mesh_light.shader_flag = kemitter->mesh_light.shader_flag;
-    mesh_light.object_id = object_id;
+    mesh_light.object_id = ls->object;
   }
   else
 #endif
   {
     ccl_global const KernelLightDistribution *kdistribution = &kernel_data_fetch(
-        light_distribution, emitter_index);
+        light_distribution, ls->emitter_id);
     prim = kdistribution->prim;
     mesh_light = kdistribution->mesh_light;
   }
 
-  /* A different value would be assigned in `triangle_light_sample()` if `!use_light_tree`. */
-  ls->pdf_selection = pdf_selection;
-
   if (prim >= 0) {
     /* Mesh light. */
     const int object = mesh_light.object_id;

intern/cycles/kernel/light/sample.h

@@ -329,17 +329,25 @@ ccl_device_inline bool light_sample_from_volume_segment(KernelGlobals kg,
                                                         const uint32_t path_flag,
                                                         ccl_private LightSample *ls)
 {
+  const int shader_flags = SD_BSDF_HAS_TRANSMISSION;
+
 #ifdef __LIGHT_TREE__
   if (kernel_data.integrator.use_light_tree) {
-    return light_tree_sample<true>(
-        kg, rand, time, P, D, t, object_receiver, SD_BSDF_HAS_TRANSMISSION, bounce, path_flag, ls);
+    if (!light_tree_sample<true>(kg, rand.z, P, D, t, object_receiver, shader_flags, ls)) {
+      return false;
+    }
   }
   else
 #endif
   {
-    return light_distribution_sample<true>(
-        kg, rand, time, P, D, object_receiver, SD_BSDF_HAS_TRANSMISSION, bounce, path_flag, ls);
+    if (!light_distribution_sample(kg, rand.z, ls)) {
+      return false;
     }
+  }
+
+  /* Sample position on the selected light. */
+  return light_sample<true>(
+      kg, rand, time, P, D, object_receiver, shader_flags, bounce, path_flag, ls);
 }
 
 ccl_device bool light_sample_from_position(KernelGlobals kg,
@@ -354,17 +362,24 @@ ccl_device bool light_sample_from_position(KernelGlobals kg,
                                            const uint32_t path_flag,
                                            ccl_private LightSample *ls)
 {
+  /* Randomly select a light. */
 #ifdef __LIGHT_TREE__
   if (kernel_data.integrator.use_light_tree) {
-    return light_tree_sample<false>(
-        kg, rand, time, P, N, 0.0f, object_receiver, shader_flags, bounce, path_flag, ls);
+    if (!light_tree_sample<false>(kg, rand.z, P, N, 0.0f, object_receiver, shader_flags, ls)) {
+      return false;
+    }
   }
   else
 #endif
   {
-    return light_distribution_sample<false>(
-        kg, rand, time, P, N, object_receiver, shader_flags, bounce, path_flag, ls);
+    if (!light_distribution_sample(kg, rand.z, ls)) {
+      return false;
     }
+  }
+
+  /* Sample position on the selected light. */
+  return light_sample<false>(
+      kg, rand, time, P, N, object_receiver, shader_flags, bounce, path_flag, ls);
 }
 
 /* Update light sample with new shading point position for MNEE. The position on the light is fixed
@@ -415,13 +430,15 @@ ccl_device_inline float light_sample_mis_weight_forward_surface(KernelGlobals kg
 #ifdef __LIGHT_TREE__
   if (kernel_data.integrator.use_light_tree) {
     float3 ray_P = INTEGRATOR_STATE(state, ray, P);
+    const float dt = INTEGRATOR_STATE(state, ray, previous_dt);
     const float3 N = INTEGRATOR_STATE(state, path, mis_origin_n);
+
     uint lookup_offset = kernel_data_fetch(object_lookup_offset, sd->object);
     uint prim_offset = kernel_data_fetch(object_prim_offset, sd->object);
     uint triangle = kernel_data_fetch(triangle_to_tree, sd->prim - prim_offset + lookup_offset);
 
     pdf *= light_tree_pdf(
-        kg, ray_P, N, path_flag, sd->object, triangle, light_link_receiver_forward(kg, state));
+        kg, ray_P, N, dt, path_flag, sd->object, triangle, light_link_receiver_forward(kg, state));
   }
   else
 #endif
@@ -445,9 +462,11 @@ ccl_device_inline float light_sample_mis_weight_forward_lamp(KernelGlobals kg,
 #ifdef __LIGHT_TREE__
   if (kernel_data.integrator.use_light_tree) {
     const float3 N = INTEGRATOR_STATE(state, path, mis_origin_n);
+    const float dt = INTEGRATOR_STATE(state, ray, previous_dt);
     pdf *= light_tree_pdf(kg,
                           P,
                           N,
+                          dt,
                           path_flag,
                           0,
                           kernel_data_fetch(light_to_tree, ls->lamp),
@@ -485,9 +504,10 @@ ccl_device_inline float light_sample_mis_weight_forward_background(KernelGlobals
 #ifdef __LIGHT_TREE__
   if (kernel_data.integrator.use_light_tree) {
     const float3 N = INTEGRATOR_STATE(state, path, mis_origin_n);
+    const float dt = INTEGRATOR_STATE(state, ray, previous_dt);
     uint light = kernel_data_fetch(light_to_tree, kernel_data.background.light_index);
     pdf *= light_tree_pdf(
-        kg, ray_P, N, path_flag, 0, light, light_link_receiver_forward(kg, state));
+        kg, ray_P, N, dt, path_flag, 0, light, light_link_receiver_forward(kg, state));
   }
   else
 #endif

intern/cycles/kernel/light/spot.h

@@ -265,6 +265,24 @@ ccl_device_inline bool spot_light_sample_from_intersection(
   return true;
 }
 
+/* Find the ray segment lit by the spot light. */
+ccl_device_inline bool spot_light_valid_ray_segment(const ccl_global KernelLight *klight,
+                                                    const float3 P,
+                                                    const float3 D,
+                                                    ccl_private float2 *t_range)
+{
+  /* Convert to local space of the spot light. */
+  const Transform itfm = klight->itfm;
+  float3 local_P = P + klight->spot.dir * klight->spot.ray_segment_dp;
+  local_P = transform_point(&itfm, local_P);
+  const float3 local_D = transform_direction(&itfm, D);
+  const float3 axis = make_float3(0.0f, 0.0f, -1.0f);
+
+  /* Intersect the ray with the smallest enclosing cone of the light spread. */
+  return ray_cone_intersect(
+      axis, local_P, local_D, sqr(klight->spot.cos_half_spot_angle), t_range);
+}
+
 template<bool in_volume_segment>
 ccl_device_forceinline bool spot_light_tree_parameters(const ccl_global KernelLight *klight,
                                                        const float3 centroid,

intern/cycles/kernel/light/tree.h

@@ -148,10 +148,7 @@ ccl_device void light_tree_importance(const float3 N_or_D,
   float cos_min_incidence_angle = 1.0f;
   float cos_max_incidence_angle = 1.0f;
 
-  /* When sampling the light tree for the second time in `shade_volume.h` and when query the pdf in
-   * `sample.h`. */
-  const bool in_volume = is_zero(N_or_D);
-  if (!in_volume_segment && !in_volume) {
+  if (!in_volume_segment) {
     const float3 N = N_or_D;
     const float cos_theta_i = has_transmission ? fabsf(dot(point_to_centroid, N)) :
                                                  dot(point_to_centroid, N);
@@ -221,9 +218,9 @@ ccl_device void light_tree_importance(const float3 N_or_D,
   max_importance = fabsf(f_a * cos_min_incidence_angle * energy * cos_min_outgoing_angle /
                          (in_volume_segment ? min_distance : sqr(min_distance)));
 
-  /* TODO: also min importance for volume? */
+  /* TODO: compute proper min importance for volume. */
   if (in_volume_segment) {
-    min_importance = max_importance;
+    min_importance = 0.0f;
     return;
   }
 
@@ -270,10 +267,10 @@ ccl_device bool compute_emitter_centroid_and_dir(KernelGlobals kg,
         /* Arbitrary centroid and direction. */
         centroid = make_float3(0.0f, 0.0f, 1.0f);
         dir = make_float3(0.0f, 0.0f, -1.0f);
-        return !in_volume_segment;
+        break;
       case LIGHT_DISTANT:
         dir = centroid;
-        return !in_volume_segment;
+        break;
       default:
         return false;
     }
@@ -323,12 +320,13 @@ ccl_device void light_tree_node_importance(KernelGlobals kg,
   float cos_theta_u;
   float distance;
   if (knode->type == LIGHT_TREE_DISTANT) {
-    if (in_volume_segment) {
-      return;
-    }
     point_to_centroid = -bcone.axis;
     cos_theta_u = fast_cosf(bcone.theta_o + bcone.theta_e);
     distance = 1.0f;
+    if (t == FLT_MAX) {
+      /* In world volume, distant light has no contribution. */
+      return;
+    }
   }
   else {
     const float3 centroid = 0.5f * (bbox.min + bbox.max);
@@ -339,6 +337,9 @@ ccl_device void light_tree_node_importance(KernelGlobals kg,
       /* Minimal distance of the ray to the cluster. */
       distance = len(centroid - closest_point);
       point_to_centroid = -compute_v(centroid, P, D, bcone.axis, t);
+      /* FIXME(weizhen): it is not clear from which point the `cos_theta_u` should be computed in
+       * volume segment. We could use `closest_point` as a conservative measure, but then
+       * `point_to_centroid` should also use `closest_point`. */
       cos_theta_u = light_tree_cos_bounding_box_angle(bbox, closest_point, point_to_centroid);
     }
     else {
@@ -697,17 +698,16 @@ ccl_device int light_tree_root_node_index(KernelGlobals kg, const int object_rec
   return 0;
 }
 
+/* Pick a random light from the light tree from a given shading point P, write to the picked light
+ * index and the probability of picking the light. */
 template<bool in_volume_segment>
 ccl_device_noinline bool light_tree_sample(KernelGlobals kg,
-                                           const float3 rand,
-                                           const float time,
+                                           const float rand,
                                            const float3 P,
                                            float3 N_or_D,
                                            float t,
                                            const int object_receiver,
                                            const int shader_flags,
-                                           const int bounce,
-                                           const uint32_t path_flag,
                                            ccl_private LightSample *ls)
 {
   if (!kernel_data.integrator.use_direct_light) {
@@ -718,10 +718,8 @@ ccl_device_noinline bool light_tree_sample(KernelGlobals kg,
   float pdf_leaf = 1.0f;
   float pdf_selection = 1.0f;
   int selected_emitter = -1;
-  int object_emitter = 0;
   int node_index = light_tree_root_node_index(kg, object_receiver);
-  /* The first two dimensions of the Sobol sequence have better stratification. */
-  float rand_selection = rand.z;
+  float rand_selection = rand;
 
   float3 local_P = P;
 
@@ -743,7 +741,7 @@ ccl_device_noinline bool light_tree_sample(KernelGlobals kg,
       }
 
       /* Continue with the picked mesh light. */
-      object_emitter = kernel_data_fetch(light_tree_emitters, selected_emitter).mesh.object_id;
+      ls->object = kernel_data_fetch(light_tree_emitters, selected_emitter).mesh.object_id;
       continue;
     }
 
@@ -766,27 +764,18 @@ ccl_device_noinline bool light_tree_sample(KernelGlobals kg,
     pdf_leaf *= (node_index == left_index) ? left_prob : (1.0f - left_prob);
   }
 
-  pdf_selection *= pdf_leaf;
+  ls->emitter_id = selected_emitter;
+  ls->pdf_selection = pdf_selection * pdf_leaf;
 
-  return light_sample<in_volume_segment>(kg,
-                                         float3_to_float2(rand),
-                                         time,
-                                         P,
-                                         N_or_D,
-                                         object_receiver,
-                                         shader_flags,
-                                         bounce,
-                                         path_flag,
-                                         selected_emitter,
-                                         object_emitter,
-                                         pdf_selection,
-                                         ls);
+  return true;
 }
 
 /* We need to be able to find the probability of selecting a given light for MIS. */
+template<bool in_volume_segment>
 ccl_device float light_tree_pdf(KernelGlobals kg,
                                 float3 P,
                                 float3 N,
+                                const float dt,
                                 const int path_flag,
                                 const int object_emitter,
                                 const uint index_emitter,
@@ -796,7 +785,7 @@ ccl_device float light_tree_pdf(KernelGlobals kg,
 
   ccl_global const KernelLightTreeEmitter *kemitter = &kernel_data_fetch(light_tree_emitters,
                                                                          index_emitter);
-  int root_index;
+  int subtree_root_index;
   uint bit_trail, target_emitter;
 
   if (is_triangle(kemitter)) {
@@ -805,16 +794,17 @@ ccl_device float light_tree_pdf(KernelGlobals kg,
     target_emitter = kernel_data_fetch(object_to_tree, object_emitter);
     ccl_global const KernelLightTreeEmitter *kmesh = &kernel_data_fetch(light_tree_emitters,
                                                                         target_emitter);
-    root_index = kmesh->mesh.node_id;
-    ccl_global const KernelLightTreeNode *kroot = &kernel_data_fetch(light_tree_nodes, root_index);
+    subtree_root_index = kmesh->mesh.node_id;
+    ccl_global const KernelLightTreeNode *kroot = &kernel_data_fetch(light_tree_nodes,
+                                                                     subtree_root_index);
     bit_trail = kroot->bit_trail;
 
     if (kroot->type == LIGHT_TREE_INSTANCE) {
-      root_index = kroot->instance.reference;
+      subtree_root_index = kroot->instance.reference;
     }
   }
   else {
-    root_index = 0;
+    subtree_root_index = -1;
     bit_trail = kemitter->bit_trail;
     target_emitter = index_emitter;
   }
@@ -836,8 +826,8 @@ ccl_device float light_tree_pdf(KernelGlobals kg,
       for (int i = 0; i < knode->num_emitters; i++) {
         const int emitter = knode->leaf.first_emitter + i;
         float max_importance, min_importance;
-        light_tree_emitter_importance<false>(
-            kg, P, N, 0, has_transmission, emitter, max_importance, min_importance);
+        light_tree_emitter_importance<in_volume_segment>(
+            kg, P, N, dt, has_transmission, emitter, max_importance, min_importance);
         num_has_importance += (max_importance > 0);
         if (emitter == target_emitter) {
           target_max_importance = max_importance;
@@ -856,13 +846,13 @@ ccl_device float light_tree_pdf(KernelGlobals kg,
         return 0.0f;
       }
 
-      if (root_index) {
+      if (subtree_root_index != -1) {
         /* Arrived at the mesh light. Continue with the subtree. */
         float unused;
-        light_tree_to_local_space<false>(kg, object_emitter, P, N, unused);
+        light_tree_to_local_space<in_volume_segment>(kg, object_emitter, P, N, unused);
 
-        node_index = root_index;
-        root_index = 0;
+        node_index = subtree_root_index;
+        subtree_root_index = -1;
         target_emitter = index_emitter;
         bit_trail = kemitter->bit_trail;
         continue;
@@ -877,8 +867,8 @@ ccl_device float light_tree_pdf(KernelGlobals kg,
     const int right_index = knode->inner.right_child;
 
     float left_prob;
-    if (!get_left_probability<false>(
-            kg, P, N, 0, has_transmission, left_index, right_index, left_prob))
+    if (!get_left_probability<in_volume_segment>(
+            kg, P, N, dt, has_transmission, left_index, right_index, left_prob))
     {
       return 0.0f;
     }
@@ -896,4 +886,27 @@ ccl_device float light_tree_pdf(KernelGlobals kg,
   }
 }
 
+/* If the function is called in volume, retrieve the previous point in volume segment, and compute
+ * pdf from there. Otherwise compute from the current shading point. */
+ccl_device_inline float light_tree_pdf(KernelGlobals kg,
+                                       float3 P,
+                                       float3 N,
+                                       const float dt,
+                                       const int path_flag,
+                                       const int emitter_object,
+                                       const uint emitter_id,
+                                       const int object_receiver)
+{
+  if (path_flag & PATH_RAY_VOLUME_SCATTER) {
+    const float3 D_times_t = N;
+    const float3 D = normalize(D_times_t);
+    P = P - D_times_t;
+    return light_tree_pdf<true>(
+        kg, P, D, dt, path_flag, emitter_object, emitter_id, object_receiver);
+  }
+
+  return light_tree_pdf<false>(
+      kg, P, N, 0.0f, path_flag, emitter_object, emitter_id, object_receiver);
+}
+
 CCL_NAMESPACE_END

intern/cycles/kernel/light/triangle.h

@@ -269,6 +269,26 @@ ccl_device_forceinline bool triangle_light_sample(KernelGlobals kg,
   return (ls->pdf > 0.0f);
 }
 
+/* Find the ray segment lit by the triangle light. */
+ccl_device_inline bool triangle_light_valid_ray_segment(KernelGlobals kg,
+                                                        const float3 P,
+                                                        const float3 D,
+                                                        ccl_private float2 *t_range,
+                                                        const ccl_private LightSample *ls)
+{
+  const int shader_flag = kernel_data_fetch(shaders, ls->shader & SHADER_MASK).flags;
+  const int SD_MIS_BOTH = SD_MIS_BACK | SD_MIS_FRONT;
+  if ((shader_flag & SD_MIS_BOTH) == SD_MIS_BOTH) {
+    /* Both sides are sampled, the complete ray segment is visible. */
+    return true;
+  }
+
+  /* Only one side is sampled, intersect the ray and the triangle light plane to find the visible
+   * ray segment. Flip normal if Emission Sampling is set to back. */
+  const float3 N = ls->Ng;
+  return ray_plane_intersect((shader_flag & SD_MIS_BACK) ? -N : N, P, D, t_range);
+}
+
 template<bool in_volume_segment>
 ccl_device_forceinline bool triangle_light_tree_parameters(
     KernelGlobals kg,
@@ -307,9 +327,8 @@ ccl_device_forceinline bool triangle_light_tree_parameters(
   }
 
   const bool front_facing = bcone.theta_o != 0.0f || dot(bcone.axis, point_to_centroid) < 0;
-  const bool in_volume = is_zero(N);
 
-  return (front_facing && shape_above_surface) || in_volume;
+  return front_facing && shape_above_surface;
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/types.h

@@ -45,6 +45,7 @@ CCL_NAMESPACE_BEGIN
 #define OBJECT_NONE (~0)
 #define PRIM_NONE (~0)
 #define LAMP_NONE (~0)
+#define EMITTER_NONE (~0)
 #define ID_NONE (0.0f)
 #define PASS_UNUSED (~0)
 #define LIGHTGROUP_NONE (~0)
@@ -1376,6 +1377,8 @@ typedef struct KernelSpotLight {
   int is_sphere;
   /* For non-uniform object scaling, the actual spread might be different. */
   float cos_half_larger_spread;
+  /* Distance from the apex of the smallest enclosing cone of the light spread to light center. */
+  float ray_segment_dp;
 } KernelSpotLight;
 
 /* PointLight is SpotLight with only radius and invarea being used. */

intern/cycles/scene/light.cpp

@@ -1362,6 +1362,9 @@ void LightManager::device_update_lights(Device *device, DeviceScene *dscene, Sce
       /* Choose the angle which spans a larger cone. */
       klights[light_index].spot.cos_half_larger_spread = inversesqrtf(
           1.0f + tan_sq * fmaxf(len_u_sq, len_v_sq) / len_w_sq);
+      /* radius / sin(half_angle_small) */
+      klights[light_index].spot.ray_segment_dp =
+          light->size * sqrtf(1.0f + len_w_sq / (tan_sq * fminf(len_u_sq, len_v_sq)));
     }
 
     klights[light_index].shader_id = shader_id;

intern/cycles/util/math.h

@@ -1030,6 +1030,46 @@ ccl_device_inline uint32_t reverse_integer_bits(uint32_t x)
 #endif
 }
 
+/* Check if intervals (first->x, first->y) and (second.x, second.y) intersect, and replace the
+ * first interval with their intersection. */
+ccl_device_inline bool intervals_intersect(ccl_private float2 *first, const float2 second)
+{
+  first->x = fmaxf(first->x, second.x);
+  first->y = fminf(first->y, second.y);
+
+  return first->x < first->y;
+}
+
+/* Solve quadratic equation a*x^2 + b*x + c = 0, adapted from Mitsuba 3
+ * The solution is ordered so that x1 <= x2.
+ * Returns true if at least one solution is found.  */
+ccl_device_inline bool solve_quadratic(
+    const float a, const float b, const float c, ccl_private float &x1, ccl_private float &x2)
+{
+  /* If the equation is linear, the solution is -c/b, but b has to be non-zero. */
+  const bool valid_linear = (a == 0.0f) && (b != 0.0f);
+  x1 = x2 = -c / b;
+
+  const float discriminant = sqr(b) - 4.0f * a * c;
+  /* Allow slightly negative discriminant in case of numerical precision issues. */
+  const bool valid_quadratic = (a != 0.0f) && (discriminant > -1e-5f);
+
+  if (valid_quadratic) {
+    /* Numerically stable version of (-b ± sqrt(discriminant)) / (2 * a), avoiding catastrophic
+     * cancellation when `b` is very close to `sqrt(discriminant)`, by finding the solution of
+     * greater magnitude which does not suffer from loss of precision, then using the identity
+     * x1 * x2 = c / a. */
+    const float temp = -0.5f * (b + copysignf(safe_sqrtf(discriminant), b));
+    const float r1 = temp / a;
+    const float r2 = c / temp;
+
+    x1 = fminf(r1, r2);
+    x2 = fmaxf(r1, r2);
+  }
+
+  return (valid_linear || valid_quadratic);
+}
+
 CCL_NAMESPACE_END
 
 #endif /* __UTIL_MATH_H__ */

intern/cycles/util/math_intersect.h

@@ -302,6 +302,140 @@ ccl_device bool ray_quad_intersect(float3 ray_P,
   return true;
 }
 
+/* Find the ray segment that lies in the same side as the normal `N` of the plane.
+ * `P` is the vector pointing from any point on the plane to the ray origin. */
+ccl_device bool ray_plane_intersect(const float3 N,
+                                    const float3 P,
+                                    const float3 ray_D,
+                                    ccl_private float2 *t_range)
+{
+  const float DN = dot(ray_D, N);
+
+  /* Distance from P to the plane. */
+  const float t = -dot(P, N) / DN;
+
+  /* Limit the range to the positive side. */
+  if (DN > 0.0f) {
+    t_range->x = fmaxf(t_range->x, t);
+  }
+  else {
+    t_range->y = fminf(t_range->y, t);
+  }
+
+  return t_range->x < t_range->y;
+}
+
+/* Find the ray segment inside an axis-aligned bounding box. */
+ccl_device bool ray_aabb_intersect(const float3 bbox_min,
+                                   const float3 bbox_max,
+                                   const float3 ray_P,
+                                   const float3 ray_D,
+                                   ccl_private float2 *t_range)
+{
+  const float3 inv_ray_D = rcp(ray_D);
+
+  /* Absolute distances to lower and upper box coordinates; */
+  const float3 t_lower = (bbox_min - ray_P) * inv_ray_D;
+  const float3 t_upper = (bbox_max - ray_P) * inv_ray_D;
+
+  /* The four t-intervals (for x-/y-/z-slabs, and ray p(t)). */
+  const float4 tmins = float3_to_float4(min(t_lower, t_upper), t_range->x);
+  const float4 tmaxes = float3_to_float4(max(t_lower, t_upper), t_range->y);
+
+  /* Max of mins and min of maxes. */
+  const float tmin = reduce_max(tmins);
+  const float tmax = reduce_min(tmaxes);
+
+  *t_range = make_float2(tmin, tmax);
+
+  return tmin < tmax;
+}
+
+/* Find the segment of a ray defined by P + D * t that lies inside a cylinder defined by
+ * (x / len_u)^2 + (y / len_v)^2 = 1. */
+ccl_device_inline bool ray_infinite_cylinder_intersect(const float3 P,
+                                                       const float3 D,
+                                                       const float len_u,
+                                                       const float len_v,
+                                                       ccl_private float2 *t_range)
+{
+  /* Convert to a 2D problem. */
+  const float2 inv_len = 1.0f / make_float2(len_u, len_v);
+  float2 P_proj = float3_to_float2(P) * inv_len;
+  const float2 D_proj = float3_to_float2(D) * inv_len;
+
+  /* Solve quadratic equation a*t^2 + 2b*t + c = 0. */
+  const float a = dot(D_proj, D_proj);
+  float b = dot(P_proj, D_proj);
+
+  /* Move ray origin closer to the cylinder to prevent precision issue when the ray is far away. */
+  const float t_mid = -b / a;
+  P_proj += D_proj * t_mid;
+
+  /* Recompute b from the shifted origin. */
+  b = dot(P_proj, D_proj);
+  const float c = dot(P_proj, P_proj) - 1.0f;
+
+  float tmin, tmax;
+  const bool valid = solve_quadratic(a, 2.0f * b, c, tmin, tmax);
+
+  return valid && intervals_intersect(t_range, make_float2(tmin, tmax) + t_mid);
+}
+
+/* *
+ * Find the ray segment inside a single-sided cone.
+ *
+ * \param axis: a unit-length direction around which the cone has a circular symmetry
+ * \param P: the vector pointing from the cone apex to the ray origin
+ * \param D: the direction of the ray, does not need to have unit-length
+ * \param cos_angle_sq: `sqr(cos(half_aperture_of_the_cone))`
+ * \param t_range: the lower and upper bounds between which the ray lies inside the cone
+ * \return whether the intersection exists and is in the provided range
+ *
+ * See https://www.geometrictools.com/Documentation/IntersectionLineCone.pdf for illustration
+ */
+ccl_device_inline bool ray_cone_intersect(const float3 axis,
+                                          const float3 P,
+                                          float3 D,
+                                          const float cos_angle_sq,
+                                          ccl_private float2 *t_range)
+{
+  if (cos_angle_sq < 1e-4f) {
+    /* The cone is nearly a plane. */
+    return ray_plane_intersect(axis, P, D, t_range);
+  }
+
+  const float inv_len = inversesqrtf(len_squared(D));
+  D *= inv_len;
+
+  const float AD = dot(axis, D);
+  const float AP = dot(axis, P);
+
+  const float a = sqr(AD) - cos_angle_sq;
+  const float b = 2.0f * (AD * AP - cos_angle_sq * dot(D, P));
+  const float c = sqr(AP) - cos_angle_sq * dot(P, P);
+
+  float tmin = 0.0f, tmax = FLT_MAX;
+  bool valid = solve_quadratic(a, b, c, tmin, tmax);
+
+  /* Check if the intersections are in the same hemisphere as the cone. */
+  const bool tmin_valid = AP + tmin * AD > 0.0f;
+  const bool tmax_valid = AP + tmax * AD > 0.0f;
+
+  valid &= (tmin_valid || tmax_valid);
+
+  if (!tmax_valid) {
+    tmax = tmin;
+    tmin = 0.0f;
+  }
+  else if (!tmin_valid) {
+    tmin = tmax;
+    tmax = FLT_MAX;
+  }
+
+  return valid && intervals_intersect(t_range, make_float2(tmin, tmax) * inv_len);
+}
+
 CCL_NAMESPACE_END
 
 #endif /* __UTIL_MATH_INTERSECT_H__ */

intern/cycles/util/transform.h

@@ -161,6 +161,17 @@ ccl_device_inline Transform make_transform(float a,
   return t;
 }
 
+ccl_device_inline Transform make_transform(const float3 x, const float3 y, const float3 z)
+{
+  Transform t;
+
+  t.x = float3_to_float4(x, 0.0f);
+  t.y = float3_to_float4(y, 0.0f);
+  t.z = float3_to_float4(z, 0.0f);
+
+  return t;
+}
+
 ccl_device_inline Transform euler_to_transform(const float3 euler)
 {
   float cx = cosf(euler.x);

intern/ghost/intern/GHOST_WindowWayland.cc

@@ -1886,6 +1886,30 @@ GHOST_WindowWayland::GHOST_WindowWayland(GHOST_SystemWayland *system,
     gwl_window_state_set(window_, state);
   }
 
+  /* NOTE(@ideasman42): Round trips are important before committing.
+   * This is needed because setting the state is likely to resize the window
+   * (in the case of maximized & full-screen), "normal" windows may still be resized when
+   * they are too large or with tiling window-managers.
+   *
+   * The additional updates allow for the actual size to be configured by the window manager
+   * which is read back before committing the surface. This avoids displaying the buffer
+   * before it's resized (avoiding flickering).
+   *
+   * Without the round-trip here:
+   * - The window will be created and this function will return using the requested buffer size,
+   *   instead of the window size which ends up being used (causing a visible flicker).
+   *   This has the down side that Blender's internal window state has the outdated size
+   *   which then gets immediately resized, causing a noticeable glitch.
+   * - The window decorations will be displayed at the wrong size before refreshing
+   *   at the new size.
+   * - On GNOME-Shell 46 shows the previous buffer-size under some conditions, see #119871.
+   * - 2x updates are needed for RIVER & HYPRLAND.
+   */
+  for (int i = 0; i < 2; i++) {
+    wl_display_flush(system->wl_display_get());
+    wl_display_dispatch(system->wl_display_get());
+  }
+
   /* Commit after setting the buffer.
    * While postponing until after the buffer drawing is context is set
    * isn't essential, it reduces flickering. */

scripts/startup/bl_ui/properties_data_armature.py

@@ -158,7 +158,7 @@ class ARMATURE_MT_collection_tree_context_menu(Menu):
         # editable or not. That means this menu has to do the disabling for it.
         sub = layout.column()
         sub.enabled = not active_bcoll_is_locked
-        sub.operator("armature.collection_add", text="Add Child Collection")
+        sub.operator("armature.collection_add", text="Add Bone Collection")
         sub.operator("armature.collection_remove")
         sub.operator("armature.collection_remove_unused", text="Remove Unused Collections")
 

scripts/startup/bl_ui/space_view3d.py

@@ -2187,6 +2187,7 @@ class VIEW3D_MT_paint_grease_pencil(Menu):
         layout.separator()
 
         layout.menu("VIEW3D_MT_edit_greasepencil_showhide")
+        layout.menu("VIEW3D_MT_edit_greasepencil_cleanup")
 
         layout.separator()
 
@@ -5803,6 +5804,15 @@ class VIEW3D_MT_edit_greasepencil_showhide(Menu):
         layout.operator("grease_pencil.layer_hide", text="Hide Inactive Layers").unselected = True
 
 
+class VIEW3D_MT_edit_greasepencil_cleanup(Menu):
+    bl_label = "Cleanup"
+
+    def draw(self, _context):
+        layout = self.layout
+
+        layout.operator("grease_pencil.clean_loose")
+
+
 class VIEW3D_MT_edit_greasepencil(Menu):
     bl_label = "Grease Pencil"
 
@@ -5828,7 +5838,7 @@ class VIEW3D_MT_edit_greasepencil(Menu):
 
         layout.menu("VIEW3D_MT_edit_greasepencil_showhide")
         layout.operator_menu_enum("grease_pencil.separate", "mode", text="Separate")
-        layout.operator("grease_pencil.clean_loose")
+        layout.menu("VIEW3D_MT_edit_greasepencil_cleanup")
 
         layout.separator()
 
@@ -7526,14 +7536,14 @@ class VIEW3D_PT_snapping(Panel):
         layout = self.layout
         col = layout.column()
 
-        col.label(text="Snap With")
+        col.label(text="Snap Base")
         row = col.row(align=True)
         row.prop(tool_settings, "snap_target", expand=True)
 
-        col.label(text="Snap To")
+        col.label(text="Snap Target")
         col.prop(tool_settings, "snap_elements_base", expand=True)
 
-        col.label(text="Snap Individual Elements To")
+        col.label(text="Snap Target for Individual Elements")
         col.prop(tool_settings, "snap_elements_individual", expand=True)
 
         col.separator()
@@ -8988,6 +8998,7 @@ classes = (
     VIEW3D_MT_edit_gpencil_delete,
     VIEW3D_MT_edit_gpencil_showhide,
     VIEW3D_MT_edit_greasepencil_showhide,
+    VIEW3D_MT_edit_greasepencil_cleanup,
     VIEW3D_MT_weight_gpencil,
     VIEW3D_MT_gpencil_animation,
     VIEW3D_MT_gpencil_simplify,

source/blender/blenkernel/intern/attribute_access.cc

@@ -253,54 +253,6 @@ static AttributeIDRef attribute_id_from_custom_data_layer(const CustomDataLayer
   return layer.name;
 }
 
-static bool add_builtin_type_custom_data_layer_from_init(CustomData &custom_data,
-                                                         const eCustomDataType data_type,
-                                                         const int domain_num,
-                                                         const AttributeInit &initializer)
-{
-  switch (initializer.type) {
-    case AttributeInit::Type::Construct: {
-      void *data = CustomData_add_layer(&custom_data, data_type, CD_CONSTRUCT, domain_num);
-      return data != nullptr;
-    }
-    case AttributeInit::Type::DefaultValue: {
-      void *data = CustomData_add_layer(&custom_data, data_type, CD_SET_DEFAULT, domain_num);
-      return data != nullptr;
-    }
-    case AttributeInit::Type::VArray: {
-      void *data = CustomData_add_layer(&custom_data, data_type, CD_CONSTRUCT, domain_num);
-      if (data == nullptr) {
-        return false;
-      }
-      const GVArray &varray = static_cast<const AttributeInitVArray &>(initializer).varray;
-      varray.materialize_to_uninitialized(varray.index_range(), data);
-      return true;
-    }
-    case AttributeInit::Type::MoveArray: {
-      void *src_data = static_cast<const AttributeInitMoveArray &>(initializer).data;
-      const void *stored_data = CustomData_add_layer_with_data(
-          &custom_data, data_type, src_data, domain_num, nullptr);
-      if (stored_data == nullptr) {
-        return false;
-      }
-      if (stored_data != src_data) {
-        MEM_freeN(src_data);
-        return true;
-      }
-      return true;
-    }
-    case AttributeInit::Type::Shared: {
-      const AttributeInitShared &init = static_cast<const AttributeInitShared &>(initializer);
-      const void *stored_data = CustomData_add_layer_with_data(
-          &custom_data, data_type, const_cast<void *>(init.data), domain_num, init.sharing_info);
-      return stored_data != nullptr;
-    }
-  }
-
-  BLI_assert_unreachable();
-  return false;
-}
-
 static void *add_generic_custom_data_layer(CustomData &custom_data,
                                            const eCustomDataType data_type,
                                            const eCDAllocType alloctype,
@@ -393,10 +345,7 @@ static bool custom_data_layer_matches_attribute_id(const CustomDataLayer &layer,
 
 bool BuiltinCustomDataLayerProvider::layer_exists(const CustomData &custom_data) const
 {
-  if (stored_as_named_attribute_) {
-    return CustomData_get_named_layer_index(&custom_data, stored_type_, name_) != -1;
-  }
-  return CustomData_has_layer(&custom_data, stored_type_);
+  return CustomData_get_named_layer_index(&custom_data, data_type_, name_) != -1;
 }
 
 GAttributeReader BuiltinCustomDataLayerProvider::try_get_for_read(const void *owner) const
@@ -416,13 +365,7 @@ GAttributeReader BuiltinCustomDataLayerProvider::try_get_for_read(const void *ow
     return {};
   }
 
-  int index;
-  if (stored_as_named_attribute_) {
-    index = CustomData_get_named_layer_index(custom_data, stored_type_, name_);
-  }
-  else {
-    index = CustomData_get_layer_index(custom_data, stored_type_);
-  }
+  const int index = CustomData_get_named_layer_index(custom_data, data_type_, name_);
   if (index == -1) {
     return {};
   }
@@ -452,13 +395,7 @@ GAttributeWriter BuiltinCustomDataLayerProvider::try_get_for_write(void *owner)
     return {};
   }
 
-  void *data = nullptr;
-  if (stored_as_named_attribute_) {
-    data = CustomData_get_layer_named_for_write(custom_data, stored_type_, name_, element_num);
-  }
-  else {
-    data = CustomData_get_layer_for_write(custom_data, stored_type_, element_num);
-  }
+  void *data = CustomData_get_layer_named_for_write(custom_data, data_type_, name_, element_num);
   if (data == nullptr) {
     return {};
   }
@@ -475,57 +412,42 @@ bool BuiltinCustomDataLayerProvider::try_delete(void *owner) const
     return {};
   }
 
-  auto update = [&]() {
+  const int element_num = custom_data_access_.get_element_num(owner);
+  if (CustomData_free_layer_named(custom_data, name_, element_num)) {
     if (update_on_change_ != nullptr) {
       update_on_change_(owner);
     }
-  };
-
-  const int element_num = custom_data_access_.get_element_num(owner);
-  if (stored_as_named_attribute_) {
-    if (CustomData_free_layer_named(custom_data, name_, element_num)) {
-      update();
     return true;
   }
-    return false;
-  }
-
-  const int layer_index = CustomData_get_layer_index(custom_data, stored_type_);
-  if (CustomData_free_layer(custom_data, stored_type_, element_num, layer_index)) {
-    update();
-    return true;
-  }
-
   return false;
 }
 
 bool BuiltinCustomDataLayerProvider::try_create(void *owner,
                                                 const AttributeInit &initializer) const
 {
-  if (createable_ != Creatable) {
-    return false;
-  }
   CustomData *custom_data = custom_data_access_.get_custom_data(owner);
   if (custom_data == nullptr) {
     return false;
   }
 
   const int element_num = custom_data_access_.get_element_num(owner);
-  if (stored_as_named_attribute_) {
   if (CustomData_has_layer_named(custom_data, data_type_, name_)) {
     /* Exists already. */
     return false;
   }
-    return add_custom_data_layer_from_attribute_init(
-        name_, *custom_data, stored_type_, element_num, initializer);
+  if (add_custom_data_layer_from_attribute_init(
+          name_, *custom_data, data_type_, element_num, initializer))
+  {
+    if (initializer.type != AttributeInit::Type::Construct) {
+      /* Avoid calling update function when values are not initialized. In that case
+       * values must be set elsewhere anyway, which will cause a separate update tag. */
+      if (update_on_change_ != nullptr) {
+        update_on_change_(owner);
+      }
+    }
+    return true;
   }
-
-  if (CustomData_get_layer(custom_data, stored_type_) != nullptr) {
-    /* Exists already. */
   return false;
-  }
-  return add_builtin_type_custom_data_layer_from_init(
-      *custom_data, stored_type_, element_num, initializer);
 }
 
 bool BuiltinCustomDataLayerProvider::exists(const void *owner) const
@@ -534,10 +456,7 @@ bool BuiltinCustomDataLayerProvider::exists(const void *owner) const
   if (custom_data == nullptr) {
     return false;
   }
-  if (stored_as_named_attribute_) {
-    return CustomData_has_layer_named(custom_data, stored_type_, name_);
-  }
-  return CustomData_get_layer(custom_data, stored_type_) != nullptr;
+  return CustomData_has_layer_named(custom_data, data_type_, name_);
 }
 
 GAttributeReader CustomDataAttributeProvider::try_get_for_read(

source/blender/blenkernel/intern/attribute_access_intern.hh

@@ -31,14 +31,10 @@ struct CustomDataAccessInfo {
  * A #BuiltinAttributeProvider is responsible for exactly one attribute on a geometry component.
  * The attribute is identified by its name and has a fixed domain and type. Builtin attributes do
  * not follow the same loose rules as other attributes, because they are mapped to internal
- * "legacy" data structures. For example, some builtin attributes cannot be deleted. */
+ * "legacy" data structures. For example, some builtin attributes cannot be deleted.
+ */
 class BuiltinAttributeProvider {
  public:
-  /* Some utility enums to avoid hard to read booleans in function calls. */
-  enum CreatableEnum {
-    Creatable,
-    NonCreatable,
-  };
   enum DeletableEnum {
     Deletable,
     NonDeletable,
@@ -48,7 +44,6 @@ class BuiltinAttributeProvider {
   const std::string name_;
   const AttrDomain domain_;
   const eCustomDataType data_type_;
-  const CreatableEnum createable_;
   const DeletableEnum deletable_;
   const AttributeValidator validator_;
 
@@ -56,13 +51,11 @@ class BuiltinAttributeProvider {
   BuiltinAttributeProvider(std::string name,
                            const AttrDomain domain,
                            const eCustomDataType data_type,
-                           const CreatableEnum createable,
                            const DeletableEnum deletable,
                            AttributeValidator validator = {})
       : name_(std::move(name)),
         domain_(domain),
         data_type_(data_type),
-        createable_(createable),
         deletable_(deletable),
         validator_(validator)
   {
@@ -174,27 +167,21 @@ class CustomDataAttributeProvider final : public DynamicAttributesProvider {
  */
 class BuiltinCustomDataLayerProvider final : public BuiltinAttributeProvider {
   using UpdateOnChange = void (*)(void *owner);
-  const eCustomDataType stored_type_;
   const CustomDataAccessInfo custom_data_access_;
   const UpdateOnChange update_on_change_;
-  bool stored_as_named_attribute_;
 
  public:
   BuiltinCustomDataLayerProvider(std::string attribute_name,
                                  const AttrDomain domain,
-                                 const eCustomDataType attribute_type,
-                                 const eCustomDataType stored_type,
-                                 const CreatableEnum creatable,
+                                 const eCustomDataType data_type,
                                  const DeletableEnum deletable,
                                  const CustomDataAccessInfo custom_data_access,
-                                 const UpdateOnChange update_on_write,
+                                 const UpdateOnChange update_on_change,
                                  const AttributeValidator validator = {})
       : BuiltinAttributeProvider(
-            std::move(attribute_name), domain, attribute_type, creatable, deletable, validator),
-        stored_type_(stored_type),
+            std::move(attribute_name), domain, data_type, deletable, validator),
         custom_data_access_(custom_data_access),
-        update_on_change_(update_on_write),
-        stored_as_named_attribute_(data_type_ == stored_type_)
+        update_on_change_(update_on_change)
   {
   }
 

source/blender/blenkernel/intern/bake_items_serialize.cc

@@ -972,9 +972,14 @@ static std::shared_ptr<DictionaryValue> serialize_geometry_set(const GeometrySet
 
     auto io_references = io_instances->append_array("references");
     for (const InstanceReference &reference : instances.references()) {
-      BLI_assert(reference.type() == InstanceReference::Type::GeometrySet);
-      io_references->append(
-          serialize_geometry_set(reference.geometry_set(), blob_writer, blob_sharing));
+      if (reference.type() == InstanceReference::Type::GeometrySet) {
+        const GeometrySet &geometry = reference.geometry_set();
+        io_references->append(serialize_geometry_set(geometry, blob_writer, blob_sharing));
+      }
+      else {
+        /* TODO: Support serializing object and collection references. */
+        io_references->append(serialize_geometry_set({}, blob_writer, blob_sharing));
+      }
     }
 
     auto io_attributes = serialize_attributes(

source/blender/blenkernel/intern/curves_geometry.cc

@@ -58,6 +58,8 @@ CurvesGeometry::CurvesGeometry() : CurvesGeometry(0, 0) {}
 
 CurvesGeometry::CurvesGeometry(const int point_num, const int curve_num)
 {
+  this->runtime = MEM_new<CurvesGeometryRuntime>(__func__);
+
   this->point_num = point_num;
   this->curve_num = curve_num;
   CustomData_reset(&this->point_data);
@@ -67,8 +69,6 @@ CurvesGeometry::CurvesGeometry(const int point_num, const int curve_num)
   this->attributes_for_write().add<float3>(
       "position", AttrDomain::Point, AttributeInitConstruct());
 
-  this->runtime = MEM_new<CurvesGeometryRuntime>(__func__);
-
   if (curve_num > 0) {
     this->curve_offsets = static_cast<int *>(
         MEM_malloc_arrayN(this->curve_num + 1, sizeof(int), __func__));

source/blender/blenkernel/intern/editmesh.cc

@@ -91,7 +91,7 @@ void BKE_editmesh_looptris_calc_with_partial_ex(BMEditMesh *em,
                                                 const BMeshCalcTessellation_Params *params)
 {
   BLI_assert(em->looptris.size() == poly_to_tri_count(em->bm->totface, em->bm->totloop));
-  BLI_assert(!em->looptris.is_empty());
+  BLI_assert(!(em->bm->totface && em->looptris.is_empty()));
 
   BM_mesh_calc_tessellation_with_partial_ex(em->bm, em->looptris, bmpinfo, params);
 }

source/blender/blenkernel/intern/geometry_component_curves.cc

@@ -476,8 +476,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider position("position",
                                                  AttrDomain::Point,
                                                  CD_PROP_FLOAT3,
-                                                 CD_PROP_FLOAT3,
-                                                 BuiltinAttributeProvider::Creatable,
                                                  BuiltinAttributeProvider::NonDeletable,
                                                  point_access,
                                                  tag_component_positions_changed);
@@ -485,8 +483,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider radius("radius",
                                                AttrDomain::Point,
                                                CD_PROP_FLOAT,
-                                               CD_PROP_FLOAT,
-                                               BuiltinAttributeProvider::Creatable,
                                                BuiltinAttributeProvider::Deletable,
                                                point_access,
                                                tag_component_radii_changed);
@@ -494,8 +490,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider id("id",
                                            AttrDomain::Point,
                                            CD_PROP_INT32,
-                                           CD_PROP_INT32,
-                                           BuiltinAttributeProvider::Creatable,
                                            BuiltinAttributeProvider::Deletable,
                                            point_access,
                                            nullptr);
@@ -503,8 +497,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider tilt("tilt",
                                              AttrDomain::Point,
                                              CD_PROP_FLOAT,
-                                             CD_PROP_FLOAT,
-                                             BuiltinAttributeProvider::Creatable,
                                              BuiltinAttributeProvider::Deletable,
                                              point_access,
                                              tag_component_normals_changed);
@@ -512,8 +504,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider handle_right("handle_right",
                                                      AttrDomain::Point,
                                                      CD_PROP_FLOAT3,
-                                                     CD_PROP_FLOAT3,
-                                                     BuiltinAttributeProvider::Creatable,
                                                      BuiltinAttributeProvider::Deletable,
                                                      point_access,
                                                      tag_component_positions_changed);
@@ -521,8 +511,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider handle_left("handle_left",
                                                     AttrDomain::Point,
                                                     CD_PROP_FLOAT3,
-                                                    CD_PROP_FLOAT3,
-                                                    BuiltinAttributeProvider::Creatable,
                                                     BuiltinAttributeProvider::Deletable,
                                                     point_access,
                                                     tag_component_positions_changed);
@@ -536,8 +524,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider handle_type_right("handle_type_right",
                                                           AttrDomain::Point,
                                                           CD_PROP_INT8,
-                                                          CD_PROP_INT8,
-                                                          BuiltinAttributeProvider::Creatable,
                                                           BuiltinAttributeProvider::Deletable,
                                                           point_access,
                                                           tag_component_topology_changed,
@@ -546,8 +532,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider handle_type_left("handle_type_left",
                                                          AttrDomain::Point,
                                                          CD_PROP_INT8,
-                                                         CD_PROP_INT8,
-                                                         BuiltinAttributeProvider::Creatable,
                                                          BuiltinAttributeProvider::Deletable,
                                                          point_access,
                                                          tag_component_topology_changed,
@@ -556,21 +540,17 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider nurbs_weight("nurbs_weight",
                                                      AttrDomain::Point,
                                                      CD_PROP_FLOAT,
-                                                     CD_PROP_FLOAT,
-                                                     BuiltinAttributeProvider::Creatable,
                                                      BuiltinAttributeProvider::Deletable,
                                                      point_access,
                                                      tag_component_positions_changed);
 
   static const auto nurbs_order_clamp = mf::build::SI1_SO<int8_t, int8_t>(
       "NURBS Order Validate",
-      [](int8_t value) { return std::max<int8_t>(value, 0); },
+      [](int8_t value) { return std::max<int8_t>(value, 1); },
       mf::build::exec_presets::AllSpanOrSingle());
   static BuiltinCustomDataLayerProvider nurbs_order("nurbs_order",
                                                     AttrDomain::Curve,
                                                     CD_PROP_INT8,
-                                                    CD_PROP_INT8,
-                                                    BuiltinAttributeProvider::Creatable,
                                                     BuiltinAttributeProvider::Deletable,
                                                     curve_access,
                                                     tag_component_topology_changed,
@@ -585,8 +565,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider normal_mode("normal_mode",
                                                     AttrDomain::Curve,
                                                     CD_PROP_INT8,
-                                                    CD_PROP_INT8,
-                                                    BuiltinAttributeProvider::Creatable,
                                                     BuiltinAttributeProvider::Deletable,
                                                     curve_access,
                                                     tag_component_normals_changed,
@@ -595,8 +573,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider custom_normal("custom_normal",
                                                       AttrDomain::Point,
                                                       CD_PROP_FLOAT3,
-                                                      CD_PROP_FLOAT3,
-                                                      BuiltinAttributeProvider::Creatable,
                                                       BuiltinAttributeProvider::Deletable,
                                                       point_access,
                                                       tag_component_normals_changed);
@@ -610,8 +586,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider nurbs_knots_mode("knots_mode",
                                                          AttrDomain::Curve,
                                                          CD_PROP_INT8,
-                                                         CD_PROP_INT8,
-                                                         BuiltinAttributeProvider::Creatable,
                                                          BuiltinAttributeProvider::Deletable,
                                                          curve_access,
                                                          tag_component_topology_changed,
@@ -626,8 +600,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider curve_type("curve_type",
                                                    AttrDomain::Curve,
                                                    CD_PROP_INT8,
-                                                   CD_PROP_INT8,
-                                                   BuiltinAttributeProvider::Creatable,
                                                    BuiltinAttributeProvider::Deletable,
                                                    curve_access,
                                                    tag_component_curve_types_changed,
@@ -640,8 +612,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider resolution("resolution",
                                                    AttrDomain::Curve,
                                                    CD_PROP_INT32,
-                                                   CD_PROP_INT32,
-                                                   BuiltinAttributeProvider::Creatable,
                                                    BuiltinAttributeProvider::Deletable,
                                                    curve_access,
                                                    tag_component_topology_changed,
@@ -650,8 +620,6 @@ static ComponentAttributeProviders create_attribute_providers_for_curve()
   static BuiltinCustomDataLayerProvider cyclic("cyclic",
                                                AttrDomain::Curve,
                                                CD_PROP_BOOL,
-                                               CD_PROP_BOOL,
-                                               BuiltinAttributeProvider::Creatable,
                                                BuiltinAttributeProvider::Deletable,
                                                curve_access,
                                                tag_component_topology_changed);

source/blender/blenkernel/intern/geometry_component_instances.cc

@@ -133,8 +133,6 @@ static ComponentAttributeProviders create_attribute_providers_for_instances()
   static BuiltinCustomDataLayerProvider id("id",
                                            AttrDomain::Instance,
                                            CD_PROP_INT32,
-                                           CD_PROP_INT32,
-                                           BuiltinAttributeProvider::Creatable,
                                            BuiltinAttributeProvider::Deletable,
                                            instance_custom_data_access,
                                            nullptr);
@@ -142,8 +140,6 @@ static ComponentAttributeProviders create_attribute_providers_for_instances()
   static BuiltinCustomDataLayerProvider instance_transform("instance_transform",
                                                            AttrDomain::Instance,
                                                            CD_PROP_FLOAT4X4,
-                                                           CD_PROP_FLOAT4X4,
-                                                           BuiltinAttributeProvider::Creatable,
                                                            BuiltinAttributeProvider::NonDeletable,
                                                            instance_custom_data_access,
                                                            nullptr);
@@ -152,8 +148,6 @@ static ComponentAttributeProviders create_attribute_providers_for_instances()
   static BuiltinCustomDataLayerProvider reference_index(".reference_index",
                                                         AttrDomain::Instance,
                                                         CD_PROP_INT32,
-                                                        CD_PROP_INT32,
-                                                        BuiltinAttributeProvider::Creatable,
                                                         BuiltinAttributeProvider::NonDeletable,
                                                         instance_custom_data_access,
                                                         tag_component_reference_index_changed);

source/blender/blenkernel/intern/geometry_component_mesh.cc

@@ -1010,8 +1010,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider position("position",
                                                  AttrDomain::Point,
                                                  CD_PROP_FLOAT3,
-                                                 CD_PROP_FLOAT3,
-                                                 BuiltinAttributeProvider::Creatable,
                                                  BuiltinAttributeProvider::NonDeletable,
                                                  point_access,
                                                  tag_component_positions_changed);
@@ -1019,8 +1017,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider id("id",
                                            AttrDomain::Point,
                                            CD_PROP_INT32,
-                                           CD_PROP_INT32,
-                                           BuiltinAttributeProvider::Creatable,
                                            BuiltinAttributeProvider::Deletable,
                                            point_access,
                                            nullptr);
@@ -1035,8 +1031,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider material_index("material_index",
                                                        AttrDomain::Face,
                                                        CD_PROP_INT32,
-                                                       CD_PROP_INT32,
-                                                       BuiltinAttributeProvider::Creatable,
                                                        BuiltinAttributeProvider::Deletable,
                                                        face_access,
                                                        nullptr,
@@ -1049,8 +1043,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider edge_verts(".edge_verts",
                                                    AttrDomain::Edge,
                                                    CD_PROP_INT32_2D,
-                                                   CD_PROP_INT32_2D,
-                                                   BuiltinAttributeProvider::Creatable,
                                                    BuiltinAttributeProvider::NonDeletable,
                                                    edge_access,
                                                    nullptr,
@@ -1065,8 +1057,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider corner_vert(".corner_vert",
                                                     AttrDomain::Corner,
                                                     CD_PROP_INT32,
-                                                    CD_PROP_INT32,
-                                                    BuiltinAttributeProvider::Creatable,
                                                     BuiltinAttributeProvider::NonDeletable,
                                                     corner_access,
                                                     nullptr,
@@ -1074,8 +1064,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider corner_edge(".corner_edge",
                                                     AttrDomain::Corner,
                                                     CD_PROP_INT32,
-                                                    CD_PROP_INT32,
-                                                    BuiltinAttributeProvider::Creatable,
                                                     BuiltinAttributeProvider::NonDeletable,
                                                     corner_access,
                                                     nullptr,
@@ -1084,8 +1072,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider sharp_face("sharp_face",
                                                    AttrDomain::Face,
                                                    CD_PROP_BOOL,
-                                                   CD_PROP_BOOL,
-                                                   BuiltinAttributeProvider::Creatable,
                                                    BuiltinAttributeProvider::Deletable,
                                                    face_access,
                                                    tag_component_sharpness_changed);
@@ -1093,8 +1079,6 @@ static ComponentAttributeProviders create_attribute_providers_for_mesh()
   static BuiltinCustomDataLayerProvider sharp_edge("sharp_edge",
                                                    AttrDomain::Edge,
                                                    CD_PROP_BOOL,
-                                                   CD_PROP_BOOL,
-                                                   BuiltinAttributeProvider::Creatable,
                                                    BuiltinAttributeProvider::Deletable,
                                                    edge_access,
                                                    tag_component_sharpness_changed);

source/blender/blenkernel/intern/geometry_component_pointcloud.cc

@@ -147,24 +147,18 @@ static ComponentAttributeProviders create_attribute_providers_for_point_cloud()
   static BuiltinCustomDataLayerProvider position("position",
                                                  AttrDomain::Point,
                                                  CD_PROP_FLOAT3,
-                                                 CD_PROP_FLOAT3,
-                                                 BuiltinAttributeProvider::Creatable,
                                                  BuiltinAttributeProvider::NonDeletable,
                                                  point_access,
                                                  tag_component_positions_changed);
   static BuiltinCustomDataLayerProvider radius("radius",
                                                AttrDomain::Point,
                                                CD_PROP_FLOAT,
-                                               CD_PROP_FLOAT,
-                                               BuiltinAttributeProvider::Creatable,
                                                BuiltinAttributeProvider::Deletable,
                                                point_access,
                                                tag_component_radius_changed);
   static BuiltinCustomDataLayerProvider id("id",
                                            AttrDomain::Point,
                                            CD_PROP_INT32,
-                                           CD_PROP_INT32,
-                                           BuiltinAttributeProvider::Creatable,
                                            BuiltinAttributeProvider::Deletable,
                                            point_access,
                                            nullptr);

source/blender/blenkernel/intern/pointcloud.cc

@@ -62,11 +62,11 @@ static void pointcloud_init_data(ID *id)
 
   MEMCPY_STRUCT_AFTER(pointcloud, DNA_struct_default_get(PointCloud), id);
 
+  pointcloud->runtime = new blender::bke::PointCloudRuntime();
+
   CustomData_reset(&pointcloud->pdata);
   pointcloud->attributes_for_write().add<float3>(
       "position", blender::bke::AttrDomain::Point, blender::bke::AttributeInitConstruct());
-
-  pointcloud->runtime = new blender::bke::PointCloudRuntime();
 }
 
 static void pointcloud_copy_data(Main * /*bmain*/,

source/blender/compositor/CMakeLists.txt

@@ -584,20 +584,6 @@ if(WITH_COMPOSITOR_CPU)
     ${CMAKE_CURRENT_BINARY_DIR}/operations
   )
 
-  set(GENSRC_DIR ${CMAKE_CURRENT_BINARY_DIR}/operations)
-  set(GENSRC ${GENSRC_DIR}/COM_SMAAAreaTexture.h)
-  add_custom_command(
-    OUTPUT ${GENSRC}
-    COMMAND ${CMAKE_COMMAND} -E make_directory ${GENSRC_DIR}
-    COMMAND "$<TARGET_FILE:smaa_areatex>" ${GENSRC}
-    DEPENDS smaa_areatex
-  )
-  list(APPEND SRC
-    ${GENSRC}
-  )
-  unset(GENSRC)
-  unset(GENSRC_DIR)
-
   if(WITH_OPENIMAGEDENOISE)
     add_definitions(-DWITH_OPENIMAGEDENOISE)
     add_definitions(-DOIDN_STATIC_LIB)

source/blender/compositor/nodes/COM_AntiAliasingNode.cc

@@ -7,37 +7,41 @@
 
 namespace blender::compositor {
 
+/* Blender encodes the threshold in the [0, 1] range, while the SMAA algorithm expects it in
+ * the [0, 0.5] range. */
+static float get_threshold(const NodeAntiAliasingData *data)
+{
+  return data->threshold / 2.0f;
+}
+
+/* Blender encodes the local contrast adaptation factor in the [0, 1] range, while the SMAA
+ * algorithm expects it in the [0, 10] range. */
+static float get_local_contrast_adaptation_factor(const NodeAntiAliasingData *data)
+{
+  return data->contrast_limit * 10.0f;
+}
+
+/* Blender encodes the corner rounding factor in the float [0, 1] range, while the SMAA algorithm
+ * expects it in the integer [0, 100] range. */
+static int get_corner_rounding(const NodeAntiAliasingData *data)
+{
+  return int(data->corner_rounding * 100.0f);
+}
+
 void AntiAliasingNode::convert_to_operations(NodeConverter &converter,
                                              const CompositorContext & /*context*/) const
 {
   const bNode *node = this->get_bnode();
   const NodeAntiAliasingData *data = (const NodeAntiAliasingData *)node->storage;
 
-  /* Edge Detection (First Pass) */
-  SMAAEdgeDetectionOperation *operation1 = nullptr;
+  SMAAOperation *operation = new SMAAOperation();
+  operation->set_threshold(get_threshold(data));
+  operation->set_local_contrast_adaptation_factor(get_local_contrast_adaptation_factor(data));
+  operation->set_corner_rounding(get_corner_rounding(data));
+  converter.add_operation(operation);
 
-  operation1 = new SMAAEdgeDetectionOperation();
-  operation1->set_threshold(data->threshold);
-  operation1->set_local_contrast_adaptation_factor(data->contrast_limit);
-  converter.add_operation(operation1);
-
-  converter.map_input_socket(get_input_socket(0), operation1->get_input_socket(0));
-
-  /* Blending Weight Calculation Pixel Shader (Second Pass) */
-  SMAABlendingWeightCalculationOperation *operation2 =
-      new SMAABlendingWeightCalculationOperation();
-  operation2->set_corner_rounding(data->corner_rounding);
-  converter.add_operation(operation2);
-
-  converter.add_link(operation1->get_output_socket(), operation2->get_input_socket(0));
-
-  /* Neighborhood Blending Pixel Shader (Third Pass) */
-  SMAANeighborhoodBlendingOperation *operation3 = new SMAANeighborhoodBlendingOperation();
-  converter.add_operation(operation3);
-
-  converter.map_input_socket(get_input_socket(0), operation3->get_input_socket(0));
-  converter.add_link(operation2->get_output_socket(), operation3->get_input_socket(1));
-  converter.map_output_socket(get_output_socket(0), operation3->get_output_socket());
+  converter.map_input_socket(get_input_socket(0), operation->get_input_socket(0));
+  converter.map_output_socket(get_output_socket(0), operation->get_output_socket());
 }
 
 }  // namespace blender::compositor

source/blender/compositor/nodes/COM_CornerPinNode.cc

@@ -18,28 +18,13 @@ void CornerPinNode::convert_to_operations(NodeConverter &converter,
   PlaneCornerPinMaskOperation *plane_mask_operation = new PlaneCornerPinMaskOperation();
   converter.add_operation(plane_mask_operation);
 
-  SMAAEdgeDetectionOperation *smaa_edge_detection = new SMAAEdgeDetectionOperation();
-  converter.add_operation(smaa_edge_detection);
+  SMAAOperation *smaa_operation = new SMAAOperation();
+  converter.add_operation(smaa_operation);
 
   converter.add_link(plane_mask_operation->get_output_socket(),
-                     smaa_edge_detection->get_input_socket(0));
+                     smaa_operation->get_input_socket(0));
 
-  SMAABlendingWeightCalculationOperation *smaa_blending_weights =
-      new SMAABlendingWeightCalculationOperation();
-  converter.add_operation(smaa_blending_weights);
-
-  converter.add_link(smaa_edge_detection->get_output_socket(),
-                     smaa_blending_weights->get_input_socket(0));
-
-  SMAANeighborhoodBlendingOperation *smaa_neighborhood = new SMAANeighborhoodBlendingOperation();
-  converter.add_operation(smaa_neighborhood);
-
-  converter.add_link(plane_mask_operation->get_output_socket(),
-                     smaa_neighborhood->get_input_socket(0));
-  converter.add_link(smaa_blending_weights->get_output_socket(),
-                     smaa_neighborhood->get_input_socket(1));
-
-  converter.map_output_socket(this->get_output_socket(1), smaa_neighborhood->get_output_socket());
+  converter.map_output_socket(this->get_output_socket(1), smaa_operation->get_output_socket());
 
   PlaneCornerPinWarpImageOperation *warp_image_operation = new PlaneCornerPinWarpImageOperation();
   converter.add_operation(warp_image_operation);
@@ -62,7 +47,7 @@ void CornerPinNode::convert_to_operations(NodeConverter &converter,
   converter.add_operation(set_alpha_operation);
   converter.add_link(warp_image_operation->get_output_socket(),
                      set_alpha_operation->get_input_socket(0));
-  converter.add_link(smaa_neighborhood->get_output_socket(),
+  converter.add_link(smaa_operation->get_output_socket(),
                      set_alpha_operation->get_input_socket(1));
   converter.map_output_socket(this->get_output_socket(0),
                               set_alpha_operation->get_output_socket());

source/blender/compositor/nodes/COM_DilateErodeNode.cc

@@ -37,26 +37,10 @@ void DilateErodeNode::convert_to_operations(NodeConverter &converter,
     converter.map_input_socket(get_input_socket(0), operation->get_input_socket(0));
 
     if (editor_node->custom3 < 2.0f) {
-      SMAAEdgeDetectionOperation *smaa_edge_detection = new SMAAEdgeDetectionOperation();
-      converter.add_operation(smaa_edge_detection);
-
-      converter.add_link(operation->get_output_socket(), smaa_edge_detection->get_input_socket(0));
-
-      SMAABlendingWeightCalculationOperation *smaa_blending_weights =
-          new SMAABlendingWeightCalculationOperation();
-      converter.add_operation(smaa_blending_weights);
-
-      converter.add_link(smaa_edge_detection->get_output_socket(),
-                         smaa_blending_weights->get_input_socket(0));
-
-      SMAANeighborhoodBlendingOperation *smaa_neighborhood =
-          new SMAANeighborhoodBlendingOperation();
-      converter.add_operation(smaa_neighborhood);
-
-      converter.add_link(operation->get_output_socket(), smaa_neighborhood->get_input_socket(0));
-      converter.add_link(smaa_blending_weights->get_output_socket(),
-                         smaa_neighborhood->get_input_socket(1));
-      converter.map_output_socket(get_output_socket(0), smaa_neighborhood->get_output_socket());
+      SMAAOperation *smaa_operation = new SMAAOperation();
+      converter.add_operation(smaa_operation);
+      converter.add_link(operation->get_output_socket(), smaa_operation->get_input_socket(0));
+      converter.map_output_socket(get_output_socket(0), smaa_operation->get_output_socket());
     }
     else {
       converter.map_output_socket(get_output_socket(0), operation->get_output_socket(0));

source/blender/compositor/nodes/COM_IDMaskNode.cc

@@ -27,27 +27,10 @@ void IDMaskNode::convert_to_operations(NodeConverter &converter,
     converter.map_output_socket(get_output_socket(0), operation->get_output_socket(0));
   }
   else {
-    SMAAEdgeDetectionOperation *operation1 = nullptr;
-
-    operation1 = new SMAAEdgeDetectionOperation();
-    converter.add_operation(operation1);
-
-    converter.add_link(operation->get_output_socket(0), operation1->get_input_socket(0));
-
-    /* Blending Weight Calculation Pixel Shader (Second Pass). */
-    SMAABlendingWeightCalculationOperation *operation2 =
-        new SMAABlendingWeightCalculationOperation();
-    converter.add_operation(operation2);
-
-    converter.add_link(operation1->get_output_socket(), operation2->get_input_socket(0));
-
-    /* Neighborhood Blending Pixel Shader (Third Pass). */
-    SMAANeighborhoodBlendingOperation *operation3 = new SMAANeighborhoodBlendingOperation();
-    converter.add_operation(operation3);
-
-    converter.add_link(operation->get_output_socket(0), operation3->get_input_socket(0));
-    converter.add_link(operation2->get_output_socket(), operation3->get_input_socket(1));
-    converter.map_output_socket(get_output_socket(0), operation3->get_output_socket());
+    SMAAOperation *smaa_operation = new SMAAOperation();
+    converter.add_operation(smaa_operation);
+    converter.add_link(operation->get_output_socket(0), smaa_operation->get_input_socket(0));
+    converter.map_output_socket(get_output_socket(0), smaa_operation->get_output_socket());
   }
 }
 

source/blender/compositor/nodes/COM_PlaneTrackDeformNode.cc

@@ -35,28 +35,13 @@ void PlaneTrackDeformNode::convert_to_operations(NodeConverter &converter,
   }
   converter.add_operation(plane_mask_operation);
 
-  SMAAEdgeDetectionOperation *smaa_edge_detection = new SMAAEdgeDetectionOperation();
-  converter.add_operation(smaa_edge_detection);
+  SMAAOperation *smaa_operation = new SMAAOperation();
+  converter.add_operation(smaa_operation);
 
   converter.add_link(plane_mask_operation->get_output_socket(),
-                     smaa_edge_detection->get_input_socket(0));
+                     smaa_operation->get_input_socket(0));
 
-  SMAABlendingWeightCalculationOperation *smaa_blending_weights =
-      new SMAABlendingWeightCalculationOperation();
-  converter.add_operation(smaa_blending_weights);
-
-  converter.add_link(smaa_edge_detection->get_output_socket(),
-                     smaa_blending_weights->get_input_socket(0));
-
-  SMAANeighborhoodBlendingOperation *smaa_neighborhood = new SMAANeighborhoodBlendingOperation();
-  converter.add_operation(smaa_neighborhood);
-
-  converter.add_link(plane_mask_operation->get_output_socket(),
-                     smaa_neighborhood->get_input_socket(0));
-  converter.add_link(smaa_blending_weights->get_output_socket(),
-                     smaa_neighborhood->get_input_socket(1));
-
-  converter.map_output_socket(this->get_output_socket(1), smaa_neighborhood->get_output_socket());
+  converter.map_output_socket(this->get_output_socket(1), smaa_operation->get_output_socket());
 
   PlaneTrackWarpImageOperation *warp_image_operation = new PlaneTrackWarpImageOperation();
   warp_image_operation->set_movie_clip(clip);
@@ -75,7 +60,7 @@ void PlaneTrackDeformNode::convert_to_operations(NodeConverter &converter,
   converter.add_operation(set_alpha_operation);
   converter.add_link(warp_image_operation->get_output_socket(),
                      set_alpha_operation->get_input_socket(0));
-  converter.add_link(smaa_neighborhood->get_output_socket(),
+  converter.add_link(smaa_operation->get_output_socket(),
                      set_alpha_operation->get_input_socket(1));
   converter.map_output_socket(this->get_output_socket(0),
                               set_alpha_operation->get_output_socket());

source/blender/compositor/nodes/COM_ZCombineNode.cc

@@ -54,25 +54,10 @@ void ZCombineNode::convert_to_operations(NodeConverter &converter,
     converter.map_input_socket(get_input_socket(3), maskoperation->get_input_socket(1));
 
     /* Step 2 anti alias mask bit of an expensive operation, but does the trick. */
-    SMAAEdgeDetectionOperation *smaa_edge_detection = new SMAAEdgeDetectionOperation();
-    converter.add_operation(smaa_edge_detection);
+    SMAAOperation *smaa_operation = new SMAAOperation();
+    converter.add_operation(smaa_operation);
 
-    converter.add_link(maskoperation->get_output_socket(),
-                       smaa_edge_detection->get_input_socket(0));
-
-    SMAABlendingWeightCalculationOperation *smaa_blending_weights =
-        new SMAABlendingWeightCalculationOperation();
-    converter.add_operation(smaa_blending_weights);
-
-    converter.add_link(smaa_edge_detection->get_output_socket(),
-                       smaa_blending_weights->get_input_socket(0));
-
-    SMAANeighborhoodBlendingOperation *smaa_neighborhood = new SMAANeighborhoodBlendingOperation();
-    converter.add_operation(smaa_neighborhood);
-
-    converter.add_link(maskoperation->get_output_socket(), smaa_neighborhood->get_input_socket(0));
-    converter.add_link(smaa_blending_weights->get_output_socket(),
-                       smaa_neighborhood->get_input_socket(1));
+    converter.add_link(maskoperation->get_output_socket(), smaa_operation->get_input_socket(0));
 
     /* use mask to blend between the input colors. */
     ZCombineMaskOperation *zcombineoperation = this->get_bnode()->custom1 ?
@@ -80,7 +65,7 @@ void ZCombineNode::convert_to_operations(NodeConverter &converter,
                                                    new ZCombineMaskOperation();
     converter.add_operation(zcombineoperation);
 
-    converter.add_link(smaa_neighborhood->get_output_socket(),
+    converter.add_link(smaa_operation->get_output_socket(),
                        zcombineoperation->get_input_socket(0));
     converter.map_input_socket(get_input_socket(0), zcombineoperation->get_input_socket(1));
     converter.map_input_socket(get_input_socket(2), zcombineoperation->get_input_socket(2));

source/blender/compositor/operations/COM_SMAAOperation.h

@@ -4,87 +4,36 @@
 
 #pragma once
 
-#include "COM_MultiThreadedOperation.h"
+#include "COM_NodeOperation.h"
 
 namespace blender::compositor {
 
-/*-----------------------------------------------------------------------------*/
-/* Edge Detection (First Pass) */
-
-class SMAAEdgeDetectionOperation : public MultiThreadedOperation {
+class SMAAOperation : public NodeOperation {
  protected:
-  float threshold_;
-  float contrast_limit_;
+  float threshold_ = 0.1f;
+  float local_contrast_adaptation_factor_ = 2.0f;
+  int corner_rounding_ = 25;
 
  public:
-  SMAAEdgeDetectionOperation();
+  SMAAOperation();
 
-  void set_threshold(float threshold);
+  void set_threshold(float threshold)
+  {
+    threshold_ = threshold;
+  }
 
-  void set_local_contrast_adaptation_factor(float factor);
+  void set_local_contrast_adaptation_factor(float factor)
+  {
+    local_contrast_adaptation_factor_ = factor;
+  }
+
+  void set_corner_rounding(int corner_rounding)
+  {
+    corner_rounding_ = corner_rounding;
+  }
 
   void get_area_of_interest(int input_idx, const rcti &output_area, rcti &r_input_area) override;
-  void update_memory_buffer_partial(MemoryBuffer *output,
-                                    const rcti &area,
-                                    Span<MemoryBuffer *> inputs) override;
-};
-
-/*-----------------------------------------------------------------------------*/
-/*  Blending Weight Calculation (Second Pass) */
-
-class SMAABlendingWeightCalculationOperation : public MultiThreadedOperation {
- private:
-  std::function<void(int x, int y, float *out)> sample_image_fn_;
-  int corner_rounding_;
-
- public:
-  SMAABlendingWeightCalculationOperation();
-
-  void set_corner_rounding(float rounding);
-
-  void get_area_of_interest(int input_idx, const rcti &output_area, rcti &r_input_area) override;
-  void update_memory_buffer_started(MemoryBuffer *output,
-                                    const rcti &area,
-                                    Span<MemoryBuffer *> inputs) override;
-  void update_memory_buffer_partial(MemoryBuffer *output,
-                                    const rcti &area,
-                                    Span<MemoryBuffer *> inputs) override;
-
- private:
-  /* Diagonal Search Functions */
-  /**
-   * These functions allows to perform diagonal pattern searches.
-   */
-  int search_diag1(int x, int y, int dir, bool *r_found);
-  int search_diag2(int x, int y, int dir, bool *r_found);
-  /**
-   * This searches for diagonal patterns and returns the corresponding weights.
-   */
-  void calculate_diag_weights(int x, int y, const float edges[2], float weights[2]);
-  bool is_vertical_search_unneeded(int x, int y);
-
-  /* Horizontal/Vertical Search Functions */
-  int search_xleft(int x, int y);
-  int search_xright(int x, int y);
-  int search_yup(int x, int y);
-  int search_ydown(int x, int y);
-
-  /*  Corner Detection Functions */
-  void detect_horizontal_corner_pattern(
-      float weights[2], int left, int right, int y, int d1, int d2);
-  void detect_vertical_corner_pattern(
-      float weights[2], int x, int top, int bottom, int d1, int d2);
-};
-
-/*-----------------------------------------------------------------------------*/
-/* Neighborhood Blending (Third Pass) */
-
-class SMAANeighborhoodBlendingOperation : public MultiThreadedOperation {
- public:
-  SMAANeighborhoodBlendingOperation();
-
-  void get_area_of_interest(int input_idx, const rcti &output_area, rcti &r_input_area) override;
-  void update_memory_buffer_partial(MemoryBuffer *output,
+  void update_memory_buffer(MemoryBuffer *output,
                             const rcti &area,
                             Span<MemoryBuffer *> inputs) override;
 };

source/blender/compositor/operations/COM_TranslateOperation.h

@@ -7,6 +7,8 @@
 #include "COM_ConstantOperation.h"
 #include "COM_MultiThreadedOperation.h"
 
+#include <mutex>
+
 namespace blender::compositor {
 
 class TranslateOperation : public MultiThreadedOperation {
@@ -21,6 +23,8 @@ class TranslateOperation : public MultiThreadedOperation {
   bool is_delta_set_;
   bool is_relative_;
 
+  std::mutex mutex_;
+
  protected:
   MemoryBufferExtend x_extend_mode_;
   MemoryBufferExtend y_extend_mode_;
@@ -50,6 +54,11 @@ class TranslateOperation : public MultiThreadedOperation {
   inline void ensure_delta()
   {
     if (!is_delta_set_) {
+      std::unique_lock lock(mutex_);
+      if (is_delta_set_) {
+        return;
+      }
+
       delta_x_ = get_input_operation(X_INPUT_INDEX)->get_constant_value_default(0.0f);
       delta_y_ = get_input_operation(Y_INPUT_INDEX)->get_constant_value_default(0.0f);
       if (get_is_relative()) {

source/blender/compositor/realtime_compositor/cached_resources/intern/cached_image.cc

@@ -15,6 +15,7 @@
 #include "GPU_shader.hh"
 #include "GPU_texture.hh"
 
+#include "IMB_colormanagement.hh"
 #include "IMB_imbuf.hh"
 #include "IMB_imbuf_types.hh"
 
@@ -56,74 +57,6 @@ bool operator==(const CachedImageKey &a, const CachedImageKey &b)
  * Cached Image.
  */
 
-/* Returns a new texture of the given format and precision preprocessed using the given shader. The
- * input texture is freed. */
-static GPUTexture *preprocess_texture(Context &context,
-                                      GPUTexture *input_texture,
-                                      eGPUTextureFormat target_format,
-                                      ResultPrecision precision,
-                                      const char *shader_name)
-{
-  const int2 size = int2(GPU_texture_width(input_texture), GPU_texture_height(input_texture));
-
-  GPUTexture *preprocessed_texture = GPU_texture_create_2d(
-      "Cached Image", size.x, size.y, 1, target_format, GPU_TEXTURE_USAGE_GENERAL, nullptr);
-
-  GPUShader *shader = context.get_shader(shader_name, precision);
-  GPU_shader_bind(shader);
-
-  const int input_unit = GPU_shader_get_sampler_binding(shader, "input_tx");
-  GPU_texture_bind(input_texture, input_unit);
-
-  const int image_unit = GPU_shader_get_sampler_binding(shader, "output_img");
-  GPU_texture_image_bind(preprocessed_texture, image_unit);
-
-  compute_dispatch_threads_at_least(shader, size);
-
-  GPU_shader_unbind();
-  GPU_texture_unbind(input_texture);
-  GPU_texture_image_unbind(preprocessed_texture);
-  GPU_texture_free(input_texture);
-
-  return preprocessed_texture;
-}
-
-/* Compositor images are expected to be always pre-multiplied, so identify if the GPU texture
- * returned by the IMB module is straight and needs to be pre-multiplied. An exception is when
- * the image has an alpha mode of channel packed or alpha ignore, in which case, we always ignore
- * pre-multiplication. */
-static bool should_premultiply_alpha(Image *image, ImBuf *image_buffer)
-{
-  if (ELEM(image->alpha_mode, IMA_ALPHA_CHANNEL_PACKED, IMA_ALPHA_IGNORE)) {
-    return false;
-  }
-
-  return !BKE_image_has_gpu_texture_premultiplied_alpha(image, image_buffer);
-}
-
-/* Get a suitable texture format supported by the compositor given the format of the texture
- * returned by the IMB module. See imb_gpu_get_format for the formats that needs to be handled. */
-static eGPUTextureFormat get_compatible_texture_format(eGPUTextureFormat original_format)
-{
-  switch (original_format) {
-    case GPU_R16F:
-    case GPU_R32F:
-    case GPU_RGBA16F:
-    case GPU_RGBA32F:
-      return original_format;
-    case GPU_R8:
-      return GPU_R16F;
-    case GPU_RGBA8:
-    case GPU_SRGB8_A8:
-      return GPU_RGBA16F;
-    default:
-      break;
-  }
-
-  BLI_assert_unreachable();
-  return original_format;
-}
-
 /* Get the selected render layer selected assuming the image is a multilayer image. */
 static RenderLayer *get_render_layer(Image *image, ImageUser &image_user)
 {
@@ -205,6 +138,56 @@ static ImageUser compute_image_user_for_pass(Context &context,
   return image_user_for_pass;
 }
 
+/* The image buffer might be stored as an sRGB 8-bit image, while the compositor expects linear
+ * float images, so compute a linear float buffer for the image buffer. This will also do linear
+ * space conversion and alpha pre-multiplication as needed. We could store those images in sRGB GPU
+ * textures and let the GPU do the linear space conversion, but the issues is that we don't control
+ * how the GPU does the conversion and so we get tiny differences across CPU and GPU compositing,
+ * and potentially even across GPUs/Drivers. Further, if alpha pre-multiplication is needed, we
+ * would need to do it ourself, which means alpha pre-multiplication will happen before linear
+ * space conversion, which would produce yet another difference. So we just do everything on the
+ * CPU, since this is already a cached resource.
+ *
+ * To avoid conflicts with other threads, create a new image buffer and assign all the necessary
+ * information to it, with IB_DO_NOT_TAKE_OWNERSHIP for buffers since a deep copy is not needed.
+ *
+ * The caller should free the returned image buffer. */
+static ImBuf *compute_linear_buffer(ImBuf *image_buffer)
+{
+  /* Do not pass the flags to the allocation function to avoid buffer allocation, but assign them
+   * after to retain important information like precision and alpha mode. */
+  ImBuf *linear_image_buffer = IMB_allocImBuf(
+      image_buffer->x, image_buffer->y, image_buffer->planes, 0);
+  linear_image_buffer->flags = image_buffer->flags;
+
+  /* Assign the float buffer if it exists, as well as its number of channels. */
+  IMB_assign_float_buffer(
+      linear_image_buffer, image_buffer->float_buffer, IB_DO_NOT_TAKE_OWNERSHIP);
+  linear_image_buffer->channels = image_buffer->channels;
+
+  /* If no float buffer exists, assign it then compute a float buffer from it. This is the main
+   * call of this function. */
+  if (!linear_image_buffer->float_buffer.data) {
+    IMB_assign_byte_buffer(
+        linear_image_buffer, image_buffer->byte_buffer, IB_DO_NOT_TAKE_OWNERSHIP);
+    IMB_float_from_rect(linear_image_buffer);
+  }
+
+  /* If the image buffer contained compressed data, assign them as well, but only if the color
+   * space of the buffer is linear or data, since we need linear data and can't preprocess the
+   * compressed buffer. If not, we fallback to the float buffer already assigned, which is
+   * guaranteed to exist as a fallback for compressed textures. */
+  const bool is_suitable_compressed_color_space =
+      IMB_colormanagement_space_is_data(image_buffer->byte_buffer.colorspace) ||
+      IMB_colormanagement_space_is_scene_linear(image_buffer->byte_buffer.colorspace);
+  if (image_buffer->ftype == IMB_FTYPE_DDS && is_suitable_compressed_color_space) {
+    linear_image_buffer->ftype = IMB_FTYPE_DDS;
+    IMB_assign_dds_data(linear_image_buffer, image_buffer->dds_data, IB_DO_NOT_TAKE_OWNERSHIP);
+  }
+
+  return linear_image_buffer;
+}
+
 CachedImage::CachedImage(Context &context,
                          Image *image,
                          ImageUser *image_user,
@@ -227,34 +210,12 @@ CachedImage::CachedImage(Context &context,
       context, image, image_user, pass_name);
 
   ImBuf *image_buffer = BKE_image_acquire_ibuf(image, &image_user_for_pass, nullptr);
-  const bool is_premultiplied = BKE_image_has_gpu_texture_premultiplied_alpha(image, image_buffer);
-  texture_ = IMB_create_gpu_texture("Image Texture", image_buffer, true, is_premultiplied);
+  ImBuf *linear_image_buffer = compute_linear_buffer(image_buffer);
+
+  texture_ = IMB_create_gpu_texture("Image Texture", linear_image_buffer, true, true);
   GPU_texture_update_mipmap_chain(texture_);
 
-  const eGPUTextureFormat original_format = GPU_texture_format(texture_);
-  const eGPUTextureFormat target_format = get_compatible_texture_format(original_format);
-  const ResultType result_type = Result::type(target_format);
-  const ResultPrecision precision = Result::precision(target_format);
-
-  /* The GPU image returned by the IMB module can be in a format not supported by the compositor,
-   * or it might need pre-multiplication, so preprocess them first. */
-  if (result_type == ResultType::Color && should_premultiply_alpha(image, image_buffer)) {
-    texture_ = preprocess_texture(
-        context, texture_, target_format, precision, "compositor_premultiply_alpha");
-  }
-  else if (original_format != target_format) {
-    const char *conversion_shader_name = result_type == ResultType::Float ?
-                                             "compositor_convert_float_to_float" :
-                                             "compositor_convert_color_to_color";
-    texture_ = preprocess_texture(
-        context, texture_, target_format, precision, conversion_shader_name);
-  }
-
-  /* Set the alpha to 1 using swizzling if alpha is ignored. */
-  if (result_type == ResultType::Color && image->alpha_mode == IMA_ALPHA_IGNORE) {
-    GPU_texture_swizzle_set(texture_, "rgb1");
-  }
-
+  IMB_freeImBuf(linear_image_buffer);
   BKE_image_release_ibuf(image, image_buffer, nullptr);
 }
 

source/blender/editors/animation/anim_filter.cc

@@ -3674,11 +3674,11 @@ size_t ANIM_animdata_filter(bAnimContext *ac,
                             void *data,
                             eAnimCont_Types datatype)
 {
-  size_t items = 0;
+  if (!data || !anim_data) {
+    return 0;
+  }
 
-  /* only filter data if there's somewhere to put it */
-  if (data && anim_data) {
-    /* firstly filter the data */
+  size_t items = 0;
   switch (datatype) {
     /* Action-Editing Modes */
     case ANIMCONT_ACTION: /* 'Action Editor' */
@@ -3803,9 +3803,7 @@ size_t ANIM_animdata_filter(bAnimContext *ac,
   if (filter_mode & ANIMFILTER_NODUPLIS) {
     items = animdata_filter_remove_duplis(anim_data);
   }
-  }
 
-  /* return the number of items in the list */
   return items;
 }
 

source/blender/editors/asset/intern/asset_ops.cc

@@ -582,7 +582,7 @@ static void ASSET_OT_catalog_delete(wmOperatorType *ot)
 static asset_system::AssetCatalogService *get_catalog_service(bContext *C)
 {
   const SpaceFile *sfile = CTX_wm_space_file(C);
-  if (!sfile) {
+  if (!sfile || ED_fileselect_is_file_browser(sfile)) {
     return nullptr;
   }
 

source/blender/editors/grease_pencil/intern/grease_pencil_edit.cc

@@ -1406,7 +1406,7 @@ static void GREASE_PENCIL_OT_clean_loose(wmOperatorType *ot)
 
   ot->invoke = WM_operator_props_popup_confirm;
   ot->exec = grease_pencil_clean_loose_exec;
-  ot->poll = editable_grease_pencil_poll;
+  ot->poll = active_grease_pencil_layer_poll;
 
   ot->flag = OPTYPE_REGISTER | OPTYPE_UNDO;
 

source/blender/editors/io/io_cache.cc

@@ -53,8 +53,8 @@ static int cachefile_open_invoke(bContext *C, wmOperator *op, const wmEvent * /*
     char filepath[FILE_MAX];
     Main *bmain = CTX_data_main(C);
 
-    STRNCPY(filepath, BKE_main_blendfile_path(bmain));
-    BLI_path_extension_replace(filepath, sizeof(filepath), ".abc");
+    /* Default to the same directory as the blend file. */
+    BLI_path_split_dir_part(BKE_main_blendfile_path(bmain), filepath, sizeof(filepath));
     RNA_string_set(op->ptr, "filepath", filepath);
   }
 
@@ -119,7 +119,7 @@ void CACHEFILE_OT_open(wmOperatorType *ot)
   ot->cancel = open_cancel;
 
   WM_operator_properties_filesel(ot,
-                                 FILE_TYPE_ALEMBIC | FILE_TYPE_FOLDER,
+                                 FILE_TYPE_ALEMBIC | FILE_TYPE_USD | FILE_TYPE_FOLDER,
                                  FILE_BLENDER,
                                  FILE_OPENFILE,
                                  WM_FILESEL_FILEPATH | WM_FILESEL_RELPATH,
@@ -163,8 +163,8 @@ static int cachefile_layer_open_invoke(bContext *C, wmOperator *op, const wmEven
     char filepath[FILE_MAX];
     Main *bmain = CTX_data_main(C);
 
-    STRNCPY(filepath, BKE_main_blendfile_path(bmain));
-    BLI_path_extension_replace(filepath, sizeof(filepath), ".abc");
+    /* Default to the same directory as the blend file. */
+    BLI_path_split_dir_part(BKE_main_blendfile_path(bmain), filepath, sizeof(filepath));
     RNA_string_set(op->ptr, "filepath", filepath);
   }
 
@@ -215,7 +215,7 @@ void CACHEFILE_OT_layer_add(wmOperatorType *ot)
   ot->exec = cachefile_layer_add_exec;
 
   WM_operator_properties_filesel(ot,
-                                 FILE_TYPE_ALEMBIC | FILE_TYPE_FOLDER,
+                                 FILE_TYPE_ALEMBIC | FILE_TYPE_USD | FILE_TYPE_FOLDER,
                                  FILE_BLENDER,
                                  FILE_OPENFILE,
                                  WM_FILESEL_FILEPATH | WM_FILESEL_RELPATH,

source/blender/editors/mesh/editmesh_tools.cc

@@ -3584,7 +3584,10 @@ static int edbm_remove_doubles_exec(bContext *C, wmOperator *op)
     }
   }
 
-  BKE_reportf(op->reports, RPT_INFO, "Removed %d vertice(s)", count_multi);
+  BKE_reportf(op->reports,
+              RPT_INFO,
+              count_multi == 1 ? "Removed %d vertex" : "Removed %d vertices",
+              count_multi);
 
   return OPERATOR_FINISHED;
 }

source/blender/editors/object/object_edit.cc

@@ -1613,7 +1613,6 @@ static int shade_smooth_exec(bContext *C, wmOperator *op)
         const float angle = RNA_float_get(op->ptr, "angle");
         bke::mesh_sharp_edges_set_from_angle(mesh, angle, keep_sharp_edges);
       }
-      mesh.tag_sharpness_changed();
       BKE_mesh_batch_cache_dirty_tag(static_cast<Mesh *>(ob->data), BKE_MESH_BATCH_DIRTY_ALL);
       changed = true;
     }

source/blender/editors/space_view3d/view3d_gizmo_preselect_type.cc

@@ -506,7 +506,7 @@ void ED_view3d_gizmo_mesh_preselect_get_active(const bContext *C,
   /* weak, allocate an array just to access the index. */
   Base *base = nullptr;
   Object *obedit = nullptr;
-  {
+  if (object_index != -1) {
     Vector<Base *> bases = BKE_view_layer_array_from_bases_in_edit_mode(
         scene, view_layer, CTX_wm_view3d(C));
     if (object_index < bases.size()) {

source/blender/editors/transform/transform_ops.cc

@@ -705,11 +705,15 @@ void Transform_Properties(wmOperatorType *ot, int flags)
 
       RNA_def_boolean(ot->srna, "use_snap_project", false, "Project Individual Elements", "");
 
-      /* TODO(@gfxcoder): Rename `snap_target` to `snap_source` to avoid previous ambiguity of
-       * "target" (now, "source" is geometry to be moved and "target" is geometry to which moved
-       * geometry is snapped).  Use "Source snap point" and "Point on source that will snap to
-       * target" for name and description, respectively. */
-      prop = RNA_def_enum(ot->srna, "snap_target", rna_enum_snap_source_items, 0, "Snap With", "");
+      /* TODO(@gfxcoder): Rename `snap_target` to `snap_base` to avoid previous ambiguity of
+       * "target" (now, "base" or "source" is geometry to be moved and "target" is geometry to
+       * which moved geometry is snapped). */
+      prop = RNA_def_enum(ot->srna,
+                          "snap_target",
+                          rna_enum_snap_source_items,
+                          0,
+                          "Snap Base",
+                          "Point on source that will snap to target");
       RNA_def_property_flag(prop, PROP_HIDDEN);
 
       /* Target selection. */

source/blender/geometry/GEO_join_geometries.hh

@@ -13,6 +13,6 @@ bke::GeometrySet join_geometries(Span<bke::GeometrySet> geometries,
                                  const bke::AnonymousAttributePropagationInfo &propagation_info);
 
 void join_attributes(Span<const bke::GeometryComponent *> src_components,
-                     bke::GeometryComponent &r_result,
-                     Span<StringRef> ignored_attributes = {});
+                     Span<StringRef> ignored_attributes,
+                     bke::GeometryComponent &r_result);
 }  // namespace blender::geometry

source/blender/geometry/GEO_realize_instances.hh

@@ -62,16 +62,8 @@ bke::GeometrySet realize_instances(bke::GeometrySet geometry_set,
                                    const RealizeInstancesOptions &options);
 
 /**
- * Join all instances into a single geometry component for each geometry type. For example, all
- * mesh instances (including the already realized mesh) are joined into a single mesh. The output
- * geometry set does not contain any instances. If the input did not contain any instances, it is
- * returned directly.
- *
- * The `id` attribute has special handling. If there is an id attribute on any component, the
- * output will contain an `id` attribute as well. The output id is generated by mixing/hashing ids
- * of instances and of the instanced geometry data.
- *
- * Will realize only the instances chosen by varied_depth_option to there chosen depth.
+ * Same #realize_instances but will realize only the instances chosen by
+ * varied_depth_option to there chosen depth.
  */
 bke::GeometrySet realize_instances(bke::GeometrySet geometry_set,
                                    const RealizeInstancesOptions &options,

source/blender/geometry/intern/join_geometries.cc

@@ -74,8 +74,8 @@ static void fill_new_attribute(const Span<const GeometryComponent *> src_compone
 }
 
 void join_attributes(const Span<const GeometryComponent *> src_components,
-                     GeometryComponent &r_result,
-                     const Span<StringRef> ignored_attributes)
+                     const Span<StringRef> ignored_attributes,
+                     GeometryComponent &r_result)
 {
   const Map<AttributeIDRef, AttributeMetaData> info = get_final_attribute_info(src_components,
                                                                                ignored_attributes);
@@ -129,7 +129,7 @@ static void join_instances(const Span<const GeometryComponent *> src_components,
 
   r_result.replace_instances(dst_instances.release());
   auto &dst_component = r_result.get_component_for_write<bke::InstancesComponent>();
-  join_attributes(src_components, dst_component, {".reference_index"});
+  join_attributes(src_components, {".reference_index"}, dst_component);
 }
 
 static void join_volumes(const Span<const GeometryComponent *> /*src_components*/,

source/blender/geometry/intern/realize_instances.cc

@@ -298,7 +298,7 @@ struct InstanceContext {
         curves(gather_info.curves.attributes.size()),
         instances(gather_info.instances_attriubutes.size())
   {
-    //empty
+    // empty
   }
 };
 
@@ -823,12 +823,11 @@ static bool attribute_foreach(const bke::GeometrySet &geometry_set,
   return is_relevant;
 }
 
-
 /**
  * Based on bke::GeometrySet::gather_attributes_for_propagation.
  * Specialized for Specialized attribute_foreach to get:
  * current_depth, depth_target, instance_depth and selection.
-*/
+ */
 void static gather_attributes_for_propagation(
     bke::GeometrySet re_geometry_set,
     const Span<bke::GeometryComponent::Type> component_types,
@@ -869,7 +868,8 @@ void static gather_attributes_for_propagation(
 
                       AttrDomain domain = meta_data.domain;
                       if (dst_component_type != bke::GeometryComponent::Type::Instance &&
-                          domain == AttrDomain::Instance) {
+                          domain == AttrDomain::Instance)
+                      {
                         domain = AttrDomain::Point;
                       }
 
@@ -1014,7 +1014,7 @@ static void execute_instances_tasks(
   }
 
   join_attributes(
-      for_join_attributes, dst_component, {"position", ".reference_index", "instance_transform"});
+      for_join_attributes, {"position", ".reference_index", "instance_transform"}, dst_component);
 }
 
 /** \} */
@@ -1990,9 +1990,7 @@ bke::GeometrySet realize_instances(bke::GeometrySet geometry_set,
   VariedDepthOption all_instances;
   all_instances.depths = VArray<int>::ForSingle(VariedDepthOption::MAX_DEPTH,
                                                 geometry_set.get_instances()->instances_num());
-  IndexMaskMemory memory;
-  all_instances.selection = IndexMask::from_bools(
-      VArray<bool>::ForSingle(true, geometry_set.get_instances()->instances_num()), memory);
+  all_instances.selection = IndexMask(geometry_set.get_instances()->instances_num());
   return realize_instances(geometry_set, options, all_instances);
 }
 
@@ -2049,12 +2047,8 @@ bke::GeometrySet realize_instances(bke::GeometrySet geometry_set,
   const float4x4 transform = float4x4::identity();
   InstanceContext attribute_fallbacks(gather_info);
 
-  gather_realize_tasks_recursive(gather_info,
-                                 0,
-                                 VariedDepthOption::MAX_DEPTH,
-                                 geometry_set,
-                                 transform,
-                                 attribute_fallbacks);
+  gather_realize_tasks_recursive(
+      gather_info, 0, VariedDepthOption::MAX_DEPTH, geometry_set, transform, attribute_fallbacks);
 
   bke::GeometrySet new_geometry_set;
   execute_instances_tasks(gather_info.instances.instances_components_to_merge,

source/blender/gpu/metal/mtl_texture.mm

@@ -527,6 +527,8 @@ void gpu::MTLTexture::update_sub(
     }
   }
 
+  const bool is_compressed = (format_flag_ & GPU_FORMAT_COMPRESSED);
+
   @autoreleasepool {
     /* Determine totalsize of INPUT Data. */
     int num_channels = to_component_len(format_);
@@ -593,10 +595,12 @@ void gpu::MTLTexture::update_sub(
         false /* Not a clear. */
     };
 
-    /* Determine whether we can do direct BLIT or not. */
+    /* Determine whether we can do direct BLIT or not. For compressed textures,
+     * always assume a direct blit (input data pretends to be float, but it is
+     * not). */
     bool can_use_direct_blit = true;
-    if (expected_dst_bytes_per_pixel != input_bytes_per_pixel ||
-        num_channels != destination_num_channels)
+    if (!is_compressed && (expected_dst_bytes_per_pixel != input_bytes_per_pixel ||
+                           num_channels != destination_num_channels))
     {
       can_use_direct_blit = false;
     }
@@ -620,7 +624,7 @@ void gpu::MTLTexture::update_sub(
 
     /* Safety Checks. */
     if (type == GPU_DATA_UINT_24_8 || type == GPU_DATA_10_11_11_REV ||
-        type == GPU_DATA_2_10_10_10_REV)
+        type == GPU_DATA_2_10_10_10_REV || is_compressed)
     {
       BLI_assert(can_use_direct_blit &&
                  "Special input data type must be a 1-1 mapping with destination texture as it "
@@ -755,6 +759,12 @@ void gpu::MTLTexture::update_sub(
                                       extent[0] :
                                       ctx->pipeline_state.unpack_row_length);
           size_t bytes_per_image = bytes_per_row;
+          if (is_compressed) {
+            size_t block_size = to_block_size(format_);
+            size_t blocks_x = divide_ceil_u(extent[0], 4);
+            bytes_per_row = blocks_x * block_size;
+            bytes_per_image = bytes_per_row;
+          }
           int max_array_index = ((type_ == GPU_TEXTURE_1D_ARRAY) ? extent[1] : 1);
           for (int array_index = 0; array_index < max_array_index; array_index++) {
 
@@ -827,6 +837,13 @@ void gpu::MTLTexture::update_sub(
                                       extent[0] :
                                       ctx->pipeline_state.unpack_row_length);
           size_t bytes_per_image = bytes_per_row * extent[1];
+          if (is_compressed) {
+            size_t block_size = to_block_size(format_);
+            size_t blocks_x = divide_ceil_u(extent[0], 4);
+            size_t blocks_y = divide_ceil_u(extent[1], 4);
+            bytes_per_row = blocks_x * block_size;
+            bytes_per_image = bytes_per_row * blocks_y;
+          }
 
           size_t texture_array_relative_offset = 0;
           int base_slice = (type_ == GPU_TEXTURE_2D_ARRAY) ? offset[2] : 0;
@@ -1218,6 +1235,12 @@ void gpu::MTLTexture::ensure_mipmaps(int miplvl)
 
 void gpu::MTLTexture::generate_mipmap()
 {
+  /* Compressed textures allow users to provide their own custom mipmaps. And
+   * we can't generate them at runtime anyway. */
+  if (format_flag_ & GPU_FORMAT_COMPRESSED) {
+    return;
+  }
+
   /* Fetch Active Context. */
   MTLContext *ctx = static_cast<MTLContext *>(unwrap(GPU_context_active_get()));
   BLI_assert(ctx);

source/blender/gpu/metal/mtl_texture_util.mm

@@ -160,13 +160,17 @@ MTLPixelFormat gpu_texture_format_to_metal(eGPUTextureFormat tex_format)
       return MTLPixelFormatR8Snorm;
     /* Special formats, texture only. */
     case GPU_SRGB8_A8_DXT1:
+      return MTLPixelFormatBC1_RGBA_sRGB;
     case GPU_SRGB8_A8_DXT3:
+      return MTLPixelFormatBC2_RGBA_sRGB;
     case GPU_SRGB8_A8_DXT5:
+      return MTLPixelFormatBC3_RGBA_sRGB;
     case GPU_RGBA8_DXT1:
+      return MTLPixelFormatBC1_RGBA;
     case GPU_RGBA8_DXT3:
+      return MTLPixelFormatBC2_RGBA;
     case GPU_RGBA8_DXT5:
-      BLI_assert_msg(false, "Compressed texture not implemented yet!\n");
-      return MTLPixelFormatRGBA8Unorm;
+      return MTLPixelFormatBC3_RGBA;
     case GPU_SRGB8:
       /* 24-Bit pixel format are not supported. Emulate using a padded type with alpha. */
       return MTLPixelFormatRGBA8Unorm_sRGB;
@@ -247,6 +251,14 @@ size_t get_mtl_format_bytesize(MTLPixelFormat tex_format)
       return 4;
     case MTLPixelFormatDepth16Unorm:
       return 2;
+    case MTLPixelFormatBC1_RGBA:
+    case MTLPixelFormatBC1_RGBA_sRGB:
+      return 1; /* Note: not quite correct (BC1 is 0.5 bpp). */
+    case MTLPixelFormatBC2_RGBA:
+    case MTLPixelFormatBC2_RGBA_sRGB:
+    case MTLPixelFormatBC3_RGBA:
+    case MTLPixelFormatBC3_RGBA_sRGB:
+      return 1;
 
     default:
       BLI_assert_msg(false, "Unrecognised GPU pixel format!\n");
@@ -272,6 +284,12 @@ int get_mtl_format_num_components(MTLPixelFormat tex_format)
     case MTLPixelFormatRGBA8Unorm_sRGB:
     case MTLPixelFormatRGB10A2Uint:
     case MTLPixelFormatRGB10A2Unorm:
+    case MTLPixelFormatBC1_RGBA_sRGB:
+    case MTLPixelFormatBC2_RGBA_sRGB:
+    case MTLPixelFormatBC3_RGBA_sRGB:
+    case MTLPixelFormatBC1_RGBA:
+    case MTLPixelFormatBC2_RGBA:
+    case MTLPixelFormatBC3_RGBA:
       return 4;
 
     case MTLPixelFormatRG11B10Float:

source/blender/imbuf/IMB_imbuf.hh

@@ -118,6 +118,21 @@ ImBuf *IMB_allocFromBuffer(const uint8_t *byte_buffer,
 void IMB_assign_byte_buffer(ImBuf *ibuf, uint8_t *buffer_data, ImBufOwnership ownership);
 void IMB_assign_float_buffer(ImBuf *ibuf, float *buffer_data, ImBufOwnership ownership);
 
+/**
+ * Assign the content and the color space of the corresponding buffer the data from the given
+ * buffer.
+ *
+ * \note Does not modify the topology (width, height, number of channels)
+ * or the mipmaps in any way.
+ *
+ * \note The ownership of the data in the source buffer is ignored.
+ */
+void IMB_assign_byte_buffer(ImBuf *ibuf, const ImBufByteBuffer &buffer, ImBufOwnership ownership);
+void IMB_assign_float_buffer(ImBuf *ibuf,
+                             const ImBufFloatBuffer &buffer,
+                             ImBufOwnership ownership);
+void IMB_assign_dds_data(ImBuf *ibuf, const DDSData &data, ImBufOwnership ownership);
+
 /**
  * Make corresponding buffers available for modification.
  * Is achieved by ensuring that the given ImBuf is the only owner of the underlying buffer data.

source/blender/imbuf/IMB_imbuf_types.hh

@@ -34,17 +34,6 @@ struct IDProperty;
 #define IMB_MIPMAP_LEVELS 20
 #define IMB_FILEPATH_SIZE 1024
 
-struct DDSData {
-  /** DDS fourcc info */
-  unsigned int fourcc;
-  /** The number of mipmaps in the dds file */
-  unsigned int nummipmaps;
-  /** The compressed image data */
-  unsigned char *data;
-  /** The size of the compressed data */
-  unsigned int size;
-};
-
 /**
  * \ingroup imbuf
  * This is the abstraction of an image. ImBuf is the basic type used for all imbuf operations.
@@ -143,6 +132,19 @@ enum ImBufOwnership {
   IB_TAKE_OWNERSHIP = 1,
 };
 
+struct DDSData {
+  /** DDS fourcc info */
+  unsigned int fourcc;
+  /** The number of mipmaps in the dds file */
+  unsigned int nummipmaps;
+  /** The compressed image data */
+  unsigned char *data;
+  /** The size of the compressed data */
+  unsigned int size;
+  /** Who owns the data buffer. */
+  ImBufOwnership ownership;
+};
+
 /* Different storage specialization.
  *
  * NOTE: Avoid direct assignments and allocations, use the buffer utilities from the IMB_imbuf.hh

source/blender/imbuf/intern/allocimbuf.cc

@@ -84,6 +84,27 @@ template<class BufferType> static void imb_free_buffer(BufferType &buffer)
   buffer.ownership = IB_DO_NOT_TAKE_OWNERSHIP;
 }
 
+/* Free the specified DDS buffer storage, freeing memory when needed and restoring the state of the
+ * buffer to its defaults. */
+static void imb_free_dds_buffer(DDSData &dds_data)
+{
+  if (dds_data.data) {
+    switch (dds_data.ownership) {
+      case IB_DO_NOT_TAKE_OWNERSHIP:
+        break;
+
+      case IB_TAKE_OWNERSHIP:
+        /* dds_data.data is allocated by DirectDrawSurface::readData(), so don't use MEM_freeN! */
+        free(dds_data.data);
+        break;
+    }
+  }
+
+  /* Reset buffer to defaults. */
+  dds_data.data = nullptr;
+  dds_data.ownership = IB_DO_NOT_TAKE_OWNERSHIP;
+}
+
 /* Allocate pixel storage of the given buffer. The buffer owns the allocated memory.
  * Returns true of allocation succeeded, false otherwise. */
 template<class BufferType>
@@ -249,11 +270,7 @@ void IMB_freeImBuf(ImBuf *ibuf)
     IMB_free_gpu_textures(ibuf);
     IMB_metadata_free(ibuf->metadata);
     colormanage_cache_free(ibuf);
-
-    if (ibuf->dds_data.data != nullptr) {
-      /* dds_data.data is allocated by DirectDrawSurface::readData(), so don't use MEM_freeN! */
-      free(ibuf->dds_data.data);
-    }
+    imb_free_dds_buffer(ibuf->dds_data);
     MEM_freeN(ibuf);
   }
 }
@@ -472,6 +489,32 @@ void IMB_assign_float_buffer(ImBuf *ibuf, float *buffer_data, const ImBufOwnersh
   }
 }
 
+void IMB_assign_byte_buffer(ImBuf *ibuf,
+                            const ImBufByteBuffer &buffer,
+                            const ImBufOwnership ownership)
+{
+  IMB_assign_byte_buffer(ibuf, buffer.data, ownership);
+  ibuf->byte_buffer.colorspace = buffer.colorspace;
+}
+
+void IMB_assign_float_buffer(ImBuf *ibuf,
+                             const ImBufFloatBuffer &buffer,
+                             const ImBufOwnership ownership)
+{
+  IMB_assign_float_buffer(ibuf, buffer.data, ownership);
+  ibuf->float_buffer.colorspace = buffer.colorspace;
+}
+
+void IMB_assign_dds_data(ImBuf *ibuf, const DDSData &data, const ImBufOwnership ownership)
+{
+  BLI_assert(ibuf->ftype == IMB_FTYPE_DDS);
+
+  imb_free_dds_buffer(ibuf->dds_data);
+
+  ibuf->dds_data = data;
+  ibuf->dds_data.ownership = ownership;
+}
+
 ImBuf *IMB_allocFromBufferOwn(
     uint8_t *byte_buffer, float *float_buffer, uint w, uint h, uint channels)
 {

source/blender/imbuf/intern/format_dds.cc

@@ -330,6 +330,7 @@ static void LoadDXTCImage(ImBuf *ibuf, Filesystem::IOMemReader &mem_reader)
     ibuf->dds_data.size = mem_reader.size() - dds_header_size;
     ibuf->dds_data.data = (uchar *)malloc(ibuf->dds_data.size);
     mem_reader.pread(ibuf->dds_data.data, ibuf->dds_data.size, dds_header_size);
+    ibuf->dds_data.ownership = IB_TAKE_OWNERSHIP;
 
     /* Flip compressed image data to match OpenGL convention. */
     FlipDXTCImage(ibuf);

source/blender/makesrna/intern/rna_modifier.cc

@@ -7703,6 +7703,8 @@ static void rna_def_modifier_nodes_bake(BlenderRNA *brna)
   StructRNA *srna;
   PropertyRNA *prop;
 
+  RNA_define_lib_overridable(true);
+
   srna = RNA_def_struct(brna, "NodesModifierBake", nullptr);
   RNA_def_struct_ui_text(srna, "Nodes Modifier Bake", "");
 
@@ -7758,6 +7760,8 @@ static void rna_def_modifier_nodes_bake(BlenderRNA *brna)
   RNA_def_property_struct_type(prop, "NodesModifierDataBlock");
   RNA_def_property_collection_sdna(prop, nullptr, "data_blocks", "data_blocks_num");
   RNA_def_property_srna(prop, "NodesModifierBakeDataBlocks");
+
+  RNA_define_lib_overridable(false);
 }
 
 static void rna_def_modifier_nodes_bakes(BlenderRNA *brna)

source/blender/makesrna/intern/rna_scene.cc

@@ -3574,7 +3574,7 @@ static void rna_def_tool_settings(BlenderRNA *brna)
       prop, "rna_ToolSettings_snap_mode_get", "rna_ToolSettings_snap_mode_set", nullptr);
   RNA_def_property_flag(prop, PROP_ENUM_FLAG);
   RNA_def_property_ui_text(
-      prop, "Snap Element", "Type of element for the \"Snap With\" to snap to");
+      prop, "Snap Element", "Type of element for the \"Snap Base\" to snap to");
   RNA_def_property_update(prop, NC_SCENE | ND_TOOLSETTINGS, nullptr); /* header redraw */
 
   prop = RNA_def_property(srna, "snap_elements_individual", PROP_ENUM, PROP_NONE);

source/blender/nodes/geometry/nodes/node_geo_realize_instances.cc

@@ -18,13 +18,13 @@ static void node_declare(NodeDeclarationBuilder &b)
   b.add_input<decl::Bool>("Selection")
       .default_value(true)
       .hide_value()
-      .supports_field()
+      .field_on_all()
       .description("Which top-level instances to realize");
   b.add_input<decl::Bool>("Realize All")
       .default_value(true)
-      .supports_field()
+      .field_on_all()
       .description("Determine wether to realize nested instances completly");
-  b.add_input<decl::Int>("Depth").default_value(0).min(0).supports_field().description(
+  b.add_input<decl::Int>("Depth").default_value(0).min(0).field_on_all().description(
       "Number of levels of nested instances to realize for each top-level instance");
   b.add_output<decl::Geometry>("Geometry").propagate_all();
 }
@@ -38,24 +38,29 @@ static void node_geo_exec(GeoNodeExecParams params)
   }
 
   GeometryComponentEditData::remember_deformed_positions_if_necessary(geometry_set);
-  Field<bool> selection_field = params.extract_input<Field<bool>>("Selection");
-  Field<bool> realize_all_filed = params.extract_input<Field<bool>>("Realize All");
+
+  Field<bool> realize_all_field = params.extract_input<Field<bool>>("Realize All");
   Field<int> depth_field = params.extract_input<Field<int>>("Depth");
 
   static auto depth_override = mf::build::SI2_SO<int, bool, int>(
       "depth_override",
-      [](int value, bool realize) { return realize ? -1 : std::max(value, 0); },
+      [](int depth, bool realize_all_field) {
+        return realize_all_field ? geometry::VariedDepthOption::MAX_DEPTH : std::max(depth, 0);
+      },
       mf::build::exec_presets::AllSpanOrSingle());
 
+  Field<int> depth_field_overrided(FieldOperation::Create(
+      depth_override, {std::move(depth_field), std::move(realize_all_field)}));
+
+  Field<bool> selection_field = params.extract_input<Field<bool>>("Selection");
+
   static auto selection_override = mf::build::SI2_SO<int, bool, bool>(
       "selection_override",
-      [](int value, bool selection) { return value == 0 ? false : selection; },
+      [](int depth_override, bool selection) { return depth_override == 0 ? false : selection; },
       mf::build::exec_presets::AllSpanOrSingle());
 
-  Field<int> depth_field_overrided(
-      FieldOperation::Create(depth_override, {depth_field, realize_all_filed}));
-  Field<bool> selection_field_overrided(
-      FieldOperation::Create(selection_override, {depth_field_overrided, selection_field}));
+  Field<bool> selection_field_overrided(FieldOperation::Create(
+      selection_override, {depth_field_overrided, std::move(selection_field)}));
 
   const bke::Instances &instances = *geometry_set.get_instances();
   const bke::InstancesFieldContext field_context(instances);

source/blender/nodes/geometry/nodes/node_geo_set_curve_normal.cc

@@ -21,7 +21,10 @@ static void node_declare(NodeDeclarationBuilder &b)
   b.add_input<decl::Geometry>("Curve").supported_type(
       {GeometryComponent::Type::Curve, GeometryComponent::Type::GreasePencil});
   b.add_input<decl::Bool>("Selection").default_value(true).hide_value().field_on_all();
-  b.add_input<decl::Vector>("Normal").default_value({0.0f, 0.0f, 1.0f}).field_on_all();
+  b.add_input<decl::Vector>("Normal")
+      .default_value({0.0f, 0.0f, 1.0f})
+      .subtype(PROP_XYZ)
+      .field_on_all();
   b.add_output<decl::Geometry>("Curve").propagate_all();
 }
 

source/blender/python/gpu/gpu_py_vertex_buffer.cc

@@ -241,7 +241,7 @@ static PyObject *pygpu_vertbuf__tp_new(PyTypeObject * /*type*/, PyObject *args,
       PY_ARG_PARSER_HEAD_COMPAT()
       "O!" /* `format` */
       "I"  /* `len` */
-      ":blender::gpu::VertBuf.__new__",
+      ":GPUVertBuf.__new__",
       _keywords,
       nullptr,
   };
@@ -307,7 +307,7 @@ static PyObject *pygpu_vertbuf_attr_fill(BPyGPUVertBuf *self, PyObject *args, Py
     return nullptr;
   }
 
-  if (!pygpu_vertbuf_fill(self->buf, id, data, "blender::gpu::VertBuf.attr_fill")) {
+  if (!pygpu_vertbuf_fill(self->buf, id, data, "GPUVertBuf.attr_fill")) {
     return nullptr;
   }
 
@@ -340,7 +340,7 @@ static void pygpu_vertbuf__tp_dealloc(BPyGPUVertBuf *self)
 PyDoc_STRVAR(
     /* Wrap. */
     pygpu_vertbuf__tp_doc,
-    ".. class:: blender::gpu::VertBuf(format, len)\n"
+    ".. class:: GPUVertBuf(format, len)\n"
     "\n"
     "   Contains a VBO.\n"
     "\n"
@@ -350,7 +350,7 @@ PyDoc_STRVAR(
     "   :type len: int\n");
 PyTypeObject BPyGPUVertBuf_Type = {
     /*ob_base*/ PyVarObject_HEAD_INIT(nullptr, 0)
-    /*tp_name*/ "blender::gpu::VertBuf",
+    /*tp_name*/ "GPUVertBuf",
     /*tp_basicsize*/ sizeof(BPyGPUVertBuf),
     /*tp_itemsize*/ 0,
     /*tp_dealloc*/ (destructor)pygpu_vertbuf__tp_dealloc,

source/blender/render/intern/engine.cc

@@ -1314,8 +1314,8 @@ bool RE_engine_gpu_context_enable(RenderEngine *engine)
     /* Activate RenderEngine System and Blender GPU Context. */
     WM_system_gpu_context_activate(engine->system_gpu_context);
     if (engine->blender_gpu_context) {
-      GPU_context_active_set(engine->blender_gpu_context);
       GPU_render_begin();
+      GPU_context_active_set(engine->blender_gpu_context);
     }
     return true;
   }
@@ -1330,8 +1330,8 @@ void RE_engine_gpu_context_disable(RenderEngine *engine)
   else {
     if (engine->system_gpu_context) {
       if (engine->blender_gpu_context) {
-        GPU_render_end();
         GPU_context_active_set(nullptr);
+        GPU_render_end();
       }
       WM_system_gpu_context_release(engine->system_gpu_context);
       /* Restore DRW state context if previously active. */

tests/data

@@ -1 +1 @@
-Subproject commit 5038ad7165fd1a77e61e0d2d6efdadd6ea7c0dfb
+Subproject commit bf5c70830540b215a3b1df21f28e0e80ead230f7