Adaptive Sampling for Cycles.

This feature takes some inspiration from
"RenderMan: An Advanced Path Tracing Architecture for Movie Rendering" and
"A Hierarchical Automatic Stopping Condition for Monte Carlo Global Illumination"

The basic principle is as follows:
While samples are being added to a pixel, the adaptive sampler writes half
of the samples to a separate buffer. This gives it two separate estimates
of the same pixel, and by comparing their difference it estimates convergence.
Once convergence drops below a given threshold, the pixel is considered done.

When a pixel has not converged yet and needs more samples than the minimum,
its immediate neighbors are also set to take more samples. This is done in order
to more reliably detect sharp features such as caustics. A 3x3 box filter that
is run periodically over the tile buffer is used for that purpose.

After a tile has finished rendering, the values of all passes are scaled as if
they were rendered with the full number of samples. This way, any code operating
on these buffers, for example the denoiser, does not need to be changed for
per-pixel sample counts.

Reviewed By: brecht, #cycles

Differential Revision: https://developer.blender.org/D4686

1 files changed, 57 insertions, 27 deletions

diff --git a/intern/cycles/kernel/kernel_work_stealing.h b/intern/cycles/kernel/kernel_work_stealing.h
index 799561a7466..c642d227e4b 100644
--- a/intern/cycles/kernel/kernel_work_stealing.h
+++ b/intern/cycles/kernel/kernel_work_stealing.h
@@ -23,17 +23,41 @@ CCL_NAMESPACE_BEGIN
  * Utility functions for work stealing
  */
 
+/* Map global work index to tile, pixel X/Y and sample. */
+ccl_device_inline void get_work_pixel(ccl_global const WorkTile *tile,
+                                      uint global_work_index,
+                                      ccl_private uint *x,
+                                      ccl_private uint *y,
+                                      ccl_private uint *sample)
+{
+#ifdef __KERNEL_CUDA__
+  /* Keeping threads for the same pixel together improves performance on CUDA. */
+  uint sample_offset = global_work_index % tile->num_samples;
+  uint pixel_offset = global_work_index / tile->num_samples;
+#else  /* __KERNEL_CUDA__ */
+  uint tile_pixels = tile->w * tile->h;
+  uint sample_offset = global_work_index / tile_pixels;
+  uint pixel_offset = global_work_index - sample_offset * tile_pixels;
+#endif /* __KERNEL_CUDA__ */
+  uint y_offset = pixel_offset / tile->w;
+  uint x_offset = pixel_offset - y_offset * tile->w;
+
+  *x = tile->x + x_offset;
+  *y = tile->y + y_offset;
+  *sample = tile->start_sample + sample_offset;
+}
+
 #ifdef __KERNEL_OPENCL__
 #  pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
 #endif
 
 #ifdef __SPLIT_KERNEL__
 /* Returns true if there is work */
-ccl_device bool get_next_work(KernelGlobals *kg,
-                              ccl_global uint *work_pools,
-                              uint total_work_size,
-                              uint ray_index,
-                              ccl_private uint *global_work_index)
+ccl_device bool get_next_work_item(KernelGlobals *kg,
+                                   ccl_global uint *work_pools,
+                                   uint total_work_size,
+                                   uint ray_index,
+                                   ccl_private uint *global_work_index)
 {
   /* With a small amount of work there may be more threads than work due to
    * rounding up of global size, stop such threads immediately. */
@@ -56,31 +80,37 @@ ccl_device bool get_next_work(KernelGlobals *kg,
   /* Test if all work for this pool is done. */
   return (*global_work_index < total_work_size);
 }
-#endif
 
-/* Map global work index to tile, pixel X/Y and sample. */
-ccl_device_inline void get_work_pixel(ccl_global const WorkTile *tile,
-                                      uint global_work_index,
-                                      ccl_private uint *x,
-                                      ccl_private uint *y,
-                                      ccl_private uint *sample)
+ccl_device bool get_next_work(KernelGlobals *kg,
+                              ccl_global uint *work_pools,
+                              uint total_work_size,
+                              uint ray_index,
+                              ccl_private uint *global_work_index)
 {
-#ifdef __KERNEL_CUDA__
-  /* Keeping threads for the same pixel together improves performance on CUDA. */
-  uint sample_offset = global_work_index % tile->num_samples;
-  uint pixel_offset = global_work_index / tile->num_samples;
-#else  /* __KERNEL_CUDA__ */
-  uint tile_pixels = tile->w * tile->h;
-  uint sample_offset = global_work_index / tile_pixels;
-  uint pixel_offset = global_work_index - sample_offset * tile_pixels;
-#endif /* __KERNEL_CUDA__ */
-  uint y_offset = pixel_offset / tile->w;
-  uint x_offset = pixel_offset - y_offset * tile->w;
-
-  *x = tile->x + x_offset;
-  *y = tile->y + y_offset;
-  *sample = tile->start_sample + sample_offset;
+  bool got_work = false;
+  if (kernel_data.film.pass_adaptive_aux_buffer) {
+    do {
+      got_work = get_next_work_item(kg, work_pools, total_work_size, ray_index, global_work_index);
+      if (got_work) {
+        ccl_global WorkTile *tile = &kernel_split_params.tile;
+        uint x, y, sample;
+        get_work_pixel(tile, *global_work_index, &x, &y, &sample);
+        uint buffer_offset = (tile->offset + x + y * tile->stride) * kernel_data.film.pass_stride;
+        ccl_global float *buffer = kernel_split_params.tile.buffer + buffer_offset;
+        ccl_global float4 *aux = (ccl_global float4 *)(buffer +
+                                                       kernel_data.film.pass_adaptive_aux_buffer);
+        if (aux->w == 0.0f) {
+          break;
+        }
+      }
+    } while (got_work);
+  }
+  else {
+    got_work = get_next_work_item(kg, work_pools, total_work_size, ray_index, global_work_index);
+  }
+  return got_work;
 }
+#endif
 
 CCL_NAMESPACE_END