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, 231 insertions, 0 deletions

diff --git a/intern/cycles/kernel/kernel_adaptive_sampling.h b/intern/cycles/kernel/kernel_adaptive_sampling.h
new file mode 100644
index 00000000000..502b69e4f7f
--- /dev/null
+++ b/intern/cycles/kernel/kernel_adaptive_sampling.h
@@ -0,0 +1,231 @@
+/*
+ * Copyright 2019 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef __KERNEL_ADAPTIVE_SAMPLING_H__
+#define __KERNEL_ADAPTIVE_SAMPLING_H__
+
+CCL_NAMESPACE_BEGIN
+
+/* Determines whether to continue sampling a given pixel or if it has sufficiently converged. */
+
+ccl_device void kernel_do_adaptive_stopping(KernelGlobals *kg,
+                                            ccl_global float *buffer,
+                                            int sample)
+{
+  /* TODO Stefan: Is this better in linear, sRGB or something else? */
+  float4 I = *((ccl_global float4 *)buffer);
+  float4 A = *(ccl_global float4 *)(buffer + kernel_data.film.pass_adaptive_aux_buffer);
+  /* The per pixel error as seen in section 2.1 of
+   * "A hierarchical automatic stopping condition for Monte Carlo global illumination"
+   * A small epsilon is added to the divisor to prevent division by zero. */
+  float error = (fabsf(I.x - A.x) + fabsf(I.y - A.y) + fabsf(I.z - A.z)) /
+                (sample * 0.0001f + sqrtf(I.x + I.y + I.z));
+  if (error < kernel_data.integrator.adaptive_threshold * (float)sample) {
+    /* Set the fourth component to non-zero value to indicate that this pixel has converged. */
+    buffer[kernel_data.film.pass_adaptive_aux_buffer + 3] += 1.0f;
+  }
+}
+
+/* Adjust the values of an adaptively sampled pixel. */
+
+ccl_device void kernel_adaptive_post_adjust(KernelGlobals *kg,
+                                            ccl_global float *buffer,
+                                            float sample_multiplier)
+{
+  *(ccl_global float4 *)(buffer) *= sample_multiplier;
+
+  /* Scale the aux pass too, this is necessary for progressive rendering to work properly. */
+  kernel_assert(kernel_data.film.pass_adaptive_aux_buffer);
+  *(ccl_global float4 *)(buffer + kernel_data.film.pass_adaptive_aux_buffer) *= sample_multiplier;
+
+#ifdef __PASSES__
+  int flag = kernel_data.film.pass_flag;
+
+  if (flag & PASSMASK(SHADOW))
+    *(ccl_global float3 *)(buffer + kernel_data.film.pass_shadow) *= sample_multiplier;
+
+  if (flag & PASSMASK(MIST))
+    *(ccl_global float *)(buffer + kernel_data.film.pass_mist) *= sample_multiplier;
+
+  if (flag & PASSMASK(NORMAL))
+    *(ccl_global float3 *)(buffer + kernel_data.film.pass_normal) *= sample_multiplier;
+
+  if (flag & PASSMASK(UV))
+    *(ccl_global float3 *)(buffer + kernel_data.film.pass_uv) *= sample_multiplier;
+
+  if (flag & PASSMASK(MOTION)) {
+    *(ccl_global float4 *)(buffer + kernel_data.film.pass_motion) *= sample_multiplier;
+    *(ccl_global float *)(buffer + kernel_data.film.pass_motion_weight) *= sample_multiplier;
+  }
+
+  if (kernel_data.film.use_light_pass) {
+    int light_flag = kernel_data.film.light_pass_flag;
+
+    if (light_flag & PASSMASK(DIFFUSE_INDIRECT))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_diffuse_indirect) *= sample_multiplier;
+    if (light_flag & PASSMASK(GLOSSY_INDIRECT))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_glossy_indirect) *= sample_multiplier;
+    if (light_flag & PASSMASK(TRANSMISSION_INDIRECT))
+      *(ccl_global float3 *)(buffer +
+                             kernel_data.film.pass_transmission_indirect) *= sample_multiplier;
+    if (light_flag & PASSMASK(VOLUME_INDIRECT))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_volume_indirect) *= sample_multiplier;
+    if (light_flag & PASSMASK(DIFFUSE_DIRECT))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_diffuse_direct) *= sample_multiplier;
+    if (light_flag & PASSMASK(GLOSSY_DIRECT))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_glossy_direct) *= sample_multiplier;
+    if (light_flag & PASSMASK(TRANSMISSION_DIRECT))
+      *(ccl_global float3 *)(buffer +
+                             kernel_data.film.pass_transmission_direct) *= sample_multiplier;
+    if (light_flag & PASSMASK(VOLUME_DIRECT))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_volume_direct) *= sample_multiplier;
+
+    if (light_flag & PASSMASK(EMISSION))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_emission) *= sample_multiplier;
+    if (light_flag & PASSMASK(BACKGROUND))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_background) *= sample_multiplier;
+    if (light_flag & PASSMASK(AO))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_ao) *= sample_multiplier;
+
+    if (light_flag & PASSMASK(DIFFUSE_COLOR))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_diffuse_color) *= sample_multiplier;
+    if (light_flag & PASSMASK(GLOSSY_COLOR))
+      *(ccl_global float3 *)(buffer + kernel_data.film.pass_glossy_color) *= sample_multiplier;
+    if (light_flag & PASSMASK(TRANSMISSION_COLOR))
+      *(ccl_global float3 *)(buffer +
+                             kernel_data.film.pass_transmission_color) *= sample_multiplier;
+  }
+#endif
+
+#ifdef __DENOISING_FEATURES__
+
+#  define scale_float3_variance(buffer, offset, scale) \
+    *(buffer + offset) *= scale; \
+    *(buffer + offset + 1) *= scale; \
+    *(buffer + offset + 2) *= scale; \
+    *(buffer + offset + 3) *= scale * scale; \
+    *(buffer + offset + 4) *= scale * scale; \
+    *(buffer + offset + 5) *= scale * scale;
+
+#  define scale_shadow_variance(buffer, offset, scale) \
+    *(buffer + offset) *= scale; \
+    *(buffer + offset + 1) *= scale; \
+    *(buffer + offset + 2) *= scale * scale;
+
+  if (kernel_data.film.pass_denoising_data) {
+    scale_shadow_variance(
+        buffer, kernel_data.film.pass_denoising_data + DENOISING_PASS_SHADOW_A, sample_multiplier);
+    scale_shadow_variance(
+        buffer, kernel_data.film.pass_denoising_data + DENOISING_PASS_SHADOW_B, sample_multiplier);
+    if (kernel_data.film.pass_denoising_clean) {
+      scale_float3_variance(
+          buffer, kernel_data.film.pass_denoising_data + DENOISING_PASS_COLOR, sample_multiplier);
+      *(buffer + kernel_data.film.pass_denoising_clean) *= sample_multiplier;
+      *(buffer + kernel_data.film.pass_denoising_clean + 1) *= sample_multiplier;
+      *(buffer + kernel_data.film.pass_denoising_clean + 2) *= sample_multiplier;
+    }
+    else {
+      scale_float3_variance(
+          buffer, kernel_data.film.pass_denoising_data + DENOISING_PASS_COLOR, sample_multiplier);
+    }
+    scale_float3_variance(
+        buffer, kernel_data.film.pass_denoising_data + DENOISING_PASS_NORMAL, sample_multiplier);
+    scale_float3_variance(
+        buffer, kernel_data.film.pass_denoising_data + DENOISING_PASS_ALBEDO, sample_multiplier);
+    *(buffer + kernel_data.film.pass_denoising_data + DENOISING_PASS_DEPTH) *= sample_multiplier;
+    *(buffer + kernel_data.film.pass_denoising_data + DENOISING_PASS_DEPTH +
+      1) *= sample_multiplier * sample_multiplier;
+  }
+#endif /* __DENOISING_FEATURES__ */
+
+  if (kernel_data.film.cryptomatte_passes) {
+    int num_slots = 0;
+    num_slots += (kernel_data.film.cryptomatte_passes & CRYPT_OBJECT) ? 1 : 0;
+    num_slots += (kernel_data.film.cryptomatte_passes & CRYPT_MATERIAL) ? 1 : 0;
+    num_slots += (kernel_data.film.cryptomatte_passes & CRYPT_ASSET) ? 1 : 0;
+    num_slots = num_slots * 2 * kernel_data.film.cryptomatte_depth;
+    ccl_global float2 *id_buffer = (ccl_global float2 *)(buffer +
+                                                         kernel_data.film.pass_cryptomatte);
+    for (int slot = 0; slot < num_slots; slot++) {
+      id_buffer[slot].y *= sample_multiplier;
+    }
+  }
+}
+
+/* This is a simple box filter in two passes.
+ * When a pixel demands more adaptive samples, let its neighboring pixels draw more samples too. */
+
+ccl_device bool kernel_do_adaptive_filter_x(KernelGlobals *kg, int y, ccl_global WorkTile *tile)
+{
+  bool any = false;
+  bool prev = false;
+  for (int x = tile->x; x < tile->x + tile->w; ++x) {
+    int index = tile->offset + x + y * tile->stride;
+    ccl_global float *buffer = tile->buffer + index * kernel_data.film.pass_stride;
+    ccl_global float4 *aux = (ccl_global float4 *)(buffer +
+                                                   kernel_data.film.pass_adaptive_aux_buffer);
+    if (aux->w == 0.0f) {
+      any = true;
+      if (x > tile->x && !prev) {
+        index = index - 1;
+        buffer = tile->buffer + index * kernel_data.film.pass_stride;
+        aux = (ccl_global float4 *)(buffer + kernel_data.film.pass_adaptive_aux_buffer);
+        aux->w = 0.0f;
+      }
+      prev = true;
+    }
+    else {
+      if (prev) {
+        aux->w = 0.0f;
+      }
+      prev = false;
+    }
+  }
+  return any;
+}
+
+ccl_device bool kernel_do_adaptive_filter_y(KernelGlobals *kg, int x, ccl_global WorkTile *tile)
+{
+  bool prev = false;
+  bool any = false;
+  for (int y = tile->y; y < tile->y + tile->h; ++y) {
+    int index = tile->offset + x + y * tile->stride;
+    ccl_global float *buffer = tile->buffer + index * kernel_data.film.pass_stride;
+    ccl_global float4 *aux = (ccl_global float4 *)(buffer +
+                                                   kernel_data.film.pass_adaptive_aux_buffer);
+    if (aux->w == 0.0f) {
+      any = true;
+      if (y > tile->y && !prev) {
+        index = index - tile->stride;
+        buffer = tile->buffer + index * kernel_data.film.pass_stride;
+        aux = (ccl_global float4 *)(buffer + kernel_data.film.pass_adaptive_aux_buffer);
+        aux->w = 0.0f;
+      }
+      prev = true;
+    }
+    else {
+      if (prev) {
+        aux->w = 0.0f;
+      }
+      prev = false;
+    }
+  }
+  return any;
+}
+
+CCL_NAMESPACE_END
+
+#endif /* __KERNEL_ADAPTIVE_SAMPLING_H__ */