Cycles: Improve denoising speed on GPUs with small tile sizes

[blender.git] / intern / cycles / kernel / kernels / cpu / filter_cpu.h

diff --git a/intern/cycles/kernel/kernels/cpu/filter_cpu.h b/intern/cycles/kernel/kernels/cpu/filter_cpu.h
index 10007ee2635870fe7a04031bb8dd3647a06dc187..4231aba88d7b8a09bb8878afbe7133cd12c9daf0 100644 (file)
--- a/intern/cycles/kernel/kernels/cpu/filter_cpu.h
+++ b/intern/cycles/kernel/kernels/cpu/filter_cpu.h
@@ -27,8 +27,7 @@ void KERNEL_FUNCTION_FULL_NAME(filter_divide_shadow)(int sample,
                                                      float *bufferV,
                                                      int* prefilter_rect,
                                                      int buffer_pass_stride,
-                                                     int buffer_denoising_offset,
-                                                     bool use_split_variance);
+                                                     int buffer_denoising_offset);
 
 void KERNEL_FUNCTION_FULL_NAME(filter_get_feature)(int sample,
                                                    TilesInfo *tiles,
@@ -40,8 +39,15 @@ void KERNEL_FUNCTION_FULL_NAME(filter_get_feature)(int sample,
                                                    float *variance,
                                                    int* prefilter_rect,
                                                    int buffer_pass_stride,
-                                                   int buffer_denoising_offset,
-                                                   bool use_split_variance);
+                                                   int buffer_denoising_offset);
+
+void KERNEL_FUNCTION_FULL_NAME(filter_detect_outliers)(int x, int y,
+                                                       ccl_global float *image,
+                                                       ccl_global float *variance,
+                                                       ccl_global float *depth,
+                                                       ccl_global float *output,
+                                                       int *rect,
+                                                       int pass_stride);
 
 void KERNEL_FUNCTION_FULL_NAME(filter_combine_halves)(int x, int y,
                                                       float *mean,
@@ -64,64 +70,59 @@ void KERNEL_FUNCTION_FULL_NAME(filter_construct_transform)(float* buffer,
 
 void KERNEL_FUNCTION_FULL_NAME(filter_nlm_calc_difference)(int dx,
                                                            int dy,
-                                                           float *weightImage,
+                                                           float *weight_image,
                                                            float *variance,
-                                                           float *differenceImage,
+                                                           float *difference_image,
                                                            int* rect,
-                                                           int w,
+                                                           int stride,
                                                            int channel_offset,
                                                            float a,
                                                            float k_2);
 
-void KERNEL_FUNCTION_FULL_NAME(filter_nlm_blur)(float *differenceImage,
-                                                float *outImage,
+void KERNEL_FUNCTION_FULL_NAME(filter_nlm_blur)(float *difference_image,
+                                                float *out_image,
                                                 int* rect,
-                                                int w,
+                                                int stride,
                                                 int f);
 
-void KERNEL_FUNCTION_FULL_NAME(filter_nlm_calc_weight)(float *differenceImage,
-                                                       float *outImage,
+void KERNEL_FUNCTION_FULL_NAME(filter_nlm_calc_weight)(float *difference_image,
+                                                       float *out_image,
                                                        int* rect,
-                                                       int w,
+                                                       int stride,
                                                        int f);
 
 void KERNEL_FUNCTION_FULL_NAME(filter_nlm_update_output)(int dx,
                                                          int dy,
-                                                         float *differenceImage,
+                                                         float *difference_image,
                                                          float *image,
-                                                         float *outImage,
-                                                         float *accumImage,
+                                                         float *out_image,
+                                                         float *accum_image,
                                                          int* rect,
-                                                         int w,
+                                                         int stride,
                                                          int f);
 
 void KERNEL_FUNCTION_FULL_NAME(filter_nlm_construct_gramian)(int dx,
                                                              int dy,
-                                                             float *differenceImage,
+                                                             float *difference_image,
                                                              float *buffer,
-                                                             float *color_pass,
-                                                             float *variance_pass,
                                                              float *transform,
                                                              int *rank,
                                                              float *XtWX,
                                                              float3 *XtWY,
                                                              int *rect,
-                                                             int *filter_rect,
-                                                             int w,
-                                                             int h,
+                                                             int *filter_window,
+                                                             int stride,
                                                              int f,
                                                              int pass_stride);
 
-void KERNEL_FUNCTION_FULL_NAME(filter_nlm_normalize)(float *outImage,
-                                                     float *accumImage,
+void KERNEL_FUNCTION_FULL_NAME(filter_nlm_normalize)(float *out_image,
+                                                     float *accum_image,
                                                      int* rect,
-                                                     int w);
+                                                     int stride);
 
 void KERNEL_FUNCTION_FULL_NAME(filter_finalize)(int x,
                                                 int y,
                                                 int storage_ofs,
-                                                int w,
-                                                int h,
                                                 float *buffer,
                                                 int *rank,
                                                 float *XtWX,