intern/cycles/kernel/split/kernel_background_buffer_update.h

   1 /*
   2  * Copyright 2011-2015 Blender Foundation
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at
   7  *
   8  * http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include "kernel_split_common.h"
  18
  19 /* Note on kernel_background_buffer_update kernel.
  20  * This is the fourth kernel in the ray tracing logic, and the third
  21  * of the path iteration kernels. This kernel takes care of rays that hit
  22  * the background (sceneintersect kernel), and for the rays of
  23  * state RAY_UPDATE_BUFFER it updates the ray's accumulated radiance in
  24  * the output buffer. This kernel also takes care of rays that have been determined
  25  * to-be-regenerated.
  26  *
  27  * We will empty QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS queue in this kernel
  28  *
  29  * Typically all rays that are in state RAY_HIT_BACKGROUND, RAY_UPDATE_BUFFER
  30  * will be eventually set to RAY_TO_REGENERATE state in this kernel. Finally all rays of ray_state
  31  * RAY_TO_REGENERATE will be regenerated and put in queue QUEUE_ACTIVE_AND_REGENERATED_RAYS.
  32  *
  33  * The input and output are as follows,
  34  *
  35  * rng_coop ---------------------------------------------|--- kernel_background_buffer_update --|--- PathRadiance_coop
  36  * throughput_coop --------------------------------------|                                      |--- L_transparent_coop
  37  * per_sample_output_buffers ----------------------------|                                      |--- per_sample_output_buffers
  38  * Ray_coop ---------------------------------------------|                                      |--- ray_state
  39  * PathState_coop ---------------------------------------|                                      |--- Queue_data (QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS)
  40  * L_transparent_coop -----------------------------------|                                      |--- Queue_data (QUEUE_ACTIVE_AND_REGENERATED_RAYS)
  41  * ray_state --------------------------------------------|                                      |--- Queue_index (QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS)
  42  * Queue_data (QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS) ----|                                      |--- Queue_index (QUEUE_ACTIVE_AND_REGENERATED_RAYS)
  43  * Queue_index (QUEUE_ACTIVE_AND_REGENERATED_RAYS) ------|                                      |--- work_array
  44  * parallel_samples -------------------------------------|                                      |--- PathState_coop
  45  * end_sample -------------------------------------------|                                      |--- throughput_coop
  46  * kg (globals) -----------------------------------------|                                      |--- rng_coop
  47  * rng_state --------------------------------------------|                                      |--- Ray
  48  * PathRadiance_coop ------------------------------------|                                      |
  49  * sw ---------------------------------------------------|                                      |
  50  * sh ---------------------------------------------------|                                      |
  51  * sx ---------------------------------------------------|                                      |
  52  * sy ---------------------------------------------------|                                      |
  53  * stride -----------------------------------------------|                                      |
  54  * work_array -------------------------------------------|                                      |--- work_array
  55  * queuesize --------------------------------------------|                                      |
  56  * start_sample -----------------------------------------|                                      |--- work_pool_wgs
  57  * work_pool_wgs ----------------------------------------|                                      |
  58  * num_samples ------------------------------------------|                                      |
  59  *
  60  * note on sd : sd argument is neither an input nor an output for this kernel. It is just filled and consumed here itself.
  61  * Note on Queues :
  62  * This kernel fetches rays from QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS queue.
  63  *
  64  * State of queues when this kernel is called :
  65  * At entry,
  66  * QUEUE_ACTIVE_AND_REGENERATED_RAYS will be filled with RAY_ACTIVE rays
  67  * QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be filled with RAY_UPDATE_BUFFER, RAY_HIT_BACKGROUND, RAY_TO_REGENERATE rays
  68  * At exit,
  69  * QUEUE_ACTIVE_AND_REGENERATED_RAYS will be filled with RAY_ACTIVE and RAY_REGENERATED rays
  70  * QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be empty
  71  */
  72 ccl_device char kernel_background_buffer_update(
  73         KernelGlobals *kg,
  74         ccl_global float *per_sample_output_buffers,
  75         ccl_global uint *rng_state,
  76         ccl_global uint *rng_coop,             /* Required for buffer Update */
  77         ccl_global float3 *throughput_coop,    /* Required for background hit processing */
  78         PathRadiance *PathRadiance_coop,       /* Required for background hit processing and buffer Update */
  79         ccl_global Ray *Ray_coop,              /* Required for background hit processing */
  80         ccl_global PathState *PathState_coop,  /* Required for background hit processing */
  81         ccl_global float *L_transparent_coop,  /* Required for background hit processing and buffer Update */
  82         ccl_global char *ray_state,            /* Stores information on the current state of a ray */
  83         int sw, int sh, int sx, int sy, int stride,
  84         int rng_state_offset_x,
  85         int rng_state_offset_y,
  86         int rng_state_stride,
  87         ccl_global unsigned int *work_array,   /* Denotes work of each ray */
  88         int end_sample,
  89         int start_sample,
  90 #ifdef __WORK_STEALING__
  91         ccl_global unsigned int *work_pool_wgs,
  92         unsigned int num_samples,
  93 #endif
  94 #ifdef __KERNEL_DEBUG__
  95         DebugData *debugdata_coop,
  96 #endif
  97         int parallel_samples,                  /* Number of samples to be processed in parallel */
  98         int ray_index)
  99 {
 100         char enqueue_flag = 0;
 101 #ifdef __KERNEL_DEBUG__
 102         DebugData *debug_data = &debugdata_coop[ray_index];
 103 #endif
 104         ccl_global PathState *state = &PathState_coop[ray_index];
 105         PathRadiance *L = L = &PathRadiance_coop[ray_index];
 106         ccl_global Ray *ray = &Ray_coop[ray_index];
 107         ccl_global float3 *throughput = &throughput_coop[ray_index];
 108         ccl_global float *L_transparent = &L_transparent_coop[ray_index];
 109         ccl_global uint *rng = &rng_coop[ray_index];
 110
 111 #ifdef __WORK_STEALING__
 112         unsigned int my_work;
 113         ccl_global float *initial_per_sample_output_buffers;
 114         ccl_global uint *initial_rng;
 115 #endif
 116         unsigned int sample;
 117         unsigned int tile_x;
 118         unsigned int tile_y;
 119         unsigned int pixel_x;
 120         unsigned int pixel_y;
 121         unsigned int my_sample_tile;
 122
 123 #ifdef __WORK_STEALING__
 124         my_work = work_array[ray_index];
 125         sample = get_my_sample(my_work, sw, sh, parallel_samples, ray_index) + start_sample;
 126         get_pixel_tile_position(&pixel_x, &pixel_y,
 127                                 &tile_x, &tile_y,
 128                                 my_work,
 129                                 sw, sh, sx, sy,
 130                                 parallel_samples,
 131                                 ray_index);
 132         my_sample_tile = 0;
 133         initial_per_sample_output_buffers = per_sample_output_buffers;
 134         initial_rng = rng_state;
 135 #else  /* __WORK_STEALING__ */
 136         sample = work_array[ray_index];
 137         int tile_index = ray_index / parallel_samples;
 138         /* buffer and rng_state's stride is "stride". Find x and y using ray_index */
 139         tile_x = tile_index % sw;
 140         tile_y = tile_index / sw;
 141         my_sample_tile = ray_index - (tile_index * parallel_samples);
 142 #endif  /* __WORK_STEALING__ */
 143
 144         rng_state += (rng_state_offset_x + tile_x) + (rng_state_offset_y + tile_y) * rng_state_stride;
 145         per_sample_output_buffers += (((tile_x + (tile_y * stride)) * parallel_samples) + my_sample_tile) * kernel_data.film.pass_stride;
 146
 147         if(IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) {
 148                 /* eval background shader if nothing hit */
 149                 if(kernel_data.background.transparent && (state->flag & PATH_RAY_CAMERA)) {
 150                         *L_transparent = (*L_transparent) + average((*throughput));
 151 #ifdef __PASSES__
 152                         if(!(kernel_data.film.pass_flag & PASS_BACKGROUND))
 153 #endif
 154                                 ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER);
 155                 }
 156
 157                 if(IS_STATE(ray_state, ray_index, RAY_HIT_BACKGROUND)) {
 158 #ifdef __BACKGROUND__
 159                         /* sample background shader */
 160                         float3 L_background = indirect_background(kg, state, ray);
 161                         path_radiance_accum_background(L, (*throughput), L_background, state->bounce);
 162 #endif
 163                         ASSIGN_RAY_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER);
 164                 }
 165         }
 166
 167         if(IS_STATE(ray_state, ray_index, RAY_UPDATE_BUFFER)) {
 168                 float3 L_sum = path_radiance_clamp_and_sum(kg, L);
 169                 kernel_write_light_passes(kg, per_sample_output_buffers, L, sample);
 170 #ifdef __KERNEL_DEBUG__
 171                 kernel_write_debug_passes(kg, per_sample_output_buffers, state, debug_data, sample);
 172 #endif
 173                 float4 L_rad = make_float4(L_sum.x, L_sum.y, L_sum.z, 1.0f - (*L_transparent));
 174
 175                 /* accumulate result in output buffer */
 176                 kernel_write_pass_float4(per_sample_output_buffers, sample, L_rad);
 177                 path_rng_end(kg, rng_state, *rng);
 178
 179                 ASSIGN_RAY_STATE(ray_state, ray_index, RAY_TO_REGENERATE);
 180         }
 181
 182         if(IS_STATE(ray_state, ray_index, RAY_TO_REGENERATE)) {
 183 #ifdef __WORK_STEALING__
 184                 /* We have completed current work; So get next work */
 185                 int valid_work = get_next_work(work_pool_wgs, &my_work, sw, sh, num_samples, parallel_samples, ray_index);
 186                 if(!valid_work) {
 187                         /* If work is invalid, this means no more work is available and the thread may exit */
 188                         ASSIGN_RAY_STATE(ray_state, ray_index, RAY_INACTIVE);
 189                 }
 190 #else  /* __WORK_STEALING__ */
 191                 if((sample + parallel_samples) >= end_sample) {
 192                         ASSIGN_RAY_STATE(ray_state, ray_index, RAY_INACTIVE);
 193                 }
 194 #endif  /* __WORK_STEALING__ */
 195
 196                 if(IS_STATE(ray_state, ray_index, RAY_TO_REGENERATE)) {
 197 #ifdef __WORK_STEALING__
 198                         work_array[ray_index] = my_work;
 199                         /* Get the sample associated with the current work */
 200                         sample = get_my_sample(my_work, sw, sh, parallel_samples, ray_index) + start_sample;
 201                         /* Get pixel and tile position associated with current work */
 202                         get_pixel_tile_position(&pixel_x, &pixel_y, &tile_x, &tile_y, my_work, sw, sh, sx, sy, parallel_samples, ray_index);
 203                         my_sample_tile = 0;
 204
 205                         /* Remap rng_state according to the current work */
 206                         rng_state = initial_rng + ((rng_state_offset_x + tile_x) + (rng_state_offset_y + tile_y) * rng_state_stride);
 207                         /* Remap per_sample_output_buffers according to the current work */
 208                         per_sample_output_buffers = initial_per_sample_output_buffers
 209                                 + (((tile_x + (tile_y * stride)) * parallel_samples) + my_sample_tile) * kernel_data.film.pass_stride;
 210 #else  /* __WORK_STEALING__ */
 211                         work_array[ray_index] = sample + parallel_samples;
 212                         sample = work_array[ray_index];
 213
 214                         /* Get ray position from ray index */
 215                         pixel_x = sx + ((ray_index / parallel_samples) % sw);
 216                         pixel_y = sy + ((ray_index / parallel_samples) / sw);
 217 #endif  /* __WORK_STEALING__ */
 218
 219                         /* Initialize random numbers and ray. */
 220                         kernel_path_trace_setup(kg, rng_state, sample, pixel_x, pixel_y, rng, ray);
 221
 222                         if(ray->t != 0.0f) {
 223                                 /* Initialize throughput, L_transparent, Ray, PathState;
 224                                  * These rays proceed with path-iteration.
 225                                  */
 226                                 *throughput = make_float3(1.0f, 1.0f, 1.0f);
 227                                 *L_transparent = 0.0f;
 228                                 path_radiance_init(L, kernel_data.film.use_light_pass);
 229                                 path_state_init(kg, state, rng, sample, ray);
 230 #ifdef __KERNEL_DEBUG__
 231                                 debug_data_init(debug_data);
 232 #endif
 233                                 ASSIGN_RAY_STATE(ray_state, ray_index, RAY_REGENERATED);
 234                                 enqueue_flag = 1;
 235                         } else {
 236                                 /* These rays do not participate in path-iteration. */
 237                                 float4 L_rad = make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 238                                 /* Accumulate result in output buffer. */
 239                                 kernel_write_pass_float4(per_sample_output_buffers, sample, L_rad);
 240                                 path_rng_end(kg, rng_state, *rng);
 241
 242                                 ASSIGN_RAY_STATE(ray_state, ray_index, RAY_TO_REGENERATE);
 243                         }
 244                 }
 245         }
 246         return enqueue_flag;
 247 }