intern/cycles/kernel/kernels/opencl/kernel_lamp_emission.cl

   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 "split/kernel_lamp_emission.h"
  18
  19 __kernel void kernel_ocl_path_trace_lamp_emission(
  20         ccl_global char *kg,
  21         ccl_constant KernelData *data,
  22         ccl_global char *sd,                   /* Required for lamp emission */
  23         ccl_global float3 *throughput_coop,    /* Required for lamp emission */
  24         PathRadiance *PathRadiance_coop,       /* Required for lamp emission */
  25         ccl_global Ray *Ray_coop,              /* Required for lamp emission */
  26         ccl_global PathState *PathState_coop,  /* Required for lamp emission */
  27         Intersection *Intersection_coop,       /* Required for lamp emission */
  28         ccl_global char *ray_state,            /* Denotes the state of each ray */
  29         int sw, int sh,
  30         ccl_global int *Queue_data,            /* Memory for queues */
  31         ccl_global int *Queue_index,           /* Tracks the number of elements in queues */
  32         int queuesize,                         /* Size (capacity) of queues */
  33         ccl_global char *use_queues_flag,      /* Used to decide if this kernel should use
  34                                                 * queues to fetch ray index
  35                                                 */
  36         int parallel_samples)                  /* Number of samples to be processed in parallel */
  37 {
  38         int x = get_global_id(0);
  39         int y = get_global_id(1);
  40
  41         /* We will empty this queue in this kernel. */
  42         if(get_global_id(0) == 0 && get_global_id(1) == 0) {
  43                 Queue_index[QUEUE_ACTIVE_AND_REGENERATED_RAYS] = 0;
  44         }
  45         /* Fetch use_queues_flag. */
  46         ccl_local char local_use_queues_flag;
  47         if(get_local_id(0) == 0 && get_local_id(1) == 0) {
  48                 local_use_queues_flag = use_queues_flag[0];
  49         }
  50         barrier(CLK_LOCAL_MEM_FENCE);
  51
  52         int ray_index;
  53         if(local_use_queues_flag) {
  54                 int thread_index = get_global_id(1) * get_global_size(0) + get_global_id(0);
  55                 ray_index = get_ray_index(thread_index,
  56                                           QUEUE_ACTIVE_AND_REGENERATED_RAYS,
  57                                           Queue_data,
  58                                           queuesize,
  59                                           1);
  60                 if(ray_index == QUEUE_EMPTY_SLOT) {
  61                         return;
  62                 }
  63         } else {
  64                 if(x < (sw * parallel_samples) && y < sh) {
  65                         ray_index = x + y * (sw * parallel_samples);
  66                 } else {
  67                         return;
  68                 }
  69         }
  70
  71         kernel_lamp_emission((KernelGlobals *)kg,
  72                              (ShaderData *)sd,
  73                              throughput_coop,
  74                              PathRadiance_coop,
  75                              Ray_coop,
  76                              PathState_coop,
  77                              Intersection_coop,
  78                              ray_state,
  79                              sw, sh,
  80                              use_queues_flag,
  81                              parallel_samples,
  82                              ray_index);
  83 }