diff options
Diffstat (limited to 'intern/cycles/kernel/split/kernel_shader_eval.h')
| -rw-r--r-- | intern/cycles/kernel/split/kernel_shader_eval.h | 91 |
1 files changed, 46 insertions, 45 deletions
diff --git a/intern/cycles/kernel/split/kernel_shader_eval.h b/intern/cycles/kernel/split/kernel_shader_eval.h index cef64bf5f36..0f1696e34a0 100644 --- a/intern/cycles/kernel/split/kernel_shader_eval.h +++ b/intern/cycles/kernel/split/kernel_shader_eval.h @@ -14,57 +14,58 @@ * limitations under the License. */ -#include "kernel_split_common.h" +CCL_NAMESPACE_BEGIN -/* Note on kernel_shader_eval kernel - * This kernel is the 5th kernel in the ray tracing logic. This is - * the 4rd kernel in path iteration. This kernel sets up the ShaderData - * structure from the values computed by the previous kernels. It also identifies - * the rays of state RAY_TO_REGENERATE and enqueues them in QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS queue. +/* This kernel sets up the ShaderData structure from the values computed + * by the previous kernels. * - * The input and output of the kernel is as follows, - * rng_coop -------------------------------------------|--- kernel_shader_eval --|--- sd - * Ray_coop -------------------------------------------| |--- Queue_data (QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS) - * PathState_coop -------------------------------------| |--- Queue_index (QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS) - * Intersection_coop ----------------------------------| | - * Queue_data (QUEUE_ACTIVE_AND_REGENERATD_RAYS)-------| | - * Queue_index(QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS)---| | - * ray_state ------------------------------------------| | - * kg (globals) ---------------------------------------| | - * queuesize ------------------------------------------| | - * - * Note on Queues : - * This kernel reads from the QUEUE_ACTIVE_AND_REGENERATED_RAYS queue and processes - * only the rays of state RAY_ACTIVE; - * State of queues when this kernel is called, - * at entry, - * QUEUE_ACTIVE_AND_REGENERATED_RAYS will be filled with RAY_ACTIVE and RAY_REGENERATED rays - * QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be empty. - * at exit, - * QUEUE_ACTIVE_AND_REGENERATED_RAYS will be filled with RAY_ACTIVE and RAY_REGENERATED rays - * QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS will be filled with RAY_TO_REGENERATE rays + * It also identifies the rays of state RAY_TO_REGENERATE and enqueues them + * in QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS queue. */ -ccl_device void kernel_shader_eval( - KernelGlobals *kg, - ShaderData *sd, /* Output ShaderData structure to be filled */ - ccl_global uint *rng_coop, /* Required for rbsdf calculation */ - ccl_global Ray *Ray_coop, /* Required for setting up shader from ray */ - ccl_global PathState *PathState_coop, /* Required for all functions in this kernel */ - Intersection *Intersection_coop, /* Required for setting up shader from ray */ - ccl_global char *ray_state, /* Denotes the state of each ray */ - int ray_index) +ccl_device void kernel_shader_eval(KernelGlobals *kg, + ccl_local_param unsigned int *local_queue_atomics) { - if(IS_STATE(ray_state, ray_index, RAY_ACTIVE)) { - Intersection *isect = &Intersection_coop[ray_index]; - ccl_global uint *rng = &rng_coop[ray_index]; - ccl_global PathState *state = &PathState_coop[ray_index]; - Ray ray = Ray_coop[ray_index]; + /* Enqeueue RAY_TO_REGENERATE rays into QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS queue. */ + if(ccl_local_id(0) == 0 && ccl_local_id(1) == 0) { + *local_queue_atomics = 0; + } + ccl_barrier(CCL_LOCAL_MEM_FENCE); + + int ray_index = ccl_global_id(1) * ccl_global_size(0) + ccl_global_id(0); + ray_index = get_ray_index(kg, ray_index, + QUEUE_ACTIVE_AND_REGENERATED_RAYS, + kernel_split_state.queue_data, + kernel_split_params.queue_size, + 0); + + char enqueue_flag = 0; + if((ray_index != QUEUE_EMPTY_SLOT) && IS_STATE(kernel_split_state.ray_state, ray_index, RAY_TO_REGENERATE)) { + enqueue_flag = 1; + } + + enqueue_ray_index_local(ray_index, + QUEUE_HITBG_BUFF_UPDATE_TOREGEN_RAYS, + enqueue_flag, + kernel_split_params.queue_size, + local_queue_atomics, + kernel_split_state.queue_data, + kernel_split_params.queue_index); + + /* Continue on with shader evaluation. */ + if((ray_index != QUEUE_EMPTY_SLOT) && IS_STATE(kernel_split_state.ray_state, ray_index, RAY_ACTIVE)) { + Intersection isect = kernel_split_state.isect[ray_index]; + RNG rng = kernel_split_state.rng[ray_index]; + ccl_global PathState *state = &kernel_split_state.path_state[ray_index]; + Ray ray = kernel_split_state.ray[ray_index]; shader_setup_from_ray(kg, - sd, - isect, + &kernel_split_state.sd[ray_index], + &isect, &ray); - float rbsdf = path_state_rng_1D_for_decision(kg, rng, state, PRNG_BSDF); - shader_eval_surface(kg, sd, rng, state, rbsdf, state->flag, SHADER_CONTEXT_MAIN); + float rbsdf = path_state_rng_1D_for_decision(kg, &rng, state, PRNG_BSDF); + shader_eval_surface(kg, &kernel_split_state.sd[ray_index], &rng, state, rbsdf, state->flag, SHADER_CONTEXT_MAIN); + kernel_split_state.rng[ray_index] = rng; } } + +CCL_NAMESPACE_END |