Page MenuHome
Differential D13361 Diff 45286 intern/cycles/kernel/device/gpu/kernel.h

Changeset View

Standalone View

View Options

intern/cycles/kernel/device/gpu/kernel.h

Show All 15 Lines
/* Common GPU kernels. */ /* Common GPU kernels. */
#include "kernel/device/gpu/parallel_active_index.h" #include "kernel/device/gpu/parallel_active_index.h"
#include "kernel/device/gpu/parallel_prefix_sum.h" #include "kernel/device/gpu/parallel_prefix_sum.h"
#include "kernel/device/gpu/parallel_sorted_index.h" #include "kernel/device/gpu/parallel_sorted_index.h"
#include "kernel/device/gpu/work_stealing.h" #include "kernel/device/gpu/work_stealing.h"
/* Include constant tables before entering Metal's context class scope (context_begin.h) */
#include "kernel/tables.h"
#ifdef __KERNEL_METAL__ #ifdef __KERNEL_METAL__
# include "kernel/device/metal/context_begin.h" # include "kernel/device/metal/context_begin.h"
#endif #endif
#include "kernel/integrator/state.h" #include "kernel/integrator/state.h"
#include "kernel/integrator/state_flow.h" #include "kernel/integrator/state_flow.h"
#include "kernel/integrator/state_util.h" #include "kernel/integrator/state_util.h"
▲ Show 20 Lines • Show All 427 Lines • ▼ Show 20 Lines if (x < sw && y < sh) {
converged = ccl_gpu_kernel_call(kernel_adaptive_sampling_convergence_check( converged = ccl_gpu_kernel_call(kernel_adaptive_sampling_convergence_check(
nullptr, render_buffer, sx + x, sy + y, threshold, reset, offset, stride)); nullptr, render_buffer, sx + x, sy + y, threshold, reset, offset, stride));
} }
/* NOTE: All threads specified in the mask must execute the intrinsic. */ /* NOTE: All threads specified in the mask must execute the intrinsic. */
const auto num_active_pixels_mask = ccl_gpu_ballot(!converged); const auto num_active_pixels_mask = ccl_gpu_ballot(!converged);
const int lane_id = ccl_gpu_thread_idx_x % ccl_gpu_warp_size; const int lane_id = ccl_gpu_thread_idx_x % ccl_gpu_warp_size;
if (lane_id == 0) { if (lane_id == 0) {
atomic_fetch_and_add_uint32(num_active_pixels, ccl_gpu_popc(num_active_pixels_mask)); atomic_fetch_and_add_uint32(num_active_pixels, popcount(num_active_pixels_mask));
} }
} }
ccl_gpu_kernel(GPU_KERNEL_BLOCK_NUM_THREADS, GPU_KERNEL_MAX_REGISTERS) ccl_gpu_kernel(GPU_KERNEL_BLOCK_NUM_THREADS, GPU_KERNEL_MAX_REGISTERS)
ccl_gpu_kernel_signature(adaptive_sampling_filter_x, ccl_gpu_kernel_signature(adaptive_sampling_filter_x,
ccl_global float *render_buffer, ccl_global float *render_buffer,
int sx, int sx,
int sy, int sy,
▲ Show 20 Lines • Show All 411 Lines • ▼ Show 20 Linesccl_gpu_kernel(GPU_KERNEL_BLOCK_NUM_THREADS, GPU_KERNEL_MAX_REGISTERS)
if (state < num_states) { if (state < num_states) {
can_split = ccl_gpu_kernel_call(kernel_shadow_catcher_path_can_split(nullptr, state)); can_split = ccl_gpu_kernel_call(kernel_shadow_catcher_path_can_split(nullptr, state));
} }
/* NOTE: All threads specified in the mask must execute the intrinsic. */ /* NOTE: All threads specified in the mask must execute the intrinsic. */
const auto can_split_mask = ccl_gpu_ballot(can_split); const auto can_split_mask = ccl_gpu_ballot(can_split);
const int lane_id = ccl_gpu_thread_idx_x % ccl_gpu_warp_size; const int lane_id = ccl_gpu_thread_idx_x % ccl_gpu_warp_size;
if (lane_id == 0) { if (lane_id == 0) {
atomic_fetch_and_add_uint32(num_possible_splits, ccl_gpu_popc(can_split_mask)); atomic_fetch_and_add_uint32(num_possible_splits, popcount(can_split_mask));
} }
} }