The improvements over the old implementation are: - Improved history reprojection filter (catmull-rom) - Use proper velocity for history reprojection. - History clipping is now done in YCoCg color space using better algorithm. - Velocity is dilated to keep correct edge anti-aliasing on moving objects. As a result, the 3x3 blocks that made the image smoother in the previous implementation are no longer visible is replaced by correct antialiasing. This removes the velocity resolve pass in order to reduce the bandwidth usage. The velocities are just resolved as they are loadded in the film pass.
200 lines
6.1 KiB
C++
200 lines
6.1 KiB
C++
/* SPDX-License-Identifier: GPL-2.0-or-later
|
|
* Copyright 2021 Blender Foundation.
|
|
*/
|
|
|
|
/** \file
|
|
* \ingroup eevee
|
|
*
|
|
* A view is either:
|
|
* - The entire main view.
|
|
* - A fragment of the main view (for panoramic projections).
|
|
* - A shadow map view.
|
|
* - A light-probe view (either planar, cube-map, irradiance grid).
|
|
*
|
|
* A pass is a container for scene data. It is view agnostic but has specific logic depending on
|
|
* its type. Passes are shared between views.
|
|
*/
|
|
|
|
#include "BKE_global.h"
|
|
#include "DRW_render.h"
|
|
|
|
#include "eevee_instance.hh"
|
|
|
|
#include "eevee_view.hh"
|
|
|
|
namespace blender::eevee {
|
|
|
|
/* -------------------------------------------------------------------- */
|
|
/** \name ShadingView
|
|
* \{ */
|
|
|
|
void ShadingView::init()
|
|
{
|
|
// dof_.init();
|
|
// mb_.init();
|
|
}
|
|
|
|
void ShadingView::sync()
|
|
{
|
|
int2 render_extent = inst_.film.render_extent_get();
|
|
|
|
if (false /* inst_.camera.is_panoramic() */) {
|
|
int64_t render_pixel_count = render_extent.x * (int64_t)render_extent.y;
|
|
/* Divide pixel count between the 6 views. Rendering to a square target. */
|
|
extent_[0] = extent_[1] = ceilf(sqrtf(1 + (render_pixel_count / 6)));
|
|
/* TODO(@fclem): Clip unused views here. */
|
|
is_enabled_ = true;
|
|
}
|
|
else {
|
|
extent_ = render_extent;
|
|
/* Only enable -Z view. */
|
|
is_enabled_ = (StringRefNull(name_) == "negZ_view");
|
|
}
|
|
|
|
if (!is_enabled_) {
|
|
return;
|
|
}
|
|
|
|
/* Create views. */
|
|
const CameraData &cam = inst_.camera.data_get();
|
|
|
|
float4x4 viewmat, winmat;
|
|
const float(*viewmat_p)[4] = viewmat.ptr(), (*winmat_p)[4] = winmat.ptr();
|
|
if (false /* inst_.camera.is_panoramic() */) {
|
|
/* TODO(@fclem) Over-scans. */
|
|
/* For now a mandatory 5% over-scan for DoF. */
|
|
float side = cam.clip_near * 1.05f;
|
|
float near = cam.clip_near;
|
|
float far = cam.clip_far;
|
|
perspective_m4(winmat.ptr(), -side, side, -side, side, near, far);
|
|
viewmat = face_matrix_ * cam.viewmat;
|
|
}
|
|
else {
|
|
viewmat_p = cam.viewmat.ptr();
|
|
winmat_p = cam.winmat.ptr();
|
|
}
|
|
|
|
main_view_ = DRW_view_create(viewmat_p, winmat_p, nullptr, nullptr, nullptr);
|
|
sub_view_ = DRW_view_create_sub(main_view_, viewmat_p, winmat_p);
|
|
render_view_ = DRW_view_create_sub(main_view_, viewmat_p, winmat_p);
|
|
|
|
// dof_.sync(winmat_p, extent_);
|
|
// mb_.sync(extent_);
|
|
// rt_buffer_opaque_.sync(extent_);
|
|
// rt_buffer_refract_.sync(extent_);
|
|
// inst_.hiz_back.view_sync(extent_);
|
|
// inst_.hiz_front.view_sync(extent_);
|
|
// inst_.gbuffer.view_sync(extent_);
|
|
|
|
postfx_tx_.sync();
|
|
}
|
|
|
|
void ShadingView::render()
|
|
{
|
|
if (!is_enabled_) {
|
|
return;
|
|
}
|
|
|
|
/* Query temp textures and create framebuffers. */
|
|
/* HACK: View name should be unique and static.
|
|
* With this, we can reuse the same texture across views. */
|
|
DrawEngineType *owner = (DrawEngineType *)name_;
|
|
|
|
RenderBuffers &rbufs = inst_.render_buffers;
|
|
rbufs.acquire(extent_, owner);
|
|
combined_fb_.ensure(GPU_ATTACHMENT_TEXTURE(rbufs.depth_tx),
|
|
GPU_ATTACHMENT_TEXTURE(rbufs.combined_tx));
|
|
prepass_fb_.ensure(GPU_ATTACHMENT_TEXTURE(rbufs.depth_tx),
|
|
GPU_ATTACHMENT_TEXTURE(rbufs.vector_tx));
|
|
|
|
update_view();
|
|
|
|
DRW_stats_group_start(name_);
|
|
DRW_view_set_active(render_view_);
|
|
|
|
/* If camera has any motion, compute motion vector in the film pass. Otherwise, we avoid float
|
|
* precision issue by setting the motion of all static geometry to 0. */
|
|
float4 clear_velocity = float4(inst_.velocity.camera_has_motion() ? VELOCITY_INVALID : 0.0f);
|
|
|
|
GPU_framebuffer_bind(prepass_fb_);
|
|
GPU_framebuffer_clear_color(prepass_fb_, clear_velocity);
|
|
/* Alpha stores transmittance. So start at 1. */
|
|
float4 clear_color = {0.0f, 0.0f, 0.0f, 1.0f};
|
|
GPU_framebuffer_bind(combined_fb_);
|
|
GPU_framebuffer_clear_color_depth(combined_fb_, clear_color, 1.0f);
|
|
|
|
inst_.pipelines.world.render();
|
|
|
|
// inst_.pipelines.deferred.render(
|
|
// render_view_, rt_buffer_opaque_, rt_buffer_refract_, depth_tx_, combined_tx_);
|
|
|
|
// inst_.lightprobes.draw_cache_display();
|
|
|
|
// inst_.lookdev.render_overlay(view_fb_);
|
|
|
|
inst_.pipelines.forward.render(
|
|
render_view_, prepass_fb_, combined_fb_, rbufs.depth_tx, rbufs.combined_tx);
|
|
|
|
// inst_.lights.debug_draw(view_fb_);
|
|
// inst_.shadows.debug_draw(view_fb_);
|
|
|
|
// GPUTexture *final_radiance_tx = render_post(combined_tx_);
|
|
|
|
inst_.film.accumulate(sub_view_);
|
|
|
|
rbufs.release();
|
|
postfx_tx_.release();
|
|
|
|
DRW_stats_group_end();
|
|
}
|
|
|
|
GPUTexture *ShadingView::render_post(GPUTexture *input_tx)
|
|
{
|
|
#if 0
|
|
if (!dof_.postfx_enabled() && !mb_.enabled()) {
|
|
return input_tx;
|
|
}
|
|
/* HACK: View name should be unique and static.
|
|
* With this, we can reuse the same texture across views. */
|
|
postfx_tx_.acquire(extent_, GPU_RGBA16F, (void *)name_);
|
|
|
|
GPUTexture *velocity_tx = velocity_.view_vectors_get();
|
|
GPUTexture *output_tx = postfx_tx_;
|
|
|
|
/* Swapping is done internally. Actual output is set to the next input. */
|
|
dof_.render(depth_tx_, &input_tx, &output_tx);
|
|
mb_.render(depth_tx_, velocity_tx, &input_tx, &output_tx);
|
|
#endif
|
|
return input_tx;
|
|
}
|
|
|
|
void ShadingView::update_view()
|
|
{
|
|
float4x4 viewmat, winmat;
|
|
DRW_view_viewmat_get(main_view_, viewmat.ptr(), false);
|
|
DRW_view_winmat_get(main_view_, winmat.ptr(), false);
|
|
|
|
/* TODO(fclem): Mixed-resolution rendering: We need to make sure we render with exactly the same
|
|
* distances between pixels to line up render samples and target pixels.
|
|
* So if the target resolution is not a multiple of the resolution divisor, we need to make the
|
|
* projection window bigger in the +X and +Y directions. */
|
|
|
|
/* Anti-Aliasing / Super-Sampling jitter. */
|
|
float2 jitter = inst_.film.pixel_jitter_get() / float2(extent_);
|
|
|
|
window_translate_m4(winmat.ptr(), winmat.ptr(), UNPACK2(jitter));
|
|
DRW_view_update_sub(sub_view_, viewmat.ptr(), winmat.ptr());
|
|
|
|
/* FIXME(fclem): The offset may be is noticeably large and the culling might make object pop
|
|
* out of the blurring radius. To fix this, use custom enlarged culling matrix. */
|
|
// dof_.jitter_apply(winmat, viewmat);
|
|
DRW_view_update_sub(render_view_, viewmat.ptr(), winmat.ptr());
|
|
|
|
// inst_.lightprobes.set_view(render_view_, extent_);
|
|
// inst_.lights.set_view(render_view_, extent_, !inst_.use_scene_lights());
|
|
}
|
|
|
|
/** \} */
|
|
|
|
} // namespace blender::eevee
|