Several GI related optimizations and fixes

-SDFGI direct light is done over many frames
-SDFGI Changed settings for rays/frame
-SDFGI Misc optimizations
-SDFGI Bug fix on probe scroll

-GIProbe was not working, got it to work again
-GIProbe dynamic objects were not working, fixed

-Added a half size GI option.
This commit is contained in:
reduz 2021-01-22 20:50:24 -03:00
parent 6ddfc8e718
commit 6fe342478b
21 changed files with 490 additions and 263 deletions

View file

@ -516,6 +516,9 @@ void EditorNode::_notification(int p_what) {
RS::get_singleton()->environment_set_volumetric_fog_positional_shadow_shrink_size(GLOBAL_GET("rendering/volumetric_fog/positional_shadow_shrink"));
RS::get_singleton()->canvas_set_shadow_texture_size(GLOBAL_GET("rendering/quality/2d_shadow_atlas/size"));
bool use_half_res_gi = GLOBAL_DEF("rendering/quality/gi/use_half_resolution", false);
RS::get_singleton()->gi_set_use_half_resolution(use_half_res_gi);
bool snap_2d_transforms = GLOBAL_GET("rendering/quality/2d/snap_2d_transforms_to_pixel");
scene_root->set_snap_2d_transforms_to_pixel(snap_2d_transforms);
bool snap_2d_vertices = GLOBAL_GET("rendering/quality/2d/snap_2d_vertices_to_pixel");

View file

@ -171,6 +171,8 @@ static bool disable_render_loop = false;
static int fixed_fps = -1;
static bool print_fps = false;
bool profile_gpu = false;
/* Helper methods */
// Used by Mono module, should likely be registered in Engine singleton instead
@ -357,6 +359,7 @@ void Main::print_help(const char *p_binary) {
OS::get_singleton()->print(" --disable-crash-handler Disable crash handler when supported by the platform code.\n");
OS::get_singleton()->print(" --fixed-fps <fps> Force a fixed number of frames per second. This setting disables real-time synchronization.\n");
OS::get_singleton()->print(" --print-fps Print the frames per second to the stdout.\n");
OS::get_singleton()->print(" --profile-gpu Show a simple profile of the tasks that took more time during frame rendering.\n");
OS::get_singleton()->print("\n");
OS::get_singleton()->print("Standalone tools:\n");
@ -1006,6 +1009,8 @@ Error Main::setup(const char *execpath, int argc, char *argv[], bool p_second_ph
}
} else if (I->get() == "--print-fps") {
print_fps = true;
} else if (I->get() == "--profile-gpu") {
profile_gpu = true;
} else if (I->get() == "--disable-crash-handler") {
OS::get_singleton()->disable_crash_handler();
} else if (I->get() == "--skip-breakpoints") {
@ -1574,6 +1579,10 @@ Error Main::setup2(Thread::ID p_main_tid_override) {
rendering_server->init();
rendering_server->set_render_loop_enabled(!disable_render_loop);
if (profile_gpu) {
rendering_server->set_print_gpu_profile(true);
}
OS::get_singleton()->initialize_joypads();
/* Initialize Audio Driver */

View file

@ -520,7 +520,10 @@ String GIProbe::get_configuration_warning() const {
warning += "\n\n";
}
warning += TTR("GIProbes are not supported by the GLES2 video driver.\nUse a BakedLightmap instead.");
} else if (probe_data.is_null()) {
warning += TTR("No GIProbe data set, so this node is disabled. Bake static objects to enable GI.");
}
return warning;
}

View file

@ -583,19 +583,6 @@ void RendererSceneRenderForward::RenderBufferDataForward::ensure_specular() {
}
}
void RendererSceneRenderForward::RenderBufferDataForward::ensure_gi() {
if (!reflection_buffer.is_valid()) {
RD::TextureFormat tf;
tf.format = RD::DATA_FORMAT_R16G16B16A16_SFLOAT;
tf.width = width;
tf.height = height;
tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
reflection_buffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
ambient_buffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
}
}
void RendererSceneRenderForward::RenderBufferDataForward::ensure_giprobe() {
if (!giprobe_buffer.is_valid()) {
RD::TextureFormat tf;
@ -633,16 +620,6 @@ void RendererSceneRenderForward::RenderBufferDataForward::ensure_giprobe() {
}
void RendererSceneRenderForward::RenderBufferDataForward::clear() {
if (ambient_buffer != RID() && ambient_buffer != color) {
RD::get_singleton()->free(ambient_buffer);
ambient_buffer = RID();
}
if (reflection_buffer != RID() && reflection_buffer != specular) {
RD::get_singleton()->free(reflection_buffer);
reflection_buffer = RID();
}
if (giprobe_buffer != RID()) {
RD::get_singleton()->free(giprobe_buffer);
giprobe_buffer = RID();
@ -1403,6 +1380,7 @@ void RendererSceneRenderForward::_fill_render_list(const PagedArray<GeometryInst
}
inst->push_constant.gi_offset = probe0_index | (probe1_index << 16);
flags |= INSTANCE_DATA_FLAG_USE_GIPROBE;
uses_gi = true;
} else {
if (p_using_sdfgi && inst->can_sdfgi) {
@ -1553,7 +1531,6 @@ void RendererSceneRenderForward::_render_scene(RID p_render_buffer, const Transf
if (!low_end && p_gi_probes.size() > 0) {
using_giprobe = true;
render_buffer->ensure_gi();
}
if (!p_environment.is_valid() && using_giprobe) {
@ -1563,7 +1540,6 @@ void RendererSceneRenderForward::_render_scene(RID p_render_buffer, const Transf
if (environment_is_sdfgi_enabled(p_environment)) {
depth_pass_mode = using_giprobe ? PASS_MODE_DEPTH_NORMAL_ROUGHNESS_GIPROBE : PASS_MODE_DEPTH_NORMAL_ROUGHNESS; // also giprobe
using_sdfgi = true;
render_buffer->ensure_gi();
} else {
depth_pass_mode = using_giprobe ? PASS_MODE_DEPTH_NORMAL_ROUGHNESS_GIPROBE : PASS_MODE_DEPTH_NORMAL_ROUGHNESS;
}
@ -1739,7 +1715,7 @@ void RendererSceneRenderForward::_render_scene(RID p_render_buffer, const Transf
}
if (using_sdfgi || using_giprobe) {
_process_gi(p_render_buffer, render_buffer->normal_roughness_buffer, render_buffer->ambient_buffer, render_buffer->reflection_buffer, render_buffer->giprobe_buffer, p_environment, p_cam_projection, p_cam_transform, p_gi_probes);
_process_gi(p_render_buffer, render_buffer->normal_roughness_buffer, render_buffer->giprobe_buffer, p_environment, p_cam_projection, p_cam_transform, p_gi_probes);
}
_setup_environment(p_environment, p_render_buffer, p_cam_projection, p_cam_transform, p_reflection_probe, p_reflection_probe.is_valid(), screen_size, p_cluster_size, p_max_cluster_elements, p_shadow_atlas, !p_reflection_probe.is_valid(), p_default_bg_color, p_cam_projection.get_z_near(), p_cam_projection.get_z_far(), p_render_buffer.is_valid());
@ -1930,7 +1906,7 @@ void RendererSceneRenderForward::_render_material(const Transform &p_cam_transfo
render_pass++;
scene_state.ubo.dual_paraboloid_side = 0;
scene_state.ubo.material_uv2_mode = true;
scene_state.ubo.material_uv2_mode = false;
_setup_environment(RID(), RID(), p_cam_projection, p_cam_transform, RID(), true, Vector2(1, 1), 1, 32, RID(), false, Color(), 0, 0);
@ -2395,7 +2371,8 @@ RID RendererSceneRenderForward::_setup_render_pass_uniform_set(RID p_render_buff
RD::Uniform u;
u.binding = 10;
u.uniform_type = RD::UNIFORM_TYPE_TEXTURE;
RID texture = rb && rb->ambient_buffer.is_valid() ? rb->ambient_buffer : storage->texture_rd_get_default(RendererStorageRD::DEFAULT_RD_TEXTURE_BLACK);
RID ambient_buffer = p_render_buffers.is_valid() ? render_buffers_get_gi_ambient_texture(p_render_buffers) : RID();
RID texture = ambient_buffer.is_valid() ? ambient_buffer : storage->texture_rd_get_default(RendererStorageRD::DEFAULT_RD_TEXTURE_BLACK);
u.ids.push_back(texture);
uniforms.push_back(u);
}
@ -2404,7 +2381,8 @@ RID RendererSceneRenderForward::_setup_render_pass_uniform_set(RID p_render_buff
RD::Uniform u;
u.binding = 11;
u.uniform_type = RD::UNIFORM_TYPE_TEXTURE;
RID texture = rb && rb->reflection_buffer.is_valid() ? rb->reflection_buffer : storage->texture_rd_get_default(RendererStorageRD::DEFAULT_RD_TEXTURE_BLACK);
RID reflection_buffer = p_render_buffers.is_valid() ? render_buffers_get_gi_reflection_texture(p_render_buffers) : RID();
RID texture = reflection_buffer.is_valid() ? reflection_buffer : storage->texture_rd_get_default(RendererStorageRD::DEFAULT_RD_TEXTURE_BLACK);
u.ids.push_back(texture);
uniforms.push_back(u);
}
@ -2585,18 +2563,6 @@ RID RendererSceneRenderForward::_render_buffers_get_normal_texture(RID p_render_
return rb->normal_roughness_buffer;
}
RID RendererSceneRenderForward::_render_buffers_get_ambient_texture(RID p_render_buffers) {
RenderBufferDataForward *rb = (RenderBufferDataForward *)render_buffers_get_data(p_render_buffers);
return rb->ambient_buffer;
}
RID RendererSceneRenderForward::_render_buffers_get_reflection_texture(RID p_render_buffers) {
RenderBufferDataForward *rb = (RenderBufferDataForward *)render_buffers_get_data(p_render_buffers);
return rb->reflection_buffer;
}
RendererSceneRenderForward *RendererSceneRenderForward::singleton = nullptr;
void RendererSceneRenderForward::set_time(double p_time, double p_step) {

View file

@ -213,9 +213,6 @@ class RendererSceneRenderForward : public RendererSceneRenderRD {
RID normal_roughness_buffer;
RID giprobe_buffer;
RID ambient_buffer;
RID reflection_buffer;
RS::ViewportMSAA msaa;
RD::TextureSamples texture_samples;
@ -236,7 +233,6 @@ class RendererSceneRenderForward : public RendererSceneRenderRD {
RID render_sdfgi_uniform_set;
void ensure_specular();
void ensure_gi();
void ensure_giprobe();
void clear();
virtual void configure(RID p_color_buffer, RID p_depth_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa);
@ -258,8 +254,6 @@ class RendererSceneRenderForward : public RendererSceneRenderRD {
void _render_buffers_clear_uniform_set(RenderBufferDataForward *rb);
virtual void _render_buffers_uniform_set_changed(RID p_render_buffers);
virtual RID _render_buffers_get_normal_texture(RID p_render_buffers);
virtual RID _render_buffers_get_ambient_texture(RID p_render_buffers);
virtual RID _render_buffers_get_reflection_texture(RID p_render_buffers);
void _update_render_base_uniform_set();
RID _setup_sdfgi_render_pass_uniform_set(RID p_albedo_texture, RID p_emission_texture, RID p_emission_aniso_texture, RID p_geom_facing_texture);

View file

@ -1276,14 +1276,26 @@ void RendererSceneRenderRD::sdfgi_update_probes(RID p_render_buffers, RID p_envi
push_constant.multibounce = rb->sdfgi->uses_multibounce;
push_constant.y_mult = rb->sdfgi->y_mult;
push_constant.process_offset = 0;
push_constant.process_increment = 1;
for (uint32_t i = 0; i < rb->sdfgi->cascades.size(); i++) {
SDFGI::Cascade &cascade = rb->sdfgi->cascades[i];
push_constant.light_count = cascade_light_count[i];
push_constant.cascade = i;
if (rb->sdfgi->cascades[i].all_dynamic_lights_dirty || sdfgi_frames_to_update_light == RS::ENV_SDFGI_UPDATE_LIGHT_IN_1_FRAME) {
push_constant.process_offset = 0;
push_constant.process_increment = 1;
} else {
static uint32_t frames_to_update_table[RS::ENV_SDFGI_UPDATE_LIGHT_MAX] = {
1, 2, 4, 8, 16
};
uint32_t frames_to_update = frames_to_update_table[sdfgi_frames_to_update_light];
push_constant.process_offset = RSG::rasterizer->get_frame_number() % frames_to_update;
push_constant.process_increment = frames_to_update;
}
rb->sdfgi->cascades[i].all_dynamic_lights_dirty = false;
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, cascade.sdf_direct_light_uniform_set, 0);
RD::get_singleton()->compute_list_set_push_constant(compute_list, &push_constant, sizeof(SDGIShader::DirectLightPushConstant));
RD::get_singleton()->compute_list_dispatch_indirect(compute_list, cascade.solid_cell_dispatch_buffer, 0);
@ -1301,7 +1313,7 @@ void RendererSceneRenderRD::sdfgi_update_probes(RID p_render_buffers, RID p_envi
push_constant.probe_axis_size = rb->sdfgi->probe_axis_count;
push_constant.history_index = rb->sdfgi->render_pass % rb->sdfgi->history_size;
push_constant.history_size = rb->sdfgi->history_size;
static const uint32_t ray_count[RS::ENV_SDFGI_RAY_COUNT_MAX] = { 8, 16, 32, 64, 96, 128 };
static const uint32_t ray_count[RS::ENV_SDFGI_RAY_COUNT_MAX] = { 4, 8, 16, 32, 64, 96, 128 };
push_constant.ray_count = ray_count[sdfgi_ray_count];
push_constant.ray_bias = rb->sdfgi->probe_bias;
push_constant.image_size[0] = rb->sdfgi->probe_axis_count * rb->sdfgi->probe_axis_count;
@ -1492,13 +1504,35 @@ void RendererSceneRenderRD::_setup_giprobes(RID p_render_buffers, const Transfor
}
}
void RendererSceneRenderRD::_process_gi(RID p_render_buffers, RID p_normal_roughness_buffer, RID p_ambient_buffer, RID p_reflection_buffer, RID p_gi_probe_buffer, RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform, const PagedArray<RID> &p_gi_probes) {
void RendererSceneRenderRD::_process_gi(RID p_render_buffers, RID p_normal_roughness_buffer, RID p_gi_probe_buffer, RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform, const PagedArray<RID> &p_gi_probes) {
RENDER_TIMESTAMP("Render GI");
RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
ERR_FAIL_COND(rb == nullptr);
Environment *env = environment_owner.getornull(p_environment);
if (rb->ambient_buffer.is_null() || rb->using_half_size_gi != gi.half_resolution) {
if (rb->ambient_buffer.is_valid()) {
RD::get_singleton()->free(rb->ambient_buffer);
RD::get_singleton()->free(rb->reflection_buffer);
}
RD::TextureFormat tf;
tf.format = RD::DATA_FORMAT_R16G16B16A16_SFLOAT;
tf.width = rb->width;
tf.height = rb->height;
if (gi.half_resolution) {
tf.width >>= 1;
tf.height >>= 1;
}
tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
rb->reflection_buffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
rb->ambient_buffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
rb->using_half_size_gi = gi.half_resolution;
_render_buffers_uniform_set_changed(p_render_buffers);
}
GI::PushConstant push_constant;
push_constant.screen_size[0] = rb->width;
@ -1693,7 +1727,7 @@ void RendererSceneRenderRD::_process_gi(RID p_render_buffers, RID p_normal_rough
RD::Uniform u;
u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
u.binding = 9;
u.ids.push_back(p_ambient_buffer);
u.ids.push_back(rb->ambient_buffer);
uniforms.push_back(u);
}
@ -1701,7 +1735,7 @@ void RendererSceneRenderRD::_process_gi(RID p_render_buffers, RID p_normal_rough
RD::Uniform u;
u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
u.binding = 10;
u.ids.push_back(p_reflection_buffer);
u.ids.push_back(rb->reflection_buffer);
uniforms.push_back(u);
}
@ -1765,12 +1799,23 @@ void RendererSceneRenderRD::_process_gi(RID p_render_buffers, RID p_normal_rough
rb->gi_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, gi.shader.version_get_shader(gi.shader_version, 0), 0);
}
GI::Mode mode = (use_sdfgi && use_giprobes) ? GI::MODE_COMBINED : (use_sdfgi ? GI::MODE_SDFGI : GI::MODE_GIPROBE);
GI::Mode mode;
if (rb->using_half_size_gi) {
mode = (use_sdfgi && use_giprobes) ? GI::MODE_HALF_RES_COMBINED : (use_sdfgi ? GI::MODE_HALF_RES_SDFGI : GI::MODE_HALF_RES_GIPROBE);
} else {
mode = (use_sdfgi && use_giprobes) ? GI::MODE_COMBINED : (use_sdfgi ? GI::MODE_SDFGI : GI::MODE_GIPROBE);
}
RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, gi.pipelines[mode]);
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, rb->gi_uniform_set, 0);
RD::get_singleton()->compute_list_set_push_constant(compute_list, &push_constant, sizeof(GI::PushConstant));
RD::get_singleton()->compute_list_dispatch_threads(compute_list, rb->width, rb->height, 1, 8, 8, 1);
if (rb->using_half_size_gi) {
RD::get_singleton()->compute_list_dispatch_threads(compute_list, rb->width >> 1, rb->height >> 1, 1, 8, 8, 1);
} else {
RD::get_singleton()->compute_list_dispatch_threads(compute_list, rb->width, rb->height, 1, 8, 8, 1);
}
RD::get_singleton()->compute_list_end();
}
@ -3105,6 +3150,9 @@ void RendererSceneRenderRD::environment_set_sdfgi_ray_count(RS::EnvironmentSDFGI
void RendererSceneRenderRD::environment_set_sdfgi_frames_to_converge(RS::EnvironmentSDFGIFramesToConverge p_frames) {
sdfgi_frames_to_converge = p_frames;
}
void RendererSceneRenderRD::environment_set_sdfgi_frames_to_update_light(RS::EnvironmentSDFGIFramesToUpdateLight p_update) {
sdfgi_frames_to_update_light = p_update;
}
void RendererSceneRenderRD::environment_set_ssr(RID p_env, bool p_enable, int p_max_steps, float p_fade_int, float p_fade_out, float p_depth_tolerance) {
Environment *env = environment_owner.getornull(p_env);
@ -5159,6 +5207,13 @@ void RendererSceneRenderRD::_free_render_buffer_data(RenderBuffers *rb) {
RD::get_singleton()->free(rb->ssr.normal_scaled);
rb->ssr.normal_scaled = RID();
}
if (rb->ambient_buffer.is_valid()) {
RD::get_singleton()->free(rb->ambient_buffer);
RD::get_singleton()->free(rb->reflection_buffer);
rb->ambient_buffer = RID();
rb->reflection_buffer = RID();
}
}
void RendererSceneRenderRD::_process_sss(RID p_render_buffers, const CameraMatrix &p_camera) {
@ -5558,10 +5613,10 @@ void RendererSceneRenderRD::_render_buffers_debug_draw(RID p_render_buffers, RID
effects->copy_to_fb_rect(_render_buffers_get_normal_texture(p_render_buffers), storage->render_target_get_rd_framebuffer(rb->render_target), Rect2(Vector2(), rtsize), false, false);
}
if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_GI_BUFFER && _render_buffers_get_ambient_texture(p_render_buffers).is_valid()) {
if (debug_draw == RS::VIEWPORT_DEBUG_DRAW_GI_BUFFER && rb->ambient_buffer.is_valid()) {
Size2 rtsize = storage->render_target_get_size(rb->render_target);
RID ambient_texture = _render_buffers_get_ambient_texture(p_render_buffers);
RID reflection_texture = _render_buffers_get_reflection_texture(p_render_buffers);
RID ambient_texture = rb->ambient_buffer;
RID reflection_texture = rb->reflection_buffer;
effects->copy_to_fb_rect(ambient_texture, storage->render_target_get_rd_framebuffer(rb->render_target), Rect2(Vector2(), rtsize), false, false, false, true, reflection_texture);
}
}
@ -5753,6 +5808,17 @@ RID RendererSceneRenderRD::render_buffers_get_default_gi_probe_buffer() {
return default_giprobe_buffer;
}
RID RendererSceneRenderRD::render_buffers_get_gi_ambient_texture(RID p_render_buffers) {
RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
ERR_FAIL_COND_V(!rb, RID());
return rb->ambient_buffer;
}
RID RendererSceneRenderRD::render_buffers_get_gi_reflection_texture(RID p_render_buffers) {
RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
ERR_FAIL_COND_V(!rb, RID());
return rb->reflection_buffer;
}
uint32_t RendererSceneRenderRD::render_buffers_get_sdfgi_cascade_count(RID p_render_buffers) const {
const RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
ERR_FAIL_COND_V(!rb, 0);
@ -5939,6 +6005,10 @@ void RendererSceneRenderRD::render_buffers_configure(RID p_render_buffers, RID p
rb->cluster_builder->setup(Size2i(p_width, p_height), max_cluster_elements, rb->depth_texture, storage->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_NEAREST, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED), rb->texture);
}
void RendererSceneRenderRD::gi_set_use_half_resolution(bool p_enable) {
gi.half_resolution = p_enable;
}
void RendererSceneRenderRD::sub_surface_scattering_set_quality(RS::SubSurfaceScatteringQuality p_quality) {
sss_quality = p_quality;
}
@ -7609,6 +7679,9 @@ void RendererSceneRenderRD::render_sdfgi(RID p_render_buffers, int p_region, con
push_constant.scroll[1] = 0;
push_constant.scroll[2] = 0;
}
rb->sdfgi->cascades[cascade].all_dynamic_lights_dirty = true;
push_constant.grid_size = rb->sdfgi->cascade_size;
push_constant.cascade = cascade;
@ -7684,6 +7757,23 @@ void RendererSceneRenderRD::render_sdfgi(RID p_render_buffers, int p_region, con
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, sdfgi_shader.integrate_default_sky_uniform_set, 1);
RD::get_singleton()->compute_list_set_push_constant(compute_list, &ipush_constant, sizeof(SDGIShader::IntegratePushConstant));
RD::get_singleton()->compute_list_dispatch_threads(compute_list, rb->sdfgi->probe_axis_count * rb->sdfgi->probe_axis_count, rb->sdfgi->probe_axis_count, 1, 8, 8, 1);
RD::get_singleton()->compute_list_add_barrier(compute_list);
if (rb->sdfgi->uses_multibounce) {
//multibounce requires this to be stored so direct light can read from it
RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, sdfgi_shader.integrate_pipeline[SDGIShader::INTEGRATE_MODE_STORE]);
//convert to octahedral to store
ipush_constant.image_size[0] *= SDFGI::LIGHTPROBE_OCT_SIZE;
ipush_constant.image_size[1] *= SDFGI::LIGHTPROBE_OCT_SIZE;
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, rb->sdfgi->cascades[cascade].integrate_uniform_set, 0);
RD::get_singleton()->compute_list_bind_uniform_set(compute_list, sdfgi_shader.integrate_default_sky_uniform_set, 1);
RD::get_singleton()->compute_list_set_push_constant(compute_list, &ipush_constant, sizeof(SDGIShader::IntegratePushConstant));
RD::get_singleton()->compute_list_dispatch_threads(compute_list, rb->sdfgi->probe_axis_count * rb->sdfgi->probe_axis_count * SDFGI::LIGHTPROBE_OCT_SIZE, rb->sdfgi->probe_axis_count * SDFGI::LIGHTPROBE_OCT_SIZE, 1, 8, 8, 1);
}
}
//ok finally barrier
@ -7922,25 +8012,8 @@ void RendererSceneRenderRD::render_sdfgi_static_lights(RID p_render_buffers, uin
_sdfgi_update_cascades(p_render_buffers); //need cascades updated for this
RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, sdfgi_shader.direct_light_pipeline[SDGIShader::DIRECT_LIGHT_MODE_STATIC]);
SDGIShader::DirectLightPushConstant dl_push_constant;
dl_push_constant.grid_size[0] = rb->sdfgi->cascade_size;
dl_push_constant.grid_size[1] = rb->sdfgi->cascade_size;
dl_push_constant.grid_size[2] = rb->sdfgi->cascade_size;
dl_push_constant.max_cascades = rb->sdfgi->cascades.size();
dl_push_constant.probe_axis_size = rb->sdfgi->probe_axis_count;
dl_push_constant.multibounce = false; // this is static light, do not multibounce yet
dl_push_constant.y_mult = rb->sdfgi->y_mult;
//all must be processed
dl_push_constant.process_offset = 0;
dl_push_constant.process_increment = 1;
SDGIShader::Light lights[SDFGI::MAX_STATIC_LIGHTS];
uint32_t light_count[SDFGI::MAX_STATIC_LIGHTS];
for (uint32_t i = 0; i < p_cascade_count; i++) {
ERR_CONTINUE(p_cascade_indices[i] >= rb->sdfgi->cascades.size());
@ -8005,9 +8078,36 @@ void RendererSceneRenderRD::render_sdfgi_static_lights(RID p_render_buffers, uin
if (idx > 0) {
RD::get_singleton()->buffer_update(cc.lights_buffer, 0, idx * sizeof(SDGIShader::Light), lights, true);
}
dl_push_constant.light_count = idx;
}
light_count[i] = idx;
}
}
/* Static Lights */
RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, sdfgi_shader.direct_light_pipeline[SDGIShader::DIRECT_LIGHT_MODE_STATIC]);
SDGIShader::DirectLightPushConstant dl_push_constant;
dl_push_constant.grid_size[0] = rb->sdfgi->cascade_size;
dl_push_constant.grid_size[1] = rb->sdfgi->cascade_size;
dl_push_constant.grid_size[2] = rb->sdfgi->cascade_size;
dl_push_constant.max_cascades = rb->sdfgi->cascades.size();
dl_push_constant.probe_axis_size = rb->sdfgi->probe_axis_count;
dl_push_constant.multibounce = false; // this is static light, do not multibounce yet
dl_push_constant.y_mult = rb->sdfgi->y_mult;
//all must be processed
dl_push_constant.process_offset = 0;
dl_push_constant.process_increment = 1;
for (uint32_t i = 0; i < p_cascade_count; i++) {
ERR_CONTINUE(p_cascade_indices[i] >= rb->sdfgi->cascades.size());
SDFGI::Cascade &cc = rb->sdfgi->cascades[p_cascade_indices[i]];
dl_push_constant.light_count = light_count[i];
dl_push_constant.cascade = p_cascade_indices[i];
if (dl_push_constant.light_count > 0) {
@ -8299,7 +8399,10 @@ RendererSceneRenderRD::RendererSceneRenderRD(RendererStorageRD *p_storage) {
roughness_layers = GLOBAL_GET("rendering/quality/reflections/roughness_layers");
sky_ggx_samples_quality = GLOBAL_GET("rendering/quality/reflections/ggx_samples");
sky_use_cubemap_array = GLOBAL_GET("rendering/quality/reflections/texture_array_reflections");
// sky_use_cubemap_array = false;
sdfgi_ray_count = RS::EnvironmentSDFGIRayCount(CLAMP(int32_t(GLOBAL_GET("rendering/sdfgi/probe_ray_count")), 0, int32_t(RS::ENV_SDFGI_RAY_COUNT_MAX - 1)));
sdfgi_frames_to_converge = RS::EnvironmentSDFGIFramesToConverge(CLAMP(int32_t(GLOBAL_GET("rendering/sdfgi/frames_to_converge")), 0, int32_t(RS::ENV_SDFGI_CONVERGE_MAX - 1)));
sdfgi_frames_to_update_light = RS::EnvironmentSDFGIFramesToUpdateLight(CLAMP(int32_t(GLOBAL_GET("rendering/sdfgi/frames_to_update_lights")), 0, int32_t(RS::ENV_SDFGI_UPDATE_LIGHT_MAX - 1)));
uint32_t textures_per_stage = RD::get_singleton()->limit_get(RD::LIMIT_MAX_TEXTURES_PER_SHADER_STAGE);
@ -8632,9 +8735,12 @@ RendererSceneRenderRD::RendererSceneRenderRD(RendererStorageRD *p_storage) {
//calculate tables
String defines = "\n#define SDFGI_OCT_SIZE " + itos(SDFGI::LIGHTPROBE_OCT_SIZE) + "\n";
Vector<String> gi_modes;
gi_modes.push_back("\n#define USE_GIPROBE\n");
gi_modes.push_back("\n#define USE_GIPROBES\n");
gi_modes.push_back("\n#define USE_SDFGI\n");
gi_modes.push_back("\n#define USE_SDFGI\n\n#define USE_GIPROBE\n");
gi_modes.push_back("\n#define USE_SDFGI\n\n#define USE_GIPROBES\n");
gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_GIPROBES\n");
gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_SDFGI\n");
gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_SDFGI\n\n#define USE_GIPROBES\n");
gi.shader.initialize(gi_modes, defines);
gi.shader_version = gi.shader.version_create();
@ -8766,6 +8872,8 @@ RendererSceneRenderRD::RendererSceneRenderRD(RendererStorageRD *p_storage) {
environment_set_volumetric_fog_positional_shadow_shrink_size(GLOBAL_GET("rendering/volumetric_fog/positional_shadow_shrink"));
cull_argument.set_page_pool(&cull_argument_pool);
gi.half_resolution = GLOBAL_GET("rendering/quality/gi/use_half_resolution");
}
RendererSceneRenderRD::~RendererSceneRenderRD() {

View file

@ -124,8 +124,6 @@ protected:
virtual void _base_uniforms_changed() = 0;
virtual void _render_buffers_uniform_set_changed(RID p_render_buffers) = 0;
virtual RID _render_buffers_get_normal_texture(RID p_render_buffers) = 0;
virtual RID _render_buffers_get_ambient_texture(RID p_render_buffers) = 0;
virtual RID _render_buffers_get_reflection_texture(RID p_render_buffers) = 0;
void _process_ssao(RID p_render_buffers, RID p_environment, RID p_normal_buffer, const CameraMatrix &p_projection);
void _process_ssr(RID p_render_buffers, RID p_dest_framebuffer, RID p_normal_buffer, RID p_specular_buffer, RID p_metallic, const Color &p_metallic_mask, RID p_environment, const CameraMatrix &p_projection, bool p_use_additive);
@ -134,7 +132,7 @@ protected:
void _setup_sky(RID p_environment, RID p_render_buffers, const CameraMatrix &p_projection, const Transform &p_transform, const Size2i p_screen_size);
void _update_sky(RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform);
void _draw_sky(bool p_can_continue_color, bool p_can_continue_depth, RID p_fb, RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform);
void _process_gi(RID p_render_buffers, RID p_normal_roughness_buffer, RID p_ambient_buffer, RID p_reflection_buffer, RID p_gi_probe_buffer, RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform, const PagedArray<RID> &p_gi_probes);
void _process_gi(RID p_render_buffers, RID p_normal_roughness_buffer, RID p_gi_probe_buffer, RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform, const PagedArray<RID> &p_gi_probes);
// needed for a single argument calls (material and uv2)
PagedArrayPool<GeometryInstance *> cull_argument_pool;
@ -904,6 +902,16 @@ private:
RID giprobe_textures[MAX_GIPROBES];
RID giprobe_buffer;
RID ambient_buffer;
RID reflection_buffer;
bool using_half_size_gi = false;
struct GI {
RID full_buffer;
RID full_dispatch;
RID full_mask;
} gi;
};
RID default_giprobe_buffer;
@ -965,6 +973,8 @@ private:
RID scroll_occlusion_uniform_set;
RID integrate_uniform_set;
RID lights_buffer;
bool all_dynamic_lights_dirty = true;
};
//used for rendering (voxelization)
@ -1026,6 +1036,8 @@ private:
RS::EnvironmentSDFGIRayCount sdfgi_ray_count = RS::ENV_SDFGI_RAY_COUNT_16;
RS::EnvironmentSDFGIFramesToConverge sdfgi_frames_to_converge = RS::ENV_SDFGI_CONVERGE_IN_10_FRAMES;
RS::EnvironmentSDFGIFramesToUpdateLight sdfgi_frames_to_update_light = RS::ENV_SDFGI_UPDATE_LIGHT_IN_4_FRAMES;
float sdfgi_solid_cell_ratio = 0.25;
Vector3 sdfgi_debug_probe_pos;
Vector3 sdfgi_debug_probe_dir;
@ -1263,14 +1275,12 @@ private:
float z_far;
float proj_info[4];
uint32_t max_giprobes;
uint32_t high_quality_vct;
uint32_t pad2;
uint32_t orthogonal;
float ao_color[3];
uint32_t pad;
uint32_t max_giprobes;
uint32_t high_quality_vct;
uint32_t orthogonal;
uint32_t pad[2];
float cam_rotation[12];
};
@ -1280,9 +1290,13 @@ private:
MODE_GIPROBE,
MODE_SDFGI,
MODE_COMBINED,
MODE_HALF_RES_GIPROBE,
MODE_HALF_RES_SDFGI,
MODE_HALF_RES_COMBINED,
MODE_MAX
};
bool half_resolution = false;
GiShaderRD shader;
RID shader_version;
RID pipelines[MODE_MAX];
@ -1657,6 +1671,7 @@ public:
virtual void environment_set_sdfgi(RID p_env, bool p_enable, RS::EnvironmentSDFGICascades p_cascades, float p_min_cell_size, RS::EnvironmentSDFGIYScale p_y_scale, bool p_use_occlusion, bool p_use_multibounce, bool p_read_sky, float p_energy, float p_normal_bias, float p_probe_bias);
virtual void environment_set_sdfgi_ray_count(RS::EnvironmentSDFGIRayCount p_ray_count);
virtual void environment_set_sdfgi_frames_to_converge(RS::EnvironmentSDFGIFramesToConverge p_frames);
virtual void environment_set_sdfgi_frames_to_update_light(RS::EnvironmentSDFGIFramesToUpdateLight p_update);
void environment_set_ssr_roughness_quality(RS::EnvironmentSSRRoughnessQuality p_quality);
RS::EnvironmentSSRRoughnessQuality environment_get_ssr_roughness_quality() const;
@ -1955,11 +1970,14 @@ public:
*/
RID render_buffers_create();
void render_buffers_configure(RID p_render_buffers, RID p_render_target, int p_width, int p_height, RS::ViewportMSAA p_msaa, RS::ViewportScreenSpaceAA p_screen_space_aa, bool p_use_debanding);
void gi_set_use_half_resolution(bool p_enable);
RID render_buffers_get_ao_texture(RID p_render_buffers);
RID render_buffers_get_back_buffer_texture(RID p_render_buffers);
RID render_buffers_get_gi_probe_buffer(RID p_render_buffers);
RID render_buffers_get_default_gi_probe_buffer();
RID render_buffers_get_gi_ambient_texture(RID p_render_buffers);
RID render_buffers_get_gi_reflection_texture(RID p_render_buffers);
uint32_t render_buffers_get_sdfgi_cascade_count(RID p_render_buffers) const;
bool render_buffers_is_sdfgi_enabled(RID p_render_buffers) const;

View file

@ -97,13 +97,12 @@ layout(push_constant, binding = 0, std430) uniform Params {
vec4 proj_info;
uint max_giprobes;
bool high_quality_vct;
uint pad2;
bool orthogonal;
vec3 ao_color;
uint pad;
uint max_giprobes;
bool high_quality_vct;
bool orthogonal;
uint pad[2];
mat3x4 cam_rotation;
}
@ -594,28 +593,16 @@ vec4 fetch_normal_and_roughness(ivec2 pos) {
return normal_roughness;
}
void main() {
// Pixel being shaded
ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
if (any(greaterThanEqual(pos, params.screen_size))) { //too large, do nothing
return;
}
vec3 vertex = reconstruct_position(pos);
vertex.y = -vertex.y;
void process_gi(ivec2 pos, vec3 vertex, inout vec4 ambient_light, inout vec4 reflection_light) {
vec4 normal_roughness = fetch_normal_and_roughness(pos);
vec3 normal = normal_roughness.xyz;
vec4 ambient_light = vec4(0.0), reflection_light = vec4(0.0);
vec3 normal = normal_roughness.xyz;
if (normal.length() > 0.5) {
//valid normal, can do GI
float roughness = normal_roughness.w;
vertex = mat3(params.cam_rotation) * vertex;
normal = normalize(mat3(params.cam_rotation) * normal);
vec3 reflection = normalize(reflect(normalize(vertex), normal));
#ifdef USE_SDFGI
@ -646,16 +633,38 @@ void main() {
spec_accum /= blend_accum;
}
if (params.use_sdfgi) {
reflection_light = blend_color(spec_accum, reflection_light);
ambient_light = blend_color(amb_accum, ambient_light);
} else {
reflection_light = spec_accum;
ambient_light = amb_accum;
}
#ifdef USE_SDFGI
reflection_light = blend_color(spec_accum, reflection_light);
ambient_light = blend_color(amb_accum, ambient_light);
#else
reflection_light = spec_accum;
ambient_light = amb_accum;
#endif
}
#endif
}
}
void main() {
ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
#ifdef MODE_HALF_RES
pos <<= 1;
#endif
if (any(greaterThanEqual(pos, params.screen_size))) { //too large, do nothing
return;
}
vec4 ambient_light = vec4(0.0), reflection_light = vec4(0.0);
vec3 vertex = reconstruct_position(pos);
vertex.y = -vertex.y;
process_gi(pos, vertex, ambient_light, reflection_light);
#ifdef MODE_HALF_RES
pos >>= 1;
#endif
imageStore(ambient_buffer, pos, ambient_light);
imageStore(reflection_buffer, pos, reflection_light);

View file

@ -58,6 +58,117 @@ void main() {
#else
#if 1
vec4 group1;
vec4 group2;
vec4 group3;
vec4 group4;
int best_index = 0;
//2X
group1.x = texelFetch(source_depth, pos, 0).r;
group1.y = texelFetch(source_depth, pos, 1).r;
//4X
if (params.sample_count <= 4) {
group1.z = texelFetch(source_depth, pos, 2).r;
group1.w = texelFetch(source_depth, pos, 3).r;
}
//8X
if (params.sample_count <= 8) {
group2.x = texelFetch(source_depth, pos, 4).r;
group2.y = texelFetch(source_depth, pos, 5).r;
group2.z = texelFetch(source_depth, pos, 6).r;
group2.w = texelFetch(source_depth, pos, 7).r;
}
//16X
if (params.sample_count <= 16) {
group3.x = texelFetch(source_depth, pos, 8).r;
group3.y = texelFetch(source_depth, pos, 9).r;
group3.z = texelFetch(source_depth, pos, 10).r;
group3.w = texelFetch(source_depth, pos, 11).r;
group4.x = texelFetch(source_depth, pos, 12).r;
group4.y = texelFetch(source_depth, pos, 13).r;
group4.z = texelFetch(source_depth, pos, 14).r;
group4.w = texelFetch(source_depth, pos, 15).r;
}
if (params.sample_count == 2) {
best_index = (pos.x & 1) ^ ((pos.y >> 1) & 1); //not much can be done here
} else if (params.sample_count == 4) {
vec4 freq = vec4(equal(group1, vec4(group1.x)));
freq += vec4(equal(group1, vec4(group1.y)));
freq += vec4(equal(group1, vec4(group1.z)));
freq += vec4(equal(group1, vec4(group1.w)));
float min_f = freq.x;
best_index = 0;
if (freq.y < min_f) {
best_index = 1;
min_f = freq.y;
}
if (freq.z < min_f) {
best_index = 2;
min_f = freq.z;
}
if (freq.w < min_f) {
best_index = 3;
min_f = freq.w;
}
} else if (params.sample_count == 8) {
vec4 freq0 = vec4(equal(group1, vec4(group1.x)));
vec4 freq1 = vec4(equal(group2, vec4(group1.x)));
freq0 += vec4(equal(group1, vec4(group1.y)));
freq1 += vec4(equal(group2, vec4(group1.y)));
freq0 += vec4(equal(group1, vec4(group1.z)));
freq1 += vec4(equal(group2, vec4(group1.z)));
freq0 += vec4(equal(group1, vec4(group1.w)));
freq1 += vec4(equal(group2, vec4(group1.w)));
freq0 += vec4(equal(group1, vec4(group2.x)));
freq1 += vec4(equal(group2, vec4(group2.x)));
freq0 += vec4(equal(group1, vec4(group2.y)));
freq1 += vec4(equal(group2, vec4(group2.y)));
freq0 += vec4(equal(group1, vec4(group2.z)));
freq1 += vec4(equal(group2, vec4(group2.z)));
freq0 += vec4(equal(group1, vec4(group2.w)));
freq1 += vec4(equal(group2, vec4(group2.w)));
float min_f0 = freq0.x;
int best_index0 = 0;
if (freq0.y < min_f0) {
best_index0 = 1;
min_f0 = freq0.y;
}
if (freq0.z < min_f0) {
best_index0 = 2;
min_f0 = freq0.z;
}
if (freq0.w < min_f0) {
best_index0 = 3;
min_f0 = freq0.w;
}
float min_f1 = freq1.x;
int best_index1 = 4;
if (freq1.y < min_f1) {
best_index1 = 5;
min_f1 = freq1.y;
}
if (freq1.z < min_f1) {
best_index1 = 6;
min_f1 = freq1.z;
}
if (freq1.w < min_f1) {
best_index1 = 7;
min_f1 = freq1.w;
}
best_index = mix(best_index0, best_index1, min_f0 < min_f1);
}
#else
float depths[16];
int depth_indices[16];
int depth_amount[16];
@ -91,7 +202,7 @@ void main() {
depth_least = depth_amount[j];
}
}
#endif
best_depth = texelFetch(source_depth, pos, best_index).r;
best_normal_roughness = texelFetch(source_normal_roughness, pos, best_index);
#ifdef GIPROBE_RESOLVE

View file

@ -2371,17 +2371,17 @@ FRAGMENT_SHADER_CODE
if (bool(draw_call.flags & INSTANCE_FLAGS_USE_GI_BUFFERS)) { //use GI buffers
ivec2 coord;
vec2 coord;
if (scene_data.gi_upscale_for_msaa) {
ivec2 base_coord = ivec2(gl_FragCoord.xy);
ivec2 closest_coord = base_coord;
float closest_ang = dot(normal, texelFetch(sampler2D(normal_roughness_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), base_coord, 0).xyz * 2.0 - 1.0);
vec2 base_coord = screen_uv;
vec2 closest_coord = base_coord;
float closest_ang = dot(normal, textureLod(sampler2D(normal_roughness_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), base_coord, 0.0).xyz * 2.0 - 1.0);
for (int i = 0; i < 4; i++) {
const ivec2 neighbours[4] = ivec2[](ivec2(-1, 0), ivec2(1, 0), ivec2(0, -1), ivec2(0, 1));
ivec2 neighbour_coord = base_coord + neighbours[i];
float neighbour_ang = dot(normal, texelFetch(sampler2D(normal_roughness_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), neighbour_coord, 0).xyz * 2.0 - 1.0);
const vec2 neighbours[4] = vec2[](vec2(-1, 0), vec2(1, 0), vec2(0, -1), vec2(0, 1));
vec2 neighbour_coord = base_coord + neighbours[i] * scene_data.screen_pixel_size;
float neighbour_ang = dot(normal, textureLod(sampler2D(normal_roughness_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), neighbour_coord, 0.0).xyz * 2.0 - 1.0);
if (neighbour_ang > closest_ang) {
closest_ang = neighbour_ang;
closest_coord = neighbour_coord;
@ -2391,11 +2391,11 @@ FRAGMENT_SHADER_CODE
coord = closest_coord;
} else {
coord = ivec2(gl_FragCoord.xy);
coord = screen_uv;
}
vec4 buffer_ambient = texelFetch(sampler2D(ambient_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), coord, 0);
vec4 buffer_reflection = texelFetch(sampler2D(reflection_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), coord, 0);
vec4 buffer_ambient = textureLod(sampler2D(ambient_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), coord, 0.0);
vec4 buffer_reflection = textureLod(sampler2D(reflection_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), coord, 0.0);
ambient_light = mix(ambient_light, buffer_ambient.rgb, buffer_ambient.a);
specular_light = mix(specular_light, buffer_reflection.rgb, buffer_reflection.a);

View file

@ -97,6 +97,8 @@ void main() {
float blend = 0.0;
#if 1
// No interpolation
vec3 inv_dir = 1.0 / ray_dir;
float rough = 0.5;
@ -161,114 +163,11 @@ void main() {
hit_light *= (dot(max(vec3(0.0), (hit_normal * hit_aniso0)), vec3(1.0)) + dot(max(vec3(0.0), (-hit_normal * hit_aniso1)), vec3(1.0)));
if (blend > 0.0) {
light = mix(light, hit_light, blend);
blend = 0.0;
} else {
light = hit_light;
//process blend
float blend_from = (float(params.probe_axis_size - 1) / 2.0) - 2.5;
float blend_to = blend_from + 2.0;
vec3 cam_pos = params.cam_transform[3].xyz - cascades.data[i].offset;
cam_pos *= cascades.data[i].to_cell;
pos += ray_dir * min(advance, max_advance);
vec3 inner_pos = pos - cam_pos;
inner_pos = inner_pos * float(params.probe_axis_size - 1) / params.grid_size.x;
float len = length(inner_pos);
inner_pos = abs(normalize(inner_pos));
len *= max(inner_pos.x, max(inner_pos.y, inner_pos.z));
if (len >= blend_from) {
blend = smoothstep(blend_from, blend_to, len);
pos /= cascades.data[i].to_cell;
pos += cascades.data[i].offset;
ray_pos = pos;
hit = false; //continue trace for blend
continue;
}
}
light = hit_light;
break;
}
light = mix(light, vec3(0.0), blend);
#else
vec3 inv_dir = 1.0 / ray_dir;
bool hit = false;
vec4 light_accum = vec4(0.0);
float blend_size = (params.grid_size.x / float(params.probe_axis_size - 1)) * 0.5;
float radius_sizes[MAX_CASCADES];
for (uint i = 0; i < params.max_cascades; i++) {
radius_sizes[i] = (1.0 / cascades.data[i].to_cell) * (params.grid_size.x * 0.5 - blend_size);
}
float max_distance = radius_sizes[params.max_cascades - 1];
float advance = 0;
while (advance < max_distance) {
for (uint i = 0; i < params.max_cascades; i++) {
if (advance < radius_sizes[i]) {
vec3 pos = (ray_pos + ray_dir * advance) - cascades.data[i].offset;
pos *= cascades.data[i].to_cell * pos_to_uvw;
float distance = texture(sampler3D(sdf_cascades[i], linear_sampler), pos).r * 255.0 - 1.0;
vec4 hit_light = vec4(0.0);
if (distance < 1.0) {
hit_light.a = max(0.0, 1.0 - distance);
hit_light.rgb = texture(sampler3D(light_cascades[i], linear_sampler), pos).rgb;
hit_light.rgb *= hit_light.a;
}
distance /= cascades.data[i].to_cell;
if (i < (params.max_cascades - 1)) {
pos = (ray_pos + ray_dir * advance) - cascades.data[i + 1].offset;
pos *= cascades.data[i + 1].to_cell * pos_to_uvw;
float distance2 = texture(sampler3D(sdf_cascades[i + 1], linear_sampler), pos).r * 255.0 - 1.0;
vec4 hit_light2 = vec4(0.0);
if (distance2 < 1.0) {
hit_light2.a = max(0.0, 1.0 - distance2);
hit_light2.rgb = texture(sampler3D(light_cascades[i + 1], linear_sampler), pos).rgb;
hit_light2.rgb *= hit_light2.a;
}
float prev_radius = i == 0 ? 0.0 : radius_sizes[i - 1];
float blend = (advance - prev_radius) / (radius_sizes[i] - prev_radius);
distance2 /= cascades.data[i + 1].to_cell;
hit_light = mix(hit_light, hit_light2, blend);
distance = mix(distance, distance2, blend);
}
light_accum += hit_light;
advance += distance;
break;
}
}
if (light_accum.a > 0.98) {
break;
}
}
light = light_accum.rgb / light_accum.a;
#endif
imageStore(screen_buffer, screen_pos, vec4(linear_to_srgb(light), 1.0));

View file

@ -125,7 +125,10 @@ void main() {
uint voxel_index = uint(gl_GlobalInvocationID.x);
//used for skipping voxels every N frames
voxel_index = params.process_offset + voxel_index * params.process_increment;
if (params.process_increment > 1) {
voxel_index *= params.process_increment;
voxel_index += params.process_offset;
}
if (voxel_index >= dispatch_data.total_count) {
return;

View file

@ -189,14 +189,12 @@ void main() {
vec3 inv_dir = 1.0 / ray_dir;
bool hit = false;
vec3 hit_normal;
vec3 hit_light;
vec3 hit_aniso0;
vec3 hit_aniso1;
uint hit_cascade;
float bias = params.ray_bias;
vec3 abs_ray_dir = abs(ray_dir);
ray_pos += ray_dir * 1.0 / max(abs_ray_dir.x, max(abs_ray_dir.y, abs_ray_dir.z)) * bias / cascades.data[params.cascade].to_cell;
vec3 uvw;
for (uint j = params.cascade; j < params.max_cascades; j++) {
//convert to local bounds
@ -215,14 +213,12 @@ void main() {
float advance = 0.0;
vec3 uvw;
while (advance < max_advance) {
//read how much to advance from SDF
uvw = (pos + ray_dir * advance) * pos_to_uvw;
float distance = texture(sampler3D(sdf_cascades[j], linear_sampler), uvw).r * 255.0 - 1.0;
if (distance < 0.001) {
if (distance < 0.05) {
//consider hit
hit = true;
break;
@ -232,17 +228,7 @@ void main() {
}
if (hit) {
const float EPSILON = 0.001;
hit_normal = normalize(vec3(
texture(sampler3D(sdf_cascades[j], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[j], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r,
texture(sampler3D(sdf_cascades[j], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[j], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r,
texture(sampler3D(sdf_cascades[j], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[j], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r));
hit_light = texture(sampler3D(light_cascades[j], linear_sampler), uvw).rgb;
vec4 aniso0 = texture(sampler3D(aniso0_cascades[j], linear_sampler), uvw);
hit_aniso0 = aniso0.rgb;
hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[j], linear_sampler), uvw).rg);
hit_cascade = j;
break;
}
@ -255,6 +241,17 @@ void main() {
vec4 light;
if (hit) {
const float EPSILON = 0.001;
vec3 hit_normal = normalize(vec3(
texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r,
texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r,
texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r));
vec3 hit_light = texture(sampler3D(light_cascades[hit_cascade], linear_sampler), uvw).rgb;
vec4 aniso0 = texture(sampler3D(aniso0_cascades[hit_cascade], linear_sampler), uvw);
vec3 hit_aniso0 = aniso0.rgb;
vec3 hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[hit_cascade], linear_sampler), uvw).rg);
//one liner magic
light.rgb = hit_light * (dot(max(vec3(0.0), (hit_normal * hit_aniso0)), vec3(1.0)) + dot(max(vec3(0.0), (-hit_normal * hit_aniso1)), vec3(1.0)));
light.a = 1.0;
@ -490,13 +487,15 @@ void main() {
//can't scroll, must look for position in parent cascade
//to global coords
float probe_cell_size = float(params.grid_size.x / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell;
float cell_to_probe = float(params.grid_size.x / float(params.probe_axis_size - 1));
float probe_cell_size = cell_to_probe / cascades.data[params.cascade].to_cell;
vec3 probe_pos = cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size;
//to parent local coords
float probe_cell_size_next = cell_to_probe / cascades.data[params.cascade + 1].to_cell;
probe_pos -= cascades.data[params.cascade + 1].offset;
probe_pos *= cascades.data[params.cascade + 1].to_cell;
probe_pos = probe_pos * float(params.probe_axis_size - 1) / float(params.grid_size.x);
probe_pos /= probe_cell_size_next;
ivec3 probe_posi = ivec3(probe_pos);
//add up all light, no need to use occlusion here, since occlusion will do its work afterwards
@ -549,20 +548,28 @@ void main() {
}
} else {
// clear and let it re-raytrace, only for the last cascade, which happens very un-often
//scroll
//scroll at the edge of the highest cascade, just copy what is there,
//since its the closest we have anyway
for (uint j = 0; j < params.history_size; j++) {
ivec2 tex_pos;
tex_pos = probe_cell.xy;
tex_pos.x += probe_cell.z * int(params.probe_axis_size);
for (int i = 0; i < SH_SIZE; i++) {
// copy from history texture
ivec3 src_pos = ivec3(tex_pos.x, tex_pos.y * SH_SIZE + i, int(j));
ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j));
imageStore(lightprobe_history_scroll_texture, dst_pos, ivec4(0));
ivec4 value = imageLoad(lightprobe_history_texture, dst_pos);
imageStore(lightprobe_history_scroll_texture, dst_pos, value);
}
}
for (int i = 0; i < SH_SIZE; i++) {
// copy from average texture
ivec2 dst_pos = ivec2(pos.x, pos.y * SH_SIZE + i);
imageStore(lightprobe_average_scroll_texture, dst_pos, ivec4(0));
ivec2 spos = ivec2(pos.x, pos.y * SH_SIZE + i);
ivec4 average = imageLoad(lightprobe_average_texture, spos);
imageStore(lightprobe_average_scroll_texture, spos, average);
}
}

View file

@ -140,6 +140,7 @@ public:
virtual void environment_set_sdfgi_ray_count(RS::EnvironmentSDFGIRayCount p_ray_count) = 0;
virtual void environment_set_sdfgi_frames_to_converge(RS::EnvironmentSDFGIFramesToConverge p_frames) = 0;
virtual void environment_set_sdfgi_frames_to_update_light(RS::EnvironmentSDFGIFramesToUpdateLight p_update) = 0;
virtual void environment_set_tonemap(RID p_env, RS::EnvironmentToneMapper p_tone_mapper, float p_exposure, float p_white, bool p_auto_exposure, float p_min_luminance, float p_max_luminance, float p_auto_exp_speed, float p_auto_exp_scale) = 0;
@ -180,6 +181,8 @@ public:
virtual RID render_buffers_create() = 0;
virtual void render_buffers_configure(RID p_render_buffers, RID p_render_target, int p_width, int p_height, RS::ViewportMSAA p_msaa, RS::ViewportScreenSpaceAA p_screen_space_aa, bool p_use_debanding) = 0;
virtual void gi_set_use_half_resolution(bool p_enable) = 0;
virtual void set_debug_draw_mode(RS::ViewportDebugDraw p_debug_draw) = 0;
virtual TypedArray<Image> bake_render_uv2(RID p_base, const Vector<RID> &p_material_overrides, const Size2i &p_image_size) = 0;

View file

@ -795,6 +795,7 @@ public:
lightmaps.set_page_pool(p_rid_pool);
reflections.set_page_pool(p_rid_pool);
decals.set_page_pool(p_rid_pool);
gi_probes.set_page_pool(p_rid_pool);
mesh_instances.set_page_pool(p_rid_pool);
for (int i = 0; i < RendererSceneRender::MAX_DIRECTIONAL_LIGHTS; i++) {
for (int j = 0; j < RendererSceneRender::MAX_DIRECTIONAL_LIGHT_CASCADES; j++) {
@ -995,6 +996,7 @@ public:
PASS11(environment_set_sdfgi, RID, bool, RS::EnvironmentSDFGICascades, float, RS::EnvironmentSDFGIYScale, bool, bool, bool, float, float, float)
PASS1(environment_set_sdfgi_ray_count, RS::EnvironmentSDFGIRayCount)
PASS1(environment_set_sdfgi_frames_to_converge, RS::EnvironmentSDFGIFramesToConverge)
PASS1(environment_set_sdfgi_frames_to_update_light, RS::EnvironmentSDFGIFramesToUpdateLight)
PASS1RC(RS::EnvironmentBG, environment_get_background, RID)
PASS1RC(int, environment_get_canvas_max_layer, RID)
@ -1024,6 +1026,7 @@ public:
PASS0R(RID, render_buffers_create)
PASS7(render_buffers_configure, RID, RID, int, int, RS::ViewportMSAA, RS::ViewportScreenSpaceAA, bool)
PASS1(gi_set_use_half_resolution, bool)
/* Shadow Atlas */
PASS0R(RID, shadow_atlas_create)

View file

@ -136,6 +136,7 @@ public:
virtual void environment_set_sdfgi_ray_count(RS::EnvironmentSDFGIRayCount p_ray_count) = 0;
virtual void environment_set_sdfgi_frames_to_converge(RS::EnvironmentSDFGIFramesToConverge p_frames) = 0;
virtual void environment_set_sdfgi_frames_to_update_light(RS::EnvironmentSDFGIFramesToUpdateLight p_update) = 0;
virtual void environment_set_tonemap(RID p_env, RS::EnvironmentToneMapper p_tone_mapper, float p_exposure, float p_white, bool p_auto_exposure, float p_min_luminance, float p_max_luminance, float p_auto_exp_speed, float p_auto_exp_scale) = 0;
@ -208,6 +209,7 @@ public:
virtual RID render_buffers_create() = 0;
virtual void render_buffers_configure(RID p_render_buffers, RID p_render_target, int p_width, int p_height, RS::ViewportMSAA p_msaa, RS::ViewportScreenSpaceAA p_screen_space_aa, bool p_use_debanding) = 0;
virtual void gi_set_use_half_resolution(bool p_enable) = 0;
virtual void screen_space_roughness_limiter_set_active(bool p_enable, float p_amount, float p_limit) = 0;
virtual bool screen_space_roughness_limiter_is_active() const = 0;

View file

@ -162,6 +162,51 @@ void RenderingServerDefault::draw(bool p_swap_buffers, double frame_step) {
}
frame_profile_frame = RSG::storage->get_captured_timestamps_frame();
if (print_gpu_profile) {
if (print_frame_profile_ticks_from == 0) {
print_frame_profile_ticks_from = OS::get_singleton()->get_ticks_usec();
}
float total_time = 0.0;
for (int i = 0; i < frame_profile.size() - 1; i++) {
String name = frame_profile[i].name;
if (name[0] == '<' || name[0] == '>') {
continue;
}
float time = frame_profile[i + 1].gpu_msec - frame_profile[i].gpu_msec;
if (name[0] != '<' && name[0] != '>') {
if (print_gpu_profile_task_time.has(name)) {
print_gpu_profile_task_time[name] += time;
} else {
print_gpu_profile_task_time[name] = time;
}
}
}
if (frame_profile.size()) {
total_time = frame_profile[frame_profile.size() - 1].gpu_msec;
}
uint64_t ticks_elapsed = OS::get_singleton()->get_ticks_usec() - print_frame_profile_ticks_from;
print_frame_profile_frame_count++;
if (ticks_elapsed > 1000000) {
print_line("GPU PROFILE (total " + rtos(total_time) + "ms): ");
float print_threshold = 0.01;
for (OrderedHashMap<String, float>::Element E = print_gpu_profile_task_time.front(); E; E = E.next()) {
float time = E.value() / float(print_frame_profile_frame_count);
if (time > print_threshold) {
print_line("\t-" + E.key() + ": " + rtos(time) + "ms");
}
}
print_gpu_profile_task_time.clear();
print_frame_profile_ticks_from = OS::get_singleton()->get_ticks_usec();
print_frame_profile_frame_count = 0;
}
}
}
float RenderingServerDefault::get_frame_setup_time_cpu() const {
@ -232,6 +277,11 @@ void RenderingServerDefault::sdfgi_set_debug_probe_select(const Vector3 &p_posit
RSG::scene->sdfgi_set_debug_probe_select(p_position, p_dir);
}
void RenderingServerDefault::set_print_gpu_profile(bool p_enable) {
RSG::storage->capturing_timestamps = p_enable;
print_gpu_profile = p_enable;
}
RID RenderingServerDefault::get_test_cube() {
if (!test_cube.is_valid()) {
test_cube = _make_test_cube();

View file

@ -32,6 +32,7 @@
#define RENDERING_SERVER_DEFAULT_H
#include "core/math/octree.h"
#include "core/templates/ordered_hash_map.h"
#include "renderer_canvas_cull.h"
#include "renderer_scene_cull.h"
#include "renderer_viewport.h"
@ -74,6 +75,12 @@ class RenderingServerDefault : public RenderingServer {
float frame_setup_time = 0;
//for printing
bool print_gpu_profile = false;
OrderedHashMap<String, float> print_gpu_profile_task_time;
uint64_t print_frame_profile_ticks_from = 0;
uint32_t print_frame_profile_frame_count = 0;
public:
//if editor is redrawing when it shouldn't, enable this and put a breakpoint in _changes_changed()
//#define DEBUG_CHANGES
@ -616,6 +623,7 @@ public:
BIND11(environment_set_sdfgi, RID, bool, EnvironmentSDFGICascades, float, EnvironmentSDFGIYScale, bool, bool, bool, float, float, float)
BIND1(environment_set_sdfgi_ray_count, EnvironmentSDFGIRayCount)
BIND1(environment_set_sdfgi_frames_to_converge, EnvironmentSDFGIFramesToConverge)
BIND1(environment_set_sdfgi_frames_to_update_light, EnvironmentSDFGIFramesToUpdateLight)
BIND3R(Ref<Image>, environment_bake_panorama, RID, bool, const Size2i &)
@ -688,6 +696,8 @@ public:
BIND3R(TypedArray<Image>, bake_render_uv2, RID, const Vector<RID> &, const Size2i &)
BIND1(gi_set_use_half_resolution, bool)
#undef BINDBASE
//from now on, calls forwarded to this singleton
#define BINDBASE RSG::canvas
@ -865,6 +875,8 @@ public:
virtual void sdfgi_set_debug_probe_select(const Vector3 &p_position, const Vector3 &p_dir);
virtual void set_print_gpu_profile(bool p_enable);
RenderingServerDefault();
~RenderingServerDefault();

View file

@ -507,6 +507,7 @@ public:
FUNC11(environment_set_sdfgi, RID, bool, EnvironmentSDFGICascades, float, EnvironmentSDFGIYScale, bool, bool, bool, float, float, float)
FUNC1(environment_set_sdfgi_ray_count, EnvironmentSDFGIRayCount)
FUNC1(environment_set_sdfgi_frames_to_converge, EnvironmentSDFGIFramesToConverge)
FUNC1(environment_set_sdfgi_frames_to_update_light, EnvironmentSDFGIFramesToUpdateLight)
FUNC11(environment_set_glow, RID, bool, Vector<float>, float, float, float, float, EnvironmentGlowBlendMode, float, float, float)
FUNC1(environment_glow_set_use_bicubic_upscale, bool)
@ -744,6 +745,8 @@ public:
return rendering_server->get_video_adapter_vendor();
}
FUNC1(gi_set_use_half_resolution, bool)
FUNC4(set_boot_image, const Ref<Image> &, const Color &, bool, bool)
FUNC1(set_default_clear_color, const Color &)
@ -786,6 +789,10 @@ public:
rendering_server->sdfgi_set_debug_probe_select(p_position, p_dir);
}
virtual void set_print_gpu_profile(bool p_enable) {
rendering_server->set_print_gpu_profile(p_enable);
}
RenderingServerWrapMT(RenderingServer *p_contained, bool p_create_thread);
~RenderingServerWrapMT();

View file

@ -2304,6 +2304,8 @@ RenderingServer::RenderingServer() {
GLOBAL_DEF("rendering/quality/reflection_atlas/reflection_size.mobile", 128);
GLOBAL_DEF("rendering/quality/reflection_atlas/reflection_count", 64);
GLOBAL_DEF("rendering/quality/gi/use_half_resolution", false);
GLOBAL_DEF("rendering/quality/gi_probes/anisotropic", false);
GLOBAL_DEF("rendering/quality/gi_probes/quality", 1);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/gi_probes/quality", PropertyInfo(Variant::INT, "rendering/quality/gi_probes/quality", PROPERTY_HINT_ENUM, "Low (4 Cones - Fast),High (6 Cones - Slow)"));
@ -2367,10 +2369,12 @@ RenderingServer::RenderingServer() {
GLOBAL_DEF("rendering/lightmapper/probe_capture_update_speed", 15);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/lightmapper/probe_capture_update_speed", PropertyInfo(Variant::FLOAT, "rendering/lightmapper/probe_capture_update_speed", PROPERTY_HINT_RANGE, "0.001,256,0.001"));
GLOBAL_DEF("rendering/sdfgi/probe_ray_count", 2);
GLOBAL_DEF("rendering/sdfgi/probe_ray_count", 1);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/sdfgi/probe_ray_count", PropertyInfo(Variant::INT, "rendering/sdfgi/probe_ray_count", PROPERTY_HINT_ENUM, "8 (Fastest),16,32,64,96,128 (Slowest)"));
GLOBAL_DEF("rendering/sdfgi/frames_to_converge", 1);
GLOBAL_DEF("rendering/sdfgi/frames_to_converge", 4);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/sdfgi/frames_to_converge", PropertyInfo(Variant::INT, "rendering/sdfgi/frames_to_converge", PROPERTY_HINT_ENUM, "5 (Less Latency but Lower Quality),10,15,20,25,30 (More Latency but Higher Quality)"));
GLOBAL_DEF("rendering/sdfgi/frames_to_update_lights", 2);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/sdfgi/frames_to_update_lights", PropertyInfo(Variant::INT, "rendering/sdfgi/frames_to_update_lights", PROPERTY_HINT_ENUM, "1 (Slower),2,4,8,16 (Faster)"));
GLOBAL_DEF("rendering/volumetric_fog/volume_size", 64);
ProjectSettings::get_singleton()->set_custom_property_info("rendering/volumetric_fog/volume_size", PropertyInfo(Variant::INT, "rendering/volumetric_fog/volume_size", PROPERTY_HINT_RANGE, "16,512,1"));

View file

@ -986,6 +986,7 @@ public:
virtual void environment_set_sdfgi(RID p_env, bool p_enable, EnvironmentSDFGICascades p_cascades, float p_min_cell_size, EnvironmentSDFGIYScale p_y_scale, bool p_use_occlusion, bool p_use_multibounce, bool p_read_sky, float p_energy, float p_normal_bias, float p_probe_bias) = 0;
enum EnvironmentSDFGIRayCount {
ENV_SDFGI_RAY_COUNT_4,
ENV_SDFGI_RAY_COUNT_8,
ENV_SDFGI_RAY_COUNT_16,
ENV_SDFGI_RAY_COUNT_32,
@ -1009,6 +1010,17 @@ public:
virtual void environment_set_sdfgi_frames_to_converge(EnvironmentSDFGIFramesToConverge p_frames) = 0;
enum EnvironmentSDFGIFramesToUpdateLight {
ENV_SDFGI_UPDATE_LIGHT_IN_1_FRAME,
ENV_SDFGI_UPDATE_LIGHT_IN_2_FRAMES,
ENV_SDFGI_UPDATE_LIGHT_IN_4_FRAMES,
ENV_SDFGI_UPDATE_LIGHT_IN_8_FRAMES,
ENV_SDFGI_UPDATE_LIGHT_IN_16_FRAMES,
ENV_SDFGI_UPDATE_LIGHT_MAX,
};
virtual void environment_set_sdfgi_frames_to_update_light(EnvironmentSDFGIFramesToUpdateLight p_update) = 0;
virtual void environment_set_fog(RID p_env, bool p_enable, const Color &p_light_color, float p_light_energy, float p_sun_scatter, float p_density, float p_height, float p_height_density, float p_aerial_perspective) = 0;
enum EnvVolumetricFogShadowFilter {
@ -1430,6 +1442,8 @@ public:
virtual float get_frame_setup_time_cpu() const = 0;
virtual void gi_set_use_half_resolution(bool p_enable) = 0;
/* TESTING */
virtual RID get_test_cube() = 0;
@ -1462,6 +1476,8 @@ public:
virtual bool is_low_end() const = 0;
virtual void set_print_gpu_profile(bool p_enable) = 0;
RenderingDevice *create_local_rendering_device() const;
bool is_render_loop_enabled() const;