diff options
Diffstat (limited to 'servers/rendering/renderer_rd')
12 files changed, 843 insertions, 38 deletions
diff --git a/servers/rendering/renderer_rd/effects_rd.cpp b/servers/rendering/renderer_rd/effects_rd.cpp index 5cf8895c8e..5dab066262 100644 --- a/servers/rendering/renderer_rd/effects_rd.cpp +++ b/servers/rendering/renderer_rd/effects_rd.cpp @@ -383,6 +383,8 @@ void EffectsRD::set_color(RID p_dest_texture, const Color &p_color, const Rect2i } void EffectsRD::gaussian_blur(RID p_source_rd_texture, RID p_texture, RID p_back_texture, const Rect2i &p_region, bool p_8bit_dst) { + ERR_FAIL_COND_MSG(!prefer_raster_effects, "Can't use the compute version of the gaussian blur with the mobile renderer."); + memset(©.push_constant, 0, sizeof(CopyPushConstant)); uint32_t base_flags = 0; @@ -416,6 +418,8 @@ void EffectsRD::gaussian_blur(RID p_source_rd_texture, RID p_texture, RID p_back } void EffectsRD::gaussian_glow(RID p_source_rd_texture, RID p_back_texture, const Size2i &p_size, float p_strength, bool p_high_quality, bool p_first_pass, float p_luminance_cap, float p_exposure, float p_bloom, float p_hdr_bleed_treshold, float p_hdr_bleed_scale, RID p_auto_exposure, float p_auto_exposure_grey) { + ERR_FAIL_COND_MSG(prefer_raster_effects, "Can't use the compute version of the gaussian glow with the mobile renderer."); + memset(©.push_constant, 0, sizeof(CopyPushConstant)); CopyMode copy_mode = p_first_pass && p_auto_exposure.is_valid() ? COPY_MODE_GAUSSIAN_GLOW_AUTO_EXPOSURE : COPY_MODE_GAUSSIAN_GLOW; @@ -449,6 +453,57 @@ void EffectsRD::gaussian_glow(RID p_source_rd_texture, RID p_back_texture, const RD::get_singleton()->compute_list_end(); } +void EffectsRD::gaussian_glow_raster(RID p_source_rd_texture, RID p_framebuffer_half, RID p_rd_texture_half, RID p_dest_framebuffer, const Vector2 &p_pixel_size, float p_strength, bool p_high_quality, bool p_first_pass, float p_luminance_cap, float p_exposure, float p_bloom, float p_hdr_bleed_treshold, float p_hdr_bleed_scale, RID p_auto_exposure, float p_auto_exposure_grey) { + ERR_FAIL_COND_MSG(!prefer_raster_effects, "Can't use the fragment version of the gaussian glow with the clustered renderer."); + + memset(&blur_raster.push_constant, 0, sizeof(BlurRasterPushConstant)); + + BlurRasterMode blur_mode = p_first_pass && p_auto_exposure.is_valid() ? BLUR_MODE_GAUSSIAN_GLOW_AUTO_EXPOSURE : BLUR_MODE_GAUSSIAN_GLOW; + uint32_t base_flags = 0; + + blur_raster.push_constant.pixel_size[0] = p_pixel_size.x; + blur_raster.push_constant.pixel_size[1] = p_pixel_size.y; + + blur_raster.push_constant.glow_strength = p_strength; + blur_raster.push_constant.glow_bloom = p_bloom; + blur_raster.push_constant.glow_hdr_threshold = p_hdr_bleed_treshold; + blur_raster.push_constant.glow_hdr_scale = p_hdr_bleed_scale; + blur_raster.push_constant.glow_exposure = p_exposure; + blur_raster.push_constant.glow_white = 0; //actually unused + blur_raster.push_constant.glow_luminance_cap = p_luminance_cap; + + blur_raster.push_constant.glow_auto_exposure_grey = p_auto_exposure_grey; //unused also + + //HORIZONTAL + RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_framebuffer_half, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD); + RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, blur_raster.pipelines[blur_mode].get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_framebuffer_half))); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_source_rd_texture), 0); + if (p_auto_exposure.is_valid() && p_first_pass) { + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_auto_exposure), 1); + } + RD::get_singleton()->draw_list_bind_index_array(draw_list, index_array); + + blur_raster.push_constant.flags = base_flags | BLUR_FLAG_HORIZONTAL | (p_first_pass ? BLUR_FLAG_GLOW_FIRST_PASS : 0); + RD::get_singleton()->draw_list_set_push_constant(draw_list, &blur_raster.push_constant, sizeof(BlurRasterPushConstant)); + + RD::get_singleton()->draw_list_draw(draw_list, true); + RD::get_singleton()->draw_list_end(); + + blur_mode = BLUR_MODE_GAUSSIAN_GLOW; + + //VERTICAL + draw_list = RD::get_singleton()->draw_list_begin(p_dest_framebuffer, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD); + RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, blur_raster.pipelines[blur_mode].get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_dest_framebuffer))); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_rd_texture_half), 0); + RD::get_singleton()->draw_list_bind_index_array(draw_list, index_array); + + blur_raster.push_constant.flags = base_flags; + RD::get_singleton()->draw_list_set_push_constant(draw_list, &blur_raster.push_constant, sizeof(BlurRasterPushConstant)); + + RD::get_singleton()->draw_list_draw(draw_list, true); + RD::get_singleton()->draw_list_end(); +} + void EffectsRD::screen_space_reflection(RID p_diffuse, RID p_normal_roughness, RenderingServer::EnvironmentSSRRoughnessQuality p_roughness_quality, RID p_blur_radius, RID p_blur_radius2, RID p_metallic, const Color &p_metallic_mask, RID p_depth, RID p_scale_depth, RID p_scale_normal, RID p_output, RID p_output_blur, const Size2i &p_screen_size, int p_max_steps, float p_fade_in, float p_fade_out, float p_tolerance, const CameraMatrix &p_camera) { RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin(); @@ -736,6 +791,8 @@ void EffectsRD::tonemapper(RID p_source_color, RID p_dst_framebuffer, const Tone } void EffectsRD::luminance_reduction(RID p_source_texture, const Size2i p_source_size, const Vector<RID> p_reduce, RID p_prev_luminance, float p_min_luminance, float p_max_luminance, float p_adjust, bool p_set) { + ERR_FAIL_COND_MSG(prefer_raster_effects, "Can't use compute version of luminance reduction with the mobile renderer."); + luminance_reduce.push_constant.source_size[0] = p_source_size.x; luminance_reduce.push_constant.source_size[1] = p_source_size.y; luminance_reduce.push_constant.max_luminance = p_max_luminance; @@ -774,7 +831,41 @@ void EffectsRD::luminance_reduction(RID p_source_texture, const Size2i p_source_ RD::get_singleton()->compute_list_end(); } +void EffectsRD::luminance_reduction_raster(RID p_source_texture, const Size2i p_source_size, const Vector<RID> p_reduce, Vector<RID> p_fb, RID p_prev_luminance, float p_min_luminance, float p_max_luminance, float p_adjust, bool p_set) { + ERR_FAIL_COND_MSG(!prefer_raster_effects, "Can't use fragment version of luminance reduction with the clustered renderer."); + ERR_FAIL_COND_MSG(p_reduce.size() != p_fb.size(), "Incorrect frame buffer account for luminance reduction."); + + luminance_reduce_raster.push_constant.max_luminance = p_max_luminance; + luminance_reduce_raster.push_constant.min_luminance = p_min_luminance; + luminance_reduce_raster.push_constant.exposure_adjust = p_adjust; + + for (int i = 0; i < p_reduce.size(); i++) { + luminance_reduce_raster.push_constant.source_size[0] = i == 0 ? p_source_size.x : luminance_reduce_raster.push_constant.dest_size[0]; + luminance_reduce_raster.push_constant.source_size[1] = i == 0 ? p_source_size.y : luminance_reduce_raster.push_constant.dest_size[1]; + luminance_reduce_raster.push_constant.dest_size[0] = MAX(luminance_reduce_raster.push_constant.source_size[0] / 8, 1); + luminance_reduce_raster.push_constant.dest_size[1] = MAX(luminance_reduce_raster.push_constant.source_size[1] / 8, 1); + + bool final = !p_set && (luminance_reduce_raster.push_constant.dest_size[0] == 1) && (luminance_reduce_raster.push_constant.dest_size[1] == 1); + LuminanceReduceRasterMode mode = final ? LUMINANCE_REDUCE_FRAGMENT_FINAL : (i == 0 ? LUMINANCE_REDUCE_FRAGMENT_FIRST : LUMINANCE_REDUCE_FRAGMENT); + + RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_fb[i], RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD); + RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, luminance_reduce_raster.pipelines[mode].get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_fb[i]))); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(i == 0 ? p_source_texture : p_reduce[i - 1]), 0); + if (final) { + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_prev_luminance), 1); + } + RD::get_singleton()->draw_list_bind_index_array(draw_list, index_array); + + RD::get_singleton()->draw_list_set_push_constant(draw_list, &luminance_reduce_raster.push_constant, sizeof(LuminanceReduceRasterPushConstant)); + + RD::get_singleton()->draw_list_draw(draw_list, true); + RD::get_singleton()->draw_list_end(); + } +} + void EffectsRD::bokeh_dof(RID p_base_texture, RID p_depth_texture, const Size2i &p_base_texture_size, RID p_secondary_texture, RID p_halfsize_texture1, RID p_halfsize_texture2, bool p_dof_far, float p_dof_far_begin, float p_dof_far_size, bool p_dof_near, float p_dof_near_begin, float p_dof_near_size, float p_bokeh_size, RenderingServer::DOFBokehShape p_bokeh_shape, RS::DOFBlurQuality p_quality, bool p_use_jitter, float p_cam_znear, float p_cam_zfar, bool p_cam_orthogonal) { + ERR_FAIL_COND_MSG(prefer_raster_effects, "Can't use compute version of BOKEH DOF with the mobile renderer."); + bokeh.push_constant.blur_far_active = p_dof_far; bokeh.push_constant.blur_far_begin = p_dof_far_begin; bokeh.push_constant.blur_far_end = p_dof_far_begin + p_dof_far_size; @@ -924,6 +1015,78 @@ void EffectsRD::bokeh_dof(RID p_base_texture, RID p_depth_texture, const Size2i RD::get_singleton()->compute_list_end(); } +void EffectsRD::blur_dof_raster(RID p_base_texture, RID p_depth_texture, const Size2i &p_base_texture_size, RID p_base_fb, RID p_secondary_texture, RID p_secondary_fb, bool p_dof_far, float p_dof_far_begin, float p_dof_far_size, bool p_dof_near, float p_dof_near_begin, float p_dof_near_size, float p_dof_blur_amount, RS::DOFBlurQuality p_quality, float p_cam_znear, float p_cam_zfar, bool p_cam_orthogonal) { + ERR_FAIL_COND_MSG(!prefer_raster_effects, "Can't use blur DOF with the clustered renderer."); + + memset(&blur_raster.push_constant, 0, sizeof(BlurRasterPushConstant)); + + BlurRasterMode blur_mode; + int qsteps[4] = { 4, 4, 10, 20 }; + uint32_t base_flags = p_cam_orthogonal ? BLUR_FLAG_USE_ORTHOGONAL_PROJECTION : 0; + + Vector2 pixel_size = Vector2(1.0 / p_base_texture_size.width, 1.0 / p_base_texture_size.height); + + blur_raster.push_constant.dof_radius = (p_dof_blur_amount * p_dof_blur_amount) / qsteps[p_quality]; + blur_raster.push_constant.pixel_size[0] = pixel_size.x; + blur_raster.push_constant.pixel_size[1] = pixel_size.y; + blur_raster.push_constant.camera_z_far = p_cam_zfar; + blur_raster.push_constant.camera_z_near = p_cam_znear; + + if (p_dof_far || p_dof_near) { + if (p_quality == RS::DOF_BLUR_QUALITY_HIGH) { + blur_mode = BLUR_MODE_DOF_HIGH; + } else if (p_quality == RS::DOF_BLUR_QUALITY_MEDIUM) { + blur_mode = BLUR_MODE_DOF_MEDIUM; + } else { // for LOW or VERYLOW we use LOW + blur_mode = BLUR_MODE_DOF_LOW; + } + + if (p_dof_far) { + base_flags |= BLUR_FLAG_DOF_FAR; + blur_raster.push_constant.dof_far_begin = p_dof_far_begin; + blur_raster.push_constant.dof_far_end = p_dof_far_begin + p_dof_far_size; + } + + if (p_dof_near) { + base_flags |= BLUR_FLAG_DOF_NEAR; + blur_raster.push_constant.dof_near_begin = p_dof_near_begin; + blur_raster.push_constant.dof_near_end = p_dof_near_begin - p_dof_near_size; + } + + //HORIZONTAL + RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_secondary_fb, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD); + RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, blur_raster.pipelines[blur_mode].get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_secondary_fb))); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_base_texture), 0); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_depth_texture), 1); + RD::get_singleton()->draw_list_bind_index_array(draw_list, index_array); + + blur_raster.push_constant.flags = base_flags | BLUR_FLAG_HORIZONTAL; + blur_raster.push_constant.dof_dir[0] = 1.0; + blur_raster.push_constant.dof_dir[1] = 0.0; + + RD::get_singleton()->draw_list_set_push_constant(draw_list, &blur_raster.push_constant, sizeof(BlurRasterPushConstant)); + + RD::get_singleton()->draw_list_draw(draw_list, true); + RD::get_singleton()->draw_list_end(); + + //VERTICAL + draw_list = RD::get_singleton()->draw_list_begin(p_base_fb, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD); + RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, blur_raster.pipelines[blur_mode].get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_base_fb))); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_secondary_texture), 0); + RD::get_singleton()->draw_list_bind_uniform_set(draw_list, _get_uniform_set_from_texture(p_depth_texture), 1); + RD::get_singleton()->draw_list_bind_index_array(draw_list, index_array); + + blur_raster.push_constant.flags = base_flags; + blur_raster.push_constant.dof_dir[0] = 0.0; + blur_raster.push_constant.dof_dir[1] = 1.0; + + RD::get_singleton()->draw_list_set_push_constant(draw_list, &blur_raster.push_constant, sizeof(BlurRasterPushConstant)); + + RD::get_singleton()->draw_list_draw(draw_list, true); + RD::get_singleton()->draw_list_end(); + } +} + void EffectsRD::gather_ssao(RD::ComputeListID p_compute_list, const Vector<RID> p_ao_slices, const SSAOSettings &p_settings, bool p_adaptive_base_pass, RID p_gather_uniform_set, RID p_importance_map_uniform_set) { RD::get_singleton()->compute_list_bind_uniform_set(p_compute_list, p_gather_uniform_set, 0); if ((p_settings.quality == RS::ENV_SSAO_QUALITY_ULTRA) && !p_adaptive_base_pass) { @@ -1464,7 +1627,35 @@ void EffectsRD::sort_buffer(RID p_uniform_set, int p_size) { RD::get_singleton()->compute_list_end(); } -EffectsRD::EffectsRD() { +EffectsRD::EffectsRD(bool p_prefer_raster_effects) { + prefer_raster_effects = p_prefer_raster_effects; + + if (prefer_raster_effects) { + // init blur shader (on compute use copy shader) + + Vector<String> blur_modes; + blur_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n"); // BLUR_MODE_GAUSSIAN_BLUR + blur_modes.push_back("\n#define MODE_GAUSSIAN_GLOW\n"); // BLUR_MODE_GAUSSIAN_GLOW + blur_modes.push_back("\n#define MODE_GAUSSIAN_GLOW\n#define GLOW_USE_AUTO_EXPOSURE\n"); // BLUR_MODE_GAUSSIAN_GLOW_AUTO_EXPOSURE + blur_modes.push_back("\n#define MODE_DOF_BLUR\n#define DOF_QUALITY_LOW\n"); // BLUR_MODE_DOF_LOW + blur_modes.push_back("\n#define MODE_DOF_BLUR\n#define DOF_QUALITY_MEDIUM\n"); // BLUR_MODE_DOF_MEDIUM + blur_modes.push_back("\n#define MODE_DOF_BLUR\n#define DOF_QUALITY_HIGH\n"); // BLUR_MODE_DOF_HIGH + + blur_raster.shader.initialize(blur_modes); + memset(&blur_raster.push_constant, 0, sizeof(BlurRasterPushConstant)); + blur_raster.shader_version = blur_raster.shader.version_create(); + + for (int i = 0; i < BLUR_MODE_MAX; i++) { + blur_raster.pipelines[i].setup(blur_raster.shader.version_get_shader(blur_raster.shader_version, i), RD::RENDER_PRIMITIVE_TRIANGLES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), RD::PipelineDepthStencilState(), RD::PipelineColorBlendState::create_disabled(), 0); + } + + } else { + // not used in clustered + for (int i = 0; i < BLUR_MODE_MAX; i++) { + blur_raster.pipelines[i].clear(); + } + } + { // Initialize copy Vector<String> copy_modes; copy_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n"); @@ -1483,10 +1674,21 @@ EffectsRD::EffectsRD() { copy.shader.initialize(copy_modes); memset(©.push_constant, 0, sizeof(CopyPushConstant)); + + if (prefer_raster_effects) { + // disable shaders we can't use + copy.shader.set_variant_enabled(COPY_MODE_GAUSSIAN_COPY, false); + copy.shader.set_variant_enabled(COPY_MODE_GAUSSIAN_COPY_8BIT, false); + copy.shader.set_variant_enabled(COPY_MODE_GAUSSIAN_GLOW, false); + copy.shader.set_variant_enabled(COPY_MODE_GAUSSIAN_GLOW_AUTO_EXPOSURE, false); + } + copy.shader_version = copy.shader.version_create(); for (int i = 0; i < COPY_MODE_MAX; i++) { - copy.pipelines[i] = RD::get_singleton()->compute_pipeline_create(copy.shader.version_get_shader(copy.shader_version, i)); + if (copy.shader.is_variant_enabled(i)) { + copy.pipelines[i] = RD::get_singleton()->compute_pipeline_create(copy.shader.version_get_shader(copy.shader_version, i)); + } } } { @@ -1551,7 +1753,20 @@ EffectsRD::EffectsRD() { } } - { + if (prefer_raster_effects) { + Vector<String> luminance_reduce_modes; + luminance_reduce_modes.push_back("\n#define FIRST_PASS\n"); // LUMINANCE_REDUCE_FRAGMENT_FIRST + luminance_reduce_modes.push_back("\n"); // LUMINANCE_REDUCE_FRAGMENT + luminance_reduce_modes.push_back("\n#define FINAL_PASS\n"); // LUMINANCE_REDUCE_FRAGMENT_FINAL + + luminance_reduce_raster.shader.initialize(luminance_reduce_modes); + memset(&luminance_reduce_raster.push_constant, 0, sizeof(LuminanceReduceRasterPushConstant)); + luminance_reduce_raster.shader_version = luminance_reduce_raster.shader.version_create(); + + for (int i = 0; i < LUMINANCE_REDUCE_FRAGMENT_MAX; i++) { + luminance_reduce_raster.pipelines[i].setup(luminance_reduce_raster.shader.version_get_shader(luminance_reduce_raster.shader_version, i), RD::RENDER_PRIMITIVE_TRIANGLES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), RD::PipelineDepthStencilState(), RD::PipelineColorBlendState::create_disabled(), 0); + } + } else { // Initialize luminance_reduce Vector<String> luminance_reduce_modes; luminance_reduce_modes.push_back("\n#define READ_TEXTURE\n"); @@ -1565,6 +1780,10 @@ EffectsRD::EffectsRD() { for (int i = 0; i < LUMINANCE_REDUCE_MAX; i++) { luminance_reduce.pipelines[i] = RD::get_singleton()->compute_pipeline_create(luminance_reduce.shader.version_get_shader(luminance_reduce.shader_version, i)); } + + for (int i = 0; i < LUMINANCE_REDUCE_FRAGMENT_MAX; i++) { + luminance_reduce_raster.pipelines[i].clear(); + } } { @@ -1583,7 +1802,9 @@ EffectsRD::EffectsRD() { cube_to_dp.pipeline.setup(shader, RD::RENDER_PRIMITIVE_TRIANGLES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), dss, RD::PipelineColorBlendState(), 0); } - { + if (prefer_raster_effects) { + // not supported + } else { // Initialize bokeh Vector<String> bokeh_modes; bokeh_modes.push_back("\n#define MODE_GEN_BLUR_SIZE\n"); @@ -1974,13 +2195,18 @@ EffectsRD::~EffectsRD() { RD::get_singleton()->free(ssao.gather_constants_buffer); RD::get_singleton()->free(ssao.importance_map_load_counter); - bokeh.shader.version_free(bokeh.shader_version); + if (prefer_raster_effects) { + blur_raster.shader.version_free(blur_raster.shader_version); + luminance_reduce_raster.shader.version_free(luminance_reduce_raster.shader_version); + } else { + bokeh.shader.version_free(bokeh.shader_version); + luminance_reduce.shader.version_free(luminance_reduce.shader_version); + } copy.shader.version_free(copy.shader_version); copy_to_fb.shader.version_free(copy_to_fb.shader_version); cube_to_dp.shader.version_free(cube_to_dp.shader_version); cubemap_downsampler.shader.version_free(cubemap_downsampler.shader_version); filter.shader.version_free(filter.shader_version); - luminance_reduce.shader.version_free(luminance_reduce.shader_version); resolve.shader.version_free(resolve.shader_version); roughness.shader.version_free(roughness.shader_version); roughness_limiter.shader.version_free(roughness_limiter.shader_version); diff --git a/servers/rendering/renderer_rd/effects_rd.h b/servers/rendering/renderer_rd/effects_rd.h index 33d32f0c57..d072564c23 100644 --- a/servers/rendering/renderer_rd/effects_rd.h +++ b/servers/rendering/renderer_rd/effects_rd.h @@ -33,6 +33,7 @@ #include "core/math/camera_matrix.h" #include "servers/rendering/renderer_rd/pipeline_cache_rd.h" +#include "servers/rendering/renderer_rd/shaders/blur_raster.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/bokeh_dof.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/copy.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/copy_to_fb.glsl.gen.h" @@ -41,6 +42,7 @@ #include "servers/rendering/renderer_rd/shaders/cubemap_filter.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/cubemap_roughness.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/luminance_reduce.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/luminance_reduce_raster.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/resolve.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/roughness_limiter.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/screen_space_reflection.glsl.gen.h" @@ -60,6 +62,63 @@ #include "servers/rendering_server.h" class EffectsRD { + enum BlurRasterMode { + BLUR_MODE_GAUSSIAN_BLUR, + BLUR_MODE_GAUSSIAN_GLOW, + BLUR_MODE_GAUSSIAN_GLOW_AUTO_EXPOSURE, + + BLUR_MODE_DOF_LOW, + BLUR_MODE_DOF_MEDIUM, + BLUR_MODE_DOF_HIGH, + + BLUR_MODE_MAX + }; + + enum { + BLUR_FLAG_HORIZONTAL = (1 << 0), + BLUR_FLAG_USE_ORTHOGONAL_PROJECTION = (1 << 1), + BLUR_FLAG_GLOW_FIRST_PASS = (1 << 2), + BLUR_FLAG_DOF_FAR = (1 << 3), + BLUR_FLAG_DOF_NEAR = (1 << 4), + }; + + struct BlurRasterPushConstant { + float pixel_size[2]; + uint32_t flags; + uint32_t pad; + + //glow + float glow_strength; + float glow_bloom; + float glow_hdr_threshold; + float glow_hdr_scale; + + float glow_exposure; + float glow_white; + float glow_luminance_cap; + float glow_auto_exposure_grey; + + //dof + float dof_far_begin; + float dof_far_end; + float dof_near_begin; + float dof_near_end; + + float dof_radius; + float dof_pad[3]; + + float dof_dir[2]; + float camera_z_far; + float camera_z_near; + }; + + struct BlurRaster { + BlurRasterPushConstant push_constant; + BlurRasterShaderRD shader; + RID shader_version; + PipelineCacheRD pipelines[BLUR_MODE_MAX]; + } blur_raster; + enum CopyMode { COPY_MODE_GAUSSIAN_COPY, COPY_MODE_GAUSSIAN_COPY_8BIT, @@ -239,6 +298,29 @@ class EffectsRD { RID pipelines[LUMINANCE_REDUCE_MAX]; } luminance_reduce; + enum LuminanceReduceRasterMode { + LUMINANCE_REDUCE_FRAGMENT_FIRST, + LUMINANCE_REDUCE_FRAGMENT, + LUMINANCE_REDUCE_FRAGMENT_FINAL, + LUMINANCE_REDUCE_FRAGMENT_MAX + }; + + struct LuminanceReduceRasterPushConstant { + int32_t source_size[2]; + int32_t dest_size[2]; + float exposure_adjust; + float min_luminance; + float max_luminance; + float pad[1]; + }; + + struct LuminanceReduceFragment { + LuminanceReduceRasterPushConstant push_constant; + LuminanceReduceRasterShaderRD shader; + RID shader_version; + PipelineCacheRD pipelines[LUMINANCE_REDUCE_FRAGMENT_MAX]; + } luminance_reduce_raster; + struct CopyToDPPushConstant { float z_far; float z_near; @@ -656,6 +738,8 @@ class EffectsRD { RID _get_compute_uniform_set_from_texture_pair(RID p_texture, RID p_texture2, bool p_use_mipmaps = false); RID _get_compute_uniform_set_from_image_pair(RID p_texture, RID p_texture2); + bool prefer_raster_effects; + public: void copy_to_fb_rect(RID p_source_rd_texture, RID p_dest_framebuffer, const Rect2i &p_rect, bool p_flip_y = false, bool p_force_luminance = false, bool p_alpha_to_zero = false, bool p_srgb = false, RID p_secondary = RID()); void copy_to_rect(RID p_source_rd_texture, RID p_dest_texture, const Rect2i &p_rect, bool p_flip_y = false, bool p_force_luminance = false, bool p_all_source = false, bool p_8_bit_dst = false, bool p_alpha_to_one = false); @@ -666,12 +750,16 @@ public: void gaussian_blur(RID p_source_rd_texture, RID p_texture, RID p_back_texture, const Rect2i &p_region, bool p_8bit_dst = false); void set_color(RID p_dest_texture, const Color &p_color, const Rect2i &p_region, bool p_8bit_dst = false); void gaussian_glow(RID p_source_rd_texture, RID p_back_texture, const Size2i &p_size, float p_strength = 1.0, bool p_high_quality = false, bool p_first_pass = false, float p_luminance_cap = 16.0, float p_exposure = 1.0, float p_bloom = 0.0, float p_hdr_bleed_treshold = 1.0, float p_hdr_bleed_scale = 1.0, RID p_auto_exposure = RID(), float p_auto_exposure_grey = 1.0); + void gaussian_glow_raster(RID p_source_rd_texture, RID p_framebuffer_half, RID p_rd_texture_half, RID p_dest_framebuffer, const Vector2 &p_pixel_size, float p_strength = 1.0, bool p_high_quality = false, bool p_first_pass = false, float p_luminance_cap = 16.0, float p_exposure = 1.0, float p_bloom = 0.0, float p_hdr_bleed_treshold = 1.0, float p_hdr_bleed_scale = 1.0, RID p_auto_exposure = RID(), float p_auto_exposure_grey = 1.0); void cubemap_roughness(RID p_source_rd_texture, RID p_dest_framebuffer, uint32_t p_face_id, uint32_t p_sample_count, float p_roughness, float p_size); void make_mipmap(RID p_source_rd_texture, RID p_dest_texture, const Size2i &p_size); void copy_cubemap_to_dp(RID p_source_rd_texture, RID p_dest_texture, const Rect2 &p_rect, float p_z_near, float p_z_far, bool p_dp_flip); void luminance_reduction(RID p_source_texture, const Size2i p_source_size, const Vector<RID> p_reduce, RID p_prev_luminance, float p_min_luminance, float p_max_luminance, float p_adjust, bool p_set = false); + void luminance_reduction_raster(RID p_source_texture, const Size2i p_source_size, const Vector<RID> p_reduce, Vector<RID> p_fb, RID p_prev_luminance, float p_min_luminance, float p_max_luminance, float p_adjust, bool p_set = false); + void bokeh_dof(RID p_base_texture, RID p_depth_texture, const Size2i &p_base_texture_size, RID p_secondary_texture, RID p_bokeh_texture1, RID p_bokeh_texture2, bool p_dof_far, float p_dof_far_begin, float p_dof_far_size, bool p_dof_near, float p_dof_near_begin, float p_dof_near_size, float p_bokeh_size, RS::DOFBokehShape p_bokeh_shape, RS::DOFBlurQuality p_quality, bool p_use_jitter, float p_cam_znear, float p_cam_zfar, bool p_cam_orthogonal); + void blur_dof_raster(RID p_base_texture, RID p_depth_texture, const Size2i &p_base_texture_size, RID p_base_fb, RID p_secondary_texture, RID p_secondary_fb, bool p_dof_far, float p_dof_far_begin, float p_dof_far_size, bool p_dof_near, float p_dof_near_begin, float p_dof_near_size, float p_dof_blur_amount, RS::DOFBlurQuality p_quality, float p_cam_znear, float p_cam_zfar, bool p_cam_orthogonal); struct TonemapSettings { bool use_glow = false; @@ -751,7 +839,7 @@ public: void sort_buffer(RID p_uniform_set, int p_size); - EffectsRD(); + EffectsRD(bool p_prefer_raster_effects); ~EffectsRD(); }; diff --git a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp index 438bbff53c..6d1c408a32 100644 --- a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp +++ b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp @@ -96,10 +96,6 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b RD::DataFormat color_format = RenderForwardMobile::singleton->_render_buffers_get_color_format(); if (p_msaa == RS::VIEWPORT_MSAA_DISABLED) { - if (color_format == RD::DATA_FORMAT_A2B10G10R10_UNORM_PACK32) { - // @TODO add a second color buffer for alpha as this format is RGB only - } - Vector<RID> fb; fb.push_back(p_color_buffer); fb.push_back(depth); @@ -165,15 +161,12 @@ bool RenderForwardMobile::free(RID p_rid) { RD::DataFormat RenderForwardMobile::_render_buffers_get_color_format() { // Using 32bit buffers enables AFBC on mobile devices which should have a definate performance improvement (MALI G710 and newer support this on 64bit RTs) - // NO ALPHA and unsigned float. - // @TODO No alpha is an issue, recommendation here is to add a second RT for alpha return RD::DATA_FORMAT_A2B10G10R10_UNORM_PACK32; } bool RenderForwardMobile::_render_buffers_can_be_storage() { // Using 32bit buffers enables AFBC on mobile devices which should have a definate performance improvement (MALI G710 and newer support this on 64bit RTs) - // NO ALPHA and unsigned float. - // @TODO No alpha is an issue, recommendation here is to add a second RT for alpha + // Doesn't support storage return false; } diff --git a/servers/rendering/renderer_rd/renderer_compositor_rd.cpp b/servers/rendering/renderer_rd/renderer_compositor_rd.cpp index e6ae66d56f..a7ee0dd141 100644 --- a/servers/rendering/renderer_rd/renderer_compositor_rd.cpp +++ b/servers/rendering/renderer_rd/renderer_compositor_rd.cpp @@ -280,6 +280,9 @@ RendererCompositorRD::RendererCompositorRD() { // default to our high end renderer scene = memnew(RendererSceneRenderImplementation::RenderForwardClustered(storage)); } + + // now we're ready to create our effects, + storage->init_effects(!scene->_render_buffers_can_be_storage()); } RendererCompositorRD::~RendererCompositorRD() { diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp index 4870f5f397..02f3a01ec3 100644 --- a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp +++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp @@ -1383,12 +1383,20 @@ void RendererSceneRenderRD::_allocate_blur_textures(RenderBuffers *rb) { uint32_t mipmaps_required = Image::get_image_required_mipmaps(rb->width, rb->height, Image::FORMAT_RGBAH); + // TODO make sure texture_create_shared_from_slice works for multiview + RD::TextureFormat tf; - tf.format = RD::DATA_FORMAT_R16G16B16A16_SFLOAT; + tf.format = _render_buffers_get_color_format(); // RD::DATA_FORMAT_R16G16B16A16_SFLOAT; tf.width = rb->width; tf.height = rb->height; - tf.texture_type = RD::TEXTURE_TYPE_2D; - tf.usage_bits = RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_COPY_TO_BIT; + tf.texture_type = rb->view_count > 1 ? RD::TEXTURE_TYPE_2D_ARRAY : RD::TEXTURE_TYPE_2D; + tf.array_layers = rb->view_count; + tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_COPY_TO_BIT; + if (_render_buffers_can_be_storage()) { + tf.usage_bits += RD::TEXTURE_USAGE_STORAGE_BIT; + } else { + tf.usage_bits += RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT; + } tf.mipmaps = mipmaps_required; rb->blur[0].texture = RD::get_singleton()->texture_create(tf, RD::TextureView()); @@ -1408,11 +1416,40 @@ void RendererSceneRenderRD::_allocate_blur_textures(RenderBuffers *rb) { mm.width = base_width; mm.height = base_height; + if (!_render_buffers_can_be_storage()) { + Vector<RID> fb; + fb.push_back(mm.texture); + + mm.fb = RD::get_singleton()->framebuffer_create(fb); + } + + if (!_render_buffers_can_be_storage()) { + // and half texture, this is an intermediate result so just allocate a texture, is this good enough? + tf.width = MAX(1, base_width >> 1); + tf.height = base_height; + tf.mipmaps = 1; // 1 or 0? + + mm.half_texture = RD::get_singleton()->texture_create(tf, RD::TextureView()); + + Vector<RID> half_fb; + half_fb.push_back(mm.half_texture); + mm.half_fb = RD::get_singleton()->framebuffer_create(half_fb); + } + rb->blur[0].mipmaps.push_back(mm); if (i > 0) { mm.texture = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->blur[1].texture, 0, i - 1); + if (!_render_buffers_can_be_storage()) { + Vector<RID> fb; + fb.push_back(mm.texture); + + mm.fb = RD::get_singleton()->framebuffer_create(fb); + + // We can re-use the half texture here as it is an intermediate result + } + rb->blur[1].mipmaps.push_back(mm); } @@ -1435,26 +1472,48 @@ void RendererSceneRenderRD::_allocate_luminance_textures(RenderBuffers *rb) { tf.format = RD::DATA_FORMAT_R32_SFLOAT; tf.width = w; tf.height = h; - tf.usage_bits = RD::TEXTURE_USAGE_STORAGE_BIT; bool final = w == 1 && h == 1; - if (final) { - tf.usage_bits |= RD::TEXTURE_USAGE_SAMPLING_BIT; + if (_render_buffers_can_be_storage()) { + tf.usage_bits = RD::TEXTURE_USAGE_STORAGE_BIT; + if (final) { + tf.usage_bits |= RD::TEXTURE_USAGE_SAMPLING_BIT; + } + } else { + tf.usage_bits = RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT; } RID texture = RD::get_singleton()->texture_create(tf, RD::TextureView()); rb->luminance.reduce.push_back(texture); + if (!_render_buffers_can_be_storage()) { + Vector<RID> fb; + fb.push_back(texture); + + rb->luminance.fb.push_back(RD::get_singleton()->framebuffer_create(fb)); + } if (final) { rb->luminance.current = RD::get_singleton()->texture_create(tf, RD::TextureView()); + + if (!_render_buffers_can_be_storage()) { + Vector<RID> fb; + fb.push_back(rb->luminance.current); + + rb->luminance.current_fb = RD::get_singleton()->framebuffer_create(fb); + } break; } } } void RendererSceneRenderRD::_free_render_buffer_data(RenderBuffers *rb) { + if (rb->texture_fb.is_valid()) { + RD::get_singleton()->free(rb->texture_fb); + rb->texture_fb = RID(); + } + if (rb->texture.is_valid()) { RD::get_singleton()->free(rb->texture); rb->texture = RID(); @@ -1466,19 +1525,43 @@ void RendererSceneRenderRD::_free_render_buffer_data(RenderBuffers *rb) { } for (int i = 0; i < 2; i++) { + for (int m = 0; m < rb->blur[i].mipmaps.size(); m++) { + // do we free the texture slice here? or is it enough to free the main texture? + + // do free the mobile extra stuff + if (rb->blur[i].mipmaps[m].fb.is_valid()) { + RD::get_singleton()->free(rb->blur[i].mipmaps[m].fb); + } + if (rb->blur[i].mipmaps[m].half_fb.is_valid()) { + RD::get_singleton()->free(rb->blur[i].mipmaps[m].half_fb); + } + if (rb->blur[i].mipmaps[m].half_texture.is_valid()) { + RD::get_singleton()->free(rb->blur[i].mipmaps[m].half_texture); + } + } + rb->blur[i].mipmaps.clear(); + if (rb->blur[i].texture.is_valid()) { RD::get_singleton()->free(rb->blur[i].texture); rb->blur[i].texture = RID(); - rb->blur[i].mipmaps.clear(); } } + for (int i = 0; i < rb->luminance.fb.size(); i++) { + RD::get_singleton()->free(rb->luminance.fb[i]); + } + rb->luminance.fb.clear(); + for (int i = 0; i < rb->luminance.reduce.size(); i++) { RD::get_singleton()->free(rb->luminance.reduce[i]); } - rb->luminance.reduce.clear(); + if (rb->luminance.current_fb.is_valid()) { + RD::get_singleton()->free(rb->luminance.current_fb); + rb->luminance.current_fb = RID(); + } + if (rb->luminance.current.is_valid()) { RD::get_singleton()->free(rb->luminance.current); rb->luminance.current = RID(); @@ -1750,17 +1833,27 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende CameraEffects *camfx = camera_effects_owner.getornull(p_render_data->camera_effects); bool can_use_effects = rb->width >= 8 && rb->height >= 8; + bool can_use_storage = _render_buffers_can_be_storage(); + + // @TODO IMPLEMENT MULTIVIEW, all effects need to support stereo buffers or effects are only applied to the left eye if (can_use_effects && camfx && (camfx->dof_blur_near_enabled || camfx->dof_blur_far_enabled) && camfx->dof_blur_amount > 0.0) { + RD::get_singleton()->draw_command_begin_label("DOF"); if (rb->blur[0].texture.is_null()) { _allocate_blur_textures(rb); } - float bokeh_size = camfx->dof_blur_amount * 64.0; - storage->get_effects()->bokeh_dof(rb->texture, rb->depth_texture, Size2i(rb->width, rb->height), rb->blur[0].mipmaps[0].texture, rb->blur[1].mipmaps[0].texture, rb->blur[0].mipmaps[1].texture, camfx->dof_blur_far_enabled, camfx->dof_blur_far_distance, camfx->dof_blur_far_transition, camfx->dof_blur_near_enabled, camfx->dof_blur_near_distance, camfx->dof_blur_near_transition, bokeh_size, dof_blur_bokeh_shape, dof_blur_quality, dof_blur_use_jitter, p_render_data->z_near, p_render_data->z_far, p_render_data->cam_ortogonal); + if (can_use_storage) { + float bokeh_size = camfx->dof_blur_amount * 64.0; + storage->get_effects()->bokeh_dof(rb->texture, rb->depth_texture, Size2i(rb->width, rb->height), rb->blur[0].mipmaps[0].texture, rb->blur[1].mipmaps[0].texture, rb->blur[0].mipmaps[1].texture, camfx->dof_blur_far_enabled, camfx->dof_blur_far_distance, camfx->dof_blur_far_transition, camfx->dof_blur_near_enabled, camfx->dof_blur_near_distance, camfx->dof_blur_near_transition, bokeh_size, dof_blur_bokeh_shape, dof_blur_quality, dof_blur_use_jitter, p_render_data->z_near, p_render_data->z_far, p_render_data->cam_ortogonal); + } else { + storage->get_effects()->blur_dof_raster(rb->texture, rb->depth_texture, Size2i(rb->width, rb->height), rb->texture_fb, rb->blur[0].mipmaps[0].texture, rb->blur[0].mipmaps[0].fb, camfx->dof_blur_far_enabled, camfx->dof_blur_far_distance, camfx->dof_blur_far_transition, camfx->dof_blur_near_enabled, camfx->dof_blur_near_distance, camfx->dof_blur_near_transition, camfx->dof_blur_amount, dof_blur_quality, p_render_data->z_near, p_render_data->z_far, p_render_data->cam_ortogonal); + } + RD::get_singleton()->draw_command_end_label(); } if (can_use_effects && env && env->auto_exposure) { + RD::get_singleton()->draw_command_begin_label("Auto exposure"); if (rb->luminance.current.is_null()) { _allocate_luminance_textures(rb); } @@ -1769,16 +1862,26 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende rb->auto_exposure_version = env->auto_exposure_version; double step = env->auto_exp_speed * time_step; - storage->get_effects()->luminance_reduction(rb->texture, Size2i(rb->width, rb->height), rb->luminance.reduce, rb->luminance.current, env->min_luminance, env->max_luminance, step, set_immediate); - + if (can_use_storage) { + storage->get_effects()->luminance_reduction(rb->texture, Size2i(rb->width, rb->height), rb->luminance.reduce, rb->luminance.current, env->min_luminance, env->max_luminance, step, set_immediate); + } else { + storage->get_effects()->luminance_reduction_raster(rb->texture, Size2i(rb->width, rb->height), rb->luminance.reduce, rb->luminance.fb, rb->luminance.current, env->min_luminance, env->max_luminance, step, set_immediate); + } //swap final reduce with prev luminance SWAP(rb->luminance.current, rb->luminance.reduce.write[rb->luminance.reduce.size() - 1]); + if (!can_use_storage) { + SWAP(rb->luminance.current_fb, rb->luminance.fb.write[rb->luminance.fb.size() - 1]); + } + RenderingServerDefault::redraw_request(); //redraw all the time if auto exposure rendering is on + RD::get_singleton()->draw_command_end_label(); } int max_glow_level = -1; if (can_use_effects && env && env->glow_enabled) { + RD::get_singleton()->draw_command_begin_label("Gaussian Glow"); + /* see that blur textures are allocated */ if (rb->blur[1].texture.is_null()) { @@ -1804,14 +1907,26 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende if (env->auto_exposure && rb->luminance.current.is_valid()) { luminance_texture = rb->luminance.current; } - storage->get_effects()->gaussian_glow(rb->texture, rb->blur[1].mipmaps[i].texture, Size2i(vp_w, vp_h), env->glow_strength, glow_high_quality, true, env->glow_hdr_luminance_cap, env->exposure, env->glow_bloom, env->glow_hdr_bleed_threshold, env->glow_hdr_bleed_scale, luminance_texture, env->auto_exp_scale); + if (can_use_storage) { + storage->get_effects()->gaussian_glow(rb->texture, rb->blur[1].mipmaps[i].texture, Size2i(vp_w, vp_h), env->glow_strength, glow_high_quality, true, env->glow_hdr_luminance_cap, env->exposure, env->glow_bloom, env->glow_hdr_bleed_threshold, env->glow_hdr_bleed_scale, luminance_texture, env->auto_exp_scale); + } else { + storage->get_effects()->gaussian_glow_raster(rb->texture, rb->blur[1].mipmaps[i].half_fb, rb->blur[1].mipmaps[i].half_texture, rb->blur[1].mipmaps[i].fb, Size2i(vp_w, vp_h), env->glow_strength, glow_high_quality, true, env->glow_hdr_luminance_cap, env->exposure, env->glow_bloom, env->glow_hdr_bleed_threshold, env->glow_hdr_bleed_scale, luminance_texture, env->auto_exp_scale); + } } else { - storage->get_effects()->gaussian_glow(rb->blur[1].mipmaps[i - 1].texture, rb->blur[1].mipmaps[i].texture, Size2i(vp_w, vp_h), env->glow_strength, glow_high_quality); + if (can_use_storage) { + storage->get_effects()->gaussian_glow(rb->blur[1].mipmaps[i - 1].texture, rb->blur[1].mipmaps[i].texture, Size2i(vp_w, vp_h), env->glow_strength, glow_high_quality); + } else { + storage->get_effects()->gaussian_glow_raster(rb->blur[1].mipmaps[i - 1].texture, rb->blur[1].mipmaps[i].half_fb, rb->blur[1].mipmaps[i].half_texture, rb->blur[1].mipmaps[i].fb, Vector2(1.0 / vp_w, 1.0 / vp_h), env->glow_strength, glow_high_quality); + } } } + + RD::get_singleton()->draw_command_end_label(); } { + RD::get_singleton()->draw_command_begin_label("Tonemap"); + //tonemap EffectsRD::TonemapSettings tonemap; @@ -1870,6 +1985,8 @@ void RendererSceneRenderRD::_render_buffers_post_process_and_tonemap(const Rende tonemap.view_count = p_render_data->view_count; storage->get_effects()->tonemapper(rb->texture, storage->render_target_get_rd_framebuffer(rb->render_target), tonemap); + + RD::get_singleton()->draw_command_end_label(); } storage->render_target_disable_clear_request(rb->render_target); @@ -2197,6 +2314,14 @@ void RendererSceneRenderRD::render_buffers_configure(RID p_render_buffers, RID p rb->depth_texture = RD::get_singleton()->texture_create(tf, RD::TextureView()); } + if (!_render_buffers_can_be_storage()) { + // ONLY USED ON MOBILE RENDERER, ONLY USED FOR POST EFFECTS! + Vector<RID> fb; + fb.push_back(rb->texture); + + rb->texture_fb = RD::get_singleton()->framebuffer_create(fb, RenderingDevice::INVALID_ID, rb->view_count); + } + rb->data->configure(rb->texture, rb->depth_texture, p_width, p_height, p_msaa, p_view_count); if (is_clustered_enabled()) { diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.h b/servers/rendering/renderer_rd/renderer_scene_render_rd.h index 5e0281002d..bb06eb608f 100644 --- a/servers/rendering/renderer_rd/renderer_scene_render_rd.h +++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.h @@ -450,6 +450,7 @@ private: RID texture; //main texture for rendering to, must be filled after done rendering RID depth_texture; //main depth texture + RID texture_fb; // framebuffer for the main texture, ONLY USED FOR MOBILE RENDERER POST EFFECTS, DO NOT USE FOR RENDERING 3D!!! RendererSceneGIRD::SDFGI *sdfgi = nullptr; VolumetricFog *volumetric_fog = nullptr; @@ -465,6 +466,11 @@ private: RID texture; int width; int height; + + // only used on mobile renderer + RID fb; + RID half_texture; + RID half_fb; }; Vector<Mipmap> mipmaps; @@ -475,6 +481,10 @@ private: struct Luminance { Vector<RID> reduce; RID current; + + // used only on mobile renderer + Vector<RID> fb; + RID current_fb; } luminance; struct SSAO { diff --git a/servers/rendering/renderer_rd/renderer_storage_rd.cpp b/servers/rendering/renderer_rd/renderer_storage_rd.cpp index 9d898f6275..178fb60885 100644 --- a/servers/rendering/renderer_rd/renderer_storage_rd.cpp +++ b/servers/rendering/renderer_rd/renderer_storage_rd.cpp @@ -4959,7 +4959,7 @@ void RendererStorageRD::particles_set_view_axis(RID p_particles, const Vector3 & RD::get_singleton()->compute_list_dispatch_threads(compute_list, particles->amount, 1, 1); RD::get_singleton()->compute_list_end(); - effects.sort_buffer(particles->particles_sort_uniform_set, particles->amount); + effects->sort_buffer(particles->particles_sort_uniform_set, particles->amount); } copy_push_constant.total_particles *= copy_push_constant.total_particles; @@ -7535,7 +7535,7 @@ void RendererStorageRD::render_target_copy_to_back_buffer(RID p_render_target, c //single texture copy for backbuffer //RD::get_singleton()->texture_copy(rt->color, rt->backbuffer_mipmap0, Vector3(region.position.x, region.position.y, 0), Vector3(region.position.x, region.position.y, 0), Vector3(region.size.x, region.size.y, 1), 0, 0, 0, 0, true); - effects.copy_to_rect(rt->color, rt->backbuffer_mipmap0, region, false, false, false, true, true); + effects->copy_to_rect(rt->color, rt->backbuffer_mipmap0, region, false, false, false, true, true); if (!p_gen_mipmaps) { return; @@ -7551,7 +7551,7 @@ void RendererStorageRD::render_target_copy_to_back_buffer(RID p_render_target, c region.size.y = MAX(1, region.size.y >> 1); const RenderTarget::BackbufferMipmap &mm = rt->backbuffer_mipmaps[i]; - effects.gaussian_blur(prev_texture, mm.mipmap, mm.mipmap_copy, region, true); + effects->gaussian_blur(prev_texture, mm.mipmap, mm.mipmap_copy, region, true); prev_texture = mm.mipmap; } } @@ -7574,7 +7574,7 @@ void RendererStorageRD::render_target_clear_back_buffer(RID p_render_target, con } //single texture copy for backbuffer - effects.set_color(rt->backbuffer_mipmap0, p_color, region, true); + effects->set_color(rt->backbuffer_mipmap0, p_color, region, true); } void RendererStorageRD::render_target_gen_back_buffer_mipmaps(RID p_render_target, const Rect2i &p_region) { @@ -7604,7 +7604,7 @@ void RendererStorageRD::render_target_gen_back_buffer_mipmaps(RID p_render_targe region.size.y = MAX(1, region.size.y >> 1); const RenderTarget::BackbufferMipmap &mm = rt->backbuffer_mipmaps[i]; - effects.gaussian_blur(prev_texture, mm.mipmap, mm.mipmap_copy, region, true); + effects->gaussian_blur(prev_texture, mm.mipmap, mm.mipmap_copy, region, true); prev_texture = mm.mipmap; } } @@ -7925,14 +7925,14 @@ void RendererStorageRD::_update_decal_atlas() { while ((K = decal_atlas.textures.next(K))) { DecalAtlas::Texture *t = decal_atlas.textures.getptr(*K); Texture *src_tex = texture_owner.getornull(*K); - effects.copy_to_atlas_fb(src_tex->rd_texture, mm.fb, t->uv_rect, draw_list, false, t->panorama_to_dp_users > 0); + effects->copy_to_atlas_fb(src_tex->rd_texture, mm.fb, t->uv_rect, draw_list, false, t->panorama_to_dp_users > 0); } RD::get_singleton()->draw_list_end(); prev_texture = mm.texture; } else { - effects.copy_to_fb_rect(prev_texture, mm.fb, Rect2i(Point2i(), mm.size)); + effects->copy_to_fb_rect(prev_texture, mm.fb, Rect2i(Point2i(), mm.size)); prev_texture = mm.texture; } } else { @@ -8804,8 +8804,13 @@ bool RendererStorageRD::free(RID p_rid) { return true; } +void RendererStorageRD::init_effects(bool p_prefer_raster_effects) { + effects = memnew(EffectsRD(p_prefer_raster_effects)); +} + EffectsRD *RendererStorageRD::get_effects() { - return &effects; + ERR_FAIL_NULL_V_MSG(effects, nullptr, "Effects haven't been initialised yet."); + return effects; } void RendererStorageRD::capture_timestamps_begin() { @@ -9535,4 +9540,9 @@ RendererStorageRD::~RendererStorageRD() { if (decal_atlas.texture.is_valid()) { RD::get_singleton()->free(decal_atlas.texture); } + + if (effects) { + memdelete(effects); + effects = NULL; + } } diff --git a/servers/rendering/renderer_rd/renderer_storage_rd.h b/servers/rendering/renderer_rd/renderer_storage_rd.h index f471874c8e..b290c07705 100644 --- a/servers/rendering/renderer_rd/renderer_storage_rd.h +++ b/servers/rendering/renderer_rd/renderer_storage_rd.h @@ -1290,7 +1290,7 @@ private: void _update_global_variables(); /* EFFECTS */ - EffectsRD effects; + EffectsRD *effects = NULL; public: virtual bool can_create_resources_async() const; @@ -2374,6 +2374,7 @@ public: static RendererStorageRD *base_singleton; + void init_effects(bool p_prefer_raster_effects); EffectsRD *get_effects(); RendererStorageRD(); diff --git a/servers/rendering/renderer_rd/shaders/blur_raster.glsl b/servers/rendering/renderer_rd/shaders/blur_raster.glsl new file mode 100644 index 0000000000..b1d1c2365e --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/blur_raster.glsl @@ -0,0 +1,228 @@ +/* clang-format off */ +#[vertex] + +#version 450 + +#VERSION_DEFINES + +#include "blur_raster_inc.glsl" + +layout(location = 0) out vec2 uv_interp; +/* clang-format on */ + +void main() { + vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0)); + uv_interp = base_arr[gl_VertexIndex]; + + gl_Position = vec4(uv_interp * 2.0 - 1.0, 0.0, 1.0); +} + +/* clang-format off */ +#[fragment] + +#version 450 + +#VERSION_DEFINES + +#include "blur_raster_inc.glsl" + +layout(location = 0) in vec2 uv_interp; +/* clang-format on */ + +layout(set = 0, binding = 0) uniform sampler2D source_color; + +#ifdef GLOW_USE_AUTO_EXPOSURE +layout(set = 1, binding = 0) uniform sampler2D source_auto_exposure; +#endif + +layout(location = 0) out vec4 frag_color; + +//DOF +#ifdef MODE_DOF_BLUR + +layout(set = 1, binding = 0) uniform sampler2D dof_source_depth; + +#ifdef DOF_QUALITY_LOW +const int dof_kernel_size = 5; +const int dof_kernel_from = 2; +const float dof_kernel[5] = float[](0.153388, 0.221461, 0.250301, 0.221461, 0.153388); +#endif + +#ifdef DOF_QUALITY_MEDIUM +const int dof_kernel_size = 11; +const int dof_kernel_from = 5; +const float dof_kernel[11] = float[](0.055037, 0.072806, 0.090506, 0.105726, 0.116061, 0.119726, 0.116061, 0.105726, 0.090506, 0.072806, 0.055037); + +#endif + +#ifdef DOF_QUALITY_HIGH +const int dof_kernel_size = 21; +const int dof_kernel_from = 10; +const float dof_kernel[21] = float[](0.028174, 0.032676, 0.037311, 0.041944, 0.046421, 0.050582, 0.054261, 0.057307, 0.059587, 0.060998, 0.061476, 0.060998, 0.059587, 0.057307, 0.054261, 0.050582, 0.046421, 0.041944, 0.037311, 0.032676, 0.028174); +#endif + +#endif + +void main() { +#ifdef MODE_MIPMAP + + vec2 pix_size = blur.pixel_size; + vec4 color = texture(source_color, uv_interp + vec2(-0.5, -0.5) * pix_size); + color += texture(source_color, uv_interp + vec2(0.5, -0.5) * pix_size); + color += texture(source_color, uv_interp + vec2(0.5, 0.5) * pix_size); + color += texture(source_color, uv_interp + vec2(-0.5, 0.5) * pix_size); + frag_color = color / 4.0; + +#endif + +#ifdef MODE_GAUSSIAN_BLUR + + //Simpler blur uses SIGMA2 for the gaussian kernel for a stronger effect + + if (bool(blur.flags & FLAG_HORIZONTAL)) { + vec2 pix_size = blur.pixel_size; + pix_size *= 0.5; //reading from larger buffer, so use more samples + vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.214607; + color += texture(source_color, uv_interp + vec2(1.0, 0.0) * pix_size) * 0.189879; + color += texture(source_color, uv_interp + vec2(2.0, 0.0) * pix_size) * 0.131514; + color += texture(source_color, uv_interp + vec2(3.0, 0.0) * pix_size) * 0.071303; + color += texture(source_color, uv_interp + vec2(-1.0, 0.0) * pix_size) * 0.189879; + color += texture(source_color, uv_interp + vec2(-2.0, 0.0) * pix_size) * 0.131514; + color += texture(source_color, uv_interp + vec2(-3.0, 0.0) * pix_size) * 0.071303; + frag_color = color; + } else { + vec2 pix_size = blur.pixel_size; + vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.38774; + color += texture(source_color, uv_interp + vec2(0.0, 1.0) * pix_size) * 0.24477; + color += texture(source_color, uv_interp + vec2(0.0, 2.0) * pix_size) * 0.06136; + color += texture(source_color, uv_interp + vec2(0.0, -1.0) * pix_size) * 0.24477; + color += texture(source_color, uv_interp + vec2(0.0, -2.0) * pix_size) * 0.06136; + frag_color = color; + } +#endif + +#ifdef MODE_GAUSSIAN_GLOW + + //Glow uses larger sigma 1 for a more rounded blur effect + +#define GLOW_ADD(m_ofs, m_mult) \ + { \ + vec2 ofs = uv_interp + m_ofs * pix_size; \ + vec4 c = texture(source_color, ofs) * m_mult; \ + if (any(lessThan(ofs, vec2(0.0))) || any(greaterThan(ofs, vec2(1.0)))) { \ + c *= 0.0; \ + } \ + color += c; \ + } + + if (bool(blur.flags & FLAG_HORIZONTAL)) { + vec2 pix_size = blur.pixel_size; + pix_size *= 0.5; //reading from larger buffer, so use more samples + vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.174938; + GLOW_ADD(vec2(1.0, 0.0), 0.165569); + GLOW_ADD(vec2(2.0, 0.0), 0.140367); + GLOW_ADD(vec2(3.0, 0.0), 0.106595); + GLOW_ADD(vec2(-1.0, 0.0), 0.165569); + GLOW_ADD(vec2(-2.0, 0.0), 0.140367); + GLOW_ADD(vec2(-3.0, 0.0), 0.106595); + color *= blur.glow_strength; + frag_color = color; + } else { + vec2 pix_size = blur.pixel_size; + vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.288713; + GLOW_ADD(vec2(0.0, 1.0), 0.233062); + GLOW_ADD(vec2(0.0, 2.0), 0.122581); + GLOW_ADD(vec2(0.0, -1.0), 0.233062); + GLOW_ADD(vec2(0.0, -2.0), 0.122581); + color *= blur.glow_strength; + frag_color = color; + } + +#undef GLOW_ADD + + if (bool(blur.flags & FLAG_GLOW_FIRST_PASS)) { +#ifdef GLOW_USE_AUTO_EXPOSURE + + frag_color /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / blur.glow_auto_exposure_grey; +#endif + frag_color *= blur.glow_exposure; + + float luminance = max(frag_color.r, max(frag_color.g, frag_color.b)); + float feedback = max(smoothstep(blur.glow_hdr_threshold, blur.glow_hdr_threshold + blur.glow_hdr_scale, luminance), blur.glow_bloom); + + frag_color = min(frag_color * feedback, vec4(blur.glow_luminance_cap)); + } + +#endif + +#ifdef MODE_DOF_BLUR + + vec4 color_accum = vec4(0.0); + + float depth = texture(dof_source_depth, uv_interp, 0.0).r; + depth = depth * 2.0 - 1.0; + + if (bool(blur.flags & FLAG_USE_ORTHOGONAL_PROJECTION)) { + depth = ((depth + (blur.camera_z_far + blur.camera_z_near) / (blur.camera_z_far - blur.camera_z_near)) * (blur.camera_z_far - blur.camera_z_near)) / 2.0; + } else { + depth = 2.0 * blur.camera_z_near * blur.camera_z_far / (blur.camera_z_far + blur.camera_z_near - depth * (blur.camera_z_far - blur.camera_z_near)); + } + + // mix near and far blur amount + float amount = 1.0; + if (bool(blur.flags & FLAG_DOF_FAR)) { + amount *= 1.0 - smoothstep(blur.dof_far_begin, blur.dof_far_end, depth); + } + if (bool(blur.flags & FLAG_DOF_NEAR)) { + amount *= smoothstep(blur.dof_near_end, blur.dof_near_begin, depth); + } + amount = 1.0 - amount; + + if (amount > 0.0) { + float k_accum = 0.0; + + for (int i = 0; i < dof_kernel_size; i++) { + int int_ofs = i - dof_kernel_from; + vec2 tap_uv = uv_interp + blur.dof_dir * float(int_ofs) * amount * blur.dof_radius; + + float tap_k = dof_kernel[i]; + + float tap_depth = texture(dof_source_depth, tap_uv, 0.0).r; + tap_depth = tap_depth * 2.0 - 1.0; + + if (bool(blur.flags & FLAG_USE_ORTHOGONAL_PROJECTION)) { + tap_depth = ((tap_depth + (blur.camera_z_far + blur.camera_z_near) / (blur.camera_z_far - blur.camera_z_near)) * (blur.camera_z_far - blur.camera_z_near)) / 2.0; + } else { + tap_depth = 2.0 * blur.camera_z_near * blur.camera_z_far / (blur.camera_z_far + blur.camera_z_near - tap_depth * (blur.camera_z_far - blur.camera_z_near)); + } + + // mix near and far blur amount + float tap_amount = 1.0; + if (bool(blur.flags & FLAG_DOF_FAR)) { + tap_amount *= mix(1.0 - smoothstep(blur.dof_far_begin, blur.dof_far_end, tap_depth), 0.0, int_ofs == 0); + } + if (bool(blur.flags & FLAG_DOF_NEAR)) { + tap_amount *= mix(smoothstep(blur.dof_near_end, blur.dof_near_begin, tap_depth), 0.0, int_ofs == 0); + } + tap_amount = 1.0 - tap_amount; + + tap_amount *= tap_amount * tap_amount; //prevent undesired glow effect + + vec4 tap_color = texture(source_color, tap_uv, 0.0) * tap_k; + + k_accum += tap_k * tap_amount; + color_accum += tap_color * tap_amount; + } + + if (k_accum > 0.0) { + color_accum /= k_accum; + } + + frag_color = color_accum; ///k_accum; + } else { + // we are in focus, don't waste time + frag_color = texture(source_color, uv_interp, 0.0); + } + +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/blur_raster_inc.glsl b/servers/rendering/renderer_rd/shaders/blur_raster_inc.glsl new file mode 100644 index 0000000000..6ea968e595 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/blur_raster_inc.glsl @@ -0,0 +1,36 @@ +#define FLAG_HORIZONTAL (1 << 0) +#define FLAG_USE_ORTHOGONAL_PROJECTION (1 << 1) +#define FLAG_GLOW_FIRST_PASS (1 << 2) +#define FLAG_DOF_FAR (1 << 3) +#define FLAG_DOF_NEAR (1 << 4) + +layout(push_constant, binding = 1, std430) uniform Blur { + vec2 pixel_size; + uint flags; + uint pad; + + // Glow. + float glow_strength; + float glow_bloom; + float glow_hdr_threshold; + float glow_hdr_scale; + + float glow_exposure; + float glow_white; + float glow_luminance_cap; + float glow_auto_exposure_grey; + + // DOF. + float dof_far_begin; + float dof_far_end; + float dof_near_begin; + float dof_near_end; + + float dof_radius; + float dof_pad[3]; + + vec2 dof_dir; + float camera_z_far; + float camera_z_near; +} +blur; diff --git a/servers/rendering/renderer_rd/shaders/luminance_reduce_raster.glsl b/servers/rendering/renderer_rd/shaders/luminance_reduce_raster.glsl new file mode 100644 index 0000000000..29ebd74a90 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/luminance_reduce_raster.glsl @@ -0,0 +1,74 @@ +/* clang-format off */ +#[vertex] + +#version 450 + +#VERSION_DEFINES + +#include "luminance_reduce_raster_inc.glsl" + +layout(location = 0) out vec2 uv_interp; +/* clang-format on */ + +void main() { + vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0)); + uv_interp = base_arr[gl_VertexIndex]; + + gl_Position = vec4(uv_interp * 2.0 - 1.0, 0.0, 1.0); +} + +/* clang-format off */ +#[fragment] + +#version 450 + +#VERSION_DEFINES + +#include "luminance_reduce_raster_inc.glsl" + +layout(location = 0) in vec2 uv_interp; +/* clang-format on */ + +layout(set = 0, binding = 0) uniform sampler2D source_exposure; + +#ifdef FINAL_PASS +layout(set = 1, binding = 0) uniform sampler2D prev_luminance; +#endif + +layout(location = 0) out highp float luminance; + +void main() { + ivec2 dest_pos = ivec2(uv_interp * settings.dest_size); + ivec2 src_pos = ivec2(uv_interp * settings.source_size); + + ivec2 next_pos = (dest_pos + ivec2(1)) * settings.source_size / settings.dest_size; + next_pos = max(next_pos, src_pos + ivec2(1)); //so it at least reads one pixel + + highp vec3 source_color = vec3(0.0); + for (int i = src_pos.x; i < next_pos.x; i++) { + for (int j = src_pos.y; j < next_pos.y; j++) { + source_color += texelFetch(source_exposure, ivec2(i, j), 0).rgb; + } + } + + source_color /= float((next_pos.x - src_pos.x) * (next_pos.y - src_pos.y)); + +#ifdef FIRST_PASS + luminance = max(source_color.r, max(source_color.g, source_color.b)); + + // This formula should be more "accurate" but gave an overexposed result when testing. + // Leaving it here so we can revisit it if we want. + // luminance = source_color.r * 0.21 + source_color.g * 0.71 + source_color.b * 0.07; +#else + luminance = source_color.r; +#endif + +#ifdef FINAL_PASS + // Obtain our target luminance + luminance = clamp(luminance, settings.min_luminance, settings.max_luminance); + + // Now smooth to our transition + highp float prev_lum = texelFetch(prev_luminance, ivec2(0, 0), 0).r; //1 pixel previous luminance + luminance = prev_lum + (luminance - prev_lum) * clamp(settings.exposure_adjust, 0.0, 1.0); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/luminance_reduce_raster_inc.glsl b/servers/rendering/renderer_rd/shaders/luminance_reduce_raster_inc.glsl new file mode 100644 index 0000000000..ed389ffe56 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/luminance_reduce_raster_inc.glsl @@ -0,0 +1,11 @@ + +layout(push_constant, binding = 1, std430) uniform PushConstant { + ivec2 source_size; + ivec2 dest_size; + + float exposure_adjust; + float min_luminance; + float max_luminance; + float pad; +} +settings; |