tmaq
/
tornavis
1
0
Fork 0
Code Issues 2 Pull Requests 29 Packages Projects 24 Releases Wiki Activity

Workbench: Fix: Performance Regressions 

Optimize Workbench performance so it's on par with the previous
implementation.
Most of these changes are barely noticeable on powerful GPUs,
but can cause a notable performance improvement on old or low-end
hardware.

* Avoid unnecessary texture copies and draw directly to the viewport
   textures.
* Optimize-out depth/stencil reads, using stencil testing instead.
* Avoid using `Texture::clear` and use framebuffer clears instead.
* Avoid framebuffer state changes (always use the same attachments).
* Avoid constant variation of acquired `TextureFromPool`s.

Fix #113010

Pull Request: https://projects.blender.org/blender/blender/pulls/113251
This commit is contained in:
Miguel Pozo 2023-10-06 15:31:44 +02:00
parent 1a30e0597b
commit fe39456ba5
11 changed files with 214 additions and 228 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
source/blender/draw/engines/workbench/shaders/infos/workbench_composite_info.hh
View File

@ -10,10 +10,9 @@
* \{ */
GPU_SHADER_CREATE_INFO(workbench_composite)
.sampler(3, ImageType::FLOAT_2D, "normal_tx")
.sampler(4, ImageType::FLOAT_2D, "material_tx")
.sampler(5, ImageType::DEPTH_2D, "depth_tx")
.sampler(6, ImageType::UINT_2D, "stencil_tx")
.sampler(3, ImageType::DEPTH_2D, "depth_tx")
.sampler(4, ImageType::FLOAT_2D, "normal_tx")
.sampler(5, ImageType::FLOAT_2D, "material_tx")
.uniform_buf(WB_WORLD_SLOT, "WorldData", "world_data")
.typedef_source("workbench_shader_shared.h")
.push_constant(Type::BOOL, "forceShadowing")
@ -35,23 +34,32 @@ GPU_SHADER_CREATE_INFO(workbench_resolve_opaque_flat).define("WORKBENCH_LIGHTING
GPU_SHADER_CREATE_INFO(workbench_resolve_curvature)
.define("WORKBENCH_CURVATURE")
.sampler(7, ImageType::UINT_2D, "object_id_tx");
.sampler(6, ImageType::UINT_2D, "object_id_tx");
GPU_SHADER_CREATE_INFO(workbench_resolve_cavity)
.define("WORKBENCH_CAVITY")
/* TODO(@pragma37): GPU_SAMPLER_EXTEND_MODE_REPEAT is set in CavityEffect,
* it doesn't work here? */
.sampler(8, ImageType::FLOAT_2D, "jitter_tx")
.sampler(7, ImageType::FLOAT_2D, "jitter_tx")
.uniform_buf(5, "vec4", "cavity_samples[512]");
GPU_SHADER_CREATE_INFO(workbench_resolve_shadow)
.define("WORKBENCH_SHADOW")
.sampler(8, ImageType::UINT_2D, "stencil_tx");
/* Variations */
#define WORKBENCH_FINAL_VARIATION(name, ...) \
GPU_SHADER_CREATE_INFO(name).additional_info(__VA_ARGS__).do_static_compilation(true);
#define WORKBENCH_RESOLVE_SHADOW_VARIATION(prefix, ...) \
WORKBENCH_FINAL_VARIATION(prefix##_shadow, "workbench_resolve_shadow", __VA_ARGS__) \
WORKBENCH_FINAL_VARIATION(prefix##_no_shadow, __VA_ARGS__)
#define WORKBENCH_CURVATURE_VARIATIONS(prefix, ...) \
WORKBENCH_FINAL_VARIATION(prefix##_curvature, "workbench_resolve_curvature", __VA_ARGS__) \
WORKBENCH_FINAL_VARIATION(prefix##_no_curvature, __VA_ARGS__)
WORKBENCH_RESOLVE_SHADOW_VARIATION( \
prefix##_curvature, "workbench_resolve_curvature", __VA_ARGS__) \
WORKBENCH_RESOLVE_SHADOW_VARIATION(prefix##_no_curvature, __VA_ARGS__)
#define WORKBENCH_CAVITY_VARIATIONS(prefix, ...) \
WORKBENCH_CURVATURE_VARIATIONS(prefix##_cavity, "workbench_resolve_cavity", __VA_ARGS__) \

2
source/blender/draw/engines/workbench/shaders/infos/workbench_depth_info.hh
View File

@ -12,8 +12,6 @@ GPU_SHADER_CREATE_INFO(workbench_merge_depth)
.do_static_compilation(true);
GPU_SHADER_CREATE_INFO(workbench_overlay_depth)
.sampler(0, ImageType::DEPTH_2D, "depth_tx")
.sampler(1, ImageType::UINT_2D, "stencil_tx")
.fragment_source("workbench_overlay_depth_frag.glsl")
.additional_info("draw_fullscreen")
.depth_write(DepthWrite::ANY)

65
source/blender/draw/engines/workbench/shaders/workbench_composite_frag.glsl
View File

@ -12,73 +12,58 @@
void main()
{
vec2 uv = uvcoordsvar.st;
float depth = texture(depth_tx, uv).r;
if (depth == 1.0) {
/* Skip the background. */
discard;
return;
}
/* Normal and Incident vector are in view-space. Lighting is evaluated in view-space. */
vec3 V = get_view_vector_from_screen_uv(uv);
vec3 N = workbench_normal_decode(texture(normal_tx, uv));
vec4 mat_data = texture(material_tx, uv);
float depth = texture(depth_tx, uv).r;
vec3 base_color = mat_data.rgb;
vec4 color = world_data.background_color;
vec4 color = vec4(1.0);
/* Background pixels. */
if (depth != 1.0) {
#ifdef WORKBENCH_LIGHTING_MATCAP
/* When using matcaps, mat_data.a is the back-face sign. */
N = (mat_data.a > 0.0) ? N : -N;
color.rgb = get_matcap_lighting(matcap_tx, base_color, N, V);
/* When using matcaps, mat_data.a is the back-face sign. */
N = (mat_data.a > 0.0) ? N : -N;
color.rgb = get_matcap_lighting(matcap_tx, base_color, N, V);
#endif
#ifdef WORKBENCH_LIGHTING_STUDIO
float roughness, metallic;
workbench_float_pair_decode(mat_data.a, roughness, metallic);
color.rgb = get_world_lighting(base_color, roughness, metallic, N, V);
float roughness, metallic;
workbench_float_pair_decode(mat_data.a, roughness, metallic);
color.rgb = get_world_lighting(base_color, roughness, metallic, N, V);
#endif
#ifdef WORKBENCH_LIGHTING_FLAT
color.rgb = base_color;
color.rgb = base_color;
#endif
#if defined(WORKBENCH_CAVITY) || defined(WORKBENCH_CURVATURE)
float cavity = 0.0, edges = 0.0, curvature = 0.0;
float cavity = 0.0, edges = 0.0, curvature = 0.0;
# ifdef WORKBENCH_CAVITY
cavity_compute(uv, depth_tx, normal_tx, cavity, edges);
cavity_compute(uv, depth_tx, normal_tx, cavity, edges);
# endif
# ifdef WORKBENCH_CURVATURE
curvature_compute(uv, object_id_tx, normal_tx, curvature);
curvature_compute(uv, object_id_tx, normal_tx, curvature);
# endif
float final_cavity_factor = clamp(
(1.0 - cavity) * (1.0 + edges) * (1.0 + curvature), 0.0, 4.0);
float final_cavity_factor = clamp((1.0 - cavity) * (1.0 + edges) * (1.0 + curvature), 0.0, 4.0);
color.rgb *= final_cavity_factor;
color.rgb *= final_cavity_factor;
#endif
bool shadow = texture(stencil_tx, uv).r != 0;
color.rgb *= get_shadow(N, shadow);
color.a = 1.0f;
}
#ifdef WORKBENCH_OUTLINE
vec3 offset = vec3(world_data.viewport_size_inv, 0.0) * world_data.ui_scale;
uint center_id = texture(object_id_tx, uv).r;
uvec4 adjacent_ids = uvec4(texture(object_id_tx, uv + offset.zy).r,
texture(object_id_tx, uv - offset.zy).r,
texture(object_id_tx, uv + offset.xz).r,
texture(object_id_tx, uv - offset.xz).r);
float outline_opacity = 1.0 - dot(vec4(equal(uvec4(center_id), adjacent_ids)), vec4(0.25));
color = mix(color, world_data.object_outline_color, outline_opacity);
#ifdef WORKBENCH_SHADOW
bool shadow = texture(stencil_tx, uv).r != 0;
color.rgb *= get_shadow(N, shadow);
#endif
if (all(equal(color, world_data.background_color))) {
discard;
}
else {
fragColor = color;
}
fragColor = color;
}

8
source/blender/draw/engines/workbench/shaders/workbench_merge_depth_frag.glsl
View File

@ -4,11 +4,5 @@
void main()
{
float depth = texture(depth_tx, uvcoordsvar.xy).r;
if (depth != 1.0) {
gl_FragDepth = depth;
}
else {
discard;
}
gl_FragDepth = texture(depth_tx, uvcoordsvar.xy).r;
}

13
source/blender/draw/engines/workbench/shaders/workbench_overlay_depth_frag.glsl
View File

@ -6,14 +6,7 @@
void main()
{
uint stencil = texelFetch(stencil_tx, ivec2(gl_FragCoord.xy), 0).r;
if (stencil != 0) {
/* Set the depth to 0 for "In Front" objects,
* so the Overlay engine doesn't draw on top of them. */
gl_FragDepth = 0.0;
}
else {
float depth = texelFetch(depth_tx, ivec2(gl_FragCoord.xy), 0).r;
gl_FragDepth = depth;
}
/* Set the depth to 0 for "In Front" objects,
* so the Overlay engine doesn't draw on top of them. */
gl_FragDepth = 0.0;
}

125
source/blender/draw/engines/workbench/workbench_effect_antialiasing.cc
View File

@ -139,17 +139,15 @@ AntiAliasingPass::~AntiAliasingPass()
void AntiAliasingPass::init(const SceneState &scene_state)
{
enabled_ = scene_state.draw_aa;
sample_ = scene_state.sample;
samples_len_ = scene_state.samples_len;
}
void AntiAliasingPass::sync(SceneResources &resources, int2 resolution)
void AntiAliasingPass::sync(const SceneState &scene_state, SceneResources &resources)
{
overlay_depth_ps_.init();
overlay_depth_ps_.state_set(DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_ALWAYS);
overlay_depth_ps_.state_set(DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_ALWAYS |
DRW_STATE_STENCIL_EQUAL);
overlay_depth_ps_.state_stencil(0x00, 0xFF, uint8_t(StencilBits::OBJECT_IN_FRONT));
overlay_depth_ps_.shader_set(overlay_depth_sh_);
overlay_depth_ps_.bind_texture("depth_tx", &resources.depth_tx);
overlay_depth_ps_.bind_texture("stencil_tx", &stencil_tx_);
overlay_depth_ps_.draw_procedural(GPU_PRIM_TRIS, 1, 3);
if (!enabled_) {
@ -158,17 +156,23 @@ void AntiAliasingPass::sync(SceneResources &resources, int2 resolution)
return;
}
taa_accumulation_tx_.ensure_2d(
GPU_RGBA16F, resolution, GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
smaa_viewport_metrics_ = float4(float2(1.0f / float2(scene_state.resolution)),
scene_state.resolution);
smaa_mix_factor_ = 1.0f - clamp_f(scene_state.sample / 4.0f, 0.0f, 1.0f);
taa_accumulation_tx_.ensure_2d(GPU_RGBA16F,
scene_state.resolution,
GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
sample0_depth_tx_.ensure_2d(GPU_DEPTH24_STENCIL8,
resolution,
scene_state.resolution,
GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
sample0_depth_in_front_tx_.ensure_2d(
GPU_DEPTH24_STENCIL8, resolution, GPU_TEXTURE_USAGE_ATTACHMENT);
GPU_DEPTH24_STENCIL8, scene_state.resolution, GPU_TEXTURE_USAGE_ATTACHMENT);
taa_accumulation_ps_.init();
taa_accumulation_ps_.state_set(sample_ == 0 ? DRW_STATE_WRITE_COLOR :
DRW_STATE_WRITE_COLOR | DRW_STATE_BLEND_ADD_FULL);
taa_accumulation_ps_.state_set(scene_state.sample == 0 ?
DRW_STATE_WRITE_COLOR :
DRW_STATE_WRITE_COLOR | DRW_STATE_BLEND_ADD_FULL);
taa_accumulation_ps_.shader_set(taa_accumulation_sh_);
taa_accumulation_ps_.bind_texture("colorBuffer", &resources.color_tx);
taa_accumulation_ps_.push_constant("samplesWeights", weights_, 9);
@ -204,7 +208,7 @@ void AntiAliasingPass::sync(SceneResources &resources, int2 resolution)
smaa_resolve_ps_.draw_procedural(GPU_PRIM_TRIS, 1, 3);
}
void AntiAliasingPass::setup_view(View &view, int2 resolution)
void AntiAliasingPass::setup_view(View &view, const SceneState &scene_state)
{
if (!enabled_) {
return;
@ -213,22 +217,22 @@ void AntiAliasingPass::setup_view(View &view, int2 resolution)
const TaaSamples &taa_samples = get_taa_samples();
float2 sample_offset;
switch (samples_len_) {
switch (scene_state.samples_len) {
default:
case 5:
sample_offset = taa_samples.x5[sample_];
sample_offset = taa_samples.x5[scene_state.sample];
break;
case 8:
sample_offset = taa_samples.x8[sample_];
sample_offset = taa_samples.x8[scene_state.sample];
break;
case 11:
sample_offset = taa_samples.x11[sample_];
sample_offset = taa_samples.x11[scene_state.sample];
break;
case 16:
sample_offset = taa_samples.x16[sample_];
sample_offset = taa_samples.x16[scene_state.sample];
break;
case 32:
sample_offset = taa_samples.x32[sample_];
sample_offset = taa_samples.x32[scene_state.sample];
break;
}
@ -242,43 +246,58 @@ void AntiAliasingPass::setup_view(View &view, int2 resolution)
DRW_view_viewmat_get(default_view, viewmat.ptr(), false);
DRW_view_persmat_get(default_view, persmat.ptr(), false);
window_translate_m4(
winmat.ptr(), persmat.ptr(), sample_offset.x / resolution.x, sample_offset.y / resolution.y);
window_translate_m4(winmat.ptr(),
persmat.ptr(),
sample_offset.x / scene_state.resolution.x,
sample_offset.y / scene_state.resolution.y);
view.sync(viewmat, winmat);
}
void AntiAliasingPass::draw(Manager &manager,
View &view,
SceneResources &resources,
int2 resolution,
GPUTexture *depth_tx,
GPUTexture *depth_in_front_tx,
GPUTexture *color_tx)
const SceneState &scene_state,
SceneResources &resources)
{
auto draw_overlay_depth = [&](GPUTexture *target) {
stencil_tx_ = resources.stencil_view.extract(manager, resources.depth_tx);
overlay_depth_fb_.ensure(GPU_ATTACHMENT_TEXTURE(target));
if (resources.depth_in_front_tx.is_valid() && scene_state.sample == 0 &&
scene_state.overlays_enabled)
{
overlay_depth_fb_.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx));
overlay_depth_fb_.bind();
manager.submit(overlay_depth_ps_);
};
}
if (!enabled_) {
GPU_texture_copy(color_tx, resources.color_tx);
draw_overlay_depth(depth_tx);
return;
}
const bool last_sample = scene_state.sample + 1 == scene_state.samples_len;
if (scene_state.samples_len > 1 && (scene_state.overlays_enabled || DRW_state_is_scene_render()))
{
if (scene_state.sample == 0) {
GPU_texture_copy(sample0_depth_tx_, resources.depth_tx);
if (resources.depth_in_front_tx.is_valid()) {
GPU_texture_copy(sample0_depth_in_front_tx_, resources.depth_in_front_tx);
}
}
else if (!DRW_state_is_scene_render() || last_sample) {
/* Copy back the saved depth buffer for correct overlays. */
GPU_texture_copy(resources.depth_tx, sample0_depth_tx_);
if (resources.depth_in_front_tx.is_valid()) {
GPU_texture_copy(resources.depth_in_front_tx, sample0_depth_in_front_tx_);
}
}
}
/**
* We always do SMAA on top of TAA accumulation, unless the number of samples of TAA is already
* high. This ensure a smoother transition.
* If TAA accumulation is finished, we only blit the result.
*/
const bool last_sample = sample_ + 1 == samples_len_;
const bool taa_finished = sample_ >= samples_len_;
const bool taa_finished = scene_state.sample >= scene_state.samples_len;
if (!taa_finished) {
if (sample_ == 0) {
if (scene_state.sample == 0) {
weight_accum_ = 0;
}
/* Accumulate result to the TAA buffer. */
@ -288,30 +307,17 @@ void AntiAliasingPass::draw(Manager &manager,
weight_accum_ += weights_sum_;
}
if (sample_ == 0) {
draw_overlay_depth(sample0_depth_tx_);
GPU_texture_copy(sample0_depth_in_front_tx_, resources.depth_in_front_tx);
}
if (!DRW_state_is_scene_render()) {
/* Copy back the saved depth buffer for correct overlays. */
GPU_texture_copy(depth_tx, sample0_depth_tx_);
GPU_texture_copy(depth_in_front_tx, sample0_depth_in_front_tx_);
}
else if (last_sample) {
GPU_texture_copy(depth_tx, sample0_depth_tx_);
/* There's no depth_in_front_tx in scene image renders. */
}
/** Always acquire to avoid constant allocation/deallocation. */
smaa_weight_tx_.acquire(scene_state.resolution,
GPU_RGBA8,
GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
smaa_edge_tx_.acquire(scene_state.resolution,
GPU_RG8,
GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
if (!DRW_state_is_image_render() || last_sample) {
smaa_weight_tx_.acquire(
resolution, GPU_RGBA8, GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
smaa_mix_factor_ = 1.0f - clamp_f(sample_ / 4.0f, 0.0f, 1.0f);
smaa_viewport_metrics_ = float4(float2(1.0f / float2(resolution)), resolution);
/* After a certain point SMAA is no longer necessary. */
if (smaa_mix_factor_ > 0.0f) {
smaa_edge_tx_.acquire(
resolution, GPU_RG8, GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
smaa_edge_fb_.ensure(GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(smaa_edge_tx_));
smaa_edge_fb_.bind();
manager.submit(smaa_edge_detect_ps_, view);
@ -319,13 +325,14 @@ void AntiAliasingPass::draw(Manager &manager,
smaa_weight_fb_.ensure(GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(smaa_weight_tx_));
smaa_weight_fb_.bind();
manager.submit(smaa_aa_weight_ps_, view);
smaa_edge_tx_.release();
}
smaa_resolve_fb_.ensure(GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(color_tx));
smaa_resolve_fb_.ensure(GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(resources.color_tx));
smaa_resolve_fb_.bind();
manager.submit(smaa_resolve_ps_, view);
smaa_weight_tx_.release();
}
smaa_edge_tx_.release();
smaa_weight_tx_.release();
}
} // namespace blender::workbench

64
source/blender/draw/engines/workbench/workbench_engine.cc
View File

@ -69,12 +69,6 @@ class Instance {
void begin_sync()
{
const float2 viewport_size = DRW_viewport_size_get();
const int2 resolution = {int(viewport_size.x), int(viewport_size.y)};
resources.depth_tx.ensure_2d(GPU_DEPTH24_STENCIL8,
resolution,
GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT |
GPU_TEXTURE_USAGE_MIP_SWIZZLE_VIEW);
resources.material_buf.clear();
opaque_ps.sync(scene_state, resources);
@ -85,7 +79,7 @@ class Instance {
volume_ps.sync(resources);
outline_ps.sync(resources);
dof_ps.sync(resources);
anti_aliasing_ps.sync(resources, scene_state.resolution);
anti_aliasing_ps.sync(scene_state, resources);
}
void end_sync()
@ -438,38 +432,38 @@ class Instance {
if (scene_state.render_finished) {
/* Just copy back the already rendered result */
anti_aliasing_ps.draw(
manager, view, resources, resolution, depth_tx, depth_in_front_tx, color_tx);
anti_aliasing_ps.draw(manager, view, scene_state, resources);
return;
}
anti_aliasing_ps.setup_view(view, resolution);
anti_aliasing_ps.setup_view(view, scene_state);
resources.color_tx.acquire(
resolution, GPU_RGBA16F, GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
resources.color_tx.clear(resources.world_buf.background_color);
resources.depth_tx.wrap(depth_tx);
resources.color_tx.wrap(color_tx);
GPUAttachment id_attachment = GPU_ATTACHMENT_NONE;
if (scene_state.draw_object_id) {
resources.object_id_tx.acquire(
resolution, GPU_R16UI, GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT);
resources.object_id_tx.clear(uint4(0));
id_attachment = GPU_ATTACHMENT_TEXTURE(resources.object_id_tx);
}
Framebuffer fb = Framebuffer("Workbench.Clear");
fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx));
fb.bind();
GPU_framebuffer_clear_depth_stencil(fb, 1.0f, 0x00);
resources.clear_fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx),
GPU_ATTACHMENT_TEXTURE(resources.color_tx),
id_attachment);
resources.clear_fb.bind();
float4 clear_colors[2] = {scene_state.background_color, float4(0.0f)};
GPU_framebuffer_multi_clear(resources.clear_fb, reinterpret_cast<float(*)[4]>(clear_colors));
GPU_framebuffer_clear_depth_stencil(resources.clear_fb, 1.0f, 0x00);
bool needs_depth_in_front = !transparent_ps.accumulation_in_front_ps_.is_empty() ||
scene_state.sample == 0;
if (needs_depth_in_front) {
resources.depth_in_front_tx.acquire(resolution,
GPU_DEPTH24_STENCIL8,
GPU_TEXTURE_USAGE_SHADER_READ |
GPU_TEXTURE_USAGE_ATTACHMENT);
fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_in_front_tx));
fb.bind();
GPU_framebuffer_clear_depth_stencil(fb, 1.0f, 0x00);
(!opaque_ps.gbuffer_in_front_ps_.is_empty() &&
scene_state.overlays_enabled && scene_state.sample == 0);
resources.depth_in_front_tx.wrap(needs_depth_in_front ? depth_in_front_tx : nullptr);
if ((!needs_depth_in_front && scene_state.overlays_enabled) ||
(needs_depth_in_front && opaque_ps.gbuffer_in_front_ps_.is_empty()))
{
resources.clear_in_front_fb.ensure(GPU_ATTACHMENT_TEXTURE(depth_in_front_tx));
resources.clear_in_front_fb.bind();
GPU_framebuffer_clear_depth_stencil(resources.clear_in_front_fb, 1.0f, 0x00);
}
opaque_ps.draw(
@ -480,12 +474,9 @@ class Instance {
volume_ps.draw(manager, view, resources);
outline_ps.draw(manager, resources);
dof_ps.draw(manager, view, resources, resolution);
anti_aliasing_ps.draw(
manager, view, resources, resolution, depth_tx, depth_in_front_tx, color_tx);
anti_aliasing_ps.draw(manager, view, scene_state, resources);
resources.color_tx.release();
resources.object_id_tx.release();
resources.depth_in_front_tx.release();
}
void draw_viewport(Manager &manager,
@ -512,8 +503,7 @@ class Instance {
/* Re-sync anything dependent on scene_state.sample. */
resources.init(scene_state);
dof_ps.init(scene_state);
anti_aliasing_ps.init(scene_state);
anti_aliasing_ps.sync(resources, scene_state.resolution);
anti_aliasing_ps.sync(scene_state, resources);
}
this->draw(manager, depth_tx, depth_in_front_tx, color_tx);
}
@ -622,9 +612,11 @@ static bool workbench_render_framebuffers_init()
"txl.color", size.x, size.y, 1, GPU_RGBA16F, usage, nullptr);
dtxl->depth = GPU_texture_create_2d(
"txl.depth", size.x, size.y, 1, GPU_DEPTH24_STENCIL8, usage, nullptr);
dtxl->depth_in_front = GPU_texture_create_2d(
"txl.depth_in_front", size.x, size.y, 1, GPU_DEPTH24_STENCIL8, usage, nullptr);
}
if (!(dtxl->depth && dtxl->color)) {
if (!(dtxl->depth && dtxl->color && dtxl->depth_in_front)) {
return false;
}

88
source/blender/draw/engines/workbench/workbench_mesh_passes.cc
View File

@ -112,19 +112,20 @@ PassMain::Sub &MeshPass::get_subpass(
void OpaquePass::sync(const SceneState &scene_state, SceneResources &resources)
{
DRWState state = DRW_STATE_WRITE_COLOR | DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_LESS_EQUAL |
scene_state.cull_state;
DRW_STATE_WRITE_STENCIL | scene_state.cull_state;
bool clip = scene_state.clip_planes.size() > 0;
DRWState in_front_state = state | DRW_STATE_WRITE_STENCIL | DRW_STATE_STENCIL_ALWAYS;
DRWState in_front_state = state | DRW_STATE_STENCIL_ALWAYS;
gbuffer_in_front_ps_.init_pass(resources, in_front_state, scene_state.clip_planes.size());
gbuffer_in_front_ps_.state_stencil(0xFF, 0xFF, 0x00);
gbuffer_in_front_ps_.state_stencil(uint8_t(StencilBits::OBJECT_IN_FRONT), 0xFF, 0x00);
gbuffer_in_front_ps_.init_subpasses(
ePipelineType::OPAQUE, scene_state.lighting_type, clip, resources.shader_cache);
state |= DRW_STATE_STENCIL_NEQUAL;
gbuffer_ps_.init_pass(resources, state, scene_state.clip_planes.size());
gbuffer_ps_.state_stencil(0x00, 0xFF, 0xFF);
gbuffer_ps_.state_stencil(
uint8_t(StencilBits::OBJECT), 0xFF, uint8_t(StencilBits::OBJECT_IN_FRONT));
gbuffer_ps_.init_subpasses(
ePipelineType::OPAQUE, scene_state.lighting_type, clip, resources.shader_cache);
@ -133,7 +134,8 @@ void OpaquePass::sync(const SceneState &scene_state, SceneResources &resources)
deferred_ps_.shader_set(resources.shader_cache.resolve_shader_get(ePipelineType::OPAQUE,
scene_state.lighting_type,
scene_state.draw_cavity,
scene_state.draw_curvature));
scene_state.draw_curvature,
scene_state.draw_shadows));
deferred_ps_.push_constant("forceShadowing", false);
deferred_ps_.bind_ubo(WB_WORLD_SLOT, resources.world_buf);
deferred_ps_.bind_texture(WB_MATCAP_SLOT, resources.matcap_tx);
@ -165,11 +167,11 @@ void OpaquePass::draw(Manager &manager,
}
if (!gbuffer_in_front_ps_.is_empty()) {
opaque_fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_material_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_normal_tx),
object_id_attachment);
opaque_fb.bind();
gbuffer_in_front_fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_material_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_normal_tx),
object_id_attachment);
gbuffer_in_front_fb.bind();
manager.submit(gbuffer_in_front_ps_, view);
@ -179,55 +181,40 @@ void OpaquePass::draw(Manager &manager,
}
if (!gbuffer_ps_.is_empty()) {
opaque_fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_material_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_normal_tx),
object_id_attachment);
opaque_fb.bind();
gbuffer_fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_material_tx),
GPU_ATTACHMENT_TEXTURE(gbuffer_normal_tx),
object_id_attachment);
gbuffer_fb.bind();
manager.submit(gbuffer_ps_, view);
}
bool needs_stencil_copy = shadow_pass && !gbuffer_in_front_ps_.is_empty();
Texture *depth_stencil_tx = nullptr;
if (needs_stencil_copy) {
if (shadow_pass) {
shadow_depth_stencil_tx.ensure_2d(GPU_DEPTH24_STENCIL8,
resolution,
GPU_TEXTURE_USAGE_SHADER_READ |
GPU_TEXTURE_USAGE_ATTACHMENT |
GPU_TEXTURE_USAGE_MIP_SWIZZLE_VIEW);
GPU_texture_copy(shadow_depth_stencil_tx, resources.depth_tx);
clear_fb.ensure(GPU_ATTACHMENT_TEXTURE(shadow_depth_stencil_tx));
clear_fb.bind();
GPU_framebuffer_clear_stencil(clear_fb, 0);
depth_stencil_tx = shadow_depth_stencil_tx.ptr();
opaque_fb.ensure(GPU_ATTACHMENT_TEXTURE(*depth_stencil_tx));
opaque_fb.bind();
GPU_framebuffer_clear_stencil(opaque_fb, 0);
shadow_pass->draw(
manager, view, resources, **&shadow_depth_stencil_tx, !gbuffer_in_front_ps_.is_empty());
deferred_ps_stencil_tx = resources.stencil_view.extract(manager, shadow_depth_stencil_tx);
}
else {
shadow_depth_stencil_tx.free();
depth_stencil_tx = resources.depth_tx.ptr();
deferred_ps_stencil_tx = nullptr;
}
if (shadow_pass) {
shadow_pass->draw(
manager, view, resources, **depth_stencil_tx, !gbuffer_in_front_ps_.is_empty());
}
deferred_ps_stencil_tx = resources.stencil_view.extract(manager, *depth_stencil_tx);
opaque_fb.ensure(GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(resources.color_tx));
opaque_fb.bind();
deferred_fb.ensure(GPU_ATTACHMENT_NONE, GPU_ATTACHMENT_TEXTURE(resources.color_tx));
deferred_fb.bind();
manager.submit(deferred_ps_, view);
if (shadow_pass && !needs_stencil_copy) {
opaque_fb.ensure(GPU_ATTACHMENT_TEXTURE(resources.depth_tx));
opaque_fb.bind();
GPU_framebuffer_clear_stencil(opaque_fb, 0);
}
gbuffer_normal_tx.release();
gbuffer_material_tx.release();
}
@ -257,7 +244,8 @@ void TransparentPass::sync(const SceneState &scene_state, SceneResources &resour
accumulation_ps_.init_pass(
resources, state | DRW_STATE_STENCIL_NEQUAL, scene_state.clip_planes.size());
accumulation_ps_.state_stencil(0x00, 0xFF, 0xFF);
accumulation_ps_.state_stencil(
uint8_t(StencilBits::OBJECT), 0xFF, uint8_t(StencilBits::OBJECT_IN_FRONT));
accumulation_ps_.clear_color(float4(0.0f, 0.0f, 0.0f, 1.0f));
accumulation_ps_.init_subpasses(
ePipelineType::TRANSPARENT, scene_state.lighting_type, clip, resources.shader_cache);
@ -335,13 +323,13 @@ TransparentDepthPass::~TransparentDepthPass()
void TransparentDepthPass::sync(const SceneState &scene_state, SceneResources &resources)
{
DRWState state = DRW_STATE_WRITE_COLOR | DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_LESS_EQUAL |
scene_state.cull_state;
DRW_STATE_WRITE_STENCIL | scene_state.cull_state;
bool clip = scene_state.clip_planes.size() > 0;
DRWState in_front_state = state | DRW_STATE_WRITE_STENCIL | DRW_STATE_STENCIL_ALWAYS;
DRWState in_front_state = state | DRW_STATE_STENCIL_ALWAYS;
in_front_ps_.init_pass(resources, in_front_state, scene_state.clip_planes.size());
in_front_ps_.state_stencil(0xFF, 0xFF, 0x00);
in_front_ps_.state_stencil(uint8_t(StencilBits::OBJECT_IN_FRONT), 0xFF, 0x00);
in_front_ps_.init_subpasses(
ePipelineType::OPAQUE, eLightingType::FLAT, clip, resources.shader_cache);
@ -350,15 +338,17 @@ void TransparentDepthPass::sync(const SceneState &scene_state, SceneResources &r
}
merge_ps_.init();
merge_ps_.shader_set(merge_sh_);
merge_ps_.state_set(DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_ALWAYS | DRW_STATE_WRITE_STENCIL |
DRW_STATE_STENCIL_ALWAYS);
merge_ps_.state_stencil(0xFF, 0xFF, 0x00);
merge_ps_.state_set(DRW_STATE_WRITE_DEPTH | DRW_STATE_DEPTH_LESS | DRW_STATE_WRITE_STENCIL |
DRW_STATE_STENCIL_EQUAL);
merge_ps_.state_stencil(
uint8_t(StencilBits::OBJECT_IN_FRONT), 0xFF, uint8_t(StencilBits::OBJECT_IN_FRONT));
merge_ps_.bind_texture("depth_tx", &resources.depth_in_front_tx);
merge_ps_.draw_procedural(GPU_PRIM_TRIS, 1, 3);
state |= DRW_STATE_STENCIL_NEQUAL;
main_ps_.init_pass(resources, state, scene_state.clip_planes.size());
main_ps_.state_stencil(0x00, 0xFF, 0xFF);
main_ps_.state_stencil(
uint8_t(StencilBits::OBJECT), 0xFF, uint8_t(StencilBits::OBJECT_IN_FRONT));
main_ps_.init_subpasses(
ePipelineType::OPAQUE, eLightingType::FLAT, clip, resources.shader_cache);
}

41
source/blender/draw/engines/workbench/workbench_private.hh
View File

@ -32,14 +32,15 @@ class ShaderCache {
GPUShader *resolve_shader_get(ePipelineType pipeline_type,
eLightingType lighting_type,
bool cavity = false,
bool curvature = false);
bool curvature = false,
bool shadow = false);
private:
/* TODO(fclem): We might want to change to a Map since most shader will never be compiled. */
GPUShader *prepass_shader_cache_[pipeline_type_len][geometry_type_len][shader_type_len]
[lighting_type_len][2 /*clip*/] = {{{{{nullptr}}}}};
GPUShader *resolve_shader_cache_[pipeline_type_len][lighting_type_len][2 /*cavity*/]
[2 /*curvature*/] = {{{{nullptr}}}};
[2 /*curvature*/][2 /*shadow*/] = {{{{nullptr}}}};
};
struct Material {
@ -96,6 +97,8 @@ struct SceneState {
bool reset_taa_next_sample = false;
bool render_finished = false;
bool overlays_enabled = false;
/* Used when material_type == eMaterialType::SINGLE */
Material material_override = Material(float3(1.0f));
/* When r == -1.0 the shader uses the vertex color */
@ -196,10 +199,14 @@ struct SceneResources {
StringRefNull current_matcap = {};
Texture matcap_tx = "matcap_tx";
TextureFromPool color_tx = "wb_color_tx";
TextureFromPool object_id_tx = "wb_object_id_tx";
Texture depth_tx = "wb_depth_tx";
TextureFromPool depth_in_front_tx = "wb_depth_in_front_tx";
TextureRef color_tx;
TextureRef depth_tx;
TextureRef depth_in_front_tx;
Framebuffer clear_fb = {"Clear Main"};
Framebuffer clear_in_front_fb = {"Clear In Front"};
StorageVectorBuffer<Material> material_buf = {"material_buf"};
UniformBuffer<WorldData> world_buf = {};
@ -243,11 +250,16 @@ class MeshPass : public PassMain {
ImageUser *iuser = nullptr);
};
enum class StencilBits : uint8_t {
BACKGROUND = 0,
OBJECT = 1u << 0,
OBJECT_IN_FRONT = 1u << 1,
};
class OpaquePass {
public:
TextureFromPool gbuffer_normal_tx = {"gbuffer_normal_tx"};
TextureFromPool gbuffer_material_tx = {"gbuffer_material_tx"};
Framebuffer opaque_fb = {};
Texture shadow_depth_stencil_tx = {"shadow_depth_stencil_tx"};
GPUTexture *deferred_ps_stencil_tx = nullptr;
@ -256,6 +268,11 @@ class OpaquePass {
MeshPass gbuffer_in_front_ps_ = {"Opaque.GbufferInFront"};
PassSimple deferred_ps_ = {"Opaque.Deferred"};
Framebuffer gbuffer_fb = {"Opaque.Gbuffer"};
Framebuffer gbuffer_in_front_fb = {"Opaque.GbufferInFront"};
Framebuffer deferred_fb = {"Opaque.Deferred"};
Framebuffer clear_fb = {"Opaque.Clear"};
void sync(const SceneState &scene_state, SceneResources &resources);
void draw(Manager &manager,
View &view,
@ -511,7 +528,6 @@ class AntiAliasingPass {
Texture sample0_depth_tx_ = {"sample0_depth_tx"};
Texture sample0_depth_in_front_tx_ = {"sample0_depth_in_front_tx"};
GPUTexture *stencil_tx_ = nullptr;
Texture taa_accumulation_tx_ = {"taa_accumulation_tx"};
Texture smaa_search_tx_ = {"smaa_search_tx"};
@ -545,15 +561,12 @@ class AntiAliasingPass {
~AntiAliasingPass();
void init(const SceneState &scene_state);
void sync(SceneResources &resources, int2 resolution);
void setup_view(View &view, int2 resolution);
void sync(const SceneState &scene_state, SceneResources &resources);
void setup_view(View &view, const SceneState &scene_state);
void draw(Manager &manager,
View &view,
SceneResources &resources,
int2 resolution,
GPUTexture *depth_tx,
GPUTexture *depth_in_front_tx,
GPUTexture *color_tx);
const SceneState &scene_state,
SceneResources &resources);
};
} // namespace blender::workbench

10
source/blender/draw/engines/workbench/workbench_shader_cache.cc
View File

@ -23,7 +23,9 @@ ShaderCache::~ShaderCache()
for (auto j : IndexRange(lighting_type_len)) {
for (auto k : IndexRange(2) /*cavity*/) {
for (auto l : IndexRange(2) /*curvature*/) {
DRW_SHADER_FREE_SAFE(resolve_shader_cache_[i][j][k][l]);
for (auto m : IndexRange(2) /*shadow*/) {
DRW_SHADER_FREE_SAFE(resolve_shader_cache_[i][j][k][l][m]);
}
}
}
}
@ -92,10 +94,11 @@ GPUShader *ShaderCache::prepass_shader_get(ePipelineType pipeline_type,
GPUShader *ShaderCache::resolve_shader_get(ePipelineType pipeline_type,
eLightingType lighting_type,
bool cavity,
bool curvature)
bool curvature,
bool shadow)
{
GPUShader *&shader_ptr =
resolve_shader_cache_[int(pipeline_type)][int(lighting_type)][cavity][curvature];
resolve_shader_cache_[int(pipeline_type)][int(lighting_type)][cavity][curvature][shadow];
if (shader_ptr != nullptr) {
return shader_ptr;
@ -125,6 +128,7 @@ GPUShader *ShaderCache::resolve_shader_get(ePipelineType pipeline_type,
}
info_name += cavity ? "_cavity" : "_no_cavity";
info_name += curvature ? "_curvature" : "_no_curvature";
info_name += shadow ? "_shadow" : "_no_shadow";
shader_ptr = GPU_shader_create_from_info_name(info_name.c_str());
return shader_ptr;

2
source/blender/draw/engines/workbench/workbench_state.cc
View File

@ -173,6 +173,8 @@ void SceneState::init(Object *camera_ob /*=nullptr*/)
shading.flag & V3D_SHADING_DEPTH_OF_FIELD;
draw_object_id = draw_outline || draw_curvature;
overlays_enabled = v3d && !(v3d->flag2 & V3D_HIDE_OVERLAYS);
};
static const CustomData *get_loop_custom_data(const Mesh *mesh)