blob: 6d6565e92da7d7a227613bb3c15672d505d2d4b5 [file] [log] [blame] [edit]
// Copyright 2013 The Flutter Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "impeller/aiks/experimental_canvas.h"
#include "fml/logging.h"
#include "impeller/aiks/canvas.h"
#include "impeller/aiks/paint_pass_delegate.h"
#include "impeller/base/validation.h"
#include "impeller/entity/contents/framebuffer_blend_contents.h"
#include "impeller/entity/contents/text_contents.h"
#include "impeller/entity/entity.h"
#include "impeller/entity/entity_pass_clip_stack.h"
#include "impeller/geometry/color.h"
#include "impeller/renderer/render_target.h"
namespace impeller {
namespace {
static void SetClipScissor(std::optional<Rect> clip_coverage,
RenderPass& pass,
Point global_pass_position) {
// Set the scissor to the clip coverage area. We do this prior to rendering
// the clip itself and all its contents.
IRect scissor;
if (clip_coverage.has_value()) {
clip_coverage = clip_coverage->Shift(-global_pass_position);
scissor = IRect::RoundOut(clip_coverage.value());
// The scissor rect must not exceed the size of the render target.
scissor = scissor.Intersection(IRect::MakeSize(pass.GetRenderTargetSize()))
.value_or(IRect());
}
pass.SetScissor(scissor);
}
static void ApplyFramebufferBlend(Entity& entity) {
auto src_contents = entity.GetContents();
auto contents = std::make_shared<FramebufferBlendContents>();
contents->SetChildContents(src_contents);
contents->SetBlendMode(entity.GetBlendMode());
entity.SetContents(std::move(contents));
entity.SetBlendMode(BlendMode::kSource);
}
} // namespace
static const constexpr RenderTarget::AttachmentConfig kDefaultStencilConfig =
RenderTarget::AttachmentConfig{
.storage_mode = StorageMode::kDeviceTransient,
.load_action = LoadAction::kDontCare,
.store_action = StoreAction::kDontCare,
};
static std::unique_ptr<EntityPassTarget> CreateRenderTarget(
ContentContext& renderer,
ISize size,
int mip_count,
const Color& clear_color) {
const std::shared_ptr<Context>& context = renderer.GetContext();
/// All of the load/store actions are managed by `InlinePassContext` when
/// `RenderPasses` are created, so we just set them to `kDontCare` here.
/// What's important is the `StorageMode` of the textures, which cannot be
/// changed for the lifetime of the textures.
if (context->GetBackendType() == Context::BackendType::kOpenGLES) {
// TODO(https://github.com/flutter/flutter/issues/141732): Implement mip map
// generation on opengles.
mip_count = 1;
}
RenderTarget target;
if (context->GetCapabilities()->SupportsOffscreenMSAA()) {
target = renderer.GetRenderTargetCache()->CreateOffscreenMSAA(
/*context=*/*context,
/*size=*/size,
/*mip_count=*/mip_count,
/*label=*/"EntityPass",
/*color_attachment_config=*/
RenderTarget::AttachmentConfigMSAA{
.storage_mode = StorageMode::kDeviceTransient,
.resolve_storage_mode = StorageMode::kDevicePrivate,
.load_action = LoadAction::kDontCare,
.store_action = StoreAction::kMultisampleResolve,
.clear_color = clear_color},
/*stencil_attachment_config=*/
kDefaultStencilConfig);
} else {
target = renderer.GetRenderTargetCache()->CreateOffscreen(
*context, // context
size, // size
/*mip_count=*/mip_count,
"EntityPass", // label
RenderTarget::AttachmentConfig{
.storage_mode = StorageMode::kDevicePrivate,
.load_action = LoadAction::kDontCare,
.store_action = StoreAction::kDontCare,
.clear_color = clear_color,
}, // color_attachment_config
kDefaultStencilConfig // stencil_attachment_config
);
}
return std::make_unique<EntityPassTarget>(
target, renderer.GetDeviceCapabilities().SupportsReadFromResolve(),
renderer.GetDeviceCapabilities().SupportsImplicitResolvingMSAA());
}
ExperimentalCanvas::ExperimentalCanvas(ContentContext& renderer,
RenderTarget& render_target,
bool requires_readback)
: Canvas(),
renderer_(renderer),
render_target_(render_target),
requires_readback_(requires_readback),
clip_coverage_stack_(EntityPassClipStack(
Rect::MakeSize(render_target.GetRenderTargetSize()))) {
SetupRenderPass();
}
ExperimentalCanvas::ExperimentalCanvas(ContentContext& renderer,
RenderTarget& render_target,
bool requires_readback,
Rect cull_rect)
: Canvas(cull_rect),
renderer_(renderer),
render_target_(render_target),
requires_readback_(requires_readback),
clip_coverage_stack_(EntityPassClipStack(
Rect::MakeSize(render_target.GetRenderTargetSize()))) {
SetupRenderPass();
}
ExperimentalCanvas::ExperimentalCanvas(ContentContext& renderer,
RenderTarget& render_target,
bool requires_readback,
IRect cull_rect)
: Canvas(cull_rect),
renderer_(renderer),
render_target_(render_target),
requires_readback_(requires_readback),
clip_coverage_stack_(EntityPassClipStack(
Rect::MakeSize(render_target.GetRenderTargetSize()))) {
SetupRenderPass();
}
void ExperimentalCanvas::SetupRenderPass() {
renderer_.GetRenderTargetCache()->Start();
auto color0 = render_target_.GetColorAttachments().find(0u)->second;
auto& stencil_attachment = render_target_.GetStencilAttachment();
auto& depth_attachment = render_target_.GetDepthAttachment();
if (!stencil_attachment.has_value() || !depth_attachment.has_value()) {
// Setup a new root stencil with an optimal configuration if one wasn't
// provided by the caller.
render_target_.SetupDepthStencilAttachments(
*renderer_.GetContext(),
*renderer_.GetContext()->GetResourceAllocator(),
color0.texture->GetSize(),
renderer_.GetContext()->GetCapabilities()->SupportsOffscreenMSAA(),
"ImpellerOnscreen", kDefaultStencilConfig);
}
// Set up the clear color of the root pass.
color0.clear_color = Color::BlackTransparent();
render_target_.SetColorAttachment(color0, 0);
// If requires_readback is true, then there is a backdrop filter or emulated
// advanced blend in the first save layer. This requires a readback, which
// isn't supported by onscreen textures. To support this, we immediately begin
// a second save layer with the same dimensions as the onscreen. When
// rendering is completed, we must blit this saveLayer to the onscreen.
if (requires_readback_) {
entity_pass_targets_.push_back(CreateRenderTarget(
renderer_, color0.texture->GetSize(), /*mip_count=*/1,
/*clear_color=*/Color::BlackTransparent()));
} else {
entity_pass_targets_.push_back(std::make_unique<EntityPassTarget>(
render_target_,
renderer_.GetDeviceCapabilities().SupportsReadFromResolve(),
renderer_.GetDeviceCapabilities().SupportsImplicitResolvingMSAA()));
}
auto inline_pass = std::make_unique<InlinePassContext>(
renderer_, *entity_pass_targets_.back(), 0);
inline_pass_contexts_.emplace_back(std::move(inline_pass));
auto result = inline_pass_contexts_.back()->GetRenderPass(0u);
render_passes_.push_back(result.pass);
}
void ExperimentalCanvas::Save(uint32_t total_content_depth) {
auto entry = CanvasStackEntry{};
entry.transform = transform_stack_.back().transform;
entry.cull_rect = transform_stack_.back().cull_rect;
entry.clip_depth = current_depth_ + total_content_depth;
entry.distributed_opacity = transform_stack_.back().distributed_opacity;
FML_CHECK(entry.clip_depth <= transform_stack_.back().clip_depth)
<< entry.clip_depth << " <=? " << transform_stack_.back().clip_depth
<< " after allocating " << total_content_depth;
entry.clip_height = transform_stack_.back().clip_height;
entry.rendering_mode = Entity::RenderingMode::kDirect;
transform_stack_.emplace_back(entry);
}
void ExperimentalCanvas::SaveLayer(
const Paint& paint,
std::optional<Rect> bounds,
const std::shared_ptr<ImageFilter>& backdrop_filter,
ContentBoundsPromise bounds_promise,
uint32_t total_content_depth,
bool can_distribute_opacity) {
if (can_distribute_opacity && !backdrop_filter &&
Paint::CanApplyOpacityPeephole(paint)) {
Save(total_content_depth);
transform_stack_.back().distributed_opacity *= paint.color.alpha;
return;
}
// Can we always guarantee that we get a bounds? Does a lack of bounds
// indicate something?
if (!bounds.has_value()) {
bounds = Rect::MakeSize(render_target_.GetRenderTargetSize());
}
// When applying a save layer, absorb any pending distributed opacity.
Paint paint_copy = paint;
paint_copy.color.alpha *= transform_stack_.back().distributed_opacity;
transform_stack_.back().distributed_opacity = 1.0;
Rect subpass_coverage = bounds->TransformBounds(GetCurrentTransform());
auto target =
CreateRenderTarget(renderer_,
ISize::MakeWH(subpass_coverage.GetSize().width,
subpass_coverage.GetSize().height),
1u, Color::BlackTransparent());
entity_pass_targets_.push_back(std::move(target));
save_layer_state_.push_back(SaveLayerState{paint_copy, subpass_coverage});
CanvasStackEntry entry;
entry.transform = transform_stack_.back().transform;
entry.cull_rect = transform_stack_.back().cull_rect;
entry.clip_depth = current_depth_ + total_content_depth;
FML_CHECK(entry.clip_depth <= transform_stack_.back().clip_depth)
<< entry.clip_depth << " <=? " << transform_stack_.back().clip_depth
<< " after allocating " << total_content_depth;
entry.clip_height = transform_stack_.back().clip_height;
entry.rendering_mode = Entity::RenderingMode::kSubpassAppendSnapshotTransform;
transform_stack_.emplace_back(entry);
auto inline_pass = std::make_unique<InlinePassContext>(
renderer_, *entity_pass_targets_.back(), 0);
inline_pass_contexts_.emplace_back(std::move(inline_pass));
auto result = inline_pass_contexts_.back()->GetRenderPass(0u);
render_passes_.push_back(result.pass);
// Start non-collapsed subpasses with a fresh clip coverage stack limited by
// the subpass coverage. This is important because image filters applied to
// save layers may transform the subpass texture after it's rendered,
// causing parent clip coverage to get misaligned with the actual area that
// the subpass will affect in the parent pass.
clip_coverage_stack_.PushSubpass(subpass_coverage, GetClipHeight());
}
bool ExperimentalCanvas::Restore() {
FML_DCHECK(transform_stack_.size() > 0);
if (transform_stack_.size() == 1) {
return false;
}
// This check is important to make sure we didn't exceed the depth
// that the clips were rendered at while rendering any of the
// rendering ops. It is OK for the current depth to equal the
// outgoing clip depth because that means the clipping would have
// been successful up through the last rendering op, but it cannot
// be greater.
// Also, we bump the current rendering depth to the outgoing clip
// depth so that future rendering operations are not clipped by
// any of the pixels set by the expiring clips. It is OK for the
// estimates used to determine the clip depth in save/saveLayer
// to be overly conservative, but we need to jump the depth to
// the clip depth so that the next rendering op will get a
// larger depth (it will pre-increment the current_depth_ value).
FML_CHECK(current_depth_ <= transform_stack_.back().clip_depth)
<< current_depth_ << " <=? " << transform_stack_.back().clip_depth;
current_depth_ = transform_stack_.back().clip_depth;
if (transform_stack_.back().rendering_mode ==
Entity::RenderingMode::kSubpassAppendSnapshotTransform ||
transform_stack_.back().rendering_mode ==
Entity::RenderingMode::kSubpassPrependSnapshotTransform) {
auto inline_pass = std::move(inline_pass_contexts_.back());
SaveLayerState save_layer_state = save_layer_state_.back();
save_layer_state_.pop_back();
std::shared_ptr<Contents> contents =
PaintPassDelegate(save_layer_state.paint)
.CreateContentsForSubpassTarget(inline_pass->GetTexture(),
transform_stack_.back().transform);
inline_pass->EndPass();
render_passes_.pop_back();
inline_pass_contexts_.pop_back();
Entity element_entity;
element_entity.SetClipDepth(++current_depth_);
element_entity.SetContents(std::move(contents));
element_entity.SetBlendMode(save_layer_state.paint.blend_mode);
element_entity.SetTransform(Matrix::MakeTranslation(
Vector3(save_layer_state.coverage.GetOrigin())));
if (element_entity.GetBlendMode() > Entity::kLastPipelineBlendMode) {
if (renderer_.GetDeviceCapabilities().SupportsFramebufferFetch()) {
ApplyFramebufferBlend(element_entity);
} else {
VALIDATION_LOG << "Emulated advanced blends are currently unsupported.";
element_entity.SetBlendMode(BlendMode::kSourceOver);
}
}
element_entity.Render(renderer_, *render_passes_.back());
clip_coverage_stack_.PopSubpass();
transform_stack_.pop_back();
// We don't need to restore clips if a saveLayer was performed, as the clip
// state is per render target, and no more rendering operations will be
// performed as the render target workloaded is completed in the restore.
return true;
}
size_t num_clips = transform_stack_.back().num_clips;
transform_stack_.pop_back();
if (num_clips > 0) {
Entity entity;
entity.SetTransform(
Matrix::MakeTranslation(Vector3(-GetGlobalPassPosition())) *
GetCurrentTransform());
// This path is empty because ClipRestoreContents just generates a quad that
// takes up the full render target.
auto clip_restore = std::make_shared<ClipRestoreContents>();
clip_restore->SetRestoreHeight(GetClipHeight());
entity.SetContents(std::move(clip_restore));
auto current_clip_coverage = clip_coverage_stack_.CurrentClipCoverage();
if (current_clip_coverage.has_value()) {
// Entity transforms are relative to the current pass position, so we need
// to check clip coverage in the same space.
current_clip_coverage =
current_clip_coverage->Shift(-GetGlobalPassPosition());
}
auto clip_coverage = entity.GetClipCoverage(current_clip_coverage);
if (clip_coverage.coverage.has_value()) {
clip_coverage.coverage =
clip_coverage.coverage->Shift(GetGlobalPassPosition());
}
EntityPassClipStack::ClipStateResult clip_state_result =
clip_coverage_stack_.ApplyClipState(clip_coverage, entity,
GetClipHeightFloor(),
GetGlobalPassPosition());
if (clip_state_result.clip_did_change) {
// We only need to update the pass scissor if the clip state has changed.
SetClipScissor(clip_coverage_stack_.CurrentClipCoverage(),
*render_passes_.back(), GetGlobalPassPosition());
}
if (!clip_state_result.should_render) {
return true;
}
entity.Render(renderer_, *render_passes_.back());
}
return true;
}
void ExperimentalCanvas::DrawTextFrame(
const std::shared_ptr<TextFrame>& text_frame,
Point position,
const Paint& paint) {
Entity entity;
entity.SetClipDepth(GetClipHeight());
entity.SetBlendMode(paint.blend_mode);
auto text_contents = std::make_shared<TextContents>();
text_contents->SetTextFrame(text_frame);
text_contents->SetForceTextColor(paint.mask_blur_descriptor.has_value());
text_contents->SetScale(GetCurrentTransform().GetMaxBasisLengthXY());
text_contents->SetColor(paint.color);
text_contents->SetOffset(position);
text_contents->SetTextProperties(paint.color, //
paint.style == Paint::Style::kStroke, //
paint.stroke_width, //
paint.stroke_cap, //
paint.stroke_join, //
paint.stroke_miter //
);
entity.SetTransform(GetCurrentTransform() *
Matrix::MakeTranslation(position));
// TODO(bdero): This mask blur application is a hack. It will always wind up
// doing a gaussian blur that affects the color source itself
// instead of just the mask. The color filter text support
// needs to be reworked in order to interact correctly with
// mask filters.
// https://github.com/flutter/flutter/issues/133297
entity.SetContents(paint.WithFilters(paint.WithMaskBlur(
std::move(text_contents), true, GetCurrentTransform())));
AddRenderEntityToCurrentPass(std::move(entity), false);
}
void ExperimentalCanvas::AddRenderEntityToCurrentPass(Entity entity,
bool reuse_depth) {
entity.SetInheritedOpacity(transform_stack_.back().distributed_opacity);
auto transform = entity.GetTransform();
entity.SetTransform(
Matrix::MakeTranslation(Vector3(-GetGlobalPassPosition())) * transform);
if (!reuse_depth) {
++current_depth_;
}
// We can render at a depth up to and including the depth of the currently
// active clips and we will still be clipped out, but we cannot render at
// a depth that is greater than the current clips or we will not be clipped.
FML_CHECK(current_depth_ <= transform_stack_.back().clip_depth)
<< current_depth_ << " <=? " << transform_stack_.back().clip_depth;
entity.SetClipDepth(current_depth_);
if (entity.GetBlendMode() > Entity::kLastPipelineBlendMode) {
if (renderer_.GetDeviceCapabilities().SupportsFramebufferFetch()) {
ApplyFramebufferBlend(entity);
} else {
VALIDATION_LOG << "Emulated advanced blends are currently unsupported.";
return;
}
}
entity.Render(renderer_, *render_passes_.back());
}
void ExperimentalCanvas::AddClipEntityToCurrentPass(Entity entity) {
auto transform = entity.GetTransform();
entity.SetTransform(
Matrix::MakeTranslation(Vector3(-GetGlobalPassPosition())) * transform);
// Ideally the clip depth would be greater than the current rendering
// depth because any rendering calls that follow this clip operation will
// pre-increment the depth and then be rendering above our clip depth,
// but that case will be caught by the CHECK in AddRenderEntity above.
// In practice we sometimes have a clip set with no rendering after it
// and in such cases the current depth will equal the clip depth.
// Eventually the DisplayList should optimize these out, but it is hard
// to know if a clip will actually be used in advance of storing it in
// the DisplayList buffer.
// See https://github.com/flutter/flutter/issues/147021
FML_CHECK(current_depth_ <= transform_stack_.back().clip_depth)
<< current_depth_ << " <=? " << transform_stack_.back().clip_depth;
entity.SetClipDepth(transform_stack_.back().clip_depth);
auto current_clip_coverage = clip_coverage_stack_.CurrentClipCoverage();
if (current_clip_coverage.has_value()) {
// Entity transforms are relative to the current pass position, so we need
// to check clip coverage in the same space.
current_clip_coverage =
current_clip_coverage->Shift(-GetGlobalPassPosition());
}
auto clip_coverage = entity.GetClipCoverage(current_clip_coverage);
if (clip_coverage.coverage.has_value()) {
clip_coverage.coverage =
clip_coverage.coverage->Shift(GetGlobalPassPosition());
}
EntityPassClipStack::ClipStateResult clip_state_result =
clip_coverage_stack_.ApplyClipState(
clip_coverage, entity, GetClipHeightFloor(), GetGlobalPassPosition());
if (clip_state_result.clip_did_change) {
// We only need to update the pass scissor if the clip state has changed.
SetClipScissor(clip_coverage_stack_.CurrentClipCoverage(),
*render_passes_.back(), GetGlobalPassPosition());
}
if (!clip_state_result.should_render) {
return;
}
entity.Render(renderer_, *render_passes_.back());
}
bool ExperimentalCanvas::BlitToOnscreen() {
auto command_buffer = renderer_.GetContext()->CreateCommandBuffer();
command_buffer->SetLabel("EntityPass Root Command Buffer");
auto offscreen_target = entity_pass_targets_.back()->GetRenderTarget();
if (renderer_.GetContext()
->GetCapabilities()
->SupportsTextureToTextureBlits()) {
auto blit_pass = command_buffer->CreateBlitPass();
blit_pass->AddCopy(offscreen_target.GetRenderTargetTexture(),
render_target_.GetRenderTargetTexture());
if (!blit_pass->EncodeCommands(
renderer_.GetContext()->GetResourceAllocator())) {
VALIDATION_LOG << "Failed to encode root pass blit command.";
return false;
}
if (!renderer_.GetContext()
->GetCommandQueue()
->Submit({command_buffer})
.ok()) {
return false;
}
} else {
auto render_pass = command_buffer->CreateRenderPass(render_target_);
render_pass->SetLabel("EntityPass Root Render Pass");
{
auto size_rect = Rect::MakeSize(offscreen_target.GetRenderTargetSize());
auto contents = TextureContents::MakeRect(size_rect);
contents->SetTexture(offscreen_target.GetRenderTargetTexture());
contents->SetSourceRect(size_rect);
contents->SetLabel("Root pass blit");
Entity entity;
entity.SetContents(contents);
entity.SetBlendMode(BlendMode::kSource);
if (!entity.Render(renderer_, *render_pass)) {
VALIDATION_LOG << "Failed to render EntityPass root blit.";
return false;
}
}
if (!render_pass->EncodeCommands()) {
VALIDATION_LOG << "Failed to encode root pass command buffer.";
return false;
}
if (!renderer_.GetContext()
->GetCommandQueue()
->Submit({command_buffer})
.ok()) {
return false;
}
}
return true;
}
void ExperimentalCanvas::EndReplay() {
FML_DCHECK(inline_pass_contexts_.size() == 1u);
inline_pass_contexts_.back()->EndPass();
// If requires_readback_ was true, then we rendered to an offscreen texture
// instead of to the onscreen provided in the render target. Now we need to
// draw or blit the offscreen back to the onscreen.
if (requires_readback_) {
BlitToOnscreen();
}
render_passes_.clear();
inline_pass_contexts_.clear();
renderer_.GetRenderTargetCache()->End();
Reset();
Initialize(initial_cull_rect_);
}
} // namespace impeller