chromium/third_party/skia/src/gpu/graphite/DrawOrder.h

/*
 * Copyright 2021 Google LLC
 *
 * Use of this source code is governed by a BSD-style license that can be
 * found in the LICENSE file.
 */

#ifndef skgpu_graphite_DrawOrder_DEFINED
#define skgpu_graphite_DrawOrder_DEFINED

#include "include/core/SkTypes.h"

namespace skgpu::graphite {

// Helper to encapsulate an unsigned number and enforce that it can only be used to create a
// monotonic sequence. The template argument 'Sequence' is used to define different sequences
// enforced by the compiler. For simplicity, and current needs within Graphite, it's assumed the
// entire sequence can be indexed by uint16_t.
template<typename Sequence>
class MonotonicValue { … };

/**
 * CompressedPaintersOrder is an ordinal number that allows draw commands to be re-ordered so long
 * as when they are executed, the read/writes to the color|depth attachments respect the original
 * painter's order. Logical draws with the same CompressedPaintersOrder can be assumed to be
 * executed in any order, however that may have been determined (e.g. BoundsManager or relying on
 * a depth test during rasterization).
 */
struct CompressedPaintersOrderSequence { … };
CompressedPaintersOrder;

/**
 * Each DisjointStencilIndex specifies an implicit set of non-overlapping draws. Assuming that two
 * draws have the same CompressedPaintersOrder and the same DisjointStencilIndex, their substeps
 * for multi-pass rendering (stencil-then-cover, etc.) can be intermingled with each other and
 * produce the same results as if each draw's substeps were executed in order before moving on to
 * the next draw's.
 *
 * Ordering within a set can be entirely arbitrary (i.e. all stencil steps can go before all cover
 * steps). Ordering between sets is also arbitrary since all draws share the same
 * CompressedPaintersOrder, so long as one set is entirely drawn before the next.
 *
 * Two draws that have different CompressedPaintersOrders but the same DisjointStencilIndex are
 * unrelated, they may or may not overlap. The painters order scopes the disjoint sets.
 */
struct DisjointStencilIndexSequence { … };
DisjointStencilIndex;

/**
 * Every draw has an associated depth value. The value is constant across the entire draw and is
 * not related to any varying Z coordinate induced by a 4x4 transform. The painter's depth is stored
 * in the depth attachment and the GREATER depth test is used to reject or accept pixels/samples
 * relative to what has already been rendered into the depth attachment. This allows draws that do
 * not depend on the previous color to be radically re-ordered relative to their original painter's
 * order while producing correct results.
 */
struct PaintersDepthSequence { … };
PaintersDepth;

/**
 * DrawOrder aggregates the three separate sequences that Graphite uses to re-order draws and their
 * substeps as much as possible while preserving the painter's order semantics of the Skia API.
 *
 * To build the full DrawOrder for a draw, start with its assigned PaintersDepth (i.e. the original
 * painter's order of the draw call). From there, the DrawOrder can be updated to reflect
 * dependencies on previous draws, either from depth-only clip draws or because the draw is
 * transparent and must blend with the previous color values. Lastly, once the
 * CompressedPaintersOrder is finalized, the DrawOrder can be updated to reflect whether or not
 * the draw will involve the stencil buffer--and if so, specify the disjoint stencil set it
 * belongs to.
 *
 * The original and effective order that draws are executed in is defined by the PaintersDepth.
 * However, the actual execution order is defined by first the CompressedPaintersOrder and then
 * the DisjointStencilIndex. This means that draws with much higher depths can be executed earlier
 * if painter's order compression allows for it.
 */
class DrawOrder { … };

} // namespace skgpu::graphite

#endif // skgpu_graphite_DrawOrder_DEFINED