Draw Operation merging
Merge simple bitmap draw operations and text operations to avoid
issuing individual gl draws for each operation. Merging other ops to
be done eventually.
The methods are different - the bitmap merging generates a single
mesh for reused, unclipped images (esp. repeated images in a listview)
The text approach queries just defers the normal font rendering until
the last drawText in the sequence that can share the same shader.
Patches are sorted and merged, but don't yet have a multiDraw
implementation. For now, the pretending-to-merge gives better sorting
behavior by keeping similar patches together.
Change-Id: Ic300cdab0a53814cf7b09c58bf54b1bf0f58ccd6
diff --git a/libs/hwui/DeferredDisplayList.cpp b/libs/hwui/DeferredDisplayList.cpp
index d5007e1..f0084f2 100644
--- a/libs/hwui/DeferredDisplayList.cpp
+++ b/libs/hwui/DeferredDisplayList.cpp
@@ -23,6 +23,7 @@
#include "Caches.h"
#include "Debug.h"
+#include "DeferredDisplayList.h"
#include "DisplayListOp.h"
#include "OpenGLRenderer.h"
@@ -38,15 +39,27 @@
// Depth of the save stack at the beginning of batch playback at flush time
#define FLUSH_SAVE_STACK_DEPTH 2
+#define DEBUG_COLOR_BARRIER 0x1f000000
+#define DEBUG_COLOR_MERGEDBATCH 0x5f7f7fff
+#define DEBUG_COLOR_MERGEDBATCH_SOLO 0x5f7fff7f
+
/////////////////////////////////////////////////////////////////////////////////
// Operation Batches
/////////////////////////////////////////////////////////////////////////////////
-class DrawOpBatch {
+class Batch {
public:
- DrawOpBatch() { mOps.clear(); }
+ virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty, int index) = 0;
+ virtual ~Batch() {}
+};
- virtual ~DrawOpBatch() { mOps.clear(); }
+class DrawBatch : public Batch {
+public:
+ DrawBatch(int batchId, mergeid_t mergeId) : mBatchId(batchId), mMergeId(mergeId) {
+ mOps.clear();
+ }
+
+ virtual ~DrawBatch() { mOps.clear(); }
void add(DrawOp* op) {
// NOTE: ignore empty bounds special case, since we don't merge across those ops
@@ -54,7 +67,7 @@
mOps.add(op);
}
- virtual bool intersects(Rect& rect) {
+ bool intersects(Rect& rect) {
if (!rect.intersects(mBounds)) return false;
for (unsigned int i = 0; i < mOps.size(); i++) {
@@ -71,8 +84,9 @@
return false;
}
- virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty) {
- DEFER_LOGD("replaying draw batch %p", this);
+ virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty, int index) {
+ DEFER_LOGD("%d replaying DrawingBatch %p, with %d ops (batch id %x, merge id %p)",
+ index, this, mOps.size(), mOps[0]->getBatchId(), mOps[0]->getMergeId());
status_t status = DrawGlInfo::kStatusDone;
DisplayListLogBuffer& logBuffer = DisplayListLogBuffer::getInstance();
@@ -84,31 +98,127 @@
#if DEBUG_DISPLAY_LIST_OPS_AS_EVENTS
renderer.eventMark(op->name());
#endif
- status |= op->applyDraw(renderer, dirty, 0);
+ status |= op->applyDraw(renderer, dirty);
logBuffer.writeCommand(0, op->name());
+
+#if DEBUG_MERGE_BEHAVIOR
+ Rect& bounds = mOps[i]->state.mBounds;
+ int batchColor = 0x1f000000;
+ if (getBatchId() & 0x1) batchColor |= 0x0000ff;
+ if (getBatchId() & 0x2) batchColor |= 0x00ff00;
+ if (getBatchId() & 0x4) batchColor |= 0xff0000;
+ renderer.drawScreenSpaceColorRect(bounds.left, bounds.top, bounds.right, bounds.bottom,
+ batchColor);
+#endif
}
return status;
}
+ inline int getBatchId() const { return mBatchId; }
+ inline mergeid_t getMergeId() const { return mMergeId; }
inline int count() const { return mOps.size(); }
-private:
+
+protected:
Vector<DrawOp*> mOps;
Rect mBounds;
+private:
+ int mBatchId;
+ mergeid_t mMergeId;
};
-class StateOpBatch : public DrawOpBatch {
+// compare alphas approximately, with a small margin
+#define NEQ_FALPHA(lhs, rhs) \
+ fabs((float)lhs - (float)rhs) > 0.001f
+
+class MergingDrawBatch : public DrawBatch {
+public:
+ MergingDrawBatch(int batchId, mergeid_t mergeId) : DrawBatch(batchId, mergeId) {}
+
+ /*
+ * Checks if a (mergeable) op can be merged into this batch
+ *
+ * If true, the op's multiDraw must be guaranteed to handle both ops simultaneously, so it is
+ * important to consider all paint attributes used in the draw calls in deciding both a) if an
+ * op tries to merge at all, and b) if the op
+ *
+ * False positives can lead to information from the paints of subsequent merged operations being
+ * dropped, so we make simplifying qualifications on the ops that can merge, per op type.
+ */
+ bool canMergeWith(DrawOp* op) {
+ if (!op->state.mMatrix.isPureTranslate()) return false;
+
+ bool isTextBatch = getBatchId() == DeferredDisplayList::kOpBatch_Text ||
+ getBatchId() == DeferredDisplayList::kOpBatch_ColorText;
+
+ // Overlapping other operations is only allowed for text without shadow. For other ops,
+ // multiDraw isn't guaranteed to overdraw correctly
+ if (!isTextBatch || op->state.mDrawModifiers.mHasShadow) {
+ if (intersects(op->state.mBounds)) return false;
+ }
+
+ const DeferredDisplayState& lhs = op->state;
+ const DeferredDisplayState& rhs = mOps[0]->state;
+
+ if (NEQ_FALPHA(lhs.mAlpha, rhs.mAlpha)) return false;
+
+ // if paints are equal, then modifiers + paint attribs don't need to be compared
+ if (op->mPaint == mOps[0]->mPaint) return true;
+
+ if (op->getPaintAlpha() != mOps[0]->getPaintAlpha()) return false;
+
+ /* Draw Modifiers compatibility check
+ *
+ * Shadows are ignored, as only text uses them, and in that case they are drawn
+ * per-DrawTextOp, before the unified text draw. Because of this, it's always safe to merge
+ * text UNLESS a later draw's shadow should overlays a previous draw's text. This is covered
+ * above with the intersection check.
+ *
+ * OverrideLayerAlpha is also ignored, as it's only used for drawing layers, which are never
+ * merged.
+ *
+ * These ignore cases prevent us from simply memcmp'ing the drawModifiers
+ */
+
+ const DrawModifiers& lhsMod = lhs.mDrawModifiers;
+ const DrawModifiers& rhsMod = rhs.mDrawModifiers;
+ if (lhsMod.mShader != rhsMod.mShader) return false;
+ if (lhsMod.mColorFilter != rhsMod.mColorFilter) return false;
+
+ // Draw filter testing expects bit fields to be clear if filter not set.
+ if (lhsMod.mHasDrawFilter != rhsMod.mHasDrawFilter) return false;
+ if (lhsMod.mPaintFilterClearBits != rhsMod.mPaintFilterClearBits) return false;
+ if (lhsMod.mPaintFilterSetBits != rhsMod.mPaintFilterSetBits) return false;
+
+ return true;
+ }
+
+ virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty, int index) {
+ DEFER_LOGD("%d replaying DrawingBatch %p, with %d ops (batch id %x, merge id %p)",
+ index, this, mOps.size(), getBatchId(), getMergeId());
+ if (mOps.size() == 1) {
+ return DrawBatch::replay(renderer, dirty, false);
+ }
+
+ DrawOp* op = mOps[0];
+ status_t status = op->multiDraw(renderer, dirty, mOps, mBounds);
+ DisplayListLogBuffer& buffer = DisplayListLogBuffer::getInstance();
+ buffer.writeCommand(0, "multiDraw");
+ buffer.writeCommand(1, op->name());
+
+#if DEBUG_MERGE_BEHAVIOR
+ renderer.drawScreenSpaceColorRect(mBounds.left, mBounds.top, mBounds.right, mBounds.bottom,
+ DEBUG_COLOR_MERGEDBATCH);
+#endif
+ return status;
+ }
+};
+
+class StateOpBatch : public Batch {
public:
// creates a single operation batch
StateOpBatch(StateOp* op) : mOp(op) {}
- bool intersects(Rect& rect) {
- // if something checks for intersection, it's trying to go backwards across a state op,
- // something not currently supported - state ops are always barriers
- CRASH();
- return false;
- }
-
- virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty) {
+ virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty, int index) {
DEFER_LOGD("replaying state op batch %p", this);
renderer.restoreDisplayState(mOp->state);
@@ -124,18 +234,11 @@
const StateOp* mOp;
};
-class RestoreToCountBatch : public DrawOpBatch {
+class RestoreToCountBatch : public Batch {
public:
RestoreToCountBatch(StateOp* op, int restoreCount) : mOp(op), mRestoreCount(restoreCount) {}
- bool intersects(Rect& rect) {
- // if something checks for intersection, it's trying to go backwards across a state op,
- // something not currently supported - state ops are always barriers
- CRASH();
- return false;
- }
-
- virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty) {
+ virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty, int index) {
DEFER_LOGD("batch %p restoring to count %d", this, mRestoreCount);
renderer.restoreDisplayState(mOp->state);
@@ -155,14 +258,30 @@
const int mRestoreCount;
};
+#if DEBUG_MERGE_BEHAVIOR
+class BarrierDebugBatch : public Batch {
+ virtual status_t replay(OpenGLRenderer& renderer, Rect& dirty, int index) {
+ renderer.drawScreenSpaceColorRect(0, 0, 10000, 10000, DEBUG_COLOR_BARRIER);
+ return DrawGlInfo::kStatusDrew;
+ }
+};
+#endif
+
/////////////////////////////////////////////////////////////////////////////////
// DeferredDisplayList
/////////////////////////////////////////////////////////////////////////////////
void DeferredDisplayList::resetBatchingState() {
for (int i = 0; i < kOpBatch_Count; i++) {
- mBatchIndices[i] = -1;
+ mBatchLookup[i] = NULL;
+ mMergingBatches[i].clear();
}
+#if DEBUG_MERGE_BEHAVIOR
+ if (mBatches.size() != 0) {
+ mBatches.add(new BarrierDebugBatch());
+ }
+#endif
+ mEarliestBatchIndex = mBatches.size();
}
void DeferredDisplayList::clear() {
@@ -174,6 +293,7 @@
}
mBatches.clear();
mSaveStack.clear();
+ mEarliestBatchIndex = 0;
}
/////////////////////////////////////////////////////////////////////////////////
@@ -282,28 +402,35 @@
return; // quick rejected
}
- op->onDrawOpDeferred(renderer);
+ int batchId = kOpBatch_None;
+ mergeid_t mergeId = (mergeid_t) -1;
+ bool mergeable = op->onDefer(renderer, &batchId, &mergeId);
+
+ // complex clip has a complex set of expectations on the renderer state - for now, avoid taking
+ // the merge path in those cases
+ mergeable &= !recordingComplexClip();
if (CC_UNLIKELY(renderer.getCaches().drawReorderDisabled)) {
// TODO: elegant way to reuse batches?
- DrawOpBatch* b = new DrawOpBatch();
+ DrawBatch* b = new DrawBatch(batchId, mergeId);
b->add(op);
mBatches.add(b);
return;
}
- // disallowReorder isn't set, so find the latest batch of the new op's type, and try to merge
- // the new op into it
- DrawOpBatch* targetBatch = NULL;
- int batchId = op->getBatchId();
+ // find the latest batch of the new op's type, and try to merge the new op into it
+ DrawBatch* targetBatch = NULL;
+ // insertion point of a new batch, will hopefully be immediately after similar batch
+ // (eventually, should be similar shader)
+ int insertBatchIndex = mBatches.size();
if (!mBatches.isEmpty()) {
if (op->state.mBounds.isEmpty()) {
// don't know the bounds for op, so add to last batch and start from scratch on next op
- mBatches.top()->add(op);
- for (int i = 0; i < kOpBatch_Count; i++) {
- mBatchIndices[i] = -1;
- }
+ DrawBatch* b = new DrawBatch(batchId, mergeId);
+ b->add(op);
+ mBatches.add(b);
+ resetBatchingState();
#if DEBUG_DEFER
DEFER_LOGD("Warning: Encountered op with empty bounds, resetting batches");
op->output(2);
@@ -311,13 +438,36 @@
return;
}
- if (batchId >= 0 && mBatchIndices[batchId] != -1) {
- int targetIndex = mBatchIndices[batchId];
- targetBatch = mBatches[targetIndex];
+ if (mergeable) {
+ // Try to merge with any existing batch with same mergeId.
+ if (mMergingBatches[batchId].get(mergeId, targetBatch)) {
+ if (!((MergingDrawBatch*) targetBatch)->canMergeWith(op)) {
+ targetBatch = NULL;
+ }
+ }
+ } else {
+ // join with similar, non-merging batch
+ targetBatch = (DrawBatch*)mBatchLookup[batchId];
+ }
+
+ if (targetBatch || mergeable) {
// iterate back toward target to see if anything drawn since should overlap the new op
- for (int i = mBatches.size() - 1; i > targetIndex; i--) {
- DrawOpBatch* overBatch = mBatches[i];
+ // if no target, merging ops still interate to find similar batch to insert after
+ for (int i = mBatches.size() - 1; i >= mEarliestBatchIndex; i--) {
+ DrawBatch* overBatch = (DrawBatch*)mBatches[i];
+
+ if (overBatch == targetBatch) break;
+
+ // TODO: also consider shader shared between batch types
+ if (batchId == overBatch->getBatchId()) {
+ insertBatchIndex = i + 1;
+ if (!targetBatch) break; // found insert position, quit
+ }
+
if (overBatch->intersects(op->state.mBounds)) {
+ // NOTE: it may be possible to optimize for special cases where two operations
+ // of the same batch/paint could swap order, such as with a non-mergeable
+ // (clipped) and a mergeable text operation
targetBatch = NULL;
#if DEBUG_DEFER
DEFER_LOGD("op couldn't join batch %d, was intersected by batch %d",
@@ -329,13 +479,21 @@
}
}
}
+
if (!targetBatch) {
- targetBatch = new DrawOpBatch();
- mBatches.add(targetBatch);
- if (batchId >= 0) {
- mBatchIndices[batchId] = mBatches.size() - 1;
+ if (mergeable) {
+ targetBatch = new MergingDrawBatch(batchId, mergeId);
+ mMergingBatches[batchId].put(mergeId, targetBatch);
+ } else {
+ targetBatch = new DrawBatch(batchId, mergeId);
+ mBatchLookup[batchId] = targetBatch;
+ DEFER_LOGD("creating Batch %p, bid %x, at %d",
+ targetBatch, batchId, insertBatchIndex);
}
+
+ mBatches.insertAt(targetBatch, insertBatchIndex);
}
+
targetBatch->add(op);
}
@@ -363,16 +521,14 @@
// Replay / flush
/////////////////////////////////////////////////////////////////////////////////
-static status_t replayBatchList(Vector<DrawOpBatch*>& batchList,
+static status_t replayBatchList(const Vector<Batch*>& batchList,
OpenGLRenderer& renderer, Rect& dirty) {
status_t status = DrawGlInfo::kStatusDone;
- int opCount = 0;
for (unsigned int i = 0; i < batchList.size(); i++) {
- status |= batchList[i]->replay(renderer, dirty);
- opCount += batchList[i]->count();
+ status |= batchList[i]->replay(renderer, dirty, i);
}
- DEFER_LOGD("--flushed, drew %d batches (total %d ops)", batchList.size(), opCount);
+ DEFER_LOGD("--flushed, drew %d batches", batchList.size());
return status;
}
@@ -400,7 +556,6 @@
renderer.setDrawModifiers(restoreDrawModifiers);
DEFER_LOGD("--flush complete, returning %x", status);
-
clear();
return status;
}