Begin instanced rendering for simple shapes

src/gpu/instanced/GLInstancedRendering.cpp

Index: src/gpu/instanced/GLInstancedRendering.cpp
diff --git a/src/gpu/instanced/GLInstancedRendering.cpp b/src/gpu/instanced/GLInstancedRendering.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..1cb640cd5c2bb82793af9afe3004d645e0317d78
--- /dev/null
+++ b/src/gpu/instanced/GLInstancedRendering.cpp
@@ -0,0 +1,303 @@
+/*
+ * Copyright 2016 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "GLInstancedRendering.h"
+
+#include "GrResourceProvider.h"
+#include "gl/GrGLGpu.h"
+#include "instanced/InstanceProcessor.h"
+
+#define GL_CALL(X) GR_GL_CALL(this->glGpu()->glInterface(), X)
+
+namespace gr_instanced {
+
+class GLInstancedRendering::GLBatch : public InstancedRendering::Batch {
+public:
+    DEFINE_BATCH_CLASS_ID
+
+    GLBatch(GLInstancedRendering* instRendering, int instanceIdx)
+        : INHERITED(ClassID(), instRendering, instanceIdx) {
+    }
+
+private:
+    int fEmulatedBaseInstance;
+    int fGLDrawCmdsIdx;
+    int fNumGLDrawCmds;
+
+    friend class GLInstancedRendering;
+
+    typedef Batch INHERITED;
+};
+
+GLInstancedRendering* GLInstancedRendering::CreateIfSupported(GrGLGpu* gpu) {
+    const GrGLCaps& caps = gpu->glCaps();
+    AntialiasMode lastSupportedAAMode;
+    if (!caps.vertexArrayObjectSupport() ||
+        !caps.drawIndirectSupport() ||
+        !InstanceProcessor::IsSupported(*caps.glslCaps(), caps, &lastSupportedAAMode)) {
+        return nullptr;
+    }
+    return new GLInstancedRendering(gpu, lastSupportedAAMode);
+}
+
+GLInstancedRendering::GLInstancedRendering(GrGLGpu* gpu, AntialiasMode lastSupportedAAMode)
+    : INHERITED(gpu, lastSupportedAAMode),
+      fVertexArrayID(0),
+      fInstanceAttribsBufferUniqueId(SK_InvalidUniqueID) {
+}
+
+GLInstancedRendering::~GLInstancedRendering() {
+    if (fVertexArrayID) {
+        GL_CALL(DeleteVertexArrays(1, &fVertexArrayID));
+        this->glGpu()->notifyVertexArrayDelete(fVertexArrayID);
+    }
+}
+
+inline GrGLGpu* GLInstancedRendering::glGpu() const {
+    return static_cast<GrGLGpu*>(this->gpu());
+}
+
+InstancedRendering::Batch* GLInstancedRendering::createBatch(int instanceIdx) {
+    return new GLBatch(this, instanceIdx);
+}
+
+void GLInstancedRendering::onBeginFlush(GrResourceProvider* rp) {
+    // Count what there is to draw.
+    BatchList::Iter iter;
+    iter.init(this->batchList(), BatchList::Iter::kHead_IterStart);
+    int numGLInstances = 0;
+    int numGLDrawCmds = 0;
+    while (Batch* b = iter.get()) {
+        GLBatch* batch = static_cast<GLBatch*>(b);
+        iter.next();
+
+        const auto& drawCmds = batch->fDrawCmds;
+        for (int c = 0; c < drawCmds.count(); ++c) {
+            numGLInstances += drawCmds[c].fInstanceRange.fCount;
+            if (c > 0 && drawCmds[c].fGeometry == drawCmds[c - 1].fGeometry) {
+                // When two adjacent draw commands have the same geometry, we will rearrange the
+                // instances for the GL buffer and combine them into a single command.
+                continue;
+            }
+            ++numGLDrawCmds;
+        }
+    }
+    if (!numGLDrawCmds) {
+        return;
+    }
+    SkASSERT(numGLInstances);
+
+    // Lazily create a vertex array object.
+    if (!fVertexArrayID) {
+        GL_CALL(GenVertexArrays(1, &fVertexArrayID));
+        if (!fVertexArrayID) {
+            return;
+        }
+        this->glGpu()->bindVertexArray(fVertexArrayID);
+
+        // Attach our index buffer to the vertex array.
+        GL_CALL(BindBuffer(GR_GL_ELEMENT_ARRAY_BUFFER,
+                           static_cast<const GrGLBuffer*>(this->indexBuffer())->bufferID()));
+
+        // Set up the non-instanced attribs.
+        this->glGpu()->bindBuffer(kVertex_GrBufferType,
+                                  static_cast<const GrGLBuffer*>(this->vertexBuffer()));
+        GL_CALL(EnableVertexAttribArray(kShapeCoords_AttribIdx));
+        GL_CALL(VertexAttribPointer(kShapeCoords_AttribIdx, 2, GR_GL_FLOAT, GR_GL_FALSE,
+                                    sizeof(ShapeVertex), (void*) offsetof(ShapeVertex, fX)));
+        GL_CALL(EnableVertexAttribArray(kVertexAttrs_AttribIdx));
+        GL_CALL(VertexAttribIPointer(kVertexAttrs_AttribIdx, 1, GR_GL_INT, sizeof(ShapeVertex),
+                                     (void*) offsetof(ShapeVertex, fAttrs)));
+
+        SkASSERT(SK_InvalidUniqueID == fInstanceAttribsBufferUniqueId);
+    }
+
+    // Create and map instance and draw-indirect buffers.
+    SkASSERT(!fInstanceBuffer);
+    fInstanceBuffer.reset(static_cast<GrGLBuffer*>(
+        rp->createBuffer(sizeof(Instance) * numGLInstances, kVertex_GrBufferType,
+                         kDynamic_GrAccessPattern, GrResourceProvider::kNoPendingIO_Flag)));
+    if (!fInstanceBuffer) {
+        return;
+    }
+
+    SkASSERT(!fDrawIndirectBuffer);
+    fDrawIndirectBuffer.reset(static_cast<GrGLBuffer*>(
+        rp->createBuffer(sizeof(GrGLDrawElementsIndirectCommand) * numGLDrawCmds,
+                         kDrawIndirect_GrBufferType, kDynamic_GrAccessPattern,
+                         GrResourceProvider::kNoPendingIO_Flag)));
+    if (!fDrawIndirectBuffer) {
+        return;
+    }
+
+    Instance* glMappedInstances = static_cast<Instance*>(fInstanceBuffer->map());
+    int glInstancesIdx = 0;
+
+    auto* glMappedCmds = static_cast<GrGLDrawElementsIndirectCommand*>(fDrawIndirectBuffer->map());
+    int glDrawCmdsIdx = 0;
+
+    bool baseInstanceSupport = this->glGpu()->glCaps().baseInstanceSupport();
+
+    if (!baseInstanceSupport) {
+        fGLDrawCmdsInfo.reset(numGLDrawCmds);
+    }
+
+    // Generate the instance and draw-indirect buffer contents based on the batches in existence.
+    iter.init(this->batchList(), BatchList::Iter::kHead_IterStart);
+    while (Batch* b = iter.get()) {
+        GLBatch* batch = static_cast<GLBatch*>(b);
+        iter.next();
+
+        batch->fEmulatedBaseInstance = baseInstanceSupport ? 0 : glInstancesIdx;
+        batch->fGLDrawCmdsIdx = glDrawCmdsIdx;
+        batch->fNumGLDrawCmds = 0;
+
+        const auto& drawCmds = batch->fDrawCmds;
+        int cidx = 0;
+        SkASSERT(!drawCmds.empty());
+        do {
+            IndexRange geometry = drawCmds[cidx].fGeometry;
+            int instanceCount = 0;
+
+            do {
+                SkASSERT(drawCmds[cidx].isValid());
+                InstanceRange range = drawCmds[cidx].fInstanceRange;
+                memcpy(&glMappedInstances[glInstancesIdx + instanceCount],
+                       &this->instance(range.fStart), range.fCount * sizeof(Instance));
+                instanceCount += range.fCount;
+            } while (++cidx < drawCmds.count() && drawCmds[cidx].fGeometry == geometry);
+
+            GrGLDrawElementsIndirectCommand& glCmd = glMappedCmds[glDrawCmdsIdx];
+            glCmd.fCount = geometry.fCount;
+            glCmd.fInstanceCount = instanceCount;
+            glCmd.fFirstIndex = geometry.fStart;
+            glCmd.fBaseVertex = 0;
+            glCmd.fBaseInstance = baseInstanceSupport ? glInstancesIdx : 0;
+
+            if (!baseInstanceSupport) {
+                fGLDrawCmdsInfo[glDrawCmdsIdx].fInstanceCount = instanceCount;
+            }
+
+            glInstancesIdx += instanceCount;
+            ++glDrawCmdsIdx;
+            ++batch->fNumGLDrawCmds;
+        } while (cidx < drawCmds.count());
+    }
+
+    SkASSERT(glDrawCmdsIdx == numGLDrawCmds);
+    fDrawIndirectBuffer->unmap();
+
+    SkASSERT(glInstancesIdx == numGLInstances);
+    fInstanceBuffer->unmap();
+}
+
+void GLInstancedRendering::onDraw(const GrPipeline& pipeline, const InstanceProcessor& instProc,
+                                  const Batch* baseBatch) {
+    if (!fDrawIndirectBuffer) {
+        return; // beginFlush was not successful.
+    }
+    if (!this->glGpu()->flushGLState(pipeline, instProc)) {
+        return;
+    }
+
+    this->glGpu()->bindBuffer(kDrawIndirect_GrBufferType, fDrawIndirectBuffer.get());
+
+    const GrGLCaps& glCaps = this->glGpu()->glCaps();
+    const GLBatch* batch = static_cast<const GLBatch*>(baseBatch);
+    int numCommands = batch->fNumGLDrawCmds;
+
+#if 0
+    SkDebugf("Instanced batch: [");
+    for (int i = 0; i < numCommands; ++i) {
+        SkDebugf("%s%i * %s", (i ? ",   " : ""), batch->fDrawCmds[i].fInstanceRange.fCount,
+                 InstanceProcessor::GetNameOfIndexRange(batch->fDrawCmds[i].fGeometry));
+    }
+    SkDebugf("]\n");
+#endif
+
+    if (1 == numCommands || !glCaps.baseInstanceSupport() || !glCaps.multiDrawIndirectSupport()) {
+        int emulatedBaseInstance = batch->fEmulatedBaseInstance;
+        for (int i = 0; i < numCommands; ++i) {
+            int glCmdIdx = batch->fGLDrawCmdsIdx + i;
+            this->flushInstanceAttribs(emulatedBaseInstance);
+            GL_CALL(DrawElementsIndirect(GR_GL_TRIANGLES, GR_GL_UNSIGNED_BYTE,
+                                         (GrGLDrawElementsIndirectCommand*) nullptr + glCmdIdx));
+            if (!glCaps.baseInstanceSupport()) {
+                emulatedBaseInstance += fGLDrawCmdsInfo[glCmdIdx].fInstanceCount;
+            }
+        }
+    } else {
+        int glCmdsIdx = batch->fGLDrawCmdsIdx;
+        this->flushInstanceAttribs(batch->fEmulatedBaseInstance);
+        GL_CALL(MultiDrawElementsIndirect(GR_GL_TRIANGLES, GR_GL_UNSIGNED_BYTE,
+                                          (GrGLDrawElementsIndirectCommand*) nullptr + glCmdsIdx,
+                                          numCommands, 0));
+    }
+}
+
+void GLInstancedRendering::flushInstanceAttribs(int baseInstance) {
+    SkASSERT(fVertexArrayID);
+    this->glGpu()->bindVertexArray(fVertexArrayID);
+
+    SkASSERT(fInstanceBuffer);
+    if (fInstanceAttribsBufferUniqueId != fInstanceBuffer->getUniqueID() ||
+        fInstanceAttribsBaseInstance != baseInstance) {
+        Instance* offsetInBuffer = (Instance*) nullptr + baseInstance;
+
+        this->glGpu()->bindBuffer(kVertex_GrBufferType, fInstanceBuffer.get());
+
+        // Info attrib.
+        GL_CALL(EnableVertexAttribArray(kInstanceInfo_AttribIdx));
+        GL_CALL(VertexAttribIPointer(kInstanceInfo_AttribIdx, 1, GR_GL_UNSIGNED_INT,
+                                     sizeof(Instance), &offsetInBuffer->fInfo));
+        GL_CALL(VertexAttribDivisor(kInstanceInfo_AttribIdx, 1));
+
+        // Shape matrix attrib.
+        GL_CALL(EnableVertexAttribArray(kShapeMatrixX_AttribIdx));
+        GL_CALL(EnableVertexAttribArray(kShapeMatrixY_AttribIdx));
+        GL_CALL(VertexAttribPointer(kShapeMatrixX_AttribIdx, 3, GR_GL_FLOAT, GR_GL_FALSE,
+                                    sizeof(Instance), &offsetInBuffer->fShapeMatrix2x3[0]));
+        GL_CALL(VertexAttribPointer(kShapeMatrixY_AttribIdx, 3, GR_GL_FLOAT, GR_GL_FALSE,
+                                    sizeof(Instance), &offsetInBuffer->fShapeMatrix2x3[3]));
+        GL_CALL(VertexAttribDivisor(kShapeMatrixX_AttribIdx, 1));
+        GL_CALL(VertexAttribDivisor(kShapeMatrixY_AttribIdx, 1));
+
+        // Color attrib.
+        GL_CALL(EnableVertexAttribArray(kColor_AttribIdx));
+        GL_CALL(VertexAttribPointer(kColor_AttribIdx, 4, GR_GL_UNSIGNED_BYTE, GR_GL_TRUE,
+                                    sizeof(Instance), &offsetInBuffer->fColor));
+        GL_CALL(VertexAttribDivisor(kColor_AttribIdx, 1));
+
+        // Local rect attrib.
+        GL_CALL(EnableVertexAttribArray(kLocalRect_AttribIdx));
+        GL_CALL(VertexAttribPointer(kLocalRect_AttribIdx, 4, GR_GL_FLOAT, GR_GL_FALSE,
+                                    sizeof(Instance), &offsetInBuffer->fLocalRect));
+        GL_CALL(VertexAttribDivisor(kLocalRect_AttribIdx, 1));
+
+        fInstanceAttribsBufferUniqueId = fInstanceBuffer->getUniqueID();
+        fInstanceAttribsBaseInstance = baseInstance;
+    }
+}
+
+void GLInstancedRendering::onEndFlush() {
+    fInstanceBuffer.reset();
+    fDrawIndirectBuffer.reset();
+    fGLDrawCmdsInfo.reset(0);
+}
+
+void GLInstancedRendering::onResetGpuResources(ResetType resetType) {
+    if (fVertexArrayID && ResetType::kDestroy == resetType) {
+        GL_CALL(DeleteVertexArrays(1, &fVertexArrayID));
+        this->glGpu()->notifyVertexArrayDelete(fVertexArrayID);
+    }
+    fVertexArrayID = 0;
+    fInstanceBuffer.reset();
+    fDrawIndirectBuffer.reset();
+    fInstanceAttribsBufferUniqueId = SK_InvalidUniqueID;
+}
+
+}