Revert of GrBatchPrototype (patchset #30 id:570001 of https://codereview.chromium.org/845103005/)

Reason for revert:
creates large performance regression

Original issue's description:
> GrBatchPrototype
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/d15e4e45374275c045572b304c229237c4a82be4

TBR=bsalomon@google.com,joshualitt@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:

Review URL: https://codereview.chromium.org/862823004

30 files changed, 309 insertions, 957 deletions

diff --git a/src/gpu/GrAAConvexPathRenderer.cpp b/src/gpu/GrAAConvexPathRenderer.cpp
index b3010767e3..031dac3c48 100644
--- a/src/gpu/GrAAConvexPathRenderer.cpp
+++ b/src/gpu/GrAAConvexPathRenderer.cpp
@@ -632,7 +632,7 @@ public:
         return SkNEW_ARGS(GLProcessor, (*this, bt));
     }
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE {
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE {
         BatchTracker* local = bt->cast<BatchTracker>();
         local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
         local->fUsesLocalCoords = init.fUsesLocalCoords;
diff --git a/src/gpu/GrAARectRenderer.cpp b/src/gpu/GrAARectRenderer.cpp
index 6866b79aef..0ffdd4d9f9 100644
--- a/src/gpu/GrAARectRenderer.cpp
+++ b/src/gpu/GrAARectRenderer.cpp
@@ -6,9 +6,6 @@
  */
 
 #include "GrAARectRenderer.h"
-#include "GrBatch.h"
-#include "GrBatchTarget.h"
-#include "GrBufferAllocPool.h"
 #include "GrDefaultGeoProcFactory.h"
 #include "GrGeometryProcessor.h"
 #include "GrGpu.h"
@@ -20,324 +17,38 @@
 
 ///////////////////////////////////////////////////////////////////////////////
 
-static void set_inset_fan(SkPoint* pts, size_t stride,
-                          const SkRect& r, SkScalar dx, SkScalar dy) {
-    pts->setRectFan(r.fLeft + dx, r.fTop + dy,
-                    r.fRight - dx, r.fBottom - dy, stride);
-}
-
-static const uint16_t gFillAARectIdx[] = {
-    0, 1, 5, 5, 4, 0,
-    1, 2, 6, 6, 5, 1,
-    2, 3, 7, 7, 6, 2,
-    3, 0, 4, 4, 7, 3,
-    4, 5, 6, 6, 7, 4,
+namespace {
+// Should the coverage be multiplied into the color attrib or use a separate attrib.
+enum CoverageAttribType {
+    kUseColor_CoverageAttribType,
+    kUseCoverage_CoverageAttribType,
 };
+}
 
-static const int kIndicesPerAAFillRect = SK_ARRAY_COUNT(gFillAARectIdx);
-static const int kVertsPerAAFillRect = 8;
-static const int kNumAAFillRectsInIndexBuffer = 256;
-
-static const GrGeometryProcessor* create_fill_rect_gp(bool tweakAlphaForCoverage,
-                                                      const SkMatrix& localMatrix) {
+static const GrGeometryProcessor* create_rect_gp(const GrPipelineBuilder& pipelineBuilder,
+                                                 GrColor color,
+                                                 CoverageAttribType* type,
+                                                 const SkMatrix& localMatrix) {
     uint32_t flags = GrDefaultGeoProcFactory::kColor_GPType;
     const GrGeometryProcessor* gp;
-    if (tweakAlphaForCoverage) {
-        gp = GrDefaultGeoProcFactory::Create(flags, GrColor_WHITE, SkMatrix::I(), localMatrix,
-                                             false, 0xff);
+    if (pipelineBuilder.canTweakAlphaForCoverage()) {
+        gp = GrDefaultGeoProcFactory::Create(flags, color, SkMatrix::I(), localMatrix);
+        SkASSERT(gp->getVertexStride() == sizeof(GrDefaultGeoProcFactory::PositionColorAttr));
+        *type = kUseColor_CoverageAttribType;
     } else {
         flags |= GrDefaultGeoProcFactory::kCoverage_GPType;
-        gp = GrDefaultGeoProcFactory::Create(flags, GrColor_WHITE, SkMatrix::I(), localMatrix,
-                                             false, 0xff);
+        gp = GrDefaultGeoProcFactory::Create(flags, color, SkMatrix::I(), localMatrix,
+                                             GrColorIsOpaque(color));
+        SkASSERT(gp->getVertexStride()==sizeof(GrDefaultGeoProcFactory::PositionColorCoverageAttr));
+        *type = kUseCoverage_CoverageAttribType;
     }
     return gp;
 }
 
-class AAFillRectBatch : public GrBatch {
-public:
-    struct Geometry {
-        GrColor fColor;
-        SkMatrix fViewMatrix;
-        SkRect fRect;
-        SkRect fDevRect;
-    };
-
-    static GrBatch* Create(const Geometry& geometry, const GrIndexBuffer* indexBuffer) {
-        return SkNEW_ARGS(AAFillRectBatch, (geometry, indexBuffer));
-    }
-
-    const char* name() const SK_OVERRIDE { return "AAFillRectBatch"; }
-
-    void getInvariantOutputColor(GrInitInvariantOutput* out) const SK_OVERRIDE {
-        // When this is called on a batch, there is only one geometry bundle
-        if (!this->canTweakAlphaForCoverage() && GrColorIsOpaque(fGeoData[0].fColor)) {
-            out->setUnknownOpaqueFourComponents();
-        } else {
-            out->setUnknownFourComponents();
-        }
-    }
-
-    void getInvariantOutputCoverage(GrInitInvariantOutput* out) const SK_OVERRIDE {
-        if (this->canTweakAlphaForCoverage()) {
-            // uniform coverage
-            out->setKnownSingleComponent(0xff);
-        } else {
-            out->setUnknownSingleComponent();
-        }
-    }
-
-    void initBatchOpt(const GrBatchOpt& batchOpt) {
-        fBatchOpt = batchOpt;
-    }
-
-    void initBatchTracker(const GrPipelineInfo& init) SK_OVERRIDE {
-        // Handle any color overrides
-        if (init.fColorIgnored) {
-            fGeoData[0].fColor = GrColor_ILLEGAL;
-        } else if (GrColor_ILLEGAL != init.fOverrideColor) {
-            fGeoData[0].fColor = init.fOverrideColor;
-        }
-
-        // setup batch properties
-        fBatch.fColorIgnored = init.fColorIgnored;
-        fBatch.fColor = fGeoData[0].fColor;
-        fBatch.fUsesLocalCoords = init.fUsesLocalCoords;
-        fBatch.fCoverageIgnored = init.fCoverageIgnored;
-    }
-
-    void generateGeometry(GrBatchTarget* batchTarget, const GrPipeline* pipeline) SK_OVERRIDE {
-        bool canTweakAlphaForCoverage = this->canTweakAlphaForCoverage();
-
-        SkMatrix localMatrix;
-        if (!this->viewMatrix().invert(&localMatrix)) {
-            SkDebugf("Cannot invert\n");
-            return;
-        }
-
-        const GrGeometryProcessor* gp = create_fill_rect_gp(canTweakAlphaForCoverage,
-                                                            localMatrix);
-
-        batchTarget->initDraw(gp, pipeline);
-        gp->unref();
-
-        // TODO this is hacky, but the only way we have to initialize the GP is to use the
-        // GrPipelineInfo struct so we can generate the correct shader.  Once we have GrBatch
-        // everywhere we can remove this nastiness
-        GrPipelineInfo init;
-        init.fColorIgnored = fBatch.fColorIgnored;
-        init.fOverrideColor = GrColor_ILLEGAL;
-        init.fCoverageIgnored = fBatch.fCoverageIgnored;
-        init.fUsesLocalCoords = this->usesLocalCoords();
-        gp->initBatchTracker(batchTarget->currentBatchTracker(), init);
-
-        size_t vertexStride = gp->getVertexStride();
-
-        SkASSERT(canTweakAlphaForCoverage ?
-                 vertexStride == sizeof(GrDefaultGeoProcFactory::PositionColorAttr) :
-                 vertexStride == sizeof(GrDefaultGeoProcFactory::PositionColorCoverageAttr));
-
-        int instanceCount = fGeoData.count();
-        int vertexCount = kVertsPerAAFillRect * instanceCount;
-
-        const GrVertexBuffer* vertexBuffer;
-        int firstVertex;
-
-        void *vertices = batchTarget->vertexPool()->makeSpace(vertexStride,
-                                                              vertexCount,
-                                                              &vertexBuffer,
-                                                              &firstVertex);
-
-        for (int i = 0; i < instanceCount; i++) {
-            const Geometry& args = fGeoData[i];
-            this->generateAAFillRectGeometry(vertices,
-                                   i * kVertsPerAAFillRect * vertexStride,
-                                   vertexStride,
-                                   args.fColor,
-                                   args.fViewMatrix,
-                                   args.fRect,
-                                   args.fDevRect,
-                                   canTweakAlphaForCoverage);
-        }
-
-        GrDrawTarget::DrawInfo drawInfo;
-        drawInfo.setPrimitiveType(kTriangles_GrPrimitiveType);
-        drawInfo.setStartVertex(0);
-        drawInfo.setStartIndex(0);
-        drawInfo.setVerticesPerInstance(kVertsPerAAFillRect);
-        drawInfo.setIndicesPerInstance(kIndicesPerAAFillRect);
-        drawInfo.adjustStartVertex(firstVertex);
-        drawInfo.setVertexBuffer(vertexBuffer);
-        drawInfo.setIndexBuffer(fIndexBuffer);
-
-        int maxInstancesPerDraw = kNumAAFillRectsInIndexBuffer;
-
-        while (instanceCount) {
-            drawInfo.setInstanceCount(SkTMin(instanceCount, maxInstancesPerDraw));
-            drawInfo.setVertexCount(drawInfo.instanceCount() * drawInfo.verticesPerInstance());
-            drawInfo.setIndexCount(drawInfo.instanceCount() * drawInfo.indicesPerInstance());
-
-            batchTarget->draw(drawInfo);
-
-            drawInfo.setStartVertex(drawInfo.startVertex() + drawInfo.vertexCount());
-            instanceCount -= drawInfo.instanceCount();
-        }
-    }
-
-    SkSTArray<1, Geometry, true>* geoData() { return &fGeoData; }
-
-private:
-    AAFillRectBatch(const Geometry& geometry, const GrIndexBuffer* indexBuffer)
-        : fIndexBuffer(indexBuffer) {
-        this->initClassID<AAFillRectBatch>();
-        fGeoData.push_back(geometry);
-    }
-
-    GrColor color() const { return fBatch.fColor; }
-    bool usesLocalCoords() const { return fBatch.fUsesLocalCoords; }
-    bool canTweakAlphaForCoverage() const { return fBatchOpt.fCanTweakAlphaForCoverage; }
-    bool colorIgnored() const { return fBatch.fColorIgnored; }
-    const SkMatrix& viewMatrix() const { return fGeoData[0].fViewMatrix; }
-
-    bool onCombineIfPossible(GrBatch* t) SK_OVERRIDE {
-        AAFillRectBatch* that = t->cast<AAFillRectBatch>();
-        if (this->canTweakAlphaForCoverage() != that->canTweakAlphaForCoverage()) {
-            return false;
-        }
-
-        if (this->colorIgnored() != that->colorIgnored()) {
-            return false;
-        }
-
-        if (this->usesLocalCoords() != that->usesLocalCoords()) {
-            return false;
-        }
-
-        // We apply the viewmatrix to the rect points on the cpu.  However, if the pipeline uses
-        // local coords then we won't be able to batch.  We could actually upload the viewmatrix
-        // using vertex attributes in these cases, but haven't investigated that
-        if (this->usesLocalCoords() && !this->viewMatrix().cheapEqualTo(that->viewMatrix())) {
-            return false;
-        }
-
-        if (this->color() != that->color()) {
-            fBatch.fColor = GrColor_ILLEGAL;
-        }
-        fGeoData.push_back_n(that->geoData()->count(), that->geoData()->begin());
-        return true;
-    }
-
-    void generateAAFillRectGeometry(void* vertices,
-                                    uint32_t offset,
-                                    uint32_t vertexStride,
-                                    GrColor color,
-                                    const SkMatrix& viewMatrix,
-                                    const SkRect& rect,
-                                    const SkRect& devRect,
-                                    bool tweakAlphaForCoverage) const {
-        intptr_t verts = reinterpret_cast<intptr_t>(vertices) + offset;
-
-        SkPoint* fan0Pos = reinterpret_cast<SkPoint*>(verts);
-        SkPoint* fan1Pos = reinterpret_cast<SkPoint*>(verts + 4 * vertexStride);
-
-        SkScalar inset = SkMinScalar(devRect.width(), SK_Scalar1);
-        inset = SK_ScalarHalf * SkMinScalar(inset, devRect.height());
-
-        if (viewMatrix.rectStaysRect()) {
-            set_inset_fan(fan0Pos, vertexStride, devRect, -SK_ScalarHalf, -SK_ScalarHalf);
-            set_inset_fan(fan1Pos, vertexStride, devRect, inset,  inset);
-        } else {
-            // compute transformed (1, 0) and (0, 1) vectors
-            SkVector vec[2] = {
-              { viewMatrix[SkMatrix::kMScaleX], viewMatrix[SkMatrix::kMSkewY] },
-              { viewMatrix[SkMatrix::kMSkewX],  viewMatrix[SkMatrix::kMScaleY] }
-            };
-
-            vec[0].normalize();
-            vec[0].scale(SK_ScalarHalf);
-            vec[1].normalize();
-            vec[1].scale(SK_ScalarHalf);
-
-            // create the rotated rect
-            fan0Pos->setRectFan(rect.fLeft, rect.fTop,
-                                rect.fRight, rect.fBottom, vertexStride);
-            viewMatrix.mapPointsWithStride(fan0Pos, vertexStride, 4);
-
-            // Now create the inset points and then outset the original
-            // rotated points
-
-            // TL
-            *((SkPoint*)((intptr_t)fan1Pos + 0 * vertexStride)) =
-                *((SkPoint*)((intptr_t)fan0Pos + 0 * vertexStride)) + vec[0] + vec[1];
-            *((SkPoint*)((intptr_t)fan0Pos + 0 * vertexStride)) -= vec[0] + vec[1];
-            // BL
-            *((SkPoint*)((intptr_t)fan1Pos + 1 * vertexStride)) =
-                *((SkPoint*)((intptr_t)fan0Pos + 1 * vertexStride)) + vec[0] - vec[1];
-            *((SkPoint*)((intptr_t)fan0Pos + 1 * vertexStride)) -= vec[0] - vec[1];
-            // BR
-            *((SkPoint*)((intptr_t)fan1Pos + 2 * vertexStride)) =
-                *((SkPoint*)((intptr_t)fan0Pos + 2 * vertexStride)) - vec[0] - vec[1];
-            *((SkPoint*)((intptr_t)fan0Pos + 2 * vertexStride)) += vec[0] + vec[1];
-            // TR
-            *((SkPoint*)((intptr_t)fan1Pos + 3 * vertexStride)) =
-                *((SkPoint*)((intptr_t)fan0Pos + 3 * vertexStride)) - vec[0] + vec[1];
-            *((SkPoint*)((intptr_t)fan0Pos + 3 * vertexStride)) += vec[0] - vec[1];
-        }
-
-        // Make verts point to vertex color and then set all the color and coverage vertex attrs
-        // values.
-        verts += sizeof(SkPoint);
-        for (int i = 0; i < 4; ++i) {
-            if (tweakAlphaForCoverage) {
-                *reinterpret_cast<GrColor*>(verts + i * vertexStride) = 0;
-            } else {
-                *reinterpret_cast<GrColor*>(verts + i * vertexStride) = color;
-                *reinterpret_cast<float*>(verts + i * vertexStride + sizeof(GrColor)) = 0;
-            }
-        }
-
-        int scale;
-        if (inset < SK_ScalarHalf) {
-            scale = SkScalarFloorToInt(512.0f * inset / (inset + SK_ScalarHalf));
-            SkASSERT(scale >= 0 && scale <= 255);
-        } else {
-            scale = 0xff;
-        }
-
-        verts += 4 * vertexStride;
-
-        float innerCoverage = GrNormalizeByteToFloat(scale);
-        GrColor scaledColor = (0xff == scale) ? color : SkAlphaMulQ(color, scale);
-
-        for (int i = 0; i < 4; ++i) {
-            if (tweakAlphaForCoverage) {
-                *reinterpret_cast<GrColor*>(verts + i * vertexStride) = scaledColor;
-            } else {
-                *reinterpret_cast<GrColor*>(verts + i * vertexStride) = color;
-                *reinterpret_cast<float*>(verts + i * vertexStride +
-                                          sizeof(GrColor)) = innerCoverage;
-            }
-        }
-    }
-
-    struct BatchTracker {
-        GrColor fColor;
-        bool fUsesLocalCoords;
-        bool fColorIgnored;
-        bool fCoverageIgnored;
-    };
-
-    GrBatchOpt fBatchOpt;
-    BatchTracker fBatch;
-    const GrIndexBuffer* fIndexBuffer;
-    SkSTArray<1, Geometry, true> fGeoData;
-};
-
-namespace {
-// Should the coverage be multiplied into the color attrib or use a separate attrib.
-enum CoverageAttribType {
-    kUseColor_CoverageAttribType,
-    kUseCoverage_CoverageAttribType,
-};
+static void set_inset_fan(SkPoint* pts, size_t stride,
+                          const SkRect& r, SkScalar dx, SkScalar dy) {
+    pts->setRectFan(r.fLeft + dx, r.fTop + dy,
+                    r.fRight - dx, r.fBottom - dy, stride);
 }
 
 void GrAARectRenderer::reset() {
@@ -346,6 +57,18 @@ void GrAARectRenderer::reset() {
     SkSafeSetNull(fAABevelStrokeRectIndexBuffer);
 }
 
+static const uint16_t gFillAARectIdx[] = {
+    0, 1, 5, 5, 4, 0,
+    1, 2, 6, 6, 5, 1,
+    2, 3, 7, 7, 6, 2,
+    3, 0, 4, 4, 7, 3,
+    4, 5, 6, 6, 7, 4,
+};
+
+static const int kIndicesPerAAFillRect = SK_ARRAY_COUNT(gFillAARectIdx);
+static const int kVertsPerAAFillRect = 8;
+static const int kNumAAFillRectsInIndexBuffer = 256;
+
 static const uint16_t gMiterStrokeAARectIdx[] = {
     0 + 0, 1 + 0, 5 + 0, 5 + 0, 4 + 0, 0 + 0,
     1 + 0, 2 + 0, 6 + 0, 6 + 0, 5 + 0, 1 + 0,
@@ -461,21 +184,135 @@ void GrAARectRenderer::geometryFillAARect(GrDrawTarget* target,
                                           const SkMatrix& viewMatrix,
                                           const SkRect& rect,
                                           const SkRect& devRect) {
+    GrPipelineBuilder::AutoRestoreEffects are(pipelineBuilder);
+
+    SkMatrix localMatrix;
+    if (!viewMatrix.invert(&localMatrix)) {
+        SkDebugf("Cannot invert\n");
+        return;
+    }
+
+    CoverageAttribType type;
+    SkAutoTUnref<const GrGeometryProcessor> gp(create_rect_gp(*pipelineBuilder, color, &type,
+                                                              localMatrix));
+
+    size_t vertexStride = gp->getVertexStride();
+    GrDrawTarget::AutoReleaseGeometry geo(target, 8, vertexStride, 0);
+    if (!geo.succeeded()) {
+        SkDebugf("Failed to get space for vertices!\n");
+        return;
+    }
+
     if (NULL == fAAFillRectIndexBuffer) {
         fAAFillRectIndexBuffer = fGpu->createInstancedIndexBuffer(gFillAARectIdx,
                                                                   kIndicesPerAAFillRect,
                                                                   kNumAAFillRectsInIndexBuffer,
                                                                   kVertsPerAAFillRect);
     }
+    GrIndexBuffer* indexBuffer = fAAFillRectIndexBuffer;
+    if (NULL == indexBuffer) {
+        SkDebugf("Failed to create index buffer!\n");
+        return;
+    }
+
+    intptr_t verts = reinterpret_cast<intptr_t>(geo.vertices());
+
+    SkPoint* fan0Pos = reinterpret_cast<SkPoint*>(verts);
+    SkPoint* fan1Pos = reinterpret_cast<SkPoint*>(verts + 4 * vertexStride);
+
+    SkScalar inset = SkMinScalar(devRect.width(), SK_Scalar1);
+    inset = SK_ScalarHalf * SkMinScalar(inset, devRect.height());
+
+    if (viewMatrix.rectStaysRect()) {
+        // Temporarily #if'ed out. We don't want to pass in the devRect but
+        // right now it is computed in GrContext::apply_aa_to_rect and we don't
+        // want to throw away the work
+#if 0
+        SkRect devRect;
+        combinedMatrix.mapRect(&devRect, rect);
+#endif
+
+        set_inset_fan(fan0Pos, vertexStride, devRect, -SK_ScalarHalf, -SK_ScalarHalf);
+        set_inset_fan(fan1Pos, vertexStride, devRect, inset,  inset);
+    } else {
+        // compute transformed (1, 0) and (0, 1) vectors
+        SkVector vec[2] = {
+          { viewMatrix[SkMatrix::kMScaleX], viewMatrix[SkMatrix::kMSkewY] },
+          { viewMatrix[SkMatrix::kMSkewX],  viewMatrix[SkMatrix::kMScaleY] }
+        };
+
+        vec[0].normalize();
+        vec[0].scale(SK_ScalarHalf);
+        vec[1].normalize();
+        vec[1].scale(SK_ScalarHalf);
+
+        // create the rotated rect
+        fan0Pos->setRectFan(rect.fLeft, rect.fTop,
+                            rect.fRight, rect.fBottom, vertexStride);
+        viewMatrix.mapPointsWithStride(fan0Pos, vertexStride, 4);
+
+        // Now create the inset points and then outset the original
+        // rotated points
+
+        // TL
+        *((SkPoint*)((intptr_t)fan1Pos + 0 * vertexStride)) =
+            *((SkPoint*)((intptr_t)fan0Pos + 0 * vertexStride)) + vec[0] + vec[1];
+        *((SkPoint*)((intptr_t)fan0Pos + 0 * vertexStride)) -= vec[0] + vec[1];
+        // BL
+        *((SkPoint*)((intptr_t)fan1Pos + 1 * vertexStride)) =
+            *((SkPoint*)((intptr_t)fan0Pos + 1 * vertexStride)) + vec[0] - vec[1];
+        *((SkPoint*)((intptr_t)fan0Pos + 1 * vertexStride)) -= vec[0] - vec[1];
+        // BR
+        *((SkPoint*)((intptr_t)fan1Pos + 2 * vertexStride)) =
+            *((SkPoint*)((intptr_t)fan0Pos + 2 * vertexStride)) - vec[0] - vec[1];
+        *((SkPoint*)((intptr_t)fan0Pos + 2 * vertexStride)) += vec[0] + vec[1];
+        // TR
+        *((SkPoint*)((intptr_t)fan1Pos + 3 * vertexStride)) =
+            *((SkPoint*)((intptr_t)fan0Pos + 3 * vertexStride)) - vec[0] + vec[1];
+        *((SkPoint*)((intptr_t)fan0Pos + 3 * vertexStride)) += vec[0] - vec[1];
+    }
 
-    AAFillRectBatch::Geometry geometry;
-    geometry.fRect = rect;
-    geometry.fViewMatrix = viewMatrix;
-    geometry.fDevRect = devRect;
-    geometry.fColor = color;
+    // Make verts point to vertex color and then set all the color and coverage vertex attrs values.
+    verts += sizeof(SkPoint);
+    for (int i = 0; i < 4; ++i) {
+        if (kUseCoverage_CoverageAttribType == type) {
+            *reinterpret_cast<GrColor*>(verts + i * vertexStride) = color;
+            *reinterpret_cast<float*>(verts + i * vertexStride + sizeof(GrColor)) = 0;
+        } else {
+            *reinterpret_cast<GrColor*>(verts + i * vertexStride) = 0;
+        }
+    }
 
-    SkAutoTUnref<GrBatch> batch(AAFillRectBatch::Create(geometry, fAAFillRectIndexBuffer));
-    target->drawBatch(pipelineBuilder, batch, &devRect);
+    int scale;
+    if (inset < SK_ScalarHalf) {
+        scale = SkScalarFloorToInt(512.0f * inset / (inset + SK_ScalarHalf));
+        SkASSERT(scale >= 0 && scale <= 255);
+    } else {
+        scale = 0xff;
+    }
+
+    verts += 4 * vertexStride;
+
+    float innerCoverage = GrNormalizeByteToFloat(scale);
+    GrColor scaledColor = (0xff == scale) ? color : SkAlphaMulQ(color, scale);
+
+    for (int i = 0; i < 4; ++i) {
+        if (kUseCoverage_CoverageAttribType == type) {
+            *reinterpret_cast<GrColor*>(verts + i * vertexStride) = color;
+            *reinterpret_cast<float*>(verts + i * vertexStride + sizeof(GrColor)) = innerCoverage;
+        } else {
+            *reinterpret_cast<GrColor*>(verts + i * vertexStride) = scaledColor;
+        }
+    }
+
+    target->setIndexSourceToBuffer(indexBuffer);
+    target->drawIndexedInstances(pipelineBuilder,
+                                 gp,
+                                 kTriangles_GrPrimitiveType,
+                                 1,
+                                 kVertsPerAAFillRect,
+                                 kIndicesPerAAFillRect);
+    target->resetIndexSource();
 }
 
 void GrAARectRenderer::strokeAARect(GrDrawTarget* target,
@@ -545,31 +382,10 @@ void GrAARectRenderer::strokeAARect(GrDrawTarget* target,
         devOutsideAssist.outset(0, ry);
     }
 
-    this->geometryStrokeAARect(target, pipelineBuilder, color, viewMatrix, devOutside,
-                               devOutsideAssist, devInside, miterStroke);
+    this->geometryStrokeAARect(target, pipelineBuilder, color, viewMatrix, devOutside, devOutsideAssist,
+                               devInside, miterStroke);
 }
 
-static const GrGeometryProcessor* create_rect_gp(const GrPipelineBuilder& pipelneBuilder,
-                                                 GrColor color,
-                                                 CoverageAttribType* type,
-                                                 const SkMatrix& localMatrix) {
-    uint32_t flags = GrDefaultGeoProcFactory::kColor_GPType;
-    const GrGeometryProcessor* gp;
-    if (pipelneBuilder.canTweakAlphaForCoverage()) {
-        gp = GrDefaultGeoProcFactory::Create(flags, color, SkMatrix::I(), localMatrix);
-        SkASSERT(gp->getVertexStride() == sizeof(GrDefaultGeoProcFactory::PositionColorAttr));
-        *type = kUseColor_CoverageAttribType;
-    } else {
-        flags |= GrDefaultGeoProcFactory::kCoverage_GPType;
-        gp = GrDefaultGeoProcFactory::Create(flags, color, SkMatrix::I(), localMatrix,
-                                             GrColorIsOpaque(color));
-        SkASSERT(gp->getVertexStride()==sizeof(GrDefaultGeoProcFactory::PositionColorCoverageAttr));
-        *type = kUseCoverage_CoverageAttribType;
-    }
-    return gp;
-}
-
-
 void GrAARectRenderer::geometryStrokeAARect(GrDrawTarget* target,
                                             GrPipelineBuilder* pipelineBuilder,
                                             GrColor color,
@@ -578,6 +394,8 @@ void GrAARectRenderer::geometryStrokeAARect(GrDrawTarget* target,
                                             const SkRect& devOutsideAssist,
                                             const SkRect& devInside,
                                             bool miterStroke) {
+    GrPipelineBuilder::AutoRestoreEffects are(pipelineBuilder);
+
     SkMatrix localMatrix;
     if (!viewMatrix.invert(&localMatrix)) {
         SkDebugf("Cannot invert\n");
diff --git a/src/gpu/GrBatch.cpp b/src/gpu/GrBatch.cpp
deleted file mode 100644
index e1650a6bd3..0000000000
--- a/src/gpu/GrBatch.cpp
+++ /dev/null
@@ -1,35 +0,0 @@
-#include "GrBatch.h"
-
-#include "GrMemoryPool.h"
-#include "SkTLS.h"
-
-// TODO I noticed a small benefit to using a larger exclusive pool for batches.  Its very small,
-// but seems to be mostly consistent.  There is a lot in flux right now, but we should really
-// revisit this when batch is everywhere
-
-class GrBatch_Globals {
-public:
-    static GrMemoryPool* GetTLS() {
-        return (GrMemoryPool*)SkTLS::Get(CreateTLS, DeleteTLS);
-    }
-
-private:
-    static void* CreateTLS() {
-        return SkNEW_ARGS(GrMemoryPool, (16384, 16384));
-    }
-
-    static void DeleteTLS(void* pool) {
-        SkDELETE(reinterpret_cast<GrMemoryPool*>(pool));
-    }
-};
-
-int32_t GrBatch::gCurrBatchClassID =
-        GrBatch::kIllegalBatchClassID;
-
-void* GrBatch::operator new(size_t size) {
-    return GrBatch_Globals::GetTLS()->allocate(size);
-}
-
-void GrBatch::operator delete(void* target) {
-    GrBatch_Globals::GetTLS()->release(target);
-}
diff --git a/src/gpu/GrBatch.h b/src/gpu/GrBatch.h
deleted file mode 100644
index ceb2c5cc2a..0000000000
--- a/src/gpu/GrBatch.h
+++ /dev/null
@@ -1,132 +0,0 @@
-/*
- * Copyright 2015 Google Inc.
- *
- * Use of this source code is governed by a BSD-style license that can be
- * found in the LICENSE file.
- */
-
-#ifndef GrBatch_DEFINED
-#define GrBatch_DEFINED
-
-#include <new>
-// TODO remove this header when we move entirely to batch
-#include "GrGeometryProcessor.h"
-#include "SkRefCnt.h"
-#include "SkThread.h"
-#include "SkTypes.h"
-
-class GrBatchTarget;
-class GrGpu;
-class GrIndexBufferAllocPool;
-class GrPipeline;
-class GrVertexBufferAllocPool;
-
-struct GrInitInvariantOutput;
-
-/*
- * GrBatch is the base class for all Ganesh deferred geometry generators.  To facilitate
- * reorderable batching, Ganesh does not generate geometry inline with draw calls.  Instead, it
- * captures the arguments to the draw and then generates the geometry on demand.  This gives GrBatch
- * subclasses complete freedom to decide how / what they can batch.
- *
- * Batches are created when GrContext processes a draw call. Batches of the same  subclass may be
- * merged using combineIfPossible. When two batches merge, one takes on the union of the data
- * and the other is left empty. The merged batch becomes responsible for drawing the data from both
- * the original batches.
- *
- * If there are any possible optimizations which might require knowing more about the full state of
- * the draw, ie whether or not the GrBatch is allowed to tweak alpha for coverage, then this
- * information will be communicated to the GrBatch prior to geometry generation.
- */
-
-struct GrBatchOpt {
-  bool fCanTweakAlphaForCoverage;
-};
-
-class GrBatch : public SkRefCnt {
-public:
-    SK_DECLARE_INST_COUNT(GrBatch)
-    GrBatch() { SkDEBUGCODE(fUsed = false;) }
-    virtual ~GrBatch() {}
-
-    virtual const char* name() const = 0;
-    virtual void getInvariantOutputColor(GrInitInvariantOutput* out) const = 0;
-    virtual void getInvariantOutputCoverage(GrInitInvariantOutput* out) const = 0;
-
-    /*
-     * initBatchOpt is used to communicate possible optimizations to the GrBatch.  initBatchTracker
-     * is a hook for the some additional overrides from the GrXferProcessor.  This is a bit
-     * confusing but has to be like this until GrBatch is everywhere.
-     *
-     * TODO combine to a single init call when GrBatch is everywhere.
-     */
-    virtual void initBatchOpt(const GrBatchOpt&) = 0;
-    virtual void initBatchTracker(const GrPipelineInfo& init) = 0;
-
-    bool combineIfPossible(GrBatch* that) {
-        if (this->classID() != that->classID()) {
-            return false;
-        }
-
-        return onCombineIfPossible(that);
-    }
-
-    virtual bool onCombineIfPossible(GrBatch*) = 0;
-
-    virtual void generateGeometry(GrBatchTarget*, const GrPipeline*) = 0;
-
-    void* operator new(size_t size);
-    void operator delete(void* target);
-
-    void* operator new(size_t size, void* placement) {
-        return ::operator new(size, placement);
-    }
-    void operator delete(void* target, void* placement) {
-        ::operator delete(target, placement);
-    }
-
-    /**
-      * Helper for down-casting to a GrBatch subclass
-      */
-    template <typename T> const T& cast() const { return *static_cast<const T*>(this); }
-    template <typename T> T* cast() { return static_cast<T*>(this); }
-
-    uint32_t classID() const { SkASSERT(kIllegalBatchClassID != fClassID); return fClassID; }
-
-    // TODO no GrPrimitiveProcessors yet read fragment position
-    bool willReadFragmentPosition() const { return false; }
-
-    SkDEBUGCODE(bool isUsed() const { return fUsed; })
-
-protected:
-    template <typename PROC_SUBCLASS> void initClassID() {
-         static uint32_t kClassID = GenClassID();
-         fClassID = kClassID;
-    }
-
-    uint32_t fClassID;
-
-private:
-    static uint32_t GenClassID() {
-        // fCurrProcessorClassID has been initialized to kIllegalProcessorClassID. The
-        // atomic inc returns the old value not the incremented value. So we add
-        // 1 to the returned value.
-        uint32_t id = static_cast<uint32_t>(sk_atomic_inc(&gCurrBatchClassID)) + 1;
-        if (!id) {
-            SkFAIL("This should never wrap as it should only be called once for each GrBatch "
-                   "subclass.");
-        }
-        return id;
-    }
-
-    enum {
-        kIllegalBatchClassID = 0,
-    };
-    static int32_t gCurrBatchClassID;
-
-    SkDEBUGCODE(bool fUsed;)
-
-    typedef SkRefCnt INHERITED;
-};
-
-#endif
diff --git a/src/gpu/GrBatchTarget.cpp b/src/gpu/GrBatchTarget.cpp
deleted file mode 100644
index 1ae8b638e5..0000000000
--- a/src/gpu/GrBatchTarget.cpp
+++ /dev/null
@@ -1,32 +0,0 @@
-/*
- * Copyright 2015 Google Inc.
- *
- * Use of this source code is governed by a BSD-style license that can be
- * found in the LICENSE file.
- */
-
-#include "GrBatchTarget.h"
-
-#include "GrBufferAllocPool.h"
-#include "GrPipeline.h"
-
-void GrBatchTarget::flush() {
-    FlushBuffer::Iter iter(fFlushBuffer);
-    fVertexPool->unmap();
-    fIndexPool->unmap();
-
-    while (iter.next()) {
-        GrProgramDesc desc;
-        BufferedFlush* bf = iter.get();
-        const GrPipeline* pipeline = bf->fPipeline;
-        const GrPrimitiveProcessor* primProc = bf->fPrimitiveProcessor.get();
-        fGpu->buildProgramDesc(&desc, *primProc, *pipeline, pipeline->descInfo(),
-                               bf->fBatchTracker);
-
-        GrGpu::DrawArgs args(primProc, pipeline, &desc, &bf->fBatchTracker);
-        for (int i = 0; i < bf->fDraws.count(); i++) {
-            fGpu->draw(args, bf->fDraws[i]);
-        }
-    }
-    fFlushBuffer.reset();
-}
diff --git a/src/gpu/GrBatchTarget.h b/src/gpu/GrBatchTarget.h
deleted file mode 100644
index 82dad45c22..0000000000
--- a/src/gpu/GrBatchTarget.h
+++ /dev/null
@@ -1,78 +0,0 @@
-/*
- * Copyright 2015 Google Inc.
- *
- * Use of this source code is governed by a BSD-style license that can be
- * found in the LICENSE file.
- */
-
-#ifndef GrBatchBuffer_DEFINED
-#define GrBatchBuffer_DEFINED
-
-#include "GrPendingProgramElement.h"
-#include "GrGpu.h"
-#include "GrTRecorder.h"
-
-/*
- * GrBatch instances use this object to allocate space for their geometry and to issue the draws
- * that render their batch.
- */
-
-class GrBatchTarget : public SkNoncopyable {
-public:
-    GrBatchTarget(GrGpu* gpu,
-                  GrVertexBufferAllocPool* vpool,
-                  GrIndexBufferAllocPool* ipool)
-        : fGpu(gpu)
-        , fVertexPool(vpool)
-        , fIndexPool(ipool)
-        , fFlushBuffer(kFlushBufferInitialSizeInBytes) {}
-
-    typedef GrDrawTarget::DrawInfo DrawInfo;
-    void initDraw(const GrPrimitiveProcessor* primProc, const GrPipeline* pipeline) {
-        GrNEW_APPEND_TO_RECORDER(fFlushBuffer, BufferedFlush, (primProc, pipeline));
-    }
-
-    void draw(const GrDrawTarget::DrawInfo& draw) {
-        fFlushBuffer.back().fDraws.push_back(draw);
-    }
-    void flush();
-
-    // TODO This goes away when everything uses batch
-    GrBatchTracker* currentBatchTracker() {
-        SkASSERT(!fFlushBuffer.empty());
-        return &fFlushBuffer.back().fBatchTracker;
-    }
-
-    GrVertexBufferAllocPool* vertexPool() { return fVertexPool; }
-    GrIndexBufferAllocPool* indexPool() { return fIndexPool; }
-
-private:
-    GrGpu* fGpu;
-    GrVertexBufferAllocPool* fVertexPool;
-    GrIndexBufferAllocPool* fIndexPool;
-
-    typedef void* TBufferAlign; // This wouldn't be enough align if a command used long double.
-
-    struct BufferedFlush {
-        BufferedFlush(const GrPrimitiveProcessor* primProc, const GrPipeline* pipeline)
-            : fPrimitiveProcessor(primProc)
-            , fPipeline(pipeline)
-            , fDraws(kDrawRecorderInitialSizeInBytes) {}
-        typedef GrPendingProgramElement<const GrPrimitiveProcessor> ProgramPrimitiveProcessor;
-        ProgramPrimitiveProcessor fPrimitiveProcessor;
-        const GrPipeline* fPipeline;
-        GrBatchTracker fBatchTracker;
-        SkSTArray<4, DrawInfo, true> fDraws;
-    };
-
-    enum {
-        kFlushBufferInitialSizeInBytes = 8 * sizeof(BufferedFlush),
-        kDrawRecorderInitialSizeInBytes = 8 * sizeof(DrawInfo),
-    };
-
-    typedef GrTRecorder<BufferedFlush, TBufferAlign> FlushBuffer;
-
-    FlushBuffer fFlushBuffer;
-};
-
-#endif
diff --git a/src/gpu/GrDefaultGeoProcFactory.cpp b/src/gpu/GrDefaultGeoProcFactory.cpp
index c0e8ea700f..20ec66553c 100644
--- a/src/gpu/GrDefaultGeoProcFactory.cpp
+++ b/src/gpu/GrDefaultGeoProcFactory.cpp
@@ -42,7 +42,7 @@ public:
     const Attribute* inCoverage() const { return fInCoverage; }
     uint8_t coverage() const { return fCoverage; }
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE {
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE {
         BatchTracker* local = bt->cast<BatchTracker>();
         local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init,
                                                    SkToBool(fInColor));
diff --git a/src/gpu/GrDrawTarget.cpp b/src/gpu/GrDrawTarget.cpp
index b89d70ea2b..e8a758d25f 100644
--- a/src/gpu/GrDrawTarget.cpp
+++ b/src/gpu/GrDrawTarget.cpp
@@ -6,9 +6,9 @@
  * found in the LICENSE file.
  */
 
-#include "GrDrawTarget.h"
 
-#include "GrBatch.h"
+
+#include "GrDrawTarget.h"
 #include "GrContext.h"
 #include "GrDrawTargetCaps.h"
 #include "GrPath.h"
@@ -526,29 +526,6 @@ void GrDrawTarget::drawNonIndexed(GrPipelineBuilder* pipelineBuilder,
     }
 }
 
-
-void GrDrawTarget::drawBatch(GrPipelineBuilder* pipelineBuilder,
-                             GrBatch* batch,
-                             const SkRect* devBounds) {
-    SkASSERT(pipelineBuilder);
-    // TODO some kind of checkdraw, but not at this level
-
-    // Setup clip
-    GrScissorState scissorState;
-    GrPipelineBuilder::AutoRestoreEffects are;
-    GrPipelineBuilder::AutoRestoreStencil ars;
-    if (!this->setupClip(pipelineBuilder, &are, &ars, &scissorState, devBounds)) {
-        return;
-    }
-
-    GrDeviceCoordTexture dstCopy;
-    if (!this->setupDstReadIfNecessary(pipelineBuilder, &dstCopy, devBounds)) {
-        return;
-    }
-
-    this->onDrawBatch(batch, *pipelineBuilder, scissorState, dstCopy.texture() ? &dstCopy : NULL);
-}
-
 static const GrStencilSettings& winding_path_stencil_settings() {
     GR_STATIC_CONST_SAME_STENCIL_STRUCT(gSettings,
         kIncClamp_StencilOp,
diff --git a/src/gpu/GrDrawTarget.h b/src/gpu/GrDrawTarget.h
index 18265c8fb7..b66f1c6bd3 100644
--- a/src/gpu/GrDrawTarget.h
+++ b/src/gpu/GrDrawTarget.h
@@ -26,7 +26,6 @@
 #include "SkTypes.h"
 #include "SkXfermode.h"
 
-class GrBatch;
 class GrClipData;
 class GrDrawTargetCaps;
 class GrPath;
@@ -260,11 +259,6 @@ public:
                         int vertexCount,
                         const SkRect* devBounds = NULL);
 
-    // TODO devbounds should live on the batch
-    void drawBatch(GrPipelineBuilder*,
-                   GrBatch*,
-                   const SkRect* devBounds = NULL);
-
     /**
      * Draws path into the stencil buffer. The fill must be either even/odd or
      * winding (not inverse or hairline). It will respect the HW antialias flag
@@ -316,14 +310,14 @@ public:
      *                    that rectangle before it is input to GrCoordTransforms that read local
      *                    coordinates
      */
-    void drawRect(GrPipelineBuilder* pipelineBuilder,
+    void drawRect(GrPipelineBuilder* ds,
                   GrColor color,
                   const SkMatrix& viewMatrix,
                   const SkRect& rect,
                   const SkRect* localRect,
                   const SkMatrix* localMatrix) {
         AutoGeometryPush agp(this);
-        this->onDrawRect(pipelineBuilder, color, viewMatrix, rect, localRect, localMatrix);
+        this->onDrawRect(ds, color, viewMatrix, rect, localRect, localMatrix);
     }
 
     /**
@@ -533,7 +527,6 @@ public:
      */
     class DrawInfo {
     public:
-        DrawInfo() { fDevBounds = NULL; }
         DrawInfo(const DrawInfo& di) { (*this) = di; }
         DrawInfo& operator =(const DrawInfo& di);
 
@@ -546,15 +539,6 @@ public:
         int indicesPerInstance() const { return fIndicesPerInstance; }
         int instanceCount() const { return fInstanceCount; }
 
-        void setPrimitiveType(GrPrimitiveType type) { fPrimitiveType = type; }
-        void setStartVertex(int startVertex) { fStartVertex = startVertex; }
-        void setStartIndex(int startIndex) { fStartIndex = startIndex; }
-        void setVertexCount(int vertexCount) { fVertexCount = vertexCount; }
-        void setIndexCount(int indexCount) { fIndexCount = indexCount; }
-        void setVerticesPerInstance(int verticesPerI) { fVerticesPerInstance = verticesPerI; }
-        void setIndicesPerInstance(int indicesPerI) { fIndicesPerInstance = indicesPerI; }
-        void setInstanceCount(int instanceCount) { fInstanceCount = instanceCount; }
-
         bool isIndexed() const { return fIndexCount > 0; }
 #ifdef SK_DEBUG
         bool isInstanced() const; // this version is longer because of asserts
@@ -584,6 +568,8 @@ public:
         const SkRect* getDevBounds() const { return fDevBounds; }
 
     private:
+        DrawInfo() { fDevBounds = NULL; }
+
         friend class GrDrawTarget;
 
         GrPrimitiveType         fPrimitiveType;
@@ -722,10 +708,6 @@ private:
                         const DrawInfo&,
                         const GrScissorState&,
                         const GrDeviceCoordTexture* dstCopy) = 0;
-    virtual void onDrawBatch(GrBatch*,
-                             const GrPipelineBuilder&,
-                             const GrScissorState&,
-                             const GrDeviceCoordTexture* dstCopy) = 0;
     // TODO copy in order drawbuffer onDrawRect to here
     virtual void onDrawRect(GrPipelineBuilder*,
                             GrColor color,
diff --git a/src/gpu/GrFlushToGpuDrawTarget.h b/src/gpu/GrFlushToGpuDrawTarget.h
index fa06ed9630..2383c623d0 100644
--- a/src/gpu/GrFlushToGpuDrawTarget.h
+++ b/src/gpu/GrFlushToGpuDrawTarget.h
@@ -46,9 +46,6 @@ protected:
     GrGpu* getGpu() { return fGpu; }
     const GrGpu* getGpu() const{ return fGpu; }
 
-    GrVertexBufferAllocPool* getVertexAllocPool() { return fVertexPool; }
-    GrIndexBufferAllocPool* getIndexAllocPool() { return fIndexPool; }
-
 private:
     enum {
         kGeoPoolStatePreAllocCnt = 4,
diff --git a/src/gpu/GrGeometryData.h b/src/gpu/GrGeometryData.h
new file mode 100644
index 0000000000..a4e9fe6871
--- /dev/null
+++ b/src/gpu/GrGeometryData.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright 2014 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#ifndef GrGeometryData_DEFINED
+#define GrGeometryData_DEFINED
+
+#include <new>
+#include "SkTypes.h"
+
+/*
+ * A super lightweight base class for GeometryProcessor's to use to store draw data in a reorderable
+ * fashion.  Its most important feature is a pool allocator.  Its virtual, but only so subclasses
+ * will have their destructors called.
+ */
+
+class GrGeometryData : SkNoncopyable {
+public:
+    virtual ~GrGeometryData() {}
+
+    /**
+      * Helper for down-casting to a GrGeometryData subclass
+      */
+    template <typename T> const T& cast() const { return *static_cast<const T*>(this); }
+
+    void* operator new(size_t size);
+
+    void operator delete(void* target);
+
+    void* operator new(size_t size, void* placement) {
+        return ::operator new(size, placement);
+    }
+
+    void operator delete(void* target, void* placement) {
+        ::operator delete(target, placement);
+    }
+};
+
+#endif
diff --git a/src/gpu/GrGeometryProcessor.cpp b/src/gpu/GrGeometryProcessor.cpp
index c707efc6a2..75e6ed8166 100644
--- a/src/gpu/GrGeometryProcessor.cpp
+++ b/src/gpu/GrGeometryProcessor.cpp
@@ -516,7 +516,7 @@ void GrPathProcessor::getInvariantOutputCoverage(GrInitInvariantOutput* out) con
     out->setKnownSingleComponent(0xff);
 }
 
-void GrPathProcessor::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void GrPathProcessor::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     PathBatchTracker* local = bt->cast<PathBatchTracker>();
     if (init.fColorIgnored) {
         local->fInputColorType = kIgnored_GrGPInput;
diff --git a/src/gpu/GrGeometryProcessor.h b/src/gpu/GrGeometryProcessor.h
index a7fea765ab..c55b9afc79 100644
--- a/src/gpu/GrGeometryProcessor.h
+++ b/src/gpu/GrGeometryProcessor.h
@@ -9,6 +9,7 @@
 #define GrGeometryProcessor_DEFINED
 
 #include "GrColor.h"
+#include "GrGeometryData.h"
 #include "GrProcessor.h"
 #include "GrShaderVar.h"
 
@@ -45,8 +46,6 @@
 /*
  * A struct for tracking batching decisions.  While this lives on GrOptState, it is managed
  * entirely by the derived classes of the GP.
- * // TODO this was an early attempt at handling out of order batching.  It should be
- * used carefully as it is being replaced by GrBatch
  */
 class GrBatchTracker {
 public:
@@ -66,24 +65,12 @@ private:
     SkAlignedSStorage<kMaxSize> fData;
 };
 
-class GrIndexBufferAllocPool;
 class GrGLCaps;
 class GrGLPrimitiveProcessor;
-class GrVertexBufferAllocPool;
+class GrOptDrawState;
 
 struct GrInitInvariantOutput;
 
-/*
- * This struct allows the GrPipeline to communicate information about the pipeline.  Most of this
- * is overrides, but some of it is general information.  Logically it should live in GrPipeline.h,
- * but this is problematic due to circular dependencies.
- */
-struct GrPipelineInfo {
-    bool fColorIgnored;
-    bool fCoverageIgnored;
-    GrColor fOverrideColor;
-    bool fUsesLocalCoords;
-};
 
 /*
  * This enum is shared by GrPrimitiveProcessors and GrGLPrimitiveProcessors to coordinate shaders
@@ -108,7 +95,17 @@ public:
     const SkMatrix& viewMatrix() const { return fViewMatrix; }
     const SkMatrix& localMatrix() const { return fLocalMatrix; }
 
-    virtual void initBatchTracker(GrBatchTracker*, const GrPipelineInfo&) const = 0;
+    /*
+     * This struct allows the optstate to communicate requirements to the GrPrimitiveProcessor.
+     */
+    struct InitBT {
+        bool fColorIgnored;
+        bool fCoverageIgnored;
+        GrColor fOverrideColor;
+        bool fUsesLocalCoords;
+    };
+
+    virtual void initBatchTracker(GrBatchTracker*, const InitBT&) const = 0;
 
     virtual bool canMakeEqual(const GrBatchTracker& mine,
                               const GrPrimitiveProcessor& that,
@@ -307,8 +304,7 @@ protected:
      * TODO this function changes quite a bit with deferred geometry.  There the GrGeometryProcessor
      * can upload a new color via attribute if needed.
      */
-    static GrGPInput GetColorInputType(GrColor* color, GrColor primitiveColor,
-                                       const GrPipelineInfo& init,
+    static GrGPInput GetColorInputType(GrColor* color, GrColor primitiveColor, const InitBT& init,
                                        bool hasVertexColor) {
         if (init.fColorIgnored) {
             *color = GrColor_ILLEGAL;
@@ -382,7 +378,7 @@ public:
         return SkNEW_ARGS(GrPathProcessor, (color, viewMatrix, localMatrix));
     }
     
-    void initBatchTracker(GrBatchTracker*, const GrPipelineInfo&) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker*, const InitBT&) const SK_OVERRIDE;
 
     bool canMakeEqual(const GrBatchTracker& mine,
                       const GrPrimitiveProcessor& that,
diff --git a/src/gpu/GrInOrderDrawBuffer.cpp b/src/gpu/GrInOrderDrawBuffer.cpp
index 1453276e49..70d61ecfff 100644
--- a/src/gpu/GrInOrderDrawBuffer.cpp
+++ b/src/gpu/GrInOrderDrawBuffer.cpp
@@ -20,8 +20,7 @@ GrInOrderDrawBuffer::GrInOrderDrawBuffer(GrGpu* gpu,
     : INHERITED(gpu, vertexPool, indexPool)
     , fCmdBuffer(kCmdBufferInitialSizeInBytes)
     , fPrevState(NULL)
-    , fDrawID(0)
-    , fBatchTarget(gpu, vertexPool, indexPool) {
+    , fDrawID(0) {
 
     SkASSERT(vertexPool);
     SkASSERT(indexPool);
@@ -211,7 +210,6 @@ int GrInOrderDrawBuffer::concatInstancedDraw(const GrPipelineBuilder& pipelineBu
     Draw* draw = static_cast<Draw*>(&fCmdBuffer.back());
 
     if (!draw->fInfo.isInstanced() ||
-        draw->fInfo.primitiveType() != info.primitiveType() ||
         draw->fInfo.verticesPerInstance() != info.verticesPerInstance() ||
         draw->fInfo.indicesPerInstance() != info.indicesPerInstance() ||
         draw->fInfo.vertexBuffer() != info.vertexBuffer() ||
@@ -268,29 +266,6 @@ void GrInOrderDrawBuffer::onDraw(const GrPipelineBuilder& pipelineBuilder,
     this->recordTraceMarkersIfNecessary();
 }
 
-void GrInOrderDrawBuffer::onDrawBatch(GrBatch* batch,
-                                      const GrPipelineBuilder& pipelineBuilder,
-                                      const GrScissorState& scissorState,
-                                      const GrDeviceCoordTexture* dstCopy) {
-    if (!this->recordStateAndShouldDraw(batch, pipelineBuilder, scissorState, dstCopy)) {
-        return;
-    }
-
-    // Check if there is a Batch Draw we can batch with
-    if (kDrawBatch_Cmd != strip_trace_bit(fCmdBuffer.back().fType)) {
-        GrNEW_APPEND_TO_RECORDER(fCmdBuffer, DrawBatch, (batch));
-        return;
-    }
-
-    DrawBatch* draw = static_cast<DrawBatch*>(&fCmdBuffer.back());
-    if (draw->fBatch->combineIfPossible(batch)) {
-        return;
-    } else {
-        GrNEW_APPEND_TO_RECORDER(fCmdBuffer, DrawBatch, (batch));
-    }
-    this->recordTraceMarkersIfNecessary();
-}
-
 void GrInOrderDrawBuffer::onStencilPath(const GrPipelineBuilder& pipelineBuilder,
                                         const GrPathProcessor* pathProc,
                                         const GrPath* path,
@@ -435,6 +410,7 @@ void GrInOrderDrawBuffer::onFlush() {
         return;
     }
 
+
     CmdBuffer::Iter iter(fCmdBuffer);
 
     int currCmdMarker = 0;
@@ -443,10 +419,6 @@ void GrInOrderDrawBuffer::onFlush() {
     // stream.
     SetState* currentState = NULL;
 
-    // TODO to prevent flushing the batch buffer too much, we only flush when wasBatch && !isBatch
-    // In the long term we can delete this and just flush once at the end of all geometry generation
-    bool wasBatch = false;
-
     while (iter.next()) {
         GrGpuTraceMarker newMarker("", -1);
         SkString traceString;
@@ -457,30 +429,13 @@ void GrInOrderDrawBuffer::onFlush() {
             ++currCmdMarker;
         }
 
-        bool isSetState = kSetState_Cmd == strip_trace_bit(iter->fType);
-
-        if (!isSetState && kDrawBatch_Cmd != strip_trace_bit(iter->fType)) {
-            // TODO see note above, this gets deleted once everyone uses batch drawing
-            if (wasBatch) {
-                wasBatch = false;
-                fBatchTarget.flush();
-            }
-        }
-
-        if (isSetState) {
+        if (kSetState_Cmd == strip_trace_bit(iter->fType)) {
             SetState* ss = reinterpret_cast<SetState*>(iter.get());
 
-            // TODO sometimes we have a prim proc, othertimes we have a GrBatch.  Eventually we will
-            // only have GrBatch and we can delete this
-            if (ss->fPrimitiveProcessor) {
-                this->getGpu()->buildProgramDesc(&ss->fDesc, *ss->fPrimitiveProcessor,
-                                                 ss->fPipeline,
-                                                 ss->fPipeline.descInfo(),
-                                                 ss->fBatchTracker);
-            } else {
-                wasBatch = true;
-            }
+            this->getGpu()->buildProgramDesc(&ss->fDesc, *ss->fPrimitiveProcessor, ss->fPipeline,
+                                             ss->fPipeline.descInfo(), ss->fBatchTracker);
             currentState = ss;
+
         } else {
             iter->execute(this, currentState);
         }
@@ -490,11 +445,6 @@ void GrInOrderDrawBuffer::onFlush() {
         }
     }
 
-    // TODO see note above, one last catch
-    if (wasBatch) {
-        fBatchTarget.flush();
-    }
-
     SkASSERT(fGpuCmdMarkers.count() == currCmdMarker);
     ++fDrawID;
 }
@@ -534,11 +484,6 @@ void GrInOrderDrawBuffer::DrawPaths::execute(GrInOrderDrawBuffer* buf, const Set
                             fCount, fStencilSettings);
 }
 
-void GrInOrderDrawBuffer::DrawBatch::execute(GrInOrderDrawBuffer* buf, const SetState* state) {
-    SkASSERT(state);
-    fBatch->generateGeometry(buf->getBatchTarget(), &state->fPipeline);
-}
-
 void GrInOrderDrawBuffer::SetState::execute(GrInOrderDrawBuffer*, const SetState*) {}
 
 void GrInOrderDrawBuffer::Clear::execute(GrInOrderDrawBuffer* buf, const SetState*) {
@@ -586,7 +531,7 @@ bool GrInOrderDrawBuffer::recordStateAndShouldDraw(const GrPipelineBuilder& pipe
     ss->fPrimitiveProcessor->initBatchTracker(&ss->fBatchTracker,
                                               ss->fPipeline.getInitBatchTracker());
 
-    if (fPrevState && fPrevState->fPrimitiveProcessor.get() &&
+    if (fPrevState &&
         fPrevState->fPrimitiveProcessor->canMakeEqual(fPrevState->fBatchTracker,
                                                       *ss->fPrimitiveProcessor,
                                                       ss->fBatchTracker) &&
@@ -599,33 +544,6 @@ bool GrInOrderDrawBuffer::recordStateAndShouldDraw(const GrPipelineBuilder& pipe
     return true;
 }
 
-bool GrInOrderDrawBuffer::recordStateAndShouldDraw(GrBatch* batch,
-                                                   const GrPipelineBuilder& pipelineBuilder,
-                                                   const GrScissorState& scissor,
-                                                   const GrDeviceCoordTexture* dstCopy) {
-    // TODO this gets much simpler when we have batches everywhere.
-    // If the previous command is also a set state, then we check to see if it has a Batch.  If so,
-    // and we can make the two batches equal, and we can combine the states, then we make them equal
-    SetState* ss = GrNEW_APPEND_TO_RECORDER(fCmdBuffer, SetState,
-                                            (batch, pipelineBuilder, *this->getGpu()->caps(), scissor,
-                                             dstCopy));
-    if (ss->fPipeline.mustSkip()) {
-        fCmdBuffer.pop_back();
-        return false;
-    }
-
-    batch->initBatchTracker(ss->fPipeline.getInitBatchTracker());
-
-    if (fPrevState && !fPrevState->fPrimitiveProcessor.get() &&
-        fPrevState->fPipeline.isEqual(ss->fPipeline)) {
-        fCmdBuffer.pop_back();
-    } else {
-        fPrevState = ss;
-        this->recordTraceMarkersIfNecessary();
-    }
-    return true;
-}
-
 void GrInOrderDrawBuffer::recordTraceMarkersIfNecessary() {
     SkASSERT(!fCmdBuffer.empty());
     SkASSERT(!cmd_has_trace_marker(fCmdBuffer.back().fType));
diff --git a/src/gpu/GrInOrderDrawBuffer.h b/src/gpu/GrInOrderDrawBuffer.h
index a421b3c449..afa7d27394 100644
--- a/src/gpu/GrInOrderDrawBuffer.h
+++ b/src/gpu/GrInOrderDrawBuffer.h
@@ -9,9 +9,6 @@
 #define GrInOrderDrawBuffer_DEFINED
 
 #include "GrFlushToGpuDrawTarget.h"
-
-#include "GrBatch.h"
-#include "GrBatchTarget.h"
 #include "GrPipeline.h"
 #include "GrPath.h"
 #include "GrTRecorder.h"
@@ -56,14 +53,13 @@ public:
 private:
     typedef GrGpu::DrawArgs DrawArgs;
     enum {
-        kDraw_Cmd              = 1,
-        kStencilPath_Cmd       = 2,
-        kSetState_Cmd          = 3,
-        kClear_Cmd             = 4,
-        kCopySurface_Cmd       = 5,
-        kDrawPath_Cmd          = 6,
-        kDrawPaths_Cmd         = 7,
-        kDrawBatch_Cmd         = 8,
+        kDraw_Cmd           = 1,
+        kStencilPath_Cmd    = 2,
+        kSetState_Cmd       = 3,
+        kClear_Cmd          = 4,
+        kCopySurface_Cmd    = 5,
+        kDrawPath_Cmd       = 6,
+        kDrawPaths_Cmd      = 7,
     };
 
     struct SetState;
@@ -184,7 +180,6 @@ private:
 
     // TODO: rename to SetPipeline once pp, batch tracker, and desc are removed
     struct SetState : public Cmd {
-        // TODO get rid of the prim proc version of this when we use batch everywhere
         SetState(const GrPipelineBuilder& pipelineBuilder, const GrPrimitiveProcessor* primProc,
                  const GrDrawTargetCaps& caps,
                  const GrScissorState& scissor, const GrDeviceCoordTexture* dstCopy)
@@ -192,13 +187,6 @@ private:
         , fPrimitiveProcessor(primProc)
         , fPipeline(pipelineBuilder, primProc, caps, scissor, dstCopy) {}
 
-        SetState(GrBatch* batch,
-                 const GrPipelineBuilder& pipelineBuilder,
-                 const GrDrawTargetCaps& caps,
-                 const GrScissorState& scissor, const GrDeviceCoordTexture* dstCopy)
-        : Cmd(kSetState_Cmd)
-        , fPipeline(batch, pipelineBuilder, caps, scissor, dstCopy) {}
-
         void execute(GrInOrderDrawBuffer*, const SetState*) SK_OVERRIDE;
 
         typedef GrPendingProgramElement<const GrPrimitiveProcessor> ProgramPrimitiveProcessor;
@@ -208,17 +196,6 @@ private:
         GrBatchTracker              fBatchTracker;
     };
 
-    struct DrawBatch : public Cmd {
-        DrawBatch(GrBatch* batch) : Cmd(kDrawBatch_Cmd), fBatch(SkRef(batch)) {
-            SkASSERT(!batch->isUsed());
-        }
-
-        void execute(GrInOrderDrawBuffer*, const SetState*) SK_OVERRIDE;
-
-        // TODO it wouldn't be too hard to let batches allocate in the cmd buffer
-        SkAutoTUnref<GrBatch>  fBatch;
-    };
-
     typedef void* TCmdAlign; // This wouldn't be enough align if a command used long double.
     typedef GrTRecorder<Cmd, TCmdAlign> CmdBuffer;
 
@@ -231,10 +208,6 @@ private:
                 const DrawInfo&,
                 const GrScissorState&,
                 const GrDeviceCoordTexture* dstCopy) SK_OVERRIDE;
-    void onDrawBatch(GrBatch*,
-                     const GrPipelineBuilder&,
-                     const GrScissorState&,
-                     const GrDeviceCoordTexture* dstCopy) SK_OVERRIDE;
     void onDrawRect(GrPipelineBuilder*,
                     GrColor,
                     const SkMatrix& viewMatrix,
@@ -280,16 +253,10 @@ private:
     // Determines whether the current draw operation requires a new GrPipeline and if so
     // records it. If the draw can be skipped false is returned and no new GrPipeline is
     // recorded.
-    // TODO delete the primproc variant when we have batches everywhere
     bool SK_WARN_UNUSED_RESULT recordStateAndShouldDraw(const GrPipelineBuilder&,
                                                         const GrPrimitiveProcessor*,
                                                         const GrScissorState&,
                                                         const GrDeviceCoordTexture*);
-    bool SK_WARN_UNUSED_RESULT recordStateAndShouldDraw(GrBatch*,
-                                                        const GrPipelineBuilder&,
-                                                        const GrScissorState&,
-                                                        const GrDeviceCoordTexture*);
-
     // We lazily record clip changes in order to skip clips that have no effect.
     void recordClipIfNecessary();
     // Records any trace markers for a command after adding it to the buffer.
@@ -297,8 +264,6 @@ private:
 
     bool isIssued(uint32_t drawID) SK_OVERRIDE { return drawID != fDrawID; }
 
-    GrBatchTarget* getBatchTarget() { return &fBatchTarget; }
-
     // TODO: Use a single allocator for commands and records
     enum {
         kCmdBufferInitialSizeInBytes = 8 * 1024,
@@ -312,7 +277,6 @@ private:
     SkTDArray<char>                     fPathIndexBuffer;
     SkTDArray<float>                    fPathTransformBuffer;
     uint32_t                            fDrawID;
-    GrBatchTarget                       fBatchTarget;
 
     typedef GrFlushToGpuDrawTarget INHERITED;
 };
diff --git a/src/gpu/GrOvalRenderer.cpp b/src/gpu/GrOvalRenderer.cpp
index b1ef8d34bd..4b24e4542b 100644
--- a/src/gpu/GrOvalRenderer.cpp
+++ b/src/gpu/GrOvalRenderer.cpp
@@ -166,7 +166,7 @@ public:
         return SkNEW_ARGS(GLProcessor, (*this, bt));
     }
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE {
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE {
         BatchTracker* local = bt->cast<BatchTracker>();
         local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
         local->fUsesLocalCoords = init.fUsesLocalCoords;
@@ -365,7 +365,7 @@ public:
         return SkNEW_ARGS(GLProcessor, (*this, bt));
     }
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE {
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE {
         BatchTracker* local = bt->cast<BatchTracker>();
         local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
         local->fUsesLocalCoords = init.fUsesLocalCoords;
@@ -584,7 +584,7 @@ public:
         return SkNEW_ARGS(GLProcessor, (*this, bt));
     }
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE {
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE {
         BatchTracker* local = bt->cast<BatchTracker>();
         local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
         local->fUsesLocalCoords = init.fUsesLocalCoords;
diff --git a/src/gpu/GrPipeline.cpp b/src/gpu/GrPipeline.cpp
index 911169e336..c35808e033 100644
--- a/src/gpu/GrPipeline.cpp
+++ b/src/gpu/GrPipeline.cpp
@@ -7,7 +7,6 @@
 
 #include "GrPipeline.h"
 
-#include "GrBatch.h"
 #include "GrDrawTargetCaps.h"
 #include "GrGpu.h"
 #include "GrPipelineBuilder.h"
@@ -15,37 +14,13 @@
 #include "GrXferProcessor.h"
 
 GrPipeline::GrPipeline(const GrPipelineBuilder& pipelineBuilder,
-                       const GrPrimitiveProcessor* primProc,
-                       const GrDrawTargetCaps& caps,
-                       const GrScissorState& scissorState,
-                       const GrDeviceCoordTexture* dstCopy) {
+                               const GrPrimitiveProcessor* primProc,
+                               const GrDrawTargetCaps& caps,
+                               const GrScissorState& scissorState,
+                               const GrDeviceCoordTexture* dstCopy) {
     const GrProcOptInfo& colorPOI = pipelineBuilder.colorProcInfo(primProc);
     const GrProcOptInfo& coveragePOI = pipelineBuilder.coverageProcInfo(primProc);
 
-    this->internalConstructor(pipelineBuilder, colorPOI, coveragePOI, caps, scissorState, dstCopy);
-}
-
-GrPipeline::GrPipeline(GrBatch* batch,
-                       const GrPipelineBuilder& pipelineBuilder,
-                       const GrDrawTargetCaps& caps,
-                       const GrScissorState& scissorState,
-                       const GrDeviceCoordTexture* dstCopy) {
-    GrBatchOpt batchOpt;
-    batchOpt.fCanTweakAlphaForCoverage = pipelineBuilder.canTweakAlphaForCoverage();
-    batch->initBatchOpt(batchOpt);
-
-    const GrProcOptInfo& colorPOI = pipelineBuilder.colorProcInfo(batch);
-    const GrProcOptInfo& coveragePOI = pipelineBuilder.coverageProcInfo(batch);
-
-    this->internalConstructor(pipelineBuilder, colorPOI, coveragePOI, caps, scissorState, dstCopy);
-}
-
-void GrPipeline::internalConstructor(const GrPipelineBuilder& pipelineBuilder,
-                                         const GrProcOptInfo& colorPOI,
-                                         const GrProcOptInfo& coveragePOI,
-                                         const GrDrawTargetCaps& caps,
-                                         const GrScissorState& scissorState,
-                                         const GrDeviceCoordTexture* dstCopy) {
     // Create XferProcessor from DS's XPFactory
     SkAutoTUnref<GrXferProcessor> xferProcessor(
         pipelineBuilder.getXPFactory()->createXferProcessor(colorPOI, coveragePOI));
diff --git a/src/gpu/GrPipeline.h b/src/gpu/GrPipeline.h
index 24d0b62795..df654ba407 100644
--- a/src/gpu/GrPipeline.h
+++ b/src/gpu/GrPipeline.h
@@ -17,8 +17,8 @@
 #include "SkMatrix.h"
 #include "SkRefCnt.h"
 
-class GrBatch;
 class GrDeviceCoordTexture;
+class GrPathProcessor;
 class GrPipelineBuilder;
 
 /**
@@ -29,14 +29,10 @@ class GrPipeline {
 public:
     SK_DECLARE_INST_COUNT(GrPipeline)
 
-    // TODO get rid of this version of the constructor when we use batch everywhere
     GrPipeline(const GrPipelineBuilder& pipelineBuilder, const GrPrimitiveProcessor*,
                const GrDrawTargetCaps&, const GrScissorState&,
                const GrDeviceCoordTexture* dstCopy);
 
-    GrPipeline(GrBatch*, const GrPipelineBuilder&, const GrDrawTargetCaps&,
-               const GrScissorState&, const GrDeviceCoordTexture* dstCopy);
-
     /*
      * Returns true if it is possible to combine the two GrPipelines and it will update 'this'
      * to subsume 'that''s draw.
@@ -136,17 +132,9 @@ public:
 
     const GrProgramDesc::DescInfo& descInfo() const { return fDescInfo; }
 
-    const GrPipelineInfo& getInitBatchTracker() const { return fInitBT; }
+    const GrGeometryProcessor::InitBT& getInitBatchTracker() const { return fInitBT; }
 
 private:
-    // TODO we can have one constructor once GrBatch is complete
-    void internalConstructor(const GrPipelineBuilder&,
-                             const GrProcOptInfo& colorPOI,
-                             const GrProcOptInfo& coveragePOI,
-                             const GrDrawTargetCaps&,
-                             const GrScissorState&,
-                             const GrDeviceCoordTexture* dstCopy);
-
     /**
      * Alter the program desc and inputs (attribs and processors) based on the blend optimization.
      */
@@ -176,13 +164,13 @@ private:
     RenderTarget                        fRenderTarget;
     GrScissorState                      fScissorState;
     GrStencilSettings                   fStencilSettings;
-    GrPipelineBuilder::DrawFace         fDrawFace;
+    GrPipelineBuilder::DrawFace               fDrawFace;
     GrDeviceCoordTexture                fDstCopy;
     uint32_t                            fFlags;
     ProgramXferProcessor                fXferProcessor;
     FragmentStageArray                  fFragmentStages;
     GrProgramDesc::DescInfo             fDescInfo;
-    GrPipelineInfo                      fInitBT;
+    GrGeometryProcessor::InitBT         fInitBT;
 
     // This function is equivalent to the offset into fFragmentStages where coverage stages begin.
     int                                 fNumColorStages;
diff --git a/src/gpu/GrPipelineBuilder.cpp b/src/gpu/GrPipelineBuilder.cpp
index 90aee748c9..f329c05b91 100644
--- a/src/gpu/GrPipelineBuilder.cpp
+++ b/src/gpu/GrPipelineBuilder.cpp
@@ -20,7 +20,9 @@ GrPipelineBuilder::GrPipelineBuilder()
     , fColorProcInfoValid(false)
     , fCoverageProcInfoValid(false)
     , fColorCache(GrColor_ILLEGAL)
-    , fCoverageCache(GrColor_ILLEGAL) {
+    , fCoverageCache(GrColor_ILLEGAL)
+    , fColorPrimProc(NULL)
+    , fCoveragePrimProc(NULL) {
     SkDEBUGCODE(fBlockEffectRemovalCnt = 0;)
 }
 
@@ -37,6 +39,8 @@ GrPipelineBuilder& GrPipelineBuilder::operator=(const GrPipelineBuilder& that) {
     fCoverageProcInfoValid = that.fCoverageProcInfoValid;
     fColorCache = that.fColorCache;
     fCoverageCache = that.fCoverageCache;
+    fColorPrimProc = that.fColorPrimProc;
+    fCoveragePrimProc = that.fCoveragePrimProc;
     if (fColorProcInfoValid) {
         fColorProcInfo = that.fColorProcInfo;
     }
@@ -80,6 +84,9 @@ void GrPipelineBuilder::setFromPaint(const GrPaint& paint, GrRenderTarget* rt) {
 
     fColorCache = GrColor_ILLEGAL;
     fCoverageCache = GrColor_ILLEGAL;
+
+    fColorPrimProc = NULL;
+    fCoveragePrimProc = NULL;
 }
 
 ////////////////////////////////////////////////////////////////////////////////
@@ -154,29 +161,22 @@ bool GrPipelineBuilder::willBlendWithDst(const GrPrimitiveProcessor* pp) const {
 }
 
 void GrPipelineBuilder::calcColorInvariantOutput(const GrPrimitiveProcessor* pp) const {
-    fColorProcInfo.calcColorWithPrimProc(pp, fColorStages.begin(), this->numColorStages());
-    fColorProcInfoValid = false;
-
+    if (!fColorProcInfoValid || fColorPrimProc != pp) {
+        fColorProcInfo.calcColorWithPrimProc(pp, fColorStages.begin(), this->numColorStages());
+        fColorProcInfoValid = true;
+        fColorPrimProc = pp;
+    }
 }
 
 void GrPipelineBuilder::calcCoverageInvariantOutput(const GrPrimitiveProcessor* pp) const {
-    fCoverageProcInfo.calcCoverageWithPrimProc(pp, fCoverageStages.begin(),
-                                               this->numCoverageStages());
-    fCoverageProcInfoValid = false;
-}
-
-void GrPipelineBuilder::calcColorInvariantOutput(const GrBatch* batch) const {
-    fColorProcInfo.calcColorWithBatch(batch, fColorStages.begin(), this->numColorStages());
-    fColorProcInfoValid = false;
-}
-
-void GrPipelineBuilder::calcCoverageInvariantOutput(const GrBatch* batch) const {
-    fCoverageProcInfo.calcCoverageWithBatch(batch, fCoverageStages.begin(),
-                                            this->numCoverageStages());
-    fCoverageProcInfoValid = false;
+    if (!fCoverageProcInfoValid ||  fCoveragePrimProc != pp) {
+        fCoverageProcInfo.calcCoverageWithPrimProc(pp, fCoverageStages.begin(),
+                                                   this->numCoverageStages());
+        fCoverageProcInfoValid = true;
+        fCoveragePrimProc = pp;
+    }
 }
 
-
 void GrPipelineBuilder::calcColorInvariantOutput(GrColor color) const {
     if (!fColorProcInfoValid || color != fColorCache) {
         GrColorComponentFlags flags = kRGBA_GrColorComponentFlags;
diff --git a/src/gpu/GrPipelineBuilder.h b/src/gpu/GrPipelineBuilder.h
index df9a1c85ee..9579024d6a 100644
--- a/src/gpu/GrPipelineBuilder.h
+++ b/src/gpu/GrPipelineBuilder.h
@@ -8,7 +8,7 @@
 #ifndef GrPipelineBuilder_DEFINED
 #define GrPipelineBuilder_DEFINED
 
-#include "GrBatch.h"
+
 #include "GrBlend.h"
 #include "GrDrawTargetCaps.h"
 #include "GrGeometryProcessor.h"
@@ -391,15 +391,6 @@ public:
     GrPipelineBuilder& operator= (const GrPipelineBuilder& that);
 
 private:
-    // Calculating invariant color / coverage information is expensive, so we partially cache the
-    // results.
-    //
-    // canUseFracCoveragePrimProc() - Called in regular skia draw, caches results but only for a
-    //                                specific color and coverage.  May be called multiple times
-    // willBlendWithDst() - only called by Nvpr, does not cache results
-    // GrOptDrawState constructor - never caches results
-
-    // TODO delete when we have Batch
     const GrProcOptInfo& colorProcInfo(const GrPrimitiveProcessor* pp) const {
         this->calcColorInvariantOutput(pp);
         return fColorProcInfo;
@@ -410,28 +401,17 @@ private:
         return fCoverageProcInfo;
     }
 
-    const GrProcOptInfo& colorProcInfo(const GrBatch* batch) const {
-        this->calcColorInvariantOutput(batch);
-        return fColorProcInfo;
-    }
-
-    const GrProcOptInfo& coverageProcInfo(const GrBatch* batch) const {
-        this->calcCoverageInvariantOutput(batch);
-        return fCoverageProcInfo;
-    }
-
     /**
-     * Primproc variants of the calc functions
-     * TODO remove these when batch is everywhere
+     * If fColorProcInfoValid is false, function calculates the invariant output for the color
+     * stages and results are stored in fColorProcInfo.
      */
     void calcColorInvariantOutput(const GrPrimitiveProcessor*) const;
-    void calcCoverageInvariantOutput(const GrPrimitiveProcessor*) const;
 
     /**
-     * GrBatch provides the initial seed for these loops based off of its initial geometry data
+     * If fCoverageProcInfoValid is false, function calculates the invariant output for the coverage
+     * stages and results are stored in fCoverageProcInfo.
      */
-    void calcColorInvariantOutput(const GrBatch*) const;
-    void calcCoverageInvariantOutput(const GrBatch*) const;
+    void calcCoverageInvariantOutput(const GrPrimitiveProcessor*) const;
 
     /**
      * If fColorProcInfoValid is false, function calculates the invariant output for the color
@@ -465,6 +445,8 @@ private:
     mutable bool fCoverageProcInfoValid;
     mutable GrColor fColorCache;
     mutable GrColor fCoverageCache;
+    mutable const GrPrimitiveProcessor* fColorPrimProc;
+    mutable const GrPrimitiveProcessor* fCoveragePrimProc;
 
     friend class GrPipeline;
 };
diff --git a/src/gpu/GrProcOptInfo.cpp b/src/gpu/GrProcOptInfo.cpp
index dc499fa337..a350ef783d 100644
--- a/src/gpu/GrProcOptInfo.cpp
+++ b/src/gpu/GrProcOptInfo.cpp
@@ -7,29 +7,10 @@
 
 #include "GrProcOptInfo.h"
 
-#include "GrBatch.h"
 #include "GrFragmentProcessor.h"
 #include "GrFragmentStage.h"
 #include "GrGeometryProcessor.h"
 
-void GrProcOptInfo::calcColorWithBatch(const GrBatch* batch,
-                                       const GrFragmentStage* stages,
-                                       int stageCount) {
-    GrInitInvariantOutput out;
-    batch->getInvariantOutputColor(&out);
-    fInOut.reset(out);
-    this->internalCalc(stages, stageCount, batch->willReadFragmentPosition());
-}
-
-void GrProcOptInfo::calcCoverageWithBatch(const GrBatch* batch,
-                                          const GrFragmentStage* stages,
-                                          int stageCount) {
-    GrInitInvariantOutput out;
-    batch->getInvariantOutputCoverage(&out);
-    fInOut.reset(out);
-    this->internalCalc(stages, stageCount, batch->willReadFragmentPosition());
-}
-
 void GrProcOptInfo::calcColorWithPrimProc(const GrPrimitiveProcessor* primProc,
                                           const GrFragmentStage* stages,
                                           int stageCount) {
diff --git a/src/gpu/GrProcOptInfo.h b/src/gpu/GrProcOptInfo.h
index 6e8f615d2c..75d88c64e2 100644
--- a/src/gpu/GrProcOptInfo.h
+++ b/src/gpu/GrProcOptInfo.h
@@ -11,7 +11,6 @@
 #include "GrColor.h"
 #include "GrInvariantOutput.h"
 
-class GrBatch;
 class GrFragmentStage;
 class GrFragmentProcessor;
 class GrPrimitiveProcessor;
@@ -34,10 +33,6 @@ public:
     void calcWithInitialValues(const GrFragmentStage*, int stageCount, GrColor startColor,
                                GrColorComponentFlags flags, bool areCoverageStages);
 
-    void calcColorWithBatch(const GrBatch*, const GrFragmentStage*, int stagecount);
-    void calcCoverageWithBatch(const GrBatch*, const GrFragmentStage*, int stagecount);
-
-    // TODO delete these when batch is everywhere
     void calcColorWithPrimProc(const GrPrimitiveProcessor*, const GrFragmentStage*, int stagecount);
     void calcCoverageWithPrimProc(const GrPrimitiveProcessor*, const GrFragmentStage*,
                                   int stagecount);
diff --git a/src/gpu/GrProcessor.cpp b/src/gpu/GrProcessor.cpp
index 08f437d50a..9977018191 100644
--- a/src/gpu/GrProcessor.cpp
+++ b/src/gpu/GrProcessor.cpp
@@ -8,6 +8,7 @@
 #include "GrProcessor.h"
 #include "GrContext.h"
 #include "GrCoordTransform.h"
+#include "GrGeometryData.h"
 #include "GrGeometryProcessor.h"
 #include "GrInvariantOutput.h"
 #include "GrMemoryPool.h"
@@ -171,6 +172,19 @@ void GrFragmentProcessor::computeInvariantOutput(GrInvariantOutput* inout) const
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////
 
+/*
+ * GrGeometryData shares the same pool so it lives in this file too
+ */
+void* GrGeometryData::operator new(size_t size) {
+    return GrProcessor_Globals::GetTLS()->allocate(size);
+}
+
+void GrGeometryData::operator delete(void* target) {
+    GrProcessor_Globals::GetTLS()->release(target);
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////////////
+
 // Initial static variable from GrXPFactory
 int32_t GrXPFactory::gCurrXPFClassID =
         GrXPFactory::kIllegalXPFClassID;
diff --git a/src/gpu/effects/GrBezierEffect.cpp b/src/gpu/effects/GrBezierEffect.cpp
index 37e1dff2c5..1c6d469033 100644
--- a/src/gpu/effects/GrBezierEffect.cpp
+++ b/src/gpu/effects/GrBezierEffect.cpp
@@ -208,7 +208,7 @@ bool GrConicEffect::onIsEqual(const GrGeometryProcessor& other) const {
     return (ce.fEdgeType == fEdgeType);
 }
 
-void GrConicEffect::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void GrConicEffect::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     ConicBatchTracker* local = bt->cast<ConicBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
     local->fCoverageScale = fCoverageScale;
@@ -432,7 +432,7 @@ bool GrQuadEffect::onIsEqual(const GrGeometryProcessor& other) const {
     return (ce.fEdgeType == fEdgeType);
 }
 
-void GrQuadEffect::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void GrQuadEffect::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     QuadBatchTracker* local = bt->cast<QuadBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
     local->fCoverageScale = fCoverageScale;
@@ -677,7 +677,7 @@ bool GrCubicEffect::onIsEqual(const GrGeometryProcessor& other) const {
     return (ce.fEdgeType == fEdgeType);
 }
 
-void GrCubicEffect::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void GrCubicEffect::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     CubicBatchTracker* local = bt->cast<CubicBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
     local->fUsesLocalCoords = init.fUsesLocalCoords;
diff --git a/src/gpu/effects/GrBezierEffect.h b/src/gpu/effects/GrBezierEffect.h
index b0039e955f..020286fddb 100644
--- a/src/gpu/effects/GrBezierEffect.h
+++ b/src/gpu/effects/GrBezierEffect.h
@@ -105,7 +105,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker*, const GrPipelineInfo&) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker*, const InitBT&) const SK_OVERRIDE;
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
                         const GrBatchTracker&) const SK_OVERRIDE;
@@ -190,7 +190,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker*, const GrPipelineInfo&) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker*, const InitBT&) const SK_OVERRIDE;
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
                         const GrBatchTracker&) const SK_OVERRIDE;
@@ -271,7 +271,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker*, const GrPipelineInfo&) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker*, const InitBT&) const SK_OVERRIDE;
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
                         const GrBatchTracker&) const SK_OVERRIDE;
diff --git a/src/gpu/effects/GrBitmapTextGeoProc.cpp b/src/gpu/effects/GrBitmapTextGeoProc.cpp
index 313326561c..924f7cd952 100644
--- a/src/gpu/effects/GrBitmapTextGeoProc.cpp
+++ b/src/gpu/effects/GrBitmapTextGeoProc.cpp
@@ -145,7 +145,7 @@ GrBitmapTextGeoProc::createGLInstance(const GrBatchTracker& bt,
     return SkNEW_ARGS(GrGLBitmapTextGeoProc, (*this, bt));
 }
 
-void GrBitmapTextGeoProc::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void GrBitmapTextGeoProc::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     BitmapTextBatchTracker* local = bt->cast<BitmapTextBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init,
                                                SkToBool(fInColor));
diff --git a/src/gpu/effects/GrBitmapTextGeoProc.h b/src/gpu/effects/GrBitmapTextGeoProc.h
index d5684d746e..d73fba83e5 100644
--- a/src/gpu/effects/GrBitmapTextGeoProc.h
+++ b/src/gpu/effects/GrBitmapTextGeoProc.h
@@ -43,7 +43,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps& caps) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker*, const GrPipelineInfo&) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker*, const InitBT&) const SK_OVERRIDE;
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
                         const GrBatchTracker&) const SK_OVERRIDE;
diff --git a/src/gpu/effects/GrDashingEffect.cpp b/src/gpu/effects/GrDashingEffect.cpp
index 34e03011b2..4a034060de 100644
--- a/src/gpu/effects/GrDashingEffect.cpp
+++ b/src/gpu/effects/GrDashingEffect.cpp
@@ -501,7 +501,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker&,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE;
 
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
@@ -706,7 +706,7 @@ bool DashingCircleEffect::onIsEqual(const GrGeometryProcessor& other) const {
             fCenterX == dce.fCenterX);
 }
 
-void DashingCircleEffect::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void DashingCircleEffect::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     DashingCircleBatchTracker* local = bt->cast<DashingCircleBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
     local->fUsesLocalCoords = init.fUsesLocalCoords;
@@ -795,7 +795,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE;
 
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
@@ -1013,7 +1013,7 @@ bool DashingLineEffect::onIsEqual(const GrGeometryProcessor& other) const {
             fIntervalLength == de.fIntervalLength);
 }
 
-void DashingLineEffect::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void DashingLineEffect::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     DashingLineBatchTracker* local = bt->cast<DashingLineBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
     local->fUsesLocalCoords = init.fUsesLocalCoords;
diff --git a/src/gpu/effects/GrDistanceFieldTextureEffect.cpp b/src/gpu/effects/GrDistanceFieldTextureEffect.cpp
index 96dea3eb1c..28220a72ec 100755
--- a/src/gpu/effects/GrDistanceFieldTextureEffect.cpp
+++ b/src/gpu/effects/GrDistanceFieldTextureEffect.cpp
@@ -252,7 +252,7 @@ GrDistanceFieldTextureEffect::createGLInstance(const GrBatchTracker& bt,
     return SkNEW_ARGS(GrGLDistanceFieldTextureEffect, (*this, bt));
 }
 
-void GrDistanceFieldTextureEffect::initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const {
+void GrDistanceFieldTextureEffect::initBatchTracker(GrBatchTracker* bt, const InitBT& init) const {
     DistanceFieldBatchTracker* local = bt->cast<DistanceFieldBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init,
                                                SkToBool(fInColor));
@@ -508,7 +508,7 @@ GrDistanceFieldNoGammaTextureEffect::createGLInstance(const GrBatchTracker& bt,
 }
 
 void GrDistanceFieldNoGammaTextureEffect::initBatchTracker(GrBatchTracker* bt,
-                                                           const GrPipelineInfo& init) const {
+                                                           const InitBT& init) const {
     DistanceFieldNoGammaBatchTracker* local = bt->cast<DistanceFieldNoGammaBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init,
                                                SkToBool(fInColor));
@@ -824,7 +824,7 @@ GrDistanceFieldLCDTextureEffect::createGLInstance(const GrBatchTracker& bt,
 }
 
 void GrDistanceFieldLCDTextureEffect::initBatchTracker(GrBatchTracker* bt,
-                                                       const GrPipelineInfo& init) const {
+                                                       const InitBT& init) const {
     DistanceFieldLCDBatchTracker* local = bt->cast<DistanceFieldLCDBatchTracker>();
     local->fInputColorType = GetColorInputType(&local->fColor, this->color(), init, false);
     local->fUsesLocalCoords = init.fUsesLocalCoords;
diff --git a/src/gpu/effects/GrDistanceFieldTextureEffect.h b/src/gpu/effects/GrDistanceFieldTextureEffect.h
index 33209a5254..5a99d24fb1 100644
--- a/src/gpu/effects/GrDistanceFieldTextureEffect.h
+++ b/src/gpu/effects/GrDistanceFieldTextureEffect.h
@@ -83,7 +83,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE;
 
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
@@ -148,7 +148,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE;
 
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,
@@ -206,7 +206,7 @@ public:
     virtual GrGLPrimitiveProcessor* createGLInstance(const GrBatchTracker& bt,
                                                      const GrGLCaps&) const SK_OVERRIDE;
 
-    void initBatchTracker(GrBatchTracker* bt, const GrPipelineInfo& init) const SK_OVERRIDE;
+    void initBatchTracker(GrBatchTracker* bt, const InitBT& init) const SK_OVERRIDE;
 
     bool onCanMakeEqual(const GrBatchTracker&,
                         const GrGeometryProcessor&,