Use transfer buffer for BatchAtlas texture copies.

Sets up use of transfer buffer (if available) to do one-copy transfers.
Get transfer buffers working properly in GL.
Implement GrVkGpu::onTransferPixels.
Check caps to ensure we can create a transfer buffer.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1888473002

Review URL: https://codereview.chromium.org/1888473002

13 files changed, 209 insertions, 52 deletions

diff --git a/src/gpu/GrBatchAtlas.cpp b/src/gpu/GrBatchAtlas.cpp
index 40ab0e6c0e..17e82ef1fe 100644
--- a/src/gpu/GrBatchAtlas.cpp
+++ b/src/gpu/GrBatchAtlas.cpp
@@ -7,19 +7,23 @@
 
 #include "GrBatchAtlas.h"
 #include "GrBatchFlushState.h"
+#include "GrBuffer.h"
 #include "GrRectanizer.h"
+#include "GrResourceProvider.h"
 #include "GrTracing.h"
 
 ////////////////////////////////////////////////////////////////////////////////
 
 GrBatchAtlas::BatchPlot::BatchPlot(int index, uint64_t genID, int offX, int offY, int width,
-                                   int height, GrPixelConfig config)
-    : fLastUpload(GrBatchDrawToken::AlreadyFlushedToken())
+                                   int height, GrPixelConfig config, GrResourceProvider* rp)
+    : fResourceProvider(rp)
+    , fLastUpload(GrBatchDrawToken::AlreadyFlushedToken())
     , fLastUse(GrBatchDrawToken::AlreadyFlushedToken())
     , fIndex(index)
     , fGenID(genID)
     , fID(CreateId(fIndex, fGenID))
-    , fData(nullptr)
+    , fDataPtr(nullptr)
+    , fTransferBuffer(nullptr)
     , fWidth(width)
     , fHeight(height)
     , fX(offX)
@@ -36,7 +40,11 @@ GrBatchAtlas::BatchPlot::BatchPlot(int index, uint64_t genID, int offX, int offY
 }
 
 GrBatchAtlas::BatchPlot::~BatchPlot() {
-    sk_free(fData);
+    if (fTransferBuffer) {
+        fTransferBuffer->unref();
+    } else {
+        sk_free(fDataPtr);
+    }
     delete fRects;
 }
 
@@ -52,14 +60,26 @@ bool GrBatchAtlas::BatchPlot::addSubImage(int width, int height, const void* ima
         return false;
     }
 
-    if (!fData) {
-        fData = reinterpret_cast<unsigned char*>(sk_calloc_throw(fBytesPerPixel * fWidth *
-                                                                 fHeight));
+    if (!fDataPtr) {
+        if (!fTransferBuffer) {
+            fTransferBuffer = 
+                fResourceProvider->createBuffer(fBytesPerPixel * fWidth * fHeight,
+                                                kXferCpuToGpu_GrBufferType,
+                                                kDynamic_GrAccessPattern,
+                                                GrResourceProvider::kNoPendingIO_Flag);
+        }
+        if (fTransferBuffer) {
+            fDataPtr = (unsigned char*)fTransferBuffer->map();
+        } else {
+            fDataPtr = reinterpret_cast<unsigned char*>(sk_calloc_throw(fBytesPerPixel * fWidth *
+                                                                        fHeight));
+        }
     }
+
     size_t rowBytes = width * fBytesPerPixel;
     const unsigned char* imagePtr = (const unsigned char*)image;
     // point ourselves at the right starting spot
-    unsigned char* dataPtr = fData;
+    unsigned char* dataPtr = fDataPtr;
     dataPtr += fBytesPerPixel * fWidth * loc->fY;
     dataPtr += fBytesPerPixel * loc->fX;
     // copy into the data buffer
@@ -79,16 +99,25 @@ bool GrBatchAtlas::BatchPlot::addSubImage(int width, int height, const void* ima
 }
 
 void GrBatchAtlas::BatchPlot::uploadToTexture(GrDrawBatch::WritePixelsFn& writePixels,
+                                              GrDrawBatch::TransferPixelsFn& xferPixels,
                                               GrTexture* texture) {
     // We should only be issuing uploads if we are in fact dirty
-    SkASSERT(fDirty && fData && texture);
+    SkASSERT(fDirty && fDataPtr && texture);
     TRACE_EVENT0(TRACE_DISABLED_BY_DEFAULT("skia.gpu"), "GrBatchPlot::uploadToTexture");
     size_t rowBytes = fBytesPerPixel * fWidth;
-    const unsigned char* dataPtr = fData;
-    dataPtr += rowBytes * fDirtyRect.fTop;
-    dataPtr += fBytesPerPixel * fDirtyRect.fLeft;
-    writePixels(texture, fOffset.fX + fDirtyRect.fLeft, fOffset.fY + fDirtyRect.fTop,
-                fDirtyRect.width(), fDirtyRect.height(), fConfig, dataPtr, rowBytes);
+    size_t dataOffset = rowBytes * fDirtyRect.fTop + fBytesPerPixel * fDirtyRect.fLeft;
+    if (fTransferBuffer) {
+        fTransferBuffer->unmap();
+        fDataPtr = nullptr;
+        xferPixels(texture, fOffset.fX + fDirtyRect.fLeft, fOffset.fY + fDirtyRect.fTop,
+                   fDirtyRect.width(), fDirtyRect.height(), fConfig, fTransferBuffer, dataOffset, 
+                   rowBytes);
+    } else {
+        writePixels(texture, fOffset.fX + fDirtyRect.fLeft, fOffset.fY + fDirtyRect.fTop,
+                    fDirtyRect.width(), fDirtyRect.height(), fConfig, fDataPtr + dataOffset, 
+                    rowBytes);
+    }
+
     fDirtyRect.setEmpty();
     SkDEBUGCODE(fDirty = false;)
 }
@@ -102,8 +131,8 @@ void GrBatchAtlas::BatchPlot::resetRects() {
     fID = CreateId(fIndex, fGenID);
 
     // zero out the plot
-    if (fData) {
-        sk_bzero(fData, fBytesPerPixel * fWidth * fHeight);
+    if (fDataPtr) {
+        sk_bzero(fDataPtr, fBytesPerPixel * fWidth * fHeight);
     }
 
     fDirtyRect.setEmpty();
@@ -112,7 +141,7 @@ void GrBatchAtlas::BatchPlot::resetRects() {
 
 ///////////////////////////////////////////////////////////////////////////////
 
-GrBatchAtlas::GrBatchAtlas(GrTexture* texture, int numPlotsX, int numPlotsY)
+GrBatchAtlas::GrBatchAtlas(GrResourceProvider* rp, GrTexture* texture, int numPlotsX, int numPlotsY)
     : fTexture(texture)
     , fAtlasGeneration(kInvalidAtlasGeneration + 1) {
 
@@ -135,7 +164,7 @@ GrBatchAtlas::GrBatchAtlas(GrTexture* texture, int numPlotsX, int numPlotsY)
         for (int x = numPlotsX - 1, c = 0; x >= 0; --x, ++c) {
             uint32_t index = r * numPlotsX + c;
             currPlot->reset(new BatchPlot(index, 1, x, y, plotWidth, plotHeight,
-                                          texture->desc().fConfig));
+                                          texture->desc().fConfig, rp));
 
             // build LRU list
             fPlotList.addToHead(currPlot->get());
@@ -162,12 +191,13 @@ inline void GrBatchAtlas::updatePlot(GrDrawBatch::Target* target, AtlasID* id, B
     // upload. Otherwise, we already have a scheduled upload that hasn't yet ocurred.
     // This new update will piggy back on that previously scheduled update.
     if (target->hasDrawBeenFlushed(plot->lastUploadToken())) {
-        // With c+14 we could move sk_sp into lamba to only ref once.
+        // With c+14 we could move sk_sp into lambda to only ref once.
         sk_sp<BatchPlot> plotsp(SkRef(plot));
         GrTexture* texture = fTexture;
         GrBatchDrawToken lastUploadToken = target->addAsapUpload(
-            [plotsp, texture] (GrDrawBatch::WritePixelsFn& writePixels) {
-               plotsp->uploadToTexture(writePixels, texture);
+            [plotsp, texture] (GrDrawBatch::WritePixelsFn& writePixels,
+                               GrDrawBatch::TransferPixelsFn& transferPixels) {
+                plotsp->uploadToTexture(writePixels, transferPixels, texture);
             }
         );
         plot->setLastUploadToken(lastUploadToken);
@@ -235,8 +265,9 @@ bool GrBatchAtlas::addToAtlas(AtlasID* id, GrDrawBatch::Target* target,
     sk_sp<BatchPlot> plotsp(SkRef(newPlot.get()));
     GrTexture* texture = fTexture;
     GrBatchDrawToken lastUploadToken = target->addInlineUpload(
-        [plotsp, texture] (GrDrawBatch::WritePixelsFn& writePixels) {
-            plotsp->uploadToTexture(writePixels, texture);
+        [plotsp, texture] (GrDrawBatch::WritePixelsFn& writePixels,
+                           GrDrawBatch::TransferPixelsFn& transferPixels) {
+            plotsp->uploadToTexture(writePixels, transferPixels, texture);
         }
     );
     newPlot->setLastUploadToken(lastUploadToken);
diff --git a/src/gpu/GrBatchAtlas.h b/src/gpu/GrBatchAtlas.h
index 707f463d97..418739fee6 100644
--- a/src/gpu/GrBatchAtlas.h
+++ b/src/gpu/GrBatchAtlas.h
@@ -16,6 +16,8 @@
 #include "batches/GrDrawBatch.h"
 
 class GrRectanizer;
+class GrResourceProvider;
+class GrBuffer;
 
 struct GrBatchAtlasConfig {
     int numPlotsX() const { return fWidth / fPlotWidth; }
@@ -41,7 +43,7 @@ public:
     // the eviction
     typedef void (*EvictionFunc)(GrBatchAtlas::AtlasID, void*);
 
-    GrBatchAtlas(GrTexture*, int numPlotsX, int numPlotsY);
+    GrBatchAtlas(GrResourceProvider*, GrTexture*, int numPlotsX, int numPlotsY);
     ~GrBatchAtlas();
 
     // Adds a width x height subimage to the atlas. Upon success it returns
@@ -172,19 +174,21 @@ private:
         void setLastUploadToken(GrBatchDrawToken batchToken) { fLastUpload = batchToken; }
         void setLastUseToken(GrBatchDrawToken batchToken) { fLastUse = batchToken; }
 
-        void uploadToTexture(GrDrawBatch::WritePixelsFn&, GrTexture* texture);
+        void uploadToTexture(GrDrawBatch::WritePixelsFn&, GrDrawBatch::TransferPixelsFn&, 
+                             GrTexture* texture);
         void resetRects();
 
     private:
         BatchPlot(int index, uint64_t genID, int offX, int offY, int width, int height,
-                  GrPixelConfig config);
+                  GrPixelConfig config, GrResourceProvider* rp);
 
         ~BatchPlot() override;
 
         // Create a clone of this plot. The cloned plot will take the place of the
         // current plot in the atlas.
         BatchPlot* clone() const {
-            return new BatchPlot(fIndex, fGenID+1, fX, fY, fWidth, fHeight, fConfig);
+            return new BatchPlot(fIndex, fGenID+1, fX, fY, fWidth, fHeight, fConfig,
+                                 fResourceProvider);
         }
 
         static GrBatchAtlas::AtlasID CreateId(uint32_t index, uint64_t generation) {
@@ -193,13 +197,16 @@ private:
             return generation << 16 | index;
         }
 
+        // used to create transfer buffers
+        GrResourceProvider*   fResourceProvider;
         GrBatchDrawToken      fLastUpload;
         GrBatchDrawToken      fLastUse;
 
         const uint32_t        fIndex;
         uint64_t              fGenID;
         GrBatchAtlas::AtlasID fID;
-        unsigned char*        fData;
+        unsigned char*        fDataPtr;
+        GrBuffer*             fTransferBuffer;
         const int             fWidth;
         const int             fHeight;
         const int             fX;
diff --git a/src/gpu/GrBatchFlushState.h b/src/gpu/GrBatchFlushState.h
index 0b2e2bd239..ee7e3a0afa 100644
--- a/src/gpu/GrBatchFlushState.h
+++ b/src/gpu/GrBatchFlushState.h
@@ -77,7 +77,14 @@ public:
             return this->fGpu->writePixels(surface, left, top, width, height, config, buffer,
                                            rowBytes);
         };
-        upload(wp);
+        GrDrawBatch::TransferPixelsFn tp = [this](GrTexture* texture,
+                                                int left, int top, int width, int height,
+                                                GrPixelConfig config, GrBuffer* buffer,
+                                                size_t offset, size_t rowBytes) -> bool {
+            return this->fGpu->transferPixels(texture, left, top, width, height, config, buffer,
+                                              offset, rowBytes);
+        };
+        upload(wp, tp);
     }
 
     void putBackIndices(size_t indices) { fIndexPool.putBack(indices * sizeof(uint16_t)); }
diff --git a/src/gpu/GrGpu.cpp b/src/gpu/GrGpu.cpp
index d062a48129..088641b908 100644
--- a/src/gpu/GrGpu.cpp
+++ b/src/gpu/GrGpu.cpp
@@ -397,14 +397,14 @@ bool GrGpu::writePixels(GrSurface* surface,
     return this->writePixels(surface, left, top, width, height, config, texels);
 }
 
-bool GrGpu::transferPixels(GrSurface* surface,
+bool GrGpu::transferPixels(GrTexture* texture,
                            int left, int top, int width, int height,
                            GrPixelConfig config, GrBuffer* transferBuffer,
                            size_t offset, size_t rowBytes) {
     SkASSERT(transferBuffer);
 
     this->handleDirtyContext();
-    if (this->onTransferPixels(surface, left, top, width, height, config,
+    if (this->onTransferPixels(texture, left, top, width, height, config,
                                transferBuffer, offset, rowBytes)) {
         fStats.incTransfersToTexture();
         return true;
diff --git a/src/gpu/GrGpu.h b/src/gpu/GrGpu.h
index 883b7ee98a..4defa040d7 100644
--- a/src/gpu/GrGpu.h
+++ b/src/gpu/GrGpu.h
@@ -280,9 +280,9 @@ public:
                      size_t rowBytes);
 
     /**
-     * Updates the pixels in a rectangle of a surface using a buffer
+     * Updates the pixels in a rectangle of a texture using a buffer
      *
-     * @param surface          The surface to write to.
+     * @param textre           The texture to write to.
      * @param left             left edge of the rectangle to write (inclusive)
      * @param top              top edge of the rectangle to write (inclusive)
      * @param width            width of rectangle to write in pixels.
@@ -293,7 +293,7 @@ public:
      * @param rowBytes         number of bytes between consecutive rows. Zero
      *                         means rows are tightly packed.
      */
-    bool transferPixels(GrSurface* surface,
+    bool transferPixels(GrTexture* texture,
                         int left, int top, int width, int height,
                         GrPixelConfig config, GrBuffer* transferBuffer,
                         size_t offset, size_t rowBytes);
@@ -579,8 +579,8 @@ private:
                                GrPixelConfig config,
                                const SkTArray<GrMipLevel>& texels) = 0;
 
-    // overridden by backend-specific derived class to perform the surface write
-    virtual bool onTransferPixels(GrSurface*,
+    // overridden by backend-specific derived class to perform the texture transfer
+    virtual bool onTransferPixels(GrTexture*,
                                   int left, int top, int width, int height,
                                   GrPixelConfig config, GrBuffer* transferBuffer,
                                   size_t offset, size_t rowBytes) = 0;
diff --git a/src/gpu/GrResourceProvider.cpp b/src/gpu/GrResourceProvider.cpp
index 9a2fff3141..922e5be320 100644
--- a/src/gpu/GrResourceProvider.cpp
+++ b/src/gpu/GrResourceProvider.cpp
@@ -139,7 +139,8 @@ GrBatchAtlas* GrResourceProvider::createAtlas(GrPixelConfig config,
     if (!texture) {
         return nullptr;
     }
-    GrBatchAtlas* atlas = new GrBatchAtlas(texture, numPlotsX, numPlotsY);
+    GrBatchAtlas* atlas = new GrBatchAtlas(this, texture, 
+                                           numPlotsX, numPlotsY);
     atlas->registerEvictionCallback(func, data);
     return atlas;
 }
diff --git a/src/gpu/batches/GrDrawBatch.h b/src/gpu/batches/GrDrawBatch.h
index bf93cf5a72..ef335fb8ba 100644
--- a/src/gpu/batches/GrDrawBatch.h
+++ b/src/gpu/batches/GrDrawBatch.h
@@ -46,13 +46,17 @@ private:
  */
 class GrDrawBatch : public GrBatch {
 public:
-    /** Method that performs an upload on behalf of a DeferredUploadFn. */
+    /** Method that performs a texture write on behalf of a DeferredUploadFn. */
     using WritePixelsFn = std::function<bool(GrSurface* texture,
                                              int left, int top, int width, int height,
-                                             GrPixelConfig config, const void* buffer,
+                                             GrPixelConfig config, const void* buffer, 
                                              size_t rowBytes)>;
+    using TransferPixelsFn = std::function<bool(GrTexture* texture,
+                                                int left, int top, int width, int height,
+                                                GrPixelConfig config, GrBuffer* buffer,
+                                                size_t offset, size_t rowBytes)>;
     /** See comments before GrDrawBatch::Target definition on how deferred uploaders work. */
-    using DeferredUploadFn = std::function<void(WritePixelsFn&)>;
+    using DeferredUploadFn = std::function<void(WritePixelsFn&, TransferPixelsFn&)>;
 
     class Target;
 
diff --git a/src/gpu/gl/GrGLBuffer.cpp b/src/gpu/gl/GrGLBuffer.cpp
index 24fd59267d..787445e558 100644
--- a/src/gpu/gl/GrGLBuffer.cpp
+++ b/src/gpu/gl/GrGLBuffer.cpp
@@ -30,6 +30,12 @@
 
 GrGLBuffer* GrGLBuffer::Create(GrGLGpu* gpu, size_t size, GrBufferType intendedType,
                                GrAccessPattern accessPattern, const void* data) {
+    if (gpu->glCaps().transferBufferType() == GrGLCaps::kNone_TransferBufferType &&
+        (kXferCpuToGpu_GrBufferType == intendedType ||
+        kXferGpuToCpu_GrBufferType == intendedType)) {
+        return nullptr;
+    }
+
     bool cpuBacked = gpu->glCaps().useNonVBOVertexAndIndexDynamicData() &&
                      GrBufferTypeIsVertexOrIndex(intendedType) &&
                      kDynamic_GrAccessPattern == accessPattern;
diff --git a/src/gpu/gl/GrGLCaps.cpp b/src/gpu/gl/GrGLCaps.cpp
index f3e494acb9..abf2cb7d56 100644
--- a/src/gpu/gl/GrGLCaps.cpp
+++ b/src/gpu/gl/GrGLCaps.cpp
@@ -393,8 +393,9 @@ void GrGLCaps::init(const GrContextOptions& contextOptions,
     } else {
         if (version >= GR_GL_VER(3, 0) || ctxInfo.hasExtension("GL_NV_pixel_buffer_object")) {
             fTransferBufferType = kPBO_TransferBufferType;
-        } else if (ctxInfo.hasExtension("GL_CHROMIUM_pixel_transfer_buffer_object")) {
-            fTransferBufferType = kChromium_TransferBufferType;
+// TODO: get transfer buffers working in Chrome
+//        } else if (ctxInfo.hasExtension("GL_CHROMIUM_pixel_transfer_buffer_object")) {
+//            fTransferBufferType = kChromium_TransferBufferType;
         }
     }
 
diff --git a/src/gpu/gl/GrGLGpu.cpp b/src/gpu/gl/GrGLGpu.cpp
index 3a7f531476..e50f1e2fdb 100644
--- a/src/gpu/gl/GrGLGpu.cpp
+++ b/src/gpu/gl/GrGLGpu.cpp
@@ -895,13 +895,13 @@ bool GrGLGpu::onWritePixels(GrSurface* surface,
     return success;
 }
 
-bool GrGLGpu::onTransferPixels(GrSurface* surface,
+bool GrGLGpu::onTransferPixels(GrTexture* texture,
                                int left, int top, int width, int height,
                                GrPixelConfig config, GrBuffer* transferBuffer,
                                size_t offset, size_t rowBytes) {
-    GrGLTexture* glTex = static_cast<GrGLTexture*>(surface->asTexture());
+    GrGLTexture* glTex = static_cast<GrGLTexture*>(texture);
 
-    if (!check_write_and_transfer_input(glTex, surface, config)) {
+    if (!check_write_and_transfer_input(glTex, texture, config)) {
         return false;
     }
 
@@ -920,15 +920,16 @@ bool GrGLGpu::onTransferPixels(GrSurface* surface,
 
     bool success = false;
     GrMipLevel mipLevel;
-    mipLevel.fPixels = transferBuffer;
+    mipLevel.fPixels = (void*)offset;
     mipLevel.fRowBytes = rowBytes;
     SkSTArray<1, GrMipLevel> texels;
     texels.push_back(mipLevel);
     success = this->uploadTexData(glTex->desc(), glTex->target(), kTransfer_UploadType,
                                   left, top, width, height, config, texels);
+
     if (success) {
         SkIRect rect = SkIRect::MakeXYWH(left, top, width, height);
-        this->didWriteToSurface(surface, &rect);
+        this->didWriteToSurface(texture, &rect);
         return true;
     }
 
@@ -1196,6 +1197,14 @@ bool GrGLGpu::uploadTexData(const GrSurfaceDesc& desc,
 
     SkASSERT(this->caps()->isConfigTexturable(desc.fConfig));
 
+    // unbind any previous transfer buffer if not transferring
+    auto& xferBufferState = fHWBufferState[kXferCpuToGpu_GrBufferType];
+    if (kTransfer_UploadType != uploadType && 
+        SK_InvalidUniqueID != xferBufferState.fBoundBufferUniqueID) {
+        GL_CALL(BindBuffer(xferBufferState.fGLTarget, 0));
+        xferBufferState.invalidate();
+    }
+
     // texels is const.
     // But we may need to flip the texture vertically to prepare it.
     // Rather than flip in place and alter the incoming data,
diff --git a/src/gpu/gl/GrGLGpu.h b/src/gpu/gl/GrGLGpu.h
index 00ceb85241..57d076094c 100644
--- a/src/gpu/gl/GrGLGpu.h
+++ b/src/gpu/gl/GrGLGpu.h
@@ -184,7 +184,7 @@ private:
                        GrPixelConfig config,
                        const SkTArray<GrMipLevel>& texels) override;
 
-    bool onTransferPixels(GrSurface*,
+    bool onTransferPixels(GrTexture*,
                           int left, int top, int width, int height,
                           GrPixelConfig config, GrBuffer* transferBuffer,
                           size_t offset, size_t rowBytes) override;
diff --git a/src/gpu/vk/GrVkGpu.cpp b/src/gpu/vk/GrVkGpu.cpp
index 5fba475cab..8083509877 100644
--- a/src/gpu/vk/GrVkGpu.cpp
+++ b/src/gpu/vk/GrVkGpu.cpp
@@ -187,11 +187,13 @@ GrBuffer* GrVkGpu::onCreateBuffer(size_t size, GrBufferType type, GrAccessPatter
             buff = GrVkIndexBuffer::Create(this, size, kDynamic_GrAccessPattern == accessPattern);
             break;
         case kXferCpuToGpu_GrBufferType:
-            SkASSERT(kStream_GrAccessPattern == accessPattern);
+            SkASSERT(kDynamic_GrAccessPattern == accessPattern ||
+                     kStream_GrAccessPattern == accessPattern);
             buff = GrVkTransferBuffer::Create(this, size, GrVkBuffer::kCopyRead_Type);
             break;
         case kXferGpuToCpu_GrBufferType:
-            SkASSERT(kStream_GrAccessPattern == accessPattern);
+            SkASSERT(kDynamic_GrAccessPattern == accessPattern ||
+                     kStream_GrAccessPattern == accessPattern);
             buff = GrVkTransferBuffer::Create(this, size, GrVkBuffer::kCopyWrite_Type);
             break;
         default:
@@ -284,6 +286,95 @@ bool GrVkGpu::onWritePixels(GrSurface* surface,
     return false;
 }
 
+
+bool GrVkGpu::onTransferPixels(GrTexture* texture,
+                               int left, int top, int width, int height,
+                               GrPixelConfig config, GrBuffer* transferBuffer,
+                               size_t bufferOffset, size_t rowBytes) {
+    GrVkTexture* vkTex = static_cast<GrVkTexture*>(texture);
+    if (!vkTex) {
+        return false;
+    }
+    GrVkTransferBuffer* vkBuffer = static_cast<GrVkTransferBuffer*>(transferBuffer);
+    if (!vkBuffer) {
+        return false;
+    }
+
+    // We assume Vulkan doesn't do sRGB <-> linear conversions when reading and writing pixels.
+    if (GrPixelConfigIsSRGB(texture->config()) != GrPixelConfigIsSRGB(config)) {
+        return false;
+    }
+
+    // TODO: Handle y axis flip via copy to temp image, then blit to final
+    if (kBottomLeft_GrSurfaceOrigin == vkTex->origin()) {
+        return false;
+    }
+
+    bool success = false;
+    if (GrPixelConfigIsCompressed(vkTex->desc().fConfig)) {
+        // We check that config == desc.fConfig in GrGpu::getWritePixelsInfo()
+        SkASSERT(config == vkTex->desc().fConfig);
+        // TODO: add compressed texture support
+        // delete the following two lines and uncomment the two after that when ready
+        vkTex->unref();
+        return false;
+        //success = this->uploadCompressedTexData(vkTex->desc(), buffer, false, left, top, width,
+        //                                       height);
+    } else {
+        // make sure the unmap has finished
+        vkBuffer->addMemoryBarrier(this,
+                                   VK_ACCESS_HOST_WRITE_BIT,
+                                   VK_ACCESS_TRANSFER_READ_BIT,
+                                   VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
+                                   VK_PIPELINE_STAGE_TRANSFER_BIT,
+                                   false);
+
+        // Set up copy region
+        size_t bpp = GrBytesPerPixel(config);
+
+        VkBufferImageCopy region;
+        memset(&region, 0, sizeof(VkBufferImageCopy));
+        region.bufferOffset = bufferOffset;
+        region.bufferRowLength = (uint32_t)(rowBytes/bpp);
+        region.bufferImageHeight = 0;
+        region.imageSubresource = { VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1 };
+        region.imageOffset = { left, top, 0 };
+        region.imageExtent = { (uint32_t)width, (uint32_t)height, 1 };
+
+        // Change layout of our target so it can be copied to
+        VkImageLayout layout = vkTex->currentLayout();
+        VkPipelineStageFlags srcStageMask = GrVkMemory::LayoutToPipelineStageFlags(layout);
+        VkPipelineStageFlags dstStageMask = VK_PIPELINE_STAGE_TRANSFER_BIT;
+        VkAccessFlags srcAccessMask = GrVkMemory::LayoutToSrcAccessMask(layout);
+        VkAccessFlags dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
+        vkTex->setImageLayout(this,
+                              VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                              srcAccessMask,
+                              dstAccessMask,
+                              srcStageMask,
+                              dstStageMask,
+                              false);
+
+        // Copy the buffer to the image
+        fCurrentCmdBuffer->copyBufferToImage(this,
+                                             vkBuffer,
+                                             vkTex,
+                                             VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+                                             1,
+                                             &region);
+
+        // Submit the current command buffer to the Queue
+        this->submitCommandBuffer(kSkip_SyncQueue);
+    }
+
+    if (success) {
+        vkTex->texturePriv().dirtyMipMaps(true);
+        return true;
+    }
+
+    return false;
+}
+
 bool GrVkGpu::uploadTexData(GrVkTexture* tex,
                             int left, int top, int width, int height,
                             GrPixelConfig dataConfig,
diff --git a/src/gpu/vk/GrVkGpu.h b/src/gpu/vk/GrVkGpu.h
index 76b18ac71e..e738e7336a 100644
--- a/src/gpu/vk/GrVkGpu.h
+++ b/src/gpu/vk/GrVkGpu.h
@@ -151,10 +151,10 @@ private:
                        int left, int top, int width, int height,
                        GrPixelConfig config, const SkTArray<GrMipLevel>&) override;
 
-    bool onTransferPixels(GrSurface*,
+    bool onTransferPixels(GrTexture*,
                           int left, int top, int width, int height,
                           GrPixelConfig config, GrBuffer* transferBuffer,
-                          size_t offset, size_t rowBytes) override { return false; }
+                          size_t offset, size_t rowBytes) override;
 
     void onResolveRenderTarget(GrRenderTarget* target) override {}