skia2/dm/DMSrcSink.cpp

1768 lines
69 KiB
C++
Raw Normal View History

/*
* 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 "DMSrcSink.h"
#include "Resources.h"
#include "SkAndroidCodec.h"
#include "SkAutoMalloc.h"
#include "SkCodec.h"
#include "SkCodecImageGenerator.h"
#include "SkColorSpace.h"
#include "SkColorSpaceXform.h"
#include "SkColorSpace_XYZ.h"
#include "SkCommonFlags.h"
#include "SkData.h"
#include "SkDebugCanvas.h"
#include "SkDeferredCanvas.h"
#include "SkDocument.h"
#include "SkImageGenerator.h"
#include "SkImageGeneratorCG.h"
#include "SkImageGeneratorWIC.h"
#include "SkLiteDL.h"
#include "SkLiteRecorder.h"
#include "SkMallocPixelRef.h"
#include "SkMultiPictureDraw.h"
#include "SkNullCanvas.h"
#include "SkOSFile.h"
#include "SkOSPath.h"
#include "SkOpts.h"
#include "SkPictureData.h"
#include "SkPictureRecorder.h"
#include "SkPipe.h"
#include "SkRandom.h"
#include "SkRecordDraw.h"
#include "SkRecorder.h"
#include "SkSVGCanvas.h"
Make SkStream *not* ref counted. SkStream is a stateful object, so it does not make sense for it to have multiple owners. Make SkStream inherit directly from SkNoncopyable. Update methods which previously called SkStream::ref() (e.g. SkImageDecoder::buildTileIndex() and SkFrontBufferedStream::Create(), which required the existing owners to call SkStream::unref()) to take ownership of their SkStream parameters and delete when done (including on failure). Switch all SkAutoTUnref<SkStream>s to SkAutoTDelete<SkStream>s. In some cases this means heap allocating streams that were previously stack allocated. Respect ownership rules of SkTypeface::CreateFromStream() and SkImageDecoder::buildTileIndex(). Update the comments for exceptional methods which do not affect the ownership of their SkStream parameters (e.g. SkPicture::CreateFromStream() and SkTypeface::Deserialize()) to be explicit about ownership. Remove test_stream_life, which tested that buildTileIndex() behaved correctly when SkStream was a ref counted object. The test does not make sense now that it is not. In SkPDFStream, remove the SkMemoryStream member. Instead of using it, create a new SkMemoryStream to pass to fDataStream (which is now an SkAutoTDelete). Make other pdf rasterizers behave like SkPDFDocumentToBitmap. SkPDFDocumentToBitmap delete the SkStream, so do the same in the following pdf rasterizers: SkPopplerRasterizePDF SkNativeRasterizePDF SkNoRasterizePDF Requires a change to Android, which currently treats SkStreams as ref counted objects. Review URL: https://codereview.chromium.org/849103004
2015-01-21 20:09:53 +00:00
#include "SkStream.h"
#include "SkSwizzler.h"
#include "SkTLogic.h"
#include <cmath>
#include <functional>
#if defined(SK_BUILD_FOR_WIN)
#include "SkAutoCoInitialize.h"
#endif
#if defined(SK_XML)
#include "SkSVGDOM.h"
#include "SkXMLWriter.h"
#endif
DEFINE_bool(multiPage, false, "For document-type backends, render the source"
" into multiple pages");
DEFINE_bool(RAW_threading, true, "Allow RAW decodes to run on multiple threads?");
using sk_gpu_test::GrContextFactory;
namespace DM {
GMSrc::GMSrc(skiagm::GMRegistry::Factory factory) : fFactory(factory) {}
Error GMSrc::draw(SkCanvas* canvas) const {
std::unique_ptr<skiagm::GM> gm(fFactory(nullptr));
canvas->concat(gm->getInitialTransform());
gm->draw(canvas);
return "";
}
SkISize GMSrc::size() const {
std::unique_ptr<skiagm::GM> gm(fFactory(nullptr));
return gm->getISize();
}
Name GMSrc::name() const {
std::unique_ptr<skiagm::GM> gm(fFactory(nullptr));
return gm->getName();
}
void GMSrc::modifyGrContextOptions(GrContextOptions* options) const {
std::unique_ptr<skiagm::GM> gm(fFactory(nullptr));
gm->modifyGrContextOptions(options);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
BRDSrc::BRDSrc(Path path, Mode mode, CodecSrc::DstColorType dstColorType, uint32_t sampleSize)
: fPath(path)
, fMode(mode)
, fDstColorType(dstColorType)
, fSampleSize(sampleSize)
{}
bool BRDSrc::veto(SinkFlags flags) const {
// No need to test to non-raster or indirect backends.
return flags.type != SinkFlags::kRaster
|| flags.approach != SinkFlags::kDirect;
}
static SkBitmapRegionDecoder* create_brd(Path path) {
sk_sp<SkData> encoded(SkData::MakeFromFileName(path.c_str()));
if (!encoded) {
return NULL;
}
return SkBitmapRegionDecoder::Create(encoded, SkBitmapRegionDecoder::kAndroidCodec_Strategy);
}
static inline void alpha8_to_gray8(SkBitmap* bitmap) {
// Android requires kGray8 bitmaps to be tagged as kAlpha8. Here we convert
// them back to kGray8 so our test framework can draw them correctly.
if (kAlpha_8_SkColorType == bitmap->info().colorType()) {
SkImageInfo newInfo = bitmap->info().makeColorType(kGray_8_SkColorType)
.makeAlphaType(kOpaque_SkAlphaType);
*const_cast<SkImageInfo*>(&bitmap->info()) = newInfo;
}
}
Error BRDSrc::draw(SkCanvas* canvas) const {
SkColorType colorType = canvas->imageInfo().colorType();
if (kRGB_565_SkColorType == colorType &&
CodecSrc::kGetFromCanvas_DstColorType != fDstColorType) {
return Error::Nonfatal("Testing non-565 to 565 is uninteresting.");
}
switch (fDstColorType) {
case CodecSrc::kGetFromCanvas_DstColorType:
break;
case CodecSrc::kIndex8_Always_DstColorType:
colorType = kIndex_8_SkColorType;
break;
case CodecSrc::kGrayscale_Always_DstColorType:
colorType = kGray_8_SkColorType;
break;
default:
SkASSERT(false);
break;
}
std::unique_ptr<SkBitmapRegionDecoder> brd(create_brd(fPath));
if (nullptr == brd.get()) {
return Error::Nonfatal(SkStringPrintf("Could not create brd for %s.", fPath.c_str()));
}
if (!brd->conversionSupported(colorType)) {
return Error::Nonfatal("Cannot convert to color type.");
}
const uint32_t width = brd->width();
const uint32_t height = brd->height();
// Visually inspecting very small output images is not necessary.
if ((width / fSampleSize <= 10 || height / fSampleSize <= 10) && 1 != fSampleSize) {
return Error::Nonfatal("Scaling very small images is uninteresting.");
}
switch (fMode) {
case kFullImage_Mode: {
SkBitmap bitmap;
if (!brd->decodeRegion(&bitmap, nullptr, SkIRect::MakeXYWH(0, 0, width, height),
fSampleSize, colorType, false)) {
return "Cannot decode (full) region.";
}
alpha8_to_gray8(&bitmap);
canvas->drawBitmap(bitmap, 0, 0);
return "";
}
case kDivisor_Mode: {
const uint32_t divisor = 2;
if (width < divisor || height < divisor) {
return Error::Nonfatal("Divisor is larger than image dimension.");
}
// Use a border to test subsets that extend outside the image.
// We will not allow the border to be larger than the image dimensions. Allowing
// these large borders causes off by one errors that indicate a problem with the
// test suite, not a problem with the implementation.
const uint32_t maxBorder = SkTMin(width, height) / (fSampleSize * divisor);
const uint32_t scaledBorder = SkTMin(5u, maxBorder);
const uint32_t unscaledBorder = scaledBorder * fSampleSize;
// We may need to clear the canvas to avoid uninitialized memory.
// Assume we are scaling a 780x780 image with sampleSize = 8.
// The output image should be 97x97.
// Each subset will be 390x390.
// Each scaled subset be 48x48.
// Four scaled subsets will only fill a 96x96 image.
// The bottom row and last column will not be touched.
// This is an unfortunate result of our rounding rules when scaling.
// Maybe we need to consider testing scaled subsets without trying to
// combine them to match the full scaled image? Or maybe this is the
// best we can do?
canvas->clear(0);
for (uint32_t x = 0; x < divisor; x++) {
for (uint32_t y = 0; y < divisor; y++) {
// Calculate the subset dimensions
uint32_t subsetWidth = width / divisor;
uint32_t subsetHeight = height / divisor;
const int left = x * subsetWidth;
const int top = y * subsetHeight;
// Increase the size of the last subset in each row or column, when the
// divisor does not divide evenly into the image dimensions
subsetWidth += (x + 1 == divisor) ? (width % divisor) : 0;
subsetHeight += (y + 1 == divisor) ? (height % divisor) : 0;
// Increase the size of the subset in order to have a border on each side
const int decodeLeft = left - unscaledBorder;
const int decodeTop = top - unscaledBorder;
const uint32_t decodeWidth = subsetWidth + unscaledBorder * 2;
const uint32_t decodeHeight = subsetHeight + unscaledBorder * 2;
SkBitmap bitmap;
if (!brd->decodeRegion(&bitmap, nullptr, SkIRect::MakeXYWH(decodeLeft,
decodeTop, decodeWidth, decodeHeight), fSampleSize, colorType, false)) {
return "Cannot decode region.";
}
alpha8_to_gray8(&bitmap);
canvas->drawBitmapRect(bitmap,
SkRect::MakeXYWH((SkScalar) scaledBorder, (SkScalar) scaledBorder,
(SkScalar) (subsetWidth / fSampleSize),
(SkScalar) (subsetHeight / fSampleSize)),
SkRect::MakeXYWH((SkScalar) (left / fSampleSize),
(SkScalar) (top / fSampleSize),
(SkScalar) (subsetWidth / fSampleSize),
(SkScalar) (subsetHeight / fSampleSize)),
nullptr);
}
}
return "";
}
default:
SkASSERT(false);
return "Error: Should not be reached.";
}
}
SkISize BRDSrc::size() const {
std::unique_ptr<SkBitmapRegionDecoder> brd(create_brd(fPath));
if (brd) {
return SkISize::Make(SkTMax(1, brd->width() / (int) fSampleSize),
SkTMax(1, brd->height() / (int) fSampleSize));
}
return SkISize::Make(0, 0);
}
static SkString get_scaled_name(const Path& path, float scale) {
return SkStringPrintf("%s_%.3f", SkOSPath::Basename(path.c_str()).c_str(), scale);
}
Name BRDSrc::name() const {
// We will replicate the names used by CodecSrc so that images can
// be compared in Gold.
if (1 == fSampleSize) {
return SkOSPath::Basename(fPath.c_str());
}
return get_scaled_name(fPath, 1.0f / (float) fSampleSize);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
static bool serial_from_path_name(const SkString& path) {
if (!FLAGS_RAW_threading) {
static const char* const exts[] = {
"arw", "cr2", "dng", "nef", "nrw", "orf", "raf", "rw2", "pef", "srw",
"ARW", "CR2", "DNG", "NEF", "NRW", "ORF", "RAF", "RW2", "PEF", "SRW",
};
const char* actualExt = strrchr(path.c_str(), '.');
if (actualExt) {
actualExt++;
for (auto* ext : exts) {
if (0 == strcmp(ext, actualExt)) {
return true;
}
}
}
}
return false;
}
CodecSrc::CodecSrc(Path path, Mode mode, DstColorType dstColorType, SkAlphaType dstAlphaType,
float scale)
: fPath(path)
, fMode(mode)
, fDstColorType(dstColorType)
, fDstAlphaType(dstAlphaType)
, fScale(scale)
, fRunSerially(serial_from_path_name(path))
{}
bool CodecSrc::veto(SinkFlags flags) const {
// Test to direct raster backends (8888 and 565).
return flags.type != SinkFlags::kRaster || flags.approach != SinkFlags::kDirect;
}
// Allows us to test decodes to non-native 8888.
static void swap_rb_if_necessary(SkBitmap& bitmap, CodecSrc::DstColorType dstColorType) {
if (CodecSrc::kNonNative8888_Always_DstColorType != dstColorType) {
return;
}
for (int y = 0; y < bitmap.height(); y++) {
uint32_t* row = (uint32_t*) bitmap.getAddr(0, y);
SkOpts::RGBA_to_BGRA(row, row, bitmap.width());
}
}
// FIXME: Currently we cannot draw unpremultiplied sources. skbug.com/3338 and skbug.com/3339.
// This allows us to still test unpremultiplied decodes.
static void premultiply_if_necessary(SkBitmap& bitmap) {
if (kUnpremul_SkAlphaType != bitmap.alphaType()) {
return;
}
switch (bitmap.colorType()) {
case kN32_SkColorType:
for (int y = 0; y < bitmap.height(); y++) {
uint32_t* row = (uint32_t*) bitmap.getAddr(0, y);
SkOpts::RGBA_to_rgbA(row, row, bitmap.width());
}
break;
case kIndex_8_SkColorType: {
SkColorTable* colorTable = bitmap.getColorTable();
SkPMColor* colorPtr = const_cast<SkPMColor*>(colorTable->readColors());
SkOpts::RGBA_to_rgbA(colorPtr, colorPtr, colorTable->count());
break;
}
default:
// No need to premultiply kGray or k565 outputs.
break;
}
// In the kIndex_8 case, the canvas won't even try to draw unless we mark the
// bitmap as kPremul.
bitmap.setAlphaType(kPremul_SkAlphaType);
}
static bool get_decode_info(SkImageInfo* decodeInfo, SkColorType canvasColorType,
CodecSrc::DstColorType dstColorType, SkAlphaType dstAlphaType) {
switch (dstColorType) {
case CodecSrc::kIndex8_Always_DstColorType:
if (kRGB_565_SkColorType == canvasColorType) {
return false;
}
*decodeInfo = decodeInfo->makeColorType(kIndex_8_SkColorType);
break;
case CodecSrc::kGrayscale_Always_DstColorType:
if (kRGB_565_SkColorType == canvasColorType) {
return false;
}
*decodeInfo = decodeInfo->makeColorType(kGray_8_SkColorType);
break;
case CodecSrc::kNonNative8888_Always_DstColorType:
if (kRGB_565_SkColorType == canvasColorType) {
return false;
}
#ifdef SK_PMCOLOR_IS_RGBA
*decodeInfo = decodeInfo->makeColorType(kBGRA_8888_SkColorType);
#else
*decodeInfo = decodeInfo->makeColorType(kRGBA_8888_SkColorType);
#endif
break;
default:
if (kRGB_565_SkColorType == canvasColorType &&
kOpaque_SkAlphaType != decodeInfo->alphaType()) {
return false;
}
*decodeInfo = decodeInfo->makeColorType(canvasColorType);
break;
}
*decodeInfo = decodeInfo->makeAlphaType(dstAlphaType);
return true;
}
static void draw_to_canvas(SkCanvas* canvas, const SkImageInfo& info, void* pixels, size_t rowBytes,
SkPMColor* colorPtr, int colorCount, CodecSrc::DstColorType dstColorType,
SkScalar left = 0, SkScalar top = 0) {
sk_sp<SkColorTable> colorTable(new SkColorTable(colorPtr, colorCount));
SkBitmap bitmap;
bitmap.installPixels(info, pixels, rowBytes, colorTable.get(), nullptr, nullptr);
premultiply_if_necessary(bitmap);
swap_rb_if_necessary(bitmap, dstColorType);
canvas->drawBitmap(bitmap, left, top);
}
Error CodecSrc::draw(SkCanvas* canvas) const {
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
if (!encoded) {
return SkStringPrintf("Couldn't read %s.", fPath.c_str());
}
std::unique_ptr<SkCodec> codec(SkCodec::NewFromData(encoded));
if (nullptr == codec.get()) {
return SkStringPrintf("Couldn't create codec for %s.", fPath.c_str());
}
SkImageInfo decodeInfo = codec->getInfo();
if (!get_decode_info(&decodeInfo, canvas->imageInfo().colorType(), fDstColorType,
fDstAlphaType)) {
return Error::Nonfatal("Testing non-565 to 565 is uninteresting.");
}
// Try to scale the image if it is desired
SkISize size = codec->getScaledDimensions(fScale);
if (size == decodeInfo.dimensions() && 1.0f != fScale) {
return Error::Nonfatal("Test without scaling is uninteresting.");
}
// Visually inspecting very small output images is not necessary. We will
// cover these cases in unit testing.
if ((size.width() <= 10 || size.height() <= 10) && 1.0f != fScale) {
return Error::Nonfatal("Scaling very small images is uninteresting.");
}
decodeInfo = decodeInfo.makeWH(size.width(), size.height());
const int bpp = SkColorTypeBytesPerPixel(decodeInfo.colorType());
const size_t rowBytes = size.width() * bpp;
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
const size_t safeSize = decodeInfo.getSafeSize(rowBytes);
SkAutoMalloc pixels(safeSize);
SkPMColor colorPtr[256];
int colorCount = 256;
SkCodec::Options options;
if (kCodecZeroInit_Mode == fMode) {
memset(pixels.get(), 0, size.height() * rowBytes);
options.fZeroInitialized = SkCodec::kYes_ZeroInitialized;
}
SkImageInfo bitmapInfo = decodeInfo;
if (kRGBA_8888_SkColorType == decodeInfo.colorType() ||
kBGRA_8888_SkColorType == decodeInfo.colorType()) {
bitmapInfo = bitmapInfo.makeColorType(kN32_SkColorType);
}
switch (fMode) {
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
case kAnimated_Mode: {
std::vector<SkCodec::FrameInfo> frameInfos = codec->getFrameInfo();
if (frameInfos.size() <= 1) {
return SkStringPrintf("%s is not an animated image.", fPath.c_str());
}
// As in CodecSrc::size(), compute a roughly square grid to draw the frames
// into. "factor" is the number of frames to draw on one row. There will be
// up to "factor" rows as well.
const float root = sqrt((float) frameInfos.size());
const int factor = sk_float_ceil2int(root);
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
// Used to cache a frame that future frames will depend on.
SkAutoMalloc priorFramePixels;
size_t cachedFrame = SkCodec::kNone;
for (size_t i = 0; i < frameInfos.size(); i++) {
options.fFrameIndex = i;
// Check for a prior frame
const size_t reqFrame = frameInfos[i].fRequiredFrame;
if (reqFrame != SkCodec::kNone && reqFrame == cachedFrame
&& priorFramePixels.get()) {
// Copy into pixels
memcpy(pixels.get(), priorFramePixels.get(), safeSize);
options.fHasPriorFrame = true;
} else {
options.fHasPriorFrame = false;
}
Fix SkGifCodec bugs around truncated data Prior to this CL, if a GIF file was truncated before reading the local color map of a frame, incremental decode would do the wrong thing. In onStartIncrementalDecode, we would either create a color table based on the global color map, or we would create a dummy one with only one color (transparent). The dummy color table is correct if there is neither a global nor a local color map, and allows us to fill the frame with transparent. But if more data is provided, and it includes an actual color map and image data, one of the following can happen: - If the created color table is smaller than the actual one, the decoded data may include indices outside of the range of the created color table, resulting in a crash. - If we get lucky, and the created color table is large enough, it may still be the wrong colors (and most likely is). To solve this, make onStartIncrementalDecode fail if there is a local color map that has not been read yet. A future call may read more data and read the correct color map. This is done by returning kIncompleteInput in SkGifCodec::prepareToDecode if there is a local color map that has not yet been read. (It is possible that there is no color map at all, in which case we still need to support decoding that frame. Skip attempting to decode in that case.) In onGetPixels, if prepareToDecode returned kIncompleteInput, return kInvalidInput. Although the input is technically incomplete, no future call will provide more data (unlike in incremental decoding), and there is nothing interesting for the client to draw. This also prevents SkCodec from attempting to fill the data with an SkSwizzler, which has not been created. (An alternative solution would be create the dummy color table and an SkSwizzler, which would keep the current behavior. But I think the new behavior of returning kInvalidInput makes more sense.) Add tests to verify the intended behavior: - getPixels fails. - startIncrementalDecode fails, but after providing more data it will succeed and incremental decoding matches the image decoded from the full stream. - Both succeed if there is no color table at all. Change-Id: Ifb52fe7f723673406a28e80c8805a552f0ac33b6 Reviewed-on: https://skia-review.googlesource.com/5758 Reviewed-by: Matt Sarett <msarett@google.com> Commit-Queue: Leon Scroggins <scroggo@google.com>
2016-12-09 21:39:33 +00:00
SkCodec::Result result = codec->getPixels(decodeInfo, pixels.get(),
rowBytes, &options,
colorPtr, &colorCount);
if (SkCodec::kInvalidInput == result && i > 0) {
// Some of our test images have truncated later frames. Treat that
// the same as incomplete.
result = SkCodec::kIncompleteInput;
}
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
switch (result) {
case SkCodec::kSuccess:
case SkCodec::kIncompleteInput: {
SkAutoCanvasRestore acr(canvas, true);
const int xTranslate = (i % factor) * decodeInfo.width();
const int yTranslate = (i / factor) * decodeInfo.height();
canvas->translate(SkIntToScalar(xTranslate), SkIntToScalar(yTranslate));
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
draw_to_canvas(canvas, bitmapInfo, pixels.get(), rowBytes,
colorPtr, colorCount, fDstColorType);
if (result == SkCodec::kIncompleteInput) {
return "";
}
break;
}
case SkCodec::kInvalidConversion:
if (i > 0 && (decodeInfo.colorType() == kRGB_565_SkColorType
|| decodeInfo.colorType() == kIndex_8_SkColorType)) {
return Error::Nonfatal(SkStringPrintf(
"Cannot decode frame %i to 565/Index8 (%s).", i, fPath.c_str()));
}
// Fall through.
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
default:
return SkStringPrintf("Couldn't getPixels for frame %i in %s.",
i, fPath.c_str());
}
// If a future frame depends on this one, store it in priorFrame.
// (Note that if i+1 does *not* depend on i, then no future frame can.)
if (i+1 < frameInfos.size() && frameInfos[i+1].fRequiredFrame == i) {
memcpy(priorFramePixels.reset(safeSize), pixels.get(), safeSize);
cachedFrame = i;
}
}
break;
}
case kCodecZeroInit_Mode:
case kCodec_Mode: {
switch (codec->getPixels(decodeInfo, pixels.get(), rowBytes, &options,
colorPtr, &colorCount)) {
case SkCodec::kSuccess:
// We consider incomplete to be valid, since we should still decode what is
// available.
case SkCodec::kIncompleteInput:
break;
default:
// Everything else is considered a failure.
return SkStringPrintf("Couldn't getPixels %s.", fPath.c_str());
}
draw_to_canvas(canvas, bitmapInfo, pixels.get(), rowBytes, colorPtr, colorCount,
fDstColorType);
break;
}
case kScanline_Mode: {
Revert of Make SkPngCodec decode progressively. (patchset #26 id:520001 of https://codereview.chromium.org/1997703003/ ) Reason for revert: Still causing problems in Google3, e.g. https://test.corp.google.com/ui#cl=124138817&flags=CAMQBQ==&id=OCL:124138817:BASE:124139560:1465227435491:219ffbdb&t=//third_party/skia/HEAD:dm Original issue's description: > Make SkPngCodec decode progressively. > > This is a step towards using SkCodec in Chromium, where progressive > decoding is necessary. > > Switch from using png_read_row (which expects all the data to be > available) to png_process_data, which uses callbacks when rows are > available. > > Create a new API for SkCodec, which supports progressive decoding and > scanline decoding. Future changes will switch the other clients off of > startScanlineDecode and get/skip-Scanlines to the new API. > > Remove SkCodec::kNone_ScanlineOrder, which was only used for interlaced > PNG images. In the new API, interlaced PNG fits kTopDown. Also remove > updateCurrScanline(), which was only used by the old implementation for > interlaced PNG. > > DMSrcSink: > - In CodecSrc::kScanline_Mode, use the new method for scanline decoding > for the supported formats (just PNG and PNG-in-ICO for now). > > fuzz.cpp: > - Remove reference to kNone_ScanlineOrder > > SkCodec: > - Add new APIs: > - startIncrementalDecode > - incrementalDecode > - Remove kNone_SkScanlineOrder and updateCurrScanline() > > SkPngCodec: > - Implement new APIs > - Switch from sk_read_fn/png_read_row etc to png_process_data > - Expand AutoCleanPng's role to decode the header and create the > SkPngCodec > - Make the interlaced PNG decoder report how many lines were > initialized during an incomplete decode > - Make initializeSwizzler return a bool instead of an SkCodec::Result > (It only returned kSuccess or kInvalidInput anyway) > > SkIcoCodec: > - Implement the new APIs; supported for PNG in ICO > > SkSampledCodec: > - Call the new method for decoding scanlines, and fall back to the old > method if the new version is unimplemented > - Remove references to kNone_SkScanlineOrder > > tests/CodecPartial: > - Add a test which decodes part of an image, then finishes the decode, > and compares it to the straightforward method > > tests/CodecTest: > - Add a test which decodes all scanlines using the new method > - Repurpose the Codec_stripes test to decode using the new method in > sections rather than all at once > - In the method check(), add a parameter for whether the image supports > the new method of scanline decoding, and be explicit about whether an > image supports incomplete > - Test incomplete PNG decodes. We should have been doing it anyway for > non-interlaced (except for an image that is too small - one row), but > the new method supports interlaced incomplete as well > - Make test_invalid_parameters test the new method > - Add a test to ensure that it's safe to fall back to scanline decoding without > rewinding > > BUG=skia:4211 > > The new version was generally faster than the old version (but not significantly so). > > Some raw performance differences can be found at https://docs.google.com/a/google.com/spreadsheets/d/1Gis3aRCEa72qBNDRMgGDg3jD-pMgO-FXldlNF9ejo4o/ > > Design doc can be found at https://docs.google.com/a/google.com/document/d/11Mn8-ePDKwVEMCjs3nWwSjxcSpJ_Cu8DF57KNtUmgLM/ > > GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1997703003 > > Committed: https://skia.googlesource.com/skia/+/a4b09a117d4d1ba5dda372e6a2323e653766539e > > Committed: https://skia.googlesource.com/skia/+/30e78c9737ff4861dc4e3fa1e4cd010680ed6965 > > Committed: https://skia.googlesource.com/skia/+/6fb2391b2cc83ee2160b4e994faa8128975acc1f TBR=reed@google.com,msarett@google.com,scroggo@chromium.org # Not skipping CQ checks because original CL landed more than 1 days ago. BUG=skia:4211 GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2044573002 Review-Url: https://codereview.chromium.org/2044573002
2016-06-06 18:26:17 +00:00
void* dst = pixels.get();
uint32_t height = decodeInfo.height();
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
const bool useIncremental = [this]() {
auto exts = { "png", "PNG", "gif", "GIF" };
for (auto ext : exts) {
if (fPath.endsWith(ext)) {
return true;
}
}
return false;
}();
// ico may use the old scanline method or the new one, depending on whether it
// internally holds a bmp or a png.
Make SkPngCodec decode progressively. This is a step towards using SkCodec in Chromium, where progressive decoding is necessary. Switch from using png_read_row (which expects all the data to be available) to png_process_data, which uses callbacks when rows are available. Create a new API for SkCodec, which supports progressive decoding and scanline decoding. Future changes will switch the other clients off of startScanlineDecode and get/skip-Scanlines to the new API. Remove SkCodec::kNone_ScanlineOrder, which was only used for interlaced PNG images. In the new API, interlaced PNG fits kTopDown. Also remove updateCurrScanline(), which was only used by the old implementation for interlaced PNG. DMSrcSink: - In CodecSrc::kScanline_Mode, use the new method for scanline decoding for the supported formats (just PNG and PNG-in-ICO for now). fuzz.cpp: - Remove reference to kNone_ScanlineOrder SkCodec: - Add new APIs: - startIncrementalDecode - incrementalDecode - Remove kNone_SkScanlineOrder and updateCurrScanline() - Set fDstInfo and fOptions in getPixels(). This may not be necessary for all implementations, but it simplifies things for SkPngCodec. SkPngCodec: - Implement new APIs - Switch from sk_read_fn/png_read_row etc to png_process_data - Expand AutoCleanPng's role to decode the header and create the SkPngCodec - Make the interlaced PNG decoder report how many lines were initialized during an incomplete decode SkIcoCodec: - Implement the new APIs; supported for PNG in ICO SkSampledCodec: - Call the new method for decoding scanlines, and fall back to the old method if the new version is unimplemented - Remove references to kNone_SkScanlineOrder tests/CodecPartial: - Add a test which decodes part of an image, then finishes the decode, and compares it to the straightforward method tests/CodecTest: - Add a test which decodes all scanlines using the new method - Repurpose the Codec_stripes test to decode using the new method in sections rather than all at once - In the method check(), add a parameter for whether the image supports the new method of scanline decoding, and be explicit about whether an image supports incomplete - Test incomplete PNG decodes. We should have been doing it anyway for non-interlaced (except for an image that is too small - one row), but the new method supports interlaced incomplete as well - Make test_invalid_parameters test the new method - Add a test to ensure that it's safe to fall back to scanline decoding without rewinding BUG=skia:4211 The new version was generally faster than the old version (but not significantly so). Some raw performance differences can be found at https://docs.google.com/a/google.com/spreadsheets/d/1Gis3aRCEa72qBNDRMgGDg3jD-pMgO-FXldlNF9ejo4o/ Design doc can be found at https://docs.google.com/a/google.com/document/d/11Mn8-ePDKwVEMCjs3nWwSjxcSpJ_Cu8DF57KNtUmgLM/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1997703003 Review-Url: https://codereview.chromium.org/1997703003
2016-09-16 15:20:38 +00:00
const bool ico = fPath.endsWith("ico");
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
bool useOldScanlineMethod = !useIncremental && !ico;
if (useIncremental || ico) {
Make SkPngCodec decode progressively. This is a step towards using SkCodec in Chromium, where progressive decoding is necessary. Switch from using png_read_row (which expects all the data to be available) to png_process_data, which uses callbacks when rows are available. Create a new API for SkCodec, which supports progressive decoding and scanline decoding. Future changes will switch the other clients off of startScanlineDecode and get/skip-Scanlines to the new API. Remove SkCodec::kNone_ScanlineOrder, which was only used for interlaced PNG images. In the new API, interlaced PNG fits kTopDown. Also remove updateCurrScanline(), which was only used by the old implementation for interlaced PNG. DMSrcSink: - In CodecSrc::kScanline_Mode, use the new method for scanline decoding for the supported formats (just PNG and PNG-in-ICO for now). fuzz.cpp: - Remove reference to kNone_ScanlineOrder SkCodec: - Add new APIs: - startIncrementalDecode - incrementalDecode - Remove kNone_SkScanlineOrder and updateCurrScanline() - Set fDstInfo and fOptions in getPixels(). This may not be necessary for all implementations, but it simplifies things for SkPngCodec. SkPngCodec: - Implement new APIs - Switch from sk_read_fn/png_read_row etc to png_process_data - Expand AutoCleanPng's role to decode the header and create the SkPngCodec - Make the interlaced PNG decoder report how many lines were initialized during an incomplete decode SkIcoCodec: - Implement the new APIs; supported for PNG in ICO SkSampledCodec: - Call the new method for decoding scanlines, and fall back to the old method if the new version is unimplemented - Remove references to kNone_SkScanlineOrder tests/CodecPartial: - Add a test which decodes part of an image, then finishes the decode, and compares it to the straightforward method tests/CodecTest: - Add a test which decodes all scanlines using the new method - Repurpose the Codec_stripes test to decode using the new method in sections rather than all at once - In the method check(), add a parameter for whether the image supports the new method of scanline decoding, and be explicit about whether an image supports incomplete - Test incomplete PNG decodes. We should have been doing it anyway for non-interlaced (except for an image that is too small - one row), but the new method supports interlaced incomplete as well - Make test_invalid_parameters test the new method - Add a test to ensure that it's safe to fall back to scanline decoding without rewinding BUG=skia:4211 The new version was generally faster than the old version (but not significantly so). Some raw performance differences can be found at https://docs.google.com/a/google.com/spreadsheets/d/1Gis3aRCEa72qBNDRMgGDg3jD-pMgO-FXldlNF9ejo4o/ Design doc can be found at https://docs.google.com/a/google.com/document/d/11Mn8-ePDKwVEMCjs3nWwSjxcSpJ_Cu8DF57KNtUmgLM/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1997703003 Review-Url: https://codereview.chromium.org/1997703003
2016-09-16 15:20:38 +00:00
if (SkCodec::kSuccess == codec->startIncrementalDecode(decodeInfo, dst,
rowBytes, nullptr, colorPtr, &colorCount)) {
int rowsDecoded;
if (SkCodec::kIncompleteInput == codec->incrementalDecode(&rowsDecoded)) {
codec->fillIncompleteImage(decodeInfo, dst, rowBytes,
SkCodec::kNo_ZeroInitialized, height,
rowsDecoded);
}
} else {
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
if (useIncremental) {
// Error: These should support incremental decode.
Make SkPngCodec decode progressively. This is a step towards using SkCodec in Chromium, where progressive decoding is necessary. Switch from using png_read_row (which expects all the data to be available) to png_process_data, which uses callbacks when rows are available. Create a new API for SkCodec, which supports progressive decoding and scanline decoding. Future changes will switch the other clients off of startScanlineDecode and get/skip-Scanlines to the new API. Remove SkCodec::kNone_ScanlineOrder, which was only used for interlaced PNG images. In the new API, interlaced PNG fits kTopDown. Also remove updateCurrScanline(), which was only used by the old implementation for interlaced PNG. DMSrcSink: - In CodecSrc::kScanline_Mode, use the new method for scanline decoding for the supported formats (just PNG and PNG-in-ICO for now). fuzz.cpp: - Remove reference to kNone_ScanlineOrder SkCodec: - Add new APIs: - startIncrementalDecode - incrementalDecode - Remove kNone_SkScanlineOrder and updateCurrScanline() - Set fDstInfo and fOptions in getPixels(). This may not be necessary for all implementations, but it simplifies things for SkPngCodec. SkPngCodec: - Implement new APIs - Switch from sk_read_fn/png_read_row etc to png_process_data - Expand AutoCleanPng's role to decode the header and create the SkPngCodec - Make the interlaced PNG decoder report how many lines were initialized during an incomplete decode SkIcoCodec: - Implement the new APIs; supported for PNG in ICO SkSampledCodec: - Call the new method for decoding scanlines, and fall back to the old method if the new version is unimplemented - Remove references to kNone_SkScanlineOrder tests/CodecPartial: - Add a test which decodes part of an image, then finishes the decode, and compares it to the straightforward method tests/CodecTest: - Add a test which decodes all scanlines using the new method - Repurpose the Codec_stripes test to decode using the new method in sections rather than all at once - In the method check(), add a parameter for whether the image supports the new method of scanline decoding, and be explicit about whether an image supports incomplete - Test incomplete PNG decodes. We should have been doing it anyway for non-interlaced (except for an image that is too small - one row), but the new method supports interlaced incomplete as well - Make test_invalid_parameters test the new method - Add a test to ensure that it's safe to fall back to scanline decoding without rewinding BUG=skia:4211 The new version was generally faster than the old version (but not significantly so). Some raw performance differences can be found at https://docs.google.com/a/google.com/spreadsheets/d/1Gis3aRCEa72qBNDRMgGDg3jD-pMgO-FXldlNF9ejo4o/ Design doc can be found at https://docs.google.com/a/google.com/document/d/11Mn8-ePDKwVEMCjs3nWwSjxcSpJ_Cu8DF57KNtUmgLM/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1997703003 Review-Url: https://codereview.chromium.org/1997703003
2016-09-16 15:20:38 +00:00
return "Could not start incremental decode";
}
// Otherwise, this is an ICO. Since incremental failed, it must contain a BMP,
// which should work via startScanlineDecode
useOldScanlineMethod = true;
}
}
if (useOldScanlineMethod) {
if (SkCodec::kSuccess != codec->startScanlineDecode(decodeInfo, NULL, colorPtr,
&colorCount)) {
return "Could not start scanline decoder";
}
switch (codec->getScanlineOrder()) {
case SkCodec::kTopDown_SkScanlineOrder:
case SkCodec::kBottomUp_SkScanlineOrder:
// We do not need to check the return value. On an incomplete
// image, memory will be filled with a default value.
codec->getScanlines(dst, height, rowBytes);
break;
}
}
draw_to_canvas(canvas, bitmapInfo, dst, rowBytes, colorPtr, colorCount, fDstColorType);
break;
}
case kStripe_Mode: {
const int height = decodeInfo.height();
// This value is chosen arbitrarily. We exercise more cases by choosing a value that
// does not align with image blocks.
const int stripeHeight = 37;
const int numStripes = (height + stripeHeight - 1) / stripeHeight;
void* dst = pixels.get();
// Decode odd stripes
if (SkCodec::kSuccess != codec->startScanlineDecode(decodeInfo, nullptr, colorPtr,
&colorCount)) {
return "Could not start scanline decoder";
}
// This mode was designed to test the new skip scanlines API in libjpeg-turbo.
// Jpegs have kTopDown_SkScanlineOrder, and at this time, it is not interesting
// to run this test for image types that do not have this scanline ordering.
// We only run this on Jpeg, which is always kTopDown.
SkASSERT(SkCodec::kTopDown_SkScanlineOrder == codec->getScanlineOrder());
for (int i = 0; i < numStripes; i += 2) {
// Skip a stripe
const int linesToSkip = SkTMin(stripeHeight, height - i * stripeHeight);
codec->skipScanlines(linesToSkip);
// Read a stripe
const int startY = (i + 1) * stripeHeight;
const int linesToRead = SkTMin(stripeHeight, height - startY);
if (linesToRead > 0) {
codec->getScanlines(SkTAddOffset<void>(dst, rowBytes * startY), linesToRead,
rowBytes);
}
}
// Decode even stripes
Merge SkCodec with SkScanlineDecoder Benefits: - This mimics other decoding APIs (including the ones SkCodec relies on, e.g. a png_struct, which can be used to decode an entire image or one line at a time). - It allows a client to ask us to do what we can do efficiently - i.e. start from encoded data and either decode the whole thing or scanlines. - It removes the duplicate methods which appeared in both SkCodec and SkScanlineDecoder (some of which, e.g. in SkJpegScanlineDecoder, just call fCodec->sameMethod()). - It simplifies moving more checks into the base class (e.g. the examples in skbug.com/4284). BUG=skia:4175 BUG=skia:4284 ===================================================================== SkScanlineDecoder.h/.cpp: Removed. SkCodec.h/.cpp: Add methods, enums, and variables which were previously in SkScanlineDecoder. Default fCurrScanline to -1, as a sentinel that start has not been called. General changes: Convert SkScanlineDecoders to SkCodecs. General changes in SkCodec subclasses: Merge SkScanlineDecoder implementation into SkCodec. Most (all?) owned an SkCodec, so they now call this-> instead of fCodec->. SkBmpCodec.h/.cpp: Replace the unused rowOrder method with an override for onGetScanlineOrder. Make getDstRow const, since it is called by onGetY, which is const. SkCodec_libpng.h/.cpp: Make SkPngCodec an abstract class, with two subclasses which handle scanline decoding separately (they share code for decoding the entire image). Reimplement onReallyHasAlpha so that it can return the most recent result (e.g. after a scanline decode which only decoded part of the image) or a better answer (e.g. if the whole image is known to be opaque). Compute fNumberPasses early, so we know which subclass to instantiate. Make SkPngInterlaceScanlineDecoder use the base class' fCurrScanline rather than a separate variable. CodexTest.cpp: Add tests for the state changes in SkCodec (need to call start before decoding scanlines; calling getPixels means that start will need to be called again before decoding more scanlines). Add a test which decodes in stripes, currently only used for an interlaced PNG. TODO: Add tests for onReallyHasAlpha. Review URL: https://codereview.chromium.org/1365313002
2015-09-30 15:57:13 +00:00
const SkCodec::Result startResult = codec->startScanlineDecode(decodeInfo, nullptr,
colorPtr, &colorCount);
if (SkCodec::kSuccess != startResult) {
return "Failed to restart scanline decoder with same parameters.";
}
for (int i = 0; i < numStripes; i += 2) {
// Read a stripe
const int startY = i * stripeHeight;
const int linesToRead = SkTMin(stripeHeight, height - startY);
codec->getScanlines(SkTAddOffset<void>(dst, rowBytes * startY), linesToRead,
rowBytes);
// Skip a stripe
const int linesToSkip = SkTMin(stripeHeight, height - (i + 1) * stripeHeight);
if (linesToSkip > 0) {
codec->skipScanlines(linesToSkip);
}
}
draw_to_canvas(canvas, bitmapInfo, dst, rowBytes, colorPtr, colorCount, fDstColorType);
break;
}
case kCroppedScanline_Mode: {
const int width = decodeInfo.width();
const int height = decodeInfo.height();
// This value is chosen because, as we move across the image, it will sometimes
// align with the jpeg block sizes and it will sometimes not. This allows us
// to test interestingly different code paths in the implementation.
const int tileSize = 36;
SkCodec::Options opts;
SkIRect subset;
for (int x = 0; x < width; x += tileSize) {
subset = SkIRect::MakeXYWH(x, 0, SkTMin(tileSize, width - x), height);
opts.fSubset = &subset;
if (SkCodec::kSuccess != codec->startScanlineDecode(decodeInfo, &opts,
colorPtr, &colorCount)) {
return "Could not start scanline decoder.";
}
codec->getScanlines(SkTAddOffset<void>(pixels.get(), x * bpp), height, rowBytes);
}
draw_to_canvas(canvas, bitmapInfo, pixels.get(), rowBytes, colorPtr, colorCount,
fDstColorType);
break;
}
case kSubset_Mode: {
// Arbitrarily choose a divisor.
int divisor = 2;
// Total width/height of the image.
const int W = codec->getInfo().width();
const int H = codec->getInfo().height();
if (divisor > W || divisor > H) {
return Error::Nonfatal(SkStringPrintf("Cannot codec subset: divisor %d is too big "
"for %s with dimensions (%d x %d)", divisor,
fPath.c_str(), W, H));
}
// subset dimensions
// SkWebpCodec, the only one that supports subsets, requires even top/left boundaries.
const int w = SkAlign2(W / divisor);
const int h = SkAlign2(H / divisor);
SkIRect subset;
SkCodec::Options opts;
opts.fSubset = &subset;
SkBitmap subsetBm;
// We will reuse pixel memory from bitmap.
void* dst = pixels.get();
// Keep track of left and top (for drawing subsetBm into canvas). We could use
// fScale * x and fScale * y, but we want integers such that the next subset will start
// where the last one ended. So we'll add decodeInfo.width() and height().
int left = 0;
for (int x = 0; x < W; x += w) {
int top = 0;
for (int y = 0; y < H; y+= h) {
// Do not make the subset go off the edge of the image.
const int preScaleW = SkTMin(w, W - x);
const int preScaleH = SkTMin(h, H - y);
subset.setXYWH(x, y, preScaleW, preScaleH);
// And scale
// FIXME: Should we have a version of getScaledDimensions that takes a subset
// into account?
const int scaledW = SkTMax(1, SkScalarRoundToInt(preScaleW * fScale));
const int scaledH = SkTMax(1, SkScalarRoundToInt(preScaleH * fScale));
decodeInfo = decodeInfo.makeWH(scaledW, scaledH);
SkImageInfo subsetBitmapInfo = bitmapInfo.makeWH(scaledW, scaledH);
size_t subsetRowBytes = subsetBitmapInfo.minRowBytes();
const SkCodec::Result result = codec->getPixels(decodeInfo, dst, subsetRowBytes,
&opts, colorPtr, &colorCount);
switch (result) {
case SkCodec::kSuccess:
case SkCodec::kIncompleteInput:
break;
default:
return SkStringPrintf("subset codec failed to decode (%d, %d, %d, %d) "
"from %s with dimensions (%d x %d)\t error %d",
x, y, decodeInfo.width(), decodeInfo.height(),
fPath.c_str(), W, H, result);
}
draw_to_canvas(canvas, subsetBitmapInfo, dst, subsetRowBytes, colorPtr,
colorCount, fDstColorType, SkIntToScalar(left),
SkIntToScalar(top));
// translate by the scaled height.
top += decodeInfo.height();
}
// translate by the scaled width.
left += decodeInfo.width();
}
return "";
}
default:
SkASSERT(false);
return "Invalid fMode";
}
return "";
}
SkISize CodecSrc::size() const {
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
std::unique_ptr<SkCodec> codec(SkCodec::NewFromData(encoded));
if (nullptr == codec) {
return SkISize::Make(0, 0);
}
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
auto imageSize = codec->getScaledDimensions(fScale);
if (fMode == kAnimated_Mode) {
// We'll draw one of each frame, so make it big enough to hold them all
// in a grid. The grid will be roughly square, with "factor" frames per
// row and up to "factor" rows.
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
const size_t count = codec->getFrameInfo().size();
const float root = sqrt((float) count);
const int factor = sk_float_ceil2int(root);
imageSize.fWidth = imageSize.fWidth * factor;
imageSize.fHeight = imageSize.fHeight * sk_float_ceil2int((float) count / (float) factor);
Add support for multiple frames in SkCodec Add an interface to decode frames beyond the first in SkCodec, and add an implementation for SkGifCodec. Add getFrameData to SkCodec. This method reads ahead in the stream to return a vector containing meta data about each frame in the image. This is not required in order to decode frames beyond the first, but it allows a client to learn extra information: - how long the frame should be displayed - whether a frame should be blended with a prior frame, allowing the client to provide the prior frame to speed up decoding Add a new fields to SkCodec::Options: - fFrameIndex - fHasPriorFrame The API is designed so that SkCodec never caches frames. If a client wants a frame beyond the first, they specify the frame in Options.fFrameIndex. If the client does not have the frame's required frame (the frame that this frame must be blended on top of) cached, they pass false for Options.fHasPriorFrame. Unless the frame is independent, the codec will then recursively decode all frames necessary to decode fFrameIndex. If the client has the required frame cached, they can put it in the dst they pass to the codec, and the codec will only draw fFrameIndex onto it. Replace SkGifCodec's scanline decoding support with progressive decoding, and update the tests accordingly. Implement new APIs in SkGifCodec. Instead of using gif_lib, use GIFImageReader, imported from Chromium (along with its copyright headers) with the following changes: - SkGifCodec is now the client - Replace blink types - Combine GIFColorMap::buildTable and ::getTable into a method that creates and returns an SkColorTable - Input comes from an SkStream, instead of a SegmentReader. Add SkStreamBuffer, which buffers the (potentially partial) stream in order to decode progressively. (FIXME: This requires copying data that previously was read directly from the SegmentReader. Does this hurt performance? If so, can we fix it?) - Remove UMA code - Instead of reporting screen width and height to the client, allow the client to query for it - Fail earlier if the first frame AND screen have size of zero - Compute required previous frame when adding a new one - Move GIFParseQuery from GIFImageDecoder to GIFImageReader - Allow parsing up to a specific frame (to skip parsing the rest of the stream if a client only wants the first frame) - Compute whether the first frame has alpha and supports index 8, to create the SkImageInfo. This happens before reporting that the size has been decoded. Add GIFImageDecoder::haveDecodedRow to SkGifCodec, imported from Chromium (along with its copyright header), with the following changes: - Add support for sampling - Use the swizzler - Keep track of the rows decoded - Do *not* keep track of whether we've seen alpha Remove SkCodec::kOutOfOrder_SkScanlineOrder, which was only used by GIF scanline decoding. Call onRewind even if there is no stream (SkGifCodec needs to clear its decoded state so it will decode from the beginning). Add a method to SkSwizzler to access the offset into the dst, taking subsetting into account. Add a GM that animates a GIF. Add tests for the new APIs. *** Behavior changes: * Previously, we reported that an image with a subset frame and no transparent index was opaque and used the background index (if present) to fill the background. This is necessary in order to support index 8, but it does not match viewers/browsers I have seen. Examples: - Chromium and Gimp render the background transparent - Firefox, Safari, Linux Image Viewer, Safari Preview clip to the frame (for a single frame image) This CL matches Chromium's behavior and renders the background transparent. This allows us to have consistent behavior across products and simplifies the code (relative to what we would have to do to continue the old behavior on Android). It also means that we will no longer support index 8 for some GIFs. * Stop checking for GIFSTAMP - all GIFs should be either 89a or 87a. This matches Chromium. I suspect that bugs would have been reported if valid GIFs started with "GIFVER" instead of "GIF89a" or "GIF87a" (but did not decode in Chromium). *** Future work not included in this CL: * Move some checks out of haveDecodedRow, since they are the same for the entire frame e.g. - intersecting the frameRect with the full image size - whether there is a color table * Change when we write transparent pixels - In some cases, Chromium deemed this unnecessary, but I suspect it is slower than the fallback case. There will continue to be cases where we should *not* write them, but for e.g. the first pass where we have already cleared to transparent (which we may also be able to skip) writing the transparent pixels will not make anything incorrect. * Report color type and alpha type per frame - Depending on alpha values, disposal methods, frame rects, etc, subsequent frames may have different properties than the first. * Skip copies of the encoded data - We copy the encoded data in case the stream is one that cannot be rewound, so we can parse and then decode (possibly not immediately). For some input streams, this is unnecessary. - I was concerned this cause a performance regression, but on average the new code is faster than the old for the images I tested [1]. - It may cause a performance regression for Chromium, though, where we can always move back in the stream, so this should be addressed. Design doc: https://docs.google.com/a/google.com/document/d/12Qhf9T92MWfdWujQwCIjhCO3sw6pTJB5pJBwDM1T7Kc/ [1] https://docs.google.com/a/google.com/spreadsheets/d/19V-t9BfbFw5eiwBTKA1qOBkZbchjlTC5EIz6HFy-6RI/ GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=2045293002 Review-Url: https://codereview.chromium.org/2045293002
2016-10-24 16:03:26 +00:00
}
return imageSize;
}
Name CodecSrc::name() const {
if (1.0f == fScale) {
Name name = SkOSPath::Basename(fPath.c_str());
if (fMode == kAnimated_Mode) {
name.append("_animated");
}
return name;
}
SkASSERT(fMode != kAnimated_Mode);
return get_scaled_name(fPath, fScale);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
AndroidCodecSrc::AndroidCodecSrc(Path path, CodecSrc::DstColorType dstColorType,
SkAlphaType dstAlphaType, int sampleSize)
: fPath(path)
, fDstColorType(dstColorType)
, fDstAlphaType(dstAlphaType)
, fSampleSize(sampleSize)
, fRunSerially(serial_from_path_name(path))
{}
bool AndroidCodecSrc::veto(SinkFlags flags) const {
// No need to test decoding to non-raster or indirect backend.
return flags.type != SinkFlags::kRaster
|| flags.approach != SinkFlags::kDirect;
}
Error AndroidCodecSrc::draw(SkCanvas* canvas) const {
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
if (!encoded) {
return SkStringPrintf("Couldn't read %s.", fPath.c_str());
}
std::unique_ptr<SkAndroidCodec> codec(SkAndroidCodec::NewFromData(encoded));
if (nullptr == codec.get()) {
return SkStringPrintf("Couldn't create android codec for %s.", fPath.c_str());
}
SkImageInfo decodeInfo = codec->getInfo();
if (!get_decode_info(&decodeInfo, canvas->imageInfo().colorType(), fDstColorType,
fDstAlphaType)) {
return Error::Nonfatal("Testing non-565 to 565 is uninteresting.");
}
// Scale the image if it is desired.
SkISize size = codec->getSampledDimensions(fSampleSize);
// Visually inspecting very small output images is not necessary. We will
// cover these cases in unit testing.
if ((size.width() <= 10 || size.height() <= 10) && 1 != fSampleSize) {
return Error::Nonfatal("Scaling very small images is uninteresting.");
}
decodeInfo = decodeInfo.makeWH(size.width(), size.height());
int bpp = SkColorTypeBytesPerPixel(decodeInfo.colorType());
size_t rowBytes = size.width() * bpp;
SkAutoMalloc pixels(size.height() * rowBytes);
SkPMColor colorPtr[256];
int colorCount = 256;
SkBitmap bitmap;
SkImageInfo bitmapInfo = decodeInfo;
if (kRGBA_8888_SkColorType == decodeInfo.colorType() ||
kBGRA_8888_SkColorType == decodeInfo.colorType()) {
bitmapInfo = bitmapInfo.makeColorType(kN32_SkColorType);
}
// Create options for the codec.
SkAndroidCodec::AndroidOptions options;
options.fColorPtr = colorPtr;
options.fColorCount = &colorCount;
options.fSampleSize = fSampleSize;
switch (codec->getAndroidPixels(decodeInfo, pixels.get(), rowBytes, &options)) {
case SkCodec::kSuccess:
case SkCodec::kIncompleteInput:
break;
default:
return SkStringPrintf("Couldn't getPixels %s.", fPath.c_str());
}
draw_to_canvas(canvas, bitmapInfo, pixels.get(), rowBytes, colorPtr, colorCount, fDstColorType);
return "";
}
SkISize AndroidCodecSrc::size() const {
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
std::unique_ptr<SkAndroidCodec> codec(SkAndroidCodec::NewFromData(encoded));
if (nullptr == codec) {
return SkISize::Make(0, 0);
}
return codec->getSampledDimensions(fSampleSize);
}
Name AndroidCodecSrc::name() const {
// We will replicate the names used by CodecSrc so that images can
// be compared in Gold.
if (1 == fSampleSize) {
return SkOSPath::Basename(fPath.c_str());
}
return get_scaled_name(fPath, 1.0f / (float) fSampleSize);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
ImageGenSrc::ImageGenSrc(Path path, Mode mode, SkAlphaType alphaType, bool isGpu)
: fPath(path)
, fMode(mode)
, fDstAlphaType(alphaType)
, fIsGpu(isGpu)
, fRunSerially(serial_from_path_name(path))
{}
bool ImageGenSrc::veto(SinkFlags flags) const {
if (fIsGpu) {
return flags.type != SinkFlags::kGPU || flags.approach != SinkFlags::kDirect;
}
return flags.type != SinkFlags::kRaster || flags.approach != SinkFlags::kDirect;
}
Error ImageGenSrc::draw(SkCanvas* canvas) const {
if (kRGB_565_SkColorType == canvas->imageInfo().colorType()) {
return Error::Nonfatal("Uninteresting to test image generator to 565.");
}
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
if (!encoded) {
return SkStringPrintf("Couldn't read %s.", fPath.c_str());
}
#if defined(SK_BUILD_FOR_WIN)
// Initialize COM in order to test with WIC.
SkAutoCoInitialize com;
if (!com.succeeded()) {
return "Could not initialize COM.";
}
#endif
std::unique_ptr<SkImageGenerator> gen(nullptr);
switch (fMode) {
case kCodec_Mode:
gen.reset(SkCodecImageGenerator::NewFromEncodedCodec(encoded.get()));
if (!gen) {
return "Could not create codec image generator.";
}
break;
case kPlatform_Mode: {
#if defined(SK_BUILD_FOR_MAC) || defined(SK_BUILD_FOR_IOS)
gen.reset(SkImageGeneratorCG::NewFromEncodedCG(encoded.get()));
#elif defined(SK_BUILD_FOR_WIN)
gen.reset(SkImageGeneratorWIC::NewFromEncodedWIC(encoded.get()));
#endif
if (!gen) {
return "Could not create platform image generator.";
}
break;
}
default:
SkASSERT(false);
return "Invalid image generator mode";
}
// Test deferred decoding path on GPU
if (fIsGpu) {
sk_sp<SkImage> image(SkImage::MakeFromGenerator(gen.release(), nullptr));
if (!image) {
return "Could not create image from codec image generator.";
}
canvas->drawImage(image, 0, 0);
return "";
}
// Test various color and alpha types on CPU
SkImageInfo decodeInfo = gen->getInfo().makeAlphaType(fDstAlphaType);
int bpp = SkColorTypeBytesPerPixel(decodeInfo.colorType());
size_t rowBytes = decodeInfo.width() * bpp;
SkAutoMalloc pixels(decodeInfo.height() * rowBytes);
SkPMColor colorPtr[256];
int colorCount = 256;
if (!gen->getPixels(decodeInfo, pixels.get(), rowBytes, colorPtr, &colorCount)) {
return SkStringPrintf("Image generator could not getPixels() for %s\n", fPath.c_str());
}
draw_to_canvas(canvas, decodeInfo, pixels.get(), rowBytes, colorPtr, colorCount,
CodecSrc::kGetFromCanvas_DstColorType);
return "";
}
SkISize ImageGenSrc::size() const {
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
std::unique_ptr<SkCodec> codec(SkCodec::NewFromData(encoded));
if (nullptr == codec) {
return SkISize::Make(0, 0);
}
return codec->getInfo().dimensions();
}
Name ImageGenSrc::name() const {
return SkOSPath::Basename(fPath.c_str());
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
ColorCodecSrc::ColorCodecSrc(Path path, Mode mode, SkColorType colorType)
: fPath(path)
, fMode(mode)
, fColorType(colorType)
{}
bool ColorCodecSrc::veto(SinkFlags flags) const {
// Test to direct raster backends (8888 and 565).
return flags.type != SinkFlags::kRaster || flags.approach != SinkFlags::kDirect;
}
Error ColorCodecSrc::draw(SkCanvas* canvas) const {
if (kRGB_565_SkColorType == canvas->imageInfo().colorType()) {
return Error::Nonfatal("No need to test color correction to 565 backend.");
}
bool runInLegacyMode = kBaseline_Mode == fMode;
if (runInLegacyMode && canvas->imageInfo().colorSpace()) {
return Error::Nonfatal("Skipping tests that are only interesting in legacy mode.");
} else if (!runInLegacyMode && !canvas->imageInfo().colorSpace()) {
return Error::Nonfatal("Skipping tests that are only interesting in srgb mode.");
}
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
if (!encoded) {
return SkStringPrintf("Couldn't read %s.", fPath.c_str());
}
std::unique_ptr<SkCodec> codec(SkCodec::NewFromData(encoded));
if (nullptr == codec.get()) {
return SkStringPrintf("Couldn't create codec for %s.", fPath.c_str());
}
// Load the dst ICC profile. This particular dst is fairly similar to Adobe RGB.
sk_sp<SkData> dstData = SkData::MakeFromFileName(
GetResourcePath("icc_profiles/HP_ZR30w.icc").c_str());
if (!dstData) {
return "Cannot read monitor profile. Is the resource path set correctly?";
}
sk_sp<SkColorSpace> dstSpace = nullptr;
if (kDst_sRGB_Mode == fMode) {
dstSpace = SkColorSpace::MakeNamed(SkColorSpace::kSRGB_Named);
} else if (kDst_HPZR30w_Mode == fMode) {
dstSpace = SkColorSpace::MakeICC(dstData->data(), dstData->size());
}
SkImageInfo decodeInfo = codec->getInfo().makeColorType(fColorType).makeColorSpace(dstSpace);
if (kUnpremul_SkAlphaType == decodeInfo.alphaType()) {
decodeInfo = decodeInfo.makeAlphaType(kPremul_SkAlphaType);
}
if (kRGBA_F16_SkColorType == fColorType) {
SkASSERT(SkColorSpace_Base::Type::kXYZ == as_CSB(decodeInfo.colorSpace())->type());
SkColorSpace_XYZ* csXYZ = static_cast<SkColorSpace_XYZ*>(decodeInfo.colorSpace());
decodeInfo = decodeInfo.makeColorSpace(csXYZ->makeLinearGamma());
}
SkImageInfo bitmapInfo = decodeInfo;
if (kRGBA_8888_SkColorType == decodeInfo.colorType() ||
kBGRA_8888_SkColorType == decodeInfo.colorType())
{
bitmapInfo = bitmapInfo.makeColorType(kN32_SkColorType);
}
SkBitmap bitmap;
if (!bitmap.tryAllocPixels(bitmapInfo)) {
return SkStringPrintf("Image(%s) is too large (%d x %d)", fPath.c_str(),
bitmapInfo.width(), bitmapInfo.height());
}
size_t rowBytes = bitmap.rowBytes();
SkCodec::Result r = codec->getPixels(decodeInfo, bitmap.getPixels(), rowBytes);
if (SkCodec::kSuccess != r && SkCodec::kIncompleteInput != r) {
return SkStringPrintf("Couldn't getPixels %s. Error code %d", fPath.c_str(), r);
}
switch (fMode) {
case kBaseline_Mode:
case kDst_sRGB_Mode:
case kDst_HPZR30w_Mode:
canvas->drawBitmap(bitmap, 0, 0);
break;
default:
SkASSERT(false);
return "Invalid fMode";
}
return "";
}
SkISize ColorCodecSrc::size() const {
sk_sp<SkData> encoded(SkData::MakeFromFileName(fPath.c_str()));
std::unique_ptr<SkCodec> codec(SkCodec::NewFromData(encoded));
if (nullptr == codec) {
return SkISize::Make(0, 0);
}
return SkISize::Make(codec->getInfo().width(), codec->getInfo().height());
}
Name ColorCodecSrc::name() const {
return SkOSPath::Basename(fPath.c_str());
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
static const SkRect kSKPViewport = {0,0, 1000,1000};
SKPSrc::SKPSrc(Path path) : fPath(path) {}
Error SKPSrc::draw(SkCanvas* canvas) const {
std::unique_ptr<SkStream> stream = SkStream::MakeFromFile(fPath.c_str());
if (!stream) {
return SkStringPrintf("Couldn't read %s.", fPath.c_str());
}
sk_sp<SkPicture> pic(SkPicture::MakeFromStream(stream.get()));
if (!pic) {
return SkStringPrintf("Couldn't decode %s as a picture.", fPath.c_str());
}
stream = nullptr; // Might as well drop this when we're done with it.
canvas->clipRect(kSKPViewport);
canvas->drawPicture(pic);
return "";
}
SkISize SKPSrc::size() const {
std::unique_ptr<SkStream> stream = SkStream::MakeFromFile(fPath.c_str());
if (!stream) {
return SkISize::Make(0,0);
}
SkPictInfo info;
if (!SkPicture::InternalOnly_StreamIsSKP(stream.get(), &info)) {
return SkISize::Make(0,0);
}
SkRect viewport = kSKPViewport;
if (!viewport.intersect(info.fCullRect)) {
return SkISize::Make(0,0);
}
return viewport.roundOut().size();
}
Name SKPSrc::name() const { return SkOSPath::Basename(fPath.c_str()); }
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
#if defined(SK_XML)
// Used when the image doesn't have an intrinsic size.
static const SkSize kDefaultSVGSize = SkSize::Make(1000, 1000);
// Used to force-scale tiny fixed-size images.
static const SkSize kMinimumSVGSize = SkSize::Make(128, 128);
SVGSrc::SVGSrc(Path path)
: fName(SkOSPath::Basename(path.c_str()))
, fScale(1) {
SkFILEStream stream(path.c_str());
if (!stream.isValid()) {
return;
}
fDom = SkSVGDOM::MakeFromStream(stream);
if (!fDom) {
return;
}
const SkSize& sz = fDom->containerSize();
if (sz.isEmpty()) {
// no intrinsic size
fDom->setContainerSize(kDefaultSVGSize);
} else {
fScale = SkTMax(1.f, SkTMax(kMinimumSVGSize.width() / sz.width(),
kMinimumSVGSize.height() / sz.height()));
}
}
Error SVGSrc::draw(SkCanvas* canvas) const {
if (!fDom) {
return SkStringPrintf("Unable to parse file: %s", fName.c_str());
}
SkAutoCanvasRestore acr(canvas, true);
canvas->scale(fScale, fScale);
fDom->render(canvas);
return "";
}
SkISize SVGSrc::size() const {
if (!fDom) {
return SkISize::Make(0, 0);
}
return SkSize::Make(fDom->containerSize().width() * fScale,
fDom->containerSize().height() * fScale).toRound();
}
Name SVGSrc::name() const { return fName; }
bool SVGSrc::veto(SinkFlags flags) const {
// No need to test to non-(raster||gpu) or indirect backends.
bool type_ok = flags.type == SinkFlags::kRaster
|| flags.type == SinkFlags::kGPU;
return !type_ok || flags.approach != SinkFlags::kDirect;
}
#endif // defined(SK_XML)
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
MSKPSrc::MSKPSrc(Path path) : fPath(path) {
std::unique_ptr<SkStreamAsset> stream = SkStream::MakeFromFile(fPath.c_str());
(void)fReader.init(stream.get());
}
int MSKPSrc::pageCount() const { return fReader.pageCount(); }
SkISize MSKPSrc::size() const { return this->size(0); }
SkISize MSKPSrc::size(int i) const { return fReader.pageSize(i).toCeil(); }
Error MSKPSrc::draw(SkCanvas* c) const { return this->draw(0, c); }
Error MSKPSrc::draw(int i, SkCanvas* canvas) const {
std::unique_ptr<SkStreamAsset> stream = SkStream::MakeFromFile(fPath.c_str());
if (!stream) {
return SkStringPrintf("Unable to open file: %s", fPath.c_str());
}
if (fReader.pageCount() == 0) {
return SkStringPrintf("Unable to parse MultiPictureDocument file: %s", fPath.c_str());
}
if (i >= fReader.pageCount()) {
return SkStringPrintf("MultiPictureDocument page number out of range: %d", i);
}
sk_sp<SkPicture> page = fReader.readPage(stream.get(), i);
if (!page) {
return SkStringPrintf("SkMultiPictureDocumentReader failed on page %d: %s",
i, fPath.c_str());
}
canvas->drawPicture(page);
return "";
}
Name MSKPSrc::name() const { return SkOSPath::Basename(fPath.c_str()); }
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error NullSink::draw(const Src& src, SkBitmap*, SkWStream*, SkString*) const {
return src.draw(SkMakeNullCanvas().get());
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
DEFINE_bool(gpuStats, false, "Append GPU stats to the log for each GPU task?");
GPUSink::GPUSink(GrContextFactory::ContextType ct,
GrContextFactory::ContextOptions options,
int samples,
bool diText,
SkColorType colorType,
sk_sp<SkColorSpace> colorSpace,
bool threaded)
: fContextType(ct)
, fContextOptions(options)
, fSampleCount(samples)
, fUseDIText(diText)
, fColorType(colorType)
, fColorSpace(std::move(colorSpace))
, fThreaded(threaded) {}
DEFINE_bool(imm, false, "Run gpu configs in immediate mode.");
DEFINE_bool(drawOpClip, false, "Clip each GrDrawOp to its device bounds for testing.");
DEFINE_int32(opLookback, -1, "Maximum GrOp lookback for combining, negative means default.");
DEFINE_int32(opLookahead, -1, "Maximum GrOp lookahead for combining, negative means default.");
Error GPUSink::draw(const Src& src, SkBitmap* dst, SkWStream*, SkString* log) const {
GrContextOptions grOptions;
grOptions.fImmediateMode = FLAGS_imm;
grOptions.fClipDrawOpsToBounds = FLAGS_drawOpClip;
grOptions.fMaxOpCombineLookback = FLAGS_opLookback;
grOptions.fMaxOpCombineLookahead = FLAGS_opLookahead;
src.modifyGrContextOptions(&grOptions);
GrContextFactory factory(grOptions);
const SkISize size = src.size();
const SkImageInfo info =
SkImageInfo::Make(size.width(), size.height(), fColorType,
kPremul_SkAlphaType, fColorSpace);
#if SK_SUPPORT_GPU
GrContext* context = factory.getContextInfo(fContextType, fContextOptions).grContext();
const int maxDimension = context->caps()->maxTextureSize();
if (maxDimension < SkTMax(size.width(), size.height())) {
return Error::Nonfatal("Src too large to create a texture.\n");
}
#endif
auto surface(
Add config options to run different GPU APIs to dm and nanobench Add extended config specification form that can be used to run different gpu backend with different APIs. The configs can be specified with the form: gpu(api=string,dit=bool,nvpr=bool,samples=int) This replaces and removes the --gpuAPI flag. All existing configs should still work. Adds following documentation: out/Debug/dm --help config Flags: --config: type: string default: 565 8888 gpu nonrendering Options: 565 8888 debug gpu gpudebug gpudft gpunull msaa16 msaa4 nonrendering null nullgpu nvprmsaa16 nvprmsaa4 pdf pdf_poppler skp svg xps or use extended form 'backend(option=value,...)'. Extended form: 'backend(option=value,...)' Possible backends and options: gpu(api=string,dit=bool,nvpr=bool,samples=int) GPU backend api type: string default: native. Select graphics API to use with gpu backend. Options: native Use platform default OpenGL or OpenGL ES backend. gl Use OpenGL. gles Use OpenGL ES. debug Use debug OpenGL. null Use null OpenGL. dit type: bool default: false. Use device independent text. nvpr type: bool default: false. Use NV_path_rendering OpenGL and OpenGL ES extension. samples type: int default: 0. Use multisampling with N samples. Predefined configs: gpu = gpu() msaa4 = gpu(samples=4) msaa16 = gpu(samples=16) nvprmsaa4 = gpu(nvpr=true,samples=4) nvprmsaa16 = gpu(nvpr=true,samples=16) gpudft = gpu(dit=true) gpudebug = gpu(api=debug) gpunull = gpu(api=null) debug = gpu(api=debug) nullgpu = gpu(api=null) BUG=skia:2992 Committed: https://skia.googlesource.com/skia/+/e13ca329fca4c28cf4e078561f591ab27b743d23 GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1490113005 Committed: https://skia.googlesource.com/skia/+/c8b4336444e7b90382e04e33665fb3b8490b825b Committed: https://skia.googlesource.com/skia/+/9ebc3f0ee6db215dde461dc4777d85988cf272dd Review URL: https://codereview.chromium.org/1490113005
2015-12-23 09:33:00 +00:00
NewGpuSurface(&factory, fContextType, fContextOptions, info, fSampleCount, fUseDIText));
if (!surface) {
return "Could not create a surface.";
}
if (FLAGS_preAbandonGpuContext) {
factory.abandonContexts();
}
SkCanvas* canvas = surface->getCanvas();
Error err = src.draw(canvas);
if (!err.isEmpty()) {
return err;
}
canvas->flush();
if (FLAGS_gpuStats) {
canvas->getGrContext()->dumpCacheStats(log);
canvas->getGrContext()->dumpGpuStats(log);
}
dst->allocPixels(info);
canvas->readPixels(dst, 0, 0);
if (FLAGS_abandonGpuContext) {
factory.abandonContexts();
} else if (FLAGS_releaseAndAbandonGpuContext) {
factory.releaseResourcesAndAbandonContexts();
}
return "";
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
static Error draw_skdocument(const Src& src, SkDocument* doc, SkWStream* dst) {
if (src.size().isEmpty()) {
return "Source has empty dimensions";
}
SkASSERT(doc);
int pageCount = src.pageCount();
for (int i = 0; i < pageCount; ++i) {
int width = src.size(i).width(), height = src.size(i).height();
SkCanvas* canvas =
doc->beginPage(SkIntToScalar(width), SkIntToScalar(height));
if (!canvas) {
return "SkDocument::beginPage(w,h) returned nullptr";
}
Error err = src.draw(i, canvas);
if (!err.isEmpty()) {
return err;
}
doc->endPage();
}
doc->close();
dst->flush();
return "";
}
Error PDFSink::draw(const Src& src, SkBitmap*, SkWStream* dst, SkString*) const {
SkDocument::PDFMetadata metadata;
metadata.fTitle = src.name();
metadata.fSubject = "rendering correctness test";
metadata.fCreator = "Skia/DM";
sk_sp<SkDocument> doc = SkDocument::MakePDF(dst, SK_ScalarDefaultRasterDPI,
metadata, nullptr, fPDFA);
if (!doc) {
return "SkDocument::MakePDF() returned nullptr";
}
return draw_skdocument(src, doc.get(), dst);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
XPSSink::XPSSink() {}
Error XPSSink::draw(const Src& src, SkBitmap*, SkWStream* dst, SkString*) const {
sk_sp<SkDocument> doc(SkDocument::MakeXPS(dst));
if (!doc) {
return "SkDocument::MakeXPS() returned nullptr";
}
return draw_skdocument(src, doc.get(), dst);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
PipeSink::PipeSink() {}
Error PipeSink::draw(const Src& src, SkBitmap*, SkWStream* dst, SkString*) const {
return src.draw(SkPipeSerializer().beginWrite(SkRect::Make(src.size()), dst));
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
SKPSink::SKPSink() {}
Error SKPSink::draw(const Src& src, SkBitmap*, SkWStream* dst, SkString*) const {
SkSize size;
size = src.size();
SkPictureRecorder recorder;
Error err = src.draw(recorder.beginRecording(size.width(), size.height()));
if (!err.isEmpty()) {
return err;
}
recorder.finishRecordingAsPicture()->serialize(dst);
return "";
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error DebugSink::draw(const Src& src, SkBitmap*, SkWStream* dst, SkString*) const {
SkDebugCanvas debugCanvas(src.size().width(), src.size().height());
Error err = src.draw(&debugCanvas);
if (!err.isEmpty()) {
return err;
}
std::unique_ptr<SkCanvas> nullCanvas = SkMakeNullCanvas();
UrlDataManager dataManager(SkString("data"));
Json::Value json = debugCanvas.toJSON(
dataManager, debugCanvas.getSize(), nullCanvas.get());
std::string value = Json::StyledWriter().write(json);
return dst->write(value.c_str(), value.size()) ? "" : "SkWStream Error";
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
SVGSink::SVGSink() {}
Error SVGSink::draw(const Src& src, SkBitmap*, SkWStream* dst, SkString*) const {
#if defined(SK_XML)
std::unique_ptr<SkXMLWriter> xmlWriter(new SkXMLStreamWriter(dst));
return src.draw(SkSVGCanvas::Make(SkRect::MakeWH(SkIntToScalar(src.size().width()),
SkIntToScalar(src.size().height())),
xmlWriter.get()).get());
#else
return Error("SVG sink is disabled.");
#endif // SK_XML
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
RasterSink::RasterSink(SkColorType colorType, sk_sp<SkColorSpace> colorSpace)
: fColorType(colorType)
, fColorSpace(std::move(colorSpace)) {}
Error RasterSink::draw(const Src& src, SkBitmap* dst, SkWStream*, SkString*) const {
const SkISize size = src.size();
// If there's an appropriate alpha type for this color type, use it, otherwise use premul.
SkAlphaType alphaType = kPremul_SkAlphaType;
(void)SkColorTypeValidateAlphaType(fColorType, alphaType, &alphaType);
SkMallocPixelRef::ZeroedPRFactory factory;
dst->allocPixels(SkImageInfo::Make(size.width(), size.height(),
fColorType, alphaType, fColorSpace),
&factory,
nullptr/*colortable*/);
SkCanvas canvas(*dst);
return src.draw(&canvas);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
// Handy for front-patching a Src. Do whatever up-front work you need, then call draw_to_canvas(),
// passing the Sink draw() arguments, a size, and a function draws into an SkCanvas.
// Several examples below.
template <typename Fn>
static Error draw_to_canvas(Sink* sink, SkBitmap* bitmap, SkWStream* stream, SkString* log,
SkISize size, const Fn& draw) {
class ProxySrc : public Src {
public:
ProxySrc(SkISize size, const Fn& draw) : fSize(size), fDraw(draw) {}
Error draw(SkCanvas* canvas) const override { return fDraw(canvas); }
Name name() const override { return "ProxySrc"; }
SkISize size() const override { return fSize; }
private:
SkISize fSize;
const Fn& fDraw;
};
return sink->draw(ProxySrc(size, draw), bitmap, stream, log);
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
DEFINE_bool(check, true, "If true, have most Via- modes fail if they affect the output.");
// Is *bitmap identical to what you get drawing src into sink?
static Error check_against_reference(const SkBitmap* bitmap, const Src& src, Sink* sink) {
// We can only check raster outputs.
// (Non-raster outputs like .pdf, .skp, .svg may differ but still draw identically.)
if (FLAGS_check && bitmap) {
SkBitmap reference;
SkString log;
SkDynamicMemoryWStream wStream;
Error err = sink->draw(src, &reference, &wStream, &log);
// If we can draw into this Sink via some pipeline, we should be able to draw directly.
SkASSERT(err.isEmpty());
if (!err.isEmpty()) {
return err;
}
// The dimensions are a property of the Src only, and so should be identical.
SkASSERT(reference.getSize() == bitmap->getSize());
if (reference.getSize() != bitmap->getSize()) {
return "Dimensions don't match reference";
}
// All SkBitmaps in DM are pre-locked and tight, so this comparison is easy.
if (0 != memcmp(reference.getPixels(), bitmap->getPixels(), reference.getSize())) {
return "Pixels don't match reference";
}
}
return "";
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
static SkISize auto_compute_translate(SkMatrix* matrix, int srcW, int srcH) {
SkRect bounds = SkRect::MakeIWH(srcW, srcH);
matrix->mapRect(&bounds);
matrix->postTranslate(-bounds.x(), -bounds.y());
return SkISize::Make(SkScalarRoundToInt(bounds.width()), SkScalarRoundToInt(bounds.height()));
}
ViaMatrix::ViaMatrix(SkMatrix matrix, Sink* sink) : Via(sink), fMatrix(matrix) {}
Error ViaMatrix::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
SkMatrix matrix = fMatrix;
SkISize size = auto_compute_translate(&matrix, src.size().width(), src.size().height());
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) {
canvas->concat(matrix);
return src.draw(canvas);
});
}
// Undoes any flip or 90 degree rotate without changing the scale of the bitmap.
// This should be pixel-preserving.
ViaUpright::ViaUpright(SkMatrix matrix, Sink* sink) : Via(sink), fMatrix(matrix) {}
Error ViaUpright::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
Error err = fSink->draw(src, bitmap, stream, log);
if (!err.isEmpty()) {
return err;
}
SkMatrix inverse;
if (!fMatrix.rectStaysRect() || !fMatrix.invert(&inverse)) {
return "Cannot upright --matrix.";
}
SkMatrix upright = SkMatrix::I();
upright.setScaleX(SkScalarSignAsScalar(inverse.getScaleX()));
upright.setScaleY(SkScalarSignAsScalar(inverse.getScaleY()));
upright.setSkewX(SkScalarSignAsScalar(inverse.getSkewX()));
upright.setSkewY(SkScalarSignAsScalar(inverse.getSkewY()));
SkBitmap uprighted;
SkISize size = auto_compute_translate(&upright, bitmap->width(), bitmap->height());
uprighted.allocPixels(bitmap->info().makeWH(size.width(), size.height()));
SkCanvas canvas(uprighted);
canvas.concat(upright);
SkPaint paint;
paint.setBlendMode(SkBlendMode::kSrc);
canvas.drawBitmap(*bitmap, 0, 0, &paint);
*bitmap = uprighted;
bitmap->lockPixels();
return "";
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error ViaSerialization::draw(
const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
// Record our Src into a picture.
auto size = src.size();
SkPictureRecorder recorder;
Error err = src.draw(recorder.beginRecording(SkIntToScalar(size.width()),
SkIntToScalar(size.height())));
if (!err.isEmpty()) {
return err;
}
sk_sp<SkPicture> pic(recorder.finishRecordingAsPicture());
// Serialize it and then deserialize it.
sk_sp<SkPicture> deserialized(SkPicture::MakeFromData(pic->serialize().get()));
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) {
canvas->drawPicture(deserialized);
return check_against_reference(bitmap, src, fSink.get());
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
ViaTiles::ViaTiles(int w, int h, SkBBHFactory* factory, Sink* sink)
: Via(sink)
, fW(w)
, fH(h)
, fFactory(factory) {}
Error ViaTiles::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
SkPictureRecorder recorder;
Error err = src.draw(recorder.beginRecording(SkIntToScalar(size.width()),
SkIntToScalar(size.height()),
fFactory.get()));
if (!err.isEmpty()) {
return err;
}
sk_sp<SkPicture> pic(recorder.finishRecordingAsPicture());
return draw_to_canvas(fSink.get(), bitmap, stream, log, src.size(), [&](SkCanvas* canvas) {
const int xTiles = (size.width() + fW - 1) / fW,
yTiles = (size.height() + fH - 1) / fH;
SkMultiPictureDraw mpd(xTiles*yTiles);
SkTArray<sk_sp<SkSurface>> surfaces;
// surfaces.setReserve(xTiles*yTiles);
SkImageInfo info = canvas->imageInfo().makeWH(fW, fH);
for (int j = 0; j < yTiles; j++) {
for (int i = 0; i < xTiles; i++) {
// This lets our ultimate Sink determine the best kind of surface.
// E.g., if it's a GpuSink, the surfaces and images are textures.
auto s = canvas->makeSurface(info);
if (!s) {
s = SkSurface::MakeRaster(info); // Some canvases can't create surfaces.
}
surfaces.push_back(s);
SkCanvas* c = s->getCanvas();
c->translate(SkIntToScalar(-i * fW),
SkIntToScalar(-j * fH)); // Line up the canvas with this tile.
mpd.add(c, pic.get());
}
}
mpd.draw();
for (int j = 0; j < yTiles; j++) {
for (int i = 0; i < xTiles; i++) {
sk_sp<SkImage> image(surfaces[i+xTiles*j]->makeImageSnapshot());
canvas->drawImage(image, SkIntToScalar(i*fW), SkIntToScalar(j*fH));
}
}
return "";
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error ViaPicture::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) -> Error {
SkPictureRecorder recorder;
sk_sp<SkPicture> pic;
Error err = src.draw(recorder.beginRecording(SkIntToScalar(size.width()),
SkIntToScalar(size.height())));
if (!err.isEmpty()) {
return err;
}
pic = recorder.finishRecordingAsPicture();
canvas->drawPicture(pic);
return check_against_reference(bitmap, src, fSink.get());
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error ViaDefer::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) -> Error {
SkDeferredCanvas deferred(canvas);
return src.draw(&deferred);
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error ViaPipe::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) -> Error {
SkDynamicMemoryWStream tmpStream;
Error err = src.draw(SkPipeSerializer().beginWrite(SkRect::Make(size), &tmpStream));
if (!err.isEmpty()) {
return err;
}
sk_sp<SkData> data = tmpStream.detachAsData();
SkPipeDeserializer().playback(data->data(), data->size(), canvas);
return check_against_reference(bitmap, src, fSink.get());
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
// Draw the Src into two pictures, then draw the second picture into the wrapped Sink.
// This tests that any shortcuts we may take while recording that second picture are legal.
Error ViaSecondPicture::draw(
const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) -> Error {
SkPictureRecorder recorder;
sk_sp<SkPicture> pic;
for (int i = 0; i < 2; i++) {
Error err = src.draw(recorder.beginRecording(SkIntToScalar(size.width()),
SkIntToScalar(size.height())));
if (!err.isEmpty()) {
return err;
}
pic = recorder.finishRecordingAsPicture();
}
canvas->drawPicture(pic);
return check_against_reference(bitmap, src, fSink.get());
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
// Draw the Src twice. This can help exercise caching.
Error ViaTwice::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
return draw_to_canvas(fSink.get(), bitmap, stream, log, src.size(), [&](SkCanvas* canvas) -> Error {
for (int i = 0; i < 2; i++) {
SkAutoCanvasRestore acr(canvas, true/*save now*/);
canvas->clear(SK_ColorTRANSPARENT);
Error err = src.draw(canvas);
if (err.isEmpty()) {
return err;
}
}
return check_against_reference(bitmap, src, fSink.get());
});
}
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
// This is like SkRecords::Draw, in that it plays back SkRecords ops into a Canvas.
// Unlike SkRecords::Draw, it builds a single-op sub-picture out of each Draw-type op.
// This is an only-slightly-exaggerated simluation of Blink's Slimming Paint pictures.
struct DrawsAsSingletonPictures {
SkCanvas* fCanvas;
const SkDrawableList& fDrawables;
SkRect fBounds;
template <typename T>
void draw(const T& op, SkCanvas* canvas) {
// We must pass SkMatrix::I() as our initial matrix.
// By default SkRecords::Draw() uses the canvas' matrix as its initial matrix,
// which would have the funky effect of applying transforms over and over.
SkRecords::Draw d(canvas, nullptr, fDrawables.begin(), fDrawables.count(), &SkMatrix::I());
d(op);
}
// Draws get their own picture.
template <typename T>
SK_WHEN(T::kTags & SkRecords::kDraw_Tag, void) operator()(const T& op) {
SkPictureRecorder rec;
this->draw(op, rec.beginRecording(fBounds));
sk_sp<SkPicture> pic(rec.finishRecordingAsPicture());
fCanvas->drawPicture(pic);
}
// We'll just issue non-draws directly.
template <typename T>
skstd::enable_if_t<!(T::kTags & SkRecords::kDraw_Tag), void> operator()(const T& op) {
this->draw(op, fCanvas);
}
};
// Record Src into a picture, then record it into a macro picture with a sub-picture for each draw.
// Then play back that macro picture into our wrapped sink.
Error ViaSingletonPictures::draw(
const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) -> Error {
// Use low-level (Skia-private) recording APIs so we can read the SkRecord.
SkRecord skr;
SkRecorder recorder(&skr, size.width(), size.height());
Error err = src.draw(&recorder);
if (!err.isEmpty()) {
return err;
}
// Record our macro-picture, with each draw op as its own sub-picture.
SkPictureRecorder macroRec;
SkCanvas* macroCanvas = macroRec.beginRecording(SkIntToScalar(size.width()),
SkIntToScalar(size.height()));
std::unique_ptr<SkDrawableList> drawables(recorder.detachDrawableList());
const SkDrawableList empty;
DrawsAsSingletonPictures drawsAsSingletonPictures = {
macroCanvas,
drawables ? *drawables : empty,
SkRect::MakeWH((SkScalar)size.width(), (SkScalar)size.height()),
};
for (int i = 0; i < skr.count(); i++) {
skr.visit(i, drawsAsSingletonPictures);
}
sk_sp<SkPicture> macroPic(macroRec.finishRecordingAsPicture());
canvas->drawPicture(macroPic);
return check_against_reference(bitmap, src, fSink.get());
});
}
SkLite* SkLiteRecorder, a new SkCanvas, fills out SkLiteDL, a new SkDrawable. This SkDrawable is a display list similar to SkRecord and SkBigPicture / SkRecordedDrawable, but with a few new design points inspired by Android and slimming paint: 1) SkLiteDL is structured as one big contiguous array rather than the two layer structure of SkRecord. This trades away flexibility and large-op-count performance for better data locality for small to medium size pictures. 2) We keep a global freelist of SkLiteDLs, both reusing the SkLiteDL struct itself and its contiguous byte array. This keeps the expected number of mallocs per display list allocation <1 (really, ~0) for cyclical use cases. These two together mean recording is faster. Measuring against the code we use at head, SkLiteRecorder trends about ~3x faster across various size pictures, matching speed at 0 draws and beating the special-case 1-draw pictures we have today. (I.e. we won't need those special case implementations anymore, because they're slower than this new generic code.) This new strategy records 10 drawRects() in about the same time the old strategy took for 2. This strategy stays the winner until at least 500 drawRect()s on my laptop, where I stopped checking. A simpler alternative to freelisting is also possible (but not implemented here), where we allow the client to manually reset() an SkLiteDL for reuse when its refcnt is 1. That's essentially what we're doing with the freelist, except tracking what's available for reuse globally instead of making the client do it. This code is not fully capable yet, but most of the key design points are there. The internal structure of SkLiteDL is the area I expect to be most volatile (anything involving Op), but its interface and the whole of SkLiteRecorder ought to be just about done. You can run nanobench --match picture_overhead as a demo. Everything it exercises is fully fleshed out, so what it tests is an apples-to-apples comparison as far as recording costs go. I have not yet compared playback performance. It should be simple to wrap this into an SkPicture subclass if we want. I won't start proposing we replace anything old with anything new quite yet until I have more ducks in a row, but this does look pretty promising (similar to the SkRecord over old SkPicture change a couple years ago) and I'd like to land, experiment, iterate, especially with an eye toward Android. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2213333002 Review-Url: https://codereview.chromium.org/2213333002
2016-08-06 19:51:51 +00:00
/*~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~*/
Error ViaLite::draw(const Src& src, SkBitmap* bitmap, SkWStream* stream, SkString* log) const {
auto size = src.size();
SkRect bounds = {0,0, (SkScalar)size.width(), (SkScalar)size.height()};
return draw_to_canvas(fSink.get(), bitmap, stream, log, size, [&](SkCanvas* canvas) -> Error {
sk_sp<SkLiteDL> dl = SkLiteDL::New(bounds);
SkLiteRecorder rec;
rec.reset(dl.get());
Error err = src.draw(&rec);
SkLite* SkLiteRecorder, a new SkCanvas, fills out SkLiteDL, a new SkDrawable. This SkDrawable is a display list similar to SkRecord and SkBigPicture / SkRecordedDrawable, but with a few new design points inspired by Android and slimming paint: 1) SkLiteDL is structured as one big contiguous array rather than the two layer structure of SkRecord. This trades away flexibility and large-op-count performance for better data locality for small to medium size pictures. 2) We keep a global freelist of SkLiteDLs, both reusing the SkLiteDL struct itself and its contiguous byte array. This keeps the expected number of mallocs per display list allocation <1 (really, ~0) for cyclical use cases. These two together mean recording is faster. Measuring against the code we use at head, SkLiteRecorder trends about ~3x faster across various size pictures, matching speed at 0 draws and beating the special-case 1-draw pictures we have today. (I.e. we won't need those special case implementations anymore, because they're slower than this new generic code.) This new strategy records 10 drawRects() in about the same time the old strategy took for 2. This strategy stays the winner until at least 500 drawRect()s on my laptop, where I stopped checking. A simpler alternative to freelisting is also possible (but not implemented here), where we allow the client to manually reset() an SkLiteDL for reuse when its refcnt is 1. That's essentially what we're doing with the freelist, except tracking what's available for reuse globally instead of making the client do it. This code is not fully capable yet, but most of the key design points are there. The internal structure of SkLiteDL is the area I expect to be most volatile (anything involving Op), but its interface and the whole of SkLiteRecorder ought to be just about done. You can run nanobench --match picture_overhead as a demo. Everything it exercises is fully fleshed out, so what it tests is an apples-to-apples comparison as far as recording costs go. I have not yet compared playback performance. It should be simple to wrap this into an SkPicture subclass if we want. I won't start proposing we replace anything old with anything new quite yet until I have more ducks in a row, but this does look pretty promising (similar to the SkRecord over old SkPicture change a couple years ago) and I'd like to land, experiment, iterate, especially with an eye toward Android. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2213333002 Review-Url: https://codereview.chromium.org/2213333002
2016-08-06 19:51:51 +00:00
if (!err.isEmpty()) {
return err;
}
dl->draw(canvas);
return check_against_reference(bitmap, src, fSink.get());
SkLite* SkLiteRecorder, a new SkCanvas, fills out SkLiteDL, a new SkDrawable. This SkDrawable is a display list similar to SkRecord and SkBigPicture / SkRecordedDrawable, but with a few new design points inspired by Android and slimming paint: 1) SkLiteDL is structured as one big contiguous array rather than the two layer structure of SkRecord. This trades away flexibility and large-op-count performance for better data locality for small to medium size pictures. 2) We keep a global freelist of SkLiteDLs, both reusing the SkLiteDL struct itself and its contiguous byte array. This keeps the expected number of mallocs per display list allocation <1 (really, ~0) for cyclical use cases. These two together mean recording is faster. Measuring against the code we use at head, SkLiteRecorder trends about ~3x faster across various size pictures, matching speed at 0 draws and beating the special-case 1-draw pictures we have today. (I.e. we won't need those special case implementations anymore, because they're slower than this new generic code.) This new strategy records 10 drawRects() in about the same time the old strategy took for 2. This strategy stays the winner until at least 500 drawRect()s on my laptop, where I stopped checking. A simpler alternative to freelisting is also possible (but not implemented here), where we allow the client to manually reset() an SkLiteDL for reuse when its refcnt is 1. That's essentially what we're doing with the freelist, except tracking what's available for reuse globally instead of making the client do it. This code is not fully capable yet, but most of the key design points are there. The internal structure of SkLiteDL is the area I expect to be most volatile (anything involving Op), but its interface and the whole of SkLiteRecorder ought to be just about done. You can run nanobench --match picture_overhead as a demo. Everything it exercises is fully fleshed out, so what it tests is an apples-to-apples comparison as far as recording costs go. I have not yet compared playback performance. It should be simple to wrap this into an SkPicture subclass if we want. I won't start proposing we replace anything old with anything new quite yet until I have more ducks in a row, but this does look pretty promising (similar to the SkRecord over old SkPicture change a couple years ago) and I'd like to land, experiment, iterate, especially with an eye toward Android. BUG=skia: GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2213333002 Review-Url: https://codereview.chromium.org/2213333002
2016-08-06 19:51:51 +00:00
});
}
} // namespace DM