skia2/tests/ImageTest.cpp
Greg Daniel 719239cd69 Move all Ganesh source files into ganesh subdirectory.
Change-Id: I238d29ba0250224fa593845ae65192653f58faff
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/528156
Reviewed-by: Kevin Lubick <kjlubick@google.com>
Reviewed-by: Jim Van Verth <jvanverth@google.com>
Commit-Queue: Greg Daniel <egdaniel@google.com>
2022-04-07 21:06:50 +00:00

1549 lines
63 KiB
C++

/*
* 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 <functional>
#include <initializer_list>
#include <vector>
#include "include/core/SkBitmap.h"
#include "include/core/SkCanvas.h"
#include "include/core/SkData.h"
#include "include/core/SkImageEncoder.h"
#include "include/core/SkImageGenerator.h"
#include "include/core/SkPicture.h"
#include "include/core/SkPictureRecorder.h"
#include "include/core/SkRRect.h"
#include "include/core/SkSerialProcs.h"
#include "include/core/SkStream.h"
#include "include/core/SkSurface.h"
#include "include/gpu/GrContextThreadSafeProxy.h"
#include "include/gpu/GrDirectContext.h"
#include "src/core/SkAutoPixmapStorage.h"
#include "src/core/SkColorSpacePriv.h"
#include "src/core/SkImagePriv.h"
#include "src/core/SkOpts.h"
#include "src/gpu/ganesh/GrDirectContextPriv.h"
#include "src/gpu/ganesh/GrGpu.h"
#include "src/gpu/ganesh/GrImageContextPriv.h"
#include "src/gpu/ganesh/GrRecordingContextPriv.h"
#include "src/gpu/ganesh/GrResourceCache.h"
#include "src/gpu/ganesh/GrTexture.h"
#include "src/gpu/ganesh/SkGr.h"
#include "src/image/SkImage_Base.h"
#include "src/image/SkImage_GpuYUVA.h"
#include "tests/Test.h"
#include "tools/Resources.h"
#include "tools/ToolUtils.h"
#include "tools/gpu/ManagedBackendTexture.h"
#include "tools/gpu/ProxyUtils.h"
using namespace sk_gpu_test;
SkImageInfo read_pixels_info(SkImage* image) {
if (image->colorSpace()) {
return SkImageInfo::MakeS32(image->width(), image->height(), image->alphaType());
}
return SkImageInfo::MakeN32(image->width(), image->height(), image->alphaType());
}
// image `b` is assumed to be raster
static void assert_equal(skiatest::Reporter* reporter, GrDirectContext* dContextA, SkImage* a,
const SkIRect* subsetA, SkImage* b) {
const int widthA = subsetA ? subsetA->width() : a->width();
const int heightA = subsetA ? subsetA->height() : a->height();
REPORTER_ASSERT(reporter, widthA == b->width());
REPORTER_ASSERT(reporter, heightA == b->height());
// see https://bug.skia.org/3965
//REPORTER_ASSERT(reporter, a->isOpaque() == b->isOpaque());
SkAutoPixmapStorage pmapA, pmapB;
pmapA.alloc(read_pixels_info(a));
pmapB.alloc(read_pixels_info(b));
const int srcX = subsetA ? subsetA->x() : 0;
const int srcY = subsetA ? subsetA->y() : 0;
REPORTER_ASSERT(reporter, a->readPixels(dContextA, pmapA, srcX, srcY));
REPORTER_ASSERT(reporter, b->readPixels(nullptr, pmapB, 0, 0));
const size_t widthBytes = widthA * 4;
for (int y = 0; y < heightA; ++y) {
REPORTER_ASSERT(reporter, !memcmp(pmapA.addr32(0, y), pmapB.addr32(0, y), widthBytes));
}
}
static void draw_image_test_pattern(SkCanvas* canvas) {
canvas->clear(SK_ColorWHITE);
SkPaint paint;
paint.setColor(SK_ColorBLACK);
canvas->drawRect(SkRect::MakeXYWH(5, 5, 10, 10), paint);
}
static sk_sp<SkImage> create_image() {
const SkImageInfo info = SkImageInfo::MakeN32(20, 20, kOpaque_SkAlphaType);
auto surface(SkSurface::MakeRaster(info));
draw_image_test_pattern(surface->getCanvas());
return surface->makeImageSnapshot();
}
static sk_sp<SkData> create_image_data(SkImageInfo* info) {
*info = SkImageInfo::MakeN32(20, 20, kOpaque_SkAlphaType);
const size_t rowBytes = info->minRowBytes();
sk_sp<SkData> data(SkData::MakeUninitialized(rowBytes * info->height()));
{
SkBitmap bm;
bm.installPixels(*info, data->writable_data(), rowBytes);
SkCanvas canvas(bm);
draw_image_test_pattern(&canvas);
}
return data;
}
static sk_sp<SkImage> create_data_image() {
SkImageInfo info;
sk_sp<SkData> data(create_image_data(&info));
return SkImage::MakeRasterData(info, std::move(data), info.minRowBytes());
}
static sk_sp<SkImage> create_image_large(int maxTextureSize) {
const SkImageInfo info = SkImageInfo::MakeN32(maxTextureSize + 1, 32, kOpaque_SkAlphaType);
auto surface(SkSurface::MakeRaster(info));
surface->getCanvas()->clear(SK_ColorWHITE);
SkPaint paint;
paint.setColor(SK_ColorBLACK);
surface->getCanvas()->drawRect(SkRect::MakeXYWH(4000, 2, 28000, 30), paint);
return surface->makeImageSnapshot();
}
static sk_sp<SkImage> create_picture_image() {
SkPictureRecorder recorder;
SkCanvas* canvas = recorder.beginRecording(10, 10);
canvas->clear(SK_ColorCYAN);
return SkImage::MakeFromPicture(recorder.finishRecordingAsPicture(), SkISize::Make(10, 10),
nullptr, nullptr, SkImage::BitDepth::kU8,
SkColorSpace::MakeSRGB());
};
// Want to ensure that our Release is called when the owning image is destroyed
struct RasterDataHolder {
RasterDataHolder() : fReleaseCount(0) {}
sk_sp<SkData> fData;
int fReleaseCount;
static void Release(const void* pixels, void* context) {
RasterDataHolder* self = static_cast<RasterDataHolder*>(context);
self->fReleaseCount++;
self->fData.reset();
}
};
static sk_sp<SkImage> create_rasterproc_image(RasterDataHolder* dataHolder) {
SkASSERT(dataHolder);
SkImageInfo info;
dataHolder->fData = create_image_data(&info);
return SkImage::MakeFromRaster(SkPixmap(info, dataHolder->fData->data(), info.minRowBytes()),
RasterDataHolder::Release, dataHolder);
}
static sk_sp<SkImage> create_codec_image() {
SkImageInfo info;
sk_sp<SkData> data(create_image_data(&info));
SkBitmap bitmap;
bitmap.installPixels(info, data->writable_data(), info.minRowBytes());
auto src = SkEncodeBitmap(bitmap, SkEncodedImageFormat::kPNG, 100);
return SkImage::MakeFromEncoded(std::move(src));
}
static sk_sp<SkImage> create_gpu_image(GrRecordingContext* rContext,
bool withMips = false,
SkBudgeted budgeted = SkBudgeted::kYes) {
const SkImageInfo info = SkImageInfo::MakeN32(20, 20, kOpaque_SkAlphaType);
auto surface = SkSurface::MakeRenderTarget(rContext, budgeted, info, 0,
kBottomLeft_GrSurfaceOrigin, nullptr, withMips);
draw_image_test_pattern(surface->getCanvas());
return surface->makeImageSnapshot();
}
static void test_encode(skiatest::Reporter* reporter, GrDirectContext* dContext, SkImage* image) {
const SkIRect ir = SkIRect::MakeXYWH(5, 5, 10, 10);
sk_sp<SkData> origEncoded = image->encodeToData();
REPORTER_ASSERT(reporter, origEncoded);
REPORTER_ASSERT(reporter, origEncoded->size() > 0);
sk_sp<SkImage> decoded(SkImage::MakeFromEncoded(origEncoded));
if (!decoded) {
ERRORF(reporter, "failed to decode image!");
return;
}
REPORTER_ASSERT(reporter, decoded);
assert_equal(reporter, dContext, image, nullptr, decoded.get());
// Now see if we can instantiate an image from a subset of the surface/origEncoded
decoded = SkImage::MakeFromEncoded(origEncoded)->makeSubset(ir);
REPORTER_ASSERT(reporter, decoded);
assert_equal(reporter, dContext, image, &ir, decoded.get());
}
DEF_TEST(ImageEncode, reporter) {
test_encode(reporter, nullptr, create_image().get());
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(ImageEncode_Gpu, reporter, ctxInfo) {
auto dContext = ctxInfo.directContext();
test_encode(reporter, dContext, create_gpu_image(dContext).get());
}
DEF_TEST(Image_MakeFromRasterBitmap, reporter) {
const struct {
SkCopyPixelsMode fCPM;
bool fExpectSameAsMutable;
bool fExpectSameAsImmutable;
} recs[] = {
{ kIfMutable_SkCopyPixelsMode, false, true },
{ kAlways_SkCopyPixelsMode, false, false },
{ kNever_SkCopyPixelsMode, true, true },
};
for (auto rec : recs) {
SkPixmap pm;
SkBitmap bm;
bm.allocN32Pixels(100, 100);
auto img = SkMakeImageFromRasterBitmap(bm, rec.fCPM);
REPORTER_ASSERT(reporter, img->peekPixels(&pm));
const bool sameMutable = pm.addr32(0, 0) == bm.getAddr32(0, 0);
REPORTER_ASSERT(reporter, rec.fExpectSameAsMutable == sameMutable);
REPORTER_ASSERT(reporter, (bm.getGenerationID() == img->uniqueID()) == sameMutable);
bm.notifyPixelsChanged(); // force a new generation ID
bm.setImmutable();
img = SkMakeImageFromRasterBitmap(bm, rec.fCPM);
REPORTER_ASSERT(reporter, img->peekPixels(&pm));
const bool sameImmutable = pm.addr32(0, 0) == bm.getAddr32(0, 0);
REPORTER_ASSERT(reporter, rec.fExpectSameAsImmutable == sameImmutable);
REPORTER_ASSERT(reporter, (bm.getGenerationID() == img->uniqueID()) == sameImmutable);
}
}
// Test that image encoding failures do not break picture serialization/deserialization.
DEF_TEST(Image_Serialize_Encoding_Failure, reporter) {
auto surface(SkSurface::MakeRasterN32Premul(100, 100));
surface->getCanvas()->clear(SK_ColorGREEN);
sk_sp<SkImage> image(surface->makeImageSnapshot());
REPORTER_ASSERT(reporter, image);
SkPictureRecorder recorder;
SkCanvas* canvas = recorder.beginRecording(100, 100);
canvas->drawImage(image.get(), 0, 0, SkSamplingOptions());
sk_sp<SkPicture> picture(recorder.finishRecordingAsPicture());
REPORTER_ASSERT(reporter, picture);
REPORTER_ASSERT(reporter, picture->approximateOpCount() > 0);
bool was_called = false;
SkSerialProcs procs;
procs.fImageProc = [](SkImage*, void* called) {
*(bool*)called = true;
return SkData::MakeEmpty();
};
procs.fImageCtx = &was_called;
REPORTER_ASSERT(reporter, !was_called);
auto data = picture->serialize(&procs);
REPORTER_ASSERT(reporter, was_called);
REPORTER_ASSERT(reporter, data && data->size() > 0);
auto deserialized = SkPicture::MakeFromData(data->data(), data->size());
REPORTER_ASSERT(reporter, deserialized);
REPORTER_ASSERT(reporter, deserialized->approximateOpCount() > 0);
}
// Test that a draw that only partially covers the drawing surface isn't
// interpreted as covering the entire drawing surface (i.e., exercise one of the
// conditions of SkCanvas::wouldOverwriteEntireSurface()).
DEF_TEST(Image_RetainSnapshot, reporter) {
const SkPMColor red = SkPackARGB32(0xFF, 0xFF, 0, 0);
const SkPMColor green = SkPackARGB32(0xFF, 0, 0xFF, 0);
SkImageInfo info = SkImageInfo::MakeN32Premul(2, 2);
auto surface(SkSurface::MakeRaster(info));
surface->getCanvas()->clear(0xFF00FF00);
SkPMColor pixels[4];
memset(pixels, 0xFF, sizeof(pixels)); // init with values we don't expect
const SkImageInfo dstInfo = SkImageInfo::MakeN32Premul(2, 2);
const size_t dstRowBytes = 2 * sizeof(SkPMColor);
sk_sp<SkImage> image1(surface->makeImageSnapshot());
REPORTER_ASSERT(reporter, image1->readPixels(nullptr, dstInfo, pixels, dstRowBytes, 0, 0));
for (size_t i = 0; i < SK_ARRAY_COUNT(pixels); ++i) {
REPORTER_ASSERT(reporter, pixels[i] == green);
}
SkPaint paint;
paint.setBlendMode(SkBlendMode::kSrc);
paint.setColor(SK_ColorRED);
surface->getCanvas()->drawRect(SkRect::MakeXYWH(1, 1, 1, 1), paint);
sk_sp<SkImage> image2(surface->makeImageSnapshot());
REPORTER_ASSERT(reporter, image2->readPixels(nullptr, dstInfo, pixels, dstRowBytes, 0, 0));
REPORTER_ASSERT(reporter, pixels[0] == green);
REPORTER_ASSERT(reporter, pixels[1] == green);
REPORTER_ASSERT(reporter, pixels[2] == green);
REPORTER_ASSERT(reporter, pixels[3] == red);
}
/////////////////////////////////////////////////////////////////////////////////////////////////
static void make_bitmap_mutable(SkBitmap* bm) {
bm->allocN32Pixels(10, 10);
}
static void make_bitmap_immutable(SkBitmap* bm) {
bm->allocN32Pixels(10, 10);
bm->setImmutable();
}
DEF_TEST(image_newfrombitmap, reporter) {
const struct {
void (*fMakeProc)(SkBitmap*);
bool fExpectPeekSuccess;
bool fExpectSharedID;
bool fExpectLazy;
} rec[] = {
{ make_bitmap_mutable, true, false, false },
{ make_bitmap_immutable, true, true, false },
};
for (size_t i = 0; i < SK_ARRAY_COUNT(rec); ++i) {
SkBitmap bm;
rec[i].fMakeProc(&bm);
sk_sp<SkImage> image(bm.asImage());
SkPixmap pmap;
const bool sharedID = (image->uniqueID() == bm.getGenerationID());
REPORTER_ASSERT(reporter, sharedID == rec[i].fExpectSharedID);
const bool peekSuccess = image->peekPixels(&pmap);
REPORTER_ASSERT(reporter, peekSuccess == rec[i].fExpectPeekSuccess);
const bool lazy = image->isLazyGenerated();
REPORTER_ASSERT(reporter, lazy == rec[i].fExpectLazy);
}
}
///////////////////////////////////////////////////////////////////////////////////////////////////
#include "src/core/SkBitmapCache.h"
/*
* This tests the caching (and preemptive purge) of the raster equivalent of a gpu-image.
* We cache it for performance when drawing into a raster surface.
*
* A cleaner test would know if each drawImage call triggered a read-back from the gpu,
* but we don't have that facility (at the moment) so we use a little internal knowledge
* of *how* the raster version is cached, and look for that.
*/
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(SkImage_Gpu2Cpu, reporter, ctxInfo) {
SkImageInfo info = SkImageInfo::MakeN32(20, 20, kOpaque_SkAlphaType);
sk_sp<SkImage> image(create_gpu_image(ctxInfo.directContext()));
const auto desc = SkBitmapCacheDesc::Make(image.get());
auto surface(SkSurface::MakeRaster(info));
// now we can test drawing a gpu-backed image into a cpu-backed surface
{
SkBitmap cachedBitmap;
REPORTER_ASSERT(reporter, !SkBitmapCache::Find(desc, &cachedBitmap));
}
surface->getCanvas()->drawImage(image, 0, 0);
{
SkBitmap cachedBitmap;
if (SkBitmapCache::Find(desc, &cachedBitmap)) {
REPORTER_ASSERT(reporter, cachedBitmap.isImmutable());
REPORTER_ASSERT(reporter, cachedBitmap.getPixels());
} else {
// unexpected, but not really a bug, since the cache is global and this test may be
// run w/ other threads competing for its budget.
SkDebugf("SkImage_Gpu2Cpu : cachedBitmap was already purged\n");
}
}
image.reset(nullptr);
{
SkBitmap cachedBitmap;
REPORTER_ASSERT(reporter, !SkBitmapCache::Find(desc, &cachedBitmap));
}
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(SkImage_makeTextureImage, reporter, contextInfo) {
auto dContext = contextInfo.directContext();
sk_gpu_test::TestContext* testContext = contextInfo.testContext();
GrContextFactory otherFactory;
ContextInfo otherContextInfo = otherFactory.getContextInfo(contextInfo.type());
testContext->makeCurrent();
std::function<sk_sp<SkImage>()> imageFactories[] = {
create_image, create_codec_image, create_data_image,
// Create an image from a picture.
create_picture_image,
// Create a texture image.
[dContext] { return create_gpu_image(dContext, true, SkBudgeted::kYes); },
[dContext] { return create_gpu_image(dContext, false, SkBudgeted::kNo); },
// Create a texture image in a another context.
[otherContextInfo] {
auto restore = otherContextInfo.testContext()->makeCurrentAndAutoRestore();
auto otherContextImage = create_gpu_image(otherContextInfo.directContext());
otherContextInfo.directContext()->flushAndSubmit();
return otherContextImage;
}};
for (auto mipmapped : {GrMipmapped::kNo, GrMipmapped::kYes}) {
for (const auto& factory : imageFactories) {
sk_sp<SkImage> image(factory());
if (!image) {
ERRORF(reporter, "Error creating image.");
continue;
}
GrTextureProxy* origProxy = nullptr;
bool origIsMippedTexture = false;
if ((origProxy = sk_gpu_test::GetTextureImageProxy(image.get(), dContext))) {
REPORTER_ASSERT(reporter, (origProxy->mipmapped() == GrMipmapped::kYes) ==
image->hasMipmaps());
origIsMippedTexture = image->hasMipmaps();
}
for (auto budgeted : {SkBudgeted::kNo, SkBudgeted::kYes}) {
auto texImage = image->makeTextureImage(dContext, mipmapped, budgeted);
if (!texImage) {
auto imageContext = as_IB(image)->context();
// We expect to fail if image comes from a different context
if (!image->isTextureBacked() || imageContext->priv().matches(dContext)) {
ERRORF(reporter, "makeTextureImage failed.");
}
continue;
}
if (!texImage->isTextureBacked()) {
ERRORF(reporter, "makeTextureImage returned non-texture image.");
continue;
}
GrTextureProxy* copyProxy = sk_gpu_test::GetTextureImageProxy(texImage.get(),
dContext);
SkASSERT(copyProxy);
// Did we ask for MIPs on a context that supports them?
bool validRequestForMips = (mipmapped == GrMipmapped::kYes &&
dContext->priv().caps()->mipmapSupport());
// Do we expect the "copy" to have MIPs?
bool shouldBeMipped = origIsMippedTexture || validRequestForMips;
REPORTER_ASSERT(reporter, shouldBeMipped == texImage->hasMipmaps());
REPORTER_ASSERT(reporter,
shouldBeMipped == (copyProxy->mipmapped() == GrMipmapped::kYes));
// We should only make a copy of an already texture-backed image if it didn't
// already have MIPs but we asked for MIPs and the context supports it.
if (image->isTextureBacked() && (!validRequestForMips || origIsMippedTexture)) {
if (origProxy->underlyingUniqueID() != copyProxy->underlyingUniqueID()) {
ERRORF(reporter, "makeTextureImage made unnecessary texture copy.");
}
} else {
GrTextureProxy* texProxy = sk_gpu_test::GetTextureImageProxy(texImage.get(),
dContext);
REPORTER_ASSERT(reporter, !texProxy->getUniqueKey().isValid());
REPORTER_ASSERT(reporter, texProxy->isBudgeted() == budgeted);
}
if (image->width() != texImage->width() || image->height() != texImage->height()) {
ERRORF(reporter, "makeTextureImage changed the image size.");
}
if (image->alphaType() != texImage->alphaType()) {
ERRORF(reporter, "makeTextureImage changed image alpha type.");
}
}
}
}
dContext->flushAndSubmit();
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(SkImage_makeNonTextureImage, reporter, contextInfo) {
auto dContext = contextInfo.directContext();
std::function<sk_sp<SkImage>()> imageFactories[] = {
create_image,
create_codec_image,
create_data_image,
create_picture_image,
[dContext] { return create_gpu_image(dContext); },
};
for (const auto& factory : imageFactories) {
sk_sp<SkImage> image = factory();
if (!image->isTextureBacked()) {
REPORTER_ASSERT(reporter, image->makeNonTextureImage().get() == image.get());
if (!(image = image->makeTextureImage(dContext))) {
continue;
}
}
auto rasterImage = image->makeNonTextureImage();
if (!rasterImage) {
ERRORF(reporter, "makeNonTextureImage failed for texture-backed image.");
}
REPORTER_ASSERT(reporter, !rasterImage->isTextureBacked());
assert_equal(reporter, dContext, image.get(), nullptr, rasterImage.get());
}
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(GrContext_colorTypeSupportedAsImage, reporter, ctxInfo) {
auto dContext = ctxInfo.directContext();
static constexpr int kSize = 10;
for (int ct = 0; ct < kLastEnum_SkColorType; ++ct) {
SkColorType colorType = static_cast<SkColorType>(ct);
bool can = dContext->colorTypeSupportedAsImage(colorType);
auto mbet = sk_gpu_test::ManagedBackendTexture::MakeWithoutData(
dContext, kSize, kSize, colorType, GrMipmapped::kNo, GrRenderable::kNo);
sk_sp<SkImage> img;
if (mbet) {
img = SkImage::MakeFromTexture(dContext, mbet->texture(), kTopLeft_GrSurfaceOrigin,
colorType, kOpaque_SkAlphaType, nullptr);
}
REPORTER_ASSERT(reporter, can == SkToBool(img),
"colorTypeSupportedAsImage:%d, actual:%d, ct:%d", can, SkToBool(img),
colorType);
}
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(UnpremulTextureImage, reporter, ctxInfo) {
SkBitmap bmp;
bmp.allocPixels(
SkImageInfo::Make(256, 256, kRGBA_8888_SkColorType, kUnpremul_SkAlphaType, nullptr));
for (int y = 0; y < 256; ++y) {
for (int x = 0; x < 256; ++x) {
*bmp.getAddr32(x, y) =
SkColorSetARGB((U8CPU)y, 255 - (U8CPU)y, (U8CPU)x, 255 - (U8CPU)x);
}
}
auto dContext = ctxInfo.directContext();
auto texImage = bmp.asImage()->makeTextureImage(dContext);
if (!texImage || texImage->alphaType() != kUnpremul_SkAlphaType) {
ERRORF(reporter, "Failed to make unpremul texture image.");
return;
}
SkBitmap unpremul;
unpremul.allocPixels(SkImageInfo::Make(256, 256, kRGBA_8888_SkColorType,
kUnpremul_SkAlphaType, nullptr));
if (!texImage->readPixels(dContext, unpremul.info(), unpremul.getPixels(), unpremul.rowBytes(),
0, 0)) {
ERRORF(reporter, "Unpremul readback failed.");
return;
}
for (int y = 0; y < 256; ++y) {
for (int x = 0; x < 256; ++x) {
if (*bmp.getAddr32(x, y) != *unpremul.getAddr32(x, y)) {
ERRORF(reporter, "unpremul(0x%08x)->unpremul(0x%08x) at %d, %d.",
*bmp.getAddr32(x, y), *unpremul.getAddr32(x, y), x, y);
return;
}
}
}
SkBitmap premul;
premul.allocPixels(
SkImageInfo::Make(256, 256, kRGBA_8888_SkColorType, kPremul_SkAlphaType, nullptr));
if (!texImage->readPixels(dContext, premul.info(), premul.getPixels(), premul.rowBytes(),
0, 0)) {
ERRORF(reporter, "Unpremul readback failed.");
return;
}
for (int y = 0; y < 256; ++y) {
for (int x = 0; x < 256; ++x) {
uint32_t origColor = *bmp.getAddr32(x, y);
int32_t origA = (origColor >> 24) & 0xff;
float a = origA / 255.f;
int32_t origB = sk_float_round2int(((origColor >> 16) & 0xff) * a);
int32_t origG = sk_float_round2int(((origColor >> 8) & 0xff) * a);
int32_t origR = sk_float_round2int(((origColor >> 0) & 0xff) * a);
uint32_t read = *premul.getAddr32(x, y);
int32_t readA = (read >> 24) & 0xff;
int32_t readB = (read >> 16) & 0xff;
int32_t readG = (read >> 8) & 0xff;
int32_t readR = (read >> 0) & 0xff;
// We expect that alpha=1 and alpha=0 should come out exact. Otherwise allow a little
// bit of tolerance for GPU vs CPU premul math.
int32_t tol = (origA == 0 || origA == 255) ? 0 : 1;
if (origA != readA || SkTAbs(readB - origB) > tol || SkTAbs(readG - origG) > tol ||
SkTAbs(readR - origR) > tol) {
ERRORF(reporter, "unpremul(0x%08x)->premul(0x%08x) expected(0x%08x) at %d, %d.",
*bmp.getAddr32(x, y), *premul.getAddr32(x, y), origColor, x, y);
return;
}
}
}
}
DEF_GPUTEST(AbandonedContextImage, reporter, options) {
using Factory = sk_gpu_test::GrContextFactory;
for (int ct = 0; ct < Factory::kContextTypeCnt; ++ct) {
auto type = static_cast<Factory::ContextType>(ct);
std::unique_ptr<Factory> factory(new Factory);
if (!factory->get(type)) {
continue;
}
sk_sp<SkImage> img;
auto gsurf = SkSurface::MakeRenderTarget(
factory->get(type), SkBudgeted::kYes,
SkImageInfo::Make(100, 100, kRGBA_8888_SkColorType, kPremul_SkAlphaType), 1,
nullptr);
if (!gsurf) {
continue;
}
img = gsurf->makeImageSnapshot();
gsurf.reset();
auto rsurf = SkSurface::MakeRaster(SkImageInfo::MakeN32Premul(100, 100));
REPORTER_ASSERT(reporter, img->isValid(factory->get(type)));
REPORTER_ASSERT(reporter, img->isValid(rsurf->getCanvas()->recordingContext()));
factory->get(type)->abandonContext();
REPORTER_ASSERT(reporter, !img->isValid(factory->get(type)));
REPORTER_ASSERT(reporter, !img->isValid(rsurf->getCanvas()->recordingContext()));
// This shouldn't crash.
rsurf->getCanvas()->drawImage(img, 0, 0);
// Give up all other refs on the context.
factory.reset(nullptr);
REPORTER_ASSERT(reporter, !img->isValid(rsurf->getCanvas()->recordingContext()));
// This shouldn't crash.
rsurf->getCanvas()->drawImage(img, 0, 0);
}
}
class EmptyGenerator : public SkImageGenerator {
public:
EmptyGenerator() : SkImageGenerator(SkImageInfo::MakeN32Premul(0, 0)) {}
};
DEF_TEST(ImageEmpty, reporter) {
const SkImageInfo info = SkImageInfo::Make(0, 0, kN32_SkColorType, kPremul_SkAlphaType);
SkPixmap pmap(info, nullptr, 0);
REPORTER_ASSERT(reporter, nullptr == SkImage::MakeRasterCopy(pmap));
REPORTER_ASSERT(reporter, nullptr == SkImage::MakeRasterData(info, nullptr, 0));
REPORTER_ASSERT(reporter, nullptr == SkImage::MakeFromRaster(pmap, nullptr, nullptr));
REPORTER_ASSERT(reporter, nullptr == SkImage::MakeFromGenerator(
std::make_unique<EmptyGenerator>()));
}
DEF_TEST(ImageDataRef, reporter) {
SkImageInfo info = SkImageInfo::MakeN32Premul(1, 1);
size_t rowBytes = info.minRowBytes();
size_t size = info.computeByteSize(rowBytes);
sk_sp<SkData> data = SkData::MakeUninitialized(size);
REPORTER_ASSERT(reporter, data->unique());
sk_sp<SkImage> image = SkImage::MakeRasterData(info, data, rowBytes);
REPORTER_ASSERT(reporter, !data->unique());
image.reset();
REPORTER_ASSERT(reporter, data->unique());
}
static bool has_pixels(const SkPMColor pixels[], int count, SkPMColor expected) {
for (int i = 0; i < count; ++i) {
if (pixels[i] != expected) {
return false;
}
}
return true;
}
static void image_test_read_pixels(GrDirectContext* dContext, skiatest::Reporter* reporter,
SkImage* image) {
if (!image) {
ERRORF(reporter, "Failed to create image!");
return;
}
const SkPMColor expected = SkPreMultiplyColor(SK_ColorWHITE);
const SkPMColor notExpected = ~expected;
const int w = 2, h = 2;
const size_t rowBytes = w * sizeof(SkPMColor);
SkPMColor pixels[w*h];
SkImageInfo info;
info = SkImageInfo::MakeUnknown(w, h);
REPORTER_ASSERT(reporter, !image->readPixels(dContext, info, pixels, rowBytes, 0, 0));
// out-of-bounds should fail
info = SkImageInfo::MakeN32Premul(w, h);
REPORTER_ASSERT(reporter, !image->readPixels(dContext, info, pixels, rowBytes, -w, 0));
REPORTER_ASSERT(reporter, !image->readPixels(dContext, info, pixels, rowBytes, 0, -h));
REPORTER_ASSERT(reporter, !image->readPixels(dContext, info, pixels, rowBytes,
image->width(), 0));
REPORTER_ASSERT(reporter, !image->readPixels(dContext, info, pixels, rowBytes,
0, image->height()));
// top-left should succeed
sk_memset32(pixels, notExpected, w*h);
REPORTER_ASSERT(reporter, image->readPixels(dContext, info, pixels, rowBytes, 0, 0));
REPORTER_ASSERT(reporter, has_pixels(pixels, w*h, expected));
// bottom-right should succeed
sk_memset32(pixels, notExpected, w*h);
REPORTER_ASSERT(reporter, image->readPixels(dContext, info, pixels, rowBytes,
image->width() - w, image->height() - h));
REPORTER_ASSERT(reporter, has_pixels(pixels, w*h, expected));
// partial top-left should succeed
sk_memset32(pixels, notExpected, w*h);
REPORTER_ASSERT(reporter, image->readPixels(dContext, info, pixels, rowBytes, -1, -1));
REPORTER_ASSERT(reporter, pixels[3] == expected);
REPORTER_ASSERT(reporter, has_pixels(pixels, w*h - 1, notExpected));
// partial bottom-right should succeed
sk_memset32(pixels, notExpected, w*h);
REPORTER_ASSERT(reporter, image->readPixels(dContext, info, pixels, rowBytes,
image->width() - 1, image->height() - 1));
REPORTER_ASSERT(reporter, pixels[0] == expected);
REPORTER_ASSERT(reporter, has_pixels(&pixels[1], w*h - 1, notExpected));
}
DEF_TEST(ImageReadPixels, reporter) {
sk_sp<SkImage> image(create_image());
image_test_read_pixels(nullptr, reporter, image.get());
image = create_data_image();
image_test_read_pixels(nullptr, reporter, image.get());
RasterDataHolder dataHolder;
image = create_rasterproc_image(&dataHolder);
image_test_read_pixels(nullptr, reporter, image.get());
image.reset();
REPORTER_ASSERT(reporter, 1 == dataHolder.fReleaseCount);
image = create_codec_image();
image_test_read_pixels(nullptr, reporter, image.get());
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(ImageReadPixels_Gpu, reporter, ctxInfo) {
auto dContext = ctxInfo.directContext();
image_test_read_pixels(dContext, reporter, create_gpu_image(dContext).get());
}
static void check_legacy_bitmap(skiatest::Reporter* reporter, GrDirectContext* dContext,
const SkImage* image, const SkBitmap& bitmap) {
REPORTER_ASSERT(reporter, image->width() == bitmap.width());
REPORTER_ASSERT(reporter, image->height() == bitmap.height());
REPORTER_ASSERT(reporter, image->alphaType() == bitmap.alphaType());
REPORTER_ASSERT(reporter, bitmap.isImmutable());
REPORTER_ASSERT(reporter, bitmap.getPixels());
const SkImageInfo info = SkImageInfo::MakeN32(1, 1, bitmap.alphaType());
SkPMColor imageColor;
REPORTER_ASSERT(reporter, image->readPixels(dContext, info, &imageColor, sizeof(SkPMColor),
0, 0));
REPORTER_ASSERT(reporter, imageColor == *bitmap.getAddr32(0, 0));
}
static void test_legacy_bitmap(skiatest::Reporter* reporter, GrDirectContext* dContext,
const SkImage* image) {
if (!image) {
ERRORF(reporter, "Failed to create image.");
return;
}
SkBitmap bitmap;
REPORTER_ASSERT(reporter, image->asLegacyBitmap(&bitmap));
check_legacy_bitmap(reporter, dContext, image, bitmap);
// Test subsetting to exercise the rowBytes logic.
SkBitmap tmp;
REPORTER_ASSERT(reporter, bitmap.extractSubset(&tmp, SkIRect::MakeWH(image->width() / 2,
image->height() / 2)));
sk_sp<SkImage> subsetImage(tmp.asImage());
REPORTER_ASSERT(reporter, subsetImage.get());
SkBitmap subsetBitmap;
REPORTER_ASSERT(reporter, subsetImage->asLegacyBitmap(&subsetBitmap));
check_legacy_bitmap(reporter, nullptr, subsetImage.get(), subsetBitmap);
}
DEF_TEST(ImageLegacyBitmap, reporter) {
sk_sp<SkImage> image(create_image());
test_legacy_bitmap(reporter, nullptr, image.get());
image = create_data_image();
test_legacy_bitmap(reporter, nullptr, image.get());
RasterDataHolder dataHolder;
image = create_rasterproc_image(&dataHolder);
test_legacy_bitmap(reporter, nullptr, image.get());
image.reset();
REPORTER_ASSERT(reporter, 1 == dataHolder.fReleaseCount);
image = create_codec_image();
test_legacy_bitmap(reporter, nullptr, image.get());
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(ImageLegacyBitmap_Gpu, reporter, ctxInfo) {
auto dContext = ctxInfo.directContext();
sk_sp<SkImage> image(create_gpu_image(dContext));
test_legacy_bitmap(reporter, dContext, image.get());
}
static void test_peek(skiatest::Reporter* reporter, SkImage* image, bool expectPeekSuccess) {
if (!image) {
ERRORF(reporter, "Failed to create image!");
return;
}
SkPixmap pm;
bool success = image->peekPixels(&pm);
REPORTER_ASSERT(reporter, expectPeekSuccess == success);
if (success) {
const SkImageInfo& info = pm.info();
REPORTER_ASSERT(reporter, 20 == info.width());
REPORTER_ASSERT(reporter, 20 == info.height());
REPORTER_ASSERT(reporter, kN32_SkColorType == info.colorType());
REPORTER_ASSERT(reporter, kPremul_SkAlphaType == info.alphaType() ||
kOpaque_SkAlphaType == info.alphaType());
REPORTER_ASSERT(reporter, info.minRowBytes() <= pm.rowBytes());
REPORTER_ASSERT(reporter, SkPreMultiplyColor(SK_ColorWHITE) == *pm.addr32(0, 0));
}
}
DEF_TEST(ImagePeek, reporter) {
sk_sp<SkImage> image(create_image());
test_peek(reporter, image.get(), true);
image = create_data_image();
test_peek(reporter, image.get(), true);
RasterDataHolder dataHolder;
image = create_rasterproc_image(&dataHolder);
test_peek(reporter, image.get(), true);
image.reset();
REPORTER_ASSERT(reporter, 1 == dataHolder.fReleaseCount);
image = create_codec_image();
test_peek(reporter, image.get(), false);
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(ImagePeek_Gpu, reporter, ctxInfo) {
sk_sp<SkImage> image(create_gpu_image(ctxInfo.directContext()));
test_peek(reporter, image.get(), false);
}
struct TextureReleaseChecker {
TextureReleaseChecker() : fReleaseCount(0) {}
int fReleaseCount;
static void Release(void* self) {
static_cast<TextureReleaseChecker*>(self)->fReleaseCount++;
}
};
DEF_GPUTEST_FOR_GL_RENDERING_CONTEXTS(SkImage_NewFromTextureRelease, reporter, ctxInfo) {
const int kWidth = 10;
const int kHeight = 10;
auto dContext = ctxInfo.directContext();
auto mbet = sk_gpu_test::ManagedBackendTexture::MakeWithoutData(dContext,
kWidth,
kHeight,
kRGBA_8888_SkColorType,
GrMipmapped::kNo,
GrRenderable::kNo,
GrProtected::kNo);
if (!mbet) {
ERRORF(reporter, "couldn't create backend texture\n");
return;
}
TextureReleaseChecker releaseChecker;
GrSurfaceOrigin texOrigin = kBottomLeft_GrSurfaceOrigin;
sk_sp<SkImage> refImg = SkImage::MakeFromTexture(
dContext,
mbet->texture(),
texOrigin,
kRGBA_8888_SkColorType,
kPremul_SkAlphaType,
/*color space*/nullptr,
sk_gpu_test::ManagedBackendTexture::ReleaseProc,
mbet->releaseContext(TextureReleaseChecker::Release, &releaseChecker));
GrSurfaceOrigin readBackOrigin;
GrBackendTexture readBackBackendTex = refImg->getBackendTexture(false, &readBackOrigin);
if (!GrBackendTexture::TestingOnly_Equals(readBackBackendTex, mbet->texture())) {
ERRORF(reporter, "backend mismatch\n");
}
REPORTER_ASSERT(reporter,
GrBackendTexture::TestingOnly_Equals(readBackBackendTex, mbet->texture()));
if (readBackOrigin != texOrigin) {
ERRORF(reporter, "origin mismatch %d %d\n", readBackOrigin, texOrigin);
}
REPORTER_ASSERT(reporter, readBackOrigin == texOrigin);
// Now exercise the release proc
REPORTER_ASSERT(reporter, 0 == releaseChecker.fReleaseCount);
refImg.reset(nullptr); // force a release of the image
REPORTER_ASSERT(reporter, 1 == releaseChecker.fReleaseCount);
}
static void test_cross_context_image(skiatest::Reporter* reporter, const GrContextOptions& options,
const char* testName,
std::function<sk_sp<SkImage>(GrDirectContext*)> imageMaker) {
for (int i = 0; i < GrContextFactory::kContextTypeCnt; ++i) {
GrContextFactory testFactory(options);
GrContextFactory::ContextType ctxType = static_cast<GrContextFactory::ContextType>(i);
ContextInfo ctxInfo = testFactory.getContextInfo(ctxType);
auto dContext = ctxInfo.directContext();
if (!dContext) {
continue;
}
// If we don't have proper support for this feature, the factory will fallback to returning
// codec-backed images. Those will "work", but some of our checks will fail because we
// expect the cross-context images not to work on multiple contexts at once.
if (!dContext->priv().caps()->crossContextTextureSupport()) {
continue;
}
// We test three lifetime patterns for a single context:
// 1) Create image, free image
// 2) Create image, draw, flush, free image
// 3) Create image, draw, free image, flush
// ... and then repeat the last two patterns with drawing on a second* context:
// 4) Create image, draw*, flush*, free image
// 5) Create image, draw*, free iamge, flush*
// Case #1: Create image, free image
{
sk_sp<SkImage> refImg(imageMaker(dContext));
refImg.reset(nullptr); // force a release of the image
}
SkImageInfo info = SkImageInfo::MakeN32Premul(128, 128);
sk_sp<SkSurface> surface = SkSurface::MakeRenderTarget(dContext, SkBudgeted::kNo, info);
if (!surface) {
ERRORF(reporter, "SkSurface::MakeRenderTarget failed for %s.", testName);
continue;
}
SkCanvas* canvas = surface->getCanvas();
// Case #2: Create image, draw, flush, free image
{
sk_sp<SkImage> refImg(imageMaker(dContext));
canvas->drawImage(refImg, 0, 0);
surface->flushAndSubmit();
refImg.reset(nullptr); // force a release of the image
}
// Case #3: Create image, draw, free image, flush
{
sk_sp<SkImage> refImg(imageMaker(dContext));
canvas->drawImage(refImg, 0, 0);
refImg.reset(nullptr); // force a release of the image
surface->flushAndSubmit();
}
// Configure second context
sk_gpu_test::TestContext* testContext = ctxInfo.testContext();
ContextInfo otherContextInfo = testFactory.getSharedContextInfo(dContext);
auto otherCtx = otherContextInfo.directContext();
sk_gpu_test::TestContext* otherTestContext = otherContextInfo.testContext();
// Creating a context in a share group may fail
if (!otherCtx) {
continue;
}
surface = SkSurface::MakeRenderTarget(otherCtx, SkBudgeted::kNo, info);
canvas = surface->getCanvas();
// Case #4: Create image, draw*, flush*, free image
{
testContext->makeCurrent();
sk_sp<SkImage> refImg(imageMaker(dContext));
otherTestContext->makeCurrent();
canvas->drawImage(refImg, 0, 0);
surface->flushAndSubmit();
testContext->makeCurrent();
refImg.reset(nullptr); // force a release of the image
}
// Case #5: Create image, draw*, free image, flush*
{
testContext->makeCurrent();
sk_sp<SkImage> refImg(imageMaker(dContext));
otherTestContext->makeCurrent();
canvas->drawImage(refImg, 0, 0);
testContext->makeCurrent();
refImg.reset(nullptr); // force a release of the image
otherTestContext->makeCurrent();
// Sync is specifically here for vulkan to guarantee the command buffer will finish
// which is when we call the ReleaseProc.
surface->flushAndSubmit(true);
}
// Case #6: Verify that only one context can be using the image at a time
{
// Suppress warnings about trying to use a texture in two contexts.
GrRecordingContextPriv::AutoSuppressWarningMessages aswm(otherCtx);
testContext->makeCurrent();
sk_sp<SkImage> refImg(imageMaker(dContext));
GrSurfaceProxyView view, otherView, viewSecondRef;
// Any context should be able to borrow the texture at this point
std::tie(view, std::ignore) = as_IB(refImg)->asView(dContext, GrMipmapped::kNo);
REPORTER_ASSERT(reporter, view);
// But once it's borrowed, no other context should be able to borrow
otherTestContext->makeCurrent();
std::tie(otherView, std::ignore) = as_IB(refImg)->asView(otherCtx, GrMipmapped::kNo);
REPORTER_ASSERT(reporter, !otherView);
// Original context (that's already borrowing) should be okay
testContext->makeCurrent();
std::tie(viewSecondRef, std::ignore) = as_IB(refImg)->asView(dContext,
GrMipmapped::kNo);
REPORTER_ASSERT(reporter, viewSecondRef);
// Release first ref from the original context
view.reset();
// We released one proxy but not the other from the current borrowing context. Make sure
// a new context is still not able to borrow the texture.
otherTestContext->makeCurrent();
std::tie(otherView, std::ignore) = as_IB(refImg)->asView(otherCtx, GrMipmapped::kNo);
REPORTER_ASSERT(reporter, !otherView);
// Release second ref from the original context
testContext->makeCurrent();
viewSecondRef.reset();
// Now we should be able to borrow the texture from the other context
otherTestContext->makeCurrent();
std::tie(otherView, std::ignore) = as_IB(refImg)->asView(otherCtx, GrMipmapped::kNo);
REPORTER_ASSERT(reporter, otherView);
// Release everything
otherView.reset();
refImg.reset(nullptr);
}
}
}
DEF_GPUTEST(SkImage_MakeCrossContextFromPixmapRelease, reporter, options) {
SkBitmap bitmap;
SkPixmap pixmap;
if (!GetResourceAsBitmap("images/mandrill_128.png", &bitmap) || !bitmap.peekPixels(&pixmap)) {
ERRORF(reporter, "missing resource");
return;
}
test_cross_context_image(reporter, options, "SkImage_MakeCrossContextFromPixmapRelease",
[&pixmap](GrDirectContext* dContext) {
return SkImage::MakeCrossContextFromPixmap(dContext, pixmap, false);
});
}
DEF_GPUTEST(SkImage_CrossContextGrayAlphaConfigs, reporter, options) {
for (SkColorType ct : { kGray_8_SkColorType, kAlpha_8_SkColorType }) {
SkAutoPixmapStorage pixmap;
pixmap.alloc(SkImageInfo::Make(4, 4, ct, kPremul_SkAlphaType));
for (int i = 0; i < GrContextFactory::kContextTypeCnt; ++i) {
GrContextFactory testFactory(options);
GrContextFactory::ContextType ctxType = static_cast<GrContextFactory::ContextType>(i);
ContextInfo ctxInfo = testFactory.getContextInfo(ctxType);
auto dContext = ctxInfo.directContext();
if (!dContext || !dContext->priv().caps()->crossContextTextureSupport()) {
continue;
}
sk_sp<SkImage> image = SkImage::MakeCrossContextFromPixmap(dContext, pixmap, false);
REPORTER_ASSERT(reporter, image);
auto [view, viewCT] = as_IB(image)->asView(dContext, GrMipmapped::kNo);
REPORTER_ASSERT(reporter, view);
REPORTER_ASSERT(reporter, GrColorTypeToSkColorType(viewCT) == ct);
bool expectAlpha = kAlpha_8_SkColorType == ct;
GrColorType grCT = SkColorTypeToGrColorType(image->colorType());
REPORTER_ASSERT(reporter, expectAlpha == GrColorTypeIsAlphaOnly(grCT));
}
}
}
DEF_GPUTEST_FOR_GL_RENDERING_CONTEXTS(makeBackendTexture, reporter, ctxInfo) {
auto context = ctxInfo.directContext();
sk_gpu_test::TestContext* testContext = ctxInfo.testContext();
sk_sp<GrContextThreadSafeProxy> proxy = context->threadSafeProxy();
GrContextFactory otherFactory;
ContextInfo otherContextInfo = otherFactory.getContextInfo(ctxInfo.type());
testContext->makeCurrent();
REPORTER_ASSERT(reporter, proxy);
auto createLarge = [context] {
return create_image_large(context->priv().caps()->maxTextureSize());
};
struct TestCase {
std::function<sk_sp<SkImage>()> fImageFactory;
bool fExpectation;
bool fCanTakeDirectly;
};
TestCase testCases[] = {
{ create_image, true, false },
{ create_codec_image, true, false },
{ create_data_image, true, false },
{ create_picture_image, true, false },
{ [context] { return create_gpu_image(context); }, true, true },
// Create a texture image in a another context.
{ [otherContextInfo] {
auto restore = otherContextInfo.testContext()->makeCurrentAndAutoRestore();
sk_sp<SkImage> otherContextImage = create_gpu_image(otherContextInfo.directContext());
otherContextInfo.directContext()->flushAndSubmit();
return otherContextImage;
}, false, false },
// Create an image that is too large to be texture backed.
{ createLarge, false, false }
};
for (const TestCase& testCase : testCases) {
sk_sp<SkImage> image(testCase.fImageFactory());
if (!image) {
ERRORF(reporter, "Failed to create image!");
continue;
}
GrBackendTexture origBackend = image->getBackendTexture(true);
if (testCase.fCanTakeDirectly) {
SkASSERT(origBackend.isValid());
}
GrBackendTexture newBackend;
SkImage::BackendTextureReleaseProc proc;
bool result = SkImage::MakeBackendTextureFromSkImage(context, std::move(image),
&newBackend, &proc);
if (result != testCase.fExpectation) {
static const char *const kFS[] = { "fail", "succeed" };
ERRORF(reporter, "This image was expected to %s but did not.",
kFS[testCase.fExpectation]);
}
if (result) {
SkASSERT(newBackend.isValid());
}
bool tookDirectly = result && GrBackendTexture::TestingOnly_Equals(origBackend, newBackend);
if (testCase.fCanTakeDirectly != tookDirectly) {
static const char *const kExpectedState[] = { "not expected", "expected" };
ERRORF(reporter, "This backend texture was %s to be taken directly.",
kExpectedState[testCase.fCanTakeDirectly]);
}
context->flushAndSubmit();
}
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(ImageBackendAccessAbandoned_Gpu, reporter, ctxInfo) {
auto dContext = ctxInfo.directContext();
sk_sp<SkImage> image(create_gpu_image(ctxInfo.directContext()));
if (!image) {
return;
}
GrBackendTexture beTex = image->getBackendTexture(true);
REPORTER_ASSERT(reporter, beTex.isValid());
dContext->abandonContext();
// After abandoning the context the backend texture should not be valid.
beTex = image->getBackendTexture(true);
REPORTER_ASSERT(reporter, !beTex.isValid());
}
///////////////////////////////////////////////////////////////////////////////////////////////////
static sk_sp<SkImage> create_picture_image(sk_sp<SkColorSpace> space) {
SkPictureRecorder recorder;
SkCanvas* canvas = recorder.beginRecording(10, 10);
canvas->clear(SK_ColorCYAN);
return SkImage::MakeFromPicture(recorder.finishRecordingAsPicture(), SkISize::Make(10, 10),
nullptr, nullptr, SkImage::BitDepth::kU8, std::move(space));
};
DEF_TEST(Image_ColorSpace, r) {
sk_sp<SkColorSpace> srgb = SkColorSpace::MakeSRGB();
sk_sp<SkImage> image = GetResourceAsImage("images/mandrill_512_q075.jpg");
REPORTER_ASSERT(r, srgb.get() == image->colorSpace());
image = GetResourceAsImage("images/webp-color-profile-lossy.webp");
skcms_TransferFunction fn;
bool success = image->colorSpace()->isNumericalTransferFn(&fn);
REPORTER_ASSERT(r, success);
REPORTER_ASSERT(r, color_space_almost_equal(1.8f, fn.g));
sk_sp<SkColorSpace> rec2020 = SkColorSpace::MakeRGB(SkNamedTransferFn::kSRGB,
SkNamedGamut::kRec2020);
image = create_picture_image(rec2020);
REPORTER_ASSERT(r, SkColorSpace::Equals(rec2020.get(), image->colorSpace()));
SkBitmap bitmap;
SkImageInfo info = SkImageInfo::MakeN32(10, 10, kPremul_SkAlphaType, rec2020);
bitmap.allocPixels(info);
image = bitmap.asImage();
REPORTER_ASSERT(r, SkColorSpace::Equals(rec2020.get(), image->colorSpace()));
sk_sp<SkSurface> surface = SkSurface::MakeRaster(
SkImageInfo::MakeN32Premul(SkISize::Make(10, 10)));
image = surface->makeImageSnapshot();
REPORTER_ASSERT(r, nullptr == image->colorSpace());
surface = SkSurface::MakeRaster(info);
image = surface->makeImageSnapshot();
REPORTER_ASSERT(r, SkColorSpace::Equals(rec2020.get(), image->colorSpace()));
}
DEF_TEST(Image_makeColorSpace, r) {
sk_sp<SkColorSpace> p3 = SkColorSpace::MakeRGB(SkNamedTransferFn::kSRGB, SkNamedGamut::kDisplayP3);
skcms_TransferFunction fn;
fn.a = 1.f; fn.b = 0.f; fn.c = 0.f; fn.d = 0.f; fn.e = 0.f; fn.f = 0.f; fn.g = 1.8f;
sk_sp<SkColorSpace> adobeGamut = SkColorSpace::MakeRGB(fn, SkNamedGamut::kAdobeRGB);
SkBitmap srgbBitmap;
srgbBitmap.allocPixels(SkImageInfo::MakeS32(1, 1, kOpaque_SkAlphaType));
*srgbBitmap.getAddr32(0, 0) = SkSwizzle_RGBA_to_PMColor(0xFF604020);
srgbBitmap.setImmutable();
sk_sp<SkImage> srgbImage = srgbBitmap.asImage();
sk_sp<SkImage> p3Image = srgbImage->makeColorSpace(p3, nullptr);
SkBitmap p3Bitmap;
bool success = p3Image->asLegacyBitmap(&p3Bitmap);
auto almost_equal = [](int a, int b) { return SkTAbs(a - b) <= 2; };
REPORTER_ASSERT(r, success);
REPORTER_ASSERT(r, almost_equal(0x28, SkGetPackedR32(*p3Bitmap.getAddr32(0, 0))));
REPORTER_ASSERT(r, almost_equal(0x40, SkGetPackedG32(*p3Bitmap.getAddr32(0, 0))));
REPORTER_ASSERT(r, almost_equal(0x5E, SkGetPackedB32(*p3Bitmap.getAddr32(0, 0))));
sk_sp<SkImage> adobeImage = srgbImage->makeColorSpace(adobeGamut, nullptr);
SkBitmap adobeBitmap;
success = adobeImage->asLegacyBitmap(&adobeBitmap);
REPORTER_ASSERT(r, success);
REPORTER_ASSERT(r, almost_equal(0x21, SkGetPackedR32(*adobeBitmap.getAddr32(0, 0))));
REPORTER_ASSERT(r, almost_equal(0x31, SkGetPackedG32(*adobeBitmap.getAddr32(0, 0))));
REPORTER_ASSERT(r, almost_equal(0x4C, SkGetPackedB32(*adobeBitmap.getAddr32(0, 0))));
srgbImage = GetResourceAsImage("images/1x1.png");
p3Image = srgbImage->makeColorSpace(p3, nullptr);
success = p3Image->asLegacyBitmap(&p3Bitmap);
REPORTER_ASSERT(r, success);
REPORTER_ASSERT(r, almost_equal(0x8B, SkGetPackedR32(*p3Bitmap.getAddr32(0, 0))));
REPORTER_ASSERT(r, almost_equal(0x82, SkGetPackedG32(*p3Bitmap.getAddr32(0, 0))));
REPORTER_ASSERT(r, almost_equal(0x77, SkGetPackedB32(*p3Bitmap.getAddr32(0, 0))));
}
///////////////////////////////////////////////////////////////////////////////////////////////////
static void make_all_premul(SkBitmap* bm) {
bm->allocPixels(SkImageInfo::MakeN32(256, 256, kPremul_SkAlphaType));
for (int a = 0; a < 256; ++a) {
for (int r = 0; r < 256; ++r) {
// make all valid premul combinations
int c = std::min(a, r);
*bm->getAddr32(a, r) = SkPackARGB32(a, c, c, c);
}
}
}
static bool equal(const SkBitmap& a, const SkBitmap& b) {
SkASSERT(a.width() == b.width());
SkASSERT(a.height() == b.height());
for (int y = 0; y < a.height(); ++y) {
for (int x = 0; x < a.width(); ++x) {
SkPMColor pa = *a.getAddr32(x, y);
SkPMColor pb = *b.getAddr32(x, y);
if (pa != pb) {
return false;
}
}
}
return true;
}
DEF_TEST(image_roundtrip_encode, reporter) {
SkBitmap bm0;
make_all_premul(&bm0);
auto img0 = bm0.asImage();
sk_sp<SkData> data = img0->encodeToData(SkEncodedImageFormat::kPNG, 100);
auto img1 = SkImage::MakeFromEncoded(data);
SkBitmap bm1;
bm1.allocPixels(SkImageInfo::MakeN32(256, 256, kPremul_SkAlphaType));
img1->readPixels(nullptr, bm1.info(), bm1.getPixels(), bm1.rowBytes(), 0, 0);
REPORTER_ASSERT(reporter, equal(bm0, bm1));
}
DEF_TEST(image_roundtrip_premul, reporter) {
SkBitmap bm0;
make_all_premul(&bm0);
SkBitmap bm1;
bm1.allocPixels(SkImageInfo::MakeN32(256, 256, kUnpremul_SkAlphaType));
bm0.readPixels(bm1.info(), bm1.getPixels(), bm1.rowBytes(), 0, 0);
SkBitmap bm2;
bm2.allocPixels(SkImageInfo::MakeN32(256, 256, kPremul_SkAlphaType));
bm1.readPixels(bm2.info(), bm2.getPixels(), bm2.rowBytes(), 0, 0);
REPORTER_ASSERT(reporter, equal(bm0, bm2));
}
DEF_TEST(image_from_encoded_alphatype_override, reporter) {
sk_sp<SkData> data = GetResourceAsData("images/mandrill_32.png");
// Ensure that we can decode the image when we specifically request premul or unpremul, but
// not when we request kOpaque
REPORTER_ASSERT(reporter, SkImage::MakeFromEncoded(data, kPremul_SkAlphaType));
REPORTER_ASSERT(reporter, SkImage::MakeFromEncoded(data, kUnpremul_SkAlphaType));
REPORTER_ASSERT(reporter, !SkImage::MakeFromEncoded(data, kOpaque_SkAlphaType));
// Same tests as above, but using SkImageGenerator::MakeFromEncoded
REPORTER_ASSERT(reporter, SkImageGenerator::MakeFromEncoded(data, kPremul_SkAlphaType));
REPORTER_ASSERT(reporter, SkImageGenerator::MakeFromEncoded(data, kUnpremul_SkAlphaType));
REPORTER_ASSERT(reporter, !SkImageGenerator::MakeFromEncoded(data, kOpaque_SkAlphaType));
}
///////////////////////////////////////////////////////////////////////////////////////////////////
static void check_scaled_pixels(skiatest::Reporter* reporter, SkPixmap* pmap, uint32_t expected) {
// Verify that all pixels contain the original test color
for (auto y = 0; y < pmap->height(); ++y) {
for (auto x = 0; x < pmap->width(); ++x) {
uint32_t pixel = *pmap->addr32(x, y);
if (pixel != expected) {
ERRORF(reporter, "Expected scaled pixels to be the same. At %d,%d 0x%08x != 0x%08x",
x, y, pixel, expected);
return;
}
}
}
}
static void test_scale_pixels(skiatest::Reporter* reporter, const SkImage* image,
uint32_t expected) {
SkImageInfo info = SkImageInfo::MakeN32Premul(image->width() * 2, image->height() * 2);
// Make sure to test kDisallow first, so we don't just get a cache hit in that case
for (auto chint : { SkImage::kDisallow_CachingHint, SkImage::kAllow_CachingHint }) {
SkAutoPixmapStorage scaled;
scaled.alloc(info);
if (!image->scalePixels(scaled, SkSamplingOptions(SkFilterMode::kLinear), chint)) {
ERRORF(reporter, "Failed to scale image");
continue;
}
check_scaled_pixels(reporter, &scaled, expected);
}
}
DEF_TEST(ImageScalePixels, reporter) {
const SkPMColor pmRed = SkPackARGB32(0xFF, 0xFF, 0, 0);
const SkColor red = SK_ColorRED;
// Test raster image
SkImageInfo info = SkImageInfo::MakeN32Premul(1, 1);
sk_sp<SkSurface> surface = SkSurface::MakeRaster(info);
surface->getCanvas()->clear(red);
sk_sp<SkImage> rasterImage = surface->makeImageSnapshot();
test_scale_pixels(reporter, rasterImage.get(), pmRed);
// Test encoded image
sk_sp<SkData> data = rasterImage->encodeToData();
sk_sp<SkImage> codecImage = SkImage::MakeFromEncoded(data);
test_scale_pixels(reporter, codecImage.get(), pmRed);
}
DEF_GPUTEST_FOR_RENDERING_CONTEXTS(ImageScalePixels_Gpu, reporter, ctxInfo) {
const SkPMColor pmRed = SkPackARGB32(0xFF, 0xFF, 0, 0);
const SkColor red = SK_ColorRED;
SkImageInfo info = SkImageInfo::MakeN32Premul(16, 16);
sk_sp<SkSurface> surface = SkSurface::MakeRenderTarget(ctxInfo.directContext(),
SkBudgeted::kNo, info);
surface->getCanvas()->clear(red);
sk_sp<SkImage> gpuImage = surface->makeImageSnapshot();
test_scale_pixels(reporter, gpuImage.get(), pmRed);
}
static sk_sp<SkImage> any_image_will_do() {
return GetResourceAsImage("images/mandrill_32.png");
}
DEF_TEST(Image_nonfinite_dst, reporter) {
auto surf = SkSurface::MakeRasterN32Premul(10, 10);
auto img = any_image_will_do();
for (SkScalar bad : { SK_ScalarInfinity, SK_ScalarNaN}) {
for (int bits = 1; bits <= 15; ++bits) {
SkRect dst = { 0, 0, 10, 10 };
if (bits & 1) dst.fLeft = bad;
if (bits & 2) dst.fTop = bad;
if (bits & 4) dst.fRight = bad;
if (bits & 8) dst.fBottom = bad;
surf->getCanvas()->drawImageRect(img, dst, SkSamplingOptions());
// we should draw nothing
ToolUtils::PixelIter iter(surf.get());
while (void* addr = iter.next()) {
REPORTER_ASSERT(reporter, *(SkPMColor*)addr == 0);
}
}
}
}
static sk_sp<SkImage> make_yuva_image(GrDirectContext* dContext) {
SkAutoPixmapStorage pm;
pm.alloc(SkImageInfo::Make(1, 1, kAlpha_8_SkColorType, kPremul_SkAlphaType));
SkYUVAInfo yuvaInfo({1, 1},
SkYUVAInfo::PlaneConfig::kY_U_V,
SkYUVAInfo::Subsampling::k444,
kJPEG_Full_SkYUVColorSpace);
const SkPixmap pmaps[] = {pm, pm, pm};
auto yuvaPixmaps = SkYUVAPixmaps::FromExternalPixmaps(yuvaInfo, pmaps);
return SkImage::MakeFromYUVAPixmaps(dContext, yuvaPixmaps);
}
DEF_GPUTEST_FOR_ALL_CONTEXTS(ImageFlush, reporter, ctxInfo) {
auto dContext = ctxInfo.directContext();
auto ii = SkImageInfo::Make(10, 10, kRGBA_8888_SkColorType, kPremul_SkAlphaType);
auto s = SkSurface::MakeRenderTarget(dContext, SkBudgeted::kYes, ii, 1, nullptr);
s->getCanvas()->clear(SK_ColorRED);
auto i0 = s->makeImageSnapshot();
s->getCanvas()->clear(SK_ColorBLUE);
auto i1 = s->makeImageSnapshot();
s->getCanvas()->clear(SK_ColorGREEN);
// Make a YUVA image.
auto i2 = make_yuva_image(dContext);
// Flush all the setup work we did above and then make little lambda that reports the flush
// count delta since the last time it was called.
dContext->flushAndSubmit();
auto numSubmits =
[dContext,
submitCnt = dContext->priv().getGpu()->stats()->numSubmitToGpus()]() mutable {
int curr = dContext->priv().getGpu()->stats()->numSubmitToGpus();
int n = curr - submitCnt;
submitCnt = curr;
return n;
};
// Images aren't used therefore flush is ignored, but submit is still called.
i0->flushAndSubmit(dContext);
i1->flushAndSubmit(dContext);
i2->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 3);
// Syncing forces the flush to happen even if the images aren't used.
i0->flush(dContext);
dContext->submit(true);
REPORTER_ASSERT(reporter, numSubmits() == 1);
i1->flush(dContext);
dContext->submit(true);
REPORTER_ASSERT(reporter, numSubmits() == 1);
i2->flush(dContext);
dContext->submit(true);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Use image 1
s->getCanvas()->drawImage(i1, 0, 0);
// Flushing image 0 should do nothing, but submit is still called.
i0->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Flushing image 1 should flush.
i1->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Flushing image 2 should do nothing, but submit is still called.
i2->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Use image 2
s->getCanvas()->drawImage(i2, 0, 0);
// Flushing image 0 should do nothing, but submit is still called.
i0->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Flushing image 1 do nothing, but submit is still called.
i1->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Flushing image 2 should flush.
i2->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
REPORTER_ASSERT(reporter, static_cast<SkImage_GpuYUVA*>(as_IB(i2.get()))->isTextureBacked());
s->getCanvas()->drawImage(i2, 0, 0);
// Flushing image 0 should do nothing, but submit is still called.
i0->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Flushing image 1 do nothing, but submit is still called.
i1->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
// Flushing image 2 should flush.
i2->flushAndSubmit(dContext);
REPORTER_ASSERT(reporter, numSubmits() == 1);
}
#include "src/shaders/SkImageShader.h"
constexpr SkM44 gCentripetalCatmulRom
(0.0f/2, -1.0f/2, 2.0f/2, -1.0f/2,
2.0f/2, 0.0f/2, -5.0f/2, 3.0f/2,
0.0f/2, 1.0f/2, 4.0f/2, -3.0f/2,
0.0f/2, 0.0f/2, -1.0f/2, 1.0f/2);
constexpr SkM44 gMitchellNetravali
( 1.0f/18, -9.0f/18, 15.0f/18, -7.0f/18,
16.0f/18, 0.0f/18, -36.0f/18, 21.0f/18,
1.0f/18, 9.0f/18, 27.0f/18, -21.0f/18,
0.0f/18, 0.0f/18, -6.0f/18, 7.0f/18);
DEF_TEST(image_cubicresampler, reporter) {
auto diff = [reporter](const SkM44& a, const SkM44& b) {
const float tolerance = 0.000001f;
for (int r = 0; r < 4; ++r) {
for (int c = 0; c < 4; ++c) {
float d = std::abs(a.rc(r, c) - b.rc(r, c));
REPORTER_ASSERT(reporter, d <= tolerance);
}
}
};
diff(SkImageShader::CubicResamplerMatrix(1.0f/3, 1.0f/3), gMitchellNetravali);
diff(SkImageShader::CubicResamplerMatrix(0, 1.0f/2), gCentripetalCatmulRom);
}
DEF_TEST(image_subset_encode_skbug_7752, reporter) {
sk_sp<SkImage> image = GetResourceAsImage("images/mandrill_128.png");
const int W = image->width();
const int H = image->height();
auto check_roundtrip = [&](sk_sp<SkImage> img) {
auto img2 = SkImage::MakeFromEncoded(img->encodeToData());
REPORTER_ASSERT(reporter, ToolUtils::equal_pixels(img.get(), img2.get()));
};
check_roundtrip(image); // should trivially pass
check_roundtrip(image->makeSubset({0, 0, W/2, H/2}));
check_roundtrip(image->makeSubset({W/2, H/2, W, H}));
check_roundtrip(image->makeColorSpace(SkColorSpace::MakeSRGBLinear()));
}