SkWriter32::snapshotAsData() is no longer performance critical.
It's only used when we're serializing to disk.
BUG=skia:2289
Review URL: https://codereview.chromium.org/875403005
SkProxyCanvas is redundant with SkNWayCanvas, and means another class
we have to keep in sync with the SkCanvas interface.
Remove tests which use an SkProxyCanvas.
Requires a change to chromium.
BUG=skia:3279
BUG=skia:500
Review URL: https://codereview.chromium.org/886813002
SkTypeface already requires typeface streams to support SkStreamAsset
in practice, and in practice all users are already supplying them.
Review URL: https://codereview.chromium.org/869763002
Fold alpha to the inner savelayer in savelayer-savelayer-restore
patterns such as this:
SaveLayer (non-opaque)
Save
ClipRect
SaveLayer
Restore
Restore
Restore
Current blink generates these for example for SVG content such as this:
<path style="opacity:0.5 filter:url(#blur_filter)"/>
The outer save layer is due to the opacity and the inner one is due to
blur filter being implemented with picture image filter.
Reduces layers in desk_carsvg.skp testcase from 115 to 78.
BUG=skia:3119
Review URL: https://codereview.chromium.org/835973005
==7023== 5,056 (896 direct, 4,160 indirect) bytes in 8 blocks are definitely lost in loss record 947 of 2,656
==7023== at 0x4C2B1C7: operator new(unsigned long) (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==7023== by 0x82CB99: SkNewImageFromBitmapTexture(SkBitmap const&, int, SkSurface::Budgeted) (SkImage_Gpu.cpp:64)
==7023== by 0x81A1CF: SkSurface_Gpu::onNewImageSnapshot(SkSurface::Budgeted) (SkSurface_Gpu.cpp:50)
==7023== by 0x694F9D: SkSurface::newImageSnapshot(SkSurface::Budgeted) (SkSurface_Base.h:92)
==7023== by 0x55EA7C: test_Surface(skiatest::Reporter*, GrContextFactory*) (SurfaceTest.cpp:485)
==7023== by 0x407CB2: run_test(skiatest::Test*) (DM.cpp:399)
==7023== by 0x408382: run_enclave_and_gpu_tests(SkTArray<Task, false>*) (DM.cpp:411)
==7023== by 0x68B71E: SkTaskGroup::wait() (SkTaskGroup.cpp:67)
==7023== by 0x40934E: dm_main() (DM.cpp:455)
==7023== by 0x409483: main (DM.cpp:477)
http://build.chromium.org/p/client.skia/builders/Test-Ubuntu12-ShuttleA-GTX550Ti-x86_64-Release-Valgrind/builds/276/steps/dm/logs/stdioTBR=bsalomon@google.com
BUG=skia:
Review URL: https://codereview.chromium.org/872183002
I have been hacking at this test trying to understand why the N5 is
still sometimes crashy (it seems, less but not zero now). No luck so far.
But, while I've been reading and hacking at it, I think I've made a
few small improvements, mostly to readability.
BUG=skia:
Review URL: https://codereview.chromium.org/870803002
This is a speculative fix for the crashing N5 bots. It looks like the bots are
always failing in or around FloatingPointTextureTest.
It looks like sometimes they're hitting a SIGBUS, which I suspect is stack
overflow. FloatingPointTextureTest allocates ~320K on the stack, which may
be too much. This CL moves those buffers to the heap. For consistency I did
the same with the half-float tests, though they're only using ~1/8th the stack.
It looks like sometimes the bots are failing to malloc. I don't understand that,
and this CL doesn't address that directly. But it's possible this is still a stack
overflow, just trashing RAM and causing arbitrary mayhem instead of a SIGBUS.
I have no idea why this is a problem only on the N5. I have been unable to
reproduce this locally, neither with a K N5 nor an L N5, but the bots are pretty
reliable.
NOTREECHECKS=true
BUG=skia:
Review URL: https://codereview.chromium.org/871623002
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
Reason for revert:
Our Valgrind bot just spewed out a weird error. I don't know if it's related, but it looks at least like one of the stacks was in the right area, so I'm going to revert this precautionarily. Sorry if this is a false positive.
http://build.chromium.org/p/client.skia/builders/Test-Ubuntu12-ShuttleA-GTX550Ti-x86_64-Release-Valgrind/builds/266/steps/dm/logs/stdio
Original issue's description:
> Make GrScratchKey memory buffer correct size on copy
>
> Scratch key memory buffer of a copy of a key was too big. The (new) copy
> was N times uint32_t bytes instead of N bytes.
>
> Adds few tests to resource cache. These tests would not catch the too
> big buffer. This is just a precaution for too small buffers. The main
> idea of the test change is that the scratch key should contain some
> information, so that lookup with a scratch key can also return no
> match. Otherwise testing of scratch lookup result is not indicative of
> correct code (eg. no-information scratch key will always match).
>
> Committed: https://skia.googlesource.com/skia/+/711ef4831363fb8cbdf061dc2c36c65b13c0ccf2TBR=bsalomon@google.com,kkinnunen@nvidia.com
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
Review URL: https://codereview.chromium.org/864833003
Scratch key memory buffer of a copy of a key was too big. The (new) copy
was N times uint32_t bytes instead of N bytes.
Adds few tests to resource cache. These tests would not catch the too
big buffer. This is just a precaution for too small buffers. The main
idea of the test change is that the scratch key should contain some
information, so that lookup with a scratch key can also return no
match. Otherwise testing of scratch lookup result is not indicative of
correct code (eg. no-information scratch key will always match).
Review URL: https://codereview.chromium.org/860333002
skiatest::Test class is now a simple struct. Some
functionalty, such as counting errors or timing is now
handled elsewhere.
skiatest:Reporter is now a simpler abstract class. The two
implementations handle test errors.
DM and pathops_unittest updated.
Review URL: https://codereview.chromium.org/830513004
Since only XP's can read dst now, there is no reason to have this query on GrFP.
This also triggered a chain reaction of cleaning up/removing unnecessary code
elsewhere.
BUG=skia:
Review URL: https://codereview.chromium.org/851143003
- Added new classes to contain YUV planes of memory, along with the associated data.
- Used these classes in load_yuv_texture() to enable YUV planes caching
- Added a unit test for the new cache
BUG=450021
Review URL: https://codereview.chromium.org/851273003
Restructure SkGpuDevice creation:
*SkSurfaceProps are optional.
*Use SkSurfaceProps to communicate DF text rather than a flag.
*Tell SkGpuDevice::Create whether RT comes from cache or not.
Review URL: https://codereview.chromium.org/848903004
There's no gyp references to these new files,
so this should only have the effect of reducing
the size of the commit that turns this code on.
TBR=
Review URL: https://codereview.chromium.org/853223002
Fold alpha of a save layer call to the subsequent draw call paint, even when
the draw call paint color is already non-opaque.
Comparing the difference of the unoptimized and the optimized call pattern
with all (layer alpha, draw alpha) combinations, this produces
off-by-one pixels ~50% of the time.
Reduces layers of current desk_carsvg.skp (recorded with cross-process
picture image filters allowed) from 122 to 115.
BUG=skia:3119
Review URL: https://codereview.chromium.org/840483005
They're only hints. We can ignore them in this optimization.
My only hesitation is that this pattern doesn't seem to happen in our .skps.
BUG=skia:
Review URL: https://codereview.chromium.org/849243002
This avoids the problem of a newly created uncached texture causing a purge of cached resources.
BUG=chromium:445885
Review URL: https://codereview.chromium.org/846303002
This code requires fewer macros to use it (just one), has less code in macro
definitions, and has simpler synchronization code (just atomic ints, no SkOnce,
no SkMutex, etc.)
A minor downside, we lose indentation and reverse-ordering in the final report:
Leaked SkRefCntBase: 7
Leaked SkFontMgr: 1
Leaked SkWeakRefCnt: 1
Leaked SkTypeface: 1
Leaked SkFlattenable: 3
Leaked SkXfermode: 3
Leaked SkPathRef: 1
Leaked SkPixelRef: 1
Leaked SkMallocPixelRef: 1
becomes
Leaked SkXfermode: 3
Leaked SkMallocPixelRef: 1
Leaked SkPixelRef: 1
Leaked SkPathRef: 1
Leaked SkFlattenable: 3
Leaked SkTypeface: 1
Leaked SkWeakRefCnt: 1
Leaked SkFontMgr: 1
Leaked SkRefCntBase: 7
This is motivated by wanting to land https://codereview.chromium.org/806473006/,
which makes sure all static use of SkOnce are in global scope. The current
implementation of SkInstCnt uses them in function scope, which isn't safe.
BUG=skia:
No public API changes.
TBR=reed@google.com
Review URL: https://codereview.chromium.org/841263004
* Add Streamer utility class which measures the current
pdf offset by calling SkWStream::bytesWritten(). Calls
SkPDFCatalog::setFileOffset() and SkPDFObject::emit() at
the same time to guarantee that everything works out.
* SkPDFCatalog::setFileOffset() no longer calculates the
object's size.
* SkPDFCatalog::setSubstituteResourcesOffsets() removed.
* SkPDFCatalog::emitSubstituteResources() removed and
getSubstituteList() made public in its place.
* Remove SkPDFPage::getPageSize and SkPDFPage::emitPage.
Replace with SkPDFPage::getContentStream().
* SkPDFObject::getOutputSize no longer virtual, only used in
unit tests. All SkPDFObject subclasses getOutputSize()
overrides removed.
* SkPDFObject::getIndirectOutputSize removed.
* PDFPrimitivesTest updated for new functions.
Review URL: https://codereview.chromium.org/846023003
Replace virutal SkPDFObject::emitObject(s, c, true) with non-virtual
SkPDFObject::emitIndirectObject(s, c), since none of the subclasses do
(or should do) anything different.
Replace object->emitObject(s, c, false) with object->emitObject(s, c)
This is one step in simplifying the SkPDFObject interface to allow for
the next step in refactoring.
Review URL: https://codereview.chromium.org/827733004
This fixes every case where virtual and SK_OVERRIDE were on the same line,
which should be the bulk of cases. We'll have to manually clean up the rest
over time unless I level up in regexes.
for f in (find . -type f); perl -p -i -e 's/virtual (.*)SK_OVERRIDE/\1SK_OVERRIDE/g' $f; end
BUG=skia:
Review URL: https://codereview.chromium.org/806653007
There are only a handful of SkFlattenables that are not flattenable. That
there are any seems highly illogical. To make this look less like a normal
thing, this removes both macros that marked SkFlattenables as non-flattenable
(in slightly different ways).
The handful of SkFlattenables in our codebase that can't be flattened now
assert violently that they can't be flattened. They're internal or
part of animator... places where we'll never actually flatten them.
TestLooper and DummyRasterizer were so trivial that I just made them flattenable.
BUG=skia:
Review URL: https://codereview.chromium.org/841753002
This removes the SkRecords::Clear struct and everything that refers to it.
Notice there is nothing actually creating a Clear, which means this is all
dead code.
Now that all ops obey the clip, I don't think we need the weird
inflate-empty-to-epsilon hack for BBH queries.
BUG=skia:
Review URL: https://codereview.chromium.org/835813002
Added in check for sized internal format.
Made choice between RED and ALPHA orthogonal to HALF_FLOAT and
HALF_FLOAT_OES.
Enabled rendertarget support on ES 2.0.
Review URL: https://codereview.chromium.org/805033002
This CL fixes 5 bugs related to hoisting image filters:
For image filters the src layer (the one prior to filtering) often needs to be
smaller then the final layer. This requires the saveLayer's optional bounds
to be stored (in SkLayerInfo.h and SkRecordDraw.cpp) and then used in
compute_source_rect and carried around in GrCachedLayer.
The image filters can add an extra offset to the final draw operation.
This is now computed in GrLayerHoister::FilterLayer and carried around in
GrCachedLayer.
Filtered layers must use exact matches. This is now done in GrLayerCache::lock.
The filter cache requires a valid matrix so it can compute the correct offset.
This is now done in GrLayerHoister::FilterLayer.
Filtered layers need to be drawn with drawSprite while unfiltered (and therefore
hopefully atlased) layers can be drawn with drawBitmap. This is now done in
draw_replacement_bitmap.
Committed: https://skia.googlesource.com/skia/+/702eb9622102599d94ab6798e6227cf29f48c2d3
Review URL: https://codereview.chromium.org/803183003
Reason for revert:
Breaking DM
Original issue's description:
> Fix layer hoisting image filter corner cases
>
> This CL fixes 5 bugs related to hoisting image filters:
>
> For image filters the src layer (the one prior to filtering) often needs to be
> smaller then the final layer. This requires the saveLayer's optional bounds
> to be stored (in SkLayerInfo.h and SkRecordDraw.cpp) and then used in
> compute_source_rect and carried around in GrCachedLayer.
>
> The image filters can add an extra offset to the final draw operation.
> This is now computed in GrLayerHoister::FilterLayer and carried around in
> GrCachedLayer.
>
> Filtered layers must use exact matches. This is now done in GrLayerCache::lock.
>
> The filter cache requires a valid matrix so it can compute the correct offset.
> This is now done in GrLayerHoister::FilterLayer.
>
> Filtered layers need to be drawn with drawSprite while unfiltered (and therefore
> hopefully atlased) layers can be drawn with drawBitmap. This is now done in
> draw_replacement_bitmap.
>
> Committed: https://skia.googlesource.com/skia/+/702eb9622102599d94ab6798e6227cf29f48c2d3TBR=bsalomon@google.com
NOTREECHECKS=true
NOTRY=true
Review URL: https://codereview.chromium.org/790643009
This CL fixes 5 bugs related to hoisting image filters:
For image filters the src layer (the one prior to filtering) often needs to be
smaller then the final layer. This requires the saveLayer's optional bounds
to be stored (in SkLayerInfo.h and SkRecordDraw.cpp) and then used in
compute_source_rect and carried around in GrCachedLayer.
The image filters can add an extra offset to the final draw operation.
This is now computed in GrLayerHoister::FilterLayer and carried around in
GrCachedLayer.
Filtered layers must use exact matches. This is now done in GrLayerCache::lock.
The filter cache requires a valid matrix so it can compute the correct offset.
This is now done in GrLayerHoister::FilterLayer.
Filtered layers need to be drawn with drawSprite while unfiltered (and therefore
hopefully atlased) layers can be drawn with drawBitmap. This is now done in
draw_replacement_bitmap.
Review URL: https://codereview.chromium.org/803183003
Reason for revert:
Test-Win7-ShuttleA-HD2000-x86-Debug-GDI
Original issue's description:
> Use SkPaint::getFontBounds() for text bounding boxes in pictures.
>
> Now that SkTextBlobs have landed, this is a perf no-op, but it at least lets us eliminate a bunch of questionable hacks.
>
> CQ_EXTRA_TRYBOTS=client.skia:Test-Win7-ShuttleA-HD2000-x86-Debug-Trybot,Test-Mac10.8-MacMini4.1-GeForce320M-x86_64-Debug-Trybot
>
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/02d2b9831579173e783569530ab7bae08de907e9TBR=reed@google.com,mtklein@chromium.org
NOTREECHECKS=true
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/789793004
Now that SkTextBlobs have landed, this is a perf no-op, but it at least lets us eliminate a bunch of questionable hacks.
CQ_EXTRA_TRYBOTS=client.skia:Test-Win7-ShuttleA-HD2000-x86-Debug-Trybot,Test-Mac10.8-MacMini4.1-GeForce320M-x86_64-Debug-Trybot
BUG=skia:
Review URL: https://codereview.chromium.org/805983003
Remove SkLCGRandom. We already decided the new one was better, which is
why we wrote the new SkRandom.
Convert GMs that were using SkLCGRandom to use the improved SkRandom.
Motivated by the fact that these GMs draw differently on some runs. We
believe this to be a result of using the old SkLCGRandom.
Add each of the tests that were using SkLCGRandom to ignore-tests.txt,
since we expect they'll draw differently using SkRandom.
Move a trimmed down version of SkLCGRandom into SkDiscretePathEffect.
In order to preserve the old behavior, trim down SkLCGRandom to only
the methods used by SkDiscretePathEffect, and hide it in
SkDiscretePathEffect's cpp file.
BUG=skia:3241
Review URL: https://codereview.chromium.org/805963002
When we added the 64K allocation cap, the bots showed we took a perf hit
on some large .skps like desk_pokemonwiki.skp, despite not seeing a local
effect. I'm still not seeing that locally, but I'd like to try removing the cap on
the bots to see what happens. For big monolithic pictures, really packing into
memory tightly is probably not as important as it is for tiny ones.
Similarly, we're probably being too cautious about making tiny allocations.
Today we start at 16 bytes, which isn't really enough to record anything.
Even the smallest picture, say,
save
clipRect
drawRect
restore
requires ~200 bytes, so we might as well move our minimum block size up
near there.
I don't know if 16 bytes is too small to start for GrTextStrikes, so I've left the
behavior the same (though the max is still gone).
Local recording performance is neutral-to-positive:
tabl_deviantart.skp 126us -> 129us 1.02x
tabl_nytimes.skp 110us -> 112us 1.02x
tabl_cuteoverload.skp 521us -> 530us 1.02x
desk_mobilenews.skp 673us -> 682us 1.01x
desk_chalkboard.skp 843us -> 854us 1.01x
desk_sfgate.skp 528us -> 535us 1.01x
desk_silkfinance.skp 68.2us -> 69us 1.01x
desk_youtube.skp 623us -> 629us 1.01x
desk_blogger.skp 472us -> 475us 1.01x
desk_jsfiddlehumperclip.skp 42.2us -> 42.5us 1.01x
desk_espn.skp 255us -> 256us 1.01x
desk_ebay.skp 174us -> 174us 1x
desk_twitter.skp 454us -> 455us 1x
tabl_pravda.skp 200us -> 201us 1x
desk_wordpress.skp 782us -> 784us 1x
desk_samoasvg.skp 762us -> 761us 1x
tabl_mozilla.skp 1.58ms -> 1.58ms 1x
tabl_slashdot.skp 107us -> 107us 1x
tabl_techmeme.skp 102us -> 102us 0.99x
tabl_gamedeksiam.skp 729us -> 724us 0.99x
tabl_nofolo.skp 65.3us -> 64.7us 0.99x
desk_gmailthread.skp 339us -> 336us 0.99x
tabl_sahadan.skp 91us -> 90us 0.99x
desk_yahooanswers.skp 144us -> 142us 0.99x
tabl_cnet.skp 143us -> 141us 0.99x
tabl_googleblog.skp 206us -> 203us 0.99x
tabl_cnn.skp 160us -> 158us 0.99x
tabl_frantzen.skp 50.5us -> 49.6us 0.98x
desk_linkedin.skp 328us -> 323us 0.98x
tabl_digg.skp 790us -> 769us 0.97x
desk_jsfiddlebigcar.skp 40.6us -> 39.5us 0.97x
desk_mapsvg.skp 1.57ms -> 1.52ms 0.97x
tabl_gmail.skp 19.4us -> 18.6us 0.96x
tabl_hsfi.skp 9.81us -> 9.11us 0.93x
BUG=skia:
Review URL: https://codereview.chromium.org/793033002
It's unclear what params should be used for half float alpha
rendertargets, so they are disabled for the moment.
Review URL: https://codereview.chromium.org/799593002
Gives more flexibility to the caller to decide whether to use the
encoded data returned by refEncodedData().
Provides an implementation that supports the old version of
SkPicture::serialize().
TODO: Update Chrome, so we can remove SK_LEGACY_ENCODE_BITMAP entirely
BUG=skia:3190
Review URL: https://codereview.chromium.org/784643002
This includes:
-Having an actual XP stage at the end of the gl pipeline.
-All Blending work is handled by XP until actually setting GL blend states
-GLPrograms test to test XP
BUG=skia:
Review URL: https://codereview.chromium.org/764643004
In addition, NVPR makes this very complicated, and I haven't quite figured out a good way to handle it, so for now color and coverage DO live on optstate, but I will figure out some way to refactor that in future CLs.
BUG=skia:
Review URL: https://codereview.chromium.org/783763002
Reason for revert:
Compilation is failing on some bots
Original issue's description:
> Replace EncodeBitmap with an interface.
>
> Gives more flexibility to the caller to decide whether to use the
> encoded data returned by refEncodedData().
>
> Provides an implementation that supports the old version of
> SkPicture::serialize().
>
> TODO: Update Chrome, so we can remove SK_LEGACY_ENCODE_BITMAP entirely
>
> BUG=skia:3190
>
> Committed: https://skia.googlesource.com/skia/+/0c4aba6edb9900c597359dfa49d3ce4a41bc5dd1TBR=reed@google.com,scroggo@google.com
NOTREECHECKS=true
NOTRY=true
BUG=skia:3190
Review URL: https://codereview.chromium.org/787833002
Gives more flexibility to the caller to decide whether to use the
encoded data returned by refEncodedData().
Provides an implementation that supports the old version of
SkPicture::serialize().
TODO: Update Chrome, so we can remove SK_LEGACY_ENCODE_BITMAP entirely
BUG=skia:3190
Review URL: https://codereview.chromium.org/784643002
Reason for revert:
break many gm's
Original issue's description:
> Make all blending up to GrOptDrawState be handled by the xp/xp factory.
>
> In this cl the blending information is extracted for the xp and stored in the ODS
> which is then used as it currently is. In the follow up cl, an XP backend will be added
> and at that point all blending work will take place inside XP's.
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/7c66342a399b529634bed0fabfaa562db2c0dbd4TBR=bsalomon@google.com
NOTREECHECKS=true
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/766653008
In this cl the blending information is extracted for the xp and stored in the ODS
which is then used as it currently is. In the follow up cl, an XP backend will be added
and at that point all blending work will take place inside XP's.
BUG=skia:
Review URL: https://codereview.chromium.org/759713002
PathOps relies on isConvex() only returning true for trivially
convex paths. The old logic also returns true if the paths that
contain NaNs and Infinities. Return kUnknown_Convexity instead
in those cases and in cases where the convexity logic computes
intermediaries that overflow.
Review URL: https://codereview.chromium.org/784593002
This allows us to create distance field textures with better precision,
which may help text quality.
BUG=skia:3103
Review URL: https://codereview.chromium.org/762923003
This was needed for pictures before v33, and we're now requiring v35+.
Will follow up with the same for skia/ext/pixel_ref_utils_unittest.cc
BUG=skia:
Review URL: https://codereview.chromium.org/769953002
Adding the rendering canvas' CTM to the layer hoisting key (i.e., Add support for hoisting layers in pictures drawn with a matrix - https://codereview.chromium.org/748853002/) has increased the cache miss rate due to accumulated floating point error. This CL fixes part of the issue by using the chain of operation indices leading to each saveLayer as the key. The canvas' CTM must still form part of the key but should be less subject to accumulated error.
BUG=skia:2315
Review URL: https://codereview.chromium.org/753253002
