Store mipmap levels in deferred texture image

This is a follow-up to https://codereview.chromium.org/2034933003/ which
was reverted due to a memory leak.

When creating the deferred texture image, detect if using medium / high
quality. If so, generate and store mipmaps in the deferred texture
image.

When creating a texture from that be sure to read it back out.

BUG=578304
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2115023002

Review-Url: https://codereview.chromium.org/2115023002

SkPDF: SkPDFFont organization changes.

SkPDFFont:
  - SkPDFType1Font::populate() encode advances correctly.
  - break out logically independent code into new files:
    * SkPDFConvertType1FontStream
    * SkPDFMakeToUnicodeCmap
    SkPDFFont.cpp is now 380 lines smaller.
    Expose `SkPDFAppendCmapSections()` for testing.

SkPDFFontImpl.h
  - Fold into SkPDFFont.

SkPDFConvertType1FontStream:
  - Now assume given a SkStreamAsset

SkPDFFont:
  - AdvanceMetric now hidden in a anonymous namespace.

No public API changes.
TBR=reed@google.com

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221163002

Review-Url: https://codereview.chromium.org/2221163002

Include EXT_window_rectangles API

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221393004

Review-Url: https://codereview.chromium.org/2221393004

Use SkNVRefCnt for a couple common types.

These types are ref-counted, but don't otherwise need a vtable.
This makes them good candidates for SkNVRefCnt.

Destruction can be a little more direct, and if nothing else,
sizeof(T) will get a little smaller by dropping the vptr.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2232433002

Review-Url: https://codereview.chromium.org/2232433002

SkLiteDL: tiny perf tweak

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2229003003

Review-Url: https://codereview.chromium.org/2229003003

Use slave_build instead of checkout in run/api.py

BUG=skia:5620
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221423002

Review-Url: https://codereview.chromium.org/2221423002

Add RasterWindowContext_mac

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2228883003

Review-Url: https://codereview.chromium.org/2228883003

Sample app uses Ganesh before calling SkGraphics::Init().

This causes inconsistent hashing.  Before Init(), we use portable Murmur3,
after a faster SSE4.2-based hash.

To fix, call SkGraphics::Init() first.

BUG=skia:5625
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2229853002

Review-Url: https://codereview.chromium.org/2229853002

Move seg_to to a new header, define SkSegType enum there

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221203002

Review-Url: https://codereview.chromium.org/2221203002

moved code into onDrawShadowedPic, only renders into shadow maps if needed

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2220633002

Review-Url: https://codereview.chromium.org/2220633002

Rework Overstroke GM to show differences between expected and actual

Draw the perpendiculars of the curve with the stroke width. See Cary's
SampleQuadStroker.cpp for what we are attempting to mimic.

BUG=589769,skia:5405,skia:5406
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2218773003

Review-Url: https://codereview.chromium.org/2218773003

Refactor various gradient effect classes

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221133002

Review-Url: https://codereview.chromium.org/2221133002

Purge the SkLiteDL freelist when PurgeAllCaches() is called.

Seems like the polite thing to do.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2228953002

Review-Url: https://codereview.chromium.org/2228953002

SkLiteRecorder: don't tell SkCanvas about clips

If no one reads our clip, and we don't expect quickReject() to help,
we can probably get away without maintaining a proper clip stack.

This puts us at about 0.6-0.7x of previous record cost.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2227833004

Review-Url: https://codereview.chromium.org/2227833004

add flaky test option

One fuzzer generated pathops test
fails everywhere except for one
builder. Add a flaky state to the
pathops test framework to handle
this until I can investigate
further.

TBR=reed@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221153005

Review-Url: https://codereview.chromium.org/2221153005

Check more GLES versions when creating context

Unifies the context creation for GL and GLES into a single loop.

BUG=skia:5403
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2201033003

Review-Url: https://codereview.chromium.org/2201033003

template intersection fuzz fixes

Plumb in the ability to ignore asserts for out of range input
deeper into the template intersection code.

Exit gracefully when error conditions are found.

TBR=reed@google.com
BUG=632607,632628,633063,633065,634251,633608
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2224823004

Review-Url: https://codereview.chromium.org/2224823004

[SVGDom] Improve whitespace handling in style parsing

Handle whitespace-padded style names/values.

R=stephana@google.com,robertphillips@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2225623002

Review-Url: https://codereview.chromium.org/2225623002

make SpecialLineRec::addSegment assert more permissive to accomodate rounding error

Because the driving code that calls SpecialLineRec::addSegment uses double precision,
when converted back to single precision this may round such that d0 and d1 are the same.
This causes the assert to trigger. The simplest thing to do here is to modify the assert
to not be so picky and allow the rare/occasional equality.

Downstream Firefox bug report: https://bugzilla.mozilla.org/show_bug.cgi?id=1290628

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2209303004

Review-Url: https://codereview.chromium.org/2209303004

[SVGDom] Improved transform parsing

Update 'transform' attribute parsing to a more robust, SkSVGAttributeParser-based implementation.

R=robertphillips@google.com,stephana@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2220933003

Review-Url: https://codereview.chromium.org/2220933003

add missing include

CQ_INCLUDE_TRYBOTS=master.client.skia:Perf-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Debug-MSAN-Trybot

Review-Url: https://codereview.chromium.org/2226783002

Add initial CurveMeasure code

- This code is entirely private and is not being used by anything.

- In a future CL we will write a class that uses CurveMeasure to compute dash points. In order to determine whether CurveMeasure or PathMeasure should be faster, we need the dash info (the sum of the on/off intervals and how many there are)

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2187083002

Review-Url: https://codereview.chromium.org/2187083002

SkLiteDL: control reuse by a single parameter

This simply caps the number of times a display list can be reused.
As this number goes up, the average amount of memory we cache goes up
and the expected number of mallocs per SkLiteDL::New() goes down.

This strategy does not need a hard-coded cap on how many display lists
to cache, or how big they can grow.

TBR=herb@google.com

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2226813002

Review-Url: https://codereview.chromium.org/2226813002

Split distance vector into direction and magnitude components

The vector was split because the vector's users need it split, and we were creating it from the split components in the first place, so it made sense to skip that step.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2223053002

Review-Url: https://codereview.chromium.org/2223053002

[SVGDom] Add viewBox support

The main feature is <svg> viewBox and proper viewport support, but the CL
touches a few other things:

* refactor SkSVGRenderContext to auto-restore canvas state, and split the
  presentation bits into a separate CoW SkSVGPresentationContext

* introduce SkSVGNode::onPrepareToRender(), as a way for nodes to push their
  custom state before the actual onRender() call (instead of relying on
  non-virtual SkSVGNode to know about all possible state bits)

* add a "Type" suffix to SVG types, to disambiguate (e.g. SkSVGRectType vs.
  SkSVGRect)

R=robertphillips@google.com,stephana@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2222793002

Review-Url: https://codereview.chromium.org/2222793002

Fixed comment about SkLightingShader::Context memory handling

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2225093002

Review-Url: https://codereview.chromium.org/2225093002

SkTypeface_win_dw.cpp remove unused fn

TBR=bungeman@google.com
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2226653003

Review-Url: https://codereview.chromium.org/2226653003

Encapsulate GrReducedClip result in class members

Updates GrReducedClip to store its result in class members instead of
various pointer arguments. This helps clean up calling code and will
make it easier to reduce the clip higher in the stack.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2222873002

Review-Url: https://codereview.chromium.org/2222873002

Revert of Cleanup use of legacy SkSurface creation methods a bit (patchset #2 id:20001 of https://codereview.chromium.org/2223023002/ )

Reason for revert:
webkit_headless!

Original issue's description:
> Cleanup use of legacy SkSurface creation methods a bit
>
> This:
>    removes the SK_SUPPORT_LEGACY_NEW_SURFACE_API flag from the bzl build
>    cleans up some remaining uses of the old API
>
> GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2223023002
>
> Committed: https://skia.googlesource.com/skia/+/f21cd16228c20927fa4e9b937d6951471cfaa37a

TBR=benjaminwagner@google.com,egdaniel@google.com
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true

Review-Url: https://codereview.chromium.org/2220933004

SkPDF/SkAdvancedTypefaceMetrics: simplify ATM, PDF takes over

No public API changes.
TBR=reed@google.com

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2222523003

Review-Url: https://codereview.chromium.org/2222523003

Use sse4.2 CRC32 instructions to hash when available.

About 9x faster than Murmur3 for long inputs.

Most of this is a mechanical change from SkChecksum::Murmur3(...) to SkOpts::hash(...).

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2208903002
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot;master.client.skia.compile:Build-Ubuntu-GCC-x86_64-Release-CMake-Trybot,Build-Mac-Clang-x86_64-Release-CMake-Trybot

Review-Url: https://codereview.chromium.org/2208903002

use onPeekBitmap instead of asLegacyBitmap

Since we know we are raster-backed at this point, it is fine, plus it avoids the copy-mutable logic in asLegacyBitmap, which is not needed here (and is slower if our bitmap was built with kNever_SkCopyPixelsMode).

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2223653002

Review-Url: https://codereview.chromium.org/2223653002

Fix alpha bug in SkColorXform

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221933002

Review-Url: https://codereview.chromium.org/2221933002

Remove SK_SUPPORT_LEGACY_DATA_FACTORIES.

The code protected by this flag is no longer used. Remove the flag and
code. This also removes SK_SUPPORT_LEGACY_TYPEFACE_PTR from Android, as
it is no longer needed.

TBR=reed
Only removes already guarded API.

Review-Url: https://codereview.chromium.org/2223933002

std::move(SkTDArray)

Since we don't support MSVC2013 anymore, we can be more
succinct when defining move constructors of compound types.

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2227673002

Review-Url: https://codereview.chromium.org/2227673002

Avoid caching resources for volatile bitmap shaders

SkBitmapProvider::isVolatile() treats all SkImages as non-volatile,
which is not what we want for temp SkImage wrappers of volatile bitmaps.

R=reed@google.com
BUG=chromium:633941
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2222783002

Review-Url: https://codereview.chromium.org/2222783002

Cleanup use of legacy SkSurface creation methods a bit

This:
   removes the SK_SUPPORT_LEGACY_NEW_SURFACE_API flag from the bzl build
   cleans up some remaining uses of the old API

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2223023002

Review-Url: https://codereview.chromium.org/2223023002

update picture recording benchmarks to allow comparison with SkLiteRecorder

Here's a demo.  The new code is still looking 2-3x faster.

~/skia (bench) $ r nanobench --match nytimes --config nonrendering --ms 2000
curr/maxrss loops min median mean max stddev samples config bench
  19/26  MB 2 146µs 147µs 151µs 422µs 9% 6615 nonrendering desk_nytimes.skp
  20/26  MB 4 46.6µs 46.9µs 48.2µs 204µs 10% 10370 nonrendering keymobi_nytimes_com_.skp

~/skia (bench) $ r nanobench --match nytimes --config nonrendering --ms 2000  --lite
curr/maxrss loops min median mean max stddev samples config bench
  19/26  MB 2 73.8µs 76.9µs 78.7µs 417µs 14% 12702 nonrendering desk_nytimes.skp
  20/26  MB 5 18.5µs 18.7µs 19.3µs 137µs 12% 20713 nonrendering keymobi_nytimes_com_.skp

Here's a quick performance diff, where <1x means --lite is faster:

    top25desk_wikipedia__1_tab_.skp  285us ->  364us 1.27x
      top25desk_games_yahoo_com.skp  302us ->  329us 1.09x
                   tabl_mozilla.skp  241us ->  260us 1.08x
                desk_chalkboard.skp  321us ->  313us 0.98x
               tabl_gamedeksiam.skp  383us ->  367us 0.96x
            top25desk_pinterest.skp  375us ->  281us 0.75x
keymobi_reddit_com_r_programmin.skp  258us ->  142us 0.55x
                   desk_nytimes.skp  149us -> 77.9us 0.52x
      keymobi_worldjournal_com_.skp  201us ->  104us 0.52x
              top25desk_blogger.skp  112us ->   55us 0.49x
    top25desk_sports_yahoo_com_.skp  186us -> 89.6us 0.48x
         desk_googlespreadsheet.skp  206us -> 97.5us 0.47x
top25desk_google_com_search_q_c.skp  192us -> 89.8us 0.47x
      keymobi_wikipedia__1_tab_.skp  170us -> 79.3us 0.47x
keymobi_wikipedia__1_tab____del.skp  170us -> 78.2us 0.46x
              desk_unicodetable.skp 6.25ms -> 2.87ms 0.46x
                    desk_carsvg.skp  138us -> 63.3us 0.46x
    top25desk_answers_yahoo_com.skp  133us -> 60.7us 0.46x
                 top25desk_espn.skp  108us -> 49.2us 0.45x
top25desk_plus_google_com_11003.skp  361us ->  162us 0.45x
                      desk_espn.skp 99.4us -> 44.5us 0.45x
              tabl_worldjournal.skp  103us -> 45.6us 0.44x
             desk_ugamsolutions.skp 56.2us -> 24.8us 0.44x
             top25desk_facebook.skp 82.7us -> 35.7us 0.43x
       keymobi_cuteoverload_com.skp  213us -> 91.9us 0.43x
             top25desk_linkedin.skp 61.3us -> 26.3us 0.43x
       top25desk_news_yahoo_com.skp  153us -> 65.6us 0.43x
               desk_gmailthread.skp 64.9us -> 27.8us 0.43x
keymobi_androidpolice_com_2012_.skp  167us -> 71.3us 0.43x
           top25desk_amazon_com.skp 77.5us -> 33.1us 0.43x
                   desk_wowwiki.skp  129us -> 54.1us 0.42x
          top25desk_weather_com.skp  113us -> 47.1us 0.42x
keymobi_facebook_com_barackobam.skp 95.2us -> 39.6us 0.42x
keymobi_shop_mobileweb_ebay_com.skp 31.5us -> 13.1us 0.42x
keymobi_amazon_com_gp_aw_s_ref_.skp 46.1us -> 18.9us 0.41x
keymobi_mobile_news_sandbox_goo.skp 90.7us ->   37us 0.41x
top25desk_google_com__hl_en_q_b.skp 52.4us -> 21.4us 0.41x
keymobi_answers_yahoo_com_quest.skp 96.5us -> 39.3us 0.41x
                    tabl_pravda.skp  126us -> 51.2us 0.41x
           keymobi_nytimes_com_.skp 46.9us ->   19us 0.4x
keymobi_ftw_usatoday_com_2014_0.skp  119us -> 48.2us 0.4x
          top25desk_youtube_com.skp  162us -> 65.3us 0.4x
         keymobi_news_yahoo_com.skp 58.1us -> 23.2us 0.4x
         keymobi_boingboing_net.skp 58.8us -> 23.4us 0.4x
         keymobi_techcrunch_com.skp 26.3us -> 10.4us 0.39x
keymobi_plus_google_com_app_bas.skp 26.9us -> 10.4us 0.38x
keymobi_google_co_uk_search_hl_.skp 35.1us -> 13.4us 0.38x
              keymobi_pinterest.skp 26.2us ->   10us 0.38x
        keymobi_deviantart_com_.skp 67.1us -> 25.4us 0.38x
                     tabl_gmail.skp 10.3us -> 3.86us 0.38x
             top25desk_ebay_com.skp 65.6us -> 24.5us 0.37x
keymobi_m_youtube_com_watch_v_9.skp 57.9us -> 21.6us 0.37x
            top25desk_wordpress.skp  138us -> 51.3us 0.37x
                 keymobi_gsp_ro.skp   17us -> 6.34us 0.37x
       top25desk_techcrunch_com.skp 93.6us -> 34.7us 0.37x
keymobi_cnn_com_2012_10_03_poli.skp  232us -> 85.5us 0.37x
                keymobi_cnn_com.skp 30.5us -> 11.1us 0.37x
keymobi_baidu_com_s_wd_barack_o.skp 39.3us -> 14.3us 0.36x
keymobi_online_wsj_com_home_pag.skp 50.3us -> 18.3us 0.36x
               keymobi_digg_com.skp 54.8us -> 19.5us 0.36x
keymobi_wowwiki_com_world_of_wa.skp 39.4us ->   14us 0.36x
keymobi_theverge_com_2012_10_28.skp  102us -> 36.4us 0.36x
                      tabl_digg.skp  105us -> 37.4us 0.36x
 top25desk_google_com_calendar_.skp 67.2us -> 23.7us 0.35x
              keymobi_wordpress.skp 65.3us ->   23us 0.35x
             desk_css3gradients.skp 56.4us -> 19.8us 0.35x
top25desk_mail_google_com_mail_.skp  119us -> 41.6us 0.35x
                desk_googlehome.skp  8.2us -> 2.85us 0.35x
top25desk_docs___1_open_documen.skp 23.8us -> 8.22us 0.35x
               keymobi_mlb_com_.skp 18.6us ->  6.3us 0.34x
          keymobi_slashdot_org_.skp   33us ->   11us 0.33x
                 desk_tiger8svg.skp 96.2us ->   32us 0.33x
              top25desk_twitter.skp  124us -> 40.7us 0.33x
keymobi_bing_com_search_q_sloth.skp 17.3us -> 5.55us 0.32x
               keymobi_linkedin.skp 6.78us -> 1.99us 0.29x
          top25desk_booking_com.skp  291us -> 83.2us 0.29x
                keymobi_blogger.skp 19.3us -> 5.47us 0.28x
            keymobi_sfgate_com_.skp 83.3us ->   23us 0.28x
            desk_jsfiddlebigcar.skp 10.8us -> 2.95us 0.27x
           keymobi_theverge_com.skp   22us -> 5.27us 0.24x
                    desk_mapsvg.skp 1.15us ->  216ns 0.19x
keymobi_iphone_capitolvolkswage.skp  121us -> 22.3us 0.18x
                 desk_wikipedia.skp 1.36us ->  244ns 0.18x
               desk_pokemonwiki.skp 1.35us ->  243ns 0.18x
                  desk_samoasvg.skp 1.39us ->  241ns 0.17x
                  desk_tigersvg.skp 1.41us ->  241ns 0.17x
keymobi_booking_com_searchresul.skp  129us -> 19.7us 0.15x

Some spot testing makes it look like everything that's not a giant speedup can be made so by tweaking my (arbitrarily set) maximum size for the free list.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2220273002

Review-Url: https://codereview.chromium.org/2220273002

Speculative fix for SKP partner-dir upload

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2229433002

Review-Url: https://codereview.chromium.org/2229433002

Test sk_linear_to_srgb() less exhaustively in Debug mode.

It's pretty slow when not optimized... it can take ~5 minutes to run.
This makes it take ~2 seconds on my laptop.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2221803002

Review-Url: https://codereview.chromium.org/2221803002

whitespace change

TBR=

Review URL: https://codereview.chromium.org/2221893002 .

[recipes] Smush the rest of builder_spec into appropriate places, other cleanup

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2219873003

Review-Url: https://codereview.chromium.org/2219873003

finish SkLiteDL

    - makeThreadsafe()
    - fix up some alignment problems
    - drawVertices / drawPatch
    - work around drawPicture() + default paint bugs?
    - turn on tests.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2219343002

Review-Url: https://codereview.chromium.org/2219343002

SkLiteDL, very nearly the rest

    fix saveLayer bounds
    all the misc calls
    a bit of refactoring

just drawVertices and drawPatch to go, plus some todos and likely bug fixes

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2226563002

Review-Url: https://codereview.chromium.org/2226563002

Update SKP version

Automatic commit by the RecreateSKPs bot.

TBR=
NO_MERGE_BUILDS
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2220833002

Review-Url: https://codereview.chromium.org/2220833002

SkLiteDL: closing in

 - optimizeFor(GrContext*)
 - images
 - basic text

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2224753002

Review-Url: https://codereview.chromium.org/2224753002

flesh out more of SkLiteDL:

    save layer, clips, simple draws

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2226513002

Review-Url: https://codereview.chromium.org/2226513002

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

Merge GrClipMaskManager into GrClipStackClip

TBR=bsalomon@google.com
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2196393007

Review-Url: https://codereview.chromium.org/2196393007

use tmp allocator for images in temp shaders

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2222593003

Review-Url: https://codereview.chromium.org/2222593003

SkPDF: Stop using kHAdvance_PerGlyphInfo

SkPDFFont:
- never request kHAdvance_PerGlyphInfo from typeface.
- set_glyph_widths() fn uses a glyph cache to get advances.
- stop expecting vertical advances that are never requested.
- composeAdvanceData() now non-templated
- appendAdvance() one-line function removed

SkPDFDevice:
- use a glyph cache for getting repeated advances.
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2219733004

Review-Url: https://codereview.chromium.org/2219733004

update Image_MakeFromRasterBitmap to check uniqueIDs

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2218223002

Review-Url: https://codereview.chromium.org/2218223002

Enable SK_DEBUG/SK_RELEASE to be defined in the user config.
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2204683005

Review-Url: https://codereview.chromium.org/2204683005

rename FoceCopyMode to SkCopyPixelMode

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2220603002

Review-Url: https://codereview.chromium.org/2220603002

SkBlendARGB32 and S32[A]_Blend_BlitRow32 are currently formulated as: SkAlphaMulQ(src, src_scale) + SkAlphaMulQ(dst, dst_scale), which boils down to ((src*src_scale)>>8) + ((dst*dst_scale)>>8). In particular, note that the intermediate precision is discarded before the two parts are added together, causing the final result to possibly inaccurate.

In Firefox, we use SkCanvas::saveLayer in combination with a backdrop that initializes the layer to the background. When this is blended back onto background using transparency, where the source and destination pixel colors are the same, the resulting color after the blend is not preserved due to the lost precision mentioned above. In cases where this operation is repeatedly performed, this causes substantially noticeable differences in color as evidenced in this downstream Firefox bug report:  https://bugzilla.mozilla.org/show_bug.cgi?id=1200684

In the test-case in the downstream report, essentially it does blend(src=0xFF2E3338, dst=0xFF2E3338, scale=217), which gives the result 0xFF2E3237, while we would expect to get back 0xFF2E3338.

This problem goes away if the blend is instead reformulated to effectively do (src*src_scale + dst*dst_scale)>>8, which keeps the intermediate precision during the addition before shifting it off.

This modifies the blending operations thusly. The performance should remain mostly unchanged, or possibly improve slightly, so there should be no real downside to doing this, with the benefit of making the results more accurate. Without this, it is currently unsafe for Firefox to blend a layer back onto itself that was initialized with a copy of its background.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2097883002
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

[mtklein adds...]
No public API changes.
TBR=reed@google.com

Review-Url: https://codereview.chromium.org/2097883002

use SkMakeBitmapShader to not force copies when possible

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2216403002

Review-Url: https://codereview.chromium.org/2216403002

Rename GrDrawingManager::drawContext to makeDrawContext

This was violating our naming convention

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2216643002

Review-Url: https://codereview.chromium.org/2216643002

Roll GN

No major new features... just good to keep up.

https://chromium.googlesource.com/chromium/buildtools.git/+log/60f7f9a..9c6ad6f

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2218063002

Review-Url: https://codereview.chromium.org/2218063002

Making a sample for shadow maps for more intensive development

Merge branch 'shadow-gm' into shadow-sample

Added variable size shadow maps. Also fixed some bugs

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2198933002

Review-Url: https://codereview.chromium.org/2198933002

Minor clean up related to blur mask filters

This is the non-substantive portion of: https://codereview.chromium.org/2201133002/ (Implement GPU occluded blur mask filter)

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2214163003

Review-Url: https://codereview.chromium.org/2214163003

Move to SkDataTable::MakeXXX and sk_sp.

Change SkDataTable::NewXXX to SkDataTable::MakeXXX and return sk_sp.
This updates users of SkDataTable to sk_sp as well.
There do not appear to be any external users of these methods.

Review-Url: https://codereview.chromium.org/2211143002

Fix vulkan resolve

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2215353003

Review-Url: https://codereview.chromium.org/2215353003

[recipes] Increase test coverage

Remove a bunch of "pragma: no cover" by removing unused code or adding test bots to cover.

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2215443003

Review-Url: https://codereview.chromium.org/2215443003

[recipes] Separate builder_spec into compile/test/perf recipes

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2205473002

Review-Url: https://codereview.chromium.org/2205473002

[recipes] Remove build environment vars from default_env

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2209423002

Review-Url: https://codereview.chromium.org/2209423002

check zero-length text draws

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2215113002

TBR=

Review-Url: https://codereview.chromium.org/2215113002

Remove color-dodge/burn from NVIDIA blacklist on 355.00+

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2212593002

Review-Url: https://codereview.chromium.org/2212593002

Implement Vulkan Resolve.

First step to getting msaa running on vulkan

BUG=skia:5127
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2210383002

Review-Url: https://codereview.chromium.org/2210383002

Add new GM for occluded blurmask filter draws

This splits out just the GM portion of https://codereview.chromium.org/2201993003/ (Add GM to test out blurmaskfilter occluders) which included some API changes

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2214033002

Review-Url: https://codereview.chromium.org/2214033002

Get Mac viewer working with SDL

Also fixes SkiaSDLExample.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2210603003

Review-Url: https://codereview.chromium.org/2210603003

This CL's base is the CL that sets up the distance vector field req. exposure: https://codereview.chromium.org/2114993002/

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2151653002

Review-Url: https://codereview.chromium.org/2151653002

Create a separate src for colorspace tests

This will allow me to run these tests in sRGB mode, while
leaving most of the image tests disabled.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2206953006

Review-Url: https://codereview.chromium.org/2206953006

[recipes] Fixes for Coverage bot

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2217523002

Review-Url: https://codereview.chromium.org/2217523002

[recipes] Stop pretending that buildbot_spec is a step, simplify

Since we no longer have to call out to obtain the buildbot_spec, merge
the two functions in vars API.  As a side effect, this applies
default_env to the sync steps as well, which shouldn't have an
appreciable effect on bot behavior.

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2209343002

Review-Url: https://codereview.chromium.org/2209343002

Move builder_spec, [dm|nanobench]_flags, builder_name_schema to recipes

- builder_name_schema becomes its own recipe module.
- builder_spec, dm, and nanobench flags move into vars module.
- recipe expectation diffs include:
  - no more buildbot_spec.py step
  - "real" dm and nanobench flags, instead of --dummy-flags
  - some inconsequential stuff in visualbench, which is removed anyway.

BUG=skia:5578
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2215803002

Review-Url: https://codereview.chromium.org/2215803002

SkRTConf: eliminate

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2212473002
DOCS_PREVIEW= https://skia.org/?cl=2212473002
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

[mtklein]
TBR=reed@google.com
Only removing unused public API.

Review-Url: https://codereview.chromium.org/2212473002

Update skimage VERSION: attempt 3

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2211973002

Review-Url: https://codereview.chromium.org/2211973002

GN: make SkPMColor BGRA on Linux.  All other platforms are auto-detected.

This technically shouldn't cause us to draw differently, but sometimes
does on buggy GMs, and also has us go down slightly different code paths.
It's good to be consistent with GYP, Chromium, Google3, etc.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2214023002

Review-Url: https://codereview.chromium.org/2214023002

Slim the CQ down to the GN bots, where they exist.

If no one objects to my "GN bot plan" email, this unblocks us from removing
a bunch of non-GN bots.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2207073002

CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Debug-GN-Trybot,Test-Ubuntu-GCC-ShuttleA-GPU-GTX660-x86_64-Release-GN-Trybot

Review-Url: https://codereview.chromium.org/2207073002

Add new bench for occluded blurmaskfilter draws

w/ occluders
  44/44  MB     6       497us   500us   500us   502us   0%      .oOOooooOO      gpu     bluroccludedrrect

w/o occluders
  41/41  MB     5       1.08ms  1.09ms  1.12ms  1.47ms  11%     .........O      gpu     bluroccludedrrect

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2203153002

Review-Url: https://codereview.chromium.org/2203153002

Remove SkSurface::MakeRenderTargetDirect

split into:
https://codereview.chromium.org/2182543003/ (Move prepareForExternalIO from GrRenderTarget to GrDrawContext)
https://codereview.chromium.org/2187573002/ (Reduce usage of MakeRenderTargetDirect)
https://codereview.chromium.org/2186073002/ (Rename GrContext's newDrawContext & drawContext to makeDrawContext)
https://codereview.chromium.org/2178353005/ (Remove use of MakeRenderTargetDirect from view system)
https://codereview.chromium.org/2198433003/ (Remove some ancillary users of SkSurface::MakeRenderTargetDirect)
https://codereview.chromium.org/2208483004/ (Remove GrRenderTarget member variable from SkGpuDevice)
https://codereview.chromium.org/2211473002/ (Move GrContext::makeDrawContext to new GrContextPriv object)

TBR=bsalomon@google.com

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2176333002

Review-Url: https://codereview.chromium.org/2176333002

Show SVG file name in SampleApp title

R=robertphillips@google.com,stephana@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2217483002

Review-Url: https://codereview.chromium.org/2217483002

Revert of Update skimage VERSION: attempt 2 (patchset #1 id:1 of https://codereview.chromium.org/2213973002/ )

Reason for revert:
Think this is going to break Windows again.

Original issue's description:
> Update skimage VERSION: attempt 2
>
> BUG=skia:
> GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2213973002
>
> Committed: https://skia.googlesource.com/skia/+/d6dec3f4a1365d34e147234acf3c4a3629146457

TBR=borenet@google.com
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:

Review-Url: https://codereview.chromium.org/2216523003

Update skimage VERSION: attempt 2

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2213973002

Review-Url: https://codereview.chromium.org/2213973002

Speculative fix for internal builds

Guard the DMSrcSink SkSVGDOM.h include.

TBR=dogben@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2210583003

Review-Url: https://codereview.chromium.org/2210583003

Remove GrRenderTarget member variable from SkGpuDevice

Split out of: https://codereview.chromium.org/2176333002/ (Remove SkSurface::MakeRenderTargetDirect)

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2208483004

Review-Url: https://codereview.chromium.org/2208483004

spin off easy stuff from Herb's windows GN CL

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2209533004

No public API changes.
TBR=reed@google.com

Review-Url: https://codereview.chromium.org/2209533004

Revert of Update to new skimage VERSION (patchset #1 id:1 of https://codereview.chromium.org/2211633002/ )

Reason for revert:
The CIPD package is broken

Original issue's description:
> Update to new skimage VERSION
> TBR=borenet@google.com
> BUG=skia:
> GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2211633002
>
> Committed: https://skia.googlesource.com/skia/+/52d1be5ca7b1ba8cc450be7dd6377ea5bb73386a

TBR=msarett@google.com
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:

Review-Url: https://codereview.chromium.org/2216613002

Add an SVG DM source

R=mtklein@google.com,robertphillips@google.com,stephana@google.com
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2209593004

Review-Url: https://codereview.chromium.org/2209593004

Revert of SkRTConf: reduce functionality to what we use, increase simplicity (patchset #8 id:150001 of https://codereview.chromium.org/2212473002/ )

Reason for revert:
missed GrVkPipelineStateCache

Original issue's description:
> SkRTConf: reduce functionality to what we use, increase simplicity
>
> GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2212473002
> DOCS_PREVIEW= https://skia.org/?cl=2212473002
> CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
>
> [mtklein]
> TBR=reed@google.com
> Only removing unused public API.
>
> Committed: https://skia.googlesource.com/skia/+/ef59974708dade6fa72fb0218d4f8a9590175c47

TBR=halcanary@google.com
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true

Review-Url: https://codereview.chromium.org/2215433003

Update to new skimage VERSION
TBR=borenet@google.com
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2211633002

Review-Url: https://codereview.chromium.org/2211633002

Fix compiler complaints about uninit vars

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2206143003

Review-Url: https://codereview.chromium.org/2206143003

SkRTConf: reduce functionality to what we use, increase simplicity

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2212473002
DOCS_PREVIEW= https://skia.org/?cl=2212473002
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

[mtklein]
TBR=reed@google.com
Only removing unused public API.

Review-Url: https://codereview.chromium.org/2212473002

GN: build sfntly, icu, harfbuzz

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2200833010

Review-Url: https://codereview.chromium.org/2200833010

Move GrContext::makeDrawContext to new GrContextPriv object

We don't want external clients to be able to call the wrapping makeDrawContext. The "creating" makeDrawContext needs to be public for external image filters.

The 3 specialized drawContext creators on GrContextPriv (to wrap backend objects) are also to be kept from public use and will be used to remove SkSurface::MakeRenderTargetDirect.

Split out of: https://codereview.chromium.org/2176333002/ (Remove SkSurface::MakeRenderTargetDirect)

TBR=bsalomon@google.com

GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2211473002

Review-Url: https://codereview.chromium.org/2211473002

Require gn format in presubmit

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2203283003

Review-Url: https://codereview.chromium.org/2203283003

SkPDF: fewer GYP_DEFINES

also: hack a unit test

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2208743003

Review-Url: https://codereview.chromium.org/2208743003

Convert SkAutoTUnref<SkData> to sk_sp<SkData>.

With the move from SkData::NewXXX to SkData::MakeXXX most
SkAutoTUnref<SkData> were changed to sk_sp<SkData>. However,
there are still a few SkAutoTUnref<SkData> around, so clean
them up.

Review-Url: https://codereview.chromium.org/2212493002

Perform color correction on png decodes

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2184543003
CQ_INCLUDE_TRYBOTS=master.client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

Review-Url: https://codereview.chromium.org/2184543003

added initial GLSL support to skslc

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search?issue=2185393003

Review-Url: https://codereview.chromium.org/2185393003