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
This enables removing the more complicated atomic shims from SkAtomics.h.
TBR=reed
This doesn't actually change any API.
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-TSAN-Trybot,Test-Ubuntu-GCC-Golo-GPU-GT610-x86_64-Release-TSAN-Trybot
Review URL: https://codereview.chromium.org/1867863002
The C++ standard library uses the name "release" for the operation we call "detach".
Rewriting each "detach(" to "release(" brings us a step closer to using standard library types directly (e.g. std::unique_ptr instead of SkAutoTDelete).
This was a fairly blind transformation. There may have been unintentional conversions in here, but it's probably for the best to have everything uniformly say "release".
BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1809733002
Review URL: https://codereview.chromium.org/1809733002
The 'element_type' typedef is to play nice with std::pointer_traits.
The full complement of operators and swap to match unique_ptr so that
sk_sp can be properly compared to nullptr and used with standard
containers.
Update to 'reset' so that calling 'unref' is the last operation.
This also adds tests for these changes, and sets the fPtr to nullptr
in debug for easier bug finding.
Review URL: https://codereview.chromium.org/1773453002
Previously, sk_sp::reset(T* t) did not release its own reference
if its internal pointer was the same as 't'. This leaks a reference.
Now always release the current reference when non-nullptr.
Review URL: https://codereview.chromium.org/1767983002
To further consolidate the various unique owning classes, this bases
SkAutoTUnref on skstd::unique_ptr. Users are updated because of two
breaking changes, swap now takes a reference and reset no longer
returns its argument.
Review URL: https://codereview.chromium.org/1370803002
SkTemplates.h contains a number of Skia specific utilities which are
not designed for external use. In addition to reducing the external
support burden, this will allow Skia to freely refactor this file.
Review URL: https://codereview.chromium.org/1272293004
- It's no longer needed to help the (2011?) transition to SkAutoTUnref.
- It prevents us from making classes that go in SkAutoTUnrefs final,
i.e. all ref-counted classes.
This had better not have been public API...
TBR=reed@google.com
BUG=skia:
Review URL: https://codereview.chromium.org/1068443002
This adds sk_memory_barrier(), implemented using sk_atomic_fetch_add() on an uninitialized variable. If that becomes a problem we can drop this to the porting layer, using std::atomic_thread_fence() / __atomic_thread_fence() / __sync_synchronize().
The big win is that ref() doesn't generate a memory barrier any more on ARM.
This is an instance of SkSafeRef() in SkPaint(const SkPaint&) after this CL:
4d0: 684a ldr r2, [r1, #4]
4d2: 6018 str r0, [r3, #0]
4d4: b13a cbz r2, 4e6 <_ZN7SkPaintC1ERKS_+0x2e>
4d6: 1d10 adds r0, r2, #4
4d8: e850 4f00 ldrex r4, [r0]
4dc: 3401 adds r4, #1
4de: e840 4500 strex r5, r4, [r0]
4e2: 2d00 cmp r5, #0
4e4: d1f8 bne.n 4d8 <_ZN7SkPaintC1ERKS_+0x20>
Here's the before, pretty much the same with two memory barriers surrounding the ref():
4d8: 684a ldr r2, [r1, #4]
4da: 6018 str r0, [r3, #0]
4dc: b15a cbz r2, 4f6 <_ZN7SkPaintC1ERKS_+0x3e>
4de: 1d10 adds r0, r2, #4
4e0: f3bf 8f5f dmb sy
4e4: e850 4f00 ldrex r4, [r0]
4e8: 3401 adds r4, #1
4ea: e840 4500 strex r5, r4, [r0]
4ee: 2d00 cmp r5, #0
4f0: d1f8 bne.n 4e4 <_ZN7SkPaintC1ERKS_+0x2c>
4f2: f3bf 8f5f dmb sy
The miscellaneous files in here are just fixups to explicitly include SkMutex.h,
instead of leeching it off SkRefCnt.h.
No public API changes.
TBR=reed@google.com
Build trybots seem hosed.
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/896803002
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
