This requires we remove NVPR from the default set of configs, as we only find
out at runtime that it's not available. All the other defaults will either be
compiled in and supported, or not compiled in and non-fatally skipped as
unknown configs.
BUG=skia:
Review URL: https://codereview.chromium.org/1100773003
Reason for revert:
Xfermode_SrcOver not looking encouraging. Up to 50% regressions.
https://perf.skia.org/#3242
Original issue's description:
> Convert Color32 code to perfect blend.
>
> Before we commit to blend_256_round_alt, let's make sure blend_perfect is
> really slower in practice (i.e. regresses on perf.skia.org).
>
> blend_perfect is really the most desirable algorithm if we can afford it. Not
> only is it correct, but it's easy to think about and break into correct pieces:
> for instance, its div255() doesn't require any coordination with the multiply.
>
> This looks like a 30% hit according to microbenches. That said, microbenches
> said my previous change would be a 20-25% perf improvement, but it didn't end
> up showing a significant effect at a high level.
>
> As for correctness, I see a bunch of off-by-1 compared to blend_256_round_alt
> (exactly what we'd expect), and one off-by-3 in a GM that looks like it has a
> bunch of overdraw.
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/61221e7f87a99765b0e034020e06bb018e2a08c2TBR=reed@google.com,fmalita@chromium.org,mtklein@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/1083923006
More nullptr checks for factories I have added.
Other checks more Yoda-like I have made. (Skia style this is.)
BUG=skia:
Review URL: https://codereview.chromium.org/1086393004
This makes all non-XMLCALL element_handler functions uniform by making
their signature
void XXX_element_handler(FamilyData* self, const char** attributes)
This makes the code somewhat easier to follow, as it makes explicit
which functions are actually logically methods of FamilyData (start
with 'FamilyData* self') and which are called directly by expat
(start with 'void* data').
Review URL: https://codereview.chromium.org/1086943007
Before we were returning non transformed bounds for the ovals instead of
device bounds from the various draw calls. This was causing us to do dst
copies from the wrong portion of the screen when there was some transform
on the matrix.
This fixes the gpu issues from the new imagefilters-xfermodes gm as well an
old bug from the srcmode GM.
BUG=skia:
Review URL: https://codereview.chromium.org/1090453003
The Android font parser prints a number of helpful messages when
something seems amiss in the configuration file. Give all of these
messages a common and more informative format.
Review URL: https://codereview.chromium.org/1096063003
Before we commit to blend_256_round_alt, let's make sure blend_perfect is
really slower in practice (i.e. regresses on perf.skia.org).
blend_perfect is really the most desirable algorithm if we can afford it. Not
only is it correct, but it's easy to think about and break into correct pieces:
for instance, its div255() doesn't require any coordination with the multiply.
This looks like a 30% hit according to microbenches. That said, microbenches
said my previous change would be a 20-25% perf improvement, but it didn't end
up showing a significant effect at a high level.
As for correctness, I see a bunch of off-by-1 compared to blend_256_round_alt
(exactly what we'd expect), and one off-by-3 in a GM that looks like it has a
bunch of overdraw.
BUG=skia:
Review URL: https://codereview.chromium.org/1098913002
Extended tests (150M+) run to completion in release in about 6 minutes; the standard test suite exceeds 100K and finishes in a few seconds on desktops.
TBR=reed
BUG=skia:3588
Review URL: https://codereview.chromium.org/1037953004
The conic, quadratic and cubic computations were using tolerance^2
instead of tolerance when computing maximum point count, causing paths
to be undertessellated when magnifying and overtessellated when minifying.
(Funny story: this bug went unnoticed back when we were tessellating
paths in screen space, when tolerance and its square were both 1.)
BUG=skia:3731
Review URL: https://codereview.chromium.org/1095773003
Previously the normal/italic style bit was obtained from scanning the
font file. With the new format the style may be stated explicitly, and
this explicit value in the configuration file should override any
information obtained from the font data itself.
This change allows the font element's style attribute to override the
font's style, but retains the default 'auto' setting for backwards
compatibility. Repecting the style bit may become more important with
variation fonts, because it will be up to the configuration writer to
determine what values of the 'slnt' variation should be considered
'normal' or 'italic'.
DOCS_PREVIEW= https://skia.org/?cl=1092093002
Committed: https://skia.googlesource.com/skia/+/673e902c9b9982a167f54f1cc175d8d9cab8bcaf
Review URL: https://codereview.chromium.org/1092093002
Reason for revert:
Failed on the compile bots.
Original issue's description:
> Respect declared font style on Android.
>
> Previously the normal/italic style bit was obtained from scanning the
> font file. With the new format the style may be stated explicitly, and
> this explicit value in the configuration file should override any
> information obtained from the font data itself.
>
> This change allows the font element's style attribute to override the
> font's style, but retains the default 'auto' setting for backwards
> compatibility. Repecting the style bit may become more important with
> variation fonts, because it will be up to the configuration writer to
> determine what values of the 'slnt' variation should be considered
> 'normal' or 'italic'.
>
> DOCS_PREVIEW= https://skia.org/?cl=1092093002
>
> Committed: https://skia.googlesource.com/skia/+/673e902c9b9982a167f54f1cc175d8d9cab8bcafTBR=mtklein@google.com,tomhudson@google.com,scroggo@google.com,bungeman@google.com
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
Review URL: https://codereview.chromium.org/1082173004
We only embed images with YUV planes. That should only grab the
subset of color JPEGs supported by PDF.
BUG=skia:3180
Review URL: https://codereview.chromium.org/1025773002
Previously the normal/italic style bit was obtained from scanning the
font file. With the new format the style may be stated explicitly, and
this explicit value in the configuration file should override any
information obtained from the font data itself.
This change allows the font element's style attribute to override the
font's style, but retains the default 'auto' setting for backwards
compatibility. Repecting the style bit may become more important with
variation fonts, because it will be up to the configuration writer to
determine what values of the 'slnt' variation should be considered
'normal' or 'italic'.
DOCS_PREVIEW= https://skia.org/?cl=1092093002
Review URL: https://codereview.chromium.org/1092093002
This algorithm changes the blend math, guarded by SK_LEGACY_COLOR32_MATH. The new math is more correct: it's never off by more than 1, and correct in all the interesting 0x00 and 0xFF edge cases, where the old math was never off by more than 2, and not always correct on the edges.
If you look at tests/BlendTest.cpp, the old code was using the `blend_256_plus1_trunc` algorithm, while the new code uses `blend_256_round_alt`. Neither uses `blend_perfect`, which is about ~35% slower than `blend_256_round_alt`.
This will require an unfathomable number of rebaselines, first to Skia, then to Blink when I remove the guard.
I plan to follow up with some integer SIMD abstractions that can unify these two implementations into a single algorithm. This was originally what I was working on here, but the correctness gains seem to be quite compelling. The only places these two algorithms really differ greatly now is the kernel function, and even there they can really both be expressed abstractly as:
- multiply 8-bits and 8-bits producing 16-bits
- add 16-bits to 16-bits, returning the top 8 bits.
All the constants are the same, except SSE is a little faster to keep 8 16-bit inverse alphas, NEON's a little faster to keep 8 8-bit inverse alphas. I may need to take this small speed win back to unify the two.
We should expect a ~25% speedup on Intel (mostly from unrolling to 8 pixels) and a ~20% speedup on ARM (mostly from using vaddhn to add `color`, round, and narrow back down to 8-bit all into one instruction.
(I am probably missing several more related bugs here.)
BUG=skia:3738,skia:420,chromium:111470
Review URL: https://codereview.chromium.org/1092433002
