If font data is used to create an SkTypeface and the font data describes
a variable font and no variation parameters are specified, the
SkTypeface created should have the default values for all axes. This is
particularaly interesting with DirectWrite since it makes this not
straight forward.
Bug: skia:10929
Cq-Include-Trybots: luci.skia.skia.primary:Test-Win10-Clang-NUC5i7RYH-CPU-AVX2-x86_64-Debug-All-NativeFonts
Change-Id: I4620deebf52142bbdffa1a283343b503cd1e6981
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/338604
Commit-Queue: Ben Wagner <bungeman@google.com>
Reviewed-by: Herb Derby <herb@google.com>
pack(x,y,bits) as an alias for x|(y<<bits) only existed originally to
implement it with the SLI arm64 instruction, but I've since realized
that was misguided.
I had thought the assumption on pack ("(x & (y << bits)) == 0"), i.e.
"no overlap between x and the shifted y", was enough to make using SLI
legal, but it's actually not strong enough a requirement.
The SLI docs say "...inserts the result into the corresponding vector
element in the destination SIMD&FP register such that the new zero bits
created by the shift are not inserted but retain their existing value."
The key thing not mentioned there happens with zero bits _not_ created
by the shift, the ones already present at the top of y. They're of
course inserted, overwriting any previous values.
This means SLI (and so pack()) become strictly order dependent in a way
I had never intended. This will work as you'd think,
skvm::I32 px = splat(0);
px = pack(px, r, 0);
px = pack(px, a, 24);
but this version swapping the two calls to pack() will overwrite alpha,
skvm::I32 px = splat(0);
px = pack(px, a, 24);
px = pack(px, r, 0);
I find that error-prone, so I've removed Op::pack and replaced it
with a simple expansion to x|(y<<bits). That of course works in either
order.
This new test can't JIT at head, but if we implement the other missing
instructions (soon, dependent CL) it would start failing when JIT'd.
The interpreter and x86 were both fine, since they're both doing what's
now the only approach to pack(), the simple x|(y<<bits).
I've left assembler support for SLI in case we want to try it again.
Change-Id: Iaf879309d3e1d0a458a688f3a62556e55ab05e23
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/337197
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Bug: skia:10914
SkAnimCodecPlayer:
- Properly handle orientation, whether the image is still or not
- Mark const methods as const
- Fix seek() so that if you seek to the duration of frame 0, it will
show frame 1
- Fix the SkImageInfo so if the first frame is opaque, but following
frames are not, those frames can still be decoded
resources:
- Rename "webp-animated.webp" to "stoplight.webp", which better
describes the animation
- Update test files accordingly
- Add "stoplight_h.webp", which is the same animation with an EXIF
that converts it to a horizontal stoplight
AnimCodecPlayer test:
- Test the new image files
- Verify SkAnimCodecPlayer::dimensions behaves as expected
- Remove extra debugging line
- Provide better error messages
AnimCodecPlayerExifGM:
- Add a new GM that shows all frames of the new animation with an EXIF
orientation
- Add a new GM that shows all frames of an animation with an opaque
first frame followed by frames with alpha
Change-Id: I43cf91c16d52aa1901eef8e13e1e644eea6058b3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/332753
Reviewed-by: Derek Sollenberger <djsollen@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
Orientation information is sometimes stored in
the SubIFD section of EXIF, so read that. This is
just a matter of searching for the SubIFD offset
value in the EXIF tags and then parsing the
values from there onwards. The data format is
the same as the EXIF data.
The images are not under any copyright as I made
them up locally specifically for these tests.
Bug: skia:10799
Change-Id: I5384ffc1c4a9a0c7d3fc8510ef4da2f278cb8b97
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/323217
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
This takes the "one pass" code path more often, using less memory, as it
does not have to allocate an intermediate width*height pixel buffer.
Wuffs v0.2 did not support SRC_OVER, only SRC, but Wuffs v0.3 does.
The gif-transparent-index.gif test file comes from the
test/data/artificial directory of the github.com/google/wuffs
repository. It was programmatically generated.
The new GifTest.cpp test passes with skia_use_wuffs true or false, with
or without the SkWuffsCodec.cpp change.
Change-Id: I46fb4c849319fbefc39f331416a8b7d3836093ce
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/320116
Commit-Queue: Leon Scroggins <scroggo@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
Auto-resizing and vertical alignment require a non-empty text box. But
currently, the presence of the text box is used to discriminate between
point text [1] and paragraph text [2].
In order to support auto-scaling and v-alignment for point text, we must
decouple the text mode encoding from the text box:
* introduce and explicit LinebreakPolicy property for skottie::Shaper,
and use it to control line breaking instead of the text box presence
* by default, the line breaking policy is initialized per existing
AE/BM semantics: non-empty text box -> paragraph mode,
empty box -> point mode
* the policy can be overridden via the PropertyObserver APIs to enable
point mode auto-resizing and vertical alignment
[1] https://helpx.adobe.com/after-effects/using/creating-editing-text-layers.html#enter_point_text
[2] https://helpx.adobe.com/after-effects/using/creating-editing-text-layers.html#enter_paragraph_text
Change-Id: I007144283a31a2faa579d7eec82af72af3d540cb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/321788
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Plumb layer size information and add machinery for retrieving layer
content as an SkPicture.
Implement the effect in SkSL.
Current limitations:
* displacement source layer must be above (on top in stacking order)
of the target layer
* if animated, the displacement layer timeline (in/out points) must
fully cover the target layer timeline
* Hue/Sat/Lightness selectors are not supported at the moment
These will be addressed in follow-up CLs.
Note: Bodymovin does not export hidden layers by default; if the
displacement source layer is hidden, one should select the "Hidden"
export option.
Change-Id: I11a5c760a9df1e75835a51371f60d5b798e7e38a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/314798
Commit-Queue: Florin Malita <fmalita@google.com>
Reviewed-by: Brian Osman <brianosman@google.com>
Refactor the shape repeater using a custom render node (instead of
duplicating per-instance SG nodes).
In the process, fix several issues:
* scale was not being composed correctly
* start/end opacity were being ignored
* non-atomic fragments were being drawn in wrong stacking order
Change-Id: I06cd3606806d1a46852a8557b27c09eb44abdadd
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/313209
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Tunnels through SkImageGenerator as well.
The new SkCodec interface doesn't assume three 8 bit planes.
New SkYUVASpec more clearly defines chroma subsampling and siting of
the planes.
The intent is to use this for other YUVA APIs as well, in particular
SkImage factories in the future.
In this change we convert to the SkYUVASpec to SkYUVASizeInfo
and SkYUVAIndex[4] representation. But the intent is to use
the SkYUVASpec representation throughout the pipeline once
legacy APIs are removed.
orientation GM is replicated to test a variety of chroma
subsampling configs.
Bug: skia:10632
Change-Id: I3fad35752b87cac16c51b24824331f2ae7d458d3
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/309658
Commit-Queue: Brian Salomon <bsalomon@google.com>
Reviewed-by: Robert Phillips <robertphillips@google.com>
Reviewed-by: Leon Scroggins <scroggo@google.com>
This reverts commit 07438b0cda.
Bug: skia:10369
Bug: skia:10371
This will allow Skia clients developing for Android 11+ to rely on
Android's NDK APIs for decoding, which will allow them to decode
without including their own decoding libraries (e.g. libjpeg-turbo).
Using these APIs also provides support for static HEIF images.
Run ImageGenSrc in kPlatform_Mode on Android to verify decoding
visually.
Add tests and a grayscale png.
Update some test bots running Android R to specify ndk_api so they will
run the new code.
Change-Id: I4ca07d832dbd6a9d8cff0faea975fd70da00718f
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308185
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
This reverts commit cfef980939.
Reason for revert: Breaking Google3 roll
Original change's description:
> Add an SkImageGenerator that uses NDK APIs
>
> Bug: skia:10369
> Bug: skia:10371
>
> This will allow Skia clients developing for Android 11+ to rely on
> Android's NDK APIs for decoding, which will allow them to decode
> without including their own decoding libraries (e.g. libjpeg-turbo).
> Using these APIs also provides support for static HEIF images.
>
> Run ImageGenSrc in kPlatform_Mode on Android to verify decoding
> visually.
>
> Add tests and a grayscale png.
>
> Update some test bots running Android R to specify ndk_api so they will
> run the new code.
>
> Change-Id: Ica782339b2414d472ede0b61729a127ce41892a5
> Reviewed-on: https://skia-review.googlesource.com/c/skia/+/305689
> Reviewed-by: Mike Klein <mtklein@google.com>
> Commit-Queue: Leon Scroggins <scroggo@google.com>
TBR=djsollen@google.com,mtklein@google.com,scroggo@google.com,brianosman@google.com,reed@google.com
Change-Id: Ifed506a76a0ff5903d101c1bf7330d319b8376a6
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: skia:10369
Bug: skia:10371
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/308180
Reviewed-by: Leon Scroggins <scroggo@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
Bug: skia:10369
Bug: skia:10371
This will allow Skia clients developing for Android 11+ to rely on
Android's NDK APIs for decoding, which will allow them to decode
without including their own decoding libraries (e.g. libjpeg-turbo).
Using these APIs also provides support for static HEIF images.
Run ImageGenSrc in kPlatform_Mode on Android to verify decoding
visually.
Add tests and a grayscale png.
Update some test bots running Android R to specify ndk_api so they will
run the new code.
Change-Id: Ica782339b2414d472ede0b61729a127ce41892a5
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/305689
Reviewed-by: Mike Klein <mtklein@google.com>
Commit-Queue: Leon Scroggins <scroggo@google.com>
BM can now export inline fonts, encoded as data-uris.
Extend our DataURIResourceProviderProxy helper to also intercept and
decode inline typefaces.
Change-Id: Iaf1be9db2fd32383af78bc351b1228fe6b3b64bf
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/305685
Commit-Queue: Florin Malita <fmalita@chromium.org>
Commit-Queue: Ben Wagner <bungeman@google.com>
Auto-Submit: Florin Malita <fmalita@chromium.org>
Reviewed-by: Ben Wagner <bungeman@google.com>
Spread/choke operates as a compression of the alpha channel towards the
low bits.
Change-Id: I82aec1321b60f7f75a79e8280e761d4629f6c923
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/305183
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Assorted bugfixes for the non-analytical mask code path.
1) SkSG modulatePaint() should only override the blend mode when
one is specified (!= kSrcOver).
2) Some modes (notably intersect) require touching pixels outside the
mask draw geometry. These modes must be applied as a layer.
Introduce an explicit layer node in SkSG, and inject for masks which
require it.
Also refactor Subtract to use more natural blend and pathops modes,
instead of always inverting geometry.
TBR=
Change-Id: I412168d1ff61eb8e59907babe8f0e091f6fffacf
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/303997
Reviewed-by: Florin Malita <fmalita@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Sksg::Merge needs to preserve the fill type of the first path appended
in the stack.
Theoretically, one could append multiple paths with different fill types
using sksg::Merge, but in practice Skottie should never do that (append
mode with invertible shape only used for the very first mask in a stack).
TBR=
Change-Id: Ie9ac9187cc1c8baaae2bef439313a7700407f04a
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/303582
Reviewed-by: Florin Malita <fmalita@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Observed semantics:
-- operates on cubic Bezier path representation
-- moves vertices towards the shape center, and control points in the
opposite direction, based on the specified amount
-- the center is determined as the vertex average
-- the amount is specified as a fraction of the transition to center
(0 -> noop, 1 -> fully collapesed to center)
-- negative and extranormal amount values are allowed
(invert direction/extrapolate)
TBR=
Change-Id: I7da81a5fe5cffd0e50bd94e6b448565b0b04ed86
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/301582
Reviewed-by: Florin Malita <fmalita@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
AE allows selecting the paint order when both fill & stroke are present.
The CL also fixes some text stroke issues: stroke width not parsed
correctly and not actually used on the paint.
Change-Id: Iec27bb65d09f689365e43b801d3844106780572b
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/301857
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Commit-Queue: Florin Malita <fmalita@chromium.org>
Two issues:
1) For static keyframes (start_value == end_value) AE yields horizontal
orientation (0 tangent). We technically have the same logic in
Skottie, but our value deduplication logic interferes: the two
consecutive equal values are consolidated, and the result ends up
holding the spatial lerp info for the next frame => our hold frames
auto-orient for the beginning of the next keyframe.
Fix: skip value deduplication when spatial lerp is present.
2) The very last keyframe is always static and holds no spatial info.
AE retains the orientation of the previous frame, but Skottie yields
0 tangent.
Fix: the easiest way to accomplish AE semantics is to detect when
we're dealing with the last keyframe, and swap with the previous
keyframe with an adjust weight of 1 (to select the end value). This
produces the same lerp result (because keyframed values are always
contiguous) and also respects the orientation of the prev frame.
TBR=
Change-Id: Id661f7804533e95b747722457489a7ef759572a4
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/301176
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Florin Malita <fmalita@google.com>
Commit-Queue: Florin Malita <fmalita@chromium.org>
Unlike other shape path effects, merge paths disables the rendering of
any preceding paints - it only extracts the merged geometry from the
stack.
Update the shape layer attacher logic to suppress paints under merge
paths.
TBR=
Change-Id: I414134839de9eaa4b0f828d8dc6d4721620242bb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/300897
Commit-Queue: Florin Malita <fmalita@google.com>
Reviewed-by: Florin Malita <fmalita@chromium.org>
Based on SkPathOps for now.
Change-Id: Id27c8a235cbd4ab5083735b67cf5d2635ee16cfc
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/300497
Commit-Queue: Florin Malita <fmalita@google.com>
Reviewed-by: Mike Reed <reed@google.com>
AE allows animating the line spacing text property [1].
Observed semantics:
- spacing is applied as an offset to all fragments in a line
- for selector/partial coverage, the spacing for a given line
is the average of the computed spacing for each fragment
- spacing is cumulative (applies to all lines following)
Plumb the new animator prop ("ls") and expand the existing line
tracking logic to also apply computed spacing offsets.
(also requires a Bodymovin update to export the line spacing property)
[1] https://helpx.adobe.com/after-effects/using/animating-text.html#text_animator_properties
Change-Id: I5517acea8dbc1b2fbae09cb0874f1e53cd2acb90
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/300377
Commit-Queue: Florin Malita <fmalita@google.com>
Reviewed-by: Ben Wagner <bungeman@google.com>
Bug: skia:10274
Change-Id: Ifb2ef8bf031e74d9d5c8183efe5aff4e6f3d2e7c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/292562
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Brian Osman <brianosman@google.com>
AE layers can be tagged for auto-orient - i.e. they observe an
additional rotation component dependent on the position property
animation (position derivative/tangent).
Augment Vec2KeyframeAnimator to optionally track orientation, for both
temporal/linear and spatial (motion-path) keyframes. Update
TransformAdapter2D to use this orientation when attached to layer
transforms.
Change-Id: I616e45a07b088e9e566b4f88450e95f9315b727c
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/291716
Commit-Queue: Florin Malita <fmalita@chromium.org>
Reviewed-by: Mike Reed <reed@google.com>
This font comes up fairly often due to being included as a primary font
in css testing. There are several Paragraph samples which already assume
it is present at this location. It is fairly small and very liberally
licensed.
This font file is ahem.ttf from
https://www.w3.org/Style/CSS/Test/Fonts/Ahem/ .
Change-Id: I6fdcb036cecca66e725ecb30e0a07fef1dfb1cdf
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/289445
Reviewed-by: Julia Lavrova <jlavrova@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
Add support for external precomp Skottie layers. This allows embedders
to seamlessly mix custom/Lottie content.
General flow:
* embedders register a PrecompInterceptor callback with
the animation builder
* at build time, Skottie invokes the callback for each pre-composed
layer
- the returned ExternalLayer implementation is used instead of the
Lottie layer payload
- (a nullptr value signals Skottie to use the usual Lottie payload)
* at render time, ExternalLayer::render() is called to defer content
rendering to the embedder
Also implement a sample PrecompInterceptor which attempts to substitute
precmp layers matching a given pattern with external Lottie animations:
precomp_name: "__foo.json" -> Animation("foo.json")
This new mechanism is a generalization of (and supersedes) the old
NestedAnimation hack - so we can remove that.
Change-Id: Id80fe11881c62b8717c2476117c7c03ad5300eef
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/288130
Commit-Queue: Florin Malita <fmalita@chromium.org>
Reviewed-by: Mike Reed <reed@google.com>
Implement non-linear contrast using a cubic polynomial approximation,
as a SkRuntimeEffect.
The effect range is significantly more constrained than the legacy
version: https://www.desmos.com/calculator/ehem0vy3ft
Change-Id: I86bdbb9cc0d30065780f87705d2d4d39385609cb
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/285840
Commit-Queue: Florin Malita <fmalita@chromium.org>
Reviewed-by: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Reed <reed@google.com>
Implement drop and inner shadow styles using explicit image filters.
Remove existing style support from DropShadowEffect.cpp, as it now
has a new cozy place with its inner sibling.
Supported properties:
- color
- opacity
- angle
- distance
- size (sigma)
Change-Id: I5b7e3c75678e036a20c1908b84c74a670a5aa196
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283918
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Commit-Queue: Florin Malita <fmalita@chromium.org>
Trace out the tree from the DAG. Trace nodes
with fan-out > 1 after all out edges have been traced.
Change-Id: Ic078d212adf95a19146fcbd9fb8d103ea23360ee
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283557
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
These are neat but mostly just a distraction for now.
I've left all the assembly in place and unit tested
to make putting these back easy when we want to.
Change-Id: Id2bd05eca363baf9c4e31125ee79e722ded54cb7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/283307
Commit-Queue: Mike Klein <mtklein@google.com>
Reviewed-by: Herb Derby <herb@google.com>
Change-Id: I2cb6255a553852a292427d6dc9ef8c5ed7f8286d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/252926
Commit-Queue: Brian Osman <brianosman@google.com>
Reviewed-by: Michael Ludwig <michaelludwig@google.com>
Reviewed-by: Mike Reed <reed@google.com>
In inverted mode (Mode::kInverted), the trim result represents the
logical segment [stop..start] (wrapping around at the path's end).
We currently emit two segments [0..start] and [stop..1], in that
exact order. This behavior breaks continuity for single closed
contour paths.
Update SkTrimPath to
1) emit the segments in the correct order ([stop..1],[0..start])
2) skip the connecting moveTo for closed paths
Bug: skia:10107
Change-Id: Icd280554ba7291c985f504793feff104df2a4a99
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/281882
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Florin Malita <fmalita@chromium.org>
Note: works for well-formed poly-to-poly (perspective) transforms, but
doesn't support AE's degenerate corners semantics (concave/inverted
polys) at this point.
Bug: skia:10100
Change-Id: I5b3492b008302495b616867c139c6e5ad6dc57df
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/281595
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Florin Malita <fmalita@chromium.org>
bit_clear is at least useful as a special case for select(),
which helps with code readability.
Add is_NaN() and use these all together in sweep gradient.
Change-Id: I57a54f8956f85e0db0662b33f8446b8dc7342d8d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/281685
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
- new this-> convention: never use it when calling common public
Builder methods like splat(), bit_and(), etc like you'd see in
normal user code, but always use it when calling private methods
like this->push(), this->isImm(), this->allImm().
- use c++17 if-statements to scope this->allImm() variables tighter.
- check for x.id == y.id cases where applicable, including a tweak
to min() and max() to make them able to hit the special case.
- add special cases for I32 +,-,*, and remove an old unimportant
unit test that assumed we didn't fold these.
- add special cases for select(), and use select() in a few more
places where it's clearer and now just as efficient.
Change-Id: Idaac9250ac5a95a48d33eeba1cc4380c8c91629d
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/281678
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
bit_clear() is just another bit_and(),
and bytes() is a way of expression pshufb
that we never really use (yet).
Can always add them back later, but there's
some extra complexity to think about for each
that I'd like to not think about now:
- common sub-expression elimination between bit_and and bit_clear
- large constant management JIT'ing bytes
Change-Id: I3a54afa963231fec1d5de949acc647e3430ed0d8
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/281557
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>
Plumb layer style parsing, and extend existing DropShadowAdapter to
support both drop shadow style and drop shadow effect.
Change-Id: Id99a419dacd06dc38dc4cf84ff4ecb92218c45f7
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/279020
Reviewed-by: Mike Reed <reed@google.com>
Commit-Queue: Florin Malita <fmalita@chromium.org>
When converting from Instructions to OptimizedInstructions
place instructions that reduce register pressure earlier in
the instruction list.
This change reduces some register pressure in SkVM, and
improves the bitmap_RGBA_8888_A_scale_bilerp benchmark by
about 5%.
Change-Id: If5f6385bd2f7720701d1c827265062b35491a790
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/276485
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Mike Klein <mtklein@google.com>
While I think trunc(mad(x, scale, 0.5)) is fine for doing our float
to fixed point conversions, round(mul(x, scale)) was kind of better
all around:
- better rounding than +0.5 and trunc
- faster when mad() is not an fma
- often now no need to use the constant 0.5f or have it in a register
- allows the mul() in to_unorm to use mul_f32_imm
Those last two points are key... this actually frees up 2 registers in
the x86 JIT when using to_unorm().
So I think maybe we can resurrect round and still guarantee our desired
intra-machine stability by committing to using instructions that follow
the current rounding mode, which is what [v]cvtps2dq inextricably uses.
Left some notes on the ARM impl... we're rounding to nearest even there,
which is probably the current mode anyway, but to be more correct we
need a slightly longer impl that rounds float->float then "truncates".
Unsure whether it matters in practice. Same deal in the unit test that
I added back, now testing negative and 0.5 cases too. The expectations
assume the current mode is nearest even.
I had the idea to resurrect this when I was looking at adding _imm Ops
for fma_f32. I noticed that the y and z arguments to an fma_f32 were by
far most likely to be constants, and when they are, they're by far likely
to both be constants, e.g. 255.0f & 0.5f from to_unorm(8,...).
llvm disassembly for SkVM_round unit test looks good:
~ $ llc -mcpu=haswell /tmp/skvm-jit-1231521224.bc -o -
.section __TEXT,__text,regular,pure_instructions
.macosx_version_min 10, 15
.globl "_skvm-jit-1231521224" ## -- Begin function skvm-jit-1231521224
.p2align 4, 0x90
"_skvm-jit-1231521224": ## @skvm-jit-1231521224
.cfi_startproc
cmpl $8, %edi
jl LBB0_3
.p2align 4, 0x90
LBB0_2: ## %loopK
## =>This Inner Loop Header: Depth=1
vcvtps2dq (%rsi), %ymm0
vmovupd %ymm0, (%rdx)
addl $-8, %edi
addq $32, %rsi
addq $32, %rdx
cmpl $8, %edi
jge LBB0_2
LBB0_3: ## %hoist1
xorl %eax, %eax
testl %edi, %edi
jle LBB0_6
.p2align 4, 0x90
LBB0_5: ## %loop1
## =>This Inner Loop Header: Depth=1
vcvtss2si (%rsi,%rax), %ecx
movl %ecx, (%rdx,%rax)
decl %edi
addq $4, %rax
testl %edi, %edi
jg LBB0_5
LBB0_6: ## %leave
vzeroupper
retq
.cfi_endproc
## -- End function
Change-Id: Ib59eb3fd8a6805397850d93226c6c6d37cc3ab84
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/276738
Auto-Submit: Mike Klein <mtklein@google.com>
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Herb Derby <herb@google.com>
- hook up fmls.4s as fnma_f32
- add fneg.4s
- use fneg.4s + fmls.4s to impl fms_f32
- more tests to exercise these
Change-Id: I60173a5e4618ab968a9361e15334a1d63c001372
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/275412
Commit-Queue: Herb Derby <herb@google.com>
Reviewed-by: Herb Derby <herb@google.com>
