patch from issue 1072303005 at patchset 40001 (http://crrev.com/1072303005#ps40001)
This looks quite launchable. radial_gradient3, min of 100 samples:
N5: 985µs -> 946µs
MBP: 395µs -> 279µs
On my MBP, most of the meat looks like it's now in reading the cache and writing to dst one color at a time. Is that something we could do in float math rather than with a lookup table?
BUG=skia:
CQ_EXTRA_TRYBOTS=client.skia.compile:Build-Mac10.8-Clang-Arm7-Debug-Android-Trybot,Build-Ubuntu-GCC-Arm7-Release-Android_NoNeon-Trybot
Committed: https://skia.googlesource.com/skia/+/abf6c5cf95e921fae59efb487480e5b5081cf0ec
Review URL: https://codereview.chromium.org/1109643002
Reason for revert:
compile failures.
Original issue's description:
> Mike's radial gradient CL with better float -> int.
>
> patch from issue 1072303005 at patchset 40001 (http://crrev.com/1072303005#ps40001)
>
> This looks quite launchable. radial_gradient3, min of 100 samples:
> N5: 985µs -> 946µs
> MBP: 395µs -> 279µs
>
> On my MBP, most of the meat looks like it's now in reading the cache and writing to dst one color at a time. Is that something we could do in float math rather than with a lookup table?
>
> BUG=skia:
>
> CQ_EXTRA_TRYBOTS=client.skia.android:Test-Android-GCC-Nexus5-CPU-NEON-Arm7-Debug-Trybot,Test-Android-GCC-Nexus9-CPU-Denver-Arm64-Debug-Trybot
>
> Committed: https://skia.googlesource.com/skia/+/abf6c5cf95e921fae59efb487480e5b5081cf0ecTBR=reed@google.com,robertphillips@google.com,mtklein@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/1109883003
patch from issue 1072303005 at patchset 40001 (http://crrev.com/1072303005#ps40001)
This looks quite launchable. radial_gradient3, min of 100 samples:
N5: 985µs -> 946µs
MBP: 395µs -> 279µs
On my MBP, most of the meat looks like it's now in reading the cache and writing to dst one color at a time. Is that something we could do in float math rather than with a lookup table?
BUG=skia:
CQ_EXTRA_TRYBOTS=client.skia.android:Test-Android-GCC-Nexus5-CPU-NEON-Arm7-Debug-Trybot,Test-Android-GCC-Nexus9-CPU-Denver-Arm64-Debug-Trybot
Review URL: https://codereview.chromium.org/1109643002
Adds an onGetBlendInfo method for GrXferProcessor subclasses to
override instead of overriding getBlendInfo directly. This gives the
base class a chance to initialize the struct with default values
before passing it on. As the BlendInfo struct grows, this will keep
things simple and less error prone.
BUG=skia:
NOTREECHECKS=true
NOTRY=true
Review URL: https://codereview.chromium.org/1049143002
This CL also adds a new parameter to SkBitmapSource which gives the user control of the filter quality.
BUG=472795
Review URL: https://codereview.chromium.org/1072603002
approx-match.
This revealed some further 0..1 texture coordinate
depedencies,specifically in SkDisplacementMapEffect and
SkXfermodeImageFilter, fixed through use of
GrTextureDomain.
Note: causes minor pixel diffs in pictureimagefilter,
testimagefilters GMs.
Also causes minor pixel diffs in some Blink layout tests,
suppressed here: https://codereview.chromium.org/1064943002
BUG=skia:3532
Review URL: https://codereview.chromium.org/1065683002
Currently, the GPU-side image filter implementation creates
exact-match textures for the offscreen backing stores for
saveLayer(). This is because several filters have GPU
implementations which depend on the texture coordinates
being 0..1.
The fix is three-fold:
1) Store the actual requested size in the SkGpuDevice, so
that when wrapping it in an SkBitmap for passing to
filterImage(), we can give it the original size.
2) Fix the filters (SkMagnifierImageFilter,
SkLightingImageFilter, SkMatrixConvolutionImageFilter,
SkMatrixImageFilter) whose GPU implementation depends on
0..1 texture coordinates.
3) Remove the exception for GPU-side image filters in
SkCanvas::internalSaveLayer().
For the lighting filters, there were two bugs which were
cancelling each other out: the sobel filter matrix was
being computed upside down, but then we'd negate the
resulting normal. This worked fine in the exact-match case,
but in the approx-match case we'd sample garbage along
the edge pixels. Also, we never implemented the edge pixels
according to spec in the GPU case. It requires a
different fragment shader for each edge of the nine-patch,
which meant we couldn't use asFragmentProcessor(), and had
to implement the drawing via a filterImageGPU() override.
In order to avoid polluting the public API, I inserted a
new base class, SkLightingImageFilterInternal above
Sk[Diffuse|Specular]LightingImageFilter to handle the
implementation.
For the SkMatrixConvolutionImageFilter, it seems the
GLSL clamp() function occasionally returns values outside
the clamped range, resulting in access of garbage
texels even in GL_NEAREST. The fix here is to clamp to a
rect inset by half a texel. There was also a bug in
the unpremultiply step when fConvolveAlpha is false.
For SkMatrixImageFilter, the fix was to make the generic
draw path be more careful about when to use texture domain.
If the bitmap already has a texture, use texture domain
if the srcRect is smaller than the entire texture (not
the entire bitmap).
N.B.: this change will cause some minor pixel diffs in the
GPU results of the following GMs (and possibly more):
matriximagefilter, matrixconvolution, imagefiltersscaled,
lighting, imagemagnifier, filterfastbounds,
complexclip_aa_Layer_invert, complexclip_aa_layer,
complexclip_bw_layer_invert, complexclip_bw_layer.
BUG=skia:3532
Committed: https://skia.googlesource.com/skia/+/b97dafefe63ea0a1bbce8e8b209f4920983fb8b9
Committed: https://skia.googlesource.com/skia/+/f5f8518fe0bbd2703e4ffc1b11ad7b4312ff7641
Committed: https://skia.googlesource.com/skia/+/46112cf2a7c7307f1c9eebb5f881cbda15aa460c
Review URL: https://codereview.chromium.org/1034733002
Reason for revert:
Spoke to Stephen about this. Reverting because failing debug builds:
https://uberchromegw.corp.google.com/i/client.skia/builders/Test-Mac10.9-Clang-MacMini6.2-GPU-HD4000-x86_64-Debug/builds/51https://uberchromegw.corp.google.com/i/client.skia/builders/Test-Ubuntu-GCC-ShuttleA-GPU-GTX660-x86_64-Debug/builds/54
Original issue's description:
> Implement approx-match support in image filter saveLayer() offscreen.
>
> Currently, the GPU-side image filter implementation creates
> exact-match textures for the offscreen backing stores for
> saveLayer(). This is because several filters have GPU
> implementations which depend on the texture coordinates
> being 0..1.
>
> The fix is three-fold:
>
> 1) Store the actual requested size in the SkGpuDevice, so
> that when wrapping it in an SkBitmap for passing to
> filterImage(), we can give it the original size.
> 2) Fix the filters (SkMagnifierImageFilter,
> SkLightingImageFilter, SkMatrixConvolutionImageFilter,
> SkMatrixImageFilter) whose GPU implementation depends on
> 0..1 texture coordinates.
> 3) Remove the exception for GPU-side image filters in
> SkCanvas::internalSaveLayer().
>
> For the lighting filters, there were two bugs which were
> cancelling each other out: the sobel filter matrix was
> being computed upside down, but then we'd negate the
> resulting normal. This worked fine in the exact-match case,
> but in the approx-match case we'd sample garbage along
> the edge pixels. Also, we never implemented the edge pixels
> according to spec in the GPU case. It requires a
> different fragment shader for each edge of the nine-patch,
> which meant we couldn't use asFragmentProcessor(), and had
> to implement the drawing via a filterImageGPU() override.
> In order to avoid polluting the public API, I inserted a
> new base class, SkLightingImageFilterInternal above
> Sk[Diffuse|Specular]LightingImageFilter to handle the
> implementation.
>
> For the SkMatrixConvolutionImageFilter, it seems the
> GLSL clamp() function occasionally returns values outside
> the clamped range, resulting in access of garbage
> texels even in GL_NEAREST. The fix here is to clamp to a
> rect inset by half a texel. There was also a bug in
> the unpremultiply step when fConvolveAlpha is false.
>
> For SkMatrixImageFilter, the fix was to make the generic
> draw path be more careful about when to use texture domain.
> If the bitmap already has a texture, use texture domain
> if the srcRect is smaller than the entire texture (not
> the entire bitmap).
>
> N.B.: this change will cause some minor pixel diffs in the
> GPU results of the following GMs (and possibly more):
> matriximagefilter, matrixconvolution, imagefiltersscaled,
> lighting, imagemagnifier, filterfastbounds,
> complexclip_aa_Layer_invert, complexclip_aa_layer,
> complexclip_bw_layer_invert, complexclip_bw_layer.
>
> BUG=skia:3532
>
> Committed: https://skia.googlesource.com/skia/+/b97dafefe63ea0a1bbce8e8b209f4920983fb8b9
>
> Committed: https://skia.googlesource.com/skia/+/f5f8518fe0bbd2703e4ffc1b11ad7b4312ff7641
>
> Committed: https://skia.googlesource.com/skia/+/46112cf2a7c7307f1c9eebb5f881cbda15aa460cTBR=bsalomon@google.com,reed@chromium.org,senorblanco@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:3532
Review URL: https://codereview.chromium.org/1057693002
Currently, the GPU-side image filter implementation creates
exact-match textures for the offscreen backing stores for
saveLayer(). This is because several filters have GPU
implementations which depend on the texture coordinates
being 0..1.
The fix is three-fold:
1) Store the actual requested size in the SkGpuDevice, so
that when wrapping it in an SkBitmap for passing to
filterImage(), we can give it the original size.
2) Fix the filters (SkMagnifierImageFilter,
SkLightingImageFilter, SkMatrixConvolutionImageFilter,
SkMatrixImageFilter) whose GPU implementation depends on
0..1 texture coordinates.
3) Remove the exception for GPU-side image filters in
SkCanvas::internalSaveLayer().
For the lighting filters, there were two bugs which were
cancelling each other out: the sobel filter matrix was
being computed upside down, but then we'd negate the
resulting normal. This worked fine in the exact-match case,
but in the approx-match case we'd sample garbage along
the edge pixels. Also, we never implemented the edge pixels
according to spec in the GPU case. It requires a
different fragment shader for each edge of the nine-patch,
which meant we couldn't use asFragmentProcessor(), and had
to implement the drawing via a filterImageGPU() override.
In order to avoid polluting the public API, I inserted a
new base class, SkLightingImageFilterInternal above
Sk[Diffuse|Specular]LightingImageFilter to handle the
implementation.
For the SkMatrixConvolutionImageFilter, it seems the
GLSL clamp() function occasionally returns values outside
the clamped range, resulting in access of garbage
texels even in GL_NEAREST. The fix here is to clamp to a
rect inset by half a texel. There was also a bug in
the unpremultiply step when fConvolveAlpha is false.
For SkMatrixImageFilter, the fix was to make the generic
draw path be more careful about when to use texture domain.
If the bitmap already has a texture, use texture domain
if the srcRect is smaller than the entire texture (not
the entire bitmap).
N.B.: this change will cause some minor pixel diffs in the
GPU results of the following GMs (and possibly more):
matriximagefilter, matrixconvolution, imagefiltersscaled,
lighting, imagemagnifier, filterfastbounds,
complexclip_aa_Layer_invert, complexclip_aa_layer,
complexclip_bw_layer_invert, complexclip_bw_layer.
BUG=skia:3532
Committed: https://skia.googlesource.com/skia/+/b97dafefe63ea0a1bbce8e8b209f4920983fb8b9
Committed: https://skia.googlesource.com/skia/+/f5f8518fe0bbd2703e4ffc1b11ad7b4312ff7641
Review URL: https://codereview.chromium.org/1034733002
Reason for revert:
Looks like this change is causing layout test failures which is blocking Skia's DEPS roll into Chromium:
https://codereview.chromium.org/1050563002/https://codereview.chromium.org/1043133005/https://codereview.chromium.org/1048273002/
Reverting to see if this fixes the DEPS roll.
Original issue's description:
> Implement approx-match support in image filter saveLayer() offscreen.
>
> Currently, the GPU-side image filter implementation creates
> exact-match textures for the offscreen backing stores for
> saveLayer(). This is because several filters have GPU
> implementations which depend on the texture coordinates
> being 0..1.
>
> The fix is three-fold:
>
> 1) Store the actual requested size in the SkGpuDevice, so
> that when wrapping it in an SkBitmap for passing to
> filterImage(), we can give it the original size.
> 2) Fix the filters (SkMagnifierImageFilter,
> SkLightingImageFilter) whose GPU implementation depends on
> 0..1 texture coordinates.
> 3) Remove the exception for GPU-side image filters in
> SkCanvas::internalSaveLayer().
>
> For the lighting filters, there were two bugs which were
> cancelling each other out: the sobel filter matrix was
> being computed upside down, but then we'd negate the
> resulting normal. This worked fine in the exact-match case,
> but in the approx-match case we'd sample garbage along
> the edge pixels. Also, we never implemented the edge pixels
> according to spec in the GPU case. It requires a
> different fragment shader for each edge of the nine-patch,
> which meant we couldn't use asFragmentProcessor(), and had
> to implement the drawing via a filterImageGPU() override.
> In order to avoid polluting the public API, I inserted a
> new base class, SkLightingImageFilterInternal above
> Sk[Diffuse|Specular]LightingImageFilter to handle the
> implementation.
>
> N.B.: this change will cause some minor pixel diffs in the
> GPU results of the following GMs (and possibly more):
> matriximagefilter, matrixconvolution, imagefiltersscaled,
> lighting, imagemagnifier, filterfastbounds,
> complexclip_aa_Layer_invert, complexclip_aa_layer,
> complexclip_bw_layer_invert, complexclip_bw_layer.
>
> BUG=skia:3532
>
> Committed: https://skia.googlesource.com/skia/+/b97dafefe63ea0a1bbce8e8b209f4920983fb8b9
>
> Committed: https://skia.googlesource.com/skia/+/f5f8518fe0bbd2703e4ffc1b11ad7b4312ff7641TBR=bsalomon@google.com,reed@chromium.org,senorblanco@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:3532
Review URL: https://codereview.chromium.org/1057443003
Currently, the GPU-side image filter implementation creates
exact-match textures for the offscreen backing stores for
saveLayer(). This is because several filters have GPU
implementations which depend on the texture coordinates
being 0..1.
The fix is three-fold:
1) Store the actual requested size in the SkGpuDevice, so
that when wrapping it in an SkBitmap for passing to
filterImage(), we can give it the original size.
2) Fix the filters (SkMagnifierImageFilter,
SkLightingImageFilter) whose GPU implementation depends on
0..1 texture coordinates.
3) Remove the exception for GPU-side image filters in
SkCanvas::internalSaveLayer().
For the lighting filters, there were two bugs which were
cancelling each other out: the sobel filter matrix was
being computed upside down, but then we'd negate the
resulting normal. This worked fine in the exact-match case,
but in the approx-match case we'd sample garbage along
the edge pixels. Also, we never implemented the edge pixels
according to spec in the GPU case. It requires a
different fragment shader for each edge of the nine-patch,
which meant we couldn't use asFragmentProcessor(), and had
to implement the drawing via a filterImageGPU() override.
In order to avoid polluting the public API, I inserted a
new base class, SkLightingImageFilterInternal above
Sk[Diffuse|Specular]LightingImageFilter to handle the
implementation.
N.B.: this change will cause some minor pixel diffs in the
GPU results of the following GMs (and possibly more):
matriximagefilter, matrixconvolution, imagefiltersscaled,
lighting, imagemagnifier, filterfastbounds,
complexclip_aa_Layer_invert, complexclip_aa_layer,
complexclip_bw_layer_invert, complexclip_bw_layer.
BUG=skia:3532
Committed: https://skia.googlesource.com/skia/+/b97dafefe63ea0a1bbce8e8b209f4920983fb8b9
Review URL: https://codereview.chromium.org/1034733002
The primary feature this delivers is SkNf and SkNd for arbitrary power-of-two N. Non-specialized types or types larger than 128 bits should now Just Work (and we can drop in a specialization to make them faster). Sk4s is now just a typedef for SkNf<4, SkScalar>; Sk4d is SkNf<4, double>, Sk2f SkNf<2, float>, etc.
This also makes implementing new specializations easier and more encapsulated. We're now using template specialization, which means the specialized versions don't have to leak out so much from SkNx_sse.h and SkNx_neon.h.
This design leaves us room to grow up, e.g to SkNf<8, SkScalar> == Sk8s, and to grown down too, to things like SkNi<8, uint16_t> == Sk8h.
To simplify things, I've stripped away most APIs (swizzles, casts, reinterpret_casts) that no one's using yet. I will happily add them back if they seem useful.
You shouldn't feel bad about using any of the typedef Sk4s, Sk4f, Sk4d, Sk2s, Sk2f, Sk2d, Sk4i, etc. Here's how you should feel:
- Sk4f, Sk4s, Sk2d: feel awesome
- Sk2f, Sk2s, Sk4d: feel pretty good
No public API changes.
TBR=reed@google.com
BUG=skia:3592
Review URL: https://codereview.chromium.org/1048593002
Currently, the GPU-side image filter implementation creates exact-match
textures for the offscreen backing stores for saveLayer(). This is
because several filters have GPU implementations which depend on the
texture coordinates being 0..1.
The fix is three-fold:
1) Store the actual requested size in the SkGpuDevice, so that when
wrapping it in an SkBitmap for passing to filterImage(), we can give
it the original size.
2) Fix the filters (SkMagnifierImageFilter, more TBD) whose GPU
implementation depends on 0..1 texture coordinates.
3) Remove the exception for GPU-side image filters in
SkCanvas::internalSaveLayer().
N.B.: this change will cause some minor pixel diffs in the
GPU results of the following GMs (and possibly more):
matriximagefilter, matrixconvolution, imagefiltersscaled,
lighting, imagemagnifier, filterfastbounds,
complexclip_aa_Layer_invert, complexclip_aa_layer,
complexclip_bw_layer_invert, complexclip_bw_layer.
BUG=skia:3532
Review URL: https://codereview.chromium.org/1034733002
(This is essentially a revert of https://codereview.chromium.org/503833002/.)
This was necessary back when SkPaint was flattened even for in-process use. Now that we only flatten SkPaint for cross-process use, there's no need to serialize UniqueIDs.
Note: SkDropShadowImageFilter is being constructed with a croprect and UniqueID (of 0) in Blink. I've made the uniqueID param default to 0 temporarily, until this rolls in and Blink can be changed. (Blink can't be changed first, since unlike the other filters, there's no constructor that takes a cropRect but not a uniqueID.)
BUG=skia:
Review URL: https://codereview.chromium.org/1019493002
Clarify asColorFilter ...
1. Rename to isColorFilterNode for DAG reduction
2. Add asAColorFilter for removing the imagefilter entirely (future use-case)
Need layouttest rebaseline suppression before this can land in chrome...
https://codereview.chromium.org/984023004/
BUG=skia:
Review URL: https://codereview.chromium.org/982933002
Reason for revert:
Need to suppress these for rebaselining, so reverting for now.
Regressions: Unexpected image-only failures (5)
css3/filters/effect-brightness-clamping-hw.html [ ImageOnlyFailure ]
css3/filters/effect-combined-hw.html [ ImageOnlyFailure ]
virtual/slimmingpaint/css3/filters/effect-brightness-clamping-hw.html [ ImageOnlyFailure ]
virtual/slimmingpaint/css3/filters/effect-combined-hw.html [ ImageOnlyFailure ]
Original issue's description:
> Use ComposeColorFilter in factory to collapse consecutive filters (when possible).
> Change asColorFilter to reflect its reliance on the new factory behavior.
>
> patch from issue 967143002 at patchset 80001 (http://crrev.com/967143002#ps80001)
>
> BUG=skia:
>
> Committed: https://skia.googlesource.com/skia/+/dac843bf046c2cd79fd955cb177aee241d7a4b0cTBR=senorblanco@chromium.org,robertphillips@google.com,bsalomon@chromium.org
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=skia:
Review URL: https://codereview.chromium.org/978923005
This makes SkComposeImageFilter::computeFastBounds compose its
filters' bounds (rather than falling back to
SkImageFilter::computeFastBounds, which takes the union of the bounds).
This also makes SkComposeImageFilter::onFilterImage correctly handle
an offset produced when applying the inner filter; such offsets were
previously ignored.
BUG=chromium:453924
Review URL: https://codereview.chromium.org/908273002
BUG=skia:1366
For the added bench, the collapsing makes the bench take:
- 70% of the time for CPU rendering of 3 consecutive matrix filters
- almost no change in the GPU rendering of the matrix filters
- 50% of the time for CPU and GPU rendering of 3 consecutive table filters
Review URL: https://codereview.chromium.org/776673002
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
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
