Added and unified usage of a helper to create primitive
sub content easily, also for visibility checking of
primitives. Some smaller cleanups, too.
Change-Id: I3c4cd3e98e62b8cf7984351dbb0b154470fd8e3c
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/145908
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
We urgently should get rid of XOR paint, modern
graphic systems allow no read access to the pixel
targets, but that's naturally a precondition for
XOR.
While we can do that for the office's
visualization, we can in principle *not* fully
avoid getting stuff that needs/defines XOR paint,
e.g. EMF/WMF imports, so we *have* to support
it (for now - sigh)...
This makes this renderer complete from the minimal
to-be-supported primitives, too.
Change-Id: Ie8fa98b777de764af0babe969296a671ca5cc7ce
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/145739
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
...which happened to go largely unnoticed (the would-be additional overload, as
seen when including drawinglayer/processor2d/SDPRProcessor2dTools.hxx first
thing in drawinglayer/source/processor2d/SDPRProcessor2dTools.cxx, was just
never defined nor used), but caused
> drawinglayer/source/processor2d/SDPRProcessor2dTools.cxx(140,6): error: externally available entity 'prepareBitmapForDirectRender' is not previously declared in an included file (if it is only used in this translation unit, make it static; otherwise, provide a declaration of it in an included file) [loplugin:external]
> bool prepareBitmapForDirectRender(
> ~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
with clang-cl
Change-Id: I9f338dd998b66fe3223af045206a423b016b93f3
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/145696
Tested-by: Jenkins
Reviewed-by: Stephan Bergmann <sbergman@redhat.com>
Added support to directly paint FillGraphicPrimitive2D, that
means object fill with multiple tiles/tiled fill. Note: This
may use not only bitmap data, but also metafile or svg.
The standard decompose creates one transformed bitmap/vector
data primitive per tile what is correct, but has some limits
(a) These may get many when tiles are small compared to
mask polyPolygon (performance/ressources)
(b) Correctness: when rendering in AA the common edges of
tiles will *not* sum up perfectly, but due to AA
multiplying when blending will leave ugly 'traces'
This direct rendering avoids both. It can use the D2D Brush
functionality to repeat. Additionally to the current
office it can also do that when content is rotated/sheared.
This may not be well known since for those fills the office
always keeps the fill unrotated (for historical reasons,
primitives can do that, but not vcl). To see it you may
convert a SdrObject to metafile and rotate that, so the
content gets rotated and shows that ugly gaps.
Also added a square-pixel step value for vector fills where
the default gets used starting from some size. The argument
is better quality for vector data fills while only a
limited number of tiles will be rendered.
Also added a buffered handling of the (old and ugly)
OffsetX/OffsetY to make that work, we will not get rid of
that soon.
Packed all of this into tooling (so started a tooling
collection for SDPRs target-system independent) to be able
to easily re-use all these preparations/tests in other
SDPR implementations in the future.
Change-Id: Iafacf4181e7c9e2d2b2e3b5cf1f7e4cdd0a43f0a
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/145466
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
All stuff used from MS with names starting with ID2D1* is COM
API stuff, so referenced and interface-based. I thought about
making this more safe since a while (shared/unique_ptr, ...)
but found no good way to do it.
Also did not want to use CComPtr from MS and expand the devenv
we need on win, so was short before doing an own small smartptr
class. Luckily I asked sberg and he pointed me to the already
existing sal::systools::COMReference which exactly does what
I need here - thanks!
Unluckily I now had to change a lot of code - sigh. I wish
I had known earlier :-/
This change contains the renderer completely adapted to using
that much safer and better ressource control.
Plus I added (even more) comments to try to make more clear
in many plasces what's going on, what is done and why.
Change-Id: Ia2aa3223d0e5f7ec6569cde176cec1fadcd921dc
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/145142
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
just some partial pieces
export TEST_SYSTEM_PRIMITIVE_RENDERER=1
and the simple drawing shapes work
Change-Id: I3e01501a1cb21ec86d6fe8f5637a23e7358ffc86
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/144632
Tested-by: Jenkins
Reviewed-by: Caolán McNamara <caolanm@redhat.com>
(SDPR == SystemDependentPrimitiveRenderer)
Make D2DPixelProcessor2D directly render the added simple
Primitives
LineRectanglePrimitive2D
FilledRectanglePrimitive2D
SingleLinePrimitive2D
to use less ressources & get faster since no conversion to
Polygons is done. These are Primitives that *may* be handled
directly, but do not have to - I checked that the decomposes
do work before adding direct handling.
Change-Id: Ifa97c595c262d84a9889f92571b2c9909bcef0e0
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/144626
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
As promised at LiBoCon'22 I have finished an example
implementation of a System-Dependent PrimitiveRenderer.
It renders on Direct2D, so Win only. I chose this with
no special thoughts, did not even know too much about
Direct2D, so had to read quite some docu. Still managed
to do this in a view days.
NOTE: To make it run you need to switch off skia on Win
probably due to some global extra stuff done there (?)
The Renderer is in a new file (ca. 1000 lines) in the
drawinglayer libary. It implements all of the needed
primitives, so capable of rendering EditViews (Draw/
Impress complete, Writer/Calc in parts, interactions,
overlay, selection(s), etc.). It is used completely
as replacement of the VclPixelProcessor2D which is
not used in the patched version *at all*.
Just one primitive is missing:
- InvertPrimitive2D: Not needed much (hopefully will
vanish), could also be added
It is not in production state, mostly intended as a
demonstration how Primitives are intended: Render
*without* using an OutputDevice at all, directly to
the traget GraphicsSystem. It can (and should) be
used as pattern for other implementations.
NOTE: Inside the implementation you can do whatever
you want, all is allowed. Only rule is to visualize
the geometry defined by the Primitives.
This version does not need any buffer allocations to
render and is thus pretty fast (uses ID2D1Layer). It
also has no buffering, but that could easily be
added. The needed classes/implementations to buffer
system-dependent representations of data exist,
so doable fast for Bitmaps, Polygons and PolyPolygons.
It runs stable and fast, but still has errors, so
it is sometimes missing to render some stuff. I
would have to debug/finetune it further, but I see
this all as doable. It's far enough to be seen as
POC and example.
I also added a not-urgently-need-to-be-implemented
primitive, the UnifiedTransparencePrimitive2D, to
demonstrate how to do that: Add a handler for it
and render the defined geometry if you can do that
more direct/better than the decomposition.
The rough to-do list would be:
- add missing Primitives
- InvertPrimitive2D
- add buffering of
- Bitmaps
- Polygons
- PolyPolygons
- add TextRendering
- using decompose to PolyPolygons currently, but
looks reasonably good with PolyPolygon AA
- general
- polishing
- bug fixing
- stabilization(s)
- speedups by adding PrimitiveHandlers
- solve problem with TransparencePrimitive2D
- had to use drawinglayer::createAlphaMask
there, see comments
Made handling of TransparencePrimitive2D work, some fixes
and changes, too.
Still not happy with processTransparencePrimitive2D, it
uses drawinglayer tooling (createAlphaMask) to do it's
job. It works, but there must be a better way. It's
okay for demonstration/POV though
Tried hard in processTransparencePrimitive2D to directly
render the alpha channel to a Direct2D
ID2D1BitmapRenderTarget using DXGI_FORMAT_A8_UNORM.
It does work theoretically as it should, but I found
no way to directly paingt to the AlphaChannel. It
seems as if Direct2D *is* hard-wired to always
alpha-blend, there seems to be no control or mode
to *select* the BlendMode to be used. Since only that
small step is missing I keep that code in (see
bTryDirectPreparationOfAlphaAsBitmap).
NOTE: No idea yet, but testing shows that the direct
but not yet complete way is ca. 7 times faster.
I have now added PointArrayPrimitive2D, so only that
InvertPrimitive2D is missing.
I also added MarkerArrayPrimitive2D to make it more
efficient (marker bitmap gets converted only once)
and better looking (not AAed anymore).
With both added, grid visualization looks good and is
fast.
Added support for BackgroundColorPrimitive2D
Added BitmapBuffering for Direct2D. This is a little
bit more complicated, theoretically all ID2D1Bitmaps
are specific to the ID2D1RenderTarget they get created
for, so theoretically they have to be (re)created
every time the ID2D1RenderTarget aka OutputDevice
changes - argh.
There is CreateSharedBitmap though that
allows re-usage of already created bitmaps for a new
RenderTaget, but this leads to needing a LO-livetime
RenderTarget to create all the Bitmaps for.
So I implemented that and it works and is fast.
Added buffering for GeometryData, ID2D1PathGeometry.
I prepared this already using the
CreateTransformedGeometry functionality of Direct2D,
so the B2DPolygon/B2DPolyPolygon can get buffered
in an unchanged form, re-used as source for
creating a ID2D1PathGeometry by calling
CreateTransformedGeometry. This could also be done by
manipulating the current transformation, but is not
necessary.
Simplified buffering by directly using derivated
helper classes (from basegfx::SystemDependentData)
as data holders, thus they can reliably call
Release() at incarnated Direct2D instances without
the need to keep record of this outside.
Completely removed OutputDevice from the
D2DPixelProcessor2D, it was anyways only used in
the constructor.
Removed the need to hand over the pixel size of the
pixel target when constructing a D2DPixelProcessor2D,
instead get that values from HDC directly. Also made
D2DPixelProcessor2D exported from drawinglayer for
further experiments.
Checked if I can use higher-level Direct2D versions
by using MS's QueryInterface to see if involved
classes are higher-level ones. This worked and
allows to paint the alpha channel for
TransparencePrimitive2D now indeed using the
BlendMode that is missing in the initial D2D stuff,
so activated that (much faster).
Aded the environment variable (for Windows only)
TEST_SYSTEM_PRIMITIVE_RENDERER to allow easy tesing
of SystemPrimitiveRenderer. Since Skia enabled
does collide in quite some places (see above) I
added hat this also disables Skia in it's central
isVCLSkiaEnabled() method. That way it is easy to
compare/quickly change between versions.
Removed maCurrentTransformation, this is just not
needed for a pixel-only renderer, see comment in
VclProcessor2D's header regarding that.
Added snap-to-pixel functionality.
Added direct support for PolygonStrokePrimitive2D,
so all lines which
- are fat or thin
- have LineCap
- have LineJoin
- use dash
can be directly rendered by Direct2D. This is especially
good for dash since no decompositions will be created
for these cases.
There is an exception for our own rarely used style
B2DLineJoin::NONE, that still needs to be decomposed,
but could also be added to the impl if needed.
There were also quite some 'conditions' for supporting
B2DLineJoin::Miter in Direct2D, but I managed to solve
them all, so Miter is fully done in Direct2D in this
example.
I also added example debug code to demonstrate how
it is possible to simply add decompose paint and the
system-specific implementation to optically compare
and play around in a running office to identify
possible problems/differences.
Improved handling of TransparencePrimitive2D massively
to a more future-proof and more common methodology by
using the Direct2D mechanism ID2D1Effect and the
CLSID_D2D1LuminanceToAlpha effect. For details, see
comment in comitted code at implCreateAlpha_Direct.
As explained in refs/changes/87/141087/22 I did add
now a tooling derivate of D2DPixelProcessor2D that can
be used to render sub-content without having to deal
with patching the current internal state of the running
instance, called D2DBitmapPixelProcessor2D. It resides
in the cxx file (so being safe from other usages), has
and creates an own ID2D1BitmapRenderTarget to work on
and will come in handy for other occasions that need
to prepare a in-between result.
Made stuff in the renderer itself more private and
added access methods to protected section, use
those in tooling.
Change-Id: Ie5803f9afa3f62dbaa8fd2f1bac8fd30a23066b5
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/141087
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
There was processorfromoutputdevice.cxx|hxx containing
createBaseProcessor2DFromOutputDevice that does exactly
the same thing as createProcessor2DFromOutputDevice in
processor2dtools.cxx|hxx, so I removed one onf them
to make handling of future changes easier.
Change-Id: Id8d49d1572a1bc39f1d330a742d4697c8c030e75
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/142320
Tested-by: Jenkins
Reviewed-by: Armin Le Grand <Armin.Le.Grand@me.com>
Rather than make all the BasePrimitive2D classes bear the cost of being an UNO
object, we just wrap the top level BasePrimitive2D in this class when we need
to pass them over UNO.
This reduces the locking overhead when doing normal drawinglayer operations,
and reduces the size of drawinglayer objects and the cost of initialising
them, which shaves 5% off the load/display time of a large barchart.
Add new drawinglayer::convertPrimitive2DContainerToBitmapEx utility
method to avoid needing to convert to Sequence<XPrimitive2D>
Change-Id: I553eaa4c16ba016b098cb21f6c55f5008f0d9b53
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/126487
Tested-by: Noel Grandin <noel.grandin@collabora.co.uk>
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
from "append" to "visit"
The current naming comes from when I created this class as replacement
for passing Primitive2DContainer around and the naming was meant to
reduce code churn.
But now that I'm using it more, the naming is an impediment to
understanding the code.
Change-Id: I4ee0156c7ffb534eac64f1c05eac6b8c67c25ab9
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/126321
Tested-by: Jenkins
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
Found with bin/find-unneeded-includes
Only removal proposals are dealt with here.
Change-Id: I8daf57df5b8f9c609d4f3ca44bae7366addceb6c
Reviewed-on: https://gerrit.libreoffice.org/69151
Tested-by: Jenkins
Reviewed-by: Miklos Vajna <vmiklos@collabora.com>
By brute force analysis, it seems to me that we can easily
remove these not used includes.
It passed "make check install"
Change-Id: I34ea2aae57e56fb8b2cf1066cf51b08068fa0b53
Reviewed-on: https://gerrit.libreoffice.org/41312
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Jochen Nitschke <j.nitschke+logerrit@ok.de>
For text layouted using EditEngine the HitTest in SVX is
identifying Field like URLs. Thus ist is better to use the
anyways more precise primitives for HitTest (rotation/shear/
mirror, ...). This was necessary since the former mechanism
which used a combination of primitive-beased HitTest and then
using an Outliner to get the position/content of the Field
landed on different positions e.g. when the layout needed to
use multiple lines for the contained URL, but there could
be more cases found.
Adapted the text decompositon, the primitive HitTest and
the TextHirearchyFieldPrimitive2D accordingly.
Change-Id: Ice559e20d02547fdcfcf9783e7cc5481706aab03
Reviewed-on: https://gerrit.libreoffice.org/40591
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Armin Le Grand <Armin.Le.Grand@cib.de>
instead of returning a Primitive2DContainer from each method which we
are then going to immediately append to another container, pass down a
single container by reference which we can append to
Change-Id: I0f28a499d2ec54f7111a7044c30099767aa079e1
Reviewed-on: https://gerrit.libreoffice.org/30258
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>
The issue of 362d4f0cd4 "Explicitly mark
overriding destructors as 'virtual'" appears to no longer be a problem with
MSVC 2013.
(The little change in the rewriting code of compilerplugins/clang/override.cxx
was necessary to prevent an endless loop when adding "override" to
OOO_DLLPUBLIC_CHARTTOOLS virtual ~CloseableLifeTimeManager();
in chart2/source/inc/LifeTime.hxx, getting stuck in the leading
OOO_DLLPUBLIC_CHARTTOOLS macro. Can't remember what that
isAtEndOfImmediateMacroExpansion thing was originally necessary for, anyway.)
Change-Id: I534c634504d7216b9bb632c2775c04eaf27e927e
A ridiculously fast way of doing this is:
for i in $(pcregrep -l -M -r --include='.*[hc]xx$' \
--exclude-dir=workdir --exclude-dir=instdir '^
{3,}' .)
do
perl -0777 -i -pe 's/^
{3,}/
/gm' $i
done
Change-Id: Iebb93eccbee9e4fc5c4380474ba595858a27ac2c
Reviewed-on: https://gerrit.libreoffice.org/22224
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Chris Sherlock <chris.sherlock79@gmail.com>
when used as a mutable data-structure. Plain std::vector halves the time
taken to display the chart dialog
Create a class to represent the std::vector we are going to be passing
around, and move some of the utility methods into it to make the code
prettier.
Also create an optimised append(&&) method for the common case of
appending small temporaries.
Change-Id: I7f5b43fb4a8a84e40e6a52fcb7e9f974091b4485
includes should be able to be included on their own
fix some of the ones that do not respect
that rule.
Change-Id: Id161224a1978461d3cea43252f232f18888a4f61
Reviewed-on: https://gerrit.libreoffice.org/19612
Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Norbert Thiebaud <nthiebaud@gmail.com>
Andrea Gelmini