Improves rendering performance, esp. for large designs by:
- using a separate glDrawArrays calls() for large buffers of contiguous vertices (e.g. large zone fills)
- pooling smaller items into a small-sized index buffer held in system RAM (has to be DMAed to the GPU by the driver anyway)
Previously, the count of segments used a magic number optimized for Pcbnew.
This is not good, and does not work well on Gerbview.
The count uses now a max error acceptable is approximation (5 microns in Gerbview and Pcbnew).
Fixes#9101https://gitlab.com/kicad/code/kicad/issues/9101
curl.h has constraints on Windows (because it includes winsocks2.h) and must be
included before any wxxxx.h.
Because only very few files need curl.h, kicad_curl.h is included only in these files
Reduce each (Accel & Fallback) to 3 options: Off, Fast, Good. Fast AA
in accelerated canvas uses SMAA with tweaks suggested by David Beinder.
Good AA is super-sampled x2.
Cairo is similarly reduced to Off, Fast, Good, which map to the Cairo
options themselves. Best is removed as it mostly affects text rendering
and not line drawing (as our text is)
Fixes https://gitlab.com/kicad/code/kicad/-/issues/8522
Conservative makes SMAA usable in Eeschema through weaker AA.
Aggressive reduces some artifacts compared to the High preset.
Restores the unused Ultra preset in the shader source to its original state.
Avoid fractional pixel offsets by forcing integer division of m_screenSize.
Shift rounded lookAtPoint onto OpenGL pixel center to ensure consistent rounding to int.
The pixel 0..1 @ 1x needs to fill the area 0..2 @ 2x so its center moves by 0.5 * (pixel size @ 1x).
This was leaking windows headers and partial wx headers to 1084 compilation units......
This also means math/util.h is leaking to 1084 compilation units which seems a bit high too.
The native cursor must be set on the GAL canvas wx object,
not its frame (otherwise on GTK the cursor is only set
when we get a request for it from the WM and not immediately).
Fixes https://gitlab.com/kicad/code/kicad/issues/6421
Vertex memory allocation uses malloc and realloc without checking the
result. Throw exceptions that can be caught in the DoRepaint() method
in EDA_DRAW_PANEL_GAL when malloc and realloc return a null pointer.
wxWidgets 3.1.5+ on Linux will compile with the Wayland EGL
canvas as the backend instead of the X11 backend. This requires a
version of GLEW compiled with the proper EGL defines and a different
header/code for certain parts that are X11 GLEW specific.
This introduces an in-tree version of GLEW that will be built with the
GLEW_EGL flag then statically linked into the KiCad executables when
EGL support is needed.
Correcting the error is more trouble than it is worth, and anyway
might not be corrected exactly the same way plotting/filling/DRC
is corrected anyway.
Use a segment count multiple of 2, because we have a control point on the
middle of the arc, and the look is better if it is on a segment junction
because there is no arc to segment approx error
The radius correction does not fix correctly approximation errors,
and creates serious issues for thin arcs:
- control point outside the arc
- HitTest active area outside the arc shown on screen
- Cairo and opengl gives different arc positions.
- the formula used in radius correction is probably incorrect.
The OpenGL missing segments (when small) are due to rounding in larger
coordinates. The GAL shader uses floats to represent int32 values for
position and distance. Thus, at larger coordinates, the smaller integer
values cannot be represented accurately. This tests for the rounding
error and, if it exists, draws a plain circle instead
Fixes https://gitlab.com/kicad/code/kicad/issues/5751
Cairo Polyline and Polygon draw functions share a common routine. While
Polygons must have 3 points to allow triangulation, a polyline (like our
text) need only have 2
This resolves a crash when passing an empty polygon and alerts debug
users to possible generation cases when bad polygons are created.
Thanks for Frans Zellman
(https://gitlab.com/kicad/code/kicad/-/merge_requests/365) for
identifying the issue.
The use of printf, wxLogDebug, and std::err/std::out causes excessive
debugging output which makes finding specific debugging messages more
difficult than it needs to be.
There is still some debugging output in test code that really needs to
be moved into a unit test.
Add debugging output section to the coding policy regarding debugging
output.
On platforms with limited framebuffer attachments, like GC7000L
driven by the open source etnaviv driver, we can still use
GPU acceleration by drawing directly to the main buffer.
EDA_DRAW_PANEL_GAL::onPaint checks if an overlay target is
available in the GAL, and if not, redraws the whole view if the
overlay target should be dirty.
Clearing of the overlay target is a no-op if there is no overlay
buffer.
Signed-off-by: Lukas F. Hartmann <lukas@mntre.com>
When drawing segments that are smaller than the segment width, OpenGL
did not use realistic values, leading to hidden, small tracks. Instead,
we set the track to only draw a segment when the length is at least 1
radius long (so that we can see the full semi-circle at the track end)
Fixes https://gitlab.com/kicad/code/kicad/issues/5009
- New public static function OPENGL_GAL::CheckFeatures() gets called in EDA_DRAW_PANEL_GAL::SwitchBackend() before switching to OPENGL_GAL
- Moved all OpenGL feature checks from OPENGL_GAL constructor to OPENGL_GAL::init()
Fixes https://gitlab.com/kicad/code/kicad/-/issues/4714
Swap syncs can limit the redraw rate as the screen waits for previously
issued syncs. Setting this to -1 allows for adaptive swapping
(resorting to unsynced if it falls behind) if it is supported by the
card but will fall back to unsynced (0) if the adaptive is not
supported.
Fixes https://gitlab.com/kicad/code/kicad/issues/4226
The scaling factor stored inside the canvas is created by a DPI_SCALING
helper, which will call the platform-specific functions if no user scaling
is specified. This change only affects OSX and Retina displays, so this
now also allows custom scaling to be used on OSX if desired (although it
shouldn't be needed, since wx has detection for it in 3.0.4).
* Split up the thirdparty code into the thirdparty folder (#3637)
* Create a new kimath static library containing all the math functions
This is part of cleaning the build system for #1906.
This update replaces the existing uses of unique pointer creation with
the C++14 std::make_unique call that provides proper memory release in
event of an exception.
Jeff Young
Wayne Stambaugh
Marek Roszko
Ola Rinta-Koski
Mike Williams
Jon Evans
jean-pierre charras
Tomasz Wlostowski
david-beinder
Mikolaj Wielgus
luz paz
Seth Hillbrand
Hajo Nils Krabbenhöft
Ian McInerney
Jonathan Haas
Dominik Wernberger
Roberto Fernandez Bautista
mntmn
Urja Rannikko
Qbort
Simon Richter
Marcus Comstedt