Godot supports many different compilers and for production releases we
have to support 3 currently: GCC8, Clang6, and MSVC2017. These compilers
all do slightly different things with -ffast-math and it is causing
issues now. See #24841, #24540, #10758, #10070. And probably other
complaints about physics differences between release and release_debug
builds.
I've done some performance comparisons on Linux x86_64. All tests are
ran 20 times.
Bunnymark: (higher is better)
(bunnies) min max stdev average
fast-math 7332 7597 71 7432
this pr 7379 7779 108 7621 (102%)
FPBench (gdscript port http://fpbench.org/) (lower is better)
(ms)
fast-math 15441 16127 192 15764
this pr 15671 16855 326 16001 (99%)
Float_add (adding floats in a tight loop) (lower is better)
(sec)
fast-math 5.49 5.78 0.07 5.65
this pr 5.65 5.90 0.06 5.76 (98%)
Float_div (dividing floats in a tight loop) (lower is better)
(sec)
fast-math 11.70 12.36 0.18 11.99
this pr 11.92 12.32 0.12 12.12 (99%)
Float_mul (multiplying floats in a tight loop) (lower is better)
(sec)
fast-math 11.72 12.17 0.12 11.93
this pr 12.01 12.62 0.17 12.26 (97%)
I have also looked at FPS numbers for tps-demo, 3d platformer, 2d
platformer, and sponza and could not find any measurable difference.
I believe that given the issues and oft-reported (physics) glitches on
release builds I believe that the couple of percent of tight-loop
floating point performance regression is well worth it.
This fixes#24540 and fixes#24841
The current system for capturing the mouse and generating motion events on X11
has issues with inaccurate and lopsided input. This is because both
XQueryPointer and XWarpPointer work in terms of integer coordinates when the
underlying X11 input driver may be tracking the mouse using subpixel
coordinates. When warping the pointer, the fractional part of the pointer
position is discarded.
To work around this issue, the fix uses raw motion events from XInput 2. These
events report relative motion and are not affected by pointer warping.
Additionally, this means Godot is able to detect motion at a higher resolution
under X11. Because this is raw mouse input, it is not affected by the user's
pointer speed and acceleration settings. This is the same system as SDL2 uses
for its relative motion.
Multitouch input on X requires XInput 2.2. Raw motion events require
XInput 2.0. Since 2.0 is old enough, this is now the minimum requirement to
use Godot on X.
Also finally move freetype to its own env and disable warnings for it.
Still needs some work to fix the awkward situation of the freetype and
svg modules used in scene/ and editor/ respectively.
We were already linking libstdc++ statically for official binaries,
protecting us against most portability issues. But apparently since
we started using GCC 7 for official builds, we also need to link
libgcc statically for at least 32-bit builds to be portable.
Fixes#16409.
This adds a separate_debug_symbols option to the x11, windows, and osx
targets. This will default to adding normal debugging symbols to the
artifacts and only splits them when separate_debug_symbols=yes on the
Scons command line.
Also made LINK and CXXFLAGS configurable as command line options.
Note that LINK currently expects the *compiler* that will be used
for linking and will call its configured linker behind the scenes
(so g++, clang++, etc., not ld.gold). See #15364 for details.
-Fixes to unwrapper (remove degenerates), makes Thekla not crash
-Added optional cancel button in EditorProgress
-Added function to force processing of events (needed for cancel button)
There are still some left in the Android Java code, even stuff to swap between
GLES1 and GLES2 support from early Godot days... would be good to see some cleanup
there too one day.
The "graphics/api" option for Android exports is removed, as only GLES 3.0 is supported.
It can be readded when GLES 2.0 support comes back. Fixes#13004.
Travis always has massive backlog of macOS builds, so we can't rely on them
too much.
The iphone build was mostly useful to spot tools=no or target=release_debug
issues, so replacing it by an appropriate X11 build.
This release hides many struct members which provides easier forward
compatibility but is a break from previous releases. A few small macros
provide compatibility between both 1.1.0 and 1.0.x.
Fixes#8624.
Now that we have a built-in stacktrace on a segfault it would be useful
to have debug information on debug_release builds so that bugreports can
include this information. Without this debug info we will still get
function names in the backtrace but not file location.
This commit will by default build all targets with minimal debug info
and then strip the information into separate files. On MacOS this is a
.dSYM file, on Linux/MingW this is a .debug file. MacOSX will
automatically load a dSYM file if it exists in its debugger. On
Linux/MingW we create a 'gnu debuglink' meaning that gdb and friends
will automatically find the debug symbols if they exist.
Existing workflow for developers does not change at all, except that we
now create two instead of one build artifact by default.
This commit also adds a 'debug_symbols' option to X11, MacOS, and MingW
targets. The default is 'yes' which corresponds to -g1. The alternatives
are 'no' (don't generate debug infos at all) or 'full' which runs with
-g2. A target=debug build will now build with -g3.
Apparently -ffast-math generates incorrect code with recent versions of
GCC and Clang. The manual page for GCC warns about this possibility.
In my tests it doesn't actually appear to be measurably slower in this
case, and this is used in a batch process so it seems safe to disable
this.
This fixes#10758 and fixes#10070
The pattern and replacement matching behaviour has been changed purely
due to the nature of switching to a standards-compliant library.
One mistake in the previous behaviour was that named groups didn't have
a number. This has been corrected.
As names are actually just an alias of numbered groups,
RegExMatch::get_name_dict() is now get_names() and is a dict
referring to the group number it represents.
Duplicate names are enabled and the with the first matching instance
used.
Due the lack of a suitable equivalent in PCRE2, RegExMatch::expand() was
removed.
Tried to organize the configure(env) calls in sections, using the same order
for all platforms whenever possible.
Apart from cosmetic changes, the following issues were fixed:
- Android: cleanup linkage, remove GLESv1_CM and GLESv2
- iPhone: Remove obsolete "ios_gles22_override" option
- OSX:
* Fix bits detection (default to 64) and remove obsolete "force_64_bits" option
(closes#9449)
* Make "fat" bits argument explicit
- Server: sync with X11
- Windows: clean up old DirectX 9 stuff
- X11:
* Do not require system OpenSSL for building (closes#9443)
* Fix typo'ed use_leak_sanitizer option
* Fix .llvm suffix overriding custom extra_suffix
Hein-Pieter van Braam
Cosmic Chip Socket
Ibrahn Sahir
Marcelo Fernandez
Rémi Verschelde
Juan Linietsky
Hugo Locurcio
bruvzg
Fabio Alessandrelli
Juan Linietsky
Colin Kinloch
Pedro J. Estébanez
Felix M. Cobos
Elliott Sales de Andrade
Zher Huei Lee