| Age | Commit message (Collapse) | Author |
|---|
|
The src/lx_memory.c file uses the xf86.h header file. This file must have
HAVE_STRNDUP defined before calling it when the building platform has a strndup
function. When using config.h, this file doesn't have that define and so fails
to compile.
The attached patch adds the conditional config.h to this file so it may compile
on the affected platforms.
The patch is trivial and may be included and used under whatever licensing
desired.
Closes: Free Desktop Bug #84541
Signed-off-by: Brian A. Lloyd <brian.lloyd@familyhonor.net>
|
|
This patch adds conditional support for the Geode MSR device on OpenBSD.
Signed-off-by: Marc Balmer <mbalmer@netbsd.org>
Signed-off-by: Matthieu Herrb <matthieu@openbsd.org>
|
|
|
|
Thanks to Maarten Lankhorst for the fix.
|
|
Signed-off-by: Adam Jackson <ajax@redhat.com>
|
|
|
|
Adds handling of negative srcX or srcY coordinates, which is used by gtk3
GtkTrayIcon implementation via cairo for centering icons on top of the
destination area.
For now just handle the drawing within source pixmap bounds in these
cases, as full handling of the out of bound zero-vector for PictOpSrc
seems to require bigger refactoring, which is not suitable for the
impending stable release.
The approach implemented here should result in correct rendering for
PictOpOver (as out of bounds with non-repeat mode ends up meaning
transparency). For PictOpSrc no black borders are rendered where
appropriate, but before this approach, no source pixmap would even be
rendered at all, so it should be a net improvement for the upcoming
release, while fixing the gtk3 icons in GNOME3 fallback.
The completely out of bounds special case check didn't work correctly for
negative srcX or srcY because exaScratch.srcWidth/Height are unsigned
types, so the negative srcX or srcY got casted to unsigned, which resulted
in a big number, so srcX >= 0 was always true (but GCC managed to not warn
of the implicit signedness cast).
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=51360
|
|
cairo-1.12 uses solid pictures instead of 1x1R pixmaps in glyph rendering
paths, so accelerate it.
In addition to acceleration, it avoids a bug in xserver-1.13.0 and earlier
which causes visible misrendering for fallback path, making cairo-1.12 a
viable and desired choice on GeodeLX systems.
Quick benchmarking suggests a 4-12% win in cairo-traces.
|
|
Became redundant with commit 32a927b8e6
|
|
Debian's 'licensecheck' tool showed these two files to have unspecified
copyright and license. Git log clarifies that these had been created by
Jordan Crouse when he was still at AMD, so their stock copyright notice
was copied as-is from other files in src/
|
|
We don't support 10bit wide color channels, BGRA and other types, which mainly
cause these fallbacks. The ErrorF's were unwarranted, these are just fallbacks
as any other, so don't output these warnings without GEODE_TRACE_FALL.
|
|
lx_check_composite has many special checks for PictOpAdd to return
successfully early, but these were done even earlier than the check
for solid fills. This resulted in lx_prepare_composite assuming
the source pixmap exists, which is not the case with solid fills.
Move the solid fill fallback checks before the others, so we are
guaranteed to always have a source pixmap in PrepareComposite and
Composite and don't crash trying to access a NULL pointer.
Also move up the mask check to do less checks in the PictOpAdd logic.
|
|
...and flip the location of the recent src/Makefile.am addition.
|
|
|
|
Preparing to compile geode on X 1.13
All compiler errors are gone as far as I can tell,
but I lack the hardware to verify it works as well.
|
|
|
|
The old version of just checking for EXA_MINOR_VERSION >= 2 would
stop working if ever EXA gets a new major version, as minor would be
reset to 0, and start unexpectedly working again when minor gets bumped
gradually up to 2 again.
Other drivers also assume that a new major would still have the vfunc we use.
|
|
geode_driver.c:372:9: warning: variable 'cpu_detected' set but not used [-Wunused-but-set-variable]
Was used only in a DEBUGMSG, which has since been made not do anything ever, so just kill
it completely, as the debug message wasn't useful anyway.
|
|
|
|
geode_ddc.c: In function 'GeodeI2CInit':
geode_ddc.c:158: error: cast to pointer from integer of different size
Signed-off-by: Dave Airlie <airlied@redhat.com>
|
|
On hybrid platforms, such as Debian GNU/KFreeBSD, that support V4L2
and yet without providing the full complement of Linux headers, ZTV
fails to compile because of this missing header.
Given how removing this header does not adversely affect compiling,
plus it might improve portability, we went ahead and removed it.
Nonetheless, as observed by Gaetan Nadon, z4l.c includes code that
explicitly disables the module on any OS other than Linux:
LoaderGetOS(&osname, NULL, NULL, NULL);
if (osname == NULL || strcmp(osname, "linux") != 0) {
if (errmaj)
*errmaj = LDR_BADOS;
if (errmin)
*errmin = 0;
return NULL;
}
Still, in the interest of OS neutrality, we decided to leave recent
autoconf magic in place and to let OS vendors figure out how to get
the module loaded on non-Linux OS themselves. Patches are welcome.
Signed-off-by: Martin-Éric Racine <martin-eric.racine@iki.fi>
|
|
|
|
Signed-off-by: Adam Jackson <ajax@redhat.com>
|
|
The author of the write() system call has deemed important for the caller
to check the return code. If not, the compiler issues a warning.
The patch handles a bad return code from write() just in the same way
as the code does for the open() calls.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
The LFS "transitional extension" API is deprecated and is not available on FreeBSD.
Large file support (64 bit) is not required on the geode 32 bit only architecture.
There are some hints that the LFS transional extension API were not used
correclty. The variable holding the value is held in an unsigned long rather
than in off_t. The msr open call did not use the O_LARGEFILE flag and did not
check for EOVERFLOW.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
Now we have a clean separation between CPP flags and C Flags
Only the C Flags apply to both geode and ztv driver, however
it is not worth doing a per target compilation.
Should some flags become mutually exclusive and if we want to have
both target in the same makefile, add the incompatible flag
on the appropriate target, example:
geode_drv_la_CFLAGS = $(AM_CFLAGS) -newCFlag
and/or
geode_drv_la_CPPFLAGS = $(AM_CPPFLAGS) -newCPPFlag
Note that both AM_CFLAGS and AM_CPPFLAGS are picked-up by both
geode and ztv driver target.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
Clean-up what goes in AM_CFLAGS and AM_CPPFLAGS
Use more descriptive names such that comments are not needed.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
For gcc compiler only at the moment.
The configuration reports if support is found or not, e.g.:
checking if gcc supports the -m32 Intel/AMD option... yes
The configuration C test relies on #include unistd.h which includes
features.h which includes gnu/stubs.h which includes gnu/stubs-32.h
which is missing on 64 bit system without the 32 bit library support.
Tested on x86_64 AMD64 CPU with/without libc6-dev-i386 which provides
32 bit support. Remains to be tested on Geode and FreeBSD 32/64 bit.
The configuration does not attempt to decide if building should proceed
or be aborted. If no 32 bit support then the build will die in the
assembly code as it always did before.
The variable M32_CFLAGS provides the flag for the makefiles.
The variable names and organization is subject to change.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
No functional changes.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
Use already existing:
--enable-strict-compilation
Enable all warnings from compiler and make them
errors (default: disabled)
which, byt the way was broken as it was not implemented in geode.
Use already existing CWARNINGFLAGS variable, so there is no need
to define GCC_WARNINGS.
Both come from util-macros version 1.4 or later.
http://cgit.freedesktop.org/xorg/util/macros/tree/xorg-macros.m4.in
?id=03b04a6b5d61aafbd2705f7d29e58e0c92553a4a
If you configure with a later version of util-macros you may see new
warnings appear as additional flags have been added over the years.
You get better support for platforms/compiler differences.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
Only the suffixes .s, .S, and .sx are recognized by automake
as being files containing assembly code.
http://www.gnu.org/software/automake/manual/automake.html#Assembly-Support
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
It is used in ztv but not geode driver.
In the wake of more platforms other than linux, might as well
remove any source of confusion.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
This macro has been removed every where from source code.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
The convention is to have the video driver (geode) in the src subdir
and any other targets (utils, apps, other drivers) in their own
directory. It makes it much simpler to right a correct makefile
and to set conditional compile.
This highlighted a few things to fix. For now, the directory includes
have not been copied as they are required. The AM_CCASFLAGS is not required
either as there is no assembler.
Other than that, the gcc command issued has no differences. Other unrequired
defines could be removed later.
Signed-off-by: Gaetan Nadon <memsize@videotron.ca>
|
|
|
|
|
|
|
|
|
|
Changing GEODE_TRACE_FALL to 1 allows to easily find the unaccelerated
code paths in a given use case while working on driver performance.
Signed-off-by: Priit Laes <plaes@plaes.org>
Signed-off-by: Mart Raudsepp <leio@gentoo.org>
|
|
First argument is the number of elements and second the size of one
element, not vice-versa.
Signed-off-by: Mart Raudsepp <leio@gentoo.org>
|
|
V4L1 support was dropped as of Linux kernel 2.6.38
|
|
Commit 5e72a00ad2 caused packed video data to corrupt glyphs and other
parts of the screen, but it turns out that the commit actually at fault
was d681a844e, incorrectly changing the size of the allocated destination
memory from the number of bytes needed to the number of lines needed.
While fixing this, I noticed that LXAllocateSurface is probably making the
same mistake, and that the height is probably not calculated correctly for
some corner cases when calling LXCopyFromSys in the packed
video path (but I'm not sure about either).
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=33004
|
|
Signed-off-by: Priit Laes <plaes@plaes.org>
|
|
The LX is a bit odd in that the palette can be used for gamma correction
of graphics data, or video data, but not both at the same time.
Right now, during X startup, X sets gamma correction (but without any
actual correction) causing the Geode driver to start using the palette
for graphics gamma correction. (this is just momentary)
Later on during startup, vg_set_custom_mode() switches to using the
palette for video gamma correction, and this is the end result.
If you later use xrandr to change the gamma, the palette starts being used
for graphics again.
If you VT switch, vg_set_custom_mode (momentarily) starts using the palette
for video, but then X jumps in with a gamma callback causing it to be used
for graphics again.
End result: no user visible bug, but this is inconsistent.
As there is no exposed mechanism for changing video gamma I suggest
we just drop that bit from vg_set_custom_mode(). Then the inconsistency
goes away.
Downside: this now becomes dependent on X setting gamma during startup.
I don't know when this was introduced but I think it is recent.
Aditionally, I think I found a (harmless) bug in
df_set_video_palette_entry, it doesn't do what the comments say.
|
|
Signed-off-by: Priit Laes <plaes@plaes.org>
|
|
NOTE: the eCafe EC-800's panel doesn't support a higher refresh rate.
|
|
Signed-off-by: Frank Huang<frankr.huang@amd.com>
|
|
This reverts commit fab58e111b6655602b4f739260a01247ff478f35.
|
|
Signed-off-by: Frank Huang<frankr.huang@amd.com>
|
