This improves aa10text performance from 74k to 569k on my 855 laptop.
This also causes my 865 to hang on aa10text like it does on rgb10text,
thanks to actually hitting render accel.
This lets the driver allocate a nice idle buffer object instead of a
busy one, reducing runtime of firefox-20090601 on my G45 from 50.7 (+/- .41%)
to 48.4 (+/- 1.1%).
This was needed when we were doing the mask computations in this pixmap,
but now they're done in a temporary and then uploaded later.
This reduces runtime of firefox-20090601 from 52.6 (+/- .96%) to 50.7
(+/- .41%) seconds on my G45.
This synchronizes the X EDID data with the kernel EDID data each time the
kernel data may have changed. Otherwise, X ends up stuck with the first EDID
data it sees, failing to accomodate to different monitors.
Signed-off-by: Keith Packard <keithp@keithp.com>
DPMS header was split into dpms.h (client) and dpmsconst.h (server). Drivers
need to include dpmsconst.h if xextproto 7.1 is available.
SHM is now shm.h instead of shmstr. Requires definition of ShmFuncs that's
not exported by the server.
Signed-off-by: Peter Hutterer <peter.hutterer@who-t.net>
Don't do it, treat this the same as every other prepare access call in uxa.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Owain Ainsworth <zerooa@googlemail.com>
The two shared i830_composite.c, so giving i830 atomic batch support
triggered anger about starting i830's atomic area while in i915's atomic
area. Instead, split the emit-a-primitive stuff from the state emission.
scrn->fbOffset may be changed when binding objects to the aperture during
server initialization or VT enter. This was accidentally removed when the
NoAlloc option was eliminated.
Signed-off-by: Keith Packard <keithp@keithp.com>
Without kernel support and explicit knowledge about where in the ring the
last rendering operation for a specific pixmap was, we must synchronize with
any outstanding rendering before accessing a pixmap which does not have a
buffer object.
Signed-off-by: Keith Packard <keithp@keithp.com>
The frame buffer only has a valid address between prepare_access and
finish_access calls, so remove all other attempts to compute an address from
the driver.
Signed-off-by: Keith Packard <keithp@keithp.com>
We only need to get static offsets for objects when not running KMS,
otherwise the kernel will manage those as needed for us.
Binding objects is done in one of two ways. For GEM buffer objects, we use
dri_bo_pin. For GART allocated memory, we bind that to the GART.
GEM requires GTT space to map objects. Under KMS, the kernel driver has
already provided all available GTT space to GEM, so the X server need not do
anything.
Signed-off-by: Keith Packard <keithp@keithp.com>
For non-DRM environments, the screen pixmap will be GART allocated memory
and not a libdrm buffer object and so uxa will only use devPrivate.ptr to
find the associated memory. Make sure devPrivate.ptr is set each time the
framebuffer is allocated so that uxa will be able to draw to it.
Signed-off-by: Keith Packard <keithp@keithp.com>
KMS mode does not call I830AccelMethodInit as that does the user
modesetting initialization (yes, it was misnamed), but that means that the DRI option
was ignored. Create a new i830_check_dri_option function to do the option
detection, then remove that from I830AccelMethodInit, which is renamed
i830_user_modesetting_init to reflect what it actually does.
Signed-off-by: Keith Packard <keithp@keithp.com>
This removes yet another 'debugging' option that hasn't seen real use in a
long time, and wasn't supported under KMS in any case.
Signed-off-by: Keith Packard <keithp@keithp.com>
The overcommit of address space combined with these buffers hitting SW
fallbacks all the time means that we're probably better off telling the
application "no" instead of likely silently failing later.
Bug #22601.
Currently we implemented basic sdvo lvds function,
But except for sdvo lvds fixed mode, we can not switch
to other modes, otherwise display get black. The patch
intends to work for all modes whose HDisplay and VDisplay
are lower than fixed mode.
Signed-off-by: Ma Ling <ling.ma@intel.com>
The bigrequests limit isn't present in current X servers (tested using
textured video on a 965 with both image and window at 2048x2048 on a
1920x1200 display, and image at 2048x2048, window at 1024x1024).
Remove the artificial limit, enabling full-screen HD video when
rotated.
When not using DRI, the screen pixmap is not in a bo, and so the usual
enable/disable access functions don't adjust the pixmap devPrivate field,
leaving it to the frame buffer allocation code to assign this correctly.
During mode setting and fb resizing, FB access is disabled, and the
screen pixmap devPrivate is stashed away by xf86EnableDisableFBAccess,
to be restored when FB access is turned back on. This means that we have to
set the pixmap devPrivate.ptr (in case xf86EnableDisableFBAccess doesn't
do this), along with storing the address in the scrn->pixmapPrivate field.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
After failing to become DRM master, the X server dies attempting to close
the master fd during free:
(EE) intel(0): [drm] failed to set drm interface version.
(EE) intel(0): Failed to become DRM master.
(EE) intel(0): failed to get resources: Bad file descriptor
(EE) intel(0): Kernel modesetting setup failed
Backtrace:
0: X(xorg_backtrace+0x3b) [0x8133a3b]
1: X(xf86SigHandler+0x55) [0x80c7945]
2: [0xb805d400]
3: /usr/lib/xorg/modules/drivers//intel_drv.so [0xb7b4bfcc]
4: X(xf86DeleteScreen+0x6b) [0x80d465b]
5: X(InitOutput+0x548) [0x80b0158]
6: X(main+0x1cb) [0x807220b]
7: /lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xe5) [0xb7d107a5]
8: X [0x8071881]
Saw signal 11. Server aborting.
ddxSigGiveUp: Closing log
ddxSigGiveUp: re-raising 11
Segmentation fault
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The DRI2 interface was changed to support both old and new drivers in an
API/ABI compatible fashion. This change syncs the intel driver with the new
version of the DRI2 API.
Signed-off-by: Keith Packard <keithp@keithp.com>
When playing a movie that is clipped on its left and right edges the Xorg
server will SEGV sometimes. This is because the intel driver ignores the
clipping info when it copies the planes out of the XV data.
The check for the optimised copy was wrong to ignore the width required.
Which leads to too much data being copied by the memcpy. It the source buffer
happens to end exactly on a page boundary the server will SEGV.
As we reviewed the code we checked the calculation of src1, src2 and src3.
The patch includes additional comments to make it clear what the elements of
the calculation are.
This bug exists in git head and we also see it in 2.4.1.
Barry
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Since the scratch pixmap header will be re-used after allocation, we
need to clear its bo attachment when we stop using it, otherwise a later
user will use a bogus bo.
Reviewed-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
To slightly clean up the implementation of i830_update_polyphase_coeffs,
introduce the two small helper functions i830_limit_coeff and
i830_store coeffs_in_overlay_regs.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
- also ensure that the most significant byte is zero
- while I was looking at the code, add the Overlay suffix to
SetPortAttribute like in the textured case.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Bug #19578. We should set private intel_crtc state according
to current, as fail to do so pipe A needs active won't be taken
care of. Also make sure pipe swap operation always set during
VT switch.
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Xiang, Haihao
Eric Anholt
Keith Packard
Carl Worth
Peter Hutterer
Owain Ainsworth
Barry Scott
ling.ma@intel.com
Simon Farnsworth
Chris Wilson
Rémi Cardona
Zhenyu Wang
Jesse Barnes
Daniel Vetter