[ Michel: Minor fixups to address compiler warnings ]
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Michel Dänzer <daenzer@vmware.com>
Acked-by: Maarten Maathuis <madman2003@gmail.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
This is more elegant and probably also slightly more correct than doing it
at FinishAccess time.
Signed-off-by: Michel Dänzer <daenzer@vmware.com>
Acked-by: Maarten Maathuis <madman2003@gmail.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
Only return TRUE if the GPU copy is being accessed, and preserve the return
value on repeated / nested calls for the same pixmap.
exaPrepareAccessReg_mixed could get inconsistent return values e.g. when the
same pixmap is both the destination and source of an operation, potentially
resulting in a crash.
Signed-off-by: Michel Dänzer <daenzer@vmware.com>
Acked-by: Maarten Maathuis <madman2003@gmail.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
- Fixup some variable names as well.
Signed-off-by: Maarten Maathuis <madman2003@gmail.com>
Acked-by: Michel Dänzer <michel@daenzer.net>
Signed-off-by: Keith Packard <keithp@keithp.com>
- Pixmaps that are created during a fallback are automatically prepared access.
- During the fallback accelerated ops are blocked to prevent new/scratch gc's
from triggering accelerated ops on mapped pixmaps.
- A few cases of incorrect wrapping (on the top level pointer instead of
between damage and (w)fb) have been removed.
Signed-off-by: Maarten Maathuis <madman2003@gmail.com>
Acked-by: Michel Dänzer <michel@daenzer.net>
Signed-off-by: Keith Packard <keithp@keithp.com>
If there are several software fallbacks affecting the system memory copy of the
same pixmap, only copy the results back to the driver pixmap when it's used for
acceleration again, or in the BlockHandler, whichever happens first.
If the PrepareAccess hook fails, use the DownloadFromScreen hook to retrieve
driver pixmap contents to a system RAM copy, perform software rendering on that
and copy the results back using the UploadToScreen hook. Use the classic
migration logic to minimize transfers (which as a bonus allows slightly
cleaning up some of the existing mixed pixmap code).
This enables things that weren't possible before with driver-allocated pixmap
storage: If some (or all) GPU pixmap storage can't be mapped directly by the
CPU, this can be handled between the PrepareAccess and
DownloadFrom/UploadToScreen hooks, e.g.:
* Radeon KMS on big endian machines can fail PrepareAccess if the pixmap
requires byte-swapping and swap bytes in DownloadFrom/UploadToScreen.
* Environments where GPU and CPU don't have a shared address space at all.
Here the driver PrepareAccess hook will always fail and leave all transfers
between GPU / CPU storage to the Download/From/UploadToScreen hooks.
Drivers which can handle all pixmaps in the PrepareAccess hook should notice
little if any difference.
- Based on driver pixmaps with some changes (completely transparent to driver).
- It helps with the problem of known software fallbacks, such as trapezoids.
- exaDoMigration is now called for all cases that provide a do_migration hook.
- exa_migration.c is renamed to exa_migration_classic.c
- Create a few seperate functions and a few private function pointers.
- Replace a few if conditions with a check for pExaPix->pDamage instead.
- This is in preperation of a third scheme that lies somewhere in between.
- Code clarity would have suffered (i started working on it and didn't like the mess).
Use reference counting and do nothing unless the reference count transitions
to/from 0.
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=22822 .
As a bonus, this avoids calling the driver Prepare/FinishAccess hooks more than
once per pixmap and operation.
Also update the Doxygen documentation for the PrepareAccess driver hook to
better match current reality.
At most once per second, under the following circumstances:
* We can't satisfy an offscreen memory allocation, but there seems to be enough
offscreen memory available in total.
or
* The server has been idle for at least 100ms, and there is more than one
available offscreen area.
Signed-off-by: Michel Dänzer <daenzer@vmware.com>
This should give the full benefits of the glyph cache even when we can't use a
mask.
This also means we no longer need to scan the glyphs to see if they overlap,
we can just use a mask or not as the client asks.
Signed-off-by: Michel Dänzer <daenzer@vmware.com>
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=18710 .
As this can't work without new EXA_PREPARE_AUX* indices, this requires a major
version bump, so we can also drop the UploadToScratch driver hook and
ExaOffscreenSwap*(). So this also fixes
http://bugs.freedesktop.org/show_bug.cgi?id=20213 .
Moreover, introduce EXA_DRIVER_KNOWN_MAJOR to break compilation of drivers
which may not be able to handle EXA_PREPARE_AUX*, giving instructions how to
make them build again in the #error message.
Signed-off-by: Michel Dänzer <daenzer@vmware.com>
- I did some testing with full fallbacks forced by the driver.
- I ran rendercheck, expedite and the (full) x11perf test suite.
- Thanks to ajax for pointing out this should be unneeded.
A grep on xorg/* revealed there's no consumer of this define.
Quote Alan Coopersmith:
"The consumer was in past versions of the headers now located
in proto/x11proto - for instance, in X11R6.0's xc/include/Xproto.h,
all the event definitions were only available if NEED_EVENTS were
defined, and all the reply definitions required NEED_REPLIES.
Looks like Xproto.h dropped them by X11R6.3, which didn't have
the #ifdef's anymore, so these are truly ancient now."
Signed-off-by: Peter Hutterer <peter.hutterer@redhat.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>
- Redo damage naming for more consistency.
- Call post submission functions only where appropriate.
- EXA can now live without it's odd damage workarounds.
There's no reason to not just dispatch this straight into the GC. As a
bonus, if you do so, damage wraps correctly, and thus swcursor works.
The side effect is it's no longer possible to override ShmPutImage with
ShmRegisterFuncs().
Also remove the (broken) damage tracking for same from EXA, since it didn't
work right, and is now superfluous.
Recording damage from other operations (e.g. creating a client damage record)
may confuse the migration code resulting in corruption.
Option "EXAOptimizeMigration" appears safe now, so enable it by default. Also
remove it from the manpage, as it should only be necessary on request in the
course of bug report diagnostics anymore.
Add a function to composite multiple independent rectangles
from the same source to the same destination in a single
operation: this is useful for building a glyph mask.
Add back exaGlyphs(); the new version copies the glyph images
onto a single large glyph pixmap and draws from their to the
destination surface. This reduces the management of small
offscreen areas and will allow us to avoid texture unit setup
between each glyph.
Replace the current score keeping algorithm with a rolling counter that's
incremented in ExaOffscreenMarkUsed, with the previous value being stored
in the area. exaOffscreenAlloc uses the difference between the counter
value and the value in the area when deciding which area to evict.
It now also takes the size of the areas into account, and favors evicting
smaller areas.
The credit for these ideas goes to Michel Dänzer.
This was an attempt to avoid scratch gc creation and validation for paintwin
because that was expensive. This is not the case in current servers, and the
danger of failure to implement it correctly (as seen in all previous
implementations) is high enough to justify removing it. No performance
difference detected with x11perf -create -move -resize -circulate on Xvfb.
Leave the screen hooks for PaintWindow* in for now to avoid ABI change.
