When selecting "CA_TWO_PASS" in glamor_composite_clipped_region() when
the hardware does not support "GL_ARB_blend_func_extended", we call
glamor_composite_choose_shader() twice in a row, which in turn calls
glamor_pixmap_ensure_fbo().
On memory pixmaps, the first call will set the FBO and the second one
will fail an assertion in glamor_upload_picture_to_texture() because
the FBO is already set.
Bail out earlier when the mask pixmap is in memory and the hardware
capabilities would require to use two pass, so that the assertion is not
failed and the rendering is correct.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99346
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
If the libEGL we are using has eglGetPlatformDisplayEXT, yet it still
returns NULL, then this very likely means that it does not support the
type (e.g. EGL_PLATFORM_GBM_MESA) passed in, and then returning NULL is
the right thing to do.
This avoids falling back to an eglGetDisplay() implementation which does
not understands the passed in gbm handle, treats it as a pointer to
something else completely, followed by a crash sooner or later.
Specifically this fixes using the nvidia binary driver, with nvidia's
libEGL + the modesetting driver on a secondary GPU crashing inside
glamor_egl_init() sometimes.
Cc: Eric Anholt <eric@anholt.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
In glamor_init(), if the minimum requirements are not met, glamor may
fail after setting up its own CloseScreen() and DestroyPixmap()
routines, leading to a crash when either of the two routines is called
if glamor failed to complete its initialization, e.g:
(EE) Backtrace:
(EE) 0: Xwayland (OsSigHandler+0x29)
(EE) 1: /lib64/libpthread.so.0 (__restore_rt+0x0)
(EE) 2: Xwayland (glamor_sync_close+0x2a)
(EE) 3: Xwayland (glamor_close_screen+0x52)
(EE) 4: Xwayland (CursorCloseScreen+0x88)
(EE) 5: Xwayland (AnimCurCloseScreen+0xa4)
(EE) 6: Xwayland (present_close_screen+0x42)
(EE) 7: Xwayland (dix_main+0x4f9)
(EE) 8: /lib64/libc.so.6 (__libc_start_main+0xf1)
(EE) 9: Xwayland (_start+0x2a)
Restore the previous CloseScreen() and DestroyPixmap() vfunc handlers in
case of failure when checking for the minimum requirements, so that if
any of the requirement is not met we don't leave the CloseScreen() and
DestroyPixmap() from glamor handlers in place.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1390018
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The extension does not exist in the registry, thus needs to know they're
using EGL 1.5 in order to determine the eglGetPlatformDisplay function
pointer is valid.
Thus brings us into some lovely circular dependency.
Since mesa won't be able (in the foreseeable future) to export the KHR
flavour of extension (another way one could assume that EGL 1.5 is
available) just drop all the heuristics and use the
EGL_EXT_platform_base extension.
In practise (checked with the Mali driver) any EGL 1.5 driver will
advertise support for EGL_EXT_platform_base.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
eglGetDisplay forces the implementation to guess which kind of display
it's been handed. glvnd does something different from Mesa, and in
general it's impossible for the library to get this right. Add a new
inline that gets the logic right, and works around a quirk in epoxy.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Commit cba28d5 - "glamor: Handle bitplane in glamor_copy_fbo_cpu"
introduced a regression as the computed pixmap offset would not match
the actual coordinates and write data elsewhere in memory causing a
segfault in fbBltOne().
Translate the pixmap coordinates so that the data is read and written at
the correct location.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97974
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
Reviewed-and-Tested-by: Michel Dänzer <michel.daenzer@amd.com>
In commit 9e9fcf5 (glamor: Add a helper function for the common
GL_QUADS fallback pattern.), the glDrawArrays count change was
accidentally changed to nbox.
Fixes xlogo with MESA_GL_VERSION_OVERRIDE=2.1 and
MESA_GLSL_VERSION_OVERRIDE=120
Signed-off-by: Mark Yao <mark.yao@rock-chips.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Current Mesa requires that the precision qualifier on uniforms matches
between stages, even if (as in this case) the uniform isn't used in
one of the stages.
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
I don't think anybody has run this code since it was pulled into the
server.
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
The extension came out in 2000, and all Mesa-supported hardware that
can do glamor supports it. We were already relying on the ARB version
being present on desktop.
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Keith Packard <keithp@keithp.com>
glUniform4ui is available starting in GL{,ES} 3.0. Technically it's
also in EXT_gpu_shader4, but that's not worth supporting. There was also
a MESA_shading_language_130 spec proposed at one point; if that ever
gets finished, we can update epoxy to know about it and fix up the
feature check.
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Even if the pixmap's storage has alpha, it may have been uploaded with
garbage in the alpha channel, so we need to force the shader to set
alpha to 1. This was broken way back in
355334fcd9.
Fixes rendercheck -t composite -f x8r8g8b8.
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
The copy optimization in d37329cba4
replicated a bug from last time we did a copy optimization: CopyArea
is only defined for matching depths. This is only a problem at 15 vs
16 depth right now (24 vs 32 would also have matching Render formats,
but they should work) but be strict in case we store other depths
differently in the future.
Fixes rendercheck -t blend -o src -f x4r4g4b4,x3r4g4b4
v2: Drop excessive src->depth == dst->depth check that snuck in.
v3: Switch back to src->depth == dst->depth
v4: Touch up commit message (s/bpp/depth).
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
This function is used by the modesetting driver to implement DRI2 and
shouldn't fail on systems that don't support DRI3.
v2: Drop stale commit message wording, fix compiler warning (by anholt)
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
v2: Fix "orignal" too (review feedback by ajax, change by anholt)_
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
When a GPU gets hotplugged while X is already running, glamor_egl_init()
gets called and changes the current egl context, without updating
lastGLContext, potentially causing the next glamor call on another GPU to
run in the wrong context.
This causes glamor to e.g. crash in the next glamor_create_pixmap() call
(called through the master's screen->CreatePixmap), note this is not the
only troublesome entry point I've seen other backtraces when using a
compositing window manager.
Set lastGLContext to NULL to force the next glamor_make_current() call
to set the right context.
Note that we cannot use glamor_make_current() here to replace the
eglMakeCurrent() call and update lastGLContext for us because
glamor_make_current takes a glamor_priv struct as argument and that
has not been created yet when glamor_egl_init() gets called.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
Add glamor_shareable_fd_from_pixmap function to get dma-buf fds suitable
for sharing across GPUs (not using GPU specific tiling).
This is necessary for the modesetting driver to correctly implement
the DRI2 SharePixmapBacking callback.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Keith Packard <keithp@keithp.com>
This can significantly speed up at least some CopyPlane cases, e.g.
indirectly for stippled fills.
v2:
* Make temporary pixmap the same size as the destination pixmap
(instead of the destination drawable size), and fix coordinate
parameters passed to fbCopyXtoX and glamor_upload_boxes. Fixes
incorrect rendering with x11perf -copyplane* and crashes with the
xscreensaver phosphor hack.
v3:
* Make the change a bit more compact and hopefully more readable by
re-using the existing src_* locals in the bitplane case as well.
Reported-by: Keith Raghubar <keith.raghubar@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Eric Anholt <eric@anholt.net>
With no users of the interface needing the readmask anymore, we can
remove it from the argument passed to these functions.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
This new libXfont API eliminates exposing internal X server symbols to
the font library, replacing those with a struct full of the entire API
needed to use that library.
v2: Use libXfont2 instead of libXfont_2
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
It is a modest performance improvement (2.7% on Intel), with the
significant downside that it keeps extra pixmap contents laying around
for 1000 BlockHandlers without the ability for the system to purge
them when under memory pressure, and tiled renderers don't know that
we could avoid reading their current contents when beginning to render
again. We could use the FB invalidate functions, but they aren't
always available, aren't hooked up well in Mesa, and would eat into
the performance gains of having the cache.
[ajax: rebased to master]
Reviewed-by: Adam Jackson <ajax@redhat.com>
Fixes incorrect clipping for redirected windows which don't happen to be
located at the top left corner of the screen.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96742
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
The screen block and wakeup handlers are the only ones which provide a
well known ordering between the wrapping layers; placing these as
close as possible to the server blocking provides a way for the driver
to control the flow of execution correctly.
Switch the shadow code to run in the screen block handler so that it
now occurrs just before the server goes to sleep.
Switch glamor to call down to the driver after it has executed its own
block handler piece, in case the driver needs to perform additional
flushing work after glamor has called glFlush.
These changes ensure that the following modules update the screen in
the correct order:
animated cursors (uses RegisterBlockAndWakeupHandlers dynamically)
composite (dynamic wrapping)
misprite (dynamic wrapping)
shadow (static wrapping)
glamor (static wrapping)
driver (static wrapping)
It looks like there's still a bit of confusion between composite and
misprite; if composite updates after misprite, then it's possible
you'd exit the block handler chain with the cursor left hidden. To fix
that, misprite should be wrapping during ScreenInit time and not
unwrapping. And composite might as well join in that fun, just to make
things consistent.
[v2] Unwrap BlockHandler in shadowCloseScreen (ajax)
[v3] ephyr: Use screen block handler for flushing changes
ephyr needs to make sure it calls glXSwapBuffers after glamor finishes
its rendering. As the screen block handler is now called last, we have
to use that instead of a registered block/wakeup handler to make sure
the GL rendering is done before we copy it to the front buffer.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Nice of FreePicture to take a void * instead of a PicturPtr so that
this error wasn't caught by the compiler.
Noticed when resetting the X server left a dangling pixmap around.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Current Mesa Git master checks that the EGL display actually supports
the API passed to eglBindAPI, which can only succeed after
eglInitialize.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96344
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Commit b64108fa ("glamor: Check for composite operations which are
equivalent to copies") failed to copy conditions from exaComposite which
ensure that the composite operation doesn't access outside of the source
picture.
This fixes rendercheck regressions from the commit above.
Reviewed-by: Keith Packard <keithp@keithp.com>
Patch b64108fa30 added a short cut that
identifies composite operations that can be performed with a simple
copy instead.
glamor_copy works in absolute coordinates, so the dx and dy values
passed in need to be converted from drawable-relative to absolute by
adding the drawable x/y values.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
A1 and A8 pixmaps are usually stored in the Red channel to conform
with more recent GL versions. When using these pixmaps as mask values,
that works great. When using these pixmaps as source values, then the
value we want depends on what the destination looks like.
For RGBA or RGB destinations, then we want to use the Red channel
for A values and leave RGB all set to zero.
For A destinations, then we want to leave the R values in the Red
channel so that they end up in the Red channel of the output.
This patch adds a helper function, glamor_bind_texture, which performs
the glBindTexture call along with setting the swizzle parameter
correctly for the Red channel. The swizzle parameter for the Alpha
channel doesn't depend on the destination as it's safe to leave it
always swizzled from the Red channel.
This fixes incorrect rendering in firefox for this page:
https://gfycat.com/HoarseCheapAmericankestrel
while not breaking rendering for this page:
https://feedly.com
v2: Add change accidentally left in patch for missing
glDisable(GL_COLOR_LOGIC_OP).
Found by Emil Velikov <emil.l.velikov@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=63397
Signed-off-by: Keith Packard <keithp@keithp.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
glamor_make_current is supposed to be called before any GL APIs.
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
If the logic op gets left enabled, it overrides the blending
operation, causing incorrect contents on the display.
v2: Disable only on non-ES2, but disable even for PictOpSrc
v3: Found another place this is needed in
glamor_composite_set_shader_blend
v4: Remove change dependent on new glamor_set_composite_texture
API. This belongs in a different patch.
Found by Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Increases x11perf -compwinwin500 numbers by a factor of 10 for me with
radeonsi.
Conditions copied from exaComposite().
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
RENDER requires that sampling outside of any source/mask picture results
in alpha == 0.0.
The OpenGL border colour cannot set alpha = 0.0 if the texture format
doesn't have an alpha channel, so we have to use the RepeatFix handling
in that case.
Also, only force alpha = 1.0 when sampling inside of RGBx source/mask
pictures.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94514
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Some drivers are calling glFinish, they really should be doing this.
This also is needed for some reverse prime scenarios.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
When using PictOpSrc, the destination is wrongly shifted back to (0, 0).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94568
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Olivier Fourdan <ofourdan@redhat.com>
I'm pretty sure Eric suspected this could cause a problem, and we
couldn't find a test. Well loading feedly in firefox seems to trigger
badness that this solves.
bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=94554
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Tested-by: Timo Aaltonen <tjaalton@ubuntu.com>
Without this, we may be manipulating the context of another screen.
In a system with two GPUs using glamor, this fixes lots of
(EE) glamor256: GL error: GL_INVALID_OPERATION in glBindTexture(non-gen name)
spew since 0b4c0c75 ('glamor: Replace "finish access" shader with texture
swizzling').
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
For pictures without alpha, and for most other formats for GLES2, we
would make a temporary FBO, make another temporary texture, upload our
GLAMOR_MEMORY pixmap to the texture, then run the "finish access" shader
across it to swizzle its values around into the temporary FBO (which we
would use for a single Render operation and then throw away).
We can simplify everything by using GL_ARB_texture_swizzle (or its
GLES3 counterpart). It's just not worth the complexity to try to
improve the performance of this already low-performance path (SHM
pixmaps + Render) on GLES2.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
The glamor_pixmap_ensure_fbo() in glamor_pixmap_upload_prepare() will
always fail on a large pixmap, so we can just be explicit about
bailing out here and then dump the rest of this garbage.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
Pixman is quite qualified to allocate our temporary memory, and all we
need to do is decide what formats to convert from and to.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
There was only a pretty special case that could have even worked --
you've got a GLES2 renderer, you've got a SHM pixmap, it's 1555 (not
the usual 565 for 16-bit), and you're little endian (BE was broken,
since GL's 5_5_5_1 picks the 1 bit from the lowest bit of the short,
and on BE we weren't doing the conversion path that swaps around the
channels). This is just not worth the complexity.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
These just smash your 2_10_10_10 data into 8888, despite what the
comments said. That's not valid rendering, so just ditch this path
and fall back to software. One might also note in the code being
removed here that the REVERT_UPLOADING_10_10_10_2 path wasn't even
connected.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
It clarifies what the difference is between the two paths, and would
potentially encourage us to handle GLES extensions that expose
additional types.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Eric Anholt <eric@anholt.net>
