| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
| |
With threaded gallium, the driver may currently be running in another
thread. In that case, we will execute all remaining commands in that
thread instead of syncing, which should be better for cache locality.
Reviewed-by: Andres Rodriguez <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Asynchronous flushes require a proper implementation of
st_server_wait_sync, because we could have the following with
threaded Gallium:
Context 1 app Context 1 driver Context 2
------------- ---------------- ---------
f = glFenceSync
glFlush
<-- app sync --> <-- app sync -->
glWaitSync(f)
.. draw calls ..
pipe_context::flush
for glFenceSync
pipe_context::flush
for glFlush
Reviewed-by: Andres Rodriguez <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Almost all of our BO export paths were already properly marked the BO as
external and added it to the handle table. Most export use-cases go
through a prime fd or flink where we have a brw_bo export helper that
does the right thing. The one missing one happens when you call
queryImage and ask for __DRI_IMAGE_ATTRIB_HANDLE. We just grabbed the
gem handle out of the BO (because it's really easy to do that) and
handed it off to the client; what could go wrong? As it turns out, this
path is used by basically every compositor that wants to turn around and
call drmModeAddFB2 on it so it can hand it off to display. The result,
as of 4b1e70cc57d7ff5f465544644b2180dee1490cee, is that we no longer set
MOCS_PTE on those surfaces and the kernel's attempts to disable caching
fail and we scanout gets corruption.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103759
Fixes: 4b1e70cc57d7ff5f465544644b2180dee1490cee
Reviewed-by: Kenneth Graunke <[email protected]>
Cc: [email protected]
|
|
|
|
|
| |
Reviewed-by: Kenneth Graunke <[email protected]>
Cc: [email protected]
|
|
|
|
|
|
|
| |
This apparently causes hangs on Broadwell, so let's back it out for now.
I think there are other PIPE_CONTROL workarounds that we're missing.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103787
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The old code used an array to store each "instruction group" (the new,
better name than the old overloaded "annotation"), and required a
memmove() to shift elements over in the array when we needed to split a
group so that we could add an error message. This was confusing and
difficult to get right, not the least of which was because the array
has a tail sentinel not included in .ann_count.
Instead use a linked list, a data structure made for efficient
insertion.
Acked-by: Samuel Iglesias Gonsálvez <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
| |
Fixes: 6165fda59b8 ("i965: Program DWord Length in MI_FLUSH_DW")
Cc: <[email protected]>
Signed-off-by: Anuj Phogat <[email protected]>
Reviewed-by: Nanley Chery <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Decorate the params accordingly with "UNUSED".
v2: move UNUSED decoration in front of parameter declaration
Signed-off-by: Gert Wollny <[email protected]>
Reviewed-by: Brian Paul <[email protected]> (v1)
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use the proper enum types for various variables. Makes life in gdb
a little nicer. Note that the size of enum bitfields must be one
larger so the high bit is always zero (for MSVC).
v2: also increase size of image_format bitfield, per Eric Engestrom.
v3: use the new ASSERT_BITFIELD_SIZE() macro
Reviewed-by: Charmaine Lee <[email protected]>
Reviewed-by: Roland Scheidegger <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>
|
|
|
|
|
|
|
| |
There is no need to have these overlap if we support hw atomics.
Reviewed-by: Ilia Mirkin <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We want to emit invariant state at the start of a render batch. In the
past, this more or less happened: a new batch flagged BRW_NEW_CONTEXT
(because we don't have hardware contexts), which triggered the
brw_invariant_state atom. So, it would be emitted before any 3D
drawing. (Technically, there might be some BLT commands in the batch
because Gen4-5 have a single combined render/BLT ring, but that should
be harmless).
With the advent of BLORP, this broke. The first item in a batch might
be a BLORP operation, which bypasses the normal draw upload path. So,
we need to ensure invariant state happens first. To do that, we just
upload it when creating a new batch. On Gen6+ we'd need to worry about
whether it's a RENDER or BLT batch, but because we have a combined ring,
this approach should work fine on Gen4-5.
Seems to fix GPU hangs when playing hardware accelerated video with
mpv -hwdec=vaapi on Ironlake.
Cc: [email protected]
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103529
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Xorg (and possibly other things) depend on this variable to find the
path to DRI drivers.
Signed-off-by: Rafael Antognolli <[email protected]>
Cc: Dylan Baker <[email protected]>
Reviewed-by: Dylan Baker <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
| |
Otherwise we'll bail with due to -Werror=implicit-function-declaration.
It went unnoticed since the we had a bug which did consistently set the
compiler flag.
Fixes: ba8a347f932 ("mesa: split extensions overrides and glGetString(GL_EXTENSIONS)")
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Andres Gomez <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
|
|
|
|
|
|
|
|
| |
Analogous to the glGetString() case - report all the
extensions enabled via MESA_EXTENSION_OVERRIDE
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Store pointers to the tokenized strings in the gl_extensions struct.
This way we can reuse them in glGetStringi() while we construct the
really long string only in _mesa_make_extension_string.
Only 16 pointers/strings are stored for now.
v2: Warn only once when we provide more than 16 unk. extensions, rebase
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]> (v1)
|
|
|
|
|
|
|
| |
Will be needed with next commit
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
If the extra_extensions string is empty there's no need to call
atexit() - there's nothing to free.
v2: Rebase
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]> (v1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The sorting was originally added to work around broken games (comment
says Quake3 demo) that were copying the extensions list into small
buffer.
Sorting does not solve the problem, since we'll still overflow and cause
corruption/crash.
Better workaround is to actually trim the string ... as done with a
later commit which introduces the MESA_EXTENSION_MAX_YEAR env. variable.
Side note: On my machine, the existing sorting makes no changes to the
extensions string.
Cc: Jose Fonseca <[email protected]>
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
| |
We already use it for _mesa_extension_override_enables.
Improve consistency and use it for both extension lists.
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
The function get_extension_override() returns a copy of a string,
only for it to be copied again ...
Drop the unneeded calloc/strdup/free dance.
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
While parsing MESA_EXTENSION_OVERRIDE we keep track of the disabled
extensions, twice - in _mesa_extension_override_disables and
disabled_extensions.
Upon context creation, we use the former to modify the extensions list.
Yet, we still check the updated list against disabled_extensions.
Remove disabled_extensions, it's obsolete.
Cc: Jordan Justen <[email protected]>
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As of previous commit we removed the extension overrides from this
function.
Thus we no longer need to call it during MakeCurrent, so we can
construct the extensions string when needed - _mesa_GetString.
This commit effectively reverts a879d14ecf8 ("mesa: initialize extension
string when context is first bound")
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently we apply the extension overrides and construct the extensions
string upon MakeCurrent.
They are two distinct things, so let's slit the two while pushing the
overrides management _before_ _mesa_compute_version(). This ensures that
the version is updated to reflect the enabled/disabled extensions.
Cc: Jordan Justen <[email protected]>
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Checking the override was useful in the early stages of developing the
extension.
Now that everything is wired, where possible, we can drop the check.
Doing so allows us to simplify some of the related code.
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
Reviewed-by: Jordan Justen <[email protected]>
|
|
|
|
|
|
|
| |
Use the helper over opencoding the check.
Signed-off-by: Emil Velikov <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
...and provide a better citation for the existing one.
v2:
- Apply the workaround to Gen8 too, as intended (caught by Topi).
- Restructure to add bits instead of an extra flush (based on a similar
patch by Rafael Antognolli).
Cc: [email protected]
Reviewed-by: Rafael Antognolli <[email protected]>
|
|
|
|
| |
Trivial.
|
|
|
|
| |
Trivial.
|
|
|
|
|
|
| |
To avoid forward declaration.
Reviewed-By: Gert Wollny <[email protected]>
|
|
|
|
|
|
| |
To get rid of forward declaration.
Reviewed-By: Gert Wollny <[email protected]>
|
|
|
|
|
|
| |
BATCH_RESERVED was deleted in commit 2c46a67b4138631217 (i965: Delete
BATCH_RESERVED handling.) The reserved_space field is dead code, and
the comments aren't useful these days.
|
|
|
|
|
|
|
|
|
|
|
| |
Having this separate could potentially make programs that rebind atomics
but no other surfaces ever so slightly faster. But it's a tiny amount
of code to add to the existing UBO/SSBO atom, and very related.
The extra atoms have a cost on every draw call, and so dropping some of
them would be nice. This also reclaims a dirty bit.
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
| |
We use the same hardware mechanism for both atomic counters and SSBO
atomics, so there's really no benefit to maintaining separate code to
handle each case. Instead, we can just use Rob's shiny new NIR pass to
convert atomic_uints to SSBOs, and delete piles of code.
The ssbo_start section of the binding table becomes a combined ABO and
SSBO section, with ABOs first, then SSBOs.
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
|
|
|
|
|
| |
This fixes the missing AutomaticSize handling in the ABO code, removes
a bunch of duplicated code, and drops an extra layer of wrapping around
brw_emit_buffer_surface_state().
Reviewed-by: Tapani Pälli <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <[email protected]>
|
|
|
|
| |
Reviewed-by: Nicolai Hähnle <[email protected]>
|
|
|
|
|
| |
Signed-off-by: Anuj Phogat <[email protected]>
Cc: Nanley Chery <[email protected]>
|
|
|
|
|
|
|
| |
Number of dwords in MI_FLUSH_DW changed from 4 to 5 in gen8+.
Signed-off-by: Anuj Phogat <[email protected]>
Cc: <[email protected]>
|
|
|
|
|
| |
Signed-off-by: Anuj Phogat <[email protected]>
Cc: <[email protected]>
|
|
|
|
|
| |
Signed-off-by: Andres Rodriguez <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>
|
|
|
|
|
| |
Signed-off-by: Andres Rodriguez <[email protected]>
Reviewed-by: Nicolai Hähnle <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Already implemented for Gallium drivers.
Useful for gbm_bo_(un)map.
Tests:
By porting wayland/weston/clients/simple-dmabuf-drm.c to GBM.
kmscube --mode=rgba
kmscube --mode=nv12-1img
kmscube --mode=nv12-2img
piglit ext_image_dma_buf_import-refcount -auto
piglit ext_image_dma_buf_import-transcode-nv12-as-r8-gr88 -auto
piglit ext_image_dma_buf_import-sample_rgb -fmt=XR24 -alpha-one -auto
piglit ext_image_dma_buf_import-sample_rgb -fmt=AR24 -auto
piglit ext_image_dma_buf_import-sample_yuv -fmt=NV12 -auto
piglit ext_image_dma_buf_import-sample_yuv -fmt=YU12 -auto
piglit ext_image_dma_buf_import-sample_yuv -fmt=YV12 -auto
v2: add early return if (flag & MAP_INTERNAL_MASK)
v3: take input rect into account and test with kmscube and piglit.
v4: handle wraparound and bo reference.
v5: indent, exclude 0 width and height on the boundary, map bo
independently of the image.
Signed-off-by: Julien Isorce <[email protected]>
Reviewed-by: Chris Wilson <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, we just had one hash set for tracking depth and render
caches called brw_context::render_cache. This is less than ideal
because the depth and render caches are separate and we can't track
moves between the depth and the render caches. This limitation led
to some unnecessary flushing around the depth cache. There are cases
(mostly with BLORP) where we can end up touching a depth or stencil
buffer through the render cache. To guard against this, blorp would
unconditionally do a render_cache_set_check_flush on it's destination
which meant that if you did any rendering (including a BLORP operation)
to a given surface and then used it as a blorp destination, you would
end up flushing it out of the render cache before rendering into it.
Things get worse when you dig into the depth/stencil state code for
regular GL draw calls. Because we may end up rendering to a depth
or stencil buffer via BLORP, we did a render_cache_set_check_flush on
all depth and stencil buffers in brw_emit_depthbuffer to ensure that
they got flushed out of the render cache prior to using them for depth
or stencil testing. However, because we also need to track dirtiness
for depth and stencil so that we can implement depth and stencil
texturing correctly, we were adding all depth and stencil buffers to the
render cache set in brw_postdraw_set_buffers_need_resolve. This meant
that, if anything caused 3DSTATE_DEPTH_BUFFER to get re-emitted
(currently _NEW_BUFFERS, BRW_NEW_BATCH, and BRW_NEW_BLORP), we would
almost always do a full pipeline stall and render/depth cache flush.
The root cause of both of these problems is that we can't tell the
difference between the render and depth caches in our tracking. This
commit splits our cache tracking into two sets, one for render and one
for depth, and properly handles transitioning between the two. We still
flush all the caches whenever anything needs to be flushed. The idea is
that if we're going to take the hit of a flush and stall, we may as well
flush everything in the hopes that we can avoid a flush by something
else later.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
| |
Right now we just always flush the destination for render and aren't
particularly careful about depth or stencil. Soon, flush_for_render
isn't going to do the same thing as flush_for_depth and we may be doing
a good deal less depth flushing so we should be a bit more precise.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
| |
In theory, this will let us track the depth and render caches
separately. Right now, they're just wrappers around
brw_render_cache_set_*
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
| |
We may access them as a texture using blorp regardless of whether or not
stencil texturing is enabled.
Reviewed-by: Kenneth Graunke <[email protected]>
Cc: [email protected]
|
|
|
|
| |
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We were already using PTE for all render targets in case one happened to
get scanned out. However, this still wasn't 100% correct because there
are still possibly cases where we may want to texture from an external
buffer even though we don't know the caching mode. This can happen, for
instance, on buffers imported from another GPU via prime.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101691
Cc: "17.3" <[email protected]>
Tested-by: Lyude Paul <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
| |
This makes our MOCS settings significantly more flexible.
Cc: "17.3" <[email protected]>
Tested-by: Lyude Paul <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
| |
Signed-off-by: Rob Clark <[email protected]>
|