Because of nir_to_tgsi, LLVM is not required here.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Fixes: d0f8fe5909 ("softpipe: Switch to using NIR as the shader format from mesa/st.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10072>
We can only use 16 OW reads/writes on SLM.
v2: Update comment (Curro)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
BSpec: 47652
Fixes: 369eab9420 ("intel/fs: Emit code for Gen12-HP indirect compute data")
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10082>
since surfaces are cached and deduplicated, we no longer have to
do deep comparisons to determine whether two surfaces are equal and can
just compare the pointers
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10087>
it turns out there's not actually a requirement that resources be unmapped,
which means that a ton of overhead can be saved both in the unmap codepath
(the cpu overhead here is pretty insane) and then also when mapping cached
resource memory, as the map can now be added to the cache for immediate reuse
as seen in radeonsi
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9980>
this was not copied directly and changed the old conditional, which
was intended to catch only non-patch tess io
fixes tess io with legacy builtins, e.g., spec@!opengl 2.0@vertex-program-two-side back front2 back2
Fixes: 2d98efd323 ("zink: pre-populate locations in variables")
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9996>
race conditions here have no impact because it cannot change the evaluation
of this condition
also no need to check zs stuff if there's no update happening
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9961>
similar to amd/radv driver debug modes for sqtt, this specifies a filename
which is checked on every flush(PIPE_FLUSH_END_OF_FRAME); when it exists,
the next frame (and only that frame) is captured into the trace
to use, specify a file with the env var, run your app, and 'touch /path/to/file'
when you want to capture a trace
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10058>
a common usage for hash tables is for tracking exactly one instance of a pointer
for a given period of time, after which the table's entries are purged and it
is reused
this macro enables the purge phase of such usage to reset the table to a
pristine state, avoiding future rehashing due to ballooning of deleted entries
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8498>
a common usage for sets is for tracking exactly one instance of a pointer
for a given period of time, after which the set's entries are purged and it
is reused
this macro enables the purge phase of such usage to reset the table to a
pristine state, avoiding future rehashing due to ballooning of deleted entries
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8498>
if zink+lavapipe is enabled at compile-time, special handling is needed
to ensure that only the intended screen is traced, otherwise malformed
xml will be generated
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10059>
Don't be smarter than state tracker here, of d3d10 wants to do
something the state tracker should hard code that. Since lavapipe
wants to use clip_halfz and depth clipping independently.
This fixes some issues blitting Z that zink was seeing
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10068>
When depth clamp is disabled the viewport values aren't meaningful,
however the value is about to be converted to a unorm so needs
to still be clamped to 0/1.
This might not be the best place for this, maybe it should be in
the write swizzled code.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10068>
The driver now supports DCC stores on GFX10 that means that we can
keep DCC compressed on all layouts/queues.
This should help games which use aync compute or which declare
all images as concurrent like Youngblood.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9850>
If the image is marked as concurrent, each element of
pQueueFamilyIndices must be unique, so queue_family_mask is
different than 1.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9850>
the system_values_read bit might not be set if the value isn't read or is
DCE, but it still needs to be emitted
probably.
fixes#4591
Reviewed-by: Witold Baryluk <witold.baryluk@gmail.com>
Tested-by: Witold Baryluk <witold.baryluk@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10079>