GitLab

Available on Gen11+.

v2: Order shading rate in correct order (Samuel)

v3: Move CPS_STATE emission to genX_state.c

v4: Don't override various output structures (Jason)

v5: Rebase on top master (Lionel)

v6: Fix invalid VkPhysicalDeviceFragmentShadingRatePropertiesKHR
    (min|max)FragmentShadingRateAttachmentTexelSize values (Ken)
    Drop #endif comment

v7: Limit extension to Gfx11+ (Lionel)
    Support conservative raster (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Those helpers exist primarily to sort out some of the weirdness around
Gen4-6 dataport access.  On Gen5 and earlier, everything was called
"dataport" and, instead of the SFID we have today there was a "target
cache" parameter in the descriptor.  There are also some bits that moved
around on various gens depending on read vs. write.  Starting with Gen6,
most things which target one of the data cache SFIDs should use
brw_dp_desc() instead.

v2: Drop backward comment (Ken)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

It's a Gen6 XFB thing.  It's never used for anything else so there's no
point in having a target cache switch.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

v2: Update to BITSET_TEST()
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Gfx12.5 has a slightly different code path.

v2: Document the oddness
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

v2: Drop new internal opcodes (Jason)
    Simplify code (Jason)

v3: Add Z computation for coarse pixels

v4: Document things a little
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

v2: Drop old register type initializers (Jason)
    Simplify instruction snippet (Jason)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

v2: Use the new inst->ex_desc field (Jason)

v3: Drop CPS LoD compensation from sampler messages (Lionel)

v4: Drop useless uses_rate_shading (Ken)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

v2: Use new RT descriptor helper
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Render target message descriptors are slightly different from the
dataport ones. In particular the msg_type field is on bits 14:17 for
RT while bits 14:18 for DP.

v2: Drop unused send_commit_msg field in brw_fb_write_desc() (Ken)

v3: Rebase on top renaming (Lionel)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

v2: Add Gen12.5
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <!7455>

Grabbing the fence value in fd_fence_repopulate() without waiting on
fd_submit_fence::ready doesn't work with async flushes, since we are
waiting for the first flush to complete (ie. we don't have the kernel-
side fence value yet).  Just simplify it and make the "repopulated"
fence delagate to the original fence.

Fixes: e9a9ac6f ("freedreno/drm: Async submit support")
Closes: mesa/mesa#4726

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <mesa/mesa!10567>

Don't rely on getting a zero'd out buffer, we could hit the bo-cache.

Fixes: 7dabd624

 ("freedreno/drm: Userspace fences")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <mesa/mesa!10567>

Overwriting entire resource multiple times indicates streaming and in this
case it's more efficient to use linear layout to avoid expensive linear->tiled
conversions.
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <!10572>

In file included from ../src/glx/apple/apple_glx_context.c:49:
../src/glx/glxclient.h:56:10: fatal error: 'loader.h' file not found
         ^~~~~~~~~~

Fixes: 1cb664c1 ("glx: s/dri_message/glx_message/")
Closes: #4702

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Part-of: <!10568>

Since commit f39fd3dc a new GLX error is issued in case context creation
fails. This broke wine on certain hardware: While wine installs an error handler
to ignore this kind of error, it does not function because it expects the
dpy->request serial number of the error to be incremented since the installation
of the handler.

Workaround this by artificially increasing the request number. This also
guarantees a unique serial number for the error.

Fixes: f39fd3dc
Closes: #3969

Signed-off-by: Bastian Beranek <bastian.beischer@rwth-aachen.de>
Part-of: <!10565>

There's no hardware support for anything indirect, so just read the
parameters out.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <!10164>

This requires an address that's safe to read from.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <mesa/mesa!10164>

This allows us to use up to 15 images or buffers (but not both). GL
supports the concept of combined resource maximums though.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

This is a purely software counter alongside the other hardware counters
for ease of use and consistency. However we have to make room for it in
the allocated query space. Use this opportunity to make the nv50 queries
work like the nvc0 ones in terms of space allocation.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <!10164>

This is best-effort for pre-nva0 ... works with a single invocation,
i.e. no locking.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

This handles BUFQ, SUQ, as well as all the various texture types and
formats, driven by data supplied by the driver (and shader itself).

TODO:
 - 2d linear surfaces
 - format via key for writeonly

These will be included in a later change. ES3.1 doesn't require
writeonly, and it's very hard to generate a 2d linear surface.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

These are needed to implement things like imageSize() as well as feed
data into lowering logic for various access types not handled by the
hardware.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Otherwise the final argument doesn't get emitted for CAS in the nv50
emitter.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Evidence suggests that having it anywhere, even as a regular e.g. atom
argument, causes issues.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Acked-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Note that shared memory loads can actually do offsets. The restrictions
vary by generation, this will be added in a later change.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Ideally this should include the size of the inputs as well. This will be
updated when we add support for kernels which take actual inputs.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Pierre Moreau <dev@pmoreau.org>
Part-of: <mesa/mesa!10164>

Let's use the 4 bytes of unused padding usefully in pipe_transfer.
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <mesa/mesa!10527>

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <mesa/mesa!10527>

We could change the type into 16 bits if needed.

PB_USAGE flags need to match PIPE_MAP flags due to static assertions.
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <!10527>

Add spec@arb_pixel_buffer_object@texsubimage cube_map_array pbo to a530
fails for the same reason as spec@arb_texture_cube_map_array@texsubimage cube_map_array
(it is sometimes triggering gpu hangs that cause other flakes).

And remove two a630 xfails that started showing up as UnexpectedPass.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <mesa/mesa!10530>

This one has shown up a couple times since fd/go-fast, I'm still trying
to reproduce/debug.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <mesa/mesa!10530>

pctx->flush_resource() has the same expectations that the resource can
be shared with an external client as pctx->flush(), but without the
convenience of a fence to know *when* the resource must be visible to
that external client.  So we need to ensure the batch is flushed all the
way to the kernel so that implicit-sync can do it's job.

Fixes: e9a9ac6f

 ("freedreno/drm: Async submit support")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <!10530>

This provides the upper layer (gallium, etc) a way to ensure that
rendering involving the bo has been flushed all the way to the kernel.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <mesa/mesa!10530>

Looks like this was unused, and only served to segfault when unbinding
textures.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <!10530>