mesa.git
4 years agodocs: add release notes for 20.0.3
Eric Engestrom [Wed, 1 Apr 2020 21:24:57 +0000 (23:24 +0200)]
docs: add release notes for 20.0.3

(cherry picked from commit b04ae1f964c977035d9c8fd4144424387e0d868e)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4413>

4 years agogallium/llvmpipe: add an optimised 32-bit memset
Dave Airlie [Tue, 31 Mar 2020 22:44:08 +0000 (08:44 +1000)]
gallium/llvmpipe: add an optimised 32-bit memset

This might have other users beyond filling/clearing buffers,

increase a fullscreen 4k gears from 68->74 fps on my Ryzen
since gears is really just a clear benchmark, and this helps
clearing.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4394>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4394>

4 years agonir: place aligned members after bitfields in shader_info.tess
Mark Janes [Tue, 31 Mar 2020 23:41:28 +0000 (16:41 -0700)]
nir: place aligned members after bitfields in shader_info.tess

The placement of new shader_info.tess members unnecessarily wastes
space by interspersing 64bit members between bitfields.

Fixes: f1dd81ae104 ("nir: Collect if shader uses cross-invocation or indirect I/O.")
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4408>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4408>

4 years agonir: check shader type before writing to shaderinfo.tess union
Mark Janes [Tue, 31 Mar 2020 23:40:57 +0000 (16:40 -0700)]
nir: check shader type before writing to shaderinfo.tess union

If the shader is not a tesselation shader, then writing to the tess
member of the shaderinfo union will overwrite other members and crash.

Closes: #2722
Fixes: f1dd81ae104 ("nir: Collect if shader uses cross-invocation or indirect I/O.")
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4408>

4 years agoanv: Do not sample from 3d depth image with HiZ
Danylo Piliaiev [Wed, 1 Apr 2020 15:41:03 +0000 (18:41 +0300)]
anv: Do not sample from 3d depth image with HiZ

For Gen8-11, there are some restrictions around sampling from HiZ.
The Skylake PRM docs for RENDER_SURFACE_STATE::AuxiliarySurfaceMode
say:

    "If this field is set to AUX_HIZ, Number of Multisamples must
    be MULTISAMPLECOUNT_1, and Surface Type cannot be SURFTYPE_3D."

Fixes: dEQP-VK.geometry.layered.3d.*.readback
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2720
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Arcady Goldmints-Orlov <agoldmints@igalia.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4409>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4409>

4 years agogallium/swr: Re-enable scratch space for client-memory buffers
Krzysztof Raszkowski [Wed, 1 Apr 2020 15:02:06 +0000 (17:02 +0200)]
gallium/swr: Re-enable scratch space for client-memory buffers

Commit 7d33203b446cdfa11c2aaea18caf05b120a16283 fixed race condition
in freeing scratch memory mechanism but that approach creates
performance regression in some cases. This change revert previous
changes and fix freeing scratch memory mechanism.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4406>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4406>

4 years agogallium/swr: Fix array stride problem.
Krzysztof Raszkowski [Wed, 1 Apr 2020 14:57:20 +0000 (16:57 +0200)]
gallium/swr: Fix array stride problem.

Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4405>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4405>

4 years agoci: Consistently use -j4 across x86 build jobs and -j8 on ARM.
Eric Anholt [Tue, 11 Feb 2020 23:44:56 +0000 (15:44 -0800)]
ci: Consistently use -j4 across x86 build jobs and -j8 on ARM.

Our shared runners are set up for concurrent jobs ~= CPUs / 4 (x86) or 8
(ARM).  If you use more build processes than that, then jobs may be
fighting each other for shared system resources, possibly to the point of
failure (we've seen one of the runners OOM on some jobs before, though I'm
not sure if this was the cause).

To try to systematically prevent the problem, we make a ninja wrapper in
the containers that passes the -j flags, and set MAKEFLAGS in the
container builds.  This doesn't cover make in non-container builds, but I
believe we don't have any of those.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3782>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3782>

4 years agoaco: only break SMEM clauses if XNACK is enabled (mostly APUs)
Samuel Pitoiset [Fri, 27 Mar 2020 14:16:39 +0000 (15:16 +0100)]
aco: only break SMEM clauses if XNACK is enabled (mostly APUs)

According to LLVM, it seems only required for APUs like RAVEN, but
we still ensure that SMEM stores are in their own clause.

pipeline-db (VEGA10):
Totals from affected shaders:
SGPRS: 1775364 -> 1775364 (0.00 %)
VGPRS: 1287176 -> 1287176 (0.00 %)
Spilled SGPRs: 725 -> 725 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 65386620 -> 65107460 (-0.43 %) bytes
Max Waves: 287099 -> 287099 (0.00 %)

pipeline-db (POLARIS10):
Totals from affected shaders:
SGPRS: 1797743 -> 1797743 (0.00 %)
VGPRS: 1271108 -> 1271108 (0.00 %)
Spilled SGPRs: 730 -> 730 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Code Size: 64046244 -> 63782324 (-0.41 %) bytes
Max Waves: 254875 -> 254875 (0.00 %)

This only affects GFX6-GFX9 chips because the compiler uses a
different pass for GFX10.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4349>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4349>

4 years agoRevert "spirv: Implement OpCopyObject and OpCopyLogical as blind copies"
Jason Ekstrand [Wed, 1 Apr 2020 17:40:34 +0000 (12:40 -0500)]
Revert "spirv: Implement OpCopyObject and OpCopyLogical as blind copies"

This reverts commit 7a53e67816ed9baf7d825ed60ee59f0c05f9df48.

4 years agoloader: fallback to kernel name, if PCI fails
Emil Velikov [Thu, 5 Mar 2020 14:41:25 +0000 (14:41 +0000)]
loader: fallback to kernel name, if PCI fails

Currently, if the PCI machinery fails, we return a NULL driver name.
In the past this has resulted in various workarounds.

To avoid those, fallback to loader_get_kernel_driver_name(). It's not
perfect, yet perfectly reasonable.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoloader: move "using driver..." message to loader_get_kernel_driver_name
Emil Velikov [Thu, 5 Mar 2020 14:30:51 +0000 (14:30 +0000)]
loader: move "using driver..." message to  loader_get_kernel_driver_name

Move the message to the function which fetches the name.

While here use the same DEBUG/WARNING approach like in the PCI case. The
current method spam a tad much, plus isn't consistent.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoloader: simplify codeflow in drm_get_pci_id_for_fd
Emil Velikov [Thu, 5 Mar 2020 13:50:46 +0000 (13:50 +0000)]
loader: simplify codeflow in drm_get_pci_id_for_fd

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoloader: simplify loader_get_user_preferred_fd()
Emil Velikov [Tue, 19 Feb 2019 15:30:40 +0000 (15:30 +0000)]
loader: simplify loader_get_user_preferred_fd()

Reoder the function a bit to make the code-flow more obvious and short.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoloader: use a maximum of 64 drmDevices
Emil Velikov [Tue, 19 Feb 2019 15:30:38 +0000 (15:30 +0000)]
loader: use a maximum of 64 drmDevices

Currently that's the hard-coded maximum in the kernel, even though the
libdrm API allows for more. Latter is done with extendability in mind.

Allocate 64 pointers^Wdevices on stack for now. Making for shorter and
ever-so-slightly faster code.

v2: Use single MAX_DRM_DEVICES #define (Eric)

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com> (v1)
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoRevert "egl/dri2: Don't dlclose() the driver on dri2_load_driver_common failure"
Emil Velikov [Thu, 5 Mar 2020 13:05:36 +0000 (13:05 +0000)]
Revert "egl/dri2: Don't dlclose() the driver on dri2_load_driver_common failure"

This reverts commit 1b87f4058de84d7a0bb4ead0c4f4b024d4cce8fb.

dlclose() of the handle is perfectly reasonable, a follow-up NULL
assignment is missing.

As-is this causes a leak for nearly every platform, since they call
dri2_load_driver* initially, followed by a second swrast fallback call.

Some platforms even loop through the existing drivers probing.

Revert the commit and add the NULL check.

Fixes: 1b87f4058de ("egl/dri2: Don't dlclose() the driver on dri2_load_driver_common failure")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoegl/drm: reinstate (kms_)swrast support
Emil Velikov [Wed, 4 Mar 2020 18:33:09 +0000 (18:33 +0000)]
egl/drm: reinstate (kms_)swrast support

With earlier commit we've added a generic LIBGL_ALWAYS_SOFTWARE handling
yet did not consider that the existing codebase unconditionally errors
out when set. That was fixed with a latter commit, while the fix itself
added erroneous restriction for egl/drm.

As mentioned in the report - the feature was working for ages. It was a
Gnome developer who added kms_swrast support for gbm in the first place.

Admittedly kms_swrast is somewhat in the middle between traditional
swrast and HW drivers, regardless - reinstate support.

Fixes: 47273d7312c ("egl: set UseFallback if LIBGL_ALWAYS_SOFTWARE is set")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/165
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agoglx: set the loader_logger early and for everyone
Emil Velikov [Wed, 4 Mar 2020 17:52:04 +0000 (17:52 +0000)]
glx: set the loader_logger early and for everyone

Currently we set the logger only for DRI3. Even though it's used nearly
everywhere. For platforms where we don't the function is effectively a
no-op.

With this in place, LIBGL_DEBUG=verbose works across the board.

Fixes: d971a4230d5 ("loader: Factor out the common driver opening logic from each loader.")
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agomeson: glx: drop with_glx == dri check
Emil Velikov [Wed, 4 Mar 2020 17:46:54 +0000 (17:46 +0000)]
meson: glx: drop with_glx == dri check

We can get into src/glx only with with_glx == dri. Thus there's no point
in the secondary, nested, check - it's always true.

Cc: Dylan Baker <dylan@pnwbakers>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4084>

4 years agomesa/main: remove unused macro
Erik Faye-Lund [Mon, 25 Feb 2019 12:30:30 +0000 (13:30 +0100)]
mesa/main: remove unused macro

This macro is no longer used, so let's get rid of it.

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_TEXTURE_EXTERNAL
Erik Faye-Lund [Mon, 25 Feb 2019 12:20:35 +0000 (13:20 +0100)]
mesa/main: clean up extension-check for GL_TEXTURE_EXTERNAL

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_RASTERIZER_DISCARD
Erik Faye-Lund [Mon, 25 Feb 2019 12:01:58 +0000 (13:01 +0100)]
mesa/main: clean up extension-check for GL_RASTERIZER_DISCARD

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_TEXTURE_CUBE_MAP_SEAMLESS
Erik Faye-Lund [Mon, 25 Feb 2019 12:00:47 +0000 (13:00 +0100)]
mesa/main: clean up extension-check for GL_TEXTURE_CUBE_MAP_SEAMLESS

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_FRAGMENT_SHADER_ATI
Erik Faye-Lund [Mon, 25 Feb 2019 12:00:07 +0000 (13:00 +0100)]
mesa/main: clean up extension-check for GL_FRAGMENT_SHADER_ATI

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for AMD_depth_clamp_separate
Erik Faye-Lund [Mon, 25 Feb 2019 11:58:35 +0000 (12:58 +0100)]
mesa/main: clean up extension-check for AMD_depth_clamp_separate

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_DEPTH_BOUNDS_TEST
Erik Faye-Lund [Mon, 25 Feb 2019 11:57:22 +0000 (12:57 +0100)]
mesa/main: clean up extension-check for GL_DEPTH_BOUNDS_TEST

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_STENCIL_TEST_TWO_SIDE
Erik Faye-Lund [Mon, 25 Feb 2019 11:54:13 +0000 (12:54 +0100)]
mesa/main: clean up extension-check for GL_STENCIL_TEST_TWO_SIDE

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_TEXTURE_RECTANGLE
Erik Faye-Lund [Mon, 25 Feb 2019 11:52:37 +0000 (12:52 +0100)]
mesa/main: clean up extension-check for GL_TEXTURE_RECTANGLE

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_VERTEX_PROGRAM_POINT_SIZE
Erik Faye-Lund [Mon, 25 Feb 2019 11:50:08 +0000 (12:50 +0100)]
mesa/main: clean up extension-check for GL_VERTEX_PROGRAM_POINT_SIZE

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_VERTEX_PROGRAM_TWO_SIDE
Erik Faye-Lund [Mon, 25 Feb 2019 11:44:55 +0000 (12:44 +0100)]
mesa/main: clean up extension-check for GL_VERTEX_PROGRAM_TWO_SIDE

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean up extension-check for GL_VERTEX_PROGRAM
Erik Faye-Lund [Mon, 25 Feb 2019 11:25:04 +0000 (12:25 +0100)]
mesa/main: clean up extension-check for GL_VERTEX_PROGRAM

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: clean-up extension-checks for point-sprites
Erik Faye-Lund [Mon, 25 Feb 2019 10:52:59 +0000 (11:52 +0100)]
mesa/main: clean-up extension-checks for point-sprites

This is the only user of the CHECK_EXTENSION2 macro, so let's remove
that while we're at it.

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agomesa/main: correct extension-checks for GL_BLACKHOLE_RENDER_INTEL
Erik Faye-Lund [Wed, 19 Feb 2020 11:12:27 +0000 (12:12 +0100)]
mesa/main: correct extension-checks for GL_BLACKHOLE_RENDER_INTEL

KHR_blend_equation_advanced_coherent isn't exposed on OpenGL ES 1.x
nor OpenGL versions prior to 30, so we shouldn't allow to query its
enum-states there either.

Fixes: 74ec39f66d5 ("mesa: add INTEL_blackhole_render")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/329>

4 years agoloader: Warn when we fail to open a device node due to permissions.
Eric Anholt [Wed, 5 Feb 2020 01:31:28 +0000 (17:31 -0800)]
loader: Warn when we fail to open a device node due to permissions.

This is definitely not the first time I've debugged why I'm getting swrast
on a device only to find out I'm not a member of the render node's group.

This does mean that you'll get a warning print even without EGL_LOG_LEVEL
set.  This may be an issue if we expect people outside of the DRI node's
group to actually be using swrast instead of getting their permissions
fixed.  Right now surfaceless throws a "libEGL warning: No hardware driver
found, falling back to software rendering" in that case anyway, so this is
just more informative.

Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3703>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3703>

4 years agosvga: Treat forced coherent maps as maps of persistent memory
Thomas Hellstrom [Tue, 31 Mar 2020 07:10:17 +0000 (09:10 +0200)]
svga: Treat forced coherent maps as maps of persistent memory

A previous commit made sure we sent a BindGBSurface command at map time
rather than at unmap time for persistent memory. To be consistent, do the
same for forced coherent maps. This makes it possible to avoid the
explicit UpdateGBSurface at unmap time for discard maps and to instead rely
on the kernel's dirty-tracking mechanism at the cost of an additional flush.

Tested with SVGA_FORCE_COHERENT=1, piglit run quick. No regressions.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4399>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4399>

4 years agosvga, winsys/svga: Fix persistent memory discard maps
Thomas Hellstrom [Sat, 28 Mar 2020 17:02:25 +0000 (18:02 +0100)]
svga, winsys/svga: Fix persistent memory discard maps

The kernel driver requires immediate notification using a
BindGBSurface command when a graphics coherent memory resource changes
backing MOB, so that it can start dirty-tracking the new MOB.
Since we always use graphics coherent memory for persistent memory, enqueue
and flush a BindGBSurface commmand at map time rather than at unmap time.
Since we're dealing with persistent memory, It's OK to flush while mapped.

This fixes an issue with gnome-shell / Wayland which uses persistent
memory together with discard maps when we advertise ARB_buffer_storage.
XWayland clients will render incorrectly.

Fixes: 71b43490dd ("svga: Support ARB_buffer_storage")
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Neha Bhende <bhenden@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4399>

4 years agopan/bi: Fix outmod/roundmode flip
Alyssa Rosenzweig [Wed, 1 Apr 2020 02:18:03 +0000 (22:18 -0400)]
pan/bi: Fix outmod/roundmode flip

I misread the disassembler, the fields are in the other order.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bi: Handle fmov class ops
Alyssa Rosenzweig [Wed, 1 Apr 2020 02:16:55 +0000 (22:16 -0400)]
pan/bi: Handle fmov class ops

We need to lower them to something reasonable (ideally, the modifier
would be attached but we need to do something for the case it's not). We
specifically have to lower pre-sched as well, but we can do the lower
literally at schedule time for now (if this proves annoying, we can move
it earlier, but I want to leave room for modifier-aware copyprop should
that prove interesting).

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bi: Fix unused port swapping
Alyssa Rosenzweig [Tue, 31 Mar 2020 17:37:12 +0000 (13:37 -0400)]
pan/bi: Fix unused port swapping

Fixes INSTR_INVALID_ENC

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bi: Add cmdline option for verbose disassembly
Alyssa Rosenzweig [Tue, 31 Mar 2020 17:08:16 +0000 (13:08 -0400)]
pan/bi: Add cmdline option for verbose disassembly

Useful for debugging packing.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bi: Don't set the back-to-back bit yet
Alyssa Rosenzweig [Tue, 31 Mar 2020 17:05:02 +0000 (13:05 -0400)]
pan/bi: Don't set the back-to-back bit yet

This is bad for performance but we can't assume it's true without some
analysis, which we presently don't do. Leave it for future work and
don't break the present.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bi: Use STAGE srcs for scheduler nops
Alyssa Rosenzweig [Tue, 31 Mar 2020 17:04:18 +0000 (13:04 -0400)]
pan/bi: Use STAGE srcs for scheduler nops

..rather than using port 0 for the source, which may or may not actually
exist.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bi: Fix writes_component for VECTOR
Alyssa Rosenzweig [Tue, 31 Mar 2020 16:55:31 +0000 (12:55 -0400)]
pan/bi: Fix writes_component for VECTOR

I'm not convinced this is the best way and it's sort of a hack, but it
fixes RA for st_vary.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bit: Wire through I/O
Alyssa Rosenzweig [Tue, 31 Mar 2020 16:20:18 +0000 (12:20 -0400)]
pan/bit: Wire through I/O

We'd like to wire in attributes and uniforms as inputs and look at the
varying as output for automatic testing on-device, building up a test
framework for us.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agopan/bit: Add `run` mode to the cmdline
Alyssa Rosenzweig [Tue, 31 Mar 2020 01:21:49 +0000 (21:21 -0400)]
pan/bit: Add `run` mode to the cmdline

This emulates the functionality of shader_runner (built for kbase) using
the bifrost testing infrastructure so it runs on mainline. Ideally this
will let us test shaders from the assembler.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4396>

4 years agoappveyor: Remove Meson job.
Jose Fonseca [Sat, 28 Mar 2020 10:42:58 +0000 (10:42 +0000)]
appveyor: Remove Meson job.

Appveyor Meson fails misteriously some times, e.g.,
- https://ci.appveyor.com/project/mesa3d/mesa/builds/31780753/job/w8b28iahboxq4na2
- https://ci.appveyor.com/project/mesa3d/mesa/builds/31857376
and now that we have msvc coverage on gitlab ci this is no longer necessary.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4392>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4392>

4 years agofreedreno/log: fix build error
Rob Clark [Tue, 31 Mar 2020 15:31:00 +0000 (08:31 -0700)]
freedreno/log: fix build error

It seems some versions of gcc are less clever about const initializers:

```
../src/gallium/drivers/freedreno/freedreno_log.c:58:33: error: initializer element is not constant
 const unsigned msgs_per_chunk = bo_size / sizeof(uint64_t);
                                 ^~~~~~~
```

See https://gitlab.freedesktop.org/mesa/mesa/-/issues/2713

Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4390>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4390>

4 years agonir/algebraic: Remove a redundant fabs pattern
Ian Romanick [Tue, 24 Mar 2020 18:18:02 +0000 (11:18 -0700)]
nir/algebraic: Remove a redundant fabs pattern

Made redundant by 5544b2cbbd2 ("nir/algebraic: Use value range analysis
to eliminate useless unary ops").

No shader-db changes on any Intel platform.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>

4 years agonir/algebraic: Use value range analysis to convert fmax to fsat
Ian Romanick [Tue, 11 Feb 2020 02:23:54 +0000 (18:23 -0800)]
nir/algebraic: Use value range analysis to convert fmax to fsat

This is conceptually similar to the 1-fsat(a) <=> fsat(1-a) rearragement
done in:

3b747909419 ("nir/algebraic: Recognize open-coded flrp(a, b, fsat(c))")

2d259713b7 ("nir/algebraic: Commute 1-fsat(a) to fsat(1-a) for all
non-fmul instructions").

Note: this helps the Aztex Ruins shader that was hurt for spills and
fills on Braodwell in the previous commit, but it does not fix the
spills or fills. :(

All Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 14528985 -> 14526116 (-0.02%)
instructions in affected programs: 477300 -> 474431 (-0.60%)
helped: 2332
HURT: 0
helped stats (abs) min: 1 max: 18 x̄: 1.23 x̃: 1
helped stats (rel) min: 0.07% max: 8.89% x̄: 0.88% x̃: 0.64%
95% mean confidence interval for instructions value: -1.27 -1.19
95% mean confidence interval for instructions %-change: -0.92% -0.85%
Instructions are helped.

total cycles in shared programs: 203723684 -> 203692984 (-0.02%)
cycles in affected programs: 4878847 -> 4848147 (-0.63%)
helped: 1764
HURT: 324
helped stats (abs) min: 1 max: 706 x̄: 22.94 x̃: 17
helped stats (rel) min: <.01% max: 17.75% x̄: 1.94% x̃: 1.66%
HURT stats (abs)   min: 1 max: 400 x̄: 30.15 x̃: 10
HURT stats (rel)   min: <.01% max: 17.76% x̄: 1.91% x̃: 0.69%
95% mean confidence interval for cycles value: -16.55 -12.86
95% mean confidence interval for cycles %-change: -1.44% -1.24%
Cycles are helped.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>

4 years agonir/algebraic: Distribute source modifiers into instructions
Ian Romanick [Mon, 15 Jul 2019 22:55:00 +0000 (15:55 -0700)]
nir/algebraic: Distribute source modifiers into instructions

There are three main classes of cases that are helped by this change:

1. When the negation is applied to a value being type converted (e.g.,
   float(-x)).  This could possibly also be handled with more clever
   code generation.

2. When the negation is applied to a phi node source (e.g., x = -(...);
   at the end of a basic block).  This was the original case that caught
   my attention while looking at shader-db dumps.

3. When the negation is applied to the source of an instruction that
   cannot have source modifiers.  This includes texture instructions and
   math box instructions on pre-Gen7 platforms (see more details below).

In many these cases the negation can be propagated into the instructions
that generate the value (e.g., -(a*b) = (-a)*b).

In addition to the operations implemtned in this patch, I also tried:

 - frcp - Helped 6 or fewer shaders on Gen7+, and hurt just as many on
   pre-Gen7.  On Gen6 and earlier, frcp is a math box instruction, and
   math box instructions cannot have source modifiers.

   I suspect this is why so many more shaders are helped on Gen6 than on
   Gen5 or Gen7.  Gen6 supports OpenGL 3.3, so a lot more shaders
   compile on it.  A lot of these shaders may have things like cos(-x)
   or rcp(-x) that could result in an explicit negation instruction.

 - bcsel - Hurt a few shaders with none helped.  bcsel operates on
   integer sources, so the fabs or fneg cannot be a source modifier in
   the bcsel itself.

 - Integer instructions - No changes on any Intel platform.

Some notes about the shader-db results below.

 - On Tiger Lake, a single Deus Ex fragment shader is hurt for both
   spills and fills.

 - On Haswell, a different Deus Ex fragment shader is hurt for both
   spills and fills.

 - On GM45, the "LOST: 1" and "GAINED: 1" is a single Left4Dead 2
   (very high graphics settings, lol) fragment shader that upgrades
   from SIMD8 to SIMD16.

v2: Add support for fsign.  Add some patterns that remove redundant
negations and redundant absolute value rather than trying to push them
down the tree.

Tiger Lake
total instructions in shared programs: 17611333 -> 17586465 (-0.14%)
instructions in affected programs: 3033734 -> 3008866 (-0.82%)
helped: 10310
HURT: 632
helped stats (abs) min: 1 max: 35 x̄: 2.61 x̃: 1
helped stats (rel) min: 0.04% max: 16.67% x̄: 1.43% x̃: 1.01%
HURT stats (abs)   min: 1 max: 47 x̄: 3.21 x̃: 2
HURT stats (rel)   min: 0.04% max: 5.08% x̄: 0.88% x̃: 0.63%
95% mean confidence interval for instructions value: -2.33 -2.21
95% mean confidence interval for instructions %-change: -1.32% -1.27%
Instructions are helped.

total cycles in shared programs: 338365223 -> 338262252 (-0.03%)
cycles in affected programs: 125291811 -> 125188840 (-0.08%)
helped: 5224
HURT: 2031
helped stats (abs) min: 1 max: 5670 x̄: 46.73 x̃: 12
helped stats (rel) min: <.01% max: 34.78% x̄: 1.91% x̃: 0.97%
HURT stats (abs)   min: 1 max: 2882 x̄: 69.50 x̃: 14
HURT stats (rel)   min: <.01% max: 44.93% x̄: 2.35% x̃: 0.74%
95% mean confidence interval for cycles value: -18.71 -9.68
95% mean confidence interval for cycles %-change: -0.80% -0.63%
Cycles are helped.

total spills in shared programs: 8942 -> 8946 (0.04%)
spills in affected programs: 8 -> 12 (50.00%)
helped: 0
HURT: 1

total fills in shared programs: 9399 -> 9401 (0.02%)
fills in affected programs: 21 -> 23 (9.52%)
helped: 0
HURT: 1

Ice Lake
total instructions in shared programs: 16124348 -> 16102258 (-0.14%)
instructions in affected programs: 2830928 -> 2808838 (-0.78%)
helped: 11294
HURT: 2
helped stats (abs) min: 1 max: 12 x̄: 1.96 x̃: 1
helped stats (rel) min: 0.07% max: 17.65% x̄: 1.32% x̃: 0.93%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 3.45% max: 4.00% x̄: 3.72% x̃: 3.72%
95% mean confidence interval for instructions value: -1.99 -1.93
95% mean confidence interval for instructions %-change: -1.34% -1.29%
Instructions are helped.

total cycles in shared programs: 335393932 -> 335325794 (-0.02%)
cycles in affected programs: 123834609 -> 123766471 (-0.06%)
helped: 5034
HURT: 2128
helped stats (abs) min: 1 max: 3256 x̄: 43.39 x̃: 11
helped stats (rel) min: <.01% max: 35.79% x̄: 1.98% x̃: 1.00%
HURT stats (abs)   min: 1 max: 2634 x̄: 70.63 x̃: 16
HURT stats (rel)   min: <.01% max: 49.49% x̄: 2.73% x̃: 0.62%
95% mean confidence interval for cycles value: -13.66 -5.37
95% mean confidence interval for cycles %-change: -0.69% -0.48%
Cycles are helped.

LOST:   0
GAINED: 2

Skylake
total instructions in shared programs: 14949240 -> 14927930 (-0.14%)
instructions in affected programs: 2594756 -> 2573446 (-0.82%)
helped: 11000
HURT: 2
helped stats (abs) min: 1 max: 12 x̄: 1.94 x̃: 1
helped stats (rel) min: 0.07% max: 18.75% x̄: 1.39% x̃: 0.94%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 4.76% max: 4.76% x̄: 4.76% x̃: 4.76%
95% mean confidence interval for instructions value: -1.97 -1.91
95% mean confidence interval for instructions %-change: -1.42% -1.37%
Instructions are helped.

total cycles in shared programs: 324829346 -> 324821596 (<.01%)
cycles in affected programs: 121566087 -> 121558337 (<.01%)
helped: 4611
HURT: 2147
helped stats (abs) min: 1 max: 3715 x̄: 33.29 x̃: 10
helped stats (rel) min: <.01% max: 36.08% x̄: 1.94% x̃: 1.00%
HURT stats (abs)   min: 1 max: 2551 x̄: 67.88 x̃: 16
HURT stats (rel)   min: <.01% max: 53.79% x̄: 3.69% x̃: 0.89%
95% mean confidence interval for cycles value: -4.25 1.96
95% mean confidence interval for cycles %-change: -0.28% -0.02%
Inconclusive result (value mean confidence interval includes 0).

Broadwell
total instructions in shared programs: 14971203 -> 14949957 (-0.14%)
instructions in affected programs: 2635699 -> 2614453 (-0.81%)
helped: 10982
HURT: 2
helped stats (abs) min: 1 max: 12 x̄: 1.93 x̃: 1
helped stats (rel) min: 0.07% max: 18.75% x̄: 1.39% x̃: 0.94%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 4.76% max: 4.76% x̄: 4.76% x̃: 4.76%
95% mean confidence interval for instructions value: -1.97 -1.90
95% mean confidence interval for instructions %-change: -1.42% -1.37%
Instructions are helped.

total cycles in shared programs: 336215033 -> 336086458 (-0.04%)
cycles in affected programs: 127383198 -> 127254623 (-0.10%)
helped: 4884
HURT: 1963
helped stats (abs) min: 1 max: 25696 x̄: 51.78 x̃: 12
helped stats (rel) min: <.01% max: 58.28% x̄: 2.00% x̃: 1.05%
HURT stats (abs)   min: 1 max: 3401 x̄: 63.33 x̃: 16
HURT stats (rel)   min: <.01% max: 39.95% x̄: 2.20% x̃: 0.70%
95% mean confidence interval for cycles value: -29.99 -7.57
95% mean confidence interval for cycles %-change: -0.89% -0.71%
Cycles are helped.

total fills in shared programs: 24905 -> 24901 (-0.02%)
fills in affected programs: 117 -> 113 (-3.42%)
helped: 4
HURT: 0

LOST:   0
GAINED: 16

Haswell
total instructions in shared programs: 13148927 -> 13131528 (-0.13%)
instructions in affected programs: 2220941 -> 2203542 (-0.78%)
helped: 8017
HURT: 4
helped stats (abs) min: 1 max: 12 x̄: 2.17 x̃: 1
helped stats (rel) min: 0.07% max: 15.25% x̄: 1.40% x̃: 0.93%
HURT stats (abs)   min: 1 max: 7 x̄: 2.50 x̃: 1
HURT stats (rel)   min: 0.33% max: 4.76% x̄: 2.73% x̃: 2.91%
95% mean confidence interval for instructions value: -2.21 -2.13
95% mean confidence interval for instructions %-change: -1.43% -1.37%
Instructions are helped.

total cycles in shared programs: 321221791 -> 321079870 (-0.04%)
cycles in affected programs: 126886055 -> 126744134 (-0.11%)
helped: 4674
HURT: 1729
helped stats (abs) min: 1 max: 23654 x̄: 56.47 x̃: 16
helped stats (rel) min: <.01% max: 53.22% x̄: 2.13% x̃: 1.05%
HURT stats (abs)   min: 1 max: 3694 x̄: 70.58 x̃: 18
HURT stats (rel)   min: <.01% max: 63.06% x̄: 2.48% x̃: 0.90%
95% mean confidence interval for cycles value: -33.31 -11.02
95% mean confidence interval for cycles %-change: -0.99% -0.78%
Cycles are helped.

total spills in shared programs: 19872 -> 19874 (0.01%)
spills in affected programs: 21 -> 23 (9.52%)
helped: 0
HURT: 1

total fills in shared programs: 20941 -> 20941 (0.00%)
fills in affected programs: 62 -> 62 (0.00%)
helped: 1
HURT: 1

LOST:   0
GAINED: 8

Ivy Bridge
total instructions in shared programs: 11875553 -> 11853839 (-0.18%)
instructions in affected programs: 1553112 -> 1531398 (-1.40%)
helped: 7304
HURT: 3
helped stats (abs) min: 1 max: 16 x̄: 2.97 x̃: 2
helped stats (rel) min: 0.07% max: 15.25% x̄: 1.62% x̃: 1.15%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 1.05% max: 3.33% x̄: 2.44% x̃: 2.94%
95% mean confidence interval for instructions value: -3.04 -2.90
95% mean confidence interval for instructions %-change: -1.65% -1.59%
Instructions are helped.

total cycles in shared programs: 178246425 -> 178184484 (-0.03%)
cycles in affected programs: 13702146 -> 13640205 (-0.45%)
helped: 4409
HURT: 1566
helped stats (abs) min: 1 max: 531 x̄: 24.52 x̃: 13
helped stats (rel) min: <.01% max: 38.67% x̄: 2.14% x̃: 1.02%
HURT stats (abs)   min: 1 max: 356 x̄: 29.48 x̃: 10
HURT stats (rel)   min: <.01% max: 64.73% x̄: 1.87% x̃: 0.70%
95% mean confidence interval for cycles value: -11.60 -9.14
95% mean confidence interval for cycles %-change: -1.19% -0.99%
Cycles are helped.

LOST:   0
GAINED: 10

Sandy Bridge
total instructions in shared programs: 10695740 -> 10667483 (-0.26%)
instructions in affected programs: 2337607 -> 2309350 (-1.21%)
helped: 10720
HURT: 1
helped stats (abs) min: 1 max: 49 x̄: 2.64 x̃: 2
helped stats (rel) min: 0.07% max: 20.00% x̄: 1.54% x̃: 1.13%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 1.04% max: 1.04% x̄: 1.04% x̃: 1.04%
95% mean confidence interval for instructions value: -2.69 -2.58
95% mean confidence interval for instructions %-change: -1.57% -1.51%
Instructions are helped.

total cycles in shared programs: 153478839 -> 153416223 (-0.04%)
cycles in affected programs: 22050900 -> 21988284 (-0.28%)
helped: 5342
HURT: 2200
helped stats (abs) min: 1 max: 1020 x̄: 20.34 x̃: 16
helped stats (rel) min: <.01% max: 24.05% x̄: 1.51% x̃: 0.86%
HURT stats (abs)   min: 1 max: 335 x̄: 20.93 x̃: 6
HURT stats (rel)   min: <.01% max: 20.18% x̄: 1.03% x̃: 0.30%
95% mean confidence interval for cycles value: -9.18 -7.42
95% mean confidence interval for cycles %-change: -0.82% -0.71%
Cycles are helped.

Iron Lake
total instructions in shared programs: 8114882 -> 8105574 (-0.11%)
instructions in affected programs: 1232504 -> 1223196 (-0.76%)
helped: 4109
HURT: 2
helped stats (abs) min: 1 max: 6 x̄: 2.27 x̃: 1
helped stats (rel) min: 0.05% max: 8.33% x̄: 0.99% x̃: 0.66%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 0.94% max: 4.35% x̄: 2.65% x̃: 2.65%
95% mean confidence interval for instructions value: -2.31 -2.21
95% mean confidence interval for instructions %-change: -1.01% -0.96%
Instructions are helped.

total cycles in shared programs: 188504036 -> 188466296 (-0.02%)
cycles in affected programs: 31203798 -> 31166058 (-0.12%)
helped: 3447
HURT: 36
helped stats (abs) min: 2 max: 92 x̄: 11.03 x̃: 8
helped stats (rel) min: <.01% max: 5.41% x̄: 0.21% x̃: 0.13%
HURT stats (abs)   min: 2 max: 30 x̄: 7.33 x̃: 6
HURT stats (rel)   min: 0.01% max: 1.65% x̄: 0.18% x̃: 0.10%
95% mean confidence interval for cycles value: -11.16 -10.51
95% mean confidence interval for cycles %-change: -0.22% -0.20%
Cycles are helped.

LOST:   0
GAINED: 1

GM45
total instructions in shared programs: 4989697 -> 4984531 (-0.10%)
instructions in affected programs: 703952 -> 698786 (-0.73%)
helped: 2493
HURT: 2
helped stats (abs) min: 1 max: 6 x̄: 2.07 x̃: 1
helped stats (rel) min: 0.05% max: 8.33% x̄: 1.03% x̃: 0.66%
HURT stats (abs)   min: 1 max: 1 x̄: 1.00 x̃: 1
HURT stats (rel)   min: 0.95% max: 4.35% x̄: 2.65% x̃: 2.65%
95% mean confidence interval for instructions value: -2.13 -2.01
95% mean confidence interval for instructions %-change: -1.07% -0.99%
Instructions are helped.

total cycles in shared programs: 128929136 -> 128903886 (-0.02%)
cycles in affected programs: 21583096 -> 21557846 (-0.12%)
helped: 2214
HURT: 17
helped stats (abs) min: 2 max: 92 x̄: 11.44 x̃: 8
helped stats (rel) min: <.01% max: 5.41% x̄: 0.24% x̃: 0.13%
HURT stats (abs)   min: 2 max: 8 x̄: 4.24 x̃: 4
HURT stats (rel)   min: 0.01% max: 1.65% x̄: 0.20% x̃: 0.09%
95% mean confidence interval for cycles value: -11.75 -10.88
95% mean confidence interval for cycles %-change: -0.25% -0.22%
Cycles are helped.

LOST:   1
GAINED: 1

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>

4 years agonir/algebraic: Change the default cursor location when replacing a unary op
Ian Romanick [Mon, 23 Mar 2020 21:25:24 +0000 (14:25 -0700)]
nir/algebraic: Change the default cursor location when replacing a unary op

If the expression tree that is being replaced has a unary operation at
its root, set the cursor (location where new instructions are inserted)
at the source instruction instead.

This doesn't do much now because there are very few patterns that have a
unary operation as the root.  Almost all of the patterns that do have a
unary operation as the root have inot.  All of the shaders that are
affected by this commit have expression trees with an inot at the root.

This change prevents some significant, spurious caused by the next
commit.  There is further explanation in the large comment added in
the code.

I also considered a couple other options that may still be worth exploring.

1. Add some mark-up to the search pattern to denote where new
   instructions should be added.  I considered using "@" to denote the
   cursor location.  For example,

    (('fneg', ('fadd@', a, b)), ...)

2. To prevent other kinds of unintended code motion, add the ability to
   name expressions in the search pattern so that they can be reused in
   the replacement.  For example,

   (('bcsel', ('ige', ('find_lsb=b', a), 0), ('find_lsb', a), -1), b),

   An alternative would be to add some kind of CSE at the time of
   inserting the replacements.  Create a new instruction, then check to
   see if it already exists.  That option might be better overall.

Over the years I know Matt has heard me complain, "I added a pattern
that just deleted an instruction, but it added a bunch of spills!"  This
was always in large, complex shaders that are very hard to analyze.  I
always blamed these cases on the scheduler being dumb.  I am now very
suspicious that unintended code motion was the real problem.

All Gen4+ Intel platforms had similar results. (Tiger Lake shown)
total instructions in shared programs: 17611405 -> 17611333 (<.01%)
instructions in affected programs: 18613 -> 18541 (-0.39%)
helped: 41
HURT: 13
helped stats (abs) min: 1 max: 18 x̄: 4.46 x̃: 4
helped stats (rel) min: 0.27% max: 5.68% x̄: 1.29% x̃: 1.34%
HURT stats (abs)   min: 1 max: 20 x̄: 8.54 x̃: 7
HURT stats (rel)   min: 0.30% max: 4.20% x̄: 2.15% x̃: 2.38%
95% mean confidence interval for instructions value: -3.29 0.63
95% mean confidence interval for instructions %-change: -0.95% 0.02%
Inconclusive result (value mean confidence interval includes 0).

total cycles in shared programs: 338366118 -> 338365223 (<.01%)
cycles in affected programs: 257889 -> 256994 (-0.35%)
helped: 42
HURT: 15
helped stats (abs) min: 2 max: 120 x̄: 39.38 x̃: 34
helped stats (rel) min: 0.04% max: 2.55% x̄: 0.86% x̃: 0.76%
HURT stats (abs)   min: 6 max: 204 x̄: 50.60 x̃: 34
HURT stats (rel)   min: 0.11% max: 4.75% x̄: 1.12% x̃: 0.56%
95% mean confidence interval for cycles value: -30.39 -1.02
95% mean confidence interval for cycles %-change: -0.66% -0.02%
Cycles are helped.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>

4 years agointel/vec4: Allow late copy propagation on vec4
Ian Romanick [Mon, 10 Jun 2019 23:45:08 +0000 (16:45 -0700)]
intel/vec4: Allow late copy propagation on vec4

This change incurs a small amount of hurt now, but it enables a lot of
benefit on vec4 shaders on the next commit.  nir_opt_algebraic_late
converts dph, dot3, etc. to dhp_replicated, dot_replicated3, etc.  In
the process, it introduces extra moves.  If the original NIR contained

        vec1 32 ssa_45 = fdot4 ssa_51, ssa_44
        vec1 32 ssa_46 = fneg ssa_45

nir_opt_algebraic_late will produce

        vec4 32 ssa_18 = fdot_replicated4 ssa_1, ssa_15
        vec1 32 ssa_19 = mov ssa_18.x
        vec1 32 ssa_17 = fneg ssa_19

The algebraic pass added in the next commit can't see through the move
to know that the fneg applies to a fdot_replicated4.

Haswell, Ivy Bridge, and Sandybridge had similar results. (Haswell shown)
total cycles in shared programs: 187077604 -> 187079858 (<.01%)
cycles in affected programs: 350132 -> 352386 (0.64%)
helped: 174
HURT: 194
helped stats (abs) min: 2 max: 124 x̄: 23.60 x̃: 16
helped stats (rel) min: 0.12% max: 15.88% x̄: 4.98% x̃: 3.86%
HURT stats (abs)   min: 2 max: 164 x̄: 32.78 x̃: 16
HURT stats (rel)   min: 0.17% max: 22.82% x̄: 6.46% x̃: 0.86%
95% mean confidence interval for cycles value: 2.04 10.21
95% mean confidence interval for cycles %-change: 0.17% 1.93%
Cycles are HURT.

No shader-db changes on any other Intel platform.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/1359>

4 years agonir: fix crash in varying packing on interface mismatch
Timothy Arceri [Fri, 27 Mar 2020 15:17:54 +0000 (02:17 +1100)]
nir: fix crash in varying packing on interface mismatch

For example when the outputs are scalars but the inputs are struct
members.

Fixes: 26aa460940f6 ("nir: rewrite varying component packing")
Reviewed-By: Timur Kristóf <timur.kristof@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4351>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4351>

4 years agofreedreno/turnip: Use the NIR info to decide if we need helper invocations.
Eric Anholt [Tue, 24 Mar 2020 16:59:16 +0000 (09:59 -0700)]
freedreno/turnip: Use the NIR info to decide if we need helper invocations.

We had an approximation that was assuming any ddx or tex instruction
needed helper invocations, but that's not true for texelFetch() or
textureSize().  It also meant that we were setting PIXLOD on vertex and
compute shaders doing texturing, which doesn't really make sense.

shader-db (with a hack to log pixlod):
total pixlod in shared programs: 582 -> 573 (-1.55%)

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2681
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4308>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4308>

4 years agofreedreno: Drop an unnecessary include marked "this should go away"
Eric Anholt [Mon, 23 Mar 2020 21:19:48 +0000 (14:19 -0700)]
freedreno: Drop an unnecessary include marked "this should go away"

It came in with the initial import, and doesn't seem to be necessary any
more.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4289>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4289>

4 years agofreedreno/ir3: fix android build
Rob Clark [Mon, 30 Mar 2020 22:53:48 +0000 (15:53 -0700)]
freedreno/ir3: fix android build

Fixes: e5339fe4a47 ("Move compiler.h and imports.h/c from src/mesa/main into src/util")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4381>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4381>

4 years agoutil: move ALIGN/ROUND_DOWN_TO to u_math.h
Rob Clark [Mon, 30 Mar 2020 22:52:30 +0000 (15:52 -0700)]
util: move ALIGN/ROUND_DOWN_TO to u_math.h

These are less mesa specific than the rest of macros.h, and would be
nice to use outside of mesa.  Prep for next patch.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4381>

4 years agospirv: Implement OpCopyObject and OpCopyLogical as blind copies
Jason Ekstrand [Mon, 30 Mar 2020 16:25:07 +0000 (11:25 -0500)]
spirv: Implement OpCopyObject and OpCopyLogical as blind copies

Because the types etc. are required to logically match, we can just
copy-propagate the guts of the vtn_value.  This was causing issues with
some new CTS tests that are doing an OpCopyObject of a sampler which is
a special-cased type in spirv_to_nir.  Of course, this is only a partial
solution.  Ideally, we've got a bit of work to do to make all the
composite stuff able to handle all types including images, sampler, and
combined image/samplers but this gets some CTS tests passing.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4375>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4375>

4 years agoisl: don't warn in physical extent calculation for yuv formats
Lionel Landwerlin [Tue, 26 Nov 2019 14:22:55 +0000 (16:22 +0200)]
isl: don't warn in physical extent calculation for yuv formats

Those format have correct descriptions already with the exception of
the planar format. In that case we introduce an assert.

This fine because we don't use the planar format in any of our
drivers. There are restrictions on how the addresses of the 2 planes
are relative to one another which make this annoying. The sampler is
also more limited than what we can do with a shader snippet.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2999>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2999>

4 years agoisl: set bpb for Y8_UNORM
Lionel Landwerlin [Tue, 26 Nov 2019 14:20:38 +0000 (16:20 +0200)]
isl: set bpb for Y8_UNORM

This isn't a format we use in any of the drivers but for consistency
just give it a correct bpb.

We also set the luminance in the G channel. We can't actually use this
format with the 3D sampler (only media).

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2999>

4 years agoscons: prune unused Makefile.sources
Eric Engestrom [Mon, 30 Mar 2020 14:30:28 +0000 (16:30 +0200)]
scons: prune unused Makefile.sources

Fixes: 2e92d3381988a85b2a6d ("scons: Prune out unnecessary targets.")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed by: Jose Fonseca <jfonseca@vmware.com>

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4373>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4373>

4 years agotu: Return the correct alignment for images
Connor Abbott [Fri, 27 Mar 2020 11:02:59 +0000 (12:02 +0100)]
tu: Return the correct alignment for images

The alignment field was never initialized, so we were just returning an
alignment of 0. Return the alignment from fdl, and while we're here
cleanup some leftovers in tu_private.h.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4357>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4357>

4 years agofreedreno/fdl: Add base_align
Connor Abbott [Fri, 27 Mar 2020 11:00:13 +0000 (12:00 +0100)]
freedreno/fdl: Add base_align

Tell users what the base address of the image needs to be aligned to.
These values are based on experimentation via passing an offset to
vkBindImageMemory with turnip and seeing if tests still pass. Note that
r8g8 is also special in this regard, however it actually has an
increased alignment (in bytes).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4357>

4 years agoanv/allocator: Use util_dynarray for blocks in anv_state_stream
Jason Ekstrand [Thu, 26 Mar 2020 17:57:36 +0000 (12:57 -0500)]
anv/allocator: Use util_dynarray for blocks in anv_state_stream

When we originally wrote a bunch of the allocation data structures, we
re-used the GPU memory for CPU-side data structures.  It's a bit more
memory efficient and usually ok.  However, this has a couple of
problems:

 1. It makes it MUCH more likely that the GPU will accidentlly stomp
    CPU-side data structures and cause nearly impossible to debug
    crashes.

 2. With discrete GPUs, the memory will be mapped somehow and that map
    may be across the BAR so it could have horribly slow CPU access.
    This is bad for our CPU-side data structures.

In the case of anv_state_stream, it also made the data structure
massively more complex than it needed to be.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4336>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4336>

4 years agoanv: Account for the header in anv_state_stream_alloc
Jason Ekstrand [Thu, 26 Mar 2020 17:22:48 +0000 (12:22 -0500)]
anv: Account for the header in anv_state_stream_alloc

If we have an allocation that's exactly the block size, we end up
computing a new block size to allocate that's exactly the block size,
add in the header, and then assert fail.  When computing the block size,
we need to account for the header.

Fixes: 955127db937 "anv/allocator: Add support for large stream..."
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4336>

4 years agost/mesa: add environment variable pin_app_thread for faster glthread on AMD Zen
Marek Olšák [Fri, 27 Mar 2020 12:20:51 +0000 (08:20 -0400)]
st/mesa: add environment variable pin_app_thread for faster glthread on AMD Zen

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4369>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4369>

4 years agogallium/u_threaded: call the driver to pin threads to L3 immediately
Marek Olšák [Sun, 22 Mar 2020 18:07:03 +0000 (14:07 -0400)]
gallium/u_threaded: call the driver to pin threads to L3 immediately

This is thread-safe and we want it to be done immediately for good L3 cache
usage.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4369>

4 years agolima: also check tiled and depth case when import
Qiang Yu [Mon, 30 Mar 2020 02:54:08 +0000 (10:54 +0800)]
lima: also check tiled and depth case when import

We missed the tiled and depth case when check buffer alignment.

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4362>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4362>

4 years agolima: fix buffer import with offset
Qiang Yu [Sat, 28 Mar 2020 06:09:22 +0000 (14:09 +0800)]
lima: fix buffer import with offset

With EGL_EXT_image_dma_buf_import, user can import dma_buf
with offset.

This is also used by AOSP GLConsumer::updateTexImage
with HAL_PIXEL_FORMAT_YV12 buffer which store YUV planes in
the same buffer with offset. Render sample from it using
GL_OES_EGL_image_external. This should fix some video
display problem when using MediaCodec soft decoding which
generates HAL_PIXEL_FORMAT_YV12 buffer and render it on
screen.

Test program:
https://github.com/yuq/gfx/tree/master/yuv2rgb/dma-buf

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4362>

4 years agopan/bi: Fix handling of constants with COMBINE
Alyssa Rosenzweig [Tue, 31 Mar 2020 00:54:51 +0000 (20:54 -0400)]
pan/bi: Fix handling of constants with COMBINE

We should never see COMBINE constants explicitly since they'll become
moves anyway, so we can simplify that. On the other hand, we do need the
type information for the lowering to work properly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Handle fp16/abs scheduling restriction
Alyssa Rosenzweig [Mon, 30 Mar 2020 16:25:20 +0000 (12:25 -0400)]
pan/bi: Handle fp16/abs scheduling restriction

See previous commit for the packing side. Here we update the scheduler
to accomodate this. Note we don't actually hit this path yet, but it's
good to be proactive.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Handle abs packing for fp16/FMA add/min
Alyssa Rosenzweig [Sat, 28 Mar 2020 02:34:15 +0000 (22:34 -0400)]
pan/bi: Handle abs packing for fp16/FMA add/min

It's seriously quirky, and all to save a single bit. Alas. It also
introduces an edge case for the scheduler which is a bit annoying.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Handle core faddminmax16 packing
Alyssa Rosenzweig [Sat, 28 Mar 2020 02:26:32 +0000 (22:26 -0400)]
pan/bi: Handle core faddminmax16 packing

This works without the exception of absolute values, which have some...
odd properties to be handled in the next commit.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Structify fadd/min/max16
Alyssa Rosenzweig [Sat, 28 Mar 2020 02:26:09 +0000 (22:26 -0400)]
pan/bi: Structify fadd/min/max16

There is some quirky encoding here.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Add v2f16 versions of rounding ops
Alyssa Rosenzweig [Sat, 28 Mar 2020 01:29:56 +0000 (21:29 -0400)]
pan/bi: Add v2f16 versions of rounding ops

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Handle round opcodes in frontend
Alyssa Rosenzweig [Sat, 28 Mar 2020 00:28:09 +0000 (20:28 -0400)]
pan/bi: Handle round opcodes in frontend

These correspond to various ops routed through BI_ROUND

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Assert out i16 related converts for now
Alyssa Rosenzweig [Sat, 28 Mar 2020 00:11:07 +0000 (20:11 -0400)]
pan/bi: Assert out i16 related converts for now

Needs more investigation, and GLSL doesn't use it quite yet sadly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Add one-source f32->f16 op
Alyssa Rosenzweig [Sat, 28 Mar 2020 00:08:03 +0000 (20:08 -0400)]
pan/bi: Add one-source f32->f16 op

This really has a second op for vectorization but we don't handle this
quite yet...

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Add bifrost_fma_2src generic
Alyssa Rosenzweig [Sat, 28 Mar 2020 00:07:43 +0000 (20:07 -0400)]
pan/bi: Add bifrost_fma_2src generic

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Handle standard FMA conversions
Alyssa Rosenzweig [Fri, 27 Mar 2020 23:06:28 +0000 (19:06 -0400)]
pan/bi: Handle standard FMA conversions

These are plain old 1-sources so they're easy to start with.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Enumerate conversions
Alyssa Rosenzweig [Fri, 27 Mar 2020 22:57:58 +0000 (18:57 -0400)]
pan/bi: Enumerate conversions

There are lots of Bifrost conversion opcodes that can all be emitted
from BI_CONVERT, let's pattern match.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Expand out FMA conversion opcodes
Alyssa Rosenzweig [Fri, 27 Mar 2020 22:34:21 +0000 (18:34 -0400)]
pan/bi: Expand out FMA conversion opcodes

There are a *lot* of them, with lots of symmetry we can exploit to
simplify the packing logic (but not entirely). Let's add the
corresponding header structs/defines, although we don't actually poke
the disassembler at this stage.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Pack outmod and roundmode with FMA
Alyssa Rosenzweig [Fri, 27 Mar 2020 19:55:03 +0000 (15:55 -0400)]
pan/bi: Pack outmod and roundmode with FMA

The fields got missed.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Add FMA16 packing
Alyssa Rosenzweig [Fri, 27 Mar 2020 19:53:57 +0000 (15:53 -0400)]
pan/bi: Add FMA16 packing

It's like the original FMA packing but with swizzles introduced.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Fix missing type for fmul
Alyssa Rosenzweig [Fri, 27 Mar 2020 19:53:12 +0000 (15:53 -0400)]
pan/bi: Fix missing type for fmul

We add a zero argument, we want it to align with the size of whatever
the other arguments were for optimization.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Finish FMA structures
Alyssa Rosenzweig [Fri, 27 Mar 2020 19:51:20 +0000 (15:51 -0400)]
pan/bi: Finish FMA structures

There were some missing fields for the 32-bit case, and the 16-bit case
has separate packing.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Ignore swizzle in unwritten component
Alyssa Rosenzweig [Fri, 27 Mar 2020 18:40:30 +0000 (14:40 -0400)]
pan/bi: Ignore swizzle in unwritten component

Otherwise we can trip the assert for no good reason.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Handle f2f* opcodes
Alyssa Rosenzweig [Fri, 27 Mar 2020 18:40:04 +0000 (14:40 -0400)]
pan/bi: Handle f2f* opcodes

Just more converts that got missed.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopanfrost: Enable PIPE_SHADER_CAP_FP16 on Bifrost
Alyssa Rosenzweig [Fri, 27 Mar 2020 18:39:39 +0000 (14:39 -0400)]
panfrost: Enable PIPE_SHADER_CAP_FP16 on Bifrost

We don't have fp16 implemented on Midgard yet but on Bifrost we can flip
it on now.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Enable precision lowering in standalone compiler
Alyssa Rosenzweig [Fri, 27 Mar 2020 18:39:18 +0000 (14:39 -0400)]
pan/bi: Enable precision lowering in standalone compiler

..since there's no CAP to guide here.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Fix off-by-one in scoreboarding packing
Alyssa Rosenzweig [Thu, 26 Mar 2020 14:10:33 +0000 (10:10 -0400)]
pan/bi: Fix off-by-one in scoreboarding packing

Clauses actually encode the *next* clauses' dependencies.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bi: Fix overzealous write barriers
Alyssa Rosenzweig [Thu, 26 Mar 2020 14:06:49 +0000 (10:06 -0400)]
pan/bi: Fix overzealous write barriers

It's possible this triggers an INSTR_INVALID_ENC.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bit: Begin generating a vertex job
Alyssa Rosenzweig [Wed, 25 Mar 2020 18:56:06 +0000 (14:56 -0400)]
pan/bit: Begin generating a vertex job

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bit: Submit a WRITE_VALUE job as a sanity check
Alyssa Rosenzweig [Tue, 24 Mar 2020 18:08:16 +0000 (14:08 -0400)]
pan/bit: Submit a WRITE_VALUE job as a sanity check

If this fails, everything else probably will too.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopanfrost: Stub out G31/G52 quirks
Alyssa Rosenzweig [Tue, 24 Mar 2020 17:53:18 +0000 (13:53 -0400)]
panfrost: Stub out G31/G52 quirks

There are none so far, but we'll need quirks accessible for Bifrost
specific details in the future, and in the mean time we need to handle
the cases somehow to avoid the unreachable(..)

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bit: Open up the device
Alyssa Rosenzweig [Tue, 24 Mar 2020 17:48:06 +0000 (13:48 -0400)]
pan/bit: Open up the device

As a start and a sanity check.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopanfrost: Move device open/close to root panfrost
Alyssa Rosenzweig [Tue, 24 Mar 2020 17:40:12 +0000 (13:40 -0400)]
panfrost: Move device open/close to root panfrost

We need it for standalone testing too.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopan/bit: Link standalone compiler with en/decoder
Alyssa Rosenzweig [Tue, 24 Mar 2020 17:24:03 +0000 (13:24 -0400)]
pan/bit: Link standalone compiler with en/decoder

We would like to submit jobs from the standalone compiler for testing
purposes, so let's get things wired up.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopanfrost: Move pan_bo to root panfrost
Alyssa Rosenzweig [Mon, 23 Mar 2020 23:36:46 +0000 (19:36 -0400)]
panfrost: Move pan_bo to root panfrost

Now that its Gallium dependencies have been resolved, we can move this
all out to root. The only nontrivial change here is keeping the
pandecode calls in Gallium-panfrost to avoid creating a circular
dependency between encoder/decoder. This could be solved with a third
drm folder but this seems less intrusive for now and Roman would
probably appreciate if I went longer than 8 hours without breaking the
Android build.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>

4 years agopanfrost: Inline reference counting routines
Alyssa Rosenzweig [Mon, 23 Mar 2020 23:17:49 +0000 (19:17 -0400)]
panfrost: Inline reference counting routines

We use only a very small subset of the capabilities of
pipe_reference (just wrappers for atomic ints..). Let's inline it and
drop the dependency on Gallium from pan_bo.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4382>