mesa.git
6 years agointel: Fix decoding for partial STATE_BASE_ADDRESS updates.
Kenneth Graunke [Mon, 22 Oct 2018 04:41:39 +0000 (21:41 -0700)]
intel: Fix decoding for partial STATE_BASE_ADDRESS updates.

STATE_BASE_ADDRESS only modifies various bases if the "modify" bit is
set.  Otherwise, we want to keep the existing base address.

Iris uses this for updating Surface State Base Address while leaving the
others as-is.

v2: Also update aubinator_viewer_decoder (caught by Lionel)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agonir: Use nir_src_is_const and nir_src_as_* in core code
Jason Ekstrand [Sat, 20 Oct 2018 14:10:02 +0000 (09:10 -0500)]
nir: Use nir_src_is_const and nir_src_as_* in core code

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agonir/search_helpers: Use nir_src_is_const and friends
Jason Ekstrand [Sat, 20 Oct 2018 17:07:41 +0000 (12:07 -0500)]
nir/search_helpers: Use nir_src_is_const and friends

This not only makes them safe for more bit sizes but it also fixes a bug
in is_zero_to_one where it would return true for constant NaN.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agonir/search: Use nir_src_is_const and friends
Jason Ekstrand [Sat, 20 Oct 2018 17:17:30 +0000 (12:17 -0500)]
nir/search: Use nir_src_is_const and friends

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agonir: Add some new helpers for working with const sources
Jason Ekstrand [Sat, 20 Oct 2018 13:36:21 +0000 (08:36 -0500)]
nir: Add some new helpers for working with const sources

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agomesa/st: Only call nir_lower_io_to_scalar_early on scalar ISAs
Alyssa Rosenzweig [Sun, 21 Oct 2018 18:29:37 +0000 (11:29 -0700)]
mesa/st: Only call nir_lower_io_to_scalar_early on scalar ISAs

On scalar ISAs, nir_lower_io_to_scalar_early enables significant
optimizations. However, on vector ISAs, it is counterproductive and
impedes optimal codegen. This patch only calls
nir_lower_io_to_scalar_early for scalar ISAs. It appears that at present
there are no upstreamed drivers using Gallium, NIR, and a vector ISA, so
for existing code, this should be a no-op. However, this patch is
necessary for the upcoming Panfrost (Midgard) and Lima (Utgard)
compilers, which are vector.

With this patch, Panfrost is able to consume NIR directly, rather than
TGSI with the TGSI->NIR conversion.

For how this affects Lima, see
https://www.mail-archive.com/mesa-dev@lists.freedesktop.org/msg189216.html

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
6 years agomeson: don't require libelf for r600 without LLVM
Dylan Baker [Mon, 22 Oct 2018 14:26:44 +0000 (07:26 -0700)]
meson: don't require libelf for r600 without LLVM

r600 doesn't have a hard requirement on LLVM, and therefore doesn't have
a hard requirement on libelf. Currently the logic doesn't allow that
however.

Distro-bug: https://bugs.gentoo.org/669058
Fixes: 5060c51b6f4dfb0d5358bde6523285163d3faaad
       ("meson: build r600 driver")
Reviewed-by: Matt Turner <mattst88@gmail.com>
6 years agoanv,radv: Trivially expose two new VK_GOOGLE extensions
Jason Ekstrand [Sat, 13 Oct 2018 13:46:20 +0000 (08:46 -0500)]
anv,radv: Trivially expose two new VK_GOOGLE extensions

This patch exposes support for the following two extensions:

 * VK_GOOGLE_decorate_string
 * VK_GOOGLE_hlsl_functionality1

There's nothing for the driver to do; it's all handled in spirv_to_nir.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107971
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agospirv: Add no-op support for VK_GOOGLE_hlsl_functionality1
Jason Ekstrand [Sat, 13 Oct 2018 13:41:36 +0000 (08:41 -0500)]
spirv: Add no-op support for VK_GOOGLE_hlsl_functionality1

This extension adds two new decorations which carry meaning only for
HLSL shaders.  They are expected to be handled by higher level layers
and can be ignored by implementations.  However, it does save the client
a bit of work if the implementation safely ignores them instead of the
client having to strip them out of the SPIR-V in order for it to be
valid.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agospirv: Add support for SPV_GOOGLE_decorate_string
Jason Ekstrand [Sat, 13 Oct 2018 13:33:22 +0000 (08:33 -0500)]
spirv: Add support for SPV_GOOGLE_decorate_string

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoandroid: Build kms_swrast for the Android platform
Rob Herring [Tue, 24 Jul 2018 09:09:39 +0000 (11:09 +0200)]
android: Build kms_swrast for the Android platform

Signed-off-by: Rob Herring <robh@kernel.org>
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoac: Fix loading a dvec3 from an SSBO
Connor Abbott [Thu, 18 Oct 2018 13:39:13 +0000 (15:39 +0200)]
ac: Fix loading a dvec3 from an SSBO

The comment was wrong, since the loop above casts to a type with the
correct bitsize already.

Fixes: 7e7ee82698247d8f93fe37775b99f4838b0247dd ("ac: add support for 16bit buffer loads")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoac: Introduce ac_build_expand()
Connor Abbott [Thu, 18 Oct 2018 13:30:11 +0000 (15:30 +0200)]
ac: Introduce ac_build_expand()

And implement ac_bulid_expand_to_vec4() on top of it.

Fixes: 7e7ee82698247d8f93fe37775b99f4838b0247dd ("ac: add support for 16bit buffer loads")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoir3/nir: Set up image_dims consts for image_deref_size intrinsic too
Eduardo Lima Mitev [Sun, 21 Oct 2018 18:48:41 +0000 (20:48 +0200)]
ir3/nir: Set up image_dims consts for image_deref_size intrinsic too

`nir_intrinsic_image_deref_size` is not being considered during scan for
driver constants, so image constants are not emitted if a shader
only ever query the size of an image (no load, store, atomic op, etc).
This is unlikely, but possible.

Reviewed-by: Rob Clark <robdclark@gmail.com>
6 years agonv50/ir: fix ConstantFolding::createMul for 64 bit muls
Karol Herbst [Fri, 19 Oct 2018 17:26:39 +0000 (19:26 +0200)]
nv50/ir: fix ConstantFolding::createMul for 64 bit muls

Fixes: 2f52925f5c60c72c9389bfdc122c3d5f8e15b25f
       "nv50/ir: move a * b -> a << log2(b) code into createMul()"

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
6 years agoradeonsi: Disable clear_state with radeon kernel driver
Sonny Jiang [Fri, 19 Oct 2018 20:16:41 +0000 (16:16 -0400)]
radeonsi: Disable clear_state with radeon kernel driver

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
6 years agomeson: Add -Werror=return-type when supported.
Kenneth Graunke [Tue, 30 Jan 2018 09:32:07 +0000 (01:32 -0800)]
meson: Add -Werror=return-type when supported.

This warning detects non-void functions with a missing return statement,
return statements with a value in void functions, and functions with an
bogus return type that ends up defaulting to int.  It's already enabled
by default with -Wall.  Generally, these are fairly serious bugs in the
code, which developers would like to notice and fix immediately.  This
patch promotes it from a warning to an error, to help developers catch
such mistakes early.

I would not expect this warning to change much based on the compiler
version, so hopefully it won't become a problem for packagers/builders.

See the GCC documentation or 'man gcc' for more details:
https://gcc.gnu.org/onlinedocs/gcc-7.3.0/gcc/Warning-Options.html#index-Wreturn-type

Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agoanv: Define trampolines as the weak functions
Jason Ekstrand [Mon, 15 Oct 2018 03:20:17 +0000 (22:20 -0500)]
anv: Define trampolines as the weak functions

Instead of having weak references to the anv functions and separate
trampoline functions with their own dispatch table, just make the
trampoline functions weak.  This gets rid of a dispatch table and
potentially lets the compiler delete the unused weak function.  The
end result is a reduction in the .text section of 5.7K and a reduction
in the .data section of 1.4K.

Before:

   text    data     bss     dec     hex filename
3190329  282232    8960 3481521  351fb1 _install/lib64/libvulkan_intel.so

After:

   text    data     bss     dec     hex filename
3184548  280792    8960 3474300  35037c _install/lib64/libvulkan_intel.so

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agodocs: fix typo in 18.2.3 release notes link
Juan A. Suarez Romero [Fri, 19 Oct 2018 16:47:45 +0000 (18:47 +0200)]
docs: fix typo in 18.2.3 release notes link

Fixes: 86b4bd52dc ("docs: update calendar, add news item and link
release notes for 18.2.3")

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
6 years agodocs: update calendar, add news item and link release notes for 18.2.3
Juan A. Suarez Romero [Fri, 19 Oct 2018 16:45:41 +0000 (18:45 +0200)]
docs: update calendar, add news item and link release notes for 18.2.3

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
6 years agodocs: add sha256 checksums for 18.2.3
Juan A. Suarez Romero [Fri, 19 Oct 2018 16:43:26 +0000 (18:43 +0200)]
docs: add sha256 checksums for 18.2.3

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
(cherry picked from commit 27fd12857b53ec22c0e918eee6c4c009643fccbc)

6 years agodocs: add release notes for 18.2.3
Juan A. Suarez Romero [Fri, 19 Oct 2018 16:02:51 +0000 (18:02 +0200)]
docs: add release notes for 18.2.3

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
(cherry picked from commit d219361b4226944835959676d1721b2a9d29da72)

6 years agoscons: Remove gles option.
Jose Fonseca [Thu, 18 Oct 2018 14:04:49 +0000 (15:04 +0100)]
scons: Remove gles option.

It's broken, and WGL state tracker is always built with GLES support
noawadays.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agoradv: Fix WSI & PCI bus info initialization order.
Bas Nieuwenhuizen [Fri, 19 Oct 2018 09:51:47 +0000 (11:51 +0200)]
radv: Fix WSI & PCI bus info initialization order.

Trying to access the bus info before it is initialized is not going
to work.

Fixes: baa38c144f6 "vulkan/wsi: Use VK_EXT_pci_bus_info for DRM fd matching"
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108491
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Andre Heider <a.heider@gmail.com>
6 years agoradeonsi: fix a typo in a comment in emit_guardband
Marek Olšák [Thu, 18 Oct 2018 22:01:00 +0000 (18:01 -0400)]
radeonsi: fix a typo in a comment in emit_guardband

6 years agoradeonsi: fix gnome-shell crash
Marek Olšák [Thu, 18 Oct 2018 21:54:24 +0000 (17:54 -0400)]
radeonsi: fix gnome-shell crash

I wasn't expecting to get viewports with the center having
negative coordinates.

Broken by: 6cc79e4411f

6 years agoRevert "anv: Stop generating weak references for instance entrypoints"
Jason Ekstrand [Mon, 15 Oct 2018 02:56:47 +0000 (21:56 -0500)]
Revert "anv: Stop generating weak references for instance entrypoints"

This reverts commit 00bb42105d6edf6e432c0e3712ffb9d3eb0aece4.  It was
not as well thought out as I had intended and broke the build when
VK_KHR_display is disabled in the build.

6 years agoradeonsi: clamp point size to the limit
Marek Olšák [Wed, 17 Oct 2018 16:26:54 +0000 (12:26 -0400)]
radeonsi: clamp point size to the limit

This fixes dEQP-GLES2.functional.rasterization.limits.points.
Broken by: ea039f789d9b54e1bd1d644b6a29863ca3500314

Tested-by: Jakob Bornecrantz <jakob@collabora.com>
6 years agoradeonsi: fix a VGT hang with primitive restart on Polaris10 and later
Marek Olšák [Tue, 16 Oct 2018 19:10:01 +0000 (15:10 -0400)]
radeonsi: fix a VGT hang with primitive restart on Polaris10 and later

Cc: 18.1 18.2 <mesa-stable@lists.freedesktop.org>
Tested-by: Jakob Bornecrantz <jakob@collabora.com>
6 years agoradeonsi: fix a deadlock due to partially-initialized context on CI
Marek Olšák [Wed, 17 Oct 2018 16:41:38 +0000 (12:41 -0400)]
radeonsi: fix a deadlock due to partially-initialized context on CI

6 years agoradeonsi: Bump number of allowed global buffers to 32
Jan Vesely [Thu, 18 Oct 2018 19:15:06 +0000 (15:15 -0400)]
radeonsi: Bump number of allowed global buffers to 32

Fixes assertion failure/crash when running luxmark/luxball on clover.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108272
CC: mesa-stable@lists.freedesktop.org
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradv: fix check for perftest options size
Andres Rodriguez [Thu, 18 Oct 2018 19:32:31 +0000 (15:32 -0400)]
radv: fix check for perftest options size

It was using the debug options array size.

CC: mesa-stable@lists.freedesktop.org
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradeonsi: fix incorrect hw screen offset and guardband computation
Marek Olšák [Thu, 18 Oct 2018 18:42:42 +0000 (14:42 -0400)]
radeonsi: fix incorrect hw screen offset and guardband computation

It resulted in assertion failures or incorrect rendering.

Broken by: 9e182b8313c5ab952498a76495f57e8420f9e5ad

6 years agovulkan/wsi: Use VK_EXT_pci_bus_info for DRM fd matching
Jason Ekstrand [Thu, 18 Oct 2018 15:08:32 +0000 (10:08 -0500)]
vulkan/wsi: Use VK_EXT_pci_bus_info for DRM fd matching

This lets us avoid passing the DRM fd around all over the place and gets
us closer to layer utopia.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agoloader/dri3: Also wait for front buffer fence if we triggered it
Michel Dänzer [Mon, 1 Oct 2018 16:43:46 +0000 (18:43 +0200)]
loader/dri3: Also wait for front buffer fence if we triggered it

In that case, we have to wait for the fence to synchronize with the
corresponding drawing we triggered in the X server.

Fixes incorrect display with the i965 driver and some applications, e.g.
solvespace.

Bugzilla: https://bugs.freedesktop.org/108097
Fixes: aefac10fecc9 "loader/dri3: Only wait for back buffer fences in
                     dri3_get_buffer"
Tested-by: Sergii Romantsov <sergii.romantsov@globallogic.com>
6 years agoanv: Stop generating weak references for instance entrypoints
Jason Ekstrand [Mon, 15 Oct 2018 02:56:47 +0000 (21:56 -0500)]
anv: Stop generating weak references for instance entrypoints

We don't need weak references to instance entrypoints because we never
have more than one of each so we don't need the NULL fall-back.  This
also helps us avoid forgetting things because we now get link errors for
missing instance entrypoints.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agovulkan/wsi: Implement GetPhysicalDevicePresentRectanglesKHR
Jason Ekstrand [Mon, 15 Oct 2018 02:56:34 +0000 (21:56 -0500)]
vulkan/wsi: Implement GetPhysicalDevicePresentRectanglesKHR

This got missed during 1.1 enabling because it was defined as an
interaction between device groups and WSI and it wasn't obvious it was
in the delta.

The idea behind it is that it's supposed to provide a hint to the
application in a multi-GPU setup to indicate which regions of the screen
are being scanned out by which GPU so a multi-device split-screen
rendering application can render each part of the screen on the GPU that
will be presenting it and avoid extra bus traffic between GPUs.  On a
single-GPU setup or one which doesn't support this present mode, we need
to do something.  We choose to return the window size (or a max-size
rect) if the compositor, X server, or crtc is associated with the given
physical device and zero rectangles otherwise.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agovulkan/wsi: Store the instance allocator in wsi_device
Jason Ekstrand [Wed, 17 Oct 2018 19:35:16 +0000 (14:35 -0500)]
vulkan/wsi: Store the instance allocator in wsi_device

We already have wsi_device and we know the instance allocator at
wsi_device_init time so there's no need to pass it into the physical
device queries.  This also fixes a memory allocation domain bug that can
occur if CreateSwapchain gets called prior to any queries (not likely)
in which case the cached connection gets allocated off the device
instead of the instance.

Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agost/xlib: Use more appropriate include guard
Michał Janiszewski [Tue, 16 Oct 2018 21:44:22 +0000 (23:44 +0200)]
st/xlib: Use more appropriate include guard

Signed-off-by: Michał Janiszewski <janisozaur+signed@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com
6 years agogallium: Fix mismatched ifdef-guards
Michał Janiszewski [Tue, 16 Oct 2018 21:44:21 +0000 (23:44 +0200)]
gallium: Fix mismatched ifdef-guards

Signed-off-by: Michał Janiszewski <janisozaur+signed@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agosoftpipe: dynamically allocate space for immediate constants
Gert Wollny [Tue, 16 Oct 2018 08:07:49 +0000 (10:07 +0200)]
softpipe: dynamically allocate space for immediate constants

The number of immediate constants was fixed and the size check was
only done by means of an assertion. Given this a shader that emits
more immediate constants would result in a memory corruption when
mesa is build in release mode.

Instead of using this fixed limit allocate the space dynamically, let it
grow as needed, and also remove the unused ImmArray.

Fixes: dEQP-GLES31.functional.ssbo.layout.random.arrays_of_arrays.1
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agoradv: use nir_shrink_vec_array_vars()
Timothy Arceri [Wed, 17 Oct 2018 23:23:42 +0000 (10:23 +1100)]
radv: use nir_shrink_vec_array_vars()

Totals from affected shaders:
SGPRS: 1096 -> 1096 (0.00 %)
VGPRS: 1192 -> 1056 (-11.41 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 100940 -> 94384 (-6.49 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 100 -> 112 (12.00 %)
Wait states: 0 -> 0 (0.00 %)

All affected shaders are from Batman Arkham City.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: use nir_split_array_vars()
Timothy Arceri [Wed, 17 Oct 2018 23:19:16 +0000 (10:19 +1100)]
radv: use nir_split_array_vars()

We call in the opt loop in case another pass results in an
array with indirect access being turned into direct access.

Totals from affected shaders:
SGPRS: 512 -> 496 (-3.12 %)
VGPRS: 456 -> 452 (-0.88 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 40040 -> 39664 (-0.94 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 41 -> 43 (4.88 %)
Wait states: 0 -> 0 (0.00 %)

All affected shaders are from Batman Arkham City.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: use nir_opt_find_array_copies()
Timothy Arceri [Wed, 17 Oct 2018 22:42:17 +0000 (09:42 +1100)]
radv: use nir_opt_find_array_copies()

Totals from affected shaders:
SGPRS: 1112 -> 1112 (0.00 %)
VGPRS: 1492 -> 1196 (-19.84 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 112172 -> 101316 (-9.68 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 93 -> 98 (5.38 %)
Wait states: 0 -> 0 (0.00 %)

All affected shaders are from "Batman: Arkham City" over DXVK.

The pass detects that the temporary array created by DXVK for
storing TCS inputs is a copy of the input arrays and allows
us to avoid copying all of the input data and then indirecting
on it with if-ladders, instead we just do indirect indexing.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: use nir_opt_copy_prop_vars and nir_opt_dead_write_vars
Timothy Arceri [Wed, 17 Oct 2018 21:55:46 +0000 (08:55 +1100)]
radv: use nir_opt_copy_prop_vars and nir_opt_dead_write_vars

Totals from affected shaders:
SGPRS: 2856 -> 2856 (0.00 %)
VGPRS: 3236 -> 3248 (0.37 %)
Spilled SGPRs: 0 -> 0 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 236560 -> 233548 (-1.27 %) bytes
LDS: 0 -> 0 (0.00 %) blocks
Max Waves: 277 -> 283 (2.17 %)
Wait states: 0 -> 0 (0.00 %)

Even in the cases were we have increased VGPR use it appears
the NIR is improved significantly.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agovulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]
Keith Packard [Thu, 11 Oct 2018 23:05:18 +0000 (16:05 -0700)]
vulkan: Add VK_EXT_calibrated_timestamps extension (radv and anv) [v5]

Offers three clocks, device, clock monotonic and clock monotonic
raw. Could use some kernel support to reduce the deviation between
clock values.

v2:
Ensure deviation is at least as big as the GPU time interval.

v3:
Set device->lost when returning DEVICE_LOST.
Use MAX2 and DIV_ROUND_UP instead of open coding these.
Delete spurious TIMESTAMP in radv version.

Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Suggested-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
v4:
Add anv_gem_reg_read to anv_gem_stubs.c

Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
v5:
Adjust maxDeviation computation to max(sampled_clock_period) +
sample_interval.

Suggested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Signed-off-by: Keith Packard <keithp@keithp.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agointel/compiler/icl: Use invocation id bits 22:16 instead of 23:17
Topi Pohjolainen [Tue, 16 Oct 2018 11:56:51 +0000 (07:56 -0400)]
intel/compiler/icl: Use invocation id bits 22:16 instead of 23:17

Identifier bits in the dispatch header have changed. See Bspec:

SINGLE_PATCH Payload:

3D Pipeline Stages - 3D Pipeline Geometry -
Hull Shader (HS) Stage IVB+ - Payloads IVB+

Fixes: KHR-GL46.tessellation_shader.tessellation_shader_tc_barriers.barrier_guarded_read_write_calls
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
6 years agoFix setting indent-tabs-mode in the Emacs .dir-locals.el files
Neil Roberts [Wed, 17 Oct 2018 15:27:07 +0000 (17:27 +0200)]
Fix setting indent-tabs-mode in the Emacs .dir-locals.el files

Some of the .dir-locals.el had the wrong name for the truthy value so
it wasn’t setting indent-tabs-mode.

Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
6 years agofreedreno/a6xx: don't allocate binning rb
Rob Clark [Mon, 15 Oct 2018 13:56:35 +0000 (09:56 -0400)]
freedreno/a6xx: don't allocate binning rb

Now that a single cmdstream is used for both binning and draw passes, we
can skip allocation of cmdstream buffer for binning.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: single cmdstream for draw+binning
Rob Clark [Sat, 13 Oct 2018 17:56:05 +0000 (13:56 -0400)]
freedreno/a6xx: single cmdstream for draw+binning

Now that state which is different for draw vs binning pass is split out
into different state-groups with appropriate enable_mask (so the
appropriate one is chosen for draw vs binning), switch over to using a
single cmdstream for both passes.

This should significantly lower draw overhead for CPU bound benchmarks.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: split binning vs draw program stateobj's
Rob Clark [Sat, 13 Oct 2018 17:54:32 +0000 (13:54 -0400)]
freedreno/a6xx: split binning vs draw program stateobj's

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: split VBO state into binning/draw variants
Rob Clark [Sat, 13 Oct 2018 17:51:33 +0000 (13:51 -0400)]
freedreno/a6xx: split VBO state into binning/draw variants

Blob seems to manage to use same input registers for BS (binning pass)
vs VS (draw pass) shaders, so it can use the same VBO state for both.
We can't quite do that yet, so split them.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: move VBO state to stateobj
Rob Clark [Sat, 13 Oct 2018 17:24:24 +0000 (13:24 -0400)]
freedreno/a6xx: move VBO state to stateobj

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: move ZSA state to stateobj
Rob Clark [Sat, 13 Oct 2018 17:04:16 +0000 (13:04 -0400)]
freedreno/a6xx: move ZSA state to stateobj

Step towards single cmdstream, where we need different state-group-id's
for binning vs draw ZSA state.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: remove vismode param
Rob Clark [Fri, 12 Oct 2018 20:29:22 +0000 (16:29 -0400)]
freedreno/a6xx: remove vismode param

We don't need to keep this IGNORE_VISIBILITY in binning pass.  Prep work
for using single cmdstream for both draw and binning passes.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/ir3: move binning-pass fixup for a6xx+
Rob Clark [Fri, 12 Oct 2018 20:01:22 +0000 (16:01 -0400)]
freedreno/ir3: move binning-pass fixup for a6xx+

Move this to after ir3_cp (which can add lowered immediates to the const
state) for a6xx+, to ensure the uniform state matches between binning
and vertex shaders.  This way we can emit just a single VS_CONST state-
group when we re-use single cmdstream for both binning and draw passes.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: a bit more state emit cleanup
Rob Clark [Fri, 12 Oct 2018 19:10:46 +0000 (15:10 -0400)]
freedreno/a6xx: a bit more state emit cleanup

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: move framebuffer state emit to emit_mrt()
Rob Clark [Fri, 12 Oct 2018 18:47:13 +0000 (14:47 -0400)]
freedreno/a6xx: move framebuffer state emit to emit_mrt()

No point in checking this per-draw, since framebuffer change means new
batch.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: small emit_mrt() cleanup
Rob Clark [Fri, 12 Oct 2018 18:41:14 +0000 (14:41 -0400)]
freedreno/a6xx: small emit_mrt() cleanup

On a6xx, this is only used for pfb->cbufs so we can just directly pass
the pfb state.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: use program cache
Rob Clark [Wed, 10 Oct 2018 19:58:57 +0000 (15:58 -0400)]
freedreno/a6xx: use program cache

Use the in-memory cache to construct shader program state and re-use it
on subsequent draws, to lower driver overhead.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/ir3: shader variant cache
Rob Clark [Tue, 24 Jul 2018 18:12:24 +0000 (14:12 -0400)]
freedreno/ir3: shader variant cache

Cache that maps gallium hwcso (in this case, 'struct ir3_shader') plus
shader variant key to a generation specific state object.

This could eventually replace the linked list of shader variants, but
for now it lets us re-use the work currently done in fdN_program_emit()

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/ir3: move binning_pass out of shader variant key
Rob Clark [Tue, 2 Oct 2018 16:38:09 +0000 (12:38 -0400)]
freedreno/ir3: move binning_pass out of shader variant key

Prep work for a following patch, that introduces a cache to map from
program state (all shader stages) plus variant key to pre-baked hw
state (which could be emit'd via CP_SET_DRAW_STATE, for example).
To do that, we really want the variant key to be immutable, and to
treat the binning pass shader as an extra shader stage, rather than
as a VS variant.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/ir3: track # of samplers used by shader
Rob Clark [Tue, 2 Oct 2018 20:04:39 +0000 (16:04 -0400)]
freedreno/ir3: track # of samplers used by shader

This is useful for a6xx to avoid program state from depending on bound
tex/samp state.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: texture state obj
Rob Clark [Wed, 10 Oct 2018 19:59:29 +0000 (15:59 -0400)]
freedreno/a6xx: texture state obj

Unfortunately gallium doesn't match what the hw wants perfectly here, in
using a separate CSO for each texture/sampler.  So we have to use a hash
table to map the collection of texture/samplers to hw state object.

We probably could use separate hw state objects for texture and sampler
state, but mesa/st tends to update the tex and samp state together.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: add resource seqno
Rob Clark [Fri, 12 Oct 2018 13:24:21 +0000 (09:24 -0400)]
freedreno: add resource seqno

Intended to be something more compact than a 64b pointer, which could be
used as a key into hashtables.  Prep work for texture state objects.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: move const emit to state group
Rob Clark [Sun, 7 Oct 2018 17:59:27 +0000 (13:59 -0400)]
freedreno/a6xx: move const emit to state group

Eventually we want to move nearly everything, but no other state depends
on const state, so this is the easiest one to move first.

For webgl aquarium, this reduces GPU load by about 10%, since for each
fish it does a uniform upload plus draw.. fish frequently are visible in
only a single tile, so this skips the uniform uploads for other tiles.

The additional step of avoiding WFI's when using CP_SET_DRAW_STATE seems
to be work an additional 10% gain for aquarium.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: add infrastructure for CP_DRAW_STATE
Rob Clark [Sun, 7 Oct 2018 17:58:30 +0000 (13:58 -0400)]
freedreno/a6xx: add infrastructure for CP_DRAW_STATE

Add helper to add state-groups to emit, and code to emit CP_DRAW_STATE
packet if we have any state-groups.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: reduce resource dependency tracking overhead
Rob Clark [Mon, 1 Oct 2018 18:13:06 +0000 (14:13 -0400)]
freedreno: reduce resource dependency tracking overhead

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: Remove the Emacs mode lines
Neil Roberts [Wed, 17 Oct 2018 15:38:03 +0000 (17:38 +0200)]
freedreno: Remove the Emacs mode lines

These are not necessary because the corresponding settings are set via
the .dir-locals.el file anyway. Most of them were missing a ‘:’ after
“tab-width” which was making Emacs display an annoying warning
whenever you open the file.

This patch was made with:

sed -ri '/-\*- mode:/,/^$/d' \
    $(find src/gallium/{drivers,winsys} -name \*.\[ch\] \
               -exec grep -l -- '-\*- mode:' {} \+)

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: Fix the Emacs indentation configuration file
Neil Roberts [Wed, 17 Oct 2018 15:38:02 +0000 (17:38 +0200)]
freedreno: Fix the Emacs indentation configuration file

The .dir-locals.el had the wrong name for the truthy value so it
wasn’t setting indent-tabs-mode.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: allocate batches from the cache in launch_grid
Hyunjun Ko [Wed, 17 Oct 2018 12:57:28 +0000 (21:57 +0900)]
freedreno: allocate batches from the cache in launch_grid

Needs to allocate batches from the cache so that it could
get a valid index and make resource dependancy tracking right.

In addition this fixes assertion on debug build since the commit
1a40faa8 landed.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: adds nondraw param to fd_bc_alloc_batch
Hyunjun Ko [Wed, 17 Oct 2018 12:57:27 +0000 (21:57 +0900)]
freedreno: adds nondraw param to fd_bc_alloc_batch

Needs to specify nondraw when creating a batch through
fd_bc_alloc_batch since it'd better create a batch through
it rather than fd_batch_create.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: remove fd6_emit_render_cntl()
Rob Clark [Fri, 12 Oct 2018 20:27:59 +0000 (16:27 -0400)]
freedreno/a6xx: remove fd6_emit_render_cntl()

It was dead code carried over from a5xx

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/ir3: fix broken texcoord inputs
Rob Clark [Sat, 15 Sep 2018 18:41:07 +0000 (14:41 -0400)]
freedreno/ir3: fix broken texcoord inputs

TODO not sure if this is best solution, but current logic is broken for
texcoord inputs.  It is definitely the simplest solution.

Fixes: 1a24f519663 freedreno/ir3: ignore unused inputs
Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: fix off-by-one error in BEGIN_RING()
Rob Clark [Sat, 13 Oct 2018 16:34:09 +0000 (12:34 -0400)]
freedreno: fix off-by-one error in BEGIN_RING()

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agoutil: document a limitation of util_fast_udiv32
Marek Olšák [Sun, 23 Sep 2018 00:03:27 +0000 (20:03 -0400)]
util: document a limitation of util_fast_udiv32

trivial

6 years agoi965/fs: Add 64-bit int immediate support to dump_instructions()
Matt Turner [Thu, 6 Sep 2018 18:15:55 +0000 (11:15 -0700)]
i965/fs: Add 64-bit int immediate support to dump_instructions()

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
6 years agoradeonsi: track context rolls better for the Vega scissor bug workaround
Marek Olšák [Sun, 7 Oct 2018 02:44:36 +0000 (22:44 -0400)]
radeonsi: track context rolls better for the Vega scissor bug workaround

We should get fewer context rolls with the SET_CONTEXT_REG optimization,
but it would have been for nothing if the scissor state rolled the context
anyway. Don't emit the scissor state if there is no context roll.

6 years agoradeonsi: emit sample locations for 1xAA only when the hw bug is present
Marek Olšák [Sun, 7 Oct 2018 02:53:33 +0000 (22:53 -0400)]
radeonsi: emit sample locations for 1xAA only when the hw bug is present

6 years agoradeonsi: use compute shaders for clear_buffer & copy_buffer
Marek Olšák [Fri, 24 Aug 2018 04:29:04 +0000 (00:29 -0400)]
radeonsi: use compute shaders for clear_buffer & copy_buffer

Fast color clears should be much faster. Also, fast color clears on
evicted buffers should be 200x faster on GFX8 and older.

6 years agoradeonsi: use copy_buffer in buffer_do_flush_region directly
Marek Olšák [Sun, 7 Oct 2018 00:56:32 +0000 (20:56 -0400)]
radeonsi: use copy_buffer in buffer_do_flush_region directly

6 years agoradeonsi: use faster integer division for instance divisors
Marek Olšák [Sun, 23 Sep 2018 02:02:32 +0000 (22:02 -0400)]
radeonsi: use faster integer division for instance divisors

We know the divisors when we upload them, so instead we can precompute
and upload division factors derived from each divisor.

This fast division consists of add, mul_hi, and two shifts,
and we have to load 4 dwords intead of 1.

This probably won't affect any apps.

6 years agoac: add helpers for fast integer division by a constant
Marek Olšák [Sun, 23 Sep 2018 01:17:52 +0000 (21:17 -0400)]
ac: add helpers for fast integer division by a constant

6 years agoradeonsi: use higher subpixel precision (QUANT_MODE) for smaller viewports
Marek Olšák [Sat, 29 Sep 2018 01:43:49 +0000 (21:43 -0400)]
radeonsi: use higher subpixel precision (QUANT_MODE) for smaller viewports

6 years agoradeonsi: move emission of PA_SU_VTX_CNTL into emit_guardband
Marek Olšák [Sat, 29 Sep 2018 00:57:07 +0000 (20:57 -0400)]
radeonsi: move emission of PA_SU_VTX_CNTL into emit_guardband

We'll modify the quant mode there, which also affects the guarband
computation.

6 years agoradeonsi: don't re-upload the sample position constant buffer repeatedly
Marek Olšák [Sat, 29 Sep 2018 00:38:26 +0000 (20:38 -0400)]
radeonsi: don't re-upload the sample position constant buffer repeatedly

6 years agoradeonsi: set PA_SU_PRIM_FILTER_CNTL optimally
Marek Olšák [Sat, 29 Sep 2018 00:16:13 +0000 (20:16 -0400)]
radeonsi: set PA_SU_PRIM_FILTER_CNTL optimally

6 years agoradeonsi: center viewport to improve guardband clipping for high resolutions
Marek Olšák [Fri, 28 Sep 2018 22:49:29 +0000 (18:49 -0400)]
radeonsi: center viewport to improve guardband clipping for high resolutions

This will be more useful when we change the quant mode to increase subpixel
precision and decrease the viewport range (which might not be possible
if the viewport is not centered in the viewport range).

6 years agoradeonsi: save raster config in screen, add se_tile_repeat
Marek Olšák [Sat, 29 Sep 2018 23:28:20 +0000 (19:28 -0400)]
radeonsi: save raster config in screen, add se_tile_repeat

6 years agoradeonsi: switch back to standard DX sample positions
Marek Olšák [Fri, 28 Sep 2018 04:38:10 +0000 (00:38 -0400)]
radeonsi: switch back to standard DX sample positions

Apps may rely on them.

6 years agoradeonsi: add GDS support to CP DMA
Marek Olšák [Sun, 11 Sep 2016 20:15:04 +0000 (22:15 +0200)]
radeonsi: add GDS support to CP DMA

6 years agoradeonsi: rename si_gfx_* functions to si_cp_*
Marek Olšák [Fri, 21 Sep 2018 07:41:18 +0000 (03:41 -0400)]
radeonsi: rename si_gfx_* functions to si_cp_*

and write_event_eop -> release_mem

6 years agoradeonsi: make si_gfx_write_event_eop more configurable
Marek Olšák [Fri, 21 Sep 2018 07:36:32 +0000 (03:36 -0400)]
radeonsi: make si_gfx_write_event_eop more configurable

6 years agoanv/skylake: disable ForceThreadDispatchEnable
Sergii Romantsov [Wed, 19 Sep 2018 16:21:11 +0000 (19:21 +0300)]
anv/skylake: disable ForceThreadDispatchEnable

On Skylake enabling of ForceThreadDispatchEnable causes gpu-hang.

-v2: enabling of  ForceThreadDispatchEnable is only for gen8, for
     gen9 and higher reverted enabling of PixelShaderHasUAV.

-v3 (Jason Ekstrand): Rework the comments a bit.

CC: Jason Ekstrand <jason.ekstrand@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107941
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107760
Fixes: 79270d2140ec (anv: Stop setting 3DSTATE_PS_EXTRA::PixelShaderHasUAV)
Signed-off-by: Sergii Romantsov <sergii.romantsov@globallogic.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
6 years agoanv: Implement VK_EXT_pci_bus_info
Lionel Landwerlin [Sun, 14 Oct 2018 12:12:50 +0000 (13:12 +0100)]
anv: Implement VK_EXT_pci_bus_info

Even though the Intel GPU are always at the same PCI location, all the
info we need is already provided by libdrm. Let's be future proof.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
6 years agoappveyor: Cache pip's cache files.
Jose Fonseca [Fri, 12 Oct 2018 09:21:38 +0000 (10:21 +0100)]
appveyor: Cache pip's cache files.

It should speed up the Python packages installation.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agoappveyor: Update to newer Mako/winflexbison versions.
Jose Fonseca [Fri, 12 Oct 2018 09:09:07 +0000 (10:09 +0100)]
appveyor: Update to newer Mako/winflexbison versions.

As that's what most people are bound to use.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agoappveyor: Update to MSVC 2017.
Jose Fonseca [Fri, 12 Oct 2018 08:52:52 +0000 (09:52 +0100)]
appveyor: Update to MSVC 2017.

That's what we (and I suppose most people out there) are using now.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agoradv: disable VK_SUBGROUP_FEATURE_VOTE_BIT
Samuel Pitoiset [Tue, 16 Oct 2018 07:42:42 +0000 (09:42 +0200)]
radv: disable VK_SUBGROUP_FEATURE_VOTE_BIT

This feature isn't used for now, so disable it until
wwm is fixed in LLVM.

Fixes dEQP-VK.subgroups.vote.graphics.subgroupallequal*

https://bugs.freedesktop.org/show_bug.cgi?id=108115
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: implement buffer to image operations for R32G32B32
Samuel Pitoiset [Fri, 12 Oct 2018 09:30:13 +0000 (11:30 +0200)]
radv: implement buffer to image operations for R32G32B32

This should fix rendering issues with Batman Arkham City.
We will probably need to implement itob and itoi at some
point, but currently nothing hits these paths.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107765
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>