mesa.git
4 years agointel/fs: Add support for copy-propagating a block of multiple FIXED_GRFs.
Francisco Jerez [Mon, 30 Dec 2019 08:36:48 +0000 (00:36 -0800)]
intel/fs: Add support for copy-propagating a block of multiple FIXED_GRFs.

In cases where a LOAD_PAYLOAD instruction copies a single block of
sequential GRF registers into the destination (see
is_identity_payload()), splitting the block copy into a number of ACP
entries (one for each LOAD_PAYLOAD source) is undesirable, because
that prevents copy propagation into any instructions which read
multiple components at once with the same source (the barycentric
source of the LINTERP instruction is going to be the overwhelmingly
most common example).

Technically it would also be possible to do this for VGRF sources, but
there is little benefit from that since register coalesce already
covers many of those cases -- There is no way for a block of
FIXED_GRFs to be coalesced into a VGRF though.

This prevents the following shader-db regressions (including SIMD32
programs) in combination with the interpolation rework part of this
series.  On SKL:

   total instructions in shared programs: 18595160 -> 18828562 (1.26%)
   instructions in affected programs: 13374946 -> 13608348 (1.75%)
   helped: 7
   HURT: 108977

   total spills in shared programs: 9116 -> 9106 (-0.11%)
   spills in affected programs: 404 -> 394 (-2.48%)
   helped: 7
   HURT: 9

   total fills in shared programs: 8994 -> 9176 (2.02%)
   fills in affected programs: 898 -> 1080 (20.27%)
   helped: 7
   HURT: 9

   LOST:   469
   GAINED: 220

On SNB:

   total instructions in shared programs: 13996898 -> 14096222 (0.71%)
   instructions in affected programs: 8088546 -> 8187870 (1.23%)
   helped: 2
   HURT: 66520

   total spills in shared programs: 2985 -> 2961 (-0.80%)
   spills in affected programs: 632 -> 608 (-3.80%)
   helped: 2
   HURT: 0

   total fills in shared programs: 3144 -> 3128 (-0.51%)
   fills in affected programs: 1515 -> 1499 (-1.06%)
   helped: 2
   HURT: 0

   LOST:   0
   GAINED: 4

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agointel/fs: Add partial support for copy-propagating FIXED_GRFs.
Francisco Jerez [Mon, 30 Dec 2019 08:38:08 +0000 (00:38 -0800)]
intel/fs: Add partial support for copy-propagating FIXED_GRFs.

This will be useful for eliminating redundant copies from the FS
thread payload, particularly in SIMD32 programs.  For the moment we
only allow FIXED_GRFs with identity strides in order to avoid dealing
with composing the arbitrary bidimensional strides that FIXED_GRF
regions potentially have, which are rarely used at the IR level
anyway.

This enables the following commit allowing block-propagation of
FIXED_GRF LOAD_PAYLOAD copies, and prevents the following shader-db
regressions (including SIMD32 programs) in combination with the
interpolation rework part of this series.  On ICL:

   total instructions in shared programs: 20484665 -> 20529650 (0.22%)
   instructions in affected programs: 6031235 -> 6076220 (0.75%)
   helped: 5
   HURT: 42073

   total spills in shared programs: 8748 -> 8925 (2.02%)
   spills in affected programs: 186 -> 363 (95.16%)
   helped: 5
   HURT: 9

   total fills in shared programs: 8663 -> 8960 (3.43%)
   fills in affected programs: 647 -> 944 (45.90%)
   helped: 5
   HURT: 9

On SKL:

   total instructions in shared programs: 18937442 -> 19128162 (1.01%)
   instructions in affected programs: 8378187 -> 8568907 (2.28%)
   helped: 39
   HURT: 68176

   LOST:   1
   GAINED: 4

On SNB:

   total instructions in shared programs: 14094685 -> 14243499 (1.06%)
   instructions in affected programs: 7751062 -> 7899876 (1.92%)
   helped: 623
   HURT: 53586

   LOST:   7
   GAINED: 25

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agointel/fs: Extend copy propagation dataflow analysis to copies with FIXED_GRF source.
Francisco Jerez [Fri, 3 Jan 2020 02:54:13 +0000 (18:54 -0800)]
intel/fs: Extend copy propagation dataflow analysis to copies with FIXED_GRF source.

This involves indexing the ACP tables used internally by
fs_copy_prop_dataflow::setup_initial_values() by reg_space() instead
of register number.  Both are nearly equivalent for virtual GRFs
(barring the single bit of entropy lost in the hash), and this makes
handling FIXED_GRFs straightforward.

Because we're only going to support FIXED_GRFs for the source of a
copy, this change is only strictly necessary during the second pass
that checks for source interference, but we also apply the same change
to the first pass for consistency.

Note that this shouldn't change the behavior of the copy propagation
pass until we start inserting FIXED_GRF entries into the ACP.  Even
then FIXED_GRF writes are extremely rare so this change will hardly
ever have an effect, but they aren't completely non-existing so we
need to handle them for correctness.

No functional nor shader-db changes.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agointel/fs: Rework fs_inst::is_copy_payload() into multiple classification helpers.
Francisco Jerez [Tue, 31 Dec 2019 08:10:28 +0000 (00:10 -0800)]
intel/fs: Rework fs_inst::is_copy_payload() into multiple classification helpers.

This reworks the current fs_inst::is_copy_payload() method into a
number of classification helpers with well-defined semantics.  This
will be useful later on in order to optimize LOAD_PAYLOAD instructions
more aggressively in cases where we can determine it's safe to do so.

The closest equivalent of the present fs_inst::is_copy_payload()
method is the is_coalescing_payload() helper introduced here.

No functional nor shader-db changes.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agointel/fs: Generalize fs_reg::is_contiguous() to register files other than VGRF.
Francisco Jerez [Thu, 2 Jan 2020 23:32:56 +0000 (15:32 -0800)]
intel/fs: Generalize fs_reg::is_contiguous() to register files other than VGRF.

No functional nor shader-db changes.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agointel/fs: Try to vectorize header setup in lower_load_payload().
Francisco Jerez [Mon, 30 Dec 2019 02:17:10 +0000 (18:17 -0800)]
intel/fs: Try to vectorize header setup in lower_load_payload().

In cases where LOAD_PAYLOAD is provided a pair of contiguous registers
as header sources, try to use a single SIMD16 instruction in order to
initialize them.  This is unlikely to affect the overall cycle count
of the shader, since the compressed instruction has twice the issue
time, except due to the reduced pressure on the instruction cache.

Main motivation is avoiding instruction-count regressions in
combination with the following copy propagation improvements, which
will allow the SIMD16 g0-1 header setup emitted for framebuffer writes
to be copy-propagated into its LOAD_PAYLOAD, leading to the emission
of two SIMD8 MOV instructions instead of a single SIMD16 MOV.

Reverting this commit on top of the copy propagation changes would
lead to the following shader-db regressions on SKL and other
platforms:

 total instructions in shared programs: 14926738 -> 14935415 (0.06%)
 instructions in affected programs: 1892445 -> 1901122 (0.46%)
 helped: 0
 HURT: 8676

Without the following copy propagation changes this doesn't have any
effect on shader-db on Gen7+, because we would typically set up the FB
write header with a separate SIMD16 MOV that isn't currently
copy-propagated into the LOAD_PAYLOAD, so the individual SIMD8 MOVs
result of LOAD_PAYLOAD lowering would get register-coalesced away
under normal circumstances.  However that wasn't the case for MRF
LOAD_PAYLOAD destinations on Gen6 and earlier, because register
coalesce only kicks in for GRFs, leaving a number of redundant SIMD8
MOVs lying around.  On SNB this leads to the following shader-db
improvements:

 total instructions in shared programs: 10770538 -> 10734681 (-0.33%)
 instructions in affected programs: 2700655 -> 2664798 (-1.33%)
 helped: 17791
 HURT: 0

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agost/dri: do FLUSH_VERTICES before calling flush_resource
Marek Olšák [Tue, 10 Dec 2019 20:45:14 +0000 (15:45 -0500)]
st/dri: do FLUSH_VERTICES before calling flush_resource

4 years agogallium: add st_context_iface::flush_resource to call FLUSH_VERTICES
Marek Olšák [Tue, 10 Dec 2019 20:35:10 +0000 (15:35 -0500)]
gallium: add st_context_iface::flush_resource to call FLUSH_VERTICES

4 years agoanv: enable VK_KHR_swapchain_mutable_format
Lionel Landwerlin [Thu, 24 Jan 2019 12:03:56 +0000 (12:03 +0000)]
anv: enable VK_KHR_swapchain_mutable_format

Enable new tests in dEQP-VK.image.swapchain_mutable.*

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agovulkan/wsi: Implement VK_KHR_swapchain_mutable_format
Jason Ekstrand [Thu, 16 Jan 2020 20:39:58 +0000 (14:39 -0600)]
vulkan/wsi: Implement VK_KHR_swapchain_mutable_format

This is only the core WSI code for the extension.  It adds the image
format list and the flags to vkCreateImage as well as handling things
properly in the modifier queries.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agovulkan/wsi: Filter modifiers with ImageFormatProperties
Jason Ekstrand [Mon, 1 Oct 2018 21:00:32 +0000 (16:00 -0500)]
vulkan/wsi: Filter modifiers with ImageFormatProperties

Just because a modifier is returned for the given format, that doesn't
mean it works with all usages and flags.  We need to filter the list by
calling vkGetPhysicalDeviceImageFormatProperties2.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agovulkan/wsi: Use the interface from the real modifiers extension
Jason Ekstrand [Mon, 1 Oct 2018 21:14:24 +0000 (16:14 -0500)]
vulkan/wsi: Use the interface from the real modifiers extension

The anv implementation still isn't quite complete, but we can at least
start using the structs from the real extension.

v2: Fix circular pNext list (Lionel)

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agovulkan/wsi: Move the ImageCreateInfo higher up
Jason Ekstrand [Mon, 1 Oct 2018 21:04:20 +0000 (16:04 -0500)]
vulkan/wsi: Move the ImageCreateInfo higher up

Future changes will be easier if we can modify it based on whether or
not we're using modifiers.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agoanv: Support modifiers in GetImageFormatProperties2
Jason Ekstrand [Thu, 16 Jan 2020 19:38:55 +0000 (13:38 -0600)]
anv: Support modifiers in GetImageFormatProperties2

Images with modifiers come with restrictions:

 1. They have to be simple 2D images right now

 2. They need to have a sensible format (not compressed, multi-plane, or
    non-power-of-two)

 3. If a CCS modifier is being requested, they have to actually support
    CCS_E and be CCS-compatible with any other formats the client may
    wish to use for image views.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agoanv: Drop some VK_IMAGE_TILING_OPTIMAL checks
Jason Ekstrand [Mon, 1 Oct 2018 21:17:47 +0000 (16:17 -0500)]
anv: Drop some VK_IMAGE_TILING_OPTIMAL checks

The DRM format modifiers extension adds a TILING_DRM_FORMAT_MODIFIER
which will be used for modifiers so we can no longer use OPTIMAL to
indicate tiled inside the driver.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3434>

4 years agoaco: print assembly with CLRXdisasm for GFX6-GFX7 if found on the system
Samuel Pitoiset [Fri, 17 Jan 2020 08:49:44 +0000 (09:49 +0100)]
aco: print assembly with CLRXdisasm for GFX6-GFX7 if found on the system

LLVM only supports GFX8+. Using CLRXdisasm works most of the time,
so it's useful to add support for it.

Original patch by Daniel Schürmann.

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/3439>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3439>

4 years agovulkan/wsi: disable the hardware cursor
Andres Rodriguez [Tue, 10 Sep 2019 18:30:35 +0000 (14:30 -0400)]
vulkan/wsi: disable the hardware cursor

Ensure the hardware cursor is disabled when we set the mode for a
VkDisplayKHR object. The extension doesn't expose any mechanisms to
program the hardware cursor, so we need to ensure it is hidden.

Currently, it seems like X is responsible for disabling the cursor
before handing over the lease. But that seems a little frail, and we
should be disabling the cursor ourselves so it works correctly
independently of how the lease was prepared for us.

Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/1922>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/1922>

4 years agogallium/swr: Disable showing detected arch message.
Krzysztof Raszkowski [Fri, 17 Jan 2020 15:43:33 +0000 (16:43 +0100)]
gallium/swr: Disable showing detected arch message.

When swr driver is in use it print detected architecture
message to std::err. It can be harmfull when swr is using
in multinodes environments.
It can be enabled setting env var SWR_PRINT_INFO to 1.

Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
4 years agoaco: fix emitting slc for MUBUF instructions on GFX6-GFX7
Samuel Pitoiset [Fri, 17 Jan 2020 07:22:48 +0000 (08:22 +0100)]
aco: fix emitting slc for MUBUF instructions on GFX6-GFX7

Same as GFX10, only GFX8/GFX9 moved that bit near the opcode.

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/3437>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3437>

4 years agopanfrost/midgard: Fix swizzle for store instructions
Boris Brezillon [Thu, 16 Jan 2020 10:20:06 +0000 (11:20 +0100)]
panfrost/midgard: Fix swizzle for store instructions

The current logic considers that the nir_intrinsic_component(store_intr)
encodes the source components start, but it actually encodes the
destination one. Source component offset adjustment is taken care of in
install_registers_instr(), when offset_swizzle() is called.

This fixes dEQP-GLES2.functional.shaders.random.all_features.fragment.45
when PAN_MESA_DEBUG=deqp (looks like exposing GLES3 features has an
impact on the varyings layout).

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3429>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3429>

4 years agodocs: do not double-close link tag
Erik Faye-Lund [Thu, 16 Jan 2020 16:56:13 +0000 (17:56 +0100)]
docs: do not double-close link tag

Fixes: f8148d0cc17 "docs: remove mailing list as way of submitting patches"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: remove double-closed definition-list
Erik Faye-Lund [Thu, 16 Jan 2020 16:45:55 +0000 (17:45 +0100)]
docs: remove double-closed definition-list

Fixes: bc17ac58661 "docs: add documentation for building with meson"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: move paragraph closing tag
Erik Faye-Lund [Thu, 16 Jan 2020 16:39:39 +0000 (17:39 +0100)]
docs: move paragraph closing tag

The pre-tag right before is a block-level tag, which means it implicitly
terminates the paragraph. So there's no paragraph to close after this.
Instead, move the paragraph-closing before the pre-tag, to explicitly
close the paragraph.

Fixes: 41b3eb08d9f "docs: update meson docs for windows"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: use code-tags instead of pre-tags
Erik Faye-Lund [Thu, 16 Jan 2020 17:01:41 +0000 (18:01 +0100)]
docs: use code-tags instead of pre-tags

Similar to the previous two commits, it seems more appropriate to use
code-tags here than pre-tag.

Fixes: 9af6c38deff "docs: Add use of Closes: tag for closing gitlab issues"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: use code-tags instead of pre-tags
Erik Faye-Lund [Thu, 16 Jan 2020 16:49:34 +0000 (17:49 +0100)]
docs: use code-tags instead of pre-tags

Similar to the previous commit, code-tags seems more appropriate than
pre-tags here. So let's change it.

Fixes: ca0c1e69cab "docs: update releasing process to use new scripts and gitlab"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: use code-tag instead of pre-tag
Erik Faye-Lund [Thu, 16 Jan 2020 16:32:19 +0000 (17:32 +0100)]
docs: use code-tag instead of pre-tag

It's unlikely the author meant to use <pre>-here, as that starts a whole
new block. Instead, the inline code-tag seems more appropriate here.

Fixes: 41b3eb08d9f "docs: update meson docs for windows"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: open paragraph before closing it
Erik Faye-Lund [Thu, 16 Jan 2020 16:27:16 +0000 (17:27 +0100)]
docs: open paragraph before closing it

Fixes: 44c5e634a5c "docs: update meson docs for windows"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: fix paragraphs
Erik Faye-Lund [Thu, 16 Jan 2020 16:21:50 +0000 (17:21 +0100)]
docs: fix paragraphs

Paragraphs are terminated by pre-tags, so the latter one closes a new,
empty one. Let's split the paragraph in two around the pre-tag instead.

Fixes: c0dfe8c6dfd "docs: do not use div for line-breaking"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agodocs: fix typo in html tag name
Erik Faye-Lund [Thu, 16 Jan 2020 16:13:12 +0000 (17:13 +0100)]
docs: fix typo in html tag name

Fixes: 5d11a828e10 "docs: update install docs for meson"
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3431>

4 years agoutil: call bind_sampler_states before setting sampler_views
Pierre-Eric Pelloux-Prayer [Sat, 15 Feb 2020 22:05:53 +0000 (23:05 +0100)]
util: call bind_sampler_states before setting sampler_views

Fixes the following valgrind error:

    Invalid read of size 16
       at 0x28F458A1: si_set_sampler_view_desc (in radeonsi_drv_video.so)
       by 0x28F4657E: si_set_sampler_views (in radeonsi_drv_video.so)
       by 0x28D62BF5: util_compute_blit (in radeonsi_drv_video.so)
       by 0x28D3A944: vlVaHandleVAProcPipelineParameterBufferType (in radeonsi_drv_video.so)
       by 0x28D34EE1: vlVaRenderPicture (in radeonsi_drv_video.so)
       by 0x4B2582B: vaRenderPicture (in libva.so.2.500.0)
     Address 0x18142a10 is 0 bytes inside a block of size 48 free'd
       at 0x48369AB: free (vg_replace_malloc.c:540)
       by 0x28D62D51: util_compute_blit (in radeonsi_drv_video.so)
       by 0x28D3A944: vlVaHandleVAProcPipelineParameterBufferType (in radeonsi_drv_video.so)
       by 0x28D34EE1: vlVaRenderPicture (in radeonsi_drv_video.so)
       by 0x4B2582B: vaRenderPicture (in libva.so.2.500.0)
     Block was alloc'd at
       at 0x4837B65: calloc (vg_replace_malloc.c:762)
       by 0x28EFB2EC: si_create_sampler_state (in radeonsi_drv_video.so)
       by 0x28D62C30: util_compute_blit (in radeonsi_drv_video.so)
       by 0x28D3A944: vlVaHandleVAProcPipelineParameterBufferType (in radeonsi_drv_video.so)
       by 0x28D34EE1: vlVaRenderPicture (in radeonsi_drv_video.so)
       by 0x4B2582B: vaRenderPicture (in libva.so.2.500.0)

Fixes: 69430d7e59e ("va: use a compute shader for the blit")
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2321
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3428>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3428>

4 years agonir: Fix printing of ~0 .locations.
Eric Anholt [Thu, 19 Dec 2019 00:46:41 +0000 (16:46 -0800)]
nir: Fix printing of ~0 .locations.

I kept wondering what "429" meant in variable declarations, when it was
just a truncated ~0 snprintf.

Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3423>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3423>

4 years agomeson: use github URL for wraps instead of completely unreliable wrapdb
Eric Engestrom [Tue, 14 Jan 2020 15:53:21 +0000 (15:53 +0000)]
meson: use github URL for wraps instead of completely unreliable wrapdb

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3391>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3391>

4 years agodocs: Update release calendar for 20.0
Dylan Baker [Wed, 15 Jan 2020 20:37:52 +0000 (12:37 -0800)]
docs: Update release calendar for 20.0

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3417>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3417>

4 years agolima: Fix alpha blending
Andreas Baierl [Wed, 15 Jan 2020 14:31:39 +0000 (15:31 +0100)]
lima: Fix alpha blending

Introduce separate helper functions to set the blendfactor bits.

Lima uses bits 0-2 for the type, bit 3 sets the inverted function
and bit 4 is set if alpha is used.
alpha_src_factor and alpha_dst_factor don't need the alpha bit, so
they are masked with 0xf. There is only place for 4 bits anyway.
If alpha_src_factor is PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE, we need
to change it to PIPE_BLENDFACTOR_ONE first.
This is exactly what the blob does and we pass all
dEQP-GLES2.functional.fragment_ops.blend.* tests now.
Better than the blob btw...

Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Signed-off-by: Andreas Baierl <ichgeh@imkreisrum.de>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3411>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3411>

4 years agoaco: ignore parallelcopies to the same register on jump threading
Daniel Schürmann [Tue, 14 Jan 2020 12:14:38 +0000 (13:14 +0100)]
aco: ignore parallelcopies to the same register on jump threading

The more conservative lowering to CSSA inserts unnecessary parallelcopies
which might get coalesced and can be ignored on jump threading.

v2: outline is_empty_block() check.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3385>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3385>

4 years agoaco: handle phi affinities transitively through parallelcopies
Daniel Schürmann [Mon, 13 Jan 2020 14:13:19 +0000 (15:13 +0100)]
aco: handle phi affinities transitively through parallelcopies

This can coalesce most unnecessarily inserted parallelcopies
from lowering to CSSA.

v2: refactor loop a bit to make it more efficient and readable.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3385>

4 years agoaco: rework lower_to_cssa()
Daniel Schürmann [Mon, 13 Jan 2020 16:35:11 +0000 (17:35 +0100)]
aco: rework lower_to_cssa()

This patch changes lower_to_cssa to be much more conservative
about assumptions which phi operands might interfere.
Previously, this pass wasn't exhaustive and could miss some corner cases.

v2: remove optimizations to find better insertion points as it's hard
to guarantee that they are always correct and have overall no benefit.

Fixes: 0b8216b2cdbcaccfd2bd1a65be6b8ac5654e3067 ('aco: Lower to CSSA')
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3385>

4 years agoaco: implement stream output with vec3 on GFX6
Samuel Pitoiset [Wed, 15 Jan 2020 13:44:26 +0000 (14:44 +0100)]
aco: implement stream output with vec3 on GFX6

GFX6 doesn't support vec3.

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/3412>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3412>

4 years agoaco: do not combine additions of DS instructions on GFX6
Samuel Pitoiset [Wed, 15 Jan 2020 09:47:17 +0000 (10:47 +0100)]
aco: do not combine additions of DS instructions on GFX6

The offset field doesn't work as expected on GFX6.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3412>

4 years agoaco: do not select 96-bit/128-bit variants for ds_read/ds_write on GFX6
Samuel Pitoiset [Tue, 14 Jan 2020 17:46:36 +0000 (18:46 +0100)]
aco: do not select 96-bit/128-bit variants for ds_read/ds_write on GFX6

Only GFX7 and later support large ds_read/ds_write.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3412>

4 years agointel/perf: report query split for mdapi
Lionel Landwerlin [Mon, 16 Dec 2019 13:42:55 +0000 (15:42 +0200)]
intel/perf: report query split for mdapi

Also forgotten in the initial implementation.

v2: Report begin timestamp scaled by the timestamp frequency (Windows
    behavior)

v3: Rename split to disjoint to match GL terminology (Tapani)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: <mesa-stable@lists.freedesktop.org>
Acked-by: Tapani Pälli <tapani.palli@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3112>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3112>

4 years agointel/perf: expose timestamp begin for mdapi
Lionel Landwerlin [Mon, 16 Dec 2019 13:36:24 +0000 (15:36 +0200)]
intel/perf: expose timestamp begin for mdapi

This was forgotten in the initial implementation.

v2: ensure the value is written for both GL & Vulkan queries

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: <mesa-stable@lists.freedesktop.org>
Acked-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3112>

4 years agoanv: set depth stall enabled when depth flush enabled on gen12
Tapani Pälli [Tue, 14 Jan 2020 08:03:21 +0000 (10:03 +0200)]
anv: set depth stall enabled when depth flush enabled on gen12

This implements HW workaround #1409600907 for anv driver.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3378>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3378>

4 years agoiris: set depth stall enabled when depth flush enabled on gen12
Tapani Pälli [Tue, 14 Jan 2020 08:02:05 +0000 (10:02 +0200)]
iris: set depth stall enabled when depth flush enabled on gen12

This implements HW workaround #1409600907 for iris driver.

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3378>

4 years agoanv: implement another workaround for non pipelined states
Lionel Landwerlin [Wed, 15 Jan 2020 13:14:23 +0000 (15:14 +0200)]
anv: implement another workaround for non pipelined states

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3408>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3408>

4 years agoiris: implement another workaround for non pipelined states
Lionel Landwerlin [Wed, 15 Jan 2020 13:14:10 +0000 (15:14 +0200)]
iris: implement another workaround for non pipelined states

v2: add comment (Ken)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3408>

4 years agoiris: handle new PIPE_CONTROL field
Lionel Landwerlin [Wed, 15 Jan 2020 13:13:43 +0000 (15:13 +0200)]
iris: handle new PIPE_CONTROL field

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3408>

4 years agogenxml: add new Gen11+ PIPE_CONTROL field
Lionel Landwerlin [Wed, 15 Jan 2020 13:11:08 +0000 (15:11 +0200)]
genxml: add new Gen11+ PIPE_CONTROL field

PIPE_CONTROL gained a new field in its first DWORD on Gen11. We had no
use for it so far, but we start using it on Gen12.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3408>

4 years agost/mesa: Allocate full miplevels if MaxLevel is explicitly set
Kenneth Graunke [Wed, 15 Jan 2020 00:25:11 +0000 (16:25 -0800)]
st/mesa: Allocate full miplevels if MaxLevel is explicitly set

Some applications explicitly call glTex[ture]Parameteri[v] to set
GL_TEXTURE_MAX_LEVEL and GL_TEXTURE_BASE_LEVEL before uploading any
texture data.  Core Mesa initializes MaxLevel to 1000, so if it isn't
that, we know they've set it.  (We check for < TEXTURE_MAX_LEVELS to
avoid hardcoding that value, however.)

If MaxLevel - BaseLevel > 0, then the app is trying to tell us that
this texture is going to have multiple miplevels.  In that case, go
ahead and allocate the space for it.

Avoids many resource_copy_region calls at texture finalization time
in the Civilization VI benchmark.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3401>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3401>

4 years agoaco: fix emitting SMEM instructions with no operands on GFX6-GFX7
Samuel Pitoiset [Wed, 15 Jan 2020 12:08:17 +0000 (13:08 +0100)]
aco: fix emitting SMEM instructions with no operands on GFX6-GFX7

Like s_memtime.

Fixes dEQP-VK.glsl.shader_clock.* on GFX6.

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/3407>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3407>

4 years agolima: fix handling of reverse depth range
Vasily Khoruzhick [Wed, 15 Jan 2020 03:53:29 +0000 (19:53 -0800)]
lima: fix handling of reverse depth range

Looks like we need to handle cases when near > far and near == far.
In first case we just need to swap near and far, and in second we
need subtract epsilon from near if it's not zero.

Fixes 10 tests in dEQP-GLES2.functional.depth_range.*

Reviewed-by: Qiang Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3400>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3400>

4 years agonvc0: disable xfb's which don't have a stride
Ilia Mirkin [Thu, 16 Jan 2020 00:46:59 +0000 (19:46 -0500)]
nvc0: disable xfb's which don't have a stride

No stride / no attributes means that nothing is being written to the
buffer. However it might still prevent primitives from being written out
to the other buffers. Disabling it entirely seems to fix it.

Fixes GTF-GL45.gtf30.GL3Tests.transform_feedback.transform_feedback_overflow

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
4 years agolima/ppir: implement full liveness analysis for regalloc
Erico Nunes [Sun, 12 Jan 2020 14:11:55 +0000 (15:11 +0100)]
lima/ppir: implement full liveness analysis for regalloc

The existing liveness analysis in ppir still ultimately relies on a
single continuous live_in and live_out range per register and was
observed to be the bottleneck for register allocation on complicated
examples with several control flow blocks.
The use of live_in and live_out ranges was fine before ppir got control
flow, but now it ends up creating unnecessary interferences as live_in
and live_out ranges may span across entire blocks after blocks get
placed sequentially.

This new liveness analysis implementation generates a set of live
variables at each program point; before and after each instruction and
beginning and end of each block.
This is a global analysis and propagates the sets of live registers
across blocks independently of their sequence.
The resulting sets optimally represent all variables that cannot share a
register at each program point, so can be directly translated as
interferences to the register allocator.

Special care has to be taken with non-ssa registers. In order to
properly define their live range, their alive components also need to be
tracked. Therefore ppir can't use simple bitsets to keep track of live
registers.

The algorithm uses an auxiliary set data structure to keep track of the
live registers. The initial implementation used only trivial arrays,
however regalloc execution time was then prohibitive (>1minute on
Cortex-A53) on extreme benchmarks with hundreds of instructions,
hundreds of registers and several spilling iterations, mostly due to the
n^2 complexity to generate the interferences from the live sets. Since
the live registers set are only a very sparse subset of all registers at
each instruction, iterating only over this subset allows it to run very
fast again (a couple of seconds for the same benchmark).

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3358>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3358>

4 years agolima/ppir: remove orphan load node after cloning
Erico Nunes [Sun, 12 Jan 2020 13:30:26 +0000 (14:30 +0100)]
lima/ppir: remove orphan load node after cloning

There are some cases in shades using control flow where the varying load
is cloned to every block, and then the original node is left orphan.
This is not harmful for program execution, but it complicates analysis
for register allocation as there is now a case of writing to a register
that is never read.
While ppir doesn't have a dead code elimination pass for its own
optimizations and it is not hard to detect when we cloned the last load,
let's remove it early.

Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3358>

4 years agoiris: Print warning and return *out = NULL when fd to syncobj fails
Kristian H. Kristensen [Wed, 15 Jan 2020 00:56:41 +0000 (16:56 -0800)]
iris: Print warning and return *out = NULL when fd to syncobj fails

Signed-off-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agoiris: Advertise PIPE_CAP_NATIVE_FENCE_FD
Kristian H. Kristensen [Thu, 19 Dec 2019 18:56:03 +0000 (10:56 -0800)]
iris: Advertise PIPE_CAP_NATIVE_FENCE_FD

Enables EGL_ANDROID_native_fence_sync.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
4 years agoiris: Fix export of fences that have already completed.
Kenneth Graunke [Thu, 19 Dec 2019 21:51:07 +0000 (13:51 -0800)]
iris: Fix export of fences that have already completed.

After flushing batches, iris_fence_flush() asks the kernel whether
each batch's last_syncpt has already signalled or not.  (The idea is
that either the compute or render batch may not have actually had any
work queued up, so last_syncpt there might have been signalled a long
time ago.)  If it's already completed, we don't bother to record it.

A strange corner is the case of repeated flushes.  For example, we
might flush for some reason, and hit a glFlush(), and hit SwapBuffers.
It's possible for all the batches to have been flushed previously, -and-
for them to have actually completed.  In this case, we'll see that there
are no syncobj's to wait on, and record fence->count == 0.

This works fine internally - fence_finish can see count == 0 and realize
that it doesn't need to wait, for example.  But when working with native
FDs, we may be asked to export a fence with count == 0.  So we need an
actual synchronization primitive we can hand off.  Because all of the
relevant batches had been signalled when creating the fence, we want the
new dummy fence to be signalled as well.

So we just make a signalled syncobj and export it.

Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
4 years agoandroid: Fix whitespace issue
Robert Foss [Wed, 15 Jan 2020 00:11:23 +0000 (01:11 +0100)]
android: Fix whitespace issue

Signed-off-by: Robert Foss <robert.foss@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
4 years agopanfrost: Prefix schedule_program to prevent collision
Robert Foss [Wed, 15 Jan 2020 00:14:16 +0000 (01:14 +0100)]
panfrost: Prefix schedule_program to prevent collision

Currently the schedule_program implementation being used is picked
at compile time, which on the Android platform means that the
bifrost compiler & scheduler is used for all targets, including
midgard based hardware.

This commit disambiguates between the two schedule_program functions.

Signed-off-by: Robert Foss <robert.foss@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
4 years agoradeonsi: merge si_compile_llvm and si_llvm_compile functions
Marek Olšák [Wed, 15 Jan 2020 01:40:51 +0000 (20:40 -0500)]
radeonsi: merge si_compile_llvm and si_llvm_compile functions

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: remove useless #includes
Marek Olšák [Wed, 15 Jan 2020 01:27:24 +0000 (20:27 -0500)]
radeonsi: remove useless #includes

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: move code for shader resources into si_shader_llvm_resources.c
Marek Olšák [Wed, 15 Jan 2020 01:17:08 +0000 (20:17 -0500)]
radeonsi: move code for shader resources into si_shader_llvm_resources.c

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: move geometry shader code into si_shader_llvm_gs.c
Marek Olšák [Wed, 15 Jan 2020 01:03:48 +0000 (20:03 -0500)]
radeonsi: move geometry shader code into si_shader_llvm_gs.c

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: remove llvm_type_is_64bit
Marek Olšák [Wed, 15 Jan 2020 00:29:34 +0000 (19:29 -0500)]
radeonsi: remove llvm_type_is_64bit

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: move tessellation shader code into si_shader_llvm_tess.c
Marek Olšák [Wed, 15 Jan 2020 00:13:42 +0000 (19:13 -0500)]
radeonsi: move tessellation shader code into si_shader_llvm_tess.c

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: move si_insert_input_* functions
Marek Olšák [Tue, 24 Dec 2019 00:53:46 +0000 (19:53 -0500)]
radeonsi: move si_insert_input_* functions

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3399>

4 years agoradeonsi: work around an LLVM crash when using llvm.amdgcn.icmp.i64.i1
Marek Olšák [Thu, 9 Jan 2020 02:52:26 +0000 (21:52 -0500)]
radeonsi: work around an LLVM crash when using llvm.amdgcn.icmp.i64.i1

Cc: 19.2 19.3 <mesa-stable@lists.freedesktop.org>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3338>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3338>

4 years agoradeonsi: fix si_build_wrapper_function for compute-based primitive culling
Marek Olšák [Thu, 9 Jan 2020 02:51:23 +0000 (21:51 -0500)]
radeonsi: fix si_build_wrapper_function for compute-based primitive culling

Fixes: 3b143369a55 "ac/nir, radv, radeonsi: Switch to using ac_shader_args"
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3338>

4 years agoradeonsi/gfx10: separate code for determining the number of vertices for NGG
Marek Olšák [Fri, 3 Jan 2020 23:02:30 +0000 (18:02 -0500)]
radeonsi/gfx10: separate code for determining the number of vertices for NGG

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
4 years agoradeonsi/gfx10: separate code for getting edgeflags from the gs_invocation_id VGPR
Marek Olšák [Fri, 3 Jan 2020 21:25:48 +0000 (16:25 -0500)]
radeonsi/gfx10: separate code for getting edgeflags from the gs_invocation_id VGPR

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
4 years agoradeonsi: move VS_STATE.LS_OUT_PATCH_SIZE a few bits higher to make space there
Marek Olšák [Tue, 24 Dec 2019 01:17:46 +0000 (20:17 -0500)]
radeonsi: move VS_STATE.LS_OUT_PATCH_SIZE a few bits higher to make space there

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
4 years agoradeonsi: make si_insert_input_* functions non-static
Marek Olšák [Tue, 24 Dec 2019 00:53:46 +0000 (19:53 -0500)]
radeonsi: make si_insert_input_* functions non-static

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
4 years agoac/cull: don't read Position.Z if it's not needed for culling
Marek Olšák [Tue, 7 Jan 2020 00:19:40 +0000 (19:19 -0500)]
ac/cull: don't read Position.Z if it's not needed for culling

It could be NULL.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
4 years agoradeonsi: separate code computing info for small primitive culling
Marek Olšák [Tue, 24 Dec 2019 03:16:42 +0000 (22:16 -0500)]
radeonsi: separate code computing info for small primitive culling

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
4 years agointel/compiler: Fix illegal mutation in get_nir_image_intrinsic_image
Kenneth Graunke [Wed, 15 Jan 2020 08:31:49 +0000 (00:31 -0800)]
intel/compiler: Fix illegal mutation in get_nir_image_intrinsic_image

get_nir_image_intrinsic_image() was incorrectly mutating the value held
by the register which holds the intrinsic's first source (image index).

If this happened to be the register for an SSA def which is also used
elsewhere in the program, this meant that we would clobber that value
in subsequent uses.

Note that this only affects i965, because neither anv nor iris use the
binding table start sections, so nothing is ever added here.

Fixes KHR-GL46.compute_shader.resources-max on i965 with Eric Anholt's
MR !3240 applied.  That MR reorders SSBOs and ABOs, so that test uses
image 0 and SSBO 0, causing this code to brilliantly add binding table
index 45 to both the image (correct) and the SSBO (bzzt, wrong!).

Fixes: 09f1de97a76 ("anv,i965: Lower away image derefs in the driver")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3404>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3404>

4 years agogitlab-ci: fix missing caselist.css/xsl
Rob Clark [Wed, 15 Jan 2020 16:50:55 +0000 (08:50 -0800)]
gitlab-ci: fix missing caselist.css/xsl

My best guess is that this was broken by d62dd8b0

Signed-off-by: Rob Clark <robdclark@chromium.org>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3413>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/merge_requests/3413>

4 years agorelnotes: Add Vulkan 1.2
Jason Ekstrand [Wed, 15 Jan 2020 15:25:51 +0000 (09:25 -0600)]
relnotes: Add Vulkan 1.2

4 years agoradv: enable Vulkan 1.2
Samuel Pitoiset [Mon, 25 Nov 2019 20:29:01 +0000 (21:29 +0100)]
radv: enable Vulkan 1.2

This bumps the Vulkan version to 1.2.128.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: implement Vulkan 1.2 features and properties
Samuel Pitoiset [Tue, 26 Nov 2019 07:34:33 +0000 (08:34 +0100)]
radv: implement Vulkan 1.2 features and properties

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: implement Vulkan 1.1 features and properties
Samuel Pitoiset [Tue, 26 Nov 2019 07:13:14 +0000 (08:13 +0100)]
radv: implement Vulkan 1.1 features and properties

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_timeline_semaphore for Vulkan 1.2
Samuel Pitoiset [Mon, 25 Nov 2019 19:50:12 +0000 (20:50 +0100)]
radv: update VK_KHR_timeline_semaphore for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_uniform_buffer_standard_layout for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 08:21:20 +0000 (10:21 +0200)]
radv: update VK_KHR_uniform_buffer_standard_layout for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_shader_subgroup_extended_types for Vulkan 1.2
Samuel Pitoiset [Tue, 26 Nov 2019 13:08:00 +0000 (14:08 +0100)]
radv: update VK_KHR_shader_subgroup_extended_types for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_shader_float_controls for Vulkan 1.2
Samuel Pitoiset [Tue, 26 Nov 2019 13:07:07 +0000 (14:07 +0100)]
radv: update VK_KHR_shader_float_controls for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_shader_float16_int8 for Vulkan 1.2
Samuel Pitoiset [Tue, 26 Nov 2019 13:05:13 +0000 (14:05 +0100)]
radv: update VK_KHR_shader_float16_int8 for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_shader_atomic_int64 for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 08:16:37 +0000 (10:16 +0200)]
radv: update VK_KHR_shader_atomic_int64 for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_imageless_framebuffer for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 08:15:04 +0000 (10:15 +0200)]
radv: update VK_KHR_imageless_framebuffer for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_image_format_list for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:26:20 +0000 (09:26 +0200)]
radv: update VK_KHR_image_format_list for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_driver_properties for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:24:07 +0000 (09:24 +0200)]
radv: update VK_KHR_driver_properties for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_draw_indirect_count for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:22:04 +0000 (09:22 +0200)]
radv: update VK_KHR_draw_indirect_count for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_depth_stencil_resolve for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:20:48 +0000 (09:20 +0200)]
radv: update VK_KHR_depth_stencil_resolve for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_create_renderpass2 for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:09:41 +0000 (09:09 +0200)]
radv: update VK_KHR_create_renderpass2 for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_buffer_device_address for Vulkan 1.2
Samuel Pitoiset [Tue, 26 Nov 2019 08:59:27 +0000 (09:59 +0100)]
radv: update VK_KHR_buffer_device_address for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_KHR_8bit_storage for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:03:28 +0000 (09:03 +0200)]
radv: update VK_KHR_8bit_storage for Vulkan 1.2

Promoted to Vulkan 1.2 with the KHR suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_EXT_scalar_block_layout for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 07:00:19 +0000 (09:00 +0200)]
radv: update VK_EXT_scalar_block_layout for Vulkan 1.2

Promoted to Vulkan 1.2 with the EXT suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_EXT_sampler_filter_minmax for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 06:58:55 +0000 (08:58 +0200)]
radv: update VK_EXT_sampler_filter_minmax for Vulkan 1.2

Promoted to Vulkan 1.2 with the EXT suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_EXT_host_query_reset for Vulkan 1.2
Samuel Pitoiset [Tue, 15 Oct 2019 06:55:07 +0000 (08:55 +0200)]
radv: update VK_EXT_host_query_reset for Vulkan 1.2

Promoted to Vulkan 1.2 with the EXT suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoradv: update VK_EXT_descriptor_indexing for Vulkan 1.2
Samuel Pitoiset [Tue, 26 Nov 2019 13:01:43 +0000 (14:01 +0100)]
radv: update VK_EXT_descriptor_indexing for Vulkan 1.2

Promoted to Vulkan 1.2 with the EXT suffix omitted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
4 years agoanv: Enable Vulkan 1.2 support
Iván Briano [Mon, 16 Sep 2019 22:41:45 +0000 (15:41 -0700)]
anv: Enable Vulkan 1.2 support

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
4 years agoanv: Implement the new core version property queries
Jason Ekstrand [Mon, 13 Jan 2020 19:44:16 +0000 (13:44 -0600)]
anv: Implement the new core version property queries

Vulkan 1.2 introduces some new structures to get the properties and
features of a device from extensions that were promoted to core in 1.1
and 1.2.  This commit implements the new property queries and makes all
of the corresponding extension queries map to them.

Reviewed-by: Iván Briano <ivan.briano@intel.com>