mesa.git
4 years agoi965: Implement ARB_compute_variable_group_size
Plamena Manolova [Mon, 12 Nov 2018 14:29:51 +0000 (16:29 +0200)]
i965: Implement ARB_compute_variable_group_size

This patch adds the implementation of ARB_compute_variable_group_size
for i965. We do this by storing the local group size in a push constant.

Additional changes made by Caio Marcelo de Oliveira Filho.

Signed-off-by: Plamena Manolova <plamena.manolova@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4504>

4 years agointel/compiler: Add support for variable workgroup size
Plamena Manolova [Mon, 12 Nov 2018 14:29:51 +0000 (06:29 -0800)]
intel/compiler: Add support for variable workgroup size

Add new builtin parameters that are used to keep track of the group
size.  This will be used to implement ARB_compute_variable_group_size.

The compiler will use the maximum group size supported to pick a
suitable SIMD variant.  A later improvement will be to keep all SIMD
variants (like FS) so the driver can select the best one at dispatch
time.

When variable workgroup size is used, the small workgroup optimization
is disabled as it we can't prove at compile time that the barriers
won't be needed.

Extracted from original i965 patch with additional changes by
Caio Marcelo de Oliveira Filho.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4504>

4 years agointel/compiler: Replace cs_prog_data->push.total with a helper
Caio Marcelo de Oliveira Filho [Sat, 21 Mar 2020 04:02:06 +0000 (21:02 -0700)]
intel/compiler: Replace cs_prog_data->push.total with a helper

The push.total field had three values but only one was directly
used (size).  Replace it with a helper function that explicitly takes
the cs_prog_data and the number of threads -- and use that in the
drivers.

This is a preparation for ARB_compute_variable_group_size where the
number of threads (hence the total size for push constants) is not
defined at compile time (not cs_prog_data->threads).

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4504>

4 years agoswr/rasterizer: Use private functions for min/max to avoid namespace issues.
Vinson Lee [Tue, 17 Mar 2020 05:49:39 +0000 (22:49 -0700)]
swr/rasterizer: Use private functions for min/max to avoid namespace issues.

This is a similiar fix as bb2287ccdf46 ("gallivm/tessellator: use
private functions for min/max to avoid namespace issues").

Fixes: ab5570820071 ("swr/rasterizer: Add tessellator implementation to the rasterizer")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Jan Zielinski <jan.zielinski@intel.com>
Tested-by: Jan Zielinski <jan.zielinski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4208>

4 years agotu: Implement descriptor set update templates
Connor Abbott [Tue, 24 Mar 2020 17:01:15 +0000 (18:01 +0100)]
tu: Implement descriptor set update templates

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agotu: Add missing code for immutable samplers
Connor Abbott [Tue, 24 Mar 2020 17:00:21 +0000 (18:00 +0100)]
tu: Add missing code for immutable samplers

Actually fill out the samplers, based on the radv implementation.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agotu: Emit CP_LOAD_STATE6 for descriptors
Connor Abbott [Mon, 23 Mar 2020 16:23:32 +0000 (17:23 +0100)]
tu: Emit CP_LOAD_STATE6 for descriptors

This restores the pre-loading of descriptor state, using the new
SS6_BINDLESS method that allows us to pre-load bindless resources.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agotu: Switch to the bindless descriptor model
Connor Abbott [Mon, 16 Mar 2020 10:49:19 +0000 (11:49 +0100)]
tu: Switch to the bindless descriptor model

Under the bindless model, there are 5 "base" registers programmed with a
64-bit address, and sam/ldib/ldc and so on each specify a base register
and an offset, in units of 16 dwords. The base registers correspond to
descriptor sets in Vulkan. We allocate a buffer at descriptor set
creation time, hopefully outside the main rendering loop, and then
switching descriptor sets is just a matter of programming the base
registers differently. Note, however, that some kinds of descriptors
need to be patched at command recording time, in particular dynamic
UBO's and SSBO's, which need to be patched at CmdBindDescriptorSets
time, and input attachments which need to be patched at draw time based
on the the pipeline that's bound. We reserve the fifth base register
(which seems to be unused by the blob driver) for these, creating a
descriptor set on-the-fly and combining all the dynamic descriptors from
all the different descriptor sets. This way, we never have to copy the
rest of the descriptor set at draw time like the blob seems to do. I
mostly chose to do this because the infrastructure was already there in
the form of dynamic_descriptors, and other drivers (at least radv) don't
cheat either when implementing this.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoir3: Rewrite UBO push analysis to support bindless
Connor Abbott [Mon, 23 Mar 2020 17:54:57 +0000 (18:54 +0100)]
ir3: Rewrite UBO push analysis to support bindless

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoir3: Plumb through bindless support
Connor Abbott [Thu, 19 Mar 2020 13:15:26 +0000 (14:15 +0100)]
ir3: Plumb through bindless support

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoir3: LDC also has a destination
Connor Abbott [Fri, 20 Mar 2020 14:25:59 +0000 (15:25 +0100)]
ir3: LDC also has a destination

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoir3: Also don't propagate immediate offset with LDC
Connor Abbott [Fri, 20 Mar 2020 14:25:11 +0000 (15:25 +0100)]
ir3: Also don't propagate immediate offset with LDC

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoir3: Plumb through support for a1.x
Connor Abbott [Wed, 18 Mar 2020 17:06:41 +0000 (18:06 +0100)]
ir3: Plumb through support for a1.x

This will need to be used in some cases for the upcoming bindless
support, plus ldc.k instructions which push data from a UBO to const
registers.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoir3: Add bindless instruction encoding
Connor Abbott [Fri, 6 Mar 2020 17:06:06 +0000 (18:06 +0100)]
ir3: Add bindless instruction encoding

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agofreedreno/a6xx: Add registers for the bindless model
Connor Abbott [Fri, 6 Mar 2020 10:29:54 +0000 (11:29 +0100)]
freedreno/a6xx: Add registers for the bindless model

In Vulkan, descriptors for samplers, SSBO's, etc. are collected into
descriptor sets, and shaders can use multiple descriptor sets. At
command-recording time, users can swap out only some of the descriptor
sets, and the driver is supposed to do the minimum amount necessary to
update any internal binding tables, knowing that only some of the
descriptors have changed.

With the old binding model, focused on GL, where there are separate
tables for each type of resource, we can do somewhat better than now by
preserving descriptors from lower descriptor sets when switching higher
descriptor sets. However we still have to copy around descriptors before
each draw.

At least for a6xx, qualcomm went further, essentially copying the Vulkan
binding model as an alternate way to load resources. There's an array of
registers (actually an array for compute and one for everything else),
where each register holds a pointer to a descriptor set that can contain
various different descriptor types. The descriptors are padded out to 16
dwords, so that every instruction can use an index instead of a dword
offset. It's called "bindless", I think, because it can also be used to
implement the old GL bindless extensions (presumably it allows more
samplers and textures than the old model).

This commit adds the register and cmdstream parts. Next up will be the
instruction encoding.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agofreedreno/a6xx: Add UBO size field
Connor Abbott [Fri, 6 Mar 2020 10:27:46 +0000 (11:27 +0100)]
freedreno/a6xx: Add UBO size field

Verified with the vulkan blob, which uses ldc and UBO descriptors, and
turnip will too soon.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agotu: ir3: Emit push constants directly
Connor Abbott [Wed, 18 Mar 2020 12:12:31 +0000 (13:12 +0100)]
tu: ir3: Emit push constants directly

Carve out some space at the beginning for push constants, and push them
directly, rather than remapping them to a UBO and then relying on the
UBO pushing code. Remapping to a UBO is easy now, where there's a single
table of UBO's, but with the bindless model it'll be a lot harder. I
haven't removed all the code to move the remaining UBO's over by 1,
though, because it's going to all get rewritten with bindless anyways.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agotu: Dump out shader assembly when requested
Connor Abbott [Thu, 26 Mar 2020 14:35:11 +0000 (15:35 +0100)]
tu: Dump out shader assembly when requested

We don't use the ir3 variant machinery, so we have to do this ourselves.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4358>

4 years agoaco: RA - move all std::function objects into proper functions
Daniel Schürmann [Tue, 7 Apr 2020 16:15:35 +0000 (17:15 +0100)]
aco: RA - move all std::function objects into proper functions

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: move all needed helper containers to ra_ctx
Daniel Schürmann [Tue, 7 Apr 2020 15:46:58 +0000 (16:46 +0100)]
aco: move all needed helper containers to ra_ctx

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: change live_out variables to std::unordered_set
Daniel Schürmann [Wed, 11 Mar 2020 10:02:20 +0000 (11:02 +0100)]
aco: change live_out variables to std::unordered_set

Improves performance of live_var_analysis for larger shaders

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: change some std::map to std::unordered_map in register_allocation
Daniel Schürmann [Wed, 11 Mar 2020 09:47:07 +0000 (10:47 +0100)]
aco: change some std::map to std::unordered_map in register_allocation

This improves compile times slightly for larger shaders

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: refactor try_remove_trivial_phi() in RA
Daniel Schürmann [Wed, 11 Mar 2020 07:38:48 +0000 (08:38 +0100)]
aco: refactor try_remove_trivial_phi() in RA

Minor refactoring to avoid some pointer chasing.
This patch also changes the live_out argument to be
passed by reference to avoid an unnecessary copy.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: improve speed of live_var_analysis
Daniel Schürmann [Tue, 10 Mar 2020 12:39:42 +0000 (13:39 +0100)]
aco: improve speed of live_var_analysis

by merging live_sgprs and live_vgprs sets.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: during RA only insert into renames table if a variable got renamed
Daniel Schürmann [Tue, 10 Mar 2020 11:41:02 +0000 (12:41 +0100)]
aco: during RA only insert into renames table if a variable got renamed

This improves the speed of register allocation.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: replace assignment hashmap by std::vector in register allocation
Daniel Schürmann [Tue, 10 Mar 2020 10:47:30 +0000 (11:47 +0100)]
aco: replace assignment hashmap by std::vector in register allocation

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: improve register assignment when live-range splits are necessary
Daniel Schürmann [Tue, 10 Mar 2020 10:50:41 +0000 (11:50 +0100)]
aco: improve register assignment when live-range splits are necessary

When finding a good place for a register, we can ignore
killed operands.

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: improve hashing for value numbering
Daniel Schürmann [Tue, 10 Mar 2020 09:00:32 +0000 (10:00 +0100)]
aco: improve hashing for value numbering

An improved hashing greatly reduces the number of collisions,
and thus, increases the speed for lookups in the hash table.
The hash function now uses Murmur3 written by Austin Appleby.

This patch also pre-reserves space for the hashmap to avoid rehashing.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: add explicit padding for all Instruction sub-structs
Daniel Schürmann [Mon, 30 Mar 2020 16:25:00 +0000 (17:25 +0100)]
aco: add explicit padding for all Instruction sub-structs

This patch also adds static_asserts on the size of Instructions
to ensure no internal padding is present.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoaco: guarantee that Temp fits in 4 bytes
Daniel Schürmann [Wed, 11 Mar 2020 12:12:08 +0000 (13:12 +0100)]
aco: guarantee that Temp fits in 4 bytes

Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4130>

4 years agoturnip: new clear/blit implementation with shader path fallback
Jonathan Marek [Fri, 13 Mar 2020 15:57:23 +0000 (11:57 -0400)]
turnip: new clear/blit implementation with shader path fallback

The shader path is used to implement the following cases:
* stencil aspect mask on D24S8 (for image_to_buffer,buffer_to_image)
* clear/copy msaa destination (2D engine can't have msaa dest)

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: add vk_format_is_snorm/is_float
Jonathan Marek [Wed, 8 Apr 2020 14:56:30 +0000 (10:56 -0400)]
turnip: add vk_format_is_snorm/is_float

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: rework format helpers
Jonathan Marek [Wed, 8 Apr 2020 14:56:16 +0000 (10:56 -0400)]
turnip: rework format helpers

* Take tile_mode as input directly
* tu6_format_gmem to tu6_base_format, use may not be limited to GMEM
* Add new helpers that will return the correct tile_mode as for image level
  as part of the format.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: use dirty bits for dynamic viewport/scissor state
Jonathan Marek [Wed, 8 Apr 2020 03:25:12 +0000 (23:25 -0400)]
turnip: use dirty bits for dynamic viewport/scissor state

CmdClearAttachments shader path will overwrite this state, so it needs to
be re-emitted with dirty bits in that case.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: save attachment samples in renderpass state
Jonathan Marek [Wed, 8 Apr 2020 02:23:27 +0000 (22:23 -0400)]
turnip: save attachment samples in renderpass state

This is needed to be able to know the number of samples during
CmdClearAttachments which can be used while the framebuffer is unknown.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: disable 8x msaa
Jonathan Marek [Wed, 8 Apr 2020 02:20:10 +0000 (22:20 -0400)]
turnip: disable 8x msaa

Not everything supports 8x msaa, and the blob doesn't support it at all.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: fix nir validate failure from push constant lowering
Jonathan Marek [Wed, 8 Apr 2020 01:39:40 +0000 (21:39 -0400)]
turnip: fix nir validate failure from push constant lowering

Fixes newly added checks in nir validate failing.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: split up gmem/tile alignment
Jonathan Marek [Tue, 18 Feb 2020 13:54:15 +0000 (08:54 -0500)]
turnip: split up gmem/tile alignment

Note: the x1/y1 align in tu6_emit_blit_scissor was broken

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoturnip: RB_CCU_CNTL fixes
Jonathan Marek [Fri, 13 Mar 2020 15:47:15 +0000 (11:47 -0400)]
turnip: RB_CCU_CNTL fixes

* Correct bypass value for a618
* Bypass value for blitter
* Don't set RB_CCU_CNTL again unnecessarily in tu6_emit_binning_pass

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agofreedreno/a6xx: set bypass RB_CCU_CNTL value for blitter
Jonathan Marek [Fri, 13 Mar 2020 14:20:23 +0000 (10:20 -0400)]
freedreno/a6xx: set bypass RB_CCU_CNTL value for blitter

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agofreedreno/registers: add RB_CCU_CNTL bitfields
Jonathan Marek [Fri, 13 Mar 2020 14:09:11 +0000 (10:09 -0400)]
freedreno/registers: add RB_CCU_CNTL bitfields

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3783>

4 years agoradv: allow TC-compat HTILE with GENERAL outside of render loops
Samuel Pitoiset [Thu, 9 Apr 2020 09:37:27 +0000 (11:37 +0200)]
radv: allow TC-compat HTILE with GENERAL outside of render loops

This gives +8% with Wolfeinstein Youngblood on my Vega64, and
according to someone else, it also improves performance with Doom
2016 and Wolfenstein 2 (and probably other ID Tech games).

This improvement is because Youngblood uses GENERAL for the main
depth-only pass and TC-compat HTILE is now enabled with GENERAL if
we know that we are outside of a render loop. This obviously also
reduces the number of HTILE decompressions from/to GENERAL.

Note that Youngblood violates the Vulkan spec regarding render loops
because they are only allowed with input attachments. Expect possible
rendering issues if apps use render loops with the wrong way (ie.
without input attachmens) because HTILE might not be coherent if
a depth-stencil texture is sampled and rendered in the same draw.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2704
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4391>

4 years agoradv: only enable TC-compat HTILE for images readable by a shader
Samuel Pitoiset [Tue, 31 Mar 2020 08:35:00 +0000 (10:35 +0200)]
radv: only enable TC-compat HTILE for images readable by a shader

If no texture fetches happen it's useless to enable TC-compat HTILE.

Because the driver currently doesn't support TC-compat HTILE for
storage images we don't have to check.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4497>

4 years agoradv: only expose fp16 control features for chips with double rate fp16
Samuel Pitoiset [Sun, 5 Apr 2020 07:42:50 +0000 (09:42 +0200)]
radv: only expose fp16 control features for chips with double rate fp16

This disables all fp16 shader control features on GFX8 because only
GFX9+ supports double rate packed math.

This improves consistency regarding other AMD Vulkan drivers.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4453>

4 years agoradv: only expose storageInputOutput16 for chips with double rate fp16
Samuel Pitoiset [Sun, 5 Apr 2020 07:33:43 +0000 (09:33 +0200)]
radv: only expose storageInputOutput16 for chips with double rate fp16

This feature allows to use both 16-bit integers and 16-bit floats
as inputs/outputs.

This disables storageInputOutput16 on GFX8 because only GFX9+ supports
double rate packed math.

This improves consistency regarding other AMD Vulkan drivers.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4453>

4 years agoradv: only expose shaderFloat16 for chips with double rate fp16
Samuel Pitoiset [Sun, 5 Apr 2020 07:25:18 +0000 (09:25 +0200)]
radv: only expose shaderFloat16 for chips with double rate fp16

This disables shaderFloat16 on GFX8 because only GFX9+ supports
double rate packed math.

This improves consistency regarding other AMD Vulkan drivers and
it makes no sense to enable that feature without packed math.

This also reduces performance with Wolfeinstein Youngblood if
fp16 is forced enabled on GFX8, while it's similar on GFX9.

We might re-introduce that feature in the future with ACO support
if it ends up being faster and correct.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4453>

4 years agoac,radv: add ac_gpu_info::has_double_rate_fp16
Samuel Pitoiset [Sun, 5 Apr 2020 07:23:16 +0000 (09:23 +0200)]
ac,radv: add ac_gpu_info::has_double_rate_fp16

Only GFX9+ support double rate packed math instructions.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4453>

4 years agoturnip: use buffer size instead of bo size for VFD_FETCH_SIZE
Jonathan Marek [Wed, 8 Apr 2020 23:43:24 +0000 (19:43 -0400)]
turnip: use buffer size instead of bo size for VFD_FETCH_SIZE

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4224>

4 years agoturnip: improve vertex input handling
Jonathan Marek [Wed, 18 Mar 2020 02:28:38 +0000 (22:28 -0400)]
turnip: improve vertex input handling

Emit vertexBindingDescriptionCount bindings, instead of one per attribute.

Verified with dEQP-VK.pipeline.vertex_input.*

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4224>

4 years agoradeonsi: fix Segmentation fault during vaapi enc test
James Zhu [Mon, 6 Apr 2020 20:34:01 +0000 (20:34 +0000)]
radeonsi: fix Segmentation fault during vaapi enc test

Fix Segmentation fault during vaapi enc test on Arcturus.

Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by Leo Liu <leo.liu@amd.com>

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4472>

4 years agoradv: Use correct buffer count with variable descriptor set sizes.
Bas Nieuwenhuizen [Wed, 8 Apr 2020 11:53:47 +0000 (13:53 +0200)]
radv: Use correct buffer count with variable descriptor set sizes.

Fixes dEQP-VK.binding_model.descriptorset_random.sets16.noarray.ubolimitlow.sbolimitlow.imglimitlow.iublimitlow.frag.ialimitlow.0

CC: <mesa-stable@lists.freedesktop.org>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2607
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4489>

4 years agoradv: Whitespace fixup.
Bas Nieuwenhuizen [Wed, 8 Apr 2020 10:51:34 +0000 (12:51 +0200)]
radv: Whitespace fixup.

Review comment that I did, but forgot to git add before amending ...

From https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4334

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4488>

4 years agoradv: set sparseAddressSpaceSize to RADV_MAX_MEMORY_ALLOCATION_SIZE
Samuel Iglesias Gonsálvez [Wed, 8 Apr 2020 08:57:28 +0000 (10:57 +0200)]
radv: set sparseAddressSpaceSize to RADV_MAX_MEMORY_ALLOCATION_SIZE

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4487>

4 years agoradv: check buffer size in vkCreateBuffer()
Samuel Iglesias Gonsálvez [Wed, 8 Apr 2020 08:55:37 +0000 (10:55 +0200)]
radv: check buffer size in vkCreateBuffer()

Fixes:

   dEQP-VK.api.buffer.basic.size_max_uint64

Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4487>

4 years agoradv: Consider maximum sample distances for entire grid.
Bas Nieuwenhuizen [Tue, 7 Apr 2020 20:23:09 +0000 (22:23 +0200)]
radv: Consider maximum sample distances for entire grid.

The other pixels in the grid might have samples with a larger
distance than the (0,0) pixel.

Fixes dEQP-VK.pipeline.multisample.sample_locations_ext.verify_location.samples_8_packed
when CTS is compiled with clang.

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4480>

4 years agoradv: enable lowering of GS intrinsics for the LLVM backend
Samuel Pitoiset [Fri, 13 Mar 2020 10:23:07 +0000 (11:23 +0100)]
radv: enable lowering of GS intrinsics for the LLVM backend

This replaces emit_vertex with:
   if (vertex_count < max_vertices) {
      emit_vertex_with_counter vertex_count ...
      vertex_count += 1
   }

Which is exactly what NIR->LLVM was doing but at NIR level. This
pass is already called by ACO.

pipeline-db changes on GFX10:
Totals from affected shaders:
SGPRS: 1952 -> 1912 (-2.05 %)
VGPRS: 2112 -> 2044 (-3.22 %)
Code Size: 189368 -> 185620 (-1.98 %) bytes
Max Waves: 494 -> 491 (-0.61 %)

No pipeline-db changes on other generations.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4182>

4 years agoradv: remove radv_layout_has_htile() helper
Samuel Pitoiset [Tue, 31 Mar 2020 13:26:00 +0000 (15:26 +0200)]
radv: remove radv_layout_has_htile() helper

The goal of this function was to return whether a depth-stencil image
has HTILE, in comparison to radv_layout_is_htile_compressed() which
is used to know whether a depth-stencil image has HTILE compressed.

These two functions are actually similar and they have never been
used for what they were supposed to. Remove radv_layout_has_htile()
in favour of radv_layout_is_htile_compressed() for now. If it's
needed in the future, I will re-introduce this concept properly.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4389>

4 years agoradv: cleanup creating the decompress/resummarize pipelines
Samuel Pitoiset [Tue, 31 Mar 2020 13:22:01 +0000 (15:22 +0200)]
radv: cleanup creating the decompress/resummarize pipelines

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4389>

4 years agoradv: rename extra graphics pipeline decompress/resummarize fields
Samuel Pitoiset [Tue, 31 Mar 2020 13:14:37 +0000 (15:14 +0200)]
radv: rename extra graphics pipeline decompress/resummarize fields

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4389>

4 years agoradv: rename decompress/resummarize depth/stencil functions
Samuel Pitoiset [Tue, 31 Mar 2020 13:09:58 +0000 (15:09 +0200)]
radv: rename decompress/resummarize depth/stencil functions

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4389>

4 years agoturnip: fix compute shaders crashing after geometry shader change
Jonathan Marek [Wed, 8 Apr 2020 01:13:31 +0000 (21:13 -0400)]
turnip: fix compute shaders crashing after geometry shader change

Fixes: 1af71bee734da7d8 ("turnip: Set has_gs in ir3_shader_key")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4483>

4 years agonir: make opt_if_loop_terminator() less strict
Timothy Arceri [Tue, 7 Apr 2020 13:33:55 +0000 (23:33 +1000)]
nir: make opt_if_loop_terminator() less strict

nir_cf_{extract,reinsert}() can't stitch a block together
if the block we are extracting ends in a jump but other jumps
nested in further ifs should be fine to move.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4477>

4 years agoradeonsi: don't lower constant arrays to uniforms in GLSL IR
Timothy Arceri [Tue, 7 Apr 2020 01:28:32 +0000 (11:28 +1000)]
radeonsi: don't lower constant arrays to uniforms in GLSL IR

This re-enables the change made in 2f5783bc2b82 which was
incorrectly disabled by 3e1dd99adca5.

For radeonsi, we will prefer the NIR pass as it'll generate better code
(some index calculation and a single load vs. a load, then index
calculation, then another load) and oftentimes NIR optimization can kick
in and make all the access indices constant.

Fixes: 3e1dd99adca5 ("radeonsi: Remove a bunch of default handling of pipe caps.")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4474>

4 years agomeson: fix debug build on Android
Dominik Behr [Tue, 22 Oct 2019 01:13:08 +0000 (18:13 -0700)]
meson: fix debug build on Android

debug_stack functions are implemented in another file for Android.
Also add backtrace library dependency.

Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-bu: Kristian H. Kristensen <hoegsber@google.com>
Signed-off-by: Dominik Behr <dbehr@chromium.org>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2435>

4 years agoradv: Store 64-bit availability bools if requested.
Bas Nieuwenhuizen [Thu, 26 Mar 2020 14:19:37 +0000 (15:19 +0100)]
radv: Store 64-bit availability bools if requested.

Fixes dEQP-VK.query_pool.*.reset_before_copy.* on RAVEN.

CC: <mesa-stable@lists.freedesktop.org>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2296
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4334>

4 years agogallivm: Add missing header for powf.
Vinson Lee [Tue, 7 Apr 2020 01:21:54 +0000 (18:21 -0700)]
gallivm: Add missing header for powf.

Fix build error after llvm-11 commit 3a29393b4709 ("Remove
math.h/cmath include from DataTypes.h").

src/gallium/auxiliary/gallivm/lp_bld_format_srgb.c: In function ‘lp_build_linear_to_srgb’:
src/gallium/auxiliary/gallivm/lp_bld_format_srgb.c:194:44: error: implicit declaration of function ‘powf’ [-Werror=implicit-function-declaration]
  194 |                                  exp2f_c * powf(coeff_f, 1.0f / exp_f));
      |                                            ^~~~
src/gallium/auxiliary/gallivm/lp_bld_format_srgb.c:194:44: warning: incompatible implicit declaration of built-in function ‘powf’
src/gallium/auxiliary/gallivm/lp_bld_format_srgb.c:78:1: note: include ‘<math.h>’ or provide a declaration of ‘powf’
   77 | #include "lp_bld_format.h"
  +++ |+#include <math.h>
   78 |

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4473>

4 years agoturnip: Drop dep_llvm from dependencies
Kristian H. Kristensen [Tue, 7 Apr 2020 16:34:42 +0000 (09:34 -0700)]
turnip: Drop dep_llvm from dependencies

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4478>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4478>

4 years agoturnip: Make Android platform build
Kristian H. Kristensen [Tue, 7 Apr 2020 16:04:00 +0000 (09:04 -0700)]
turnip: Make Android platform build

We still don't have a way to keep this from breaking, but I don't
think this ever built.  Let's call it progress.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4478>

4 years agoturnip: Stub out VK_KHR_external_{fence,semaphore}_fd
Kristian H. Kristensen [Tue, 7 Apr 2020 15:57:10 +0000 (08:57 -0700)]
turnip: Stub out VK_KHR_external_{fence,semaphore}_fd

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4478>

4 years agoturnip: Add missing VKAPI_ATTR annotations
Kristian H. Kristensen [Tue, 7 Apr 2020 15:45:03 +0000 (08:45 -0700)]
turnip: Add missing VKAPI_ATTR annotations

Make sure the types match.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4478>

4 years agotracie: Reformat code to fix indentation
Rohan Garg [Tue, 7 Apr 2020 11:17:07 +0000 (13:17 +0200)]
tracie: Reformat code to fix indentation

Signed-off-by: Rohan Garg <rohan.garg@collabora.com>
Reviewed-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4435>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4435>

4 years agotracie: Print results in a machine readable format
Rohan Garg [Wed, 1 Apr 2020 16:06:49 +0000 (18:06 +0200)]
tracie: Print results in a machine readable format

Signed-off-by: Rohan Garg <rohan.garg@collabora.com>
Reviewed-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4435>

4 years agofreedreno/a6xx: Set a level's pitch based on minified level0 pitch, not width0.
Eric Anholt [Tue, 25 Feb 2020 22:03:36 +0000 (14:03 -0800)]
freedreno/a6xx: Set a level's pitch based on minified level0 pitch, not width0.

Found from piglit fbo-generatemipmaps failures, then tracked down with the
texturator test.  The piece that really revealed things was finding that
1024x1 linear RGBA8 on the older blob drivers would have a pitch of 5120
instead of 4096, and the following levels minified that pitch.

Fixes ~124 piglit tests (~8.5% of piglit failures) on cheza.

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>

4 years agofreedreno: Add the outline of a test for a6xx texture layout.
Eric Anholt [Wed, 26 Feb 2020 19:27:04 +0000 (11:27 -0800)]
freedreno: Add the outline of a test for a6xx texture layout.

Trying to work out texture layout by remembering what things looked like
in texturator is hard.  Instead, let's use texture layouts from tracing
the blob as a source of truth to make sure that we pick the same layouts
they do (and don't break known-good ones).  More testcases will be added
as I fix layout bugs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>

4 years agofreedreno/a6xx: Drop the "alignment" layout temporary.
Eric Anholt [Tue, 25 Feb 2020 22:42:33 +0000 (14:42 -0800)]
freedreno/a6xx: Drop the "alignment" layout temporary.

It's just 1 for !3d, which means that the align we're doing in that case
is pointless.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>

4 years agofreedreno/a6xx: Remove the "aligned_height" temporary.
Eric Anholt [Tue, 25 Feb 2020 22:40:37 +0000 (14:40 -0800)]
freedreno/a6xx: Remove the "aligned_height" temporary.

Now that we're not incrementally minifying height, we can just modify it.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>

4 years agofreedreno/a6xx: Sink the per-level size temps inside the loop.
Eric Anholt [Thu, 27 Feb 2020 20:35:32 +0000 (12:35 -0800)]
freedreno/a6xx: Sink the per-level size temps inside the loop.

u_minify(n, 1) is no cheaper than u_minify(n, level), and this makes the
logic a lot simpler to follow.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3987>

4 years agogitlab-ci: Run merge request pipelines automatically only for Marge Bot
Michel Dänzer [Fri, 3 Apr 2020 09:46:12 +0000 (11:46 +0200)]
gitlab-ci: Run merge request pipelines automatically only for Marge Bot

MR pipelines not triggered by Marge Bot can still be triggered manually.

Motivation: The main & forked Mesa project CI pipelines combined are
currently generating over 1 TB of egress traffic per week. ~80% of this
is from pre-merge pipelines. Assuming this corresponds to 4 pre-merge
and one post-merge pipeline per MR on average, this change could
potentially eliminate up to ~60% of the overall traffic (by preventing
3 of the 4 pre-merge pipelines from running automatically).

(Of course, this could be subverted if all jobs of the other pipelines
were triggered manually anyway... In most cases, manually triggering
just a few jobs should suffice)

v2:
* $GITLAB_USER_NAME was the wrong variable, $GITLAB_USER_LOGIN should
  do the trick.

Suggested-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4432>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4432>

4 years agogitlab-ci: Don't require triggering build/test jobs manually
Michel Dänzer [Fri, 3 Apr 2020 10:50:11 +0000 (12:50 +0200)]
gitlab-ci: Don't require triggering build/test jobs manually

Let them run automatically once all their dependencies have passed.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4432>

4 years agogitlab-ci/lava: Add needs: for container image to test jobs (again)
Michel Dänzer [Fri, 3 Apr 2020 13:59:31 +0000 (15:59 +0200)]
gitlab-ci/lava: Add needs: for container image to test jobs (again)

Without this, the test jobs could spuriously run after the container
job failed or was cancelled, even if the build job didn't run at all.

(I already did this in 94cfe590703018cf3d34a0c1f8667064919bf843, but it
got dropped accidentally in 22d976454f4e50142116f4544c0bbf11134ce991)

Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4432>

4 years agogitlab-ci: Rename "paths" YAML anchor to "all_paths"
Michel Dänzer [Fri, 3 Apr 2020 09:17:48 +0000 (11:17 +0200)]
gitlab-ci: Rename "paths" YAML anchor to "all_paths"

To avoid confusion with `paths:` elements.

Reviewed-by: Adam Jackson <ajax@redhat.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4432>

4 years agoanv/gen12: Lower VK_KHR_multiview using Primitive Replication
Caio Marcelo de Oliveira Filho [Tue, 27 Mar 2018 17:10:34 +0000 (10:10 -0700)]
anv/gen12: Lower VK_KHR_multiview using Primitive Replication

Identify if view_index is used only for position calculation, and use
Primitive Replication to implement Multiview in Gen12.  This feature
allows storing per-view position information in a single execution of
the shader, treating position as an array.

The shader is transformed by adding a for-loop around it, that have an
iteration per active view (in the view_mask).  Stores to the position
now store into the position array for the current index in the loop,
and load_view_index() will return the view index corresponding to the
current index in the loop.

The feature is controlled by setting the environment variable
ANV_PRIMITIVE_REPLICATION_MAX_VIEWS, which defaults to 2 if unset.
For pipelines with view counts larger than that, the regular
instancing will be used instead of Primitive Replication.  To disable
it completely set the variable to 0.

v2: Don't assume position is set in vertex shader; remove only stores
    for position; don't apply optimizations since other passes will
    do; clone shader body without extract/reinsert; don't use
    last_block (potentially stale). (Jason)

    Fix view_index immediate to contain the view index, not its order.
    Check for maximum number of views supported.
    Add guard for gen12.

v3: Clone the entire shader function and change it before reinsert;
    disable optimization when shader has memory writes. (Jason)

    Use a single environment variable with _DEBUG on the name.

v4: Change to use new nir_deref_instr.
    When removing stores, look for mode nir_var_shader_out instead
    of the walking the list of outputs.
    Ensure unused derefs are removed in the non-position part of the
    shader.
    Remove dead control flow when identifying if can use or not
    primitive replication.

v5: Consider all the active shaders (including fragment) when deciding
    that Primitive Replication can be used.
    Change environment variable to ANV_PRIMITIVE_REPLICATION.
    Squash the emission of 3DSTATE_PRIMITIVE_REPLICATION into this patch.
    Disable Prim Rep in blorp_exec_3d.

v6: Use a loop around the shader, instead of manually unrolling, since
    the regular unroll pass will kick in.
    Document that we don't expect to see copy_deref or load_deref
    involving the position variable.
    Recover use_primitive_replication value when loading pipeline from
    the cache.
    Set VARYING_SLOT_LAYER to 0 in the shader.  Earlier versions were
    relying on ForceZeroRTAIndexEnable but that might not be
    sufficient.
    Disable Prim Rep in cmd_buffer_so_memcpy.

v7: Don't use Primitive Replication if position is not set, fallback
    to instancing; change environment variable to be
    ANV_PRIMITVE_REPLICATION_MAX_VIEWS and default it to 2 based on
    experiments.

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2313>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2313>

4 years agointel/fs: Allow multiple slots for position
Caio Marcelo de Oliveira Filho [Fri, 21 Sep 2018 23:07:38 +0000 (16:07 -0700)]
intel/fs: Allow multiple slots for position

Change brw_compute_vue_map() to also take the number of pos slots.  If
more than one slot is used, the VARYING_SLOT_POS is treated as an
array.

When using Primitive Replication, instead of a single position, the
VUE must contain an array of positions.  Padding might be
necessary (after clip distance) to ensure rest of attributes start
aligned.

v2: Add note about array in the commit message and assert that
    pos_slots >= 1 to make clear 0 is invalid. (Jason)
    Move padding to be after the clip distance.

v3: Apply the correct offset when gathering the sources from outputs.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> [v2]
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2313>

4 years agointel/gen12: Add XML description for 3DSTATE_PRIMITIVE_REPLICATION
Caio Marcelo de Oliveira Filho [Sat, 12 Oct 2019 00:04:36 +0000 (17:04 -0700)]
intel/gen12: Add XML description for 3DSTATE_PRIMITIVE_REPLICATION

v2: Use groups for the 16-element arrays "Viewport Offset"
    and "RTAI Offset". (Ken)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2313>

4 years agonir: Add per_view attribute to nir_variable
Caio Marcelo de Oliveira Filho [Tue, 11 Feb 2020 22:41:05 +0000 (14:41 -0800)]
nir: Add per_view attribute to nir_variable

If a nir_variable is tagged with per_view, it must be an array with
size corresponding to the number of views.  For slot-tracking, it is
considered to take just the slot for a single element -- drivers will
take care of expanding this appropriately.

This will be used to implement the ability of having per-view position
in a vertex shader in Intel platforms.

Acked-by: Rafael Antognolli <rafael.antognolli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2313>

4 years agomesa: add support for NV_pixel_buffer_object
Simon Ser [Thu, 2 Apr 2020 20:53:41 +0000 (22:53 +0200)]
mesa: add support for NV_pixel_buffer_object

Signed-off-by: Simon Ser <contact@emersion.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4422>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4422>

4 years agoturnip: implement timestamp query
Jonathan Marek [Tue, 3 Mar 2020 01:52:15 +0000 (20:52 -0500)]
turnip: implement timestamp query

Passes tests in:
dEQP-VK.pipeline.timestamp.*

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4027>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4027>

4 years agoturnip: Enable geometryShader device feature
Brian Ho [Thu, 2 Apr 2020 18:01:54 +0000 (11:01 -0700)]
turnip: Enable geometryShader device feature

Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Enable geometry shaders for CP_DRAWs
Brian Ho [Wed, 1 Apr 2020 20:36:31 +0000 (13:36 -0700)]
turnip: Enable geometry shaders for CP_DRAWs

Enable geometry shading on draw if the pipeline has a geometry
stage.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Populate tu_pipeline.active_stages
Brian Ho [Mon, 6 Apr 2020 20:38:04 +0000 (13:38 -0700)]
turnip: Populate tu_pipeline.active_stages

This can be used to determine if the pipeline has a specific shader
stage (e.g. geometry shader).

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Update maxGeometryShaderInvocations to match blob
Brian Ho [Fri, 3 Apr 2020 14:59:47 +0000 (07:59 -0700)]
turnip: Update maxGeometryShaderInvocations to match blob

Geometry shaders support an invocations parameter up to a limit
defined by maxGeometryShaderInvocations. This was set to 127, but
executing with invocations > 32 causes a crash. As it turns out, the
blob only advertises a max of 32 invocations, so we set that in
turnip as well.

Fixes dEQP-VK.geometry.instanced.draw_*_instances_{127, 64}_geometry_invocations

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Selectively configure GRAS_LAYER_CNTL
Brian Ho [Fri, 3 Apr 2020 14:57:25 +0000 (07:57 -0700)]
turnip: Selectively configure GRAS_LAYER_CNTL

One of the features of geometry shaders is the ability to render to
different layers by assigning to the gl_Layer (Layer in SPIR-V)
builtin.

While have already plumbed the layer regid to the geometry shader,
we also need to GRAS_LAYER_CNTL to actually use layered rendering.
In addition, gmem does not support layered rendering, so we need to
force sysmem.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Set up REG_A6XX_SP_GS_CONFIG
Brian Ho [Wed, 1 Apr 2020 20:26:17 +0000 (13:26 -0700)]
turnip: Set up REG_A6XX_SP_GS_CONFIG

Updates GS_CONFIG and HLSQ_GS_CNTL registers to match those emitted
by the blob and fd.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Configure VFD_CONTROL with gsheader and primitiveid
Brian Ho [Wed, 1 Apr 2020 20:21:26 +0000 (13:21 -0700)]
turnip: Configure VFD_CONTROL with gsheader and primitiveid

This commit updates VFD_CONTROL to use the GS header and primitive
ID sysvals if a geometry shader stage is present in the pipeline.
Like in the case of VPC, the code here is adapted from fd6_program.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Configure VPC for geometry shaders
Brian Ho [Wed, 1 Apr 2020 18:50:55 +0000 (11:50 -0700)]
turnip: Configure VPC for geometry shaders

This commit updates tu6_emit_vpc to selectively emit GS-specifc
configuration. Most of this is repurposed from fd6_program.c.

This also refactors `link_geometry_stages` to ir3_nir_lower_tess.c
so it can be shared between fd and tu.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Emit geometry shader obj and related consts
Brian Ho [Wed, 1 Apr 2020 18:09:48 +0000 (11:09 -0700)]
turnip: Emit geometry shader obj and related consts

Like with other shader types, we need to emit the geometry shader
object and the consts it uses. In addition, we need to emit
additional geometry-specific consts that link primitive/vertex stride
between the vs and gs. In conjunction with the gsheader, these are
used by the vs to determine where to stlw outputs and used by the gs
to determine where to ldlw those outputs from.

FD emits these consts in the draw call because in GL, you can mix
and match shaders in different programs. In Vulkan, however, we
compile and link the shaders at pipeline creation, so we can emit
these in the pipeline IB instead.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoturnip: Set has_gs in ir3_shader_key
Brian Ho [Wed, 1 Apr 2020 16:01:52 +0000 (09:01 -0700)]
turnip: Set has_gs in ir3_shader_key

The ir3 compiler only lowers the VS and GS for geometry shading if
the corresponding has_gs key is set in the shader key. Without it,
GS-specific intrinsics like load_per_vertex_input won't get lowered
and the GS header will be initialized with invalid values.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4436>

4 years agoradv: Print shader stage before disassembly.
Timur Kristóf [Wed, 1 Apr 2020 11:43:50 +0000 (13:43 +0200)]
radv: Print shader stage before disassembly.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-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/3576>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3576>

4 years agoaco: Print shader stage in aco_print_program.
Timur Kristóf [Wed, 1 Apr 2020 11:39:25 +0000 (13:39 +0200)]
aco: Print shader stage in aco_print_program.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3576>

4 years agoradv: Enable ACO for NGG VS/TES, but disable NGG for ACO GS.
Timur Kristóf [Tue, 31 Mar 2020 08:41:01 +0000 (10:41 +0200)]
radv: Enable ACO for NGG VS/TES, but disable NGG for ACO GS.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3576>