v3dv: use imm-helpers

Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23855>

broadcom/compiler: use imm-helpers

Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23855>

vulkan: avoid needless constant-folding

While we're at it, also switch to the nir_f{add,mul}_imm helpers.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23855>

mesa/st: use nir_ineg

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23855>

microsoft/compiler: use nir_imm_zero

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23855>

mapi: Remove dead struct _glapi_function in glapi/glapi_getproc.c

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23879>

mapi: Merge get_static_proc_address into _glapi_get_proc_address

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23879>

mapi: Style fixes in glapi/glapi_getproc.c

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23879>

util: sizeof bucket are always 32bit width, use align instead align64

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23732>

util: Do not use align64 over unsigned int in register_allocate.c

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23732>

util: Replace the usage of redundant u_align_u32 with align and remove u_align_u32

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23732>

util: Getting align and align64 consistence with ALIGN

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23732>

util: use uint32_t instead of unsigned in bitscan.h

uint32_t is more exact than unsigned for these functions

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23732>

util: Add function util_is_power_of_two_nonzero64 in bitscan.h

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23732>

pvr: Rename rogue_fw.xml -> rogue_kmd_stream.xml.

The UMD does not care if firmware is used, and the current name isn't
very informative either.

Signed-off-by: Donald Robson <donald.robson@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23872>

pvr: Rename transfer 3D heap to transfer frag heap

This better matches the naming throughout the rest of the driver.

Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23872>

pvr: Merge main and extension command streams

Signed-off-by: Sarah Walker <sarah.walker@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23872>

pvr: Rename heap reserved area to static data carveout

Signed-off-by: Sarah Walker <sarah.walker@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23872>

pvr: use pvr_csb_pack() to setup CR_FB_CDC_ZLS

Signed-off-by: Sarah Walker <sarah.walker@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23872>

pvr: Fragment register fb_cdc_zls is feature dependent

Signed-off-by: Sarah Walker <sarah.walker@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23872>

microsoft/clc: Don't build compiler test if build-tests is false

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8161

Cc: mesa-stable
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23890>

d3d12: Fix usage of D3D12_VIDEO_ENCODER_RATE_CONTROL_FLAG, was using D3D12_VIDEO_ENCODER_SUPPORT_FLAG wrongly instead

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23904>

d3d12: Only set reduced_tx_set when supported by D3D12 caps (no libva caps for reduced_tx_set to map to)

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23904>

d3d12: Correct tx_mode_support reporting as specified in libva spec

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23904>

compiler: set alignment=1 by default for handling empty struct/interface in glsl_types.cpp

When there is no elements in struct/interface, the alignment of it should be 1 instead of 0.

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23841>

radv: Do not enable robustness for push constants with robustBufferAccess2

There is no spec text requiring this behaviour, it is only for buffers.

Signed-off-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23885>

nir/print: Print more representations in load_const

In addition to the hexadecimal and float (when applicable), print the
signed and unsigned representations.  Representations may be omitted based
on information about the value:

- If gather types has unambiguous information, we use it;
- Float is omitted for 8 bit values;
- Signed decimal is omitted for positive values;
- Unsigned decimal is omitted for small values (representation is same as hex);

Note for now the "terse form" that appear in SSA uses is unchaged.

Based on a patch by Mike Blumenkrantz.

Examples:

```
// Just used as float. Omitted decimals.
vec4 32 ssa_81 = load_const (0x3f800000, 0x3f800000, 0x3e4ccccd, 0x3f800000) = (1.000000, 1.000000, 0.200000, 1.000000)
vec1 32 ssa_28 = load_const (0x3e4ccccd = 0.200000)

// Just a small integer. Omitted float and decimal.
vec1 32 ssa_45 = load_const (0x00000001)

// Larger positive integers. Omitted float.
vec1 32 ssa_39 = load_const (0x00002000 = 8192)
vec1 32 ssa_30 = load_const (0x000000ff = 255)
vec1 32 ssa_28 = load_const (0x00000010 = 16)

// Integers with negative values.
load_const (0xff = -1 = 255)
load_const (0xff80 = -128 = 65408)
load_const (0xffff = -1 = 65535)

// Same value, in the first case we know is used as an integer.
load_const (0xffffffe0 = -32 = 4294967264)
load_const (0xffffffe0 = -nan = -32 = 4294967264)
```

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

nir/print: Use src_type when printing consts in SSA uses

If the src_type is not available, untie by looking at the results from
nir_gather_ssa_types(). If that is ambiguous, just pick uint.

Now in print_const_from_load() when the type is invalid, print the full
constant form (with both padded hex and float); when the passed type
is valid, print the terse form based on it.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

nir: Extract logic to get dest and srcs types from intrinsic

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

nir: Make a const-friendly way to get the offset_src and arrayed_io_src from intrinsic

The existing helper returns a `nir_src *` so expects a non-const instr.

We plan to use this function in queries that don't modify the shader, so
create (and use internally) a variant that returns the index instead.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

nir/print: Make NIR_DEBUG=print_consts behavior the default

Now there's a NIR_DEBUG=print_no_inline_consts to omit them.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

nir/print: Improve NIR_PRINT=print_consts by using nir_gather_ssa_types()

The two representations are *always* used for `load_const`, but when
inlining the value as SSA source, use just a single terse
representation.

The choice between integer or float is based on the result of
nir_gather_ssa_types(), with a bias for integer when in doubt.

Also remove extra comment `/* */` syntax since the value is already
enclosed by parenthesis.

---

For illustration, here's some instructions from crucible test
func.shader.averageRounded.uint64_t with NIR_DEBUG=print_consts:

BEFORE:

```
vec1 32 con ssa_23 = load_const (0xfffffffc = -nan)
vec1 32 div ssa_24 = iand ssa_13, ssa_23 /*(0xfffffffc = -nan)*/
vec1 32 con ssa_25 = load_const (0x00000024 = 0.000000)
vec1 32 con ssa_26 = intrinsic load_ubo (ssa_1 /*(0x00000002 = 0.000000)*/, ssa_25 /*(0x00000024 = 0.000000)*/) (access=0, align_mul=1073741824, align_offset=36, range_base=0, range=-1)
vec1 32 con ssa_27 = load_const (0x00000008 = 0.000000)
vec1 32 con ssa_28 = load_const (0x00000007 = 0.000000)
vec1 32 con ssa_29 = iand ssa_4.y, ssa_1 /*(0x00000002 = 0.000000)*/
vec1 32 con ssa_30 = ishl ssa_29, ssa_28 /*(0x00000007 = 0.000000)*/
vec1 32 con ssa_31 = load_const (0x7b000808 = 664776890994587263929995856502063104.000000)
vec1 32 con ssa_32 = ior ssa_31 /*(0x7b000808 = 664776890994587263929995856502063104.000000)*/, ssa_30
```

AFTER:

```
vec1 32 con ssa_23 = load_const (0xfffffffc = -nan)
vec1 32 div ssa_24 = iand ssa_13, ssa_23 (0xfffffffc)
vec1 32 con ssa_25 = load_const (0x00000024 = 0.000000)
vec1 32 con ssa_26 = intrinsic load_ubo (ssa_1 (0x2), ssa_25 (0x24)) (access=0, align_mul=1073741824, align_offset=36, range_base=0, range=-1)
vec1 32 con ssa_27 = load_const (0x00000008 = 0.000000)
vec1 32 con ssa_28 = load_const (0x00000007 = 0.000000)
vec1 32 con ssa_29 = iand ssa_4.y, ssa_1 (0x2)
vec1 32 con ssa_30 = ishl ssa_29, ssa_28 (0x7)
vec1 32 con ssa_31 = load_const (0x7b000808 = 664776890994587263929995856502063104.000000)
vec1 32 con ssa_32 = ior ssa_31 (0x7b000808), ssa_30
```

and some instructions from crucible test func.gs.basic with NIR_DEBUG=print_consts,
now showing float representation being selected:

BEFORE:

```
vec4 32 ssa_10 = load_const (0x3e4ccccd, 0x3e4ccccd, 0x00000000, 0x00000000) = (0.200000, 0.200000, 0.000000, 0.000000)
vec4 32 ssa_9 = intrinsic load_deref (ssa_42) (access=0)
vec4 32 ssa_11 = fadd ssa_9, ssa_10 /*(0x3e4ccccd, 0x3e4ccccd, 0x00000000, 0x00000000) = (0.200000, 0.200000, 0.000000, 0.000000)*/
```

AFTER:

```
vec4 32 ssa_10 = load_const (0x3e4ccccd, 0x3e4ccccd, 0x00000000, 0x00000000) = (0.200000, 0.200000, 0.000000, 0.000000)
vec4 32 ssa_9 = intrinsic load_deref (ssa_42) (access=0)
vec4 32 ssa_11 = fadd ssa_9, ssa_10 (0.200000, 0.200000, 0.000000, 0.000000)
```

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

nir: Allow nir_gather_ssa_types() to ignore regs instead of assert

If we infer a type for a reg, just ignore and keep going.  This will allow
to use this pass even when registers are present.

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23562>

radv/rt: Hash stages using radv_hash_shaders

The hash also depends on the radv_pipeline_key as well as the flags. The
pipeline layout will also play a role when we implement inline
descriptor sets and push constants.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23747>

radv/rt: Fix caching non-recursive stages

The hash used for insertion is calculated in a different way than the
hash used for lookup.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23747>

util: Do not include immintrin.h in half_float.h

The files included are extremely large and hurt compile time of
everything that inludes half_float.h directly or indirectly.

Compile time of a fresh RADV build:
before 32.477s 32.661s 32.625s
after  25.116s 24.928s 25.114s

v2: Include xmmintrin instead (Marek Olšák)
after  25.552s 25.811s 25.678s

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23871>

amd/ci: add another dEQP-VK.multiview.renderpass2.multisample.* flake

https://gitlab.freedesktop.org/mesa/mesa/-/jobs/44557372

Signed-off-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23910>

vulkan/wsi: check for dri3 buffer initialization failure

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8427
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Lina Versace <linyaa@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23081>

freedreno/registers: add bitfield for DSI wide bus enablement

Add a bitfield controlling wide bus enablement for DPU<->DSI interface.

Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23829>

draw/i915: move hwfmt array to i915 specific struct

There's no point in bloating the vertex_info struct everywhere with
information that's only used by i915 in a single place. Let's explicitly
store the hwinfo when needed, instead of piggy-backing on vertex_info.

Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23851>

radv: inline more values in radv_emit_fb_ds_state()

These are no longer adjusted.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23887>

radv: stop emitting TILE_SURFACE_ENABLE for the ZRANGE_PRECISION workaround

The only case that matters is when the fb is emitted, but HTILE is
already disabled there using DB_RENDER_CONTROL.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23887>

d3d12: Fixes unused-variable compile error

The compile error message is:
../../src/gallium/drivers/d3d12/d3d12_video_screen.cpp:481:70: error: unused variable ‘sliceData’ [-Werror=unused-variable]
  481 |    D3D12_VIDEO_ENCODER_PICTURE_CONTROL_SUBREGIONS_LAYOUT_DATA_SLICES sliceData = { };
      |                                                                      ^~~~~~~~~
cc1plus: all warnings being treated as errors

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23900>

pvr: Setup ZLS depth and stencil load/store separately

Previously the code assumed that you could only have depth-stencil
attachments so no stencil only or depth only, for ZLS load/stores.
This isn't true as we can have stencil only attachments so the
ZLS depth and stencil store/load enable have to be set separately.

Other ZLSCTL setup has also been adjusted for separate depth-stencil.
E.g. the z{load,store}format, and {load,store}twiddled.

Co-Authored-By: Soroush Kashani <soroush.kashani@imgtec.com>
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Signed-off-by: Soroush Kashani <soroush.kashani@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23830>

v3dv: add a linear images to buffer copy codepath

Called copy_image_to_buffer_texel_buffer, that reuses
copy_image_linear_texel_buffer, by setting up a image destination from
the buffer destination.

This fixes new ycbcr tests added recently (1.3.6.0) like:
dEQP-VK.ycbcr.copy.*.*.*buffer*

that were failing due lack of a codepath handling them.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23864>

v3dv: refactor copy_image_to_buffer_blit

In order to have common code to create a image from a buffer, that we
plan to use later on a new codepath.

This refactor adds three new methods:
 * One that gathers all the info required to create the structures and
   implement the operation
 * One that creates the image from the buffer, based on that info
 * One that creates a BlitRegion from that info

This seems like too much splitting, but we needed to do it in this
way, because we can't ensure that future uses of this common code
would use a BlitRegion.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23864>

radv: allow NV_device_generated_commands with RADV_DEBUG=noibs

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23791>

radv/amdgpu: add support for executing DGC cmdbuf with RADV_DEBUG=noibs

This contains some preliminary work to be able to execute DGC cmdbuf
on the compute queue because IB2 doesn't exist.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23791>

radv/amdgpu: add more small helpers for managing CS

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23791>

CI/windows: Update headers and Agility redist to 1.711.3-preview

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23811>

d3d12: AV1 Encode

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23811>

frontends/va: Extend AV1 Encode params

Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23811>

nir: Convert to nir_foreach_function_impl

Done by hand at each call site but going very quickly with funny Vim motions and
common regexes. This is a very common idiom in NIR.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23807>

nir: Add nir_foreach_function_impl helper

Most users of nir_foreach_function actually want the nir_function_impl, not the
nir_function, and want to skip empty functions (though some graphics-specific
passes sometimes fail to do that part). Add a nir_foreach_function_impl macro
to make that case more ergonomic.

   nir_foreach_function_impl(impl, shader) {
      ...
      foo(impl)
   }

is equivalent to:

   nir_foreach_function(func, shader) {
      if (func->impl) {
         ...
         foo(func->impl);
      }
   }

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23807>

docs/rusticl: add Enabling section

Signed-off-by: Karol Herbst <git@karolherbst.de>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23859>

docs/rusticl: mark building section as such

Signed-off-by: Karol Herbst <git@karolherbst.de>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23859>

intel/devinfo/i915: Set has_set_pat_uapi for MTL+

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

anv: Use set PAT extension on BO creation for MTL

Reworks:
 * Drop local pat_index var (suggested by José)

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

iris: Use set PAT extension on BO creation for MTL

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

drm-uapi/i915_drm.h: Update from drm-next (2023-06-09)

git://anongit.freedesktop.org/drm/drm 2222dcb0775d36de28992f56455ab3967b30d380

The motivation for this change in to get the uapi changes from:

commit 81b1b599dfd71c958418dad586fa72c8d30d1065
Author: Fei Yang <fei.yang@intel.com>
Date:   Tue Jun 6 12:00:42 2023 +0200

    drm/i915: Allow user to set cache at BO creation

Specifically, the I915_GEM_CREATE_EXT_SET_PAT extension.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

anv: Swap ordering of memory types on non-LLC platforms to work around application bugs.

The Vulkan specification indicates that if memory types have
properties which are a strict subset of another type's, then they
should appear before that memory type.  Otherwise the specification
does not require a specific ordering of memory types.

But, it appears that Aztec Ruins and the Vulkan CTS make an assumption
that the first host-accessible memory type is host-coherent and select
it when they expect data written by the CPU to become visible without
calling vkFlushMappedMemoryRanges(), even though flushing is required
by the spec, which leads to misrendering and hangs on MTL platforms.

We found that other drivers also put a host-coherent, but not cached
memory type as the first host-accessible memory type, so let's do the
same in order to match the expectations of such broken applications.

Host-coherent uncached memory types are currently implemented with a
WC CPU map on non-LLC platforms, so there shouldn't be a huge
performance penalty from this: If an application intends to do heavy
R/W CPU access on a memory range it's expected to loop over the
available memory types and select one marked as host-cached -- If an
application fails to do that and simply selects the first available
type it seems more robust to stay on the safe side and give them a
host-coherent type rather than a cached one.

Rework:
 * Jordan: Add initial explanation to body of commmit message.
 * Curro: Add additional comments to commit message.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

iris: Map aux-map with WC on MTL+ (has_set_pat_uapi)

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

iris/bufmgr: Skip bucket allocation if not using writeback cache PAT index

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

iris/bufmgr: Add iris_pat_index_for_bo_flags()

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Suggested-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

intel/devinfo: Define PAT indices used on MTL

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

intel/devinfo: Add has_set_pat_uapi

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

iris: Allocate coherent buffers for resources flagged as persistent/coherent

If the application requests a coherent resource, we should honor that.
We technically don't need to ensure coherency for persistent mappings,
but we would have to handle PIPE_BARRIER_MAPPED_BUFFER to ensure that
data became visible at the right times. Instead, we just opt for the
easy plan and mark them coherent too.

Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22878>

intel/dev: update mesa_defs.json from defect database

These modifications represent:

 * changes to defects made since May 24, 2023
 * changes to handling of defects which were manually cloned

Acked-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23865>

dzn: VK_EXT_external_memory_host

When ID3D12Device13 is available, we can support importing host memory.
Imported host memory can be used to back buffers and linear textures.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23886>

anv: Flush untyped dataport cache DC flush is requested on compute

Although the following is based on this observations for OpenGL, we
probably need this for Vulkan as well.

KHR-GL46.texture_buffer.texture_buffer_operations_ssbo_writes writes
to an SSBO in a compute program, then issues a memory-barrier, which
causes us to add a DC-flush. Then a second compute program samples
from the SSBO written by the first compute program.

Although we expected the DC-flush to make the writes available to the
second compute program, on MTL this wasn't the case. Adding the
"Untyped Data-Port Cache Flush" fixes this.

The PRM indicates that compute programs must set "Untyped Data-Port
Cache Flush" to flush some LSC writes when flushing HDC. Although we
are setting DC-flush, and not HDC-flush, it does appear that the
following reference might also apply to DC-flush.

In the Intel(R) Arc(tm) A-Series Graphics and Intel Data Center GPU
Flex Series Open-Source Programmer's Reference Manual, Vol 2a: Command
Reference: Instructions, PIPE_CONTROL, HDC Pipeline Flush (DWord 0,
Bit 9), there is a programming note:

> When the "Pipeline Select" mode is set to "GPGPU", the LSC Untyped
> L1 cache flush is controlled by "Untyped Data-Port Cache Flush" bit
> in the PIPE_CONTROL command.

Ref: a8108f1d444 ("anv: Add missing untyped data port flush on PIPELINE_SELECT")
Ref: bd8e8d204db ("iris: Add missing untyped data port flush on PIPELINE_SELECT")
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23176>

anv: Flush untyped dataport cache when HDC flush is requested on compute

In the Intel(R) Arc(tm) A-Series Graphics and Intel Data Center GPU
Flex Series Open-Source Programmer's Reference Manual, Vol 2a: Command
Reference: Instructions, PIPE_CONTROL, HDC Pipeline Flush (DWord 0,
Bit 9), there is a programming note:

> When the "Pipeline Select" mode is set to "GPGPU", the LSC Untyped
> L1 cache flush is controlled by "Untyped Data-Port Cache Flush" bit
> in the PIPE_CONTROL command.

Ref: a8108f1d444 ("anv: Add missing untyped data port flush on PIPELINE_SELECT")
Ref: bd8e8d204db ("iris: Add missing untyped data port flush on PIPELINE_SELECT")
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23176>

anv: Clear untyped dataport cache flush bit if not in GPGPU mode

This should be equivalent, but refactoring the code will allow the
next two patches to use an else block for this check.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23176>

iris: Flush untyped dataport cache DC flush is requested on compute

KHR-GL46.texture_buffer.texture_buffer_operations_ssbo_writes writes
to an SSBO in a compute program, then issues a memory-barrier, which
causes us to add a DC-flush. Then a second compute program samples
from the SSBO written by the first compute program.

Although we expected the DC-flush to make the writes available to the
second compute program, on MTL this wasn't the case. Adding the
"Untyped Data-Port Cache Flush" fixes this.

The PRM indicates that compute programs must set "Untyped Data-Port
Cache Flush" to flush some LSC writes when flushing HDC. Although we
are setting DC-flush, and not HDC-flush, it does appear that the
following reference might also apply to DC-flush.

In the Intel(R) Arc(tm) A-Series Graphics and Intel Data Center GPU
Flex Series Open-Source Programmer's Reference Manual, Vol 2a: Command
Reference: Instructions, PIPE_CONTROL, HDC Pipeline Flush (DWord 0,
Bit 9), there is a programming note:

> When the "Pipeline Select" mode is set to "GPGPU", the LSC Untyped
> L1 cache flush is controlled by "Untyped Data-Port Cache Flush" bit
> in the PIPE_CONTROL command.

Ref: bd8e8d204db ("iris: Add missing untyped data port flush on PIPELINE_SELECT")
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23176>

iris: Flush untyped dataport cache when HDC flush is requested on compute

In the Intel(R) Arc(tm) A-Series Graphics and Intel Data Center GPU
Flex Series Open-Source Programmer's Reference Manual, Vol 2a: Command
Reference: Instructions, PIPE_CONTROL, HDC Pipeline Flush (DWord 0,
Bit 9), there is a programming note:

> When the "Pipeline Select" mode is set to "GPGPU", the LSC Untyped
> L1 cache flush is controlled by "Untyped Data-Port Cache Flush" bit
> in the PIPE_CONTROL command.

Ref: bd8e8d204db ("iris: Add missing untyped data port flush on PIPELINE_SELECT")
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23176>

ci: Testing -D shared-glapi=disabled with debian-clang-release

Acked-by: Eric Engestrom <eric@igalia.com>
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23409>

mapi: Hide OpenGL functions to be exported when shared-glapi is disabled

Fixes the following test error:
135/154 mesa:gallium / osmesa-symbols-check                                                                       FAIL             0.07s   exit status 1

```
src/gallium/targets/osmesa/libOSMesa.so.8.0.0: unknown symbol exported: glAreTexturesResidentEXT
src/gallium/targets/osmesa/libOSMesa.so.8.0.0: unknown symbol exported: glDeleteTexturesEXT
src/gallium/targets/osmesa/libOSMesa.so.8.0.0: unknown symbol exported: glGenTexturesEXT
src/gallium/targets/osmesa/libOSMesa.so.8.0.0: unknown symbol exported: glIsTextureEXT
```
The build options is:
```
-D glx=xlib -D gles1=disabled -D gles2=disabled -D shared-glapi=disabled
```

Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Corentin Noël <corentin.noel@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23409>

intel/aubinator_error_decode: add ccs support

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23867>

genxml: enable decoding on compute engine

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23867>

radv: use nir_opt_intrinsics

No fossil-db changes (navi21).

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23621>

nir/opt_intrinsic: optimize quad vote

Optimizes a quadAll()/quadAny() pattern created by dxil-spirv:
https://github.com/HansKristian-Work/dxil-spirv/commit/7adc87d4deaba8078bcdef8dfbebdda0165cd7bc

dxil-spirv can't use clustered reductions because they are not guaranteed
to include helper invocations.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23621>

nir,aco: add INCLUDE_HELPERS index to reduce intrinsic

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23621>

aco: include helpers in emit_uniform_{reduce,scan}

Found by inspection.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23621>

nir/peephole_select: allow some invocation broadcast intrinsics

fossil-db (navi21):
Totals from 3 (0.00% of 133428) affected shaders:
Instrs: 2074 -> 2083 (+0.43%)
CodeSize: 10596 -> 10692 (+0.91%)
Latency: 75754 -> 75946 (+0.25%)
InvThroughput: 16900 -> 16975 (+0.44%)
Copies: 312 -> 309 (-0.96%)
Branches: 150 -> 132 (-12.00%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23621>

treewide: Remove unused builders

-Wunused-variables kicks in now that it can see through the init.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23860>

treewide: Use nir_builder_create more

perl -p0e 's/nir_builder_init\(&([^,]*), /\1 = nir_builder_create(/g' -i $(git grep -l nir_builder_init)

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23860>

nir: Use nir_builder_create

perl -p0e 's/nir_builder ([^;]*);\s*nir_builder_init\(&\1, /nir_builder \1 = nir_builder_create(/g' -i $(git grep -l nir_builder_init)

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23860>

nir: Add nir_builder_create returning nir_builder

More ergonomic.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23860>

nir/builder_opcodes: Remove nir_build_ prefixed helpers

This patch decreases the size of nir_builder_opcodes.h from 14292 loc to
13763 loc.

nir_build_ versions are still needed if the nir_ is a custom helper.
Intrinsics which need such a helper have to be added to
build_prefixed_intrinsics.

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

nir: Use nir_ instead of nir_build_ helpers

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

vtn: Use nir_ instead of nir_build_ helpers

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

freedreno: Use nir_ instead of nir_build_ helpers

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

intel: Use nir_ instead of nir_build_ helpers

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

microsoft: Use nir_ instead of nir_build_ helpers

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

amd: Use nir_ instead of nir_build_ helpers

Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23858>

nir/lower_blend: Optimize masked out RTs

While debugging KHR-GLES31.core.draw_buffers_indexed.color_masks, the noise from
piles of store_output(load_output) instructions got in the way. Optimize it out.

This does not fix the test, but if this case ever happened in a real app it
would improve performance. This is only load bearing on Asahi (and PanVK?),
since Panfrost wouldn't call nir_lower_blend at all in this case.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23836>

asahi: Use txf for background program

More straightforward (txf instead of tex, with integer coords). No discrernible
performance difference.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23836>

agx: Use nir_lower_frag_coord_to_pixel_coord

Instead of open-coding the logic.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23836>

pan/bi: Use lower_frag_coord_to_pixel_coord

Instead of vendoring the logic. This has a side benefit of letting NIR
optimize the generated code a bit.

total instructions in shared programs: 2687284 -> 2687281 (<.01%)
instructions in affected programs: 532 -> 529 (-0.56%)
helped: 3
HURT: 1
Inconclusive result (value mean confidence interval includes 0).

total cycles in shared programs: 140711.33 -> 140711.31 (<.01%)
cycles in affected programs: 2.53 -> 2.52 (-0.62%)
helped: 1
HURT: 0

total fma in shared programs: 22059.44 -> 22059.39 (<.01%)
fma in affected programs: 2.69 -> 2.64 (-1.74%)
helped: 3
HURT: 0

total cvt in shared programs: 14659.09 -> 14659.09 (0.00%)
cvt in affected programs: 1.56 -> 1.56 (0.00%)
helped: 1
HURT: 1

total quadwords in shared programs: 1455408 -> 1455416 (<.01%)
quadwords in affected programs: 128 -> 136 (6.25%)
helped: 0
HURT: 1

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23836>

nir: Add lower_frag_coord_to_pixel_coord pass

We've open coded this in a few backends.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23836>

nir: Add pixel_coord, frag_coord_zw intrinsics

On some architectures, gl_FragCoord.xy is available as an integer but
gl_FragCoord.zw requires interpolation. Add dedicated intrinsics so we can
lower it all in NIR.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23836>