I'm not sure why we need to set this magic bit, but this fixes the
non-depth_component portion of
dEQP-GLES3.functional.texture.format.sized.*, e.g
dEQP-GLES3.functional.texture.format.sized.cube.srgb_rg8_pot
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18167>
Move tiling.c into ail, using ail data structures and helpers to manage
the tiling. This fixes a staggering number of issues with the tiling
routines:
* NPOT block sizes defeatured. The hardware only supports POT block
sizes. There's no need to handle anything else.
* Use ail to determine tile sizes, instead of the broken
agx_select_tile_shift routine that didn't work for non-square tile
sizes (for instance).
* Handle up to 128x128 tiles, as required by 8bpp textures.
* Handle non-square tiles. If the block size is not a multiple of 4, the
tile size will be of the form 2n x n. This is easy with the ail_tile
data structure, but not possible architecturally with
agx_select_tile_shift. This is required for 16bpp and 64bpp textures.
* Express in terms of elements instead of pixels, using unit
suffixes to make the dimensional analysis obvious. In particular this
handles tiling of block-compressed textures by tiling the blocks
themselves. This is required for block-compressed textures (internally handled
like smaller 64bpp textures).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18167>
Introduce ail, a small library for working with the image (and buffer)
layouts encountered with AGX hardware. Its design is inspired by isl. In
particular, ail strives to use isl unit suffixes and to represent
quantities in a canonical, API-agnostic fashion [1].
ail replaces the old miptree code (based on some ad hoc heuristics that
passed a few dEQP tests). It is based on a thorough reverse-engineering
of AGX's twiddled format, courtesy of Asahi Lina, Dougall Johnson, and
me. This corrects our handling of many common cases that were totally
wrong in the old code, leading to GPU faults.
Unlike the code, ail differentiates between pixels and elements
consistently, allowing block-compressed formats like ETC2 to be
supported correctly. These formats will be enabled later in the series.
This commit fixes Inochi2D, glmark2 -brefract and -bterrain, and who
knows what else.
ail stands for { Asahi, AGX } Image { Layout, Library } at your
convenience. ail is best served warm.
Liberal use of ail is recommended. Yum!
[1] https://docs.mesa3d.org/isl/units.html
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18167>
The ASTC enum only encodes the block width/height. By contrast the
LDR/HDR/sRGB distinction is encoded as UNORM/Float and via the sRGB bit.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18167>
We won't be implementing AGX compression for a while, but this gets some
unknowns out of the way when looking at dumps from Metal that use it.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18167>
There are many subtleties to laying out twiddled images on AGX. Document
the known ones in a central place. Many of the hairy details of
twiddling were discovered by Asahi Lina and Dougall Johnson, many thanks
to them.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18167>
There is a limit to how vector widths (eg due to RA), larger vectors should be
broken up in NIR so we shouldn't ever see such vectors.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17220>
These are used with OpenCL, particularly with 8-bit types. Luckily, they are
pretty easy to implement with our existing infrastructure. We just need to hit
backspace enough times and we're good to go.
Fixes a subtest of test_basic hiloeo.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17220>
We need to be careful to ensure the output of constant lowering still selects a
single lane, otherwise we lower something like:
44 = MKVEC.v2i8 43.b1, #0x0.b0, #0x0
to
44 = MKVEC.v2i8 24.b1, u256, u256
which loses the .b0 swizzle, and thus renders the MKVEC.v2i8 unable to pack.
Similarly, we need to allow zero-extension with fields marked "lane". Currently
only MKVEC.v2i8 hits this path, as the other instructions using "lane"
modifiers are unary and therefore will have been constant folded.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17220>
Refactor the 16-bit vector construction code to separate the collect of separate
32-bit words from the construction of single 32-bit vectors from sub-32-bit
parts. Then add a path to combine (up to 4) bytes into a 32-bit word using
Valhall's MKVEC.v2i8 instruction.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17220>
This is similar to what RADV and several other drivers have been doing
for quite some time. It's good to have it common.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18324>
This is little more than a wrapper around a function pointer at present
so it doesn't really gain the driver much to use it. The new callback
will, however, be put to good use in the next commit.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18324>
This is the standard pattern in the kernel for providing vfunc tables
for C objects. We're using it in the pipeline cache code but we're
about to start adding more stuff and so it really helps if we have it
for command buffers as well.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18324>
Most other init functions follow the Vulkan API convention of putting
the parent object first.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18324>
This is not necessarily a full re-initialization of the object but is
whatever is necessary/expedient for the client to see it as a new object
and not the one it has seen before. For vk_base_object, this removes
any private data and resets the object name to NULL.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18324>
This is a counterpart to the previous commit. When we have multiple
depth attachments, in the secondary we currently don't disable LRZ and
so we may need a valid LRZ fast-clear base.
Fixes: 4b5f0d98 ("tu: Overhaul LRZ, implement on-GPU dir tracking and LRZ fast-clear")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18369>
In
dEQP-VK.renderpass.dedicated_allocation.attachment_allocation.input_output.94
we have the following:
- There is more than one subpass, but only one depth attachment.
- The first subpass doesn't use depth.
- The subpass that does use depth has a draw call in a secondary.
We wouldn't hit the case where there's more than one depth attachment,
but because tu_begin_resumed_renderpass() only looked at the first
subpass it wouldn't find the depth attachment and would leave LRZ
invalid and thus a NULL LRZ fast-clear base. Then
tu_begin_secondary_cmdbuf() would leave LRZ enabled and the draw would
have LRZ enabled, leading to a hang.
Fix this by making tu_begin_resumed_renderpass() match
tu_begin_renderpass() with how it finds the depth attachment.
Fixes: 4b5f0d98 ("tu: Overhaul LRZ, implement on-GPU dir tracking and LRZ fast-clear")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18369>
To use the similar producer/consumer logic we do when linking shaders.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18278>
It's a link step somehow, except for VS only on GFX10+ but keep it
there anyways.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18278>
This doesn't make sense and it was assigned because the shader info
stuff was a complete mess.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18278>
This doesn't make sense and it was assigned because the shader info
stuff was a complete mess. LS is only a thing on GFX6-8, on GFX9+
it's a merged VS+TCS to HS.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18278>
