This hardware hang workaround (PAL waMiscPopsMissedOverlap) is needed only
on some Vega chips, and only for 8 or more samples per pixel. It has a
significant performance cost (around 1.5x-2x in
nvpro-samples/vk_order_independent_transparency), so it should be precisely
configured when setting up Primitive Ordered Pixel Shading.
It was added in 47b780be21, when POPS was not
used in Mesa, with the change being described as "this may not be needed
yet, but let's set it now".
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
When letting the overlapping waves enter their ordered sections, there must
be no memory accesses to resources which need primitive-ordered access that
are still pending, or there would be a race between the current wave and
the overlapping waves.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
If the wave has set the Primitive Ordered Pixel Shading packer ID hardware
register, it must send MSG_ORDERED_PS_DONE once before the program ends.
It's also safe to send the message if the packer ID register hasn't been
set yet, therefore the message may be sent conservatively. For simplicity,
to ensure that it's sent on all execution paths after setting the packer ID
register, always sending it from a top-level block. This is required for
GFX9-10.3 POPS.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
Implementing the acquire/release semantics of fragment shader interlock
ordered section in Vulkan, and preventing reordering of memory accesses
requiring primitive ordering out of the ordered section.
Also, the ordered section should be as short as possible, so not reordering
the instructions awaiting overlapped waves upwards, and the exit from the
ordered section downwards.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
Waits are needed for early exits from inside a Primitive Ordered Pixel
Shading ordered section, but that code doesn't insert them reliably anyway
because it doesn't obtain the counters for the exact locations of the
jumps, which may be anywhere inside the predecessor blocks.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
A wait for export_ready (if the corresponding bit is not set in the
instruction) is done to enter the Primitive Ordered Pixel Shading ordered
section on GFX11.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
pops_exiting_wave_id is a volatile ALU source operand containing the ID of
the latest wave that hasn't exited yet, for comparing with the newest
overlapped wave ID in overlapping waves.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22250>
For executing IB on the compute queue (ie. IB2 isn't supported), we
will need to break chaining, this is a first step towards this.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23727>
This was only necessary for preambles/postambles, let's clarify this
by determining the IB info from the first IB in the array instead.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23727>
The current IB size is copied when radv_amdgpu_cs_add_old_ib_buffer()
is called, which might not be the real IB size because we might still
pad the CS with NOP packets after.
Found by inspection.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23727>
The preprocess buffer is the buffer used to generate the cmdbuf. It
was aligned to 256 bytes but the correct alignment is actually
ac_gpu_info::ib_alignment.
Otherwise, if a DGC IB is executed like a IB1, this hits an assertion
in radv_amdgpu_cs_submit() because the alignment is incorrect.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23764>
Backends never see these instructions.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Suggested-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23831>
Cleanup of various fields with redundant information.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23597>
With this change, ACO is going to rely on the caller to set
the HW stage and will no longer guess it from the input shaders.
This will help enable compiling merged shaders separately,
but that will need further changes in instruction selection.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23597>
ACO will rely on this field instead of guessing
the stage internally.
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/23597>
Unused in this commit, but this is going to replace the shader
stage selection inside ACO after the drivers set it correctly.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23597>
The new ac_hw_stage is going to be used by drivers as well.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23597>
This is going to be shared between RADV, RadeonSI and ACO.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23597>
Shaders are the largest thing we hash now, so they benefit from a faster
hash.
Change the field name from `sha1` to `hash` to avoid tying the definition
to a particular algorithm. This doubles down as a precaution against
callers still assuming a 20-byte hash (in which case the compilation will
error out).
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22571>
We are trying to re-enable the valve CI... but doing so runs all the
jobs, including the manual ones.
Since some can take over an hour to run, let's disable them, and
re-enable them in another MR by reverting this commit.
Sorry for the noise!
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23789>
Move most of the DB_SHADER_CONTROL fields from the pipeline to the pixel
shader for preparation for shader objects.
Also, the GFX11 export conflict bug workaround doesn't need to be enabled
for non-1x sample counts or if blending is not enabled, so make the
application of DB_SHADER_CONTROL consider the current sample count and
blending state even if they're dynamic.
Having access to the exact sample count in DB_SHADER_CONTROL setup is also
necessary for good performance in SampleInterlock execution modes of
fragment shader interlock, for configuration of POPS_OVERLAP_NUM_SAMPLES
(GFX9-10.3) or OVERRIDE_INTRINSIC_RATE (GFX11), as PixelInterlock is
massively slower with multisampling due to overlap between adjacent
polygons sharing covered pixels among the common edge.
The name of the dynamic state controlling DB_SHADER_CONTROL is now
unambiguous - previously line rasterization mode had effect on attachment
feedback loop state emission.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23474>
We can not use it for all stages because RADV has some special stages
and shader-db doesn't handle all stages.
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/23524>
