For fixed point Z the offset_units passed to glPolygonOffset
must be scaled by 2, and we do that. However, this scaling is not
needed for floating point output, and actually produces incorrect Z
values. The detailed description of the depth offset is in
e.g. section 13.7.2 of the OpenGL ES 3.2 spec. Note that equation (13.11)
has no factor of 2. That factor is what we get as the minimum
resolvable difference in a 24 bit fixed point format.
After this patch the output Z for a 32 bit floating point depth
buffer exactly match what the software renderer produces, and are extremely
close (off by 1 ULP) to the amdgpu hardware renderer; before the
patch they were significantly different.
Signed-off-by: Eric R. Smith <eric.smith@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28477>
The WRITE_SAMPLE_COUNT_DIFF attribute of CP_EVENT_WRITE7 calculates the
difference between the begin and end sample-count values, but it in fact
accumulates that difference in the desired location, so it's renamed
to reflect that.
When writing out the ending-point samples count for the occlusion query
through CP_EVENT_WRITE7::ZPASS_DONE, enable the SAMPLE_COUNT_END_OFFSET and
WRITE_ACCUM_SAMPLE_COUNT_DIFF attributes on the event struct and use the
address of the beginning-count value. The latter adjustment is necessary
since the difference will be written 8 bytes into the buffer and the
ending-count will be written 16 bytes into the buffer.
With the occlusion query result now being handily accumulated in the query
buffer as long as the hardware supports it, the occlusion_query_slot struct
is adjusted to reflect that. On pre-a740 hardware the difference is still
calculated and stored in the result location, whereas on newer hardware
CP_EVENT_WRITE7::ZPASS_DONE can do it for us.
Signed-off-by: Zan Dobersek <zdobersek@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28610>
Remove the 'A6XX' variant attribute from the RB_SAMPLE_COUNT_ADDR register
definition since this register is also used up to a730, being displaced
from a740 onwards by CP_EVENT_WRITE7::WRITE_SAMPLE_COUNT.
Signed-off-by: Zan Dobersek <zdobersek@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28610>
Use the SAMPLE_COUNT_END_OFFSET attribute when writing out the sample count
at the end of ZPASS_DONE event, if the GPU version supports it. The write
is done 16 bytes into the specified iova address, which is already where
the tu_renderpass_samples struct on top of that address expects the end
value.
Signed-off-by: Zan Dobersek <zdobersek@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28610>
The header was missing after the definitions were moved.
Fixes: 82114ac02a ("r600g: switch to a common formats.h file since they are in different regs")
Acked-by: Gert Wollny <gert.wollny@collabora.com>
Signed-off-by: David Heidelberg <david@ixit.cz>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28395>
As we were checking the bind count after decreasing the ref count,
we could hit a situation where the worker thread decreases the bind count
just after the ref count is decreased, but before the bind count is checked
which would lead to both threads calling the resource dtor.
Cc: mesa-stable
Signed-off-by: Axel Davy <davyaxel0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28232>
llvmpipe does not support ray tracing stages, so update_pcbuf needs to
be aware of the Vulkan and the llvmpipe stage.
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28187>
the location has to be preserved when lowering them to scratch using
nir_lower_vars_to_explicit_types and nir_lower_explicit_io.
cc: mesa-stable
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28187>
We add a units generic parameter to Extent4D and Offset4D and then make
every helper take and return the right type. The mul and div_* helpers
only half-enforce types but it's better than nothing. We leave them
non-public so no one outside extent.rs uses them.
Reviewed-by: Lina Versace <linyaa@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28639>
We're about to add a PhantomData to Extent/Offset4D and that'll make it
harder to construct. Adding a helper lets us more easily construct them
elsewhere in the code.
Reviewed-by: Lina Versace <linyaa@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28639>
