When we print asm for merged VS+TCS, we need to record asm for VS
even when only TCS is being printed because they are printed together.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35620>
The problem was that SGPRs returned from LS were treated as VGPRs by LLVM
because LS executes conditionally. The workaround is to use input SGPRs
initialized by the hw (i.e. parameters of the wrapper function) instead of
the ones returned by LS.
Fixes: 80236f2367 - ac/nir/tess: add if/endif for HS threads in NIR instead of ACO/LLVM
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35620>
For non-ONE_TIME_SUBMIT command buffers we have to create CS that
copies u_trace timestamps. Having cs pool allows us to avoid
allocations in most cases.
Together with previous commit, BO allocations on submit would be
rare during normal u_trace tracing.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35637>
For non-ONE_TIME_SUBMIT command buffers we have to copy tracepoints
which leads to BO allocations. That adds overhead we don't want,
especially when there are many tiny submissions.
While suballocator for u_trace BOs is allocated only when needed,
it's not deallocated afterwards, since the point where it's not needed
is tricky to determine.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35637>
should slightly help proton
requires reordering the uvs lowering to be after tes lowering since that can
insert psiz writes.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35658>
No longer need the defaults, and can properly bail on real errors. The
behavior change of chroma sitting has been mentioned in the prior
commit, and doesn't affect gallium side.
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35613>
Gralloc is the central piece to align defaults and expectations across
different APIs. Doing so simplifies client side handling. To be noted,
midpoint chroma location is preferred as the default to skip the chroma
lowering pass. The same has been suggested by all Vulkan drivers with
AHB implemented and practiced for years under venus.
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35613>
Clients are expecting the color info to be fully filled when the api
exists. Give proper defaults for the metadata to stay aligned with
legacy backends.
Also amend the missing ChromaSiting cases.
Fixes: ee42e2166d ("android: Introduce the Android buffer info abstraction")
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35613>
Clients are expecting the color info to be fully filled when the api
exists. Give proper defaults for the metadata to stay aligned with
legacy backends.
Fixes: 122fd46b15 ("Android15 support gralloc IMapper5")
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35613>
The msm driver hasn't yet passed conformance, so it's not part of the
auto set at this point in time. But this will allow distribution to
enable it by default, if they find it suitable.
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35618>
When applying the sync flags after legalize, we might run into the issue
of the first instruction not supporting (ss)/(sy). We cannot insert a
nop at that point anymore since jump targets have already been
calculated. Fix this by applying the sync flags during legalize, where
all other sync flags are also applied.
Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35653>
This isn't necessary and since predt is a terminator and we (mistakenly)
didn't propagate force_ss across blocks, it wasn't actually applied.
Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35653>
