If a then/else block ends in a jump, the phi nodes do not necessarily
have to reference the always taken branch because they are dead code.
Avoid crashing in this case by only rewriting phis, if the block does
not end in a jump.
cc: mesa-stable
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23150>
The common state code expects you to use a different struct for state in
graphics pipelines and in pipeline libraries. This means we need to
copy the approach radv uses in order to be compatible. This also allows
us to shrink the structs a bit by moving compute-only things to the
compute pipeline and library-only things to the library pipeline.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
This field also affects triangle strip and triangle fan ordering, so we
would get the incorrect (D3D) order with tessellation and geometry
shaders both enabled. Instead flip clockwise/counterclockwise when
the domain origin is upper-left, as radv does.
Because the register is only emitted when tessellation is active which
forces sysmem, it shouldn't regress performance to emit it directly
instead of using a draw state. We're already very tight on draw states.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
This was relying on cb being NULL instead of just gracefully handling
it, and it will stop being NULL once we start tracking attachment count
as state. Moreover is was broken in the case where only the blend enable
is dynamic.
Reviewed-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
Previously, drivers have either not supported some dynamic state (like
vertex input or sample locations) at all or it's been always dynamic. In
order to be able to set dynamic state sometimes and other times leave it
up to driver-specific state packets, we need a few helpers.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
On turnip we support dynamic vertex input, but static vertex input is
precompiled and so we will copy from a source without VI to a
destination with VI and it's valid in this case to do nothing. On the
other hand, it should never be valid if VI state is set but the pointer
isn't there, which the code previously silently skipped over. There's a
similar issue with sample locations.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
This is a mirror of vi::bindings_valid, but we can track it and set it
properly even when vertex input state is precompiled, because it is also
needed on turnip for knowing the size of the vertex buffer and vertex
stride state packets even when vertex input state is precompiled.
Previously drivers that could pre-bake vertex input state were expected
to handle this themselves, but this would've been complicated for turnip
because we can handle both pre-baked and dynamic vertex input state. Now
we have the one field which is correctly set in all circumstances and we
never have to setup space for vertex input state in the pipeline.
Reviewed-by: Faith Ekstrand <faith.ekstrand@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
On turnip, there are two cases for feedback loops:
- For feedback loops that involve input attachments, everything works as
normal in GMEM mode but have to do a workaround in sysmem.
- For feedback loops that may involve any texture, GMEM mode is
impossible and we have to disable it.
Currently we track this through a special flag on the pipeline, but this
won't be practical in the future. Add a flag to the common renderpass
state struct to patch this info through when using our own renderpass.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
It only has a subset of the renderpass state, whereas with turnip we
need to use pretty much all of it at one point or another. Just allow
the driver to pass in the entire vk_render_pass_state if it's using its
own renderpass implementation.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22301>
