This was a good idea, but ended up tying the build systems in knots.
We can revisit this later, in particular if we can put in place dummy
implementations of cell_create_screen(), llvmpipe_create_screen()
which just return NULL if the driver isn't available.
In the meantime, just duplicate this smallish function in the two
places it was being called.
Components such as state trackers, drivers, etc, should be free to be
recombined in arbtrary ways to build driver stacks. They should not
be reaching out and trying to build the stack themselves - this is now
expected to be handled by the "target" abstraction.
Add a helper gallium_wrap_screen() for injecting the commonly
used extra layers into a gallium stack. Currently that's just the
trace module and identity layer, but there could be more in the
future, eg. a validation layer.
CSO can often be null.
For example:
1. at initialization
2. using an util module (u_blit) right after initialization (it will push
state and pop the previous null state)
3. at shutdown time (state shouldn't be bound when being destroyed)
Glean was hitting 2.
This makes draw_arrays()/draw_arrays_instanced() do the right thing and
not require the (probably broken anyway) flush_notify() usage.
It also fixes a potential bug in the behaviour of reading InstanceID from
shaders, where 0 should be read for non-instanced drawing, previously it
was possible to read non-0 ids if mixing instanced/non-instanced.
Really though, using this at all is just not a good idea in the 3D driver.
I'm almost certain the hardware will not like a reloc appearing between
begin()/end().. Anyways, this is still better than before, more fixes
to come..
Also allows the nv50_state_validate() caller to request a minimum amount
of space that itself requires, not all callers accurately use this yet
but the simple cases are now accounted for.
Rendering will also be dropped on the floor if validate fails now.
We were doing it ad-hoc before, as instructions with potential
aliasing problems were identified. But thanks to swizzling basically
anything can have aliasing, so just do it generally at source reg
setup time. This is somewhat inefficient, because sometimes an
operation doesn't need unaliasing protection if the swizzling is safe,
but the unaliasing before didn't cover those cases either.
Fixes piglit glsl-fs-loop.
We were patching up all the break and continues between the start of
our loop and the end of our loop, even if they were breaks/continues
for an inner loop. Avoiding patching already patched breaks/continues
fixes piglit glsl-vs-loop-nested.
José Fonseca
Keith Whitwell
Ben Skeggs
Corbin Simpson
Chia-I Wu
Eric Anholt
Maciej Cencora