Unfortunately, this requires some non-trivial changes to the driver. Now
that the primitive restart index isn't given explicitly by the client, we
always use ~0 for everything like D3D does. Unfortunately, our hardware is
awesome and a 32-bit version of ~0 doesn't match any 16-bit values. This
means, we have to set it to either UINT16_MAX or UINT32_MAX depending on
the size of the index type. Since we get the index type from
CmdBindIndexBuffer and the rest of the VF packet from the pipeline, we need
to lazy-emit the VF packet.
This was missing and was causing the driver to not work with
execlists. Presumably we get a different initial hw context with
execlists enabled, that has sample mask 0 initially.
Set this to 0xffff for now. When we add MS support, we need to take the
value from VkPipelineMsStateCreateInfo::sampleMask.
The new headers use stdbool for enable/disable fields which
implicitly converts expressions like (flags & 8) to 0 or 1.
Also handles MBO (must-be-one) fields by setting them to one,
corrects a bspec typo (_3DPRIM_LISTSTRIP_ADJ -> LINESTRIP) and
makes a few enum values less clashy.
I misinterpreted anv_format::format as a VkFormat. Instead, it is
a hardware surface format (RENDER_SURFACE_STATE.SurfaceFormat). Rename
the field to 'surface_format' to make it unambiguous.
We compute the right mask and thread width max parameters as part of
pipeline creation and set them accordingly at vkCmdDispatch() and
vkCmdDispatchIndirect() time. These parameters depend only on the local
group size and the dispatch width of the program so we can figure this
out at pipeline create time.
This mega-commit primarily does two things. First, is to turn anv_batch
into a better abstraction of a batch. Instead of actually having a BO, it
now has a few pointers to some piece of memory that are used to add data to
the "batch". If it gets to the end, there is a function pointer that it
can call to attempt to grow the batch.
The second change is to start using chained batch buffers. When the end of
the current batch BO is reached, it automatically creates a new one and
ineserts an MI_BATCH_BUFFER_START command to chain to it. In this way, our
batch buffers are effectively infinite in length.
Previously, we just blasted out whatever VB's we had marked as "dirty"
regardless of which ones were used by the pipeline. Given that the stride
of the VB is embedded in the pipeline this can cause problems. One problem
is if the pipeline doesn't use the given VB binding we emit a bogus stride.
Another problem is that we weren't properly resetting the dirty bits when
the pipeline changed.
This changes the way descriptor sets and layouts work so that we fill
out binding table contents at the time we bind descriptor sets. We
manipulate the binding table contents and sampler state in a shadow-copy
in anv_cmd_buffer. At draw time, we allocate the actual binding table
and sampler state and flush the anv_cmd_buffer copies.
This way we can pass in a vertex shader and yet have the pipeline emit an
empty 3DSTATE_VS packet. We need this for meta because we need to trick
the compiler into not deleting our inputs but at the same time disable the
VS so that we can use a rectlist. This should go away once we actually get
SPIR-V.
