Two potential problems, batch re-ordering doesn't really play nicely
with fence_server_sync(), so when we switch away from one batch, detect
the case that we need to sync, and if so flush. The alternative of
trying to track that later batches depend on an earlier batch that had
an in-fence is hairy, and the normal use-case would be to sync at the
beginning of the frame.
But this brings up the second problem, which is that typically we'll get
told to sync on an in-fence before the first draw, which means before
mesa/st flushes down the framebuffer state to the driver. Which means
we don't yet have the correct batch to attach the fence to. So we need
to track the in-fence on the context, and transfer it to the batch
before draws, etc.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6575>
When we have softpin, we know the address of the shader constant data at
shader upload time because it's sitting at the end of the shader. This
commit changes ANV to use patch constants to embed the address in the
shader patch the right address in at upload time. This allows us to
avoid having to set up a UBO binding on-the-fly for shader constants.
This commit uses an A64 message but it's quite possible that we could
also use an A32 message and make the dataport do the 64-bit add for us.
However, load_global is what we have right now so it was easier to just
use that.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6244>
With discrete GPUs, it's going to be possible to have GPUs from two
different hardware generations in the machine at the same time. Global
singletons like this aren't going to fly. Have a struct containing the
pointers which gets initialized once per shader disassemble instead.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6244>
With the kernel timeline sysncobj changes, the kernel submits do
not necessarily happen in global vkQueueSubmit order. Which should
be fine, we added the appropriate waits for that. (See
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_FOR_SUBMIT in the winsys)
However, all kernel submissions take a lock on the bo_list mutex,
and since we do the wait in the winsys, we wait while having the
bo_list mutex held. This means that as soon as a wait and a signal
submission are out of order we have a deadlock on the bo_list mutex
and the wait.
Solution is to use a shared reader lock during the kernel submission,
as we only need read access for the submission.
Fixes: 6bc5ce7a91 "radv: Add timeline syncobj for timeline semaphores."
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3446
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6478>
If decrement == 0 then:
- it isn't safe to access the submission
- even if it is, checking that the result of the atomic_sub is 0
doesn't given an unique owner anymore.
So skip it. The submission always starts out with refcount >= 1,
so first one to decrement to 0 still get dibs on executing it.
Fixes: 4aa75bb3bd "radv: Add wait-before-submit support for timelines."
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6478>
WAIT takes a notification register as a destination and a src0 argument.
Since the same notification register is specified in both fields, we
treat it as a special case and disassemble it only once.
If we disassemble it as if it is a source register, its scalar region
will be printed as <0,1,0>. This causes difficulties round-tripping
through the assembler <-> disassembler because that is not an acceptable
destination region. If we instead disassemble the destination, we
instead get a <1> region which is an acceptable and equivalent region
for source and destination.
The test .asm files are regenerated by round-tripping them through the
assembler/disassembler. Note that the <0> region in the tests was a
harmless mistake: the compiler translated it to a <0,1,0> source region
and a <1> destination region, since <0> isn't valid.
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6543>
We can get duplicate declarations for an index (for example dvec3 + float
packed into 2 vec4s, the second one won't pack into the first's array
decl), and we'd end up stepping by the wrong amount in GS vtx/prim emit.
Fixes vs-gs-fs-double, sso-vs-gs-fs-array-interleave piglit tests.
Fixes: 49155c3264 ("draw/tgsi: fix geometry shader input/output swizzling")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6567>
For NIR-to-TGSI, we don't want to revectorize 64-bit ops that we split to
scalar beyond vec2 width. We even have some ops that we would rather
retain as scalar due to TGSI opcodes being scalar, or having more unusual
requirements.
This could be used to do the vectorize_vec2_16bit filtering, but that
shader compiler option is also used in algebraic so leave it in place for
now.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6567>
It would be nice if we could do swizzling of an expression on the
replacement side so that we could have a single ieq/ine of the vector
after CSE. However, if you do want vector operations, nir_opt_vectorize()
does just fine.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6567>
