Create a new stage called nouveau-postmerge and move the full and manual
jobs over there, to avoid entanglement with the pre-merge jobs.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30784>
Create a new stage called software-backends-postmerge and move the full
and manual jobs for both llvmpipe and lavapipe over there, to avoid
entanglement with the pre-merge jobs.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30784>
Create a new stage called layered-backends-postmerge and move the full
and manual jobs over there, to avoid entanglement with the pre-merge
jobs.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30784>
Create new stages called broadcom-postmerge and freedreno-postmerge, and
move the full and manual jobs over there, to avoid entanglement with the
pre-merge jobs.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30784>
Create a new stage called intel-postmerge and move the full and manual
jobs over there, to avoid entanglement with the pre-merge jobs.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30784>
Create a new stage called amd-postmerge and move the full and manual
jobs over there, to avoid entanglement with the pre-merge jobs.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30784>
We used the logical cfg in helper_sched under the assumption that (eq)
only kills active helpers. This could lead to something like this:
br p0.x, else
sam
(eq)nop
jump end
else:
sam
end:
It turns out this is not correct: (eq) kills *all* helpers so the
snippet above could produce wrong results.
To fix this, simply switch from using the logical cfg in helper_sched to
using the physical cfg.
Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29409>
opt_jump destroys the physical cfg but it turns out we need it
afterwards in helper_sched. Keeping it correctly up-to-date during
opt_jump turned out to be tricky and complex. Simply recalculate it
afterwards to keep things simple.
Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29409>
When calling ir3_calc_reconvergence after opt_jump, it may happen that
there are multiple divergent branches to the same block. This was
handled incorrectly since first_divergent_pred was unconditionally
overwritten for successors of divergent branches. This patch fixes this
by only making sure only the earliest divergent predecessor is used.
Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29409>
After opt_jump, we might end up with weird control flow likes this:
block0: br block2
[more blocks]
block1: br block3
block2: ...
block3: ...
Since block2 is a logical successor of block1 (due to the fall through),
a physical edge will be added. However, another one will be added
because the branch to block3 crosses block2 (which is a reconvergence
point due to the branch from block0). This currently results in an
assert. This commit fixes this by ignoring successors of the start block
of edges that cross reconvergence points.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29409>
Instead of handling backedges after a physical edge is added that extends
the divergence range, handle it at the beginning of the loop. This
handles the TODO by making sure that we also check for backedges after a
block is added to the worklist at the beginning due to a divergent
branch. For example, this fixes cases where opt_jump optimizes
"if (foo) { break; }" to remove the break block and we have a divergent
branch across the loop backedge.
While we're here, also fix an off-by-one that meant we wouldn't handle
backedges that jumped to the immediate previous block (which is again
possible after opt_jump) or the same block.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29409>
Blocks with two terminators have branches to both successors (instead of
a fall through to one) and hence need to mark both successors as
reconvergence points.
Signed-off-by: Job Noorman <jnoorman@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29409>
Planned downtime in the farm:
* Start: 2024-08-22 06:00 UTC
* End: 2024-08-22 12:00 UTC
Risk notice due to electrical maintenance in the building where the farm
is hosted:
* Start: 2024-08-19 07:00 UTC
* End: 2024-08-22 16:00 UTC
So, there is a time window when the farm will be disabled. But there is
a bigger window when there is a change for this commit to be merged.
Signed-off-by: Sergi Blanch Torne <sergi.blanch.torne@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30534>
On Ampere and later, this instruction allows to handle packing of F32x2
to F16x2.
Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30765>
We were specifying align_offset = 64 and align_mul = 64, which is
invalid. nir_combined_align() asserts that align_offset < align_mul.
Our intention here is to perform cacheline-aligned (64B-aligned) block
loads, so we should set align_mul = 64 and can leave align_offset = 0.
Fixes: fbafa9cabd ("intel/nir: remove load_global_const_block_intel intrinsic")
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30755>
At least one ray tracing shader in cp2077 is over 4MB on Xe2. There
isn't a memory pool large enough for the allocation, so the driver
crashes instead. This commit adds 8MB and 16MB pools.
I intend this as a stop gap fix. I would prefer to figure out why this
shader is so much larger than on previous platforms. The shader in
question has 3824 spills and 8625 fills. That is not good. I suspect
dealing with that will also solve the problem, but that will require a
bit more time.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11739
Suggested-by: Lionel Landwerlin
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30751>
this avoids a scenario where invalidate happens after a non-winsys blit
for a renderpass and the driver skips storing framebuffer contents
because the invalidate flag is set
cc: mesa-stable
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30750>
This fix
"dEQP-VK.dynamic_state.*.general_state.{bind_order, state_persistence, state_switch}"
Fixes: 1f57aae4e4 ("panvk: Move vkCmdDraw* functions to their own file")
Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30754>
Previous changes of the helper invocation pass fixed missing conditional
control flow tracking but this is not enough.
Propagation of the dependency chain also need to handle value outside of
direct predecessors.
This fix "dEQP-VK.graphicsfuzz.cov-nested-loops-sample-opposite-corners"
for real this time.
Fixes: 33fef27356 ("bi: Do not mark tex ops as skip when dest is used by control flow")
Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30754>
This is faster than a pile of 1-dword NOPs. Note that GFX6 actually
supports type-3 NOP as long as the size is more than the header which
is always the case for the DGC preamble.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30748>
this allows deleting the fd check which is really what this was
intending to do (but is still broken because llvmpipe supports this)
Fixes: 991cc686a5f (egl: really fix kopper fd passing)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30647>
this was added to avoid deadlocking the device-select layer, but
the layer should just be disabled since it is already handled elsewhere
Fixes: 991cc686a5f (egl: really fix kopper fd passing)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30647>
After the transition to vk_meta, all clear tests seem to pass.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29451>
