We actually want to use util_vma to handle this. But fortunately core
drm-shim alredy does this for mem offset, we can just delete a bunch of
code and re-use that.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16250>
When HW binning is used tile loads/stores could be skipped
if there is no geometry in the tile.
Loads could be skipped when:
- The attachment won't be resolved, otherwise if load is skipped
there would be holes in the resolved attachment;
- There is no vkCmdClearAttachments afterwards since it is likely
a partial clear done via 2d blit (2d blit doesn't produce geometry).
Stores could be skipped when:
- The attachment was not cleared, which may happen by load_op or
vkCmdClearAttachments;
- When store is not a resolve.
I chose to predicate each load/store separately to allow them to be
skipped when only some attachments are cleared or resolved.
Gmem loads are moved into separate cs because whether to emit
CP_COND_REG_EXEC depends on HW binning being enabled and usage of
vkCmdClearAttachments.
CP_COND_REG_EXEC predicate could be changed during draw_cs only
by perf query, in such case the predicate should be re-emitted.
(At the moment it is always re-emitted before stores)
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15974>
This controls the whole lowering of "make tex ops with implicit
derivatives on non-implicit-derivative stages be tex ops with an explicit
lod of 0 instead", but it's really hard to describe that in a git commit
summary.
All existing callers get it added except:
- nir_to_tgsi which didn't want it.
- nouveau, which didn't want it (fixes regressions in shadowcube and
shadow2darray with NIR, since the shading languages don't expose txl of
those sampler types and thus it's not supported in HW)
- optional lowering passes in mesa/st (lower_rect, YUV lowering, etc)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16156>
This is needed for the VIRTGPU_WAIT ioctl to work.
TODO we could perhaps limit this, since it is not needed for residency,
but only fencing. Ie. we could omit cmdstream, and probably anything
that has FD_BO_NOMAP flag.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16086>
This syncs up with the protocol of what eventually landed in virglrender.
1) Move all static params to capset to avoid having to query host
(reduce synchronous round trips at startup)
2) Use res_id instead of host_handle.. costs extra hashtable lookups in
host during submit, but this lets us (with userspace allocated IOVA)
make bo alloc and import completely async.
3) Require userspace allocated IOVA to simplify the protocol and not
have to deal with GEM_NEW/GEM_INFO potentially being synchronous.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16086>
These paths should be corner cases, but still it is a bad idea to block
in the host (because it is single threaded), so instead just turn waits
in the host into polling in the guest.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16086>
If supported by host virglrenderer and host kernel, use userspace
allocated GPU virtual addresses. This lets us avoid stalling on
waiting for response from host kernel until we need to know the
host handle (which is usually not until submit time).
Handling the async response from host to get host_handle is done
thru the submit_queue, so that in the submit path (hot) we do not
need any additional synchronization to know that the host_handle
is valid.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16086>
ring_idx zero is the CPU ring, others map to the priority level, as each
priority level for a given drm_file on the host kernel side maps to a
single fence timeline.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16086>
gfxbench sets these between the gbuffer subpass and the following ones.
They should be no-ops as subpass dependencies. gfxbench vk-5-debug perf
12.8 -> 14.6 fps thanks to getting gmem on the gbuffer rendering.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15982>
gfxbench vk-5-normal has a shader that sampels into a texels[] array at
the top, then in a loop calls a GLSL function passing texels[] in by
value. This resulted in a copy to a temp inside the loop, which got
lowered to scratch stores since it was pretty big.
By doing find_array_copies(), we notice that it's equivalent to
copy_deref, then get to copy-propagate from the array at the top. Then we
only have to set up the scratch array outside of the loop and load_scratch
from it in the called function inside the loop. This also causes there to
be less spilling, stps 1144 -> 354 and ldps 826->36.
However, it doesn't seem to change performance on the test. So, while
this seems to be an improvement for the shader, and we could maybe even do
better by rematerializing the txl samples inside the loop instead of
storing the texture fetches to scratch in the first place, it doesn't
currently seem worth pursuing more optimization of this shader.
No change on freedreno shader-db.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15982>
This was useful for comparing image allocations between gfxbench
gl_5_normal and vk_5_normal to see if rendering was generally equivalent
(formats, MSAA, UBWC choices, and notably gfxbench vk was choosing DXT5
instead of ASTC on non-android builds!)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15982>
Refactored tu6_calculate_lrz_state and added comments.
1) If there is no depth write we could keep LRZ valid with any
compare op, we just have to temporary disable LRZ for incompatible
ops in such case.
2) Found that VK_COMPARE_OP_EQUAL is not compatible with LRZ,
and since it doesn't change LRZ buffer - LRZ could be just
temporary disabled. This fixes rendering of grass/trees in
PUBG mobile on angle.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6127
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16014>
Fix invalid usage of meson objects which violates official meson
specification and thus breaks muon, an implementation of meson
written in C.
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15715>
We've wanted to remove destructors from ralloc's API for a long time (it's
an extra storage cost per ralloc for a rarely-used feature), and for the
suballoc change we'd need to spend more storage on storing the tu_device
pointer per result since destructors don't get anything else but the
pointer passed into them.
Fixes use-after-frees:
=================================================================
==2383==ERROR: AddressSanitizer: heap-use-after-free on address 0xffff88fe1940 at pc 0xffff934f427c bp 0xfffff5481e90 sp 0xfffff5481ea8
WRITE of size 8 at 0xffff88fe1940 thread T0
#0 0xffff934f4278 in list_del ../src/util/list.h:108
#1 0xffff934f4278 in result_destructor ../src/freedreno/vulkan/tu_autotune.c:237
#2 0xffff9377793c in unsafe_free ../src/util/ralloc.c:300
#3 0xffff9377793c in ralloc_free ../src/util/ralloc.c:265
#4 0xffff934f4368 in history_destructor ../src/freedreno/vulkan/tu_autotune.c:229
#5 0xffff9377793c in unsafe_free ../src/util/ralloc.c:300
#6 0xffff9377793c in ralloc_free ../src/util/ralloc.c:265
#7 0xffff934f5990 in tu_autotune_on_submit ../src/freedreno/vulkan/tu_autotune.c:442
[...]
0xffff88fe1940 is located 80 bytes inside of 112-byte region [0xffff88fe18f0,0xffff88fe1960)
freed by thread T0 here:
#0 0xffff9c1c90d8 in __interceptor_free ../../../../src/libsanitizer/asan/asan_malloc_linux.cpp:127
#1 0xffff934f4368 in history_destructor ../src/freedreno/vulkan/tu_autotune.c:229
#2 0xffff9377793c in unsafe_free ../src/util/ralloc.c:300
#3 0xffff9377793c in ralloc_free ../src/util/ralloc.c:265
#4 0xffff934f5990 in tu_autotune_on_submit ../src/freedreno/vulkan/tu_autotune.c:442
#5 0xffff935cf2ac in tu_queue_submit_locked ../src/freedreno/vulkan/tu_drm.c:997
[...]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15038>
We don't return unused space to the suballocator, so it's a little useful
to limit how much we overallocate to reduce memory footprint. I took a
look through the tu_cs_emit_array() calls and accounted for a couple of
them in the variant-specific space calculation, then dropped the base
allocation by factors of 2 until we started throwing asserts.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15038>
Allocating a BO for each pipeline meant that for apps with many pipelines
(such as Asphalt9 under ANGLE), we would end up spending too much time in
the kernel tracking the BO references.
Looking at CS:Source on zink, before we had 85 BOs for the pipelines for a
total of 1036 kb, and now we have 7 BOs for a total of 896 kb.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15038>
Bit 20 isn't actually MERGEDREGS, the mode for the entire geometry
pipeline is controlled by SP_VS_CTRL_REG0::MERGEDREGS and it appears to
be something preamble-related instead since writing any register in the
preamble hangs if it's set. This fixes those hangs on freedreno and
turnip since we no longer set it.
Fixes: fccc35c2de ("ir3: Add preamble optimization pass")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15801>
We don't need to restrict our timeout to 5 seconds, because the kernel's
hangcheck will ensure that the wait completes in finite time if the GPU
gets wedged. If the GPU is making progress, we don't want to time out
early and have pipe_transfer_map() return an error, causing glReadPixels()
to throw a confusing GL_OOM even though we're not out memory.
The INFINITE arg to this function isn't actually infinite, it's limited to
an hour. But an hour of GPU processing to wait on is probably plenty.
This 5s timeout has caused problems with the CTS on freedreno at high
parallelism, and I suspect is the cause of recent issues in the closed
traces replay jobs.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15805>
Similar to pipeline statistics but done for a single counter.
We use REG_A6XX_RBBM_PRIMCTR_7 to get generated primitives
and not PRIMCTR_8 because PRIMCTR_7 counts pre-clipped prims
while PRIMCTR_8 counts them after clipping.
OpenGL spec for GL_PRIMITIVES_GENERATED says:
"Subsequent rendering will increment the counter once for every
vertex that is emitted from the geometry shader, or from the
vertex shader if no geometry shader is present."
Passes tests:
dEQP-VK.transform_feedback.primitives_generated_query.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15746>
deqp-runner uprevved to reduce memory usage on HW runners, let us
experiment with shader cache on tmpfs, and hopefully provide a tool for
virgl to be able to plausibly run piglit under crosvm instead of vtest.
piglit uprevved to avoid a flake in softpipe in glx-multithread-texture,
and improve performance of the test, too. This also brings in the
fbo-blending-format-quirks fix to properly initialize the buffers, fixing
some fails/flakes.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15419>
Unaligned gmem store is a mostly untested path since most
of the times faster path is chosen. We have to force unaligned
store to really test it.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15773>
Cube descriptors require a different sampling instruction in shader,
however we don't know whether image is a cube or not until the start
of a renderpass. We have to patch the descriptor to make it compatible
with how it is sampled in shader.
For the reference subpassLoad is currently translated into isaml.a
Blob v615 also doesn't handle this case correctly.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15734>
Vulkan spec says:
"When an image view of a depth/stencil image is used as a depth/stencil
framebuffer attachment, the aspectMask is ignored and both depth and
stencil image subresources are used."
Since we use two planes for D32S8 format we have to add a special
case for depth in addition to already existing case for stencil.
Fixes hang in CTS:
dEQP-VK.renderpass.depth_stencil_write_conditions.stencil_kill_write_d32sf_s8ui
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15532>
- When resolving d32s8 to s8 we stored stencil with a wrong format.
- For unaligned multi-sample store we used wrong gmem offset for stencil.
If unaligined store is forced this change fixes a hang in:
dEQP-VK.renderpass2.depth_stencil_resolve.image_2d_32_32.samples_2.d32_sfloat_s8_uint_separate_layouts.compatibility_depth_zero_stencil_zero_testing_stencil
Fixes: b157a5d0d6
("tu: Implement non-aligned multisample GMEM STORE_OP_STORE")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15532>
Signed-off-by: Omar Akkila <omar.akkila@collabora.com>
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15668>
If we know the height is 1, then it would be a waste to align each
miplevel to tile height. For non-mipmapped textures, it doesn't save us
memory (since you still align to 4 on the last miplevel), but it should be
better cache locality by not loading those unused lines.
Incidentally, this gets us some more coverage of swap != WZYX cases in CTS
tests, which often use optimal tiling without also testing linear.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15293>
This field does appear to work as expected: with 1D/1DArray turnip storage
images switched to be always linear, it fixes the dEQP-VK.image.*store*
tests using a color swapped format (once we allow color swap).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15293>
This reports all of our storage formats as supporting read/write without
format, since we don't have any in-shader format conversions. Similarly,
shadow comparisons were already supported on all the depth formats.
This extension is required for VK 1.3.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15293>
Display Stream compression (DSC) compresses the display stream in
host which is later decoded by panel. This requires addition of 3 new
DSI registers to support DSC over DSI.
Signed-off-by: Vinod Koul <vkoul@kernel.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14967>
Moving duplicate vk_format helper functions to common
vulkan/util/vk_format.h and also renaming
vk_format_get_component_size_in_bits to match how amd and
freedreno name the same function. Not moving this function
to common code as freedreno's implementation is a bit different.
Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15696>
It seems that a660 has the same bug. Without the workaround there
are a lot of flakes with depth-stencil tests, e.g. in:
dEQP-VK.pipeline.extended_dynamic_state.*
dEQP-VK.renderpass.depth_stencil_write_conditions.*
dEQP-VK.pipeline.stencil.format.d24_unorm_s8_uint.states.*
Or guaranteed failures like of:
dEQP-VK.pipeline.render_to_image.core.2d.huge.width.r8g8b8a8_unorm_d32_sfloat_s8_uint
Enabling the workaround fixes all of them.
cc: mesa-stable
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15548>
This test was never supposed to be skipped, and the referenced commit
just exposed a bug in turnip fixed by the previous commit. It was
hanging due to a CTS bug making the submit take way too long, which will
be fixed once the CTS change lands.
Also, add it to the a630 skips.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15563>
In this case we should relax checks based on the format, since the user
will be responsible for them when creating an image view.
This gets dEQP-VK.image.sample_texture.*_bit_compressed_format_* not
skipping again after VK-GL-CTS 736eec57dc0c ("Fix checkSupport in
compressed texture sampling tests").
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15563>
We have to add WFM to pending bits when we are flushing into CP
for indirect draw to know when they should apply WFM workaround.
Fixes CTS tests:
dEQP-VK.draw.renderpass.indirect_draw.*_data_from_compute.indirect_draw_count*
Fixes: abf0ae014a
("tu: Properly handle waiting on an earlier pipeline stage")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15577>
We're trying to replace VK_OUTARRAY_MAKE() by VK_OUTARRAY_MAKE_TYPED()
so people don't get tempted to use it and make things incompatible with
MSVC (which doesn't support typeof()).
Suggested-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15522>
Add a new backend to enable using native driver in a VM guest, via a new
virtgpu context type which (indirectly) makes host kernel interface
available in guest and handles the details of mapping buffers to guest,
etc.
Note that fence-fd's are currently a bit awkward, in that they get
signaled by the guest kernel driver (drm/virtio) once virglrenderer in
the host has processed the execbuf, not when host kernel has signaled
the submit fence. For passing buffers to the host (virtio-wl) the egl
context in virglrenderer is used to create a fence on the host side.
But use of out-fence-fd's in guest could have slightly unexpected
results. For this reason we limit all submitqueues to default priority
(so they cannot be preepmted by host egl context). AFAICT virgl and
venus have a similar problem, which will eventually be solveable once we
have RESOURCE_CREATE_SYNC.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
Call backend specific cleanup fxn earlier. This is needed if the
backend has things like bo's to delete, otherwise the handle_table
will already be destroyed causing problems in bo_del()
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
We are going to want basically the identical thing, other than
flush_submit_list, for virtio backend. Now that we've moved various
other dependencies into the base classes, extract out an abstract base
class for submit/ringbuffer.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
The virtio backend will want this too, and it will make it easier to
share most of the submit/ringbuffer implementation with the virtio
backend.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
With userspace fences, if we know definitely that the buffer is idle
(which implies that it is not shared with other processes, etc), then
skip the ioctl.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
There are some buffers that we mmap just to write to them a single time.
Add the possibility of the drm backend to provide an alternate upload
path to avoid these mmap's.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
With the virtio backend we will need to pass an extra flag when
allocating buffers that will be shared cross-device (such as with
virtio-wl for passing between host and guest)
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
Add a hint for buffers that we won't need to mmap. With the virtio
backend, virglrenderer needs to create a dmabuf fd for mapping into
the host, which we want to avoid when possible.
Low hanging fruit is to use this hint for anything tiled/ubwc. There
are probably more bo's that can be flagged as such.
TODO add fd_bo_upload() for memcpy to bo.. this would be useful for
uploads, for example, shaders which we just write once and never touch
again.. for virtio this could be implemented with a TRANSFER_TO_HOST
ioctl.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
Decoupling handle and fd_bo creation simplifies things for "normal" drm
drivers, avoiding duplication for the create vs import paths. But this
is awkward for the virtio backend when wants to do multiple things in
the same guest<->host round trip.
So instead, split the paths in the interface backend and move the code
sharing for the two different paths into the msm backend itself.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14900>
R8G8 have a different block width/height and height alignment from other
formats that would normally be compatible (like R16), and so if we are
trying to, for example, sample R16 as R8G8 we need to demote to linear.
Follows the fix in Freedreno: b97e3bb2e1
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15465>
ir3_compile_shader_nir() should set local_size[] and local_size_variable
fields not only for compute shaders, but for the OpenCL kernels too.
v2: use gl_shader_stage_is_compute() instead of explicit comparison with
MESA_SHADER_[COMPUTE,KERNEL].
Signed-off-by: Andrey Konovalov <andrey.konovalov@linaro.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14863>
We get a lot of useful coverage from running graphicsfuzz with spilling
enabled, but it's also pretty slow and can cause intermittent hangcheck
failures. I thought I'd categorized them when merging !14839 (device loss
on reset), but it looks like not all of them and we're now more likely to
have flakes take out the whole test run when a single flake makes the rest
of the caselist a flake.
This is a little unfortunate in that it means our test environment is not
the same as a stock system you would want to run deqp on to submit
conformance, but I think it's an improvement in the test maintenance work
vs needing to fix things up later.
We have some other tests besides turnip that can trigger hangchecks which
we might also like this increase for (some disabled traces, for example).
However, freedreno GL has a 5-second timeout waiting for idle when
mapping, and a couple of 2-second timeouts in a row can result in spurious
failures in other tests!
Fixes: #6163
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15435>
The previous alignment of 64 bytes, which we got from the blob,
indicates that single-texel alignment isn't supported. So just do a
trivial no-op implementation that returns the same alignment as before.
This matches what newer blobs that expose this extension do.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15427>
This reuses the same UBO analysis to do the pushing in the shader
preamble via the ldc.k instruction instead of in the driver via
CP_LOAD_STATE6. The const_data UBO is exempted as it uses a different
codepath that isn't as critical.
Don't do this on gallium because there are some regressions. Aztec Ruins
in particular regresses a bit, and nothing I've benchmarked benefits.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13148>
Since the NIR pass runs very late, it needs to be aware of preambles,
and when creating the instruction we need to move it to the start block
so that RA doesn't overwrite it in the preamble.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13148>
Add in the type, even though it turns out to not be that useful. Add
in support for assembling it. Add some notes based on computerator
experiments. And add support for the indirect a1.x mode that's needed
for storing c64.x and later.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13148>
These will be used to implement the ir3-specific shader preamble
lowering in NIR. shps is conceptually similar to getone (although it
technically can't be duplicated) and shpe is similar to other barriers,
since it has to happen after any stores to the constant file in the
preamble. Add NIR intrinsics and plumbs them through ir3.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13148>
Previously we included the reserved user consts (for Vulkan push
constants) as part of the pushed UBO contents, but that led to a problem
because when calculating the worst-case space for UBOs we didn't factor
in the reserved user consts. We'll have the same problem when doing the
same thing in the preamble optimization pass. Stop including the
reserved size in ubo_state::size, and have ir3_setup_consts() add it in
instead, so we won't forget to add it anywhere.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13148>
ANGLE-on-venus-on-turnip and zink-on-turnip want real data here for EGL's
reset tests.
This required moving the remaining GPU-reset-causing tests from flakes or
xfails to skips. Otherwise, the rest of the caselist associated with them
ends up being marked as fails as well. The alternative would be to put
these tests in their own test groups with tests_per_group = 1, but that
didn't seem worth the effort. Or, we could finally do something with
https://gitlab.freedesktop.org/anholt/deqp-runner/-/issues/14.
Fixes: #5955
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14839>
If the base register is killed, it may be reused as the destination of a
ldp. In that case we should just skip resetting it afterwards.
Fixes regressions in dEQP-VK.ssbo.layout.random.scalar.38 later.
Fixes: 9912c61362 ("ir3/spill: Support larger spill slot offset")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15288>
We need to prepare for storage buffers having different sizes from
uniform buffers. This switches dynamic_offset_offset to have units of
bytes, the same as offset, and as a nice bonus we can more easily
combine the dynamic and non-dynamic paths in various different places.
This also entails rewriting the code that patches dynamic descriptors,
since we can no longer assume a linear mapping between indices in
dynamicOffsets and descriptor locations which the previous approach
heavily relied on.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15288>
Normally this wouldn't matter, but it will matter for the upcoming scan
macro because the running tally is communicated through a shared
register across a physical edge. It may also matter if a live-range
split occurs.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14107>
We weren't considering the other destinations when allocating a
destination, so we could allocate overlapping destinations. This wasn't
done before because we never had a need for it, but the subgroup
reduction macros will need it.
The trickiest part of this is that we have to rewrite the
compress_regs_left fallback, because we may have to move around the
other already-allocated destinations. We now have a list of destinations
to (re)allocate in addition to the popped live intervals. For the rest
of the destination handling, we can just bail out if the proposed spot
for something overlaps another destination, but for the fallback we have
to handle all the cases gracefully. I also added support for odd
combinations of multiple destinations where some of them are tied, which
we'll use in the next commit to handle early-clobber destinations and
which will actually be used because one of the destinations of the
subgroup reduction macro will be early-clobber. The result is that the
order of intervals to allocate is now a lot more complicated.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14107>
For pcopies we only care about the register's type, i.e. whether its a
half-register and whether it's an array (plus its size). Copying over
other flags like IR3_REG_RELATIV just leads to sadness and validator
assertions.
Fixes: 0ffcb19b9d ("ir3: Rewrite register allocation")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14107>
Before, we were careful to
1. Get the source physreg.
2. Allocate the destination.
3. Insert a copy with the source being the physreg from step 1.
and this guaranteed that if the tied source were moved in step 2 we'd
still insert a copy from the correct place. However this won't work with
multiple destinations because an earlier destination could've already
moved the tied source around. Instead flip steps 2 and 3 (we'll insert
the copy before we allocate the interval, but that's ok) and run the
first two steps in a separate loop before any destinations are
allocated.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14107>
Note: this is a behavior change for arrays, because it will count the
entire array instead of just the components written in the register
pressure calculation. However this is more accurate since this matches
how RA works.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14107>
I noticed it was using getenv directly when I tried to use 'setprop
mesa.tu.debug ..' on android. Use os_get_option() instead so we get
sysprop fallback on android.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15289>
This is to match other NIR terminology.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15103>
This doesn't actually get run in CI, but this helps track outstanding
issues / expectations. This is from a run on my IFC6540 with A420.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15251>
Experiments suggest that e.g.
add.u r0.y, hr0.x, hr0.y
will result in the summed value in both the high and low words of r0.y.
This only happens with odd registers, not even ones (r0.x works fine).
Seen in the bit_count lowering (which turns out to be unnecessary, but
this is still a larger problem).
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15251>
Enables rasterization order attachment access for all pipelines,
see VK_ARM_rasterization_order_attachment_access for details.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15262>
Mainly the workgroup id comes injected via consts by the hardware (or
CP), and we must make room for it, otherwise the driver won't know where
to put it.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14794>
No need to check error return and deal with older kernels. Older
kernels won't have this param but their default behavior allows for
systemwide perfcntr collection.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15236>
Something is slightly off in the integer values returned. It passes many
tests without the fixup, but the dEQP-GLES31 tests complain. The blob
ends up doing 3x gathers, and selects between them based on getinfo
results. Since we already have a per-sampler key with some spare bits,
just stick the bit-size info in there. And we can derive signedness from
the associated type info.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14670>
It can't be done. This just provides bad results. The blob had a
comparable approach where they fixed up coordinates, but that also can't
work with a separate texture definition with nearest filtering. By then,
might as well provide a unswizzled variant instead, and using native
functionality.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14670>
In the bindless case, we don't have to keep any shadow descriptors and
can just reuse the IBO descriptor as a texture descriptor. Now that
we're emitting the swizzle we can just flip this on.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15114>
ncoords includes the array index, and the NIR source has the array index
as its last component, so we have to insert the extra y coordinate in
the middle in this case.
Fixes: 0bb0cac ("freedreno/ir3: handle image buffer")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15114>
Since these were reverse-engineered, it's become clear that IBO
descriptors are just a subset of texture descriptors, and bindless reads
of readonly images actually use isam on the IBO descriptor, further
confirming that the two are always compatible, even if not all of the
texture fields exist for IBOs. It's pointless to have a separate type
for IBOs, and just leads to things getting out-of-sync unnecessarily
which has already happened. Just remove it and use TEX_CONST insted.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15114>
CAN_REORDER takes volatile into account, and is closer to what we
actually require to use texture instructions, which is that we can
arbitrarily reorder loads.
Fixes: aa93896 ("freedreno/ir3: adjust condition for when to use ldib")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15114>
From Vulkan spec for VkDrmFormatModifierProperties2EXT:
"drmFormatModifierTilingFeatures is a bitmask of VkFormatFeatureFlagBits
that are supported by any image created with format and drmFormatModifier."
"The returned drmFormatModifierTilingFeatures must contain at least one bit."
"Therefore, if the returned drmFormatModifier is DRM_FORMAT_MOD_LINEAR,
then drmFormatModifierPlaneCount must equal the format planecount, and
drmFormatModifierTilingFeatures must be identical to the
VkFormatProperties2::linearTilingFeatures returned in the same pNext chain."
Relevant tests: dEQP-VK.drm_format_modifiers.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15032>
This commit adds support for Vulkan backend on a630_skqp job.
= Needed changes
- Needed to install libvulkan-dev package on system
- Refactored the way the available skqp reports are printed
tested in development builds with skia tools
Piglit expectations had to be updated in various drivers due to !14750 not
having bumped the tags when it tried to uprev.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14686>
The Android CTS 10 version is relative old when compared with skia main
branch, which was being used before. Some modifications in the skqp
build/runner scripts were needed to make it run on CI.
- skqp versions from android-cts have already all assets inside
platform_tools folder.
- along with the assets, are the render and unit files which are
expected to pass in the Android CTS execution.
- removed custom test files from the a630 folder, to make it comply
with the CTS expectations.
- include new patches to remove Python2 dependencies and avoid the
installation of it in rootfs.
- strip binariesthe built binaries `skqp` and `list_gpu_unit_tests`, as
`is_debug = false` gn argument did not work, maybe it is not well
tested in development builds with skia tools
- use Clang instead of GCC. The GCC support is not so graceful as it is
in the skia main branch, some NEON instructions needs to be turned off
in the GCC compilation, causing different tests result. This change
does not imply a bigger rootfs, since the built skqp binary uses GCC
libc++ and other library runtimes. So clang is just a build
dependency.
= Changes in skqp results =
Some errors were found for GL backend and unit tests. GLES and VK tests are green.
All the failed tests were classified as expected to fail in the render and unit tests list.
```
gl_blur2rectsnonninepatch
gl_bug339297_as_clip
gl_bug6083
gl_dashtextcaps
```
```
SRGBReadWritePixels (../../tests/SRGBReadWritePixelsTest.cpp:214 Could not create sRGB surface context. [OpenGL])
```
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14686>
Add a global-level variable that allows disabling all jobs that would
have gone to the Collabora lab, to be used in case of outages.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15150>
5nm PHY is a variation of 7nm PHY, they use the same register
definitions. To remove duplication, drop 5nm defs.
Cc: Robert Foss <robert.foss@linaro.org>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15051>
For ordinary feedback loops GMEM is a lot faster than sysmem since
we don't set SINGLE_PRIM mode.
For feedback loops with ordered rasterization GMEM should also be
faster.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15106>
Trivially implemented by using A6XX_GRAS_SC_CNTL_SINGLE_PRIM_MODE.
This extension is useful for emulators e.g. AetherSX2 PS2 emulator and
could drastically improve performance when blending is emulated.
Relevant tests:
dEQP-VK.rasterization.rasterization_order_attachment_access.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15106>
Otherwise a shader invocation would read the value which should have
been set AFTER this shader invocation.
Fixes tests:
dEQP-VK.rasterization.rasterization_order_attachment_access.depth.samples_1.multi_draw_barriers
dEQP-VK.rasterization.rasterization_order_attachment_access.stencil.samples_1.multi_draw_barriers
Fixes: 71595a189a
("tu: Fix feedback loops in sysmem mode")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15106>
I chose to implement this as a global flag in the device, because
otherwise we would end up with extra draw overhead trying to avoid it in
the implicit-sync WSI case, and you're probably going to end up needing
implicit sync anyway because you used one of the BOs in any of the
submitted cmdbufs. To do better than this, we would probably want a
skip-implicit-sync flag on the BOs in the BO list, rather than global on
the submit.
Reports about venus on turnip say that this flag reduces worst-case
QueueSubmit time in a game workload from ~10ms to ~4ms.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14838>
The zw were already known, but throw them in here too. I'm not extremely
happy with the description of "y", feels like there's a simpler
explanation there, but I couldn't find it.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14672>
Until now we have lived without a refcount mechanism in the driver
because in Vulkan the user is responsible for handling the life
span of memory allocations for all Vulkan objects, however,
imported BOs are tricky because the kernel doesn't refcount
so user-space needs to make sure that:
1. When importing a BO into the same device used to create it
(self-importing) it does not double free the same BO.
2. Frees imported BOs that were not allocated through the same
device.
Our initial implementation always freed BOs when requested,
so we handled 2) correctly but not 1) on drm and we would
double-free self-imported BOs because kernel doesn't return
a unique gem_handle on each import.
Beside this the submit ioctl checks for duplicates in the
BO list and returns an error if there is one.
This fixes the problem for good by adding refcounts to BOs
so that self-imported BOs have a refcnt > 1 and are only freed
when all references are freed.
KGSL on the other hand does not have the same problems,
at least not with ION buffers which are used for exportable
BOs on pre 5.10 android kernels.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5936
Fixes CTS tests: dEQP-VK.drm_format_modifiers.export_import.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15031>
We've got lots of VK coverage on 618, so take some of the load off (but
leave a little bit of testing just to make sure we don't totally break
630). This should help with our Marge times since we've added some other
coverage to 630 that's started overloading the runners.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15085>
Commit 38800b38 changed nir_opcodes.py, but that doesn't seem to have
triggered nir_opt_algebraic.py. The change in 75ef5991 depends on
opt_algebraic lowering 16-bit versions of slt, but if opt_algebraic is
not rebuilt, this may not happen. This resulted in some people seeing
assertion failures in, for example,
dEQP-VK.spirv_assembly.instruction.compute.float16.arithmetic_3.step,
due to the backend seeing nir_op_slt that it didn't know how to handle.
v2: Add nir_opcodes.py to nir_algebraic_py so that all the per-driver
algebraic passes pick up the dependency too. Rename it to
nir_algebraic_depends. Suggested by Emma.
Closes: #6047
Fixes: d1992255bb ("meson: Add build Intel "anv" vulkan driver")
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15050>
Once we simplified a phi node, we never updated the definition it points
to, which meant that it could become out of date if that definition were
also simplified, and we didn't check that when rewriting sources. That
could happen when there are multiple nested loops with phi nodes at the
header.
Fix it by updating the phi's pointer. Since we always update sources
after visiting the definition it points to, when we go to rewrite a
source, if that source points to a simplified phi, the phi's pointer
can't be pointing to a simplified phi because we already visited the phi
earlier in the pass and updated it, or else it's been simplified in the
meantime and this isn't the last pass. This way we don't need to
keep recursing when rewriting sources.
Fixes: 613eaac7b5 ("ir3: Initial support for spilling non-shared registers")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15035>
Both Venus and Android AHB requires this extension.
Turnip ignores VK_SHARING_MODE_EXCLUSIVE so this is a no-op.
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Acked-by: Rob Clark <robdclark@chromium.org>
Acked-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14836>
STL/LDL have 13 bits to store imm offset.
Fixes crash in CS compilation in Monster Hunter World.
Fixes: b024102d7c
("freedreno/ir3: Use nir_opt_offset for removing constant adds for shared vars.")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14968>
Some bits are slightly different on a4xx. Use the encodings that work.
Perhaps these can be combined at some point if we get a proper
understanding of what they mean.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14789>
This is necessary for some ops which have slightly different encoding on
a4xx/a5xx, but are otherwise identical. This helps keeping the compiler
from having to worry about these details and creating separate ops.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14789>
Looks like 3 implementations already have that field in their private
command_buffer struct, and having it at the vk_command_buffer opens the
door for generic (but suboptimal) secondary command buffer support.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14917>
Expose most of the counters exposed by blob. By faking the value of
counters returned from kgsl I found the exact underlying counters and
constant coefficients being used.
Note, coefficients for counters that depend on time are NOT verified.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14323>
The main thing is VK 1.3 testing, but also includes test bugfixes. The
1.3 CTS required an uprev of deqp-runner to handle a new style of test
output, and that deqp-runner brings in some neat new features, too (piglit
in your deqp-runner suite, and extension list checking).
A bunch of VK tests got renamed, so I replaced panvk's custom test list
with simple include filters on the main test list.
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com> (panvk)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14920>
Graphics Flight Recorder is:
"The Graphics Flight Recorder (GFR) is a Vulkan layer to help
trackdown and identify the cause of GPU hangs and crashes.
It works by instrumenting command buffers with completion tags."
This is a nice little tool which could help quickly identify the call
which hanged. Or if command buffer is executed for too long.
The tiling nature of our GPU shouldn't be a big issue aside from
lower performance.
For non-segfault case, if:
- Hang happens at the same place in cmdbuf and draw/dispatch is not
finished at that point - it is likely that there is an infinite
loop in some of the shaders in this draw.
- Hang happens always in different place - likely there is nothing
wrong and command buffer just takes too long to execute and you
should try increasing hangcheck_period_ms. If it doesn't help
it is likely a synchronization issue.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13553>
We didn't remove desc set from the pool's list if pool was
host_memory_base. On the other hand in there is no point in removing
desc set from the list in DestroyDescriptorPool/ResetDescriptorPool.
Fixes: da7a4751
("turnip: Drop references to layout of all sets on pool reset/destruction")
Fixes cts tests:
dEQP-VK.api.buffer_marker.graphics.default_mem.bottom_of_pipe.memory_dep.draw
dEQP-VK.api.buffer_marker.graphics.default_mem.bottom_of_pipe.memory_dep.dispatch
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14855>
Meson devenv is a feature added in meson 0.58 (thus the features is
version guarded) that allows creating a shell environment with
environment variables automatically setup for running the project inside
the build dir. Some variables (such as LD_LIBRARY_PATH and PATH) are set
automatically, others must be added by the project.
For vulkan is is relativley simple, we create a new, uninstalled, icd
file for each driver and set the VK_ICD_FILENAMES variable
appropriately. This can be used with:
```sh
meson devenv -C $builddir
```
then, vulkan applications will automatically use the uninstall vulkan
driver, no need to install.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14826>
This is entirely implemented in the SPIR-V frontend.
Relevant CTS tests:
dEQP-VK.spirv_assembly.instruction.compute.non_semantic_info.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14829>
Moved nir_lower_compute_system_values to lower
load_local_invocation_index which could be emitted by
nir_zero_initialize_shared_memory.
Relevant CTS tests:
dEQP-VK.compute.zero_initialize_workgroup_memory.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14829>
VK_EXT_image_robustness is a strict subset of VK_EXT_robustness2
so we could just expose it.
Relevant CTS tests: dEQP-VK.robustness.image_robustness.*
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14829>
This brings in some interesting new vulkan tests and fixes for the
spurious KHR-GL TF failures. Also, reduces the runtime of
dEQP-GLES31.functional.ssbo.layout.random.all_shared_buffer.36 so that it
should stop timing out.
Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13779>
This ran into OOM-kills with the CTS uprev. Looking at caselists at the
time of fail, some had 500MB of system memory used by the CTS (mostly
spirv string codegen), plus whatever BOs were allocated, and the lazors
are only 4GB it looks like.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13779>
Between adding features and increased test coverage, we're hitting the
1-hour job limit. !13441 tried to increase the full run timeout for LAVA,
but by having not bumped the gitlab-ci timeout value it ended up just
letting the job keep running in LAVA after gitlab had given up on it.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13779>
In the autotune merge, we added another 1/15th run of a configuration
knob, thinking that was small enough to be in the noise. But actually the
main run is only 1/9th, so another 1/15th took us from nearly hitting the
job runtime target, to totally missing it. Crank things back down to keep
MRs flowing.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13779>
We need to check the source of the copy, not the destination. That means
this we need to move this check inside the ifs, where we know that the
source is a copy.
Fixes: 590efd180b ("Prevent propagating shared regs out of loops")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14412>
The old code was both too aggressive and not aggressive enough in
inserting (jp), and it wasn't based at all on a solid understanding of
how the hardware operates. It inserted an extra unnecessary (jp) at the
beginning of an if statement right after the conditional branch, which
is unnecessary. On the other hand, the only case where it didn't
insert a (jp) was a block with one predecessor that has only one
successor. We usually don't emit these kinds of blocks, or if we do then
one of the blocks is empty, but there is a case where we *do* and the
difference actually matters: for something like
while (true) {
if (...) {
// do stuff
...
break;
}
}
The instructions inside the if could be moved below the loop, except
that they are supposed to execute before control flow is merged if the
loop trip count is nonuniform. The subgroup reduce/scan macro does
exactly this, and relies on the control flow being correct. We have to
insert a (jp) after the loop, which this code wasn't doing, breaking the
scan macro.
Since we're now using the physical edges in a non-trivial way, we have
to preserve them better when we modify control flow in ir3_legalize.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14412>
We recently had a bug of forgeting to add the buf->bo_offset. Just make
the easiest field to get be the bo->iova + buf->bo_offset already. Plus,
a little less work at emit time.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14816>
After autotuner introduction most CTS tests are running in
sysmem mode. Now we have to force gmem run and add a small
forced sysmem run since it's not guaranteed that autotuner
would select gmem in future.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12128>
The implementation is separate from Freedreno due to multithreading
support.
In Vulkan application may fill command buffer from many threads
and expect no locking to occur. We do introduce the possibility of
locking on renderpass end, however assuming that application
doesn't have a huge amount of slightly different renderpasses,
there would be minimal to none contention.
Other assumptions are:
- Application does submit command buffers soon after their creation.
Breaking the above may lead to some decrease in performance or
autotuner turning itself off.
The heuristic is too simplistic at the moment, to find a proper
one - we should run a bunch of traces with sysmem and gmem, and
build better heuristic from gathered data.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12128>
The automatic VK coverage we care about is happening on a618, which is the
HW we're shipping. Having the old 630 runners make sure we don't leak
memory is a great use for them. Still, keep one default A630 VK job to
make sure we don't totally trash it.
Reviewed-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14235>
For GL, we disable the dlclose() call on the driver in asan builds so that
leak reports get proper backtraces. For Vulkan, the dlclose() happens
from libvulkan so you need a bigger hammer to keep our drivers loaded.
Acked-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14235>
We dropped the references only for non-host_memory_base pools.
Create a list of alive descriptor to account for all of them.
Fixes: 1b513f49 ("tu: add reference counting for descriptor set layouts")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14235>
Perfetto requires time in clock snaphots to be monotonic, otherwise
the clock would be excluded.
GPU timestamps start from zero after every suspend-resume cycle
which makes them non-monotonic.
As a solution on msm we check whether GPU was just resumed and
remember previous highest timestamp to then add it to the next
timestamps.
If the functionality to get whether gpu is resumed is unavailable
or doesn't work - we fallback to a check for a discontinuity
in timestamps. For kgsl we always use fallback.
Fixes renderstage timeline disappearing in AGI.
Or you could avoid the issue altogether by preventing GPU from going to
sleep by increasing auto suspend delay e.g.:
echo 5000 > /sys/devices/platform/soc\@0/3d00000.gpu/power/autosuspend_delay_ms
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14391>
Fixes the case when last cmd buffer in submission doesn't have
tracepoints leading to flush data not being freed.
Added a few comments, renamed things, refactored allocations - now
the data flow should be a bit more clean.
Extracted submission data creation into tu_u_trace_submission_data_create
which would be later used in in tu_kgsl.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14391>
We shouldn't do ioctl to get timestamp if perfetto isn't connected.
Also it's better to sync timestamps after submission since the
call could block until GPU is resumed.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14391>
The new runner reduces the runtime by about 1/3 thanks to using rust
instead of python, and includes automatic flake handling so you don't just
have to skip flaky tests. The wrapper script also includes IRC flake
reporting (so one can track and update the flakes list to improve CI
reliability), always uploading results to CI for review (so you can
diagnose flakes and look at timings), has a prettier regressions report
and a helpful timing report, and is the same as what's used by all the HW
runners as well.
The downside is that by dropping the massive list of skips, you no longer
get flagged if Mesa refactors end up accidentally disabling extensions and
thus making tests skip. For that, I've started on
https://gitlab.freedesktop.org/anholt/deqp-runner/-/merge_requests/33 so
that hardware drivers get extension checking coverage too.
Thanks to the perf improvement, we get to drop one of the jobs for
llvmpipe.
xfail lists were mostly sed-jobs from the prior expectations lists. The
exceptions to that you'll find in the form of whitespace around the
affected test group (usually changes of capitalization or
special-characters), or an explanation for the more interesting changes
(which thankfully we can now record in the xfails lists!).
Reviewed-by: Cristian Ciocaltea <cristian.ciocaltea@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14604>
The spec states that descriptor set layouts can be destroyed almost
at any time:
"VkDescriptorSetLayout objects may be accessed by commands that operate
on descriptor sets allocated using that layout, and those descriptor
sets must not be updated with vkUpdateDescriptorSets after the descriptor
set layout has been destroyed. Otherwise, a VkDescriptorSetLayout object
passed as a parameter to create another object is not further accessed
by that object after the duration of the command it is passed into."
Copied mostly from ANV.
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5893
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14622>
This allows the exported fds to be mapped for writing. My use case is
for virtio-gpu blob resources where the fds are mapped rw and mappings
are added to the guests using KVM_SET_USER_MEMORY_REGION.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14699>
The current msm_kgsl.h header in the tree isn't sanitized and the kernel
specific macros will confuse a compiler. Copy in the sanitized version of
the header from the 4.19 kernel tree which also adds a few new API bits
that are currently unused but may be useful some day.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14651>
This was copied from the blob before we understood what it did, and it
has questionable utility: there's nothing in the GL, Vulkan, or D3D11
specs that require the result be clamped to the underlying range to
account for imprecision. And it doesn't make sense at all for cubic
filtering, because the result can legitimately be outside the range in
some scenarios. Just remove it.
This fixes a bunch of tests added in vulkan CTS 1.2.8 to test blitting
from compressed textures, which use random inputs and therefore are more
likely to hit the out-of-range condition. For example,
dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.color.2d.etc2_r8g8b8a8_unorm_block.r8g8b8a8_snorm.general_general_cubic.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14613>
This appears to do the same thing as CLAMPENABLE on a3xx. That is, it
clamps the result to [0, 1] for unorm formats and [-1, 1] for snorm
formats *after* filtering. In particular it's now more easily observable
with cubic filtering, because cubic filtering can produce values outside
the original range. Presumably this only matters with linear filtering
due to rounding errors when computing the weighted average.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14613>
With the proper version checking in the common vulkan instance code
(commit 88b9b68) it is now possible to bring the reported interface
version up to v5.
Reviewed-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14563>
This lets nir-to-tgsi fold the constant offset of addressing calculations
into the CONST[] reference, which is important for D3D9-era compatibility:
HW of that age has limited uniform space, and if we do the addressing math
as math in the shader for dynamic indexing, the nir_load_consts end up
taking up uniforms we don't have available.
r300:
total instructions in shared programs: 1279699 -> 1279167 (-0.04%)
instructions in affected programs: 134796 -> 134264 (-0.39%)
total instructions in shared programs: 1279699 -> 1279167 (-0.04%)
instructions in affected programs: 134796 -> 134264 (-0.39%)
total temps in shared programs: 213912 -> 213736 (-0.08%)
temps in affected programs: 2166 -> 1990 (-8.13%)
total consts in shared programs: 953237 -> 952973 (-0.03%)
consts in affected programs: 45980 -> 45716 (-0.57%)
Acked-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14309>
This creates an internal shader_prim enum, I've fixed up most
users to use it instead of GL types.
don't store the enum in shader_info as it changes size, and confuses
other things.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14605>
To avoid dragging gl.h into places it has no business being,
defined tessellation primitive mode to an enum.
This has a lot of fallout all over the place.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14605>
The test
`KHR-GLES31.core.shader_storage_buffer_object.basic-stdLayout_UBO_SSBO-case2-cs`
was failing even before the kernel uprev
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14548>
Rob Clark
Chia-I Wu
Danylo Piliaiev
Emma Anholt
illiliti
Tomeu Vizoso
Connor Abbott
Jason Ekstrand
Omar Akkila
Vinod Koul
Rajnesh Kanwal
Boris Brezillon
Andrey Konovalov
Tapani Pälli
Timur Kristóf
Ilia Mirkin
Guilherme Gallo
Dmitry Baryshkov
Ian Romanick
Yiwei Zhang
Louis-Francis Ratté-Boulianne
Dylan Baker
Hyunjun Ko
Jordan Crouse
Bas Nieuwenhuizen
Charles Giessen
Dave Airlie
Marcin Ślusarz