Fixed:
- VK_QUERY_TYPE_PRIMITIVES_GENERATED_EXT was able to stop prim counter
when pipeline stats query is running.
- This may have happened when prim gen query was in secondary cmdbuf.
- VK_QUERY_TYPE_PRIMITIVES_GENERATED_EXT counting geometry in each tile.
- VK_QUERY_TYPE_PRIMITIVES_GENERATED_EXT counting geometry in each tile
when pipeline stats query is started inside prim gen query and inside
a renderpass.
The matter of VK_QUERY_TYPE_PRIMITIVES_GENERATED_EXT and pipeline stats
interaction is solved by tracking whether pipeline stats query is
running both on CPU (for non secondary cmdbuf case) and on GPU (for
secondary cmdbuf).
Note, prim gen query is not allowed with secondary command buffers, so
only pipeline stats query is tracked on gpu.
See https://gitlab.khronos.org/vulkan/vulkan/-/issues/3142
Counting geometry per each tile is solved by:
- Conditionally executing START/STOP_PRIMITIVE_CTRS to not run in tiling
pass. Solves the case when prim gen query is inside a renderpass.
- Stop prim counters before executing `draw_cs` and restarting them
afterwards. Solves prim gen query being outside a renderpass.
Fixes GL CTS tests with Zink + `TU_DEBUG=gmem`:
GTF-GL46.gtf30.GL3Tests.transform_feedback.transform_feedback_max_separate
GTF-GL46.gtf40.GL3Tests.transform_feedback2.transform_feedback2_basic
GTF-GL46.gtf40.GL3Tests.transform_feedback2.transform_feedback2_framebuffer
GTF-GL46.gtf40.GL3Tests.transform_feedback3.transform_feedback3_streams_overflow
GTF-GL46.gtf40.GL3Tests.transform_feedback3.transform_feedback3_streams_queried
GTF-GL46.gtf40.GL3Tests.transform_feedback2.transform_feedback2_states
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6602
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17163>
Flushing the batch midframe (splitting a renderpass) is expensive on a tiler, as
it requires the GPU to flush the framebuffer contents to main memory and read
them back. Clearing the framebuffer should not trigger a flush. Apps expect
clears to be (almost) free, flushing for a clear is at the very least unexpected
behaviour.
The only reason we previously flushed is to ensure we could always use a "fast"
clear. But a slow clear is a heck of a lot faster than a flush ;-) Instead of
flushing, we should clear with a draw (via u_blitter) in case a fast clear isn't
possible.
This fixes pathological performance for applications that rely on partial clears
within a frame. This issue was identified with Inochi2D, which repeatedly clears
the stencil buffer midframe, in order to implement masking efficiently with the
stencil buffer. In total, the all-important workload of rendering Asahi Lina is
improved from 17fps to 29fps on a panfrost device.
Fixes: c138ca80d2 ("panfrost: Make sure a clear does not re-use a pre-existing batch")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17112>
I wrote fiction about dreaming that these were supported but after
waking finding that they were not. Somehow I came to consider that
fiction as fact and I never thought to test if they did work.
While reverse engineering the polygon list format, I found that these
were supported after all.
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17371>
This maps to CL_DEVICE_MAX_COMPUTE_UNITS, which is defined as:
The number of parallel compute cores on the OpenCL device.
Since all supported Malis are unified shader cores, the number of parallel
compute cores is simply the number of shader cores.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17265>
To query the core count, the hardware has a SHADERS_PRESENT register containing
a mask of shader cores connected. The core count equals the number of 1-bits,
regardless of placement. This value is useful for public consumption (like
in clinfo).
However, internally we are interested in the range of core IDs.
We usually query core count to determine how many cores to allocate various
per-core buffers for (performance counters, occlusion queries, and the stack).
In each case, the hardware writes at the index of its core ID, so we have to
allocate enough for entire range of core IDs. If the core mask is
discontiguous, this necessarily overallocates.
Rename the existing core_count to core_id_range, better reflecting its
definition and purpose, and repurpose core_count for the actual core count.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17265>
this is for drivers (like freedreno) which need the format in the sampler
state in order to accurately handle border colors
when set, drivers MAY receive a format in the sampler state if the frontend
supports it (e.g., nine does not), and the cso sampler cache will include
the format member of the struct
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17189>
This is a rewite of the NIR backend. it adds some optimization
and a scheduler.
v2: - replace some magic numbers by constants
- make sure constructor is always used with new
- use default initialization in more places
(changes suggested by Filip Gawin)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17076>
We can't unconditionally support separable shader objects on the host,
so submit the property only if the shader is actually separable, the
host knows about the property, and supports SSO.
Without support for SSOs, the host can still compile and link the shaders,
it needs to do more work on interface matching though.
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17344>
Most drivers don't care about the property, and virgl should only handle
it if the host supports it.
This is a partial revert of b634030, i.e. we keep the definition of the
property, but we don't set it only based on the shader info.
Fixes: b634030542
tgsi: Add SEPARABLE_PROGRAM property
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17344>
Now that usage flags can be specified even when using the modifier path for
allocation and frontends like GBM and EGL wayland do this properly, we can
drop the assumption that all resources allocated through the modifier
enabled path need to be SCANOUT capable.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17364>
TGSI has no legitimate[1] notion of linked shaders, which means tgsi_to_nir
should conservatively assume everything all shaders are separable. This requires
setting nir->info.separate_shader to warn drivers that shader CSOs might be
mixed and matched. Otherwise, the driver might enable optimizations that
are invalid for separate shaders, causing issues when the shaders are
later treated as separable.
This will fix varying linking with u_blitter's shaders on Panfrost (Bifrost and
older), when util_blitter_clear is used with Panfrost.
[1] There was a TGSI property added recently to forward
nir->info.separate_shader up to virglrenderer, but it's not actually used for
anything in virglrenderer and I am still struggling to understand what the use
case would be. My gut says we should revert b634030542 ("tgsi: Add
SEPARABLE_PROGRAM property"), but I'm not interested in fighting that yak right
now. Notably, the u_blitter and hud shaders are separable but are not marked
with this property.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17282>
New tests, dEQP line rasterization test fix that lets Intel pass.
Clears out bogus xfails from 1.3.2.0 uprev on a630, which I suspect were
"we lost the device twice on a full run once, and those fails got pasted
in without checking if it happened a full run again" (since we haven't
seen them in other full run attempts).
Also clears out the a630 vk asan xfails (essentially all tests run) by
turning off leak detection which was just catching leaks in vkcts.
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17304>
Georg Lehmann