On Alchemist, the FF_MODE2 documentation says that we must set the
FF_MODE2 timer values for GS and HS to 224. The hardware performance
tuning guide also recommends setting the TDS timer to 4.
On Tigerlake, i915 applies workarounds to set the GS timer to 224
(failing to do so can cause HS/DS unit hangs), and the TDS timer to 4
(for performance). It doesn't currently apply a HS timer there, and
I'm not sure if it's strictly necessary, but given that Alchemist
needed it, and the other two settings matched, let's assume that it
ought to match as well.
Unfortunately, there has been a bug in the i915 workarounds
infrastructure for non-masked context registers where writing one
field of the register zeroes out all the others. So, I believe the
Tigerlake TDS timer value of 4 isn't being applied correctly there,
though the register is also not readable on that platform which
makes it hard to verify. So, this may also speed up tessellation.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9233
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23839>
(cherry picked from commit 1b3669a1ed)
This enables L3 partial write merging for a number of cases that seem
to be getting accidentally disabled by the kernel, which was causing a
serious performance bottleneck on DG2 and MTL platforms. The
"Compressible Partial Write Merge Enable", "Coherent Partial Write
Merge Enable" and "Cross-Tile Partial Write Merge Enable" bits in
L3SQCREG5 were expected to be enabled by default (and confusingly,
they even read off as enabled if you ran 'intel_reg read 0xb158' on an
idle system), but they are getting clobbered during 3D context
initialization by an i915 workaround.
Enabling L3 partial write merging of compressible surfaces in
particular seems to increase rendering fillrate by over 3x in some
cases (e.g. the
"VulkanFillRate/FillRateGPU/resolution:1[0-3]/format:*/blend:0"
fillrate-bound microbenchmarks). Significant improvements can also be
reproduced in most real-world workloads we've tested so far,
e.g. Counter Strike GO improves by ~11%, Shadow Of the Tomb Raider
improves by ~5.5%, and AztecRuins-VK improves by ~6.5% on DG2-512 --
Thanks a lot to Caleb Callaway for these figures. No regressions have
been observed so far.
Even though this patch might strike as surprisingly simple for such a
large payoff, it's the result of Felix DeGrood and I trying to
root-cause the rendering performance gap of DG2 on Linux vs Windows on
and off during the last year, and some of the OA statistics captured
by Felix early this month were greatly helpful for me to connect the
last few dots, so Felix deserves a big chunk of the credit for this
work.
Cc: mesa-stable
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23783>
(cherry picked from commit 427fee3507)
A global aux-context can be created on-demand for cases where we need to
(for example) blit a resource when we only have a screen ptr.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23324>
(cherry picked from commit 75193262fd)
If we consider clearing the group flag of a vec4 register that is used as
source for some instruction we have to take into account that the parent
of the register element may also be part of a group in the parent instruction.
In this case we must not clear the group flag.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/9118
Fixes: f3415cb26a (r600/sfn: copy propagate register load chains)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23813>
(cherry picked from commit 34163e19f7)
there are only 2 ubos that can be emitted, except the emitted ubos
can start at an offset based on the first-used ubo, which means this
has to support the full range of ubo indices
fixes oob access in game Beyond All Reason
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23729>
(cherry picked from commit 9803174942)
Indeed, the vertex state was restored using a specific
condition at the util_blitter_restore_vertex_states()
level. This change ensures that the condition is the
same when the vertex state is saved.
The function util_blitter_clear_buffer() is only called
by the r600 driver on pre-evergreen gpus.
This issue is triggered on a rv770 gpu with "piglit/bin/fbo-1d -auto -fbo"
or "piglit/bin/draw_buffers_gles2 -auto -fbo"
while setting GALLIUM_REFCNT_LOG=refcnt.log.
Fixes: 5f566faa46 ("radeonsi: don't save and restore vertex buffers and elements for u_blitter")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23721>
(cherry picked from commit 23c003b88c)
There is no need to split the CF if only the register ID
in a previous write is the same, we should look at the actual
slots instead, ut we have also to take writes of 0 and 1 into
account.
Cc: mesa-stable
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23710>
(cherry picked from commit 3a569fbf9b)
this fixes the case where a draw without sample shading precedes
a draw with sample shading without changes to the sample mask
Fixes: cc9e958053 ("lavapipe: fix DS3 min sample setting")
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23663>
(cherry picked from commit cb03191e82)
This reverts commit 0523607ebb.
The issue that commit worked around seems to have been fixed as of
commit 1c8b4940eb ("iris: Emit flushes for push constant source
buffers"). I could no longer reproduce it from that point onward with
this revert applied.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18725>
(cherry picked from commit bb6d300b3a)
This change fixes a buffer overflow by implementing the
special swizzles. This behavior is already available with
evergreen_convert_border_color().
For instance, this issue is triggered on a cayman gpu with
"piglit/bin/texwrap bordercolor -auto -fbo" or "piglit/bin/max-samplers -auto -fbo":
==5610==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x603000012d20 at pc 0x7fb798cb876f bp 0x7ffd78670460 sp 0x7ffd78670458
READ of size 4 at 0x603000012d20 thread T0
#0 0x7fb798cb876e in cayman_convert_border_color ../src/gallium/drivers/r600/evergreen_state.c:2444
#1 0x7fb798cb876e in evergreen_emit_sampler_states ../src/gallium/drivers/r600/evergreen_state.c:2539
#2 0x7fb7989e6cb2 in r600_emit_atom ../src/gallium/drivers/r600/r600_pipe.h:655
#3 0x7fb7989e6cb2 in r600_draw_vbo ../src/gallium/drivers/r600/r600_state_common.c:2333
#4 0x7fb7985082c7 in u_vbuf_draw_vbo ../src/gallium/auxiliary/util/u_vbuf.c:1497
#5 0x7fb796ef2eda in cso_draw_vbo ../src/gallium/auxiliary/cso_cache/cso_context.h:262
#6 0x7fb796ef2eda in st_draw_gallium_multimode ../src/mesa/state_tracker/st_draw.c:170
#7 0x7fb7970d9cfd in vbo_exec_vtx_flush ../src/mesa/vbo/vbo_exec_draw.c:341
#8 0x7fb7970d32d7 in vbo_exec_FlushVertices_internal ../src/mesa/vbo/vbo_exec_api.c:693
#9 0x7fb7970d32d7 in vbo_exec_FlushVertices ../src/mesa/vbo/vbo_exec_api.c:1193
#10 0x7fb7975f237c in enable_texture ../src/mesa/main/enable.c:337
Fixes: 923d635357 ("r600: fix some border color swizzles on CAYMAN")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23435>
(cherry picked from commit 4284705733)
Indeed, this function was not processing the linked
allocated list.
For instance, this issue is triggered with "piglit/bin/hiz-depth-read-fbo-d24-s0 -auto":
Indirect leak of 40 byte(s) in 1 object(s) allocated from:
#0 0x7f6795638987 in calloc (/usr/lib64/libasan.so.6+0xb1987)
#1 0x7f678bac13b9 in nouveau_heap_alloc ../src/gallium/drivers/nouveau/nouveau_heap.c:64
#2 0x7f678bb6c7e4 in nv50_program_upload_code ../src/gallium/drivers/nouveau/nv50/nv50_program.c:490
#3 0x7f678bb83b92 in nv50_vertprog_validate ../src/gallium/drivers/nouveau/nv50/nv50_shader_state.c:161
#4 0x7f678bba3000 in nv50_state_validate ../src/gallium/drivers/nouveau/nv50/nv50_state_validate.c:552
#5 0x7f678bba3c4d in nv50_state_validate_3d ../src/gallium/drivers/nouveau/nv50/nv50_state_validate.c:575
#6 0x7f678b9e3e92 in nv50_blit_3d ../src/gallium/drivers/nouveau/nv50/nv50_surface.c:1444
#7 0x7f678b9e3e92 in nv50_blit ../src/gallium/drivers/nouveau/nv50/nv50_surface.c:1832
#8 0x7f678a0b378a in blit_to_staging ../src/mesa/state_tracker/st_cb_readpixels.c:337
#9 0x7f678a0b7358 in st_ReadPixels ../src/mesa/state_tracker/st_cb_readpixels.c:516
#10 0x7f6789f82005 in read_pixels ../src/mesa/main/readpix.c:1178
#11 0x7f6789f82005 in _mesa_ReadnPixelsARB ../src/mesa/main/readpix.c:1195
#12 0x7f6789f82ac0 in _mesa_ReadPixels ../src/mesa/main/readpix.c:1210
...
SUMMARY: AddressSanitizer: 80 byte(s) leaked in 2 allocation(s).
Fixes: 67635a0a71 ("nouveau: get rid of tabs")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23592>
(cherry picked from commit 1980934d0d)
There was a race between the worker thread and flush, which could lead to
the last event flushed getting its status set to CL_SUCCESS before any
other event.
Just wait on all flushed events in order to solve this.
The current queue/event implementation isn't the best and we want to
rework it, so this is good enough for now.
Fixes: 47a80d7ff4 ("rusticl/event: proper eventing support")
Signed-off-by: Karol Herbst <git@karolherbst.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23578>
(cherry picked from commit da4b27452b)
ARB_occlusion_query specifies that the query is reset on BeginQueryARB,
not when the fetching the result of the query. This behavior also makes
a lot of sense for the perfmon queries.
CC: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23557>
(cherry picked from commit b6a4b988ab)
When the RS uses the pixel pipes it seems to destroy/invalidate any
content sitting in the color and depth caches from a previous draw.
Always flush the color and depth cache before using the RS to make
sure that any cache content written by the PE is properly flushed
to memory.
Fixes spec@!opengl 1.0@gl-1.0-drawpixels-depth-test and probably a
few others that are suffering from corruption of PE writes.
CC: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23530>
(cherry picked from commit b1cd5780d6)
Move the TS cache flush into the same conditional block where
the TS setup is changed. TS cache always needs to be flushed
before making any changes to the TS setup.
CC: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23530>
(cherry picked from commit cfc1be9590)
conditional render is only supposed to be enabled during renderpasses,
and this ends up doing mismatched start/stop in and out of renderpasses
affects:
GTF-GL46.gtf30.GL3Tests.conditional_render*
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23511>
(cherry picked from commit 6aa9e95021)
DS_XCHG_RET and LDS_CMP_XCHG_RET don't have a version that doesn't return
a value in the LDS red queue. so we have to read the value from the queue
and discard it.
Fixes: 79ca456b (r600/sfn: rewrite NIR backend)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23518>
(cherry picked from commit 976d6de232)
Since these were added to generic translation, lavapipe has
been annoyed.
Cc: mesa-stable
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23486>
(cherry picked from commit d99dca86ce)
So the shuffleup tests did a shuffle up with const 5,
we'd use invocation id (0..8) shuffle it down by 5,
get (-5..3), then call llvmshufflevector with that
which is totally illegal.
There might be a nicer way to fix this, but I can't see
it straight away, just bail on the fast path.
Fixes:
dEQP-VK.subgroups.shuffle.compute.subgroupshuffleup*
Cc: mesa-stable
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23484>
(cherry picked from commit c20df7e22e)
gallium has no mechanism for setting an offset to the index buffer,
so this requires a new buffer to be created+bound dynamically at
the specified offset in order to read the correct data
cc: mesa-stable
Acked-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23394>
(cherry picked from commit 3209609535)
GPUs without native integers lower idiv in lower_int_to_float and
there is no need to call nir_lower_idiv(..) for such GPUs.
Fixes nir crashes I am seeing with gc2000_gles2 CI job.
Fixes: f532202f2d ("etnaviv: use nir_lower_idiv(..) before opt loop")
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23450>
(cherry picked from commit add14d6cfb)
Instead of incrementing the seqno, just directly invalidate any existing
tex cache entries when we update a sampler view.
No reason not to just directly clear stale entries, and avoiding to re-
assign the seqno will simplify the next patch.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23470>
(cherry picked from commit 3f00f4ac30)
