Growing the batch/state buffer is a lot more dangerous than I thought.
A number of places emit multiple state buffer sections, and then write
data to the returned pointer, or save a pointer to brw->batch.state.bo
and then use it in relocations. If each call can grow, this can result
in stale map references or stale BO pointers. Furthermore, fences refer
to the old batch BO, and that reference needs to continue working.
To avoid these woes, we avoid ever swapping the brw->batch.*.bo pointer,
instead exchanging the brw_bo structures in place. That way, stale BO
references are fine - the GEM handle changes, but the brw_bo pointer
doesn't. We also defer the memcpy until a quiescent point, so callers
can write to the returned pointer - which may be in either BO - and
we'll sort it out and combine the two properly in the end.
v2/v3:
- Handle stale pointers in the shadow copy case, where realloc may or
may not move our shadow copy to a new address.
- Track the partial map explicitly, to avoid problems with buffer reuse
where multiple map modes exist (caught by Chris Wilson).
v4:
- Don't use realloc in the CPU shadow case, it isn't safe.
Fixes: 2dfc119f22 "i965: Grow the batch/state buffers if we need space and can't flush."
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com> [v3]
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
'aux' is a very generic name, suggesting it can be a bunch of things.
However, it's always the brw_*_prog_data structure. So, call it that.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part 2 of 2 (part 1 is autoconf changes, part 2 is C++ changes)
When only a single SWR architecture is being used, this allows that
architecture to be builtin rather than as a separate libswrARCH.so that
gets loaded via dlopen. Since there are now several different code
paths for each detected CPU architecture, the log output is also
adjusted to convey where the backend is getting loaded from.
This allows SWR to be used for static mesa builds which are still
important for large HPC environments where shared libraries can impose
unacceptable application startup times as hundreds of thousands of copies
of the libs are loaded from a shared parallel filesystem.
Based on an initial implementation by Tim Rowley.
v2: Refactor repetitive preprocessor checks to reduce code duplication
v3: Formatting changes per Bruce C. Also delay screen creation until end
to avoid leaks when failure conditions are hit.
Signed-off-by: Chuck Atkins <chuck.atkins@kitware.com>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
CC: Tim Rowley <timothy.o.rowley@intel.com>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
Part 1 of 2 (part 1 is autoconf changes, part 2 is C++ changes)
When only a single SWR architecture is being used, this allows that
architecture to be builtin rather than as a separate libswrARCH.so that
gets loaded via dlopen. Since there are now several different code
paths for each detected CPU architecture, the log output is also
adjusted to convey where the backend is getting loaded from.
This allows SWR to be used for static mesa builds which are still
important for large HPC environments where shared libraries can impose
unacceptable application startup times as hundreds of thousands of copies
of the libs are loaded from a shared parallel filesystem.
Based on an initial implementation by Tim Rowley.
v2: Fix comment placement pointed out by Bruce C.
Signed-off-by: Chuck Atkins <chuck.atkins@kitware.com>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
CC: Tim Rowley <timothy.o.rowley@intel.com>
Reviewed-by: Bruce Cherniak <bruce.cherniak@intel.com>
v2: Don't annotate, but remove the unused ctx parameter
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
v2: Don't annotate, but remove the unused ctx parameter
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
As a followup to the previous patch propagate the change of numSamples
from int to unsigned to gl_config::samples and consequently fix some
-Wsign-compare warnings.
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
According to the ARB_multisample num_samples is a non-negative integer.
Consequently define it as such, fail in glx/choose_visual if a negative
number is given.
v2: split patch into gallium and mesa part
Signed-off-by: Gert Wollny <gw.fossdev@gmail.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
For scons, windows/mingw dealing with LLVM_CONFIG is done before
untarring. This is also more convenient for copy and paste.
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Cc: Emil Velikov <emil.velikov@collabora.com>
Cc: Juan A. Suarez Romero <jasuarez@igalia.com>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
vk_error() is a macro that calls __vk_errorf() with instance == NULL.
Then, __vk_errorf() passes a pointer to instance->debug_report_callbacks
to vk_debug_error(), which segfaults as this pointer is invalid but not
NULL.
Fixes: e5b1bd6ab8 "vulkan: move anv VK_EXT_debug_report implementation to common code."
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
When a channel was not set we also did not increase the LDS address,
while that obviously should happen.
The output loading code was inadvertently fixed which resulted in a
mismatch causing the SaschaWillems tessellation demo to result
in corrupt rendering.
Fixes: 7898eb9a60 "ac: rework load_tcs_{inputs,outputs}"
Reviewed-by: Dave Airlie <airlied@redhat.com>
The bindings also have an index field.
Fixes: 49d035122e "radv: Add single pipeline cache key."
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104677
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Passes
dEQP-VK.api.smoke.*
dEQP-VK.wsi.android.*
with android-cts-7.1_r12 .
Unlike the initial anv implementation this does
use syncobjs instead of waiting on the CPU.
This is missing meson build coverage for now.
One possible todo is that linux 4.15 now has a
sycall that allows us to export amdgpu fence to
a sync_file, which allows us not to force all
fences and semaphores to use syncobjs. However,
I had trouble with my kernel crashing regularly
with NULL pointers, and I'm not sure how beneficial
it is in the first place given that intel uses
syncobjs for all fences if available.
Reviewed-by: Dave Airlie <airlied@redhat.com>
lp_build_interleave2_half was not doing the right thing for avx512-style
16-wide loads.
This path is hit in the swr driver with a 16-wide vertex shader. It is
called from lp_build_transpose_aos, when doing texel fetches and the
fetched data needs to be transposed to one component per output register.
Special-case the post-load swizzle operations for avx512 16x32 (16-wide
32-bit values) so that we move the xyzw components correctly to the outputs.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
The optimization in change 8e4efdc895 ("vbo: optimize some display
list drawing") missed the loopback case. This is used when the
glBegin/End primitive doesn't have a uniform set of vertex attributes.
The new Piglit gl-1.0-dlist-materials test hits this.
So check the aligned_vertex_buffer_offset(list) value and adjust the
buffer offset accordingly.
We also need to remove the 'start == 0' assertion in the loopback
code since it no longer applies.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Currently a couple of gallium targets race with xmlpool_options.h being
generated, don't do that.
Signed-off-by: Dylan Baker <dylan.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The view index user sgpr wasn't being accounted for properly,
this refactors out the code to decide if it's required and then
uses that info to account for it.
Fixes: 180c1b924e (ac/nir: Add shader support for multiviews.)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Only one piglit test fails,
sso-vs-gs-fs-array-interleave
There are 3 tests using ssbo without checking sizes failing also
but those are test bugs.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The kernel is moving to a $class$instance naming scheme in preparation
for accommodating more rings in the future in a consistent manner. It is
already using the naming scheme internally, and now we are looking at
updating some soft-ABI such as the error state to use the new naming
scheme. This of course means we need to teach aubinator_error_decode how
to map both sets of ring names onto its register maps.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Portal 2 appears to bind RGBA8888_UNORM textures to a sampler2DShadow,
and calls shadow2D() on it. This causes undefined behavior in OpenGL.
Unfortunately, our sampler appears to hang in this scenario, which is
not acceptable. Just give them a null surface instead, which returns
all zeroes.
Fixes GPU hangs in Portal 2 on Kabylake.
Huge thanks to Jason Ekstrand for noticing this crazy behavior while
sifting through crash dumps.
Cc: mesa-stable@lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104487
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
From the Vulkan spec with KHX extensions:
"If queries are used while executing a render pass instance that has
multiview enabled, the query uses N consecutive query indices
in the query pool (starting at query) where N is the number of bits
set in the view mask in the subpass the query is used in.
How the numerical results of the query are distributed among the
queries is implementation-dependent. For example, some implementations
may write each view's results to a distinct query, while other
implementations may write the total result to the first query and write
zero to the other queries. However, the sum of the results in all the
queries must accurately reflect the total result of the query summed
over all views. Applications can sum the results from all the queries to
compute the total result."
In our case we only really emit a single query (in the first query index)
that stores the aggregated result for all views, but we still need to manage
availability for all the other query indices involved, even if we don't
actually use them.
This is relevant when clients call vkGetQueryPoolResults and pass all N
queries to retrieve the results. In that scenario, without this patch,
we will never see queries other than the first being available since we
never emit them.
v2: we need the same treatment for timestamp queries.
v3 (Jason):
- Better an if instead of an early return.
- We can't write to this memory in the CPU, we should use
MI_STORE_DATA_IMM and emit_query_availability (Jason).
v4 (Jason):
- No need to take the value to write as parameter, just hard code it to 0.
Fixes test failures in some work-in-progress CTS multiview+query tests.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Since setup of ALLOW_RGB10_CONFIGS was moved to i965's own
brw_config_options.xml, this was hard-coded to false and
could not be overriden by drirc. Add some parsing into
i965's private screen->optionCache to enable drirc again.
Fixes: b391fb26df ("dri_util: remove ALLOW_RGB10_CONFIGS option (v2)")
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Cc: Marek Olšák <marek.olsak@amd.com>
Cc: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This was handled for VS, but not for GS.
Fixes for gallium drivers using nir:
spec@arb_gpu_shader5@arb_gpu_shader5-xfb-streams-without-invocations
spec@arb_gpu_shader5@arb_gpu_shader5-xfb-streams*
spec@arb_transform_feedback3@arb_transform_feedback3-ext_interleaved_two_bufs_gs*
spec@ext_transform_feedback@geometry-shaders-basic
spec@ext_transform_feedback@* use_gs
spec@glsl-1.50@execution@geometry@primitive-id*
spec@glsl-1.50@execution@geometry@tri-strip-ordering-with-prim-restart gl_triangle_strip *
spec@glsl-1.50@transform-feedback-builtins
spec@glsl-1.50@transform-feedback-type-and-size
v2: don't call st_translate_program_stream_output) for TCS
v3: drop scanning patch outputs as TCS can't output xfb
Signed-off-by: Rob Clark <robdclark@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Tested-by: Karol Herbst <kherbst@redhat.com>
if no destination:
a) convert _RET instructions to non _RET variants if no dst
b) set src0 to undefined if it's a READ, this should get DCE then.
Acked-By: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The normal ssa renumbering isn't sufficient for LDS queue access,
this uses two stacks, one for the lds queue, and one for the
lds r/w ordering.
The LDS oq values are incremented in their use in a linear
fashion.
The LDS rw values are incremented in their definitions and used
in the next lds operation to ensure reordering doesn't occur.
Acked-By: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
So LDS ops have to be SLOT_X,
and LDS OQ reads have read port restrictions so we try
and force those into only having one per slot and avoiding
bank swizzles.
Acked-By: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This tries to avoid an lds queue read getting scheduled separately
from an lds ret read, the non-sb code uses the same style of hammer,
this isn't foolproof.
We can do better, but it's a bit tricky, as you have to scan ahead
and either schedule more lds oq moves and more lds reads and that
could lead to you running out of space anyways.
Acked-By: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This adds support for tracking the lds oq read/writes
so can avoid scheduling other things in between.
This patch just adds the tracking and assert to show
problems.
Acked-By: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
You have to schedule LDS_READ_RET _, x and MOV reg, LDS_OQ_A_POP
in the same basic block/clause. This makes sure once we've issues
and MOV we don't add another block until we balance it with an
LDS read.
Acked-By: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>