Commit 64d6f56ad2 ("panfrost: Allocate syncobjs in panfrost_flush")
aimed at optimizing the fencing logic but it looks it also broke the
fence-based synchronization in subtle ways.
Indeed, now that the fence only waits on a single syncobj, we're not
guaranteed that all jobs queued in panfrost_flush_all_batches() will
be done when the fence is signaled, because jobs at the top level
(those stored in the batches hashmap) have not inter-dependencies.
Commit 9e397956b0 ("panfrost: signal syncobj if nothing is going to
be flushed") made this even more apparent by signaling the fence right
away if nothing was left to be drawn in the current context, thus
ignoring any of the batches left to flushed in the ->batches map.
If we want to keep relying the existing kernel APIs there's clearly no
ideal solution here. We can either go back to the original fencing
mechanism where each fence contained an array of syncobjs to be tested
or serialize jobs that have no explicit dependencies so we know the last
submitted job will also be the last one to return. The orginal approach
has proven to add quite a significant overhead (caused by the amount of
ioctls and the time spent in kernel space to gather dma fences attached
to those syncobjs and test them). So let's go for the simple solution
where we have a single syncobj bound to the context which we update to
point to the last job out_sync every time we submit a top-level job.
This approach implies reworking the way we create fences since we
need to capture the syncobj state at the time the fence is created.
Unfortunately, there's not SYNCOBJ_CLONE ioctl, which forces us to
export/create/import a fence so we have a new object that's not
subject to changes done to the context syncobj.
If we want to further optimize the logic, we should probably explore
some of those options:
1/ Adding array based SYNCOBJ ioctls (SYNCOBJ_{CREATE,DESTROY,CLONE}_ARRAY)
so we can mitigate the cost of ioctls when we need to manipulate
arrays of syncobjs
2/ Support synchronization jobs. That is, jobs that have a NULL job chain
but an array of sync_in and a sync_out to allow creating
synchronization points
3/ Add syncobj aggregators so we only have to wait on one syncobj from
userspace. The syncobj aggregator would wait for all sub syncobjs to
be signaled before signaling the top-level one.
Fixes: 64d6f56ad2 ("panfrost: Allocate syncobjs in panfrost_flush")
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7831>
(cherry picked from commit 29f938a0ec)
We shouldn't reset the ->writer field when a reader comes in because we
want subsequent readers to have a dependency on the writer too. Let's
add a new field encoding the last access type and use it to replace the
writer != NULL test.
Reported-by: Roman Elshin
Fixes: c6ebff3ecd ("panfrost: Remove panfrost_bo_access type")
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7831>
(cherry picked from commit 387221e4f2)
among all Android gen rules '::' was used only here to declare dependencies;
mesa development and stable branch are worth receiving the fix
Fixes the following building errors with Android 7:
obj/STATIC_LIBRARIES/libmesa_nir_intermediates/spirv/gl_spirv.P:184: *** target file
gen/STATIC_LIBRARIES/libmesa_nir_intermediates/spirv/vtn_generator_ids.h' has both : and :: entries. Stop.
Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Fixes: 1070bba19e ("android: fix SPIR-V -> NIR build")
Reported-by: youling257 <youling257@gmail.com>
(cherry picked from commit 185df8ef07)
Apparently LRZ will be read/written regardless of depth being enabled or
not, so we have to make sure these registers are zero.
Fixes: 1d83f5ae84 ("turnip: disable LRZ on vkCmdClearattachments() 3D fallback path")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>
(cherry picked from commit fa16e66a3f)
There is an early return if cmd->state.predication_active is true, so do
the LRZ invalidate before that.
Fixes: 2f79e00664 ("turnip: disable LRZ on vkCmdClearAttachments()")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>
(cherry picked from commit f24358e002)
The packet size is constant and assumes all states, except for the 2 input
attachment states. (this means we get an invalid packet if DIRTY_LRZ isn't
set when DIRTY_DRAW_STATE is set).
Fixes: 3c07a14998 ("turnip: enable LRZ")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7899>
(cherry picked from commit af6e74bca8)
In the Chrome WebGL Aquarium stress test, 20 instances of Chrome will run
Aquarium simultaneously over 20+ hours. That causes Chrome crash.
During the stress, glBeginQueryIndexed is called frequently.
1.Each query will only use 32 bytes from query_buffer_uploader. After the offset
exceed 4096, it will alloc new buffer for query_buffer_uploader->buffer
and release the old buffer.
2.But iris_begin_query will call u_upload_alloc when the offset changed, and it
will increase the query_buffer_uploader->buffer->reference.count every time
when it called u_upload_alloc.
3.So when u_upload_release_buffer try to release the resource of
query_buffer_uploader->buffer, its reference.count is
already equal to 129. pipe_reference_described will only decrease its reference
count to 128.So it never called old_dst->screen->resource_destroy.
4.The old resouce bo will never be freeed. And chrome will called mmap every time
when it alloc new resource bo.
5. Chrome process map too many vmas in its process. Its map count exceed the
sysctl_max_map_count which is 65530 defined in kernel.
6. When iris_begin_query want to alloc new resource bo, it will meet NULL pointer
because mmap return failed. Finally chrome crashed when it access this NULL resource
bo.
The fix is decrease the reference count in iris_destroy_query.
Patch is verified by chrome webgl Aquarium test case for more than 72 hours.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Yang Shi <yang.a.shi@intel.com>
Reviewed-by: Alex Zuo <alex.zuo@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7890>
(cherry picked from commit 3aaac40b12)
avoids errors seen when building on OpenBSD/amd64
../src/amd/compiler/aco_instruction_selection.cpp:1677:62: error: ambiguous conversion for functional-style cast from 'unsigned long' to 'aco::Operand'
bld.vop3(aco_opcode::v_mul_f64, Definition(dst), Operand(0x3FF0000000000000lu), tmp);
^~~~~~~~~~~~~~~~~~~~~~~~~~~
glibc uses unsigned long for uint64_t on LP64 archs and unsigned long long for
uint64_t on ILP32 archs. On OpenBSD unsigned long long is used for uint64_t
on all archs.
The Operand constructors are uint8_t uint16_t uint32_t uint64_t
use UINT64_C so lu or llu suffix will be used as needed.
Fixes: df645fa369 ("aco: implement VK_KHR_shader_float_controls")
Signed-off-by: Jonathan Gray <jsg@jsg.id.au>
Reviewed-by: Tony Wasserka <tony.wasserka@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7944>
(cherry picked from commit ebfb9e1817)
The feature was exposed but completely ignored by the driver. Other
AMD drivers don't expose it as well, probably because it's complicated
to implement alpha-to-coverage properly. Let's disable it.
Cc: mesa-stable.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7966>
(cherry picked from commit 59b1578176)
PIPE_MAX_CONSTANT_BUFFERS is 32, however many Vulkan implementations
has maxPerStageDescriptorUniformBuffers that exceeds it, for example:
radv 8388606,
anv 64
nvidia 1048580 for RTX 2000 and up.
and, together with the current zink logic, the returned value
will exceed the maximum allowed value for the cap.
This causes cso_destroy_context to pass big values back to zink
(via zink_set_constant_buffer), resulting in access beyond end of
allocated buffer for all UBOs.
Cap the cap to PIPE_MAX_CONSTANT_BUFFERS (32), not INT_MAX.
Add an assert to verify future drivers.
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Fixes: daaf5f1d18 ("gallium: Fix leak of currently bound UBOs at CSO context destruction.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7976>
(cherry picked from commit e2b4247e40)
This looks like a typo. Packed vulkan formats should always map to the
inverse order of the corresponding gallium notation. Besides, it makes
no sense that unsigned and signed formats have different ordering.
Fixes: cdfb1d925f ("zink: add last few format maps for ARB_vertex_type_2_10_10_10_rev")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7964>
(cherry picked from commit 8bb4a76add)
There's no good reason why drivers that doesn't grok geometry,
tesselation or compute shaders needs to deal with them.
This fixes a crash on a lot of Piglit tests for Zink.
Fixes: daaf5f1d18 ("gallium: Fix leak of currently bound UBOs at CSO context destruction.")
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7971>
(cherry picked from commit 3abac03d49)
If a subpass uses multiview but the fragment shader doesn't load it
we still have to export it.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7815>
(cherry picked from commit fadcf13c8b)
In theory, GFX10.3 is not considered to be a conformant Vulkan
implementation because we didn't submit a conformance submission
package.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7913>
(cherry picked from commit 5cacb56041)
Fixes a use-after-free of the state on the next iteration when it was
probably just destroyed.
Fixes: 6de01faac5 ("freedreno/a6xx: invalidate tex state cache entries on rebind")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7695>
(cherry picked from commit d3c67d7e7e)
Conflicts:
src/gallium/drivers/freedreno/a6xx/fd6_texture.c
llvmpipe was reporting 32 max const buffers, while sizing its arrays to 16
according to gallivm's #define.
Fixes: 1d35f77228 ("gallivm,llvmpipe,draw: Support multiple constant buffers.")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7695>
(cherry picked from commit d90107a200)
This is only required for the DRI-path. For the swrast code-path, we
don't need this.
We also don't need to explicitly test for it in the DRI-path, because we
test for KHR_external_memory_fd, which depends on KHR_external_memory. So
no implementation will expose the former without the latter.
Fixes: f1432fd3e2 ("zink: generate extension infrastructure using a python script")
Reviewed-by: Hoe Hao Cheng <haochengho12907@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7882>
(cherry picked from commit 59a6705cce)
computerator depends on ir3_parser.h, which is a generated file, but
this dependency is not expressed in the build.
Fixes: 1e8808a4a0 ("freedreno/ir3: refactor out helper to compile shader from asm")
Signed-off-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7870>
(cherry picked from commit 9eee405484)
I think that reducing the CU mask to 1 disabled CU per SA broke the WGP mode
on VanGogh, causing a hang. To be sure, disable it on all chips.
Fixes: 9538b9a68e - radeonsi: add support for Sienna Cichlid
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7721>
(cherry picked from commit 69c927debe)
If you did:
si_pm4_set_reg(pm4, reg, val0);
si_pm4_cmd_add(pm4, val1);
si_pm4 set_reg(pm4, reg + 4, val1);
it wrote val0 to reg, val1 to reg + 4, and val2 to reg + 8.
This fixes it by clearing last_opcode in si_pm4_cmd_add, so that
si_pm4_set_reg doesn't try to combine set_reg calls across si_pm4_cmd_add.
Fixes: da78d50bc8 - radeonsi: make si_pm4_cmd_begin/end static and simplify all usages
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7721>
(cherry picked from commit 0d4f1dcd15)
In commit 00b28a50b2, Marek extended
a number of optimizations that had been 32-bit specific to work on
other bit-sizes.
Most optimizations preserve the data type across the transformation.
In other words, an optimization which generates e.g. fp64 operations
only does so when the source expression also contains fp64 operations.
These transformations are fine with respect to lowering, because we
will lower away all expressions that would trigger the search portion
of the expression, and so we'd never apply those rules.
However, a few of the rules create new operations that run afoul of
lowering passes. For example,
('bcsel', a, 1.0, 0.0) => ('b2f', a)
where the result is a double would simply be a selection between two
different 64-bit constants. The replacement expression, on the other
hand, involves a nir_op_b2f64 ALU operation. If we're run after
nir_lower_doubles, then it may not be legal to generate such an
expression anymore (at least without running lowering again, which we
don't do today).
Regressions due to this are blocking the 20.3 release, so for now, we
take the easy route and simply disallow those few rules when doing full
softfp64 lowering, which fixes the immediate problem. But it doesn't
solve the long-term problem in an extensible manner.
In the future, we may want to add a `lowered_alu_ops` bitfield to the
NIR shader, and as lowering passes are run, mark them as taboo. Then,
we could have each algebraic transformation track which operations it
creates in the replacement expression. With both of those in place,
nir_replace_instr could compare the transformation's list of ALU ops
against `lowered_alu_ops` and implicitly skip rules that generate
forbidden ALU operations.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3504
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7841>
(cherry picked from commit 531843cf2e)
This workaround fixes a hang while loading a renderdoc trace for me.
Since the workload does 1 mip per cmdbuffer it is quite hard to confirm
what exactly the conditions for the hang are but this is the most
restrictive set I found and it corresponds to a workaround in AMDVLK as
well.
CC: mesa-stable
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7210>
(cherry picked from commit 4cce4d22a7)
