Resolves the following valgrind issue when running GLES CTS:
==8318== Conditional jump or move depends on uninitialised value(s)
==8318== at 0x72AE91C: vk_format_to_pipe_format (vk_format.c:263)
==8318== by 0x7051EE7: vk_format_description (vk_format.h:176)
==8318== by 0x7052CFB: pvr_get_format_swizzle (pvr_formats.c:701)
==8318== by 0x708CC2F: pvr_init_tex_info (pvr_query_compute.c:454)
==8318== by 0x708D887: pvr_add_query_program (pvr_query_compute.c:663)
==8318== by 0x708B9BB: pvr_CmdCopyQueryPoolResults (pvr_query.c:535)
==8318== by 0x607E68F: copy_pool_results_to_buffer (zink_query.c:782)
==8318== by 0x607EB37: update_qbo (zink_query.c:853)
==8318== by 0x607FA17: zink_get_query_result (zink_query.c:1145)
==8318== by 0x5E82987: tc_get_query_result (u_threaded_context.c:1283)
==8318== by 0x59A7543: get_query_result (queryobj.c:297)
==8318== by 0x59A78A7: _mesa_check_query (queryobj.c:388)
Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
When importing the same dma-buf multiple times, the kernel will return us the
same GEM handle each time. This means that if we close the GEM handle for one BO
we'll end up making the handle invalid for all BOs. The driver wasn't taking
this into account and was therefore creating a new BO each time a dma-buf was
imported, meaning we could end up with multiple BOs with the same GEM
handle. Fix this by creating a single BO per GEM handle and taking an additional
reference on a BO when the kernel returns an existing GEM handle.
Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
When calculating the size of an image, the driver was always factoring in space
for a full mip chain. However, this isn't necessary when mipLevels is 1 and this
resulted in applications needing to allocate more memory for these images than
is strictly necessary. Fix this by calculating the size of additional mip levels
(those greater than mipLevels) when more than 1 mip level has been requested.
Fixes: 2a3aa6da50 ("pvr: Fix cubemap layer stride")
Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
Base level offseting only needs to be applied when creating non-compressed view
of a compressed image. Additionally the logic when patching offset, did not fix
up the rest of image state.
Seen in:
dEQP-VK.pipeline.monolithic.image_view.view_type.2d.format.r32_uint.subresource_range.lod_base_mip_level_array_layer_last
Signed-off-by: SoroushIMG <soroush.kashani@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
pvr_cmd_buffer_end_sub_cmd() sets the current sub-command to NULL. This was
causing list_move_to(), which is called immediately after this, to access a NULL
pointer. Fix this by storing the current sub command before calling
pve_cmd_buffer_end_sub_cmd() so that this can be used instead when modifying the
list.
Fixes: d1b17a5edc ("pvr: Implement ZLS subtile alignment")
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
The descriptor set code will undergo a rework fairly soon so instead of trying
to fixup `per_stage_descriptor_masks` and deal with possible side effects of
that, for now let's always regenerate all descriptor program data sections no
matter the stage mask.
Fixes the following:
dEQP-VK.pipeline.monolithic.dynamic_offset.graphics
.{multi,single_}set.{storage,uniform}_buffer.numcmdbuffers_1
.sameorder.numdescriptorsetbindings_2.numdynamicbindings_1
.numnondynamicbindings_0
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
In `pvr_create_renderpass_hwsetup()` we use the size of the pbe accum format to
setup a mask for the valid parts of the output regs.
For formats such as `E5B9G9R9_UFLOAT_PACK32` which aren't supported as colour
attachments, `vkCreateRenderPass()` can still be called so to avoid undefined
behavior we use `0` as the bitsize.
Fixes `Unknown pbe accum format. Implementation error` for:
dEQP-VK.api.granularity.in_render_pass.e5b9g9r9_ufloat_pack32
dEQP-VK.api.granularity.multi.e5b9g9r9_ufloat_pack32
dEQP-VK.api.granularity.random.e5b9g9r9_ufloat_pack32
dEQP-VK.api.granularity.random.r32g32b32_uint
dEQP-VK.api.granularity.random.r32g32b32a32_sint
dEQP-VK.api.granularity.single.e5b9g9r9_ufloat_pack32
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31679>
In rare cases, it could happen that during post-RA validation,
live-var-analysis sets needs_vcc = false after if was true
before register allocation.
Fixes: bb5eace0dc ('aco/live_var_analysis: check for isPrecolored flag rather than isFixed')
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31791>
Perfetto is allowed to choose it's own default clock, and before this we just assumed the presentation times reported by the compositor are the same as perfetto's internal clock, which is not always the case. I got a nasty trace where all the wayland presents were in the wrong location. This fixes that by asking the compositor which clock it uses, then passing that along to perfetto.
A workaround for my compositor was setting use_monotonic_clock=true in the perfetto config, as my compositor (and I suspect most others) use the monotonic clock for presentation timestamps. However, asking the compositor is definitely the most correct solution.
I added a clock param to `MESA_TRACE_TIMESTAMP_{BEGIN,END}`, as it's only use that I could see was in wsi_common_wayland, and in general it seems good to be careful about which clock tracing timestamps come from.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31779>
In function 'r600_init_resource_fields',
inlined from 'r600_buffer_create' at ../src/gallium/drivers/r600/r600_buffer_common.c:561:2:
../src/gallium/drivers/r600/r600_buffer_common.c:121:48: error: array subscript 'struct r600_texture[0]' is partly outside array bounds of 'unsigned char[256]' [-Werror=array-bounds=]
121 | if ((res->b.b.target != PIPE_BUFFER && !rtex->surface.is_linear) ||
| ^~~~~~~~~~~~~~~~~~~~~~~~
In file included from ../src/util/os_memory.h:37
Signed-off-by: David Heidelberg <david@ixit.cz>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31790>
This moves the generate-gfxstream-vulkan.sh script
to upstream Mesa too. Right now, Mesa is the source
of truth for both guest and host codegen.
There needs to be a simple way to invoke genvk.py
for users. The script assumes the AOSP directory
structure to find gfxstream host, but the user may
also pass the path to gfxstream as the first argument.
Please run this from the src/gfxstream/codegen directory.
Reviewed-by: Marcin Radomski <dextero@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31775>
We used to only store Temps in the stack, so undef meant exec.
Then the stack was changed to operands, and some places started storing exec
directly, drop the undef handling by replacing everything with Operand(exec, lm)
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31560>
