[Mesa-dev] [PATCH 00/12] anv: Add support for the variablePointers feature

Jason Ekstrand jason at jlekstrand.net
Tue Nov 7 03:26:24 UTC 2017


On Mon, Nov 6, 2017 at 6:33 PM, Chad Versace <chadversary at chromium.org>
wrote:

> On Mon 06 Nov 2017, Jason Ekstrand wrote:
> > On Mon, Nov 6, 2017 at 12:18 PM, Chad Versace <[1]
> chadversary at chromium.org>
> > wrote:
> >
> >     Jason, I tested this series against the khronos-internal vk-gl-cts
> and
> >     found an assertion failure in src/compiler/spirv. Any thoughts?
> >
> >     I haven't debugged yet because I don't grok these parts of Mesa.
> >
> >     vk-gl-cts
> >
> >         commit a24448cdd72ffdbd8f7f571886625b8a53100979
> >
> >     mesa
> >
> >         refs/tags/chadv/test/anv-variable-pointers-2017-11-06-r1
> >         cgit: [2]http://git.kiwitree.net/cgit/~chadv/mesa/tag/?h=chadv/
> test/
> >     anv-variable-pointers-2017-11-06-r1
> >         base: master 4bcb48b "radv: add initial copy descriptor support.
> (v2)"
> >
> >     error
> >
> >         Test case 'dEQP-VK.spirv_assembly.instruction.compute.variable_
> >     pointers.complex_types_compute.opptraccesschain_
> >     matrices_two_buffers_second_input'..
> >         INTEL-MESA: debug: anv_GetPhysicalDeviceFeatures2KHR: ignored
> >     VkStructureType VK_STRUCTURE_TYPE_PHYSICAL_
> DEVICE_16BIT_STORAGE_FEATURES_
> >     KHR(1000083000)
> >         deqp-vk: ../../../../../src/mesa/src/
> compiler/spirv/vtn_variables.
> >     c:174: vtn_ssa_offset_pointer_dereference: Assertion `offset'
> failed.
> >
> >
> > That looks very much like a test bug that I fixed.  The CLs have been
> merged
> > into vulkan-gl-cts-1.0.2 but I don't know if that's been merged forward
> yet.
> > Try that CTS version and see if it still fails.  It's CLs #1864 and
> #1863 in
> > gerrit if you're interested.
>
> Weird. I'm testing with vk-gl-cts master a24448c, which contains both of
> those CLs.
>
> To be safe, I did a clean build of libdrm, mesa, and vk-gl-cts (with
> external/fetch_sources.py, of course). The failure remained.
>
> Maybe it's a toolchain problem?
>

No, it's because I failed at finding 100% of the broken tests.  New CL
incoming.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20171106/91b6a070/attachment.html>


More information about the mesa-dev mailing list