[Bug 92760] Add FP64 support to the i965 shader backends

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Mon Jan 25 06:24:52 PST 2016


https://bugs.freedesktop.org/show_bug.cgi?id=92760

--- Comment #37 from Samuel Iglesias <siglesias at igalia.com> ---
(In reply to Samuel Iglesias from comment #36)
> (In reply to Samuel Iglesias from comment #35)
> > I found i965 driver is currently running vec4 visitor and generator for
> > geometry shaders and we have some crashed/failed tests due to its lack of
> > support for doubles. After enabling scalar GS, piglit shows no regressions.
> > 
> 
> I forgot to mention we are currently focusing on the scalar backend to fix
> failed fp64 tests and running the tests on Broadwell (gen8) for now.
> 
> By enabling scalar GS in Broadwell by default, we would take advantage of
> our previous work to add support for doubles in the scalar backend. Thanks
> to that work, scalar GS not only shows no regressions on piglit but also
> passes the crashed/failed fp64's geometry shader tests we have seen with
> vec4 backend.
> 
> Then, my question is if we can enable scalar GS in Broadwell by default or
> if there are reasons to keep it disabled.
> 
> > I would like to know if there is a strong reason to disable scalar GS by
> > default (perhaps I am missing something) because then we would need to add
> > support for doubles in vec4.

Oh, I did a wrong comparative of the piglit results:

There are 11 piglit regressions when enabling scalar GS. Those regressions are
unrelated to fp64 but they show why scalar GS was disabled by default in
Broadwell: the instanced GS support.

7 regressions are related to a failed assert in brw_compile_gs():
assert(prog_data->invocations == 1). Removing that assert, they pass. Is
instanced GS supported already in scalar backend?

The other 4 regressions seem to be a different issue, but I have not analyzed
it yet:

shader_runner: brw_fs_nir.cpp:1813: void fs_visitor::emit_gs_input_load(const
fs_reg&, const nir_src&, unsigned int, const nir_src&, unsigned int): Assertion
`imm_offset == 0' failed.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20160125/3f8f30dc/attachment.html>


More information about the intel-3d-bugs mailing list