On 2/8/19 10:09 AM, Mark Cave-Ayland wrote:
> On 08/02/2019 17:39, Richard Henderson wrote:
>
>> On 2/7/19 2:42 PM, Mark Cave-Ayland wrote:
>>> Due to a cut/paste error in the original implementation, the unsigned vector
>>> saturating arithmetic was erroneously being calculated as signed vector
>>> saturating
>>> arithmetic.
>>>
>>> Fixes: 8ffafbcec2 ("tcg/i386: Implement vector saturating arithmetic")
>>> Signed-off-by: Mark Cave-Ayland <mark.cave-ayl...@ilande.co.uk>
>>> ---
>>> tcg/i386/tcg-target.inc.c | 4 ++--
>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> Ooof. Thanks.
>
> AFAICT none of the other TCG backends currently make use of it, but it was
> this bug
> causing the graphical corruption that myself and Howard saw testing the PPC
> vector
> patchset.
I would have seen the error if I had done ARM SVE regression testing.
But I only did ARM NEON testing, which does not make use of this, and
I had forgotten that.
Like PPC, NEON sets a "saw saturation" bit. I think I'll do the same
trick as we did for ppc to compute that with vector insns.
> FWIW I've now rebased and repushed the outstanding patches from your patchset
> along
> with this fix to https://github.com/mcayland/qemu/commits/ppc-altivec-v6 as
> it would
> be great if this could make it into 4.0.
I've given all of my r-b...
r~
