On 18.05.2021 13:48, Luboš Luňák wrote:
So please leave the SSE2 version in.
But please note that the code must do the same in all cases; so with
SSE2, please also implement Neumaier, not plain Kahan summation.
Otherwise, the results would vary depending on installed HW.
--
Best regards,
On Tuesday 18 of May 2021, Dante Doménech wrote:
> That file is mine: https://gerrit.libreoffice.org/c/core/+/115675
First of all, please learn to write better mails. This should have been
included in the first mail, together with a better mail subject. It's hard to
do anything about a mail wit
That file is mine: https://gerrit.libreoffice.org/c/core/+/115675
The affected header is one of intel's intrinsics.
The file is: immintrin.h
For GCC / clang it depends on: avx512fintrin.h
This is the definition for GCC on github:
extern __inline __m512d
__attribute__ ((__gnu_inline__, __always_in
On 18/05/2021 10:58, Dante Doménech wrote:
I've stumbled with an error on the headers we are using to build (GCC at
least).
I've got this: __m512d _mm512_abs_pd(__m512)
error: cannot convert ‘__m512d {aka __vector(8) double}’ to ‘__m512 {aka
__vector(16) float}’ for argument ‘1’ to ‘__m512d _m
