Commit Graph

10 Commits

Author SHA1 Message Date
Bart Oldeman b073d759d0 x86_64: clobber all xmm registers after vzeroupper
As observed using GCC 10 using -march=native -ftree-vectorize
on Knights Landing, it is now smart enough to find clobbers inside
non-inlined static functions.

In particular, sgemv counted on a kernel to preserve the whole
%ymm2 register (since it was not in the clobber list), but the top
part was destroyed by vzeroupper. This caused many tests to fail.

This patch makes sure all xmm (and ymm/zmm by extension) registers
are listed as clobbered to avoid this happening, as most kernels
already did correctly in fact.
2020-10-20 02:16:47 +00:00
Martin Kroeker 4255a58cd2
Rename operands to put lda on the input/output constraint list 2019-02-15 10:10:04 +01:00
Martin Kroeker 46e415b140
Save and restore input argument 8 (lda4)
Fixes miscompilation with gcc9 -ftree-vectorize (related to issue #2009)
2019-02-14 22:43:18 +01:00
Martin Kroeker 723f396a20
Tag %1 and %2 as both input and output
The inline assembly modifies its input operands, so mark them as output to avoid surprises with optimization. Fixes #1292
2017-12-29 23:56:41 +01:00
Zhang Xianyi 6e7be06e07 Refs JuliaLang/julia#5728. Fix gemv performance bug on Haswell Mac OSX.
On Mac OS X, it should use .align 4 (equal to .align 16 on Linux).
I didn't get the performance benefit from .align. Thus, I deleted it.
2016-02-19 17:56:07 -05:00
Werner Saar bc5fff7085 changed inline assembler labels to short form 2014-12-07 12:38:54 +01:00
wernsaar 7c0a94ff47 bugfix in sgemv_n_microk_haswell-4.c 2014-09-08 10:54:33 +02:00
wernsaar cbbc80aad3 added optimized sgemv_t kernel for haswell 2014-09-08 10:13:39 +02:00
wernsaar cf5544b417 optimization for small size 2014-09-06 13:17:56 +02:00
wernsaar d143f84dd2 added optimized sgemv_n kernel for haswell 2014-09-06 12:08:48 +02:00