qemu-ppc
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-ppc] [PATCH v2 0/7] target/ppc: switch fpr/vsrl registers so a


From: David Gibson
Subject: Re: [Qemu-ppc] [PATCH v2 0/7] target/ppc: switch fpr/vsrl registers so all VSX registers are in host endian order
Date: Fri, 8 Mar 2019 10:38:44 +1100
User-agent: Mutt/1.11.3 (2019-02-01)

On Thu, Mar 07, 2019 at 06:05:13PM +0000, Mark Cave-Ayland wrote:
> After some investigation into Andrew's report of corruption in his ppc64le
> tests at https://lists.gnu.org/archive/html/qemu-devel/2019-02/msg07234.html, 
> I
> discovered the underlying cause was that the first 32 VSX registers are not
> stored in host endian order.
> 
> This is something that Richard and I had discussed before, but missed that 
> with
> VSX if you have source registers from different register sets then even 
> logical
> operations will give you the wrong result.
> 
> Rather than revert 7b8fe477e1 "target/ppc: convert VSX logical operations to
> vector operations" let's keep the use of the accelerated vector instructions,
> and instead fix the real problem which is to switch the first 32 VSX registers
> to host endian order matching the VMX registers.
> 
> Patches 1-5 aim to consolidate the offset calculations for both CPUPPCState
> and the associated _ptr() functions into one single place.
> 
> With this preliminary work complete, patch 6 switches the first 32 registers
> into host endian order without too much difficulty.
> 
> Finally now that all VSX registers are stored in the same way, the vsr offset
> functions and get_cpu_vsrh()/get_cpu_vsrl() can be simplified accordingly.
> 
> Signed-off-by: Mark Cave-Ayland <address@hidden>

Series applied to ppc-for-4.0, thanks.

> 
> v2:
> - Rebase onto master
> - Rework patchset set based upon av64_offset()/vsr64_offset() as suggested by
>   Richard, rather than using separate low/high accessors
> 
> 
> Mark Cave-Ayland (7):
>   target/ppc: introduce single fpr_offset() function
>   target/ppc: introduce single vsrl_offset() function
>   target/ppc: move Vsr* macros from internal.h to cpu.h
>   target/ppc: introduce avr_full_offset() function
>   target/ppc: improve avr64_offset() and use it to simplify
>     get_avr64()/set_avr64()
>   target/ppc: switch fpr/vsrl registers so all VSX registers are in host
>     endian order
>   target/ppc: introduce vsr64_offset() to simplify get_cpu_vsr{l,h}()
>     and set_cpu_vsr{l,h}()
> 
>  target/ppc/cpu.h                    | 51 
> ++++++++++++++++++++++++++++++++++---
>  target/ppc/internal.h               | 27 +++-----------------
>  target/ppc/machine.c                |  8 +++---
>  target/ppc/translate.c              | 20 +++------------
>  target/ppc/translate/vmx-impl.inc.c | 27 ++++++++------------
>  target/ppc/translate/vsx-impl.inc.c | 39 +++-------------------------
>  6 files changed, 75 insertions(+), 97 deletions(-)
> 

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

Attachment: signature.asc
Description: PGP signature


reply via email to

[Prev in Thread] Current Thread [Next in Thread]