Re: [Qemu-devel] [PATCH v3 00/19] target-mips: add MSA module

[Leon Alrae]

On 02/11/2014 23:08, Peter Maydell wrote:
> On 2 November 2014 22:14, Leon Alrae <address@hidden> wrote:
>> On 01/11/14 05:28, Yongbok Kim wrote:
>>> For 2.2 Release.
>>>
>>> The following patchset implements MIPS SIMD Architecture module.
>>> MSA adds new instructions to MIPS Architecture that allow efficient
>>> parallel processing of vector operations.
> 
>>>  disas/mips.c                 |  716 +++++++++-
>>>  target-mips/Makefile.objs    |    2 +-
>>>  target-mips/cpu.h            |   56 +-
>>>  target-mips/gdbstub.c        |    7 -
>>>  target-mips/helper.c         |   10 +
>>>  target-mips/helper.h         |  185 +++
>>>  target-mips/mips-defs.h      |    1 +
>>>  target-mips/msa_helper.c     | 3437 
>>> ++++++++++++++++++++++++++++++++++++++++++
>>>  target-mips/op_helper.c      |   89 +-
>>>  target-mips/translate.c      | 1647 +++++++++++++++++++--
>>>  target-mips/translate_init.c |   39 +-
>>>  11 files changed, 6052 insertions(+), 137 deletions(-)
>>>  create mode 100644 target-mips/msa_helper.c
>>>
>>
>> This patchset certainly looks cleaner than previous versions - thanks!
>> However, there are still some wrinkles to iron out, and also there is
>> room for improvement - for example big-endian host issue, "switch (df)"
>> present in each helper even though data format is known at translation
>> time, missing register partitioning etc.). Having said that I think this
>> patchset deserves to be merged as it introduces significant amount of
>> tested MIPS SIMD instructions. In my opinion further MSA improvements
>> and cleaning can be done as a part of separate patchset (I'll be happy
>> to help doing that), thus I'm going to apply it to mips-next branch.
> 
> Bear in mind that we're only two days from hardfreeze now, and
> this is a big lump of code (6000 lines). If you're at all
> unsure then the safe choice is to let it wait until 2.2 has
> released and schedule it for 2.3 instead.

I would like to send pull request containing MIPS64R6-privileged and
this MSA patchset, I hope it's not too late to get this into 2.2
release. I ran regression tests for MIPS - i.e. Linux images (Aurelien's
Debian, mips32/64 r1/r2/r6-candidate Linux mti), mips32/64 gcc tests as
well as module level tests covering MIPS architectural features (DSP,
microMIPS, MSA, etc.). I did that on 64-bit Linux host. Result was good,
therefore this makes me quite confident that this new stuff doesn't
break existing functionality. Do you think I should perform any other
testing?

Thanks,
Leon