[PATCH][AArch64] request to push AArch64 patches to release 3.4 branch

Wed Dec 4 19:09:34 PST 2013

Hi Bill,

We have more commitments in the past several days. Could you please help to
merge them all together to branch 3.4?

r196189 r
196190

Add missing intrinsics like vmlsl_high_n_xxx

r196191 r
196192

Add missing SISD misc integer intrinsics
r
196198

r196199

Add missing SISD pair intrinsics
r
196208

r196209

Add missing ACLE intrinsics mapping to general arithmetic operation from
VFP instructions
r
196211

r196211

Add missing floating point convert, round and misc intrinsics
r
196359 r196360

Add missing compare intrinsics
r
196361 r196362

Add missing instrinsics vceqz_f64
r
196456
                      ***
Add missing register cost calculation for big value types like v4i64 and
v8i64

1) The last one marked with "***" is critical, and would break most of -O1
and -O2 code containing vector load store, so this one should go into
branch 3.4 with high priority.
2) All the others covers all of missing intrinsic functions for AArch64
Neon. Hopefully they can go into branch 3.4 as well.

After merging those commitments, for AArch64 Neon work, the status is,
1) MClayer can pass MCHammer test.
2) For intrinsics implementation defined in ACLE 2.0, we will be having two
commitments only. A) Ana's patch for vset/vget_lane_f16, which is being
under code review in mail list. B) The patch for poly128 and related
things. Kevin will probably send out the patch by tomorrow.

Hopefully If possible, 2.A) and 2.B) can both go into branch 3.4. And
that way we would be able to claim we have a complete AArch64 Neon feature
in 3.4 release.

Thanks,
-Jiangning

2013/12/3 Bill Wendling <isanbard at gmail.com>

> Added in. Thanks!
>
> -bw
>
> On Dec 2, 2013, at 11:28 PM, Ana Pazos <apazos at codeaurora.org> wrote:
>
> Hi Bill and Tim,
>
> Thanks for merging all the patches to branch 3.4. We will have a couple
> more coming this week to complete Neon intrinsics implementation.
>
> Here are the patches from me that need to go to 3.4 branch:
> Clang:
> r196206 Fixed vcopy_laneq_f64 intrinsic implementation
> 196153: Implemented vcopy_lane patterns using scalar DUP instruction
> LLVM:
> 196151 Implemented vcopy_lane patterns using scalar DUP instruction
>
> Thank you,
> Ana.
> *From:* Bill Wendling [mailto:isanbard at gmail.com <isanbard at gmail.com>]
> *Sent:* Saturday, November 30, 2013 8:40 PM
> *To:* Jiangning Liu
> *Cc:* Ana Pazos; Jiangning Liu; llvm-commits; cfe-commits
> *Subject:* Re: [PATCH][AArch64] request to push AArch64 patches to
> release 3.4 branch
>
> Hi Jiangning,
>
> Thanks for the list! I added them into the 3.4 release branch. Please let
> me know if there are any I missed.
>
> -bw
>
> On Nov 29, 2013, at 6:36 PM, Jiangning Liu <liujiangning1 at gmail.com>
> wrote:
>
>
> Hi Bill,
>
> Now we have a couple new patches for 3.4 release, so could you please help
> to merge them into the branch?
>
> r195068
> Clean up predefined macros for AArch64 to follow ACLE 2.0.
> r195069 r195071
> All crypto intrinsics should be unavailable when crypto feature is not
> enabled
> r195073 r195074
> Implement SISD intrinsics for vget_high and vget_low
> r195713
> Implement 128 bit register copy with NEON.
> r195903
> *** Fix the problem that AArch64 Post-index LD4 lane is decoded incorrectly
> r195844 r195843 r195905 r195932
> *** Fix the failure exposed by checking constant integer argument range
> r195937
> Add missing intrinsic vcombine_f64
> r195936
> *** Fix An assertion failure caused by disassembling AArch64 SHLL
> instruction
> r195938 r195939 r195940
> Add missing intrinsic bsl_f64
> r195941 r195942
> *** The pattern match should check the range of the immediate
> r195943
> *** Some ACLE intrinsics in arm_neon.h return float32 instead of float64.
>
>
> The ones marked with "***" are critical ones because they are bug fixes
> either for MCLayer or ACLE, and they should go into branch with the high
> priority.
>
> Ideally we want all of them go into branch, because they are combined all
> together as a big feature for AArch64 NEON implementation.
> Thanks,
> -Jiangning
>
>
>
>
> 2013/11/26 Bill Wendling <isanbard at gmail.com>
> Sure. done.
>
> -bw
>
> On Nov 25, 2013, at 8:32 PM, Jiangning Liu <liujiangning1 at gmail.com>
> wrote:
>
>
> Bill,
> Can you merge r195716 to branch 3.4 as well?Thanks,-Jiangning
>
>
> 2013/11/26 Bill Wendling <isanbard at gmail.com>
> Sure. Done.
>
> -bw
>
>
>
> On Mon, Nov 25, 2013 at 9:59 AM, Ana Pazos <apazos at codeaurora.org> wrote:
> Thank you Bill,
>
> We have a fix from Hao that should also go into branch 3.4 (it's a fix of
> encoding error for ld/st instruction): r195591
>
> Can you merge it too, please.
>
> Thanks,
> Ana.
>
>
> -----Original Message-----
> From: Bill Wendling [mailto:isanbard at gmail.com]
> Sent: Sunday, November 24, 2013 9:44 PM
> To: Tim Northover
> Cc: Ana Pazos; llvm-commits; cfe-commits at cs.uiuc.edu
> Subject: Re: [PATCH][AArch64] request to push AArch64 patches to release
> 3.4
> branch
>
> Sure. Merged.
>
> -bw
>
> On Nov 22, 2013, at 10:53 AM, Tim Northover <t.p.northover at gmail.com>
> wrote:
>
> > Hi Ana & Bill,
> >
> >> Committed revision 195326
> >> Committed revision 195329
> >> Committed revision 195327
> >> Committed revision 195330
> >
> > I'm happy for these to go in. The interested parties are keen and they
> > should only affect AArch64 with -mattr=+neon, which I don't think is
> > being widely used at the moment (and itself almost certainly won't be
> > made worse).
> >
> > Cheers.
> >
> > Tim.
>
>
>
>
> _______________________________________________
> cfe-commits mailing list
> cfe-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
>
>
>
> --
> Thanks,
> -Jiangning
>
>
>
>
> --
> Thanks,
> -Jiangning
>
>
>

-- 
Thanks,
-Jiangning
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131205/fe20cdf9/attachment.html>