[llvm] r312364 - [AMDGPU] Prevent infinite recursion in DAG.computeKnownBits()

[Mekhanoshin, Stanislav via llvm-commits]

Committed test in rL312388.

Stas

-----Original Message-----
From: Mekhanoshin, Stanislav 
Sent: Friday, September 01, 2017 2:45 PM
To: Davide Italiano <davide at freebsd.org>
Cc: llvm-commits <llvm-commits at lists.llvm.org>
Subject: RE: [llvm] r312364 - [AMDGPU] Prevent infinite recursion in DAG.computeKnownBits()

I have instead modelled this with small artificial function. I will submit testcase shortly.

Stas

-----Original Message-----
From: Mekhanoshin, Stanislav
Sent: Friday, September 01, 2017 2:36 PM
To: Davide Italiano <davide at freebsd.org>
Cc: llvm-commits <llvm-commits at lists.llvm.org>
Subject: Re: [llvm] r312364 - [AMDGPU] Prevent infinite recursion in DAG.computeKnownBits()

Note that actual behavior is hung llc, so automated tools are not a good help.

Stas



--- Original message ---
From: Davide Italiano <davide at freebsd.org>
Sent: September 1, 2017 2:21:23 PM
To: Mekhanoshin, Stanislav <Stanislav.Mekhanoshin at amd.com>
CC: llvm-commits <llvm-commits at lists.llvm.org>
Subject: Re: [llvm] r312364 - [AMDGPU] Prevent infinite recursion in DAG.computeKnownBits()

> On Fri, Sep 1, 2017 at 2:11 PM, Mekhanoshin, Stanislav 
> <Stanislav.Mekhanoshin at amd.com> wrote:
>> That's a little tricky. The actual testcase is huge. I'm trying to minimize it, but actually the mistake was trivial.
>>
>> Stas
>>
>
> This is not ideal. In years, we'll probably forget, and no test may 
> result in this regression being reintroduced. In particular because 
> the mistake was trivial, reintroducing it is trivial as well.
> I recommend following this:
> https://gcc.gnu.org/wiki/A_guide_to_testcase_reduction
> It takes days at times, but it should do it. (also, Mehdi wrote a 
> bugpoint equivalent a while ago).
>
> Thanks,
>
> --
> Davide