[PATCH] D22397: MIRParser: Rewrite register info initialization; mostly NFC

[Matthias Braun via llvm-commits]

> On Jul 27, 2016, at 5:07 AM, Duncan P. N. Exon Smith via llvm-commits <llvm-commits at lists.llvm.org> wrote:
> 
> 
>> On 2016-Jul-14, at 23:17, Matthias Braun via llvm-commits <llvm-commits at lists.llvm.org> wrote:
>> 
>> This commit removes the unused feature of having nonsequential register
>> numbers. This was confusing anyway as the vreg numbers would be
>> different after parsing when you had "holes" in your numbering.
>> 
> 
> I don't want to hold up the patch (I'm on vacation and haven't taken the time to understand the change), but I want to point out that if there's a way to reproduce the state before serialization even when it's a bad state, or to catch the bad state when writing out IR, then it'll be easier to reproduce backend bugs.  (It's not clear from your description whether you're getting closer to or further from that goal.)

My first version of the patch brought us closer to this goal, it was rejected in the review so now we have an intermediate map again instead of explicitely stating the exact vreg numbers used.

While I first thought in similarly to this argument, I let my self convince by the review that we should also keep an eye on usabilty (see my response to the isSSA review for more about this).