[llvm-dev] Alive now available online

Sun Jan 8 11:26:48 PST 2017

It's very on topic, certainly. We've ported more than 300 InstCombine 
patterns to Alive ourselves.
Right now the C++ code generation is not great. Alive produces a linear 
search through the patterns, similar to how InstCombine works today.
It's on my todo list to improve this part of Alive; I just didn't have time 
to do it yet.  If this year I manage to get an intern with experience on 
graph rewriting, we'll get it done (or try, at least)! :)

Nuno

-----Original Message-----
From: Bryant Wong
Sent: Friday, January 6, 2017 7:29 PM
To: Sanjay Patel ; llvm-dev at lists.llvm.org
Cc: Nuno Lopes ; John Regehr
Subject: Re: [llvm-dev] Alive now available online

Not sure how off-topic this is, but should we consider/have we considered 
porting our InstCombines to Alive? The PLDI '15 paper even demos C++ 
extraction from Alive theorems. I think it'd be a small step from that to 
extracting tightly optimized VM code, not unlike what Tablegen emits. 
Everything would be so clean and readable and organized. And edge cases can 
still be handled manually, like ISel.

On Fri, Jan 6, 2017 at 9:38 AM, Sanjay Patel <spatel at rotateright.com> wrote:

On Fri, Jan 6, 2017 at 2:30 AM, Nuno Lopes <nuno.lopes at ist.utl.pt> wrote:

Hi Sanjay,

You used Alive correctly, of course :)

At this moment we cannot give you the best precondition. It’s on the todo 
list, but it’s not even started yet. It’s a much harder problem to solve. 
We do have a mode to compute the best set of nsw/nuw/exact attributes in the 
transformed expression, but it’s not enabled on the web interface yet 
(InstCombine was missing quite a few cases last time I checked).

I played a bit with your example and realized that it doesn’t even need a 
precondition :)  Check this out: http://rise4fun.com/Alive/j

(btw, > is signed comparison, and u> is unsigned).

Nice! I'll draft a patch for the transform.

Nuno

From: Sanjay Patel [mailto:spatel at rotateright.com]
Sent: 6 de janeiro de 2017 00:35
To: Nuno Lopes <nuno.lopes at ist.utl.pt>
Cc: llvm-dev <llvm-dev at lists.llvm.org>; John Regehr <regehr at cs.utah.edu>; 
3.14472+reviews.llvm.org at gmail.com
Subject: Re: [llvm-dev] Alive now available online

Hi Nuno,

This is great. I just stumbled onto a problem that's similar to the one that 
Bryant solved in:
https://reviews.llvm.org/rL285729

define i1 @foo(i32 %x) {
  %shl = shl nsw i32 %x, 4
  %cmp = icmp sgt i32 %shl, 1
  ret i1 %cmp
}

(For more background, see PR30773 - 
https://llvm.org/bugs/show_bug.cgi?id=30773 )

So I'm staring at that wondering why instcombine can't see that it's really 
just:
define i1 @foo(i32 %x) {
  %cmp = icmp sgt i32 %x, 0
  ret i1 %cmp
}

As a first hack, I did this:
http://rise4fun.com/Alive/qG

Name: sgt
Pre: C0 > 0
%a = shl nsw i8 %x, C1
%b = icmp sgt %a, C0
  =>
%b = icmp sgt %x, (C0 >> C1)

...and success! Assuming I used Alive correctly. :)

But is there a way to tell if I've chosen the most liberal constraint? Ie, 
is C0 > 0 the best pre-condition?

On Thu, Jan 5, 2017 at 3:42 PM, Nuno Lopes via llvm-dev 
<llvm-dev at lists.llvm.org> wrote:

Hi,

Just a short email to announce that Alive is now available online: 
http://rise4fun.com/Alive

The site includes a few examples (both correct and buggy). You can also 
create a "permalink" to send the proof to someone else.

The execution time is limited to 30 seconds for now.  You may want to 
constrain the operand's types if the tool times out, for example.

The service is still in tests.  Please let me know if you run into problems 
and/or if you have comments or feature requests.

Nuno

P.S.: The semantics implemented in this release is the one I've presented at 
the last LLVM dev meeting.  I've kept undef for now, though, but there's a 
poison value.  There's (very) limited support for branches as well.
_______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev 