[llvm-dev] Improving SCEV's behavior around IR level no-wrap flags

Sanjoy Das via llvm-dev llvm-dev at lists.llvm.org
Fri Sep 23 19:26:52 PDT 2016

Hi Sebastian,

Sebastian Pop wrote:
 > On Fri, Sep 23, 2016 at 6:08 AM, Andrew Trick<atrick at apple.com>  wrote:
 >> The alternative is to represent proofs about IR values independent from SCEV.
 > I like this for two reasons:
 > - it simplifies the SCEV analysis and representation by moving other
 > analyses results on the side,
 > - and it will make the analyses more context sensitive: proofs
 > attached to an IR location, and not to the SCEV expressions.

Can you please take a look at my reply to Andy on this (last segment
of the previous email)?  Given that both you and Andy have suggested
this scheme, I now suspect I misunderstood something.

-- Sanjoy

More information about the llvm-dev mailing list