[PATCH] D48283: [SCEV] Properly solve quadratic equations

[Krzysztof Parzyszek via Phabricator via llvm-commits]

kparzysz marked 5 inline comments as done.
kparzysz added inline comments.


================
Comment at: lib/Analysis/ScalarEvolution.cpp:8285
+    // In i1, 1 is the same as -1, so sign-extending (i1 1) would actually
+    // produce -1 in the larger type.
+    N = N.zext(NewWidth);
----------------
efriedma wrote:
> The algorithm should still work even if you treat "i1 true" as -1, I think.  Actually, you end up solving the exact same equation, because of the `A.isNegative()` check.
You're right.  The check for `A.isNegative()` was added later.


================
Comment at: lib/Analysis/ScalarEvolution.cpp:10641
+          return R;
         return SE.getCouldNotCompute(); // Something strange happened
     }
----------------
efriedma wrote:
> Please make sure we have a testcase that hits this "getCouldNotCompute".
This shouldn't really happen.  I've changed it to an assert.


Repository:
  rL LLVM

https://reviews.llvm.org/D48283