mirror of
https://github.com/llvm/llvm-project.git
synced 2025-05-19 01:16:06 +00:00
0714e3e245
I could not come up a way to test this -- I think this bug is latent today, and will not actually result in a miscompile. In `getPreStartForExtend`, SCEV constructs `PreStart` as a sum of all of `SA`'s operands except `Op`. It also uses `SA`'s no-wrap flags, and this is problematic because removing an element from an add expression can make it signed-wrap. E.g. if `SA` was `(127 + 1 + -1)`, then it could safely be `<nsw>` (since `sext(127) + sext(1) + sext(-1)` == `sext(127 + 1 + -1)`), but `(127 + 1)` (== `PreStart` if `Op` is `-1`) is not `<nsw>`. Transferring `<nuw>` from `SA` to `PreStart` is safe, as far as I can tell. llvm-svn: 251097
Analysis Opportunities:
//===---------------------------------------------------------------------===//
In test/Transforms/LoopStrengthReduce/quadradic-exit-value.ll, the
ScalarEvolution expression for %r is this:
{1,+,3,+,2}<loop>
Outside the loop, this could be evaluated simply as (%n * %n), however
ScalarEvolution currently evaluates it as
(-2 + (2 * (trunc i65 (((zext i64 (-2 + %n) to i65) * (zext i64 (-1 + %n) to i65)) /u 2) to i64)) + (3 * %n))
In addition to being much more complicated, it involves i65 arithmetic,
which is very inefficient when expanded into code.
//===---------------------------------------------------------------------===//
In formatValue in test/CodeGen/X86/lsr-delayed-fold.ll,
ScalarEvolution is forming this expression:
((trunc i64 (-1 * %arg5) to i32) + (trunc i64 %arg5 to i32) + (-1 * (trunc i64 undef to i32)))
This could be folded to
(-1 * (trunc i64 undef to i32))
//===---------------------------------------------------------------------===//