[PATCH] D22011: [SystemZ] Generate fewer instructions for (sub <constant>, x)

[Ulrich Weigand via llvm-commits]

uweigand added a comment.

So I see that we get less instructions in this particular case.  However, I'm wondering whether execution is also *faster*, given that the new code seems to create longer dependency chains.  Normally I'd like to compare benchmark runs before making a change like that ...

Also, I'm wondering why we can't have the best of both worlds: just two instructions *and* the shorter dependency chain, by creating instead something like:

  lghi %r0, 63
  sgrk %r2, %r0, %r2

(assuming we're on z196 and higher.  But then I don't think z10 is a big concern these days anyway.)


http://reviews.llvm.org/D22011