[PATCH] D26602: [DAGCombiner] do not fold (fmul (fadd X, 1), Y) -> (fmad X, Y, Y) by default

[Nicolai Hähnle via Phabricator via llvm-commits]

nhaehnle added a comment.

In https://reviews.llvm.org/D26602#606673, @spatel wrote:

> 3. There's an unanswered question of whether we can do this with FPOpFusion::Fast with TargetOptions::NoInfsFPMath. That depends on whether x * y + x is always more precise than the original x * (y + 1).


Here's an example with binary floating point numbers with two bits of mantissa.

  x = 1.01
  y = 111
  
  x * (y + 1) = 1.01 * 1000 = 1010 (this is the exact result; no rounding occurs at any step)
  
  x * y + x = 1000.11 + 1.01 =r 1000 + 1.01 = 1001.01 =r 1000 (with rounding towards zero)

The example relies on rounding towards zero at least in the second step, but it does seem to disqualify the transform at least for FMAD (i.e. with an intermediate rounding step). I suspect FMA (without intermediate rounding) is fine, but I have no proof.


https://reviews.llvm.org/D26602