[PATCH] D101844: [MicroBenchmarks] Add initial loop vectorization benchmarks.

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri added inline comments.


================
Comment at: MicroBenchmarks/LoopVectorization/MathFunctions.cpp:54-55
+      // Treat nans/infs as matching, if both versions produce nan/inf.
+      if (C[i] != CNovec[i] && !(isnan(C[i]) && isnan(CNovec[i])) &&
+          !(isinf(C[i]) && isinf(CNovec[i]))) {
+        std::cerr << "ERROR: autovec result different to scalar result " << C[i]
----------------
I'm not confident this is
1. consistent with the rules
2. will withstand the whatever compiler optimization levels test-suite is compiled with

Perhaps you want to be on the safe side and do `fpclassify()`,
fail if they mismatch, and compare value-equality if they are normal/subnormal?


Repository:
  rT test-suite

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101844/new/

https://reviews.llvm.org/D101844