Error in fused multiply add

John Pryce j.d.pryce@ntlworld.com
Fri Jul 4 07:00:00 GMT 2008


Can someone point me to where I should report this bug?

With GCC on my Mac, the FMA function in math.h does not behave according 
to the C standard when handling infinite input, as far as I can see. For 
instance the statement
  printf("\n  fma(PosInf,PosInf,0) = %g\n", fma(PosInf,PosInf,0)); // (*)

where PosInf holds +oo (positive infinity), produced the output
  fma(BiasPosInf,BiasPosInf,0) = nan

The C standard 7.12.13 says "The fma functions [ fma(x,y,z) ] compute 
(x*y)+z, rounded as one ternary operation: they compute the value (as if) to 
infinite precision and round once to the result format, according to the 
rounding mode characterized by the value of FLT_ROUNDS. A range error 
may occur."

The statement (*) is doing (+oo * +oo) + 0 which should be +oo, not NaN, 
according to my reading of the above. The new IEEE standard P754 is pretty 
clear that this should be so.

Context: I want to use the C's FMA function in the corresponding function in 
the Profil/BIAS Interval Arithmetic package, so correct operation in all 
circumstances is crucial.

Platform:
  Model Identifier:	MacBookPro2,2
  Processor Name:	Intel Core 2 Duo
  OS: Mac OS X 10.4.11 Tiger
  Compiler: i686-apple-darwin8-gcc-4.0.1 (GCC) 4.0.1

John Pryce

-----------------------------------------
Email sent from www.virginmedia.com/email
Virus-checked using McAfee(R) Software and scanned for spam



More information about the Libc-help mailing list