If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.
Enables or disables the reassociation of floating-point arithmetic. The default value is --fp_reassoc=on. Because floating-point values are of limited precision, and because floating-point operations round, floating-point arithmetic is neither associative nor distributive. For instance, (1 + 3e100) - 3e100 is not equal to 1 + (3e100 - 3e100). If strictly following IEEE 754, the compiler cannot, in general, reassociate floating-point operations. Using --fp_reassoc=on allows the compiler to perform the algebraic reassociation, at the cost of a small amount of precision for some operations. When --fp_reassoc=on, RPT MACF32 instructions may be generated. Because the RPT MACF32 instruction computes two partial sums and adds them together afterward to compute the entire accumulation, the result can vary in precision from a serial floating-point multiply accumulate loop.
