alt Hacker NewsThat's very much an issue with SIMD, especially where floating point numbers are concerned.
Matt Godbolt wrote about it recently.
https://xania.org/202512/21-vectorising-floats
TLDR, math notation and language specify particular orders in which floating point operations happen, and precision limits of IEEE float representation mean those have to be honoured by default.
Allowing compilers to reorder things in breach of that contract is an option, but it comes with risks.
Replies
Fortran requires compilers to “honor the integrity of parentheses” but otherwise doesn’t restrict compilers from rearranging expressions. Want a specific order of operations and rounding? Use parentheses to force them. This is why you’ll sometimes see parens around operations that already have arithmetic precedence, like `(x times x)-(y times y)`, to prevent the use of FMA for one of the multiplications but not the other.
I like that Zig allows using relaxed floating point rules with per block granularity to reduce the risk of breaking something else where IEEE compliance does matter. I think OpenMP simd pragmas can be used similarly for C/C++, but that's non-standard.