Re: seeking embeddable statement compiler code

If you are looking for an efficient code for some mathematical
expressions, you must worry about the classical problem of numerical
precision. Do not forget that floating point numbers are not real
numbers, so you would make some tradeoff between the computation
efficiency and the result correctness.


For instance, mathematically, ((a*b)*c)*d=(a*b)*(c*d)
The code of left expression computes 3 mult in serial. The code generated
for the right hand side is more efficient since you can use the
fine-grain parallelism between two multiplications (if you have two
available multipliers). However, the result could be slightly different
because of floating rounding treatments. The error cumul can give non
expected results.


yours,
SAAT

Return to the comp.compilers page.
Search the comp.compilers archives again.