On 27-Sep-2000 Nicolas Pitre wrote:
On Tue, 26 Sep 2000, Sean 'Shaleh' Perry wrote:
gcc -Wall -m486 -g -O -DHAVE_CONFIG_H -DFPM_INTEL -c -o fixed.o fixed.c
Even with --enable-speed, -O is the most used for Intels. Why is this?
Because the C code is already carefully written with the operation flow for the best possible performances (loops were unrolled by hand only where it should be, etc). When -O2 (or hier) is used, gcc tries to be smarter than it should and actually produces worse output than with -O. This isn't generally true for many programs but it is for this particular application.
that makes sense, the use of m486 however actually causes worse performance on Pentium II's and higher I was told. I this true? perhaps the -m<arch> should be left off? Or at least made optional.