transfer /opt/local to another machine
René J.V. Bertin
rjvbertin at gmail.com
Mon Apr 13 04:34:10 PDT 2015
On Monday April 13 2015 11:30:49 Chris Jones wrote:
>which is not a nice situation for debugging etc. I am also willing to
You've got a point - though I have a hunch that forcing -O3 instead of -O2 or -Os could have more adverse effects.
>bet the number of ports that would significantly benefit from compiling
>with higher instruction set support, AVX etc., is a very very select
>few. For most there simply will not be much in the way of an appreciable
>gain to the end user.
That's a good question, and will depend on to what extent the compiler uses an appropriate instruction set and does auto-vectorisation with the standard arguments. I'm aware I have a habit of investing some effort to scrape out the maximum performance obtainable with little effort (-mdynamic-no-pic is a favourite in that domain) because for a long time I worked in a context where every little gain counted one way or another.
This is OT, but I will never forget the instant I discovered that gcc had made auto-vectorisation default (with O3?) and its implementation was capable enough to compile the generic/scalar version of a YUV conversion function to code that was TEN times faster than a hand-coded SSE2 version (but only in 64bit; in 32bit the picture was the opposite though not by that big a factor IIRC).
All that to say that even code where you don't expect it could benefit. And I'm not really thinking of a few percents on an otherwise fast CPU; those same few percent on an otherwise slow(er) CPU can make much more of a real-world difference, IMHO.
More information about the macports-users