![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
(no subject)
May. 22nd, 2006 01:50 pmmutter mutter fucking compilers, mutter mutter, f77 vs f90 mutter mutter, faster computers mutter mutter...
So, I have a spectral line generation code.
It is very slow at the currently resolution and for an accurate comparison to some other data, I need to shift to running it on a faster machine. No problem. Shift everything over and recompile. HA! Different sources for where it goes when told to look for f77, which end up using it through f90 anyway. One lets the f77 specific command work, the other doesn't. Guess which one.
The problem is the calculation precision (as far as I can tell with my erratic computer knowledge) required. The spectral line generation program needs
-xtypemap=real:64,double:128,integer:mixed
for the data mapping - ie apparently "8-byte data but only 4-byte arithmetic, and is only available with f77." from here under -xtypemap ie way the fuck down the page.
It appears to be the integer:mixed option which is fucking the entire thing up. Unfortunately the -xtypemap bit appears rather essential with a complete removal given a scree of NaN for a spectrum. Not optimal.
Have looked at g77, but that doesn't like it anyway, even with
g77 -ff77 -xtypemap=real:64,double:128,integer:mixed -O3 ...f ...f ...f
to explicitly warn that it is dealing with f77 specfic language.
Googling with key phrases etc has helped a little but not quite enough as all my fortran/programing is self taught and very erratic. I just don't have the right vocab to ask the right questions (I think).
Any way, any suggestions other than ripping hair out, bouncing the computer out the window and running up and down the corridor laughing manically will be greatly appreciated.
I need lunch.
So, I have a spectral line generation code.
It is very slow at the currently resolution and for an accurate comparison to some other data, I need to shift to running it on a faster machine. No problem. Shift everything over and recompile. HA! Different sources for where it goes when told to look for f77, which end up using it through f90 anyway. One lets the f77 specific command work, the other doesn't. Guess which one.
The problem is the calculation precision (as far as I can tell with my erratic computer knowledge) required. The spectral line generation program needs
-xtypemap=real:64,double:128,integer:mixed
for the data mapping - ie apparently "8-byte data but only 4-byte arithmetic, and is only available with f77." from here under -xtypemap ie way the fuck down the page.
It appears to be the integer:mixed option which is fucking the entire thing up. Unfortunately the -xtypemap bit appears rather essential with a complete removal given a scree of NaN for a spectrum. Not optimal.
Have looked at g77, but that doesn't like it anyway, even with
g77 -ff77 -xtypemap=real:64,double:128,integer:mixed -O3 ...f ...f ...f
to explicitly warn that it is dealing with f77 specfic language.
Googling with key phrases etc has helped a little but not quite enough as all my fortran/programing is self taught and very erratic. I just don't have the right vocab to ask the right questions (I think).
Any way, any suggestions other than ripping hair out, bouncing the computer out the window and running up and down the corridor laughing manically will be greatly appreciated.
I need lunch.
