This is the mail archive of the gcc-help@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: 64 bit fortran unformatted on Opteron -- incompatible with32unformatted


Andreas,

No x86 is not the whole world, but Sun, SGI, etc. all have files which use a 
32 bit record for this field.  I move back and forth between these universes 
all the time, and with byteswapping i/o it works quite nicely.  

It would seem to me that being incompatible with other systems is not a good 
place to be, especially when the reasons for doing so are not compelling. 
with a 100MB/sec transfer rate, a 4GB record will take 41 seconds to read, 
assuming there is no other contention for the I/O channel, and DMA is working 
flawlessly.  This also means that you've set aside 4GB of actual memory to 
hold this monster and have memory besides this to have your program in.  And 
when you read the 2nd 4BG record...

Are the Itaniums following this path?  I can see for them, since they are more 
a clean break with the 32 bit world, but even there,  being able to re-use 
grids that have taken in some cases months to create (yes, although a complex 
grid i s now more down to a matter of weeks) would be a real benefit.  
Especially when you find yourself in the situation of not being able to 
re-create the grid, or wanting to run exactly the same grid on the new 
platform.  

This was one of the failings of the Cray, with it's proprietary format, all 
the grids I've generated are now toast(which are not that complex but still 
represent bonehead-months of work)...

Mike


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]