Hi all
First post, first question

We´re running on Unix Tru64. We are maintaining a large amount of C-ISAM files, and some of them have a size > 10 GB. So we´re currently using C to perform some of our biggest file conversions. Now I´d like to investigate the possibility of performing these conversions using assembler language instead.

Does any of you have any ideas or experience about using assembler for file conversion? Any remarks are highly appreciated.

HTH
Yann

generally file conversion in ASM is same as in C.

You should profile your C application, find which part is taking most time, and rewrite it in assembly optimized for your specific purpose.

And it takes 100 % CPU usage? The problem with files is that disks are not fast enough to get the CPU permanently busy, so if your current C program doesn't takes all the CPU time then I think it doesn't worth the efforts of writing an ASM version.

Thanks for the response guys.
Yes, the conversions are usually taking 100% of the CPU. The C programs used for this purpose are fully designed and optimized so I could probably start to identify what parts are taking most time, as you suggest.

I´m not sure one can easily translate all CISAM specific commands to assembler, unless this can be found on the internet or if you´re good at analysing that kind of thing. I´ll probably do some search first.

Any idea about the size of such a small project? Do you think such competence is beyond the average Assembler coder´s capacities?

HTH
Yann

we don't know what exactly you do.

Do you call some 3rd party libraries to parse data ("CISAM specific commands")? If yes, then probably THESE are the speed critical parts.

If entire transformation is just your code, what do you mean by "Cisam specific commands"?

We are using the isam.h library, which is heavily used for all programs using CISAM files, containing all commands to create/amend/delete a record in such files (property of Informix). Don't know if this answers your question though.

then i believe the part that eats most time is this library. You would probably have to rewrite parts of this library to speed things up