[fortran-dev, patch] Parsed format data caching

[Jerry DeLisle]

Hi folks,

This patch uses the simplest of hashing functions to hash a format string and 
save a pointer to the parsed format data and its tokenized fnode tree.

This saves repeated re-parsing of format strings for every instance of a 
formatted I/O operation, especially inside loops.

The method is probabilistic, meaning it relies on a good chance you won't 
clobber previously saved data.  If there is a collision, the previously saved 
data is discarded and replaced by the new.  The hash table size is fixed with 16 
entries.

On some test cases I have seen about 18% speed up in formatted I/O.

Regression tested on x86-64-linux-gnu.

OK for commit to fortran-dev branch?  If anyone is curious, the patch will apply 
on 4.4 and 4.5 with some fuzz.

Regards,

Jerry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: format-cache.diff
Type: text/x-patch
Size: 11342 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20090329/09969634/attachment.bin>