Hi all,

With the fixing of a bug this weekend, the LLVM backend is now at 
the same level of completion as the C backend. This includes passing
all tests, even when the library is compiled via LLVM.

One thing I had to do was increase the heap size of the ddc binary
when run under war. The increase was from 30M to 40M.

The problem seems to be that the LLVM text format seems to have
a bigger in memory footprint than the C text format. For instance,
the C backend makes extensive use of C macros while the LLVM backend
generates inline code.

It may be possible to reduce the memory footprint by converting
each function from an LLVM AST to the textual representation for
each function as it is completed. However, I think the gains from
do this will be smaller than the gains we will get when the
Data.Text module moves from UTF16 to UTF8 [0].

Erik

[0] http://jaspervdj.be/files/text.html
-- 
----------------------------------------------------------------------
Erik de Castro Lopo
http://www.mega-nerd.com/

-- 
Disciple-Cafe mailing list
http://groups.google.com/group/disciple-cafe

Reply via email to