[CELT-dev] CELT grabbing 100KB of memory right off the top

Andrew Lentvorski bsder at allcaps.org
Mon Apr 18 22:34:35 PDT 2011

I'd like to make a couple of suggestions, but I really don't want to 
make this into a pile on.  CELT was *WAY* easier to port than Tremor 
was.  Please remember that--I'm very grateful that it was as 
straightforward as it is.  Tremor was a *NIGHTMARE* to port.

The one thing that I *really* didn't like was the #include  of 
"static_modes_fixed.c".  "#include"-ing a C file caused MPLAB all manner 
of grief.  Given that it's being generated it anyway, a 
"static_modes_fixed.h" file should be generated and then 
"static_modes_fixed.c" should be treated like a normal .c file.  I 
actually had to split this up this way by hand in order to work around 
the issues.

On 4/18/11 9:54 PM, Gregory Maxwell wrote:
> I don't believe anyone was trying to suggest using configure.

To be honest, does this project really need configure?  Does it *really* 
buy anything?  A straightforward Makefile with a "config.h" that 
actually has all of the relevant switches in it is probably more than 

I realize that nothing is going to make everybody 100% happy.  So, I'm 
not going to be terribly vocal about this.  However, I did have to go 
move this to a system with a working "configure" chain, build it once, 
capture all the files and linker flags, create a simple build script, 
rip out all the configure garbage, and then move it to the embedded target.

In addition, I'd probably like for a compile to at least bark a 
"#warning" like it does when it lacks lrint() for things like ilog.  The 
fact that I could pick up 10% because the compiler couldn't figure out 
how to get at my native clz instruction is kind of a big deal.


More information about the celt-dev mailing list