Division error with long double

[codinguncut]

#include <stdio.h>

int main()
{
  unsigned long ips = 4000000;
  double m_ips;
  m_ips = ((double)ips) / 1000000.0L;
  printf("m_ips %f\n", m_ips);
}

gcc division.c && ./a.out

m_ips 4.000000

emcc division.c -o division.js && node division.js

Warning: .ll contains floating-point values with more than 64 bits. Faking values for them. If they are used, this will almost certainly break horribly!
m_ips -0.000000

I am aware a warning is displayed in above code, but in multiple places in the documentation is it noted that this warning is not known to have any averse effects ;)

System

emscripten: git repo Tue Sep 11 19:28:48 2012 -0700
llvm: 3.1
node: v0.8.12
gcc: gcc (Ubuntu/Linaro 4.7.2-1ubuntu1) 4.7.2
uname -a
Linux Lorax 3.5.0-15-generic #22-Ubuntu SMP Wed Sep 19 20:06:19 UTC 2012 i686 i686 i686 GNU/Linux
Ubuntu quantal
cpu/system: Intel(R) Core(TM)2 Duo CPU     U9600  @ 1.60GHz

[kripken]

The compiler warns on floats larger than 64 bits - there is an x86_fp80 (80-bit x86 float) there, which is what the clang frontend does to long doubles. We can't support that.

The long-term fix is to make a clang frontend that does not have this problem, that is, that makes long double mean 64 bits.

[manuels]

I just tested this bug with the upstream branch.
Looks like it's fixed.

If you can confirm it, you can close it.

[kripken]

Yes, we switched meanwhile to an llvm target without long double issues, so this should indeed be fixed.