[PATCH] Fix ldbl128ibm fmodl for subnormals

[Adhemerval Zanella]

On 06/05/2012 01:02 PM, Andreas Schwab wrote:
> Adhemerval Zanella <azanella@linux.vnet.ibm.com> writes:
>
>> Current subnormal tests for fmod (TEST_ff_f (fmod, 0x0.fffffffffffffp-1022L, 0x1p-1074L, plus_zero)) 
>> fails for ldbl-128ibm. This patch fixes it and it is related to commit 'Fix fmod for subnormals' 
>> (c5bfe3d5ba29d36563f1e4bd4f8d7336093ee6fc). 
>>
>> Tested on ppc32 and ppc64.
> How did you test it?  I get an infinite loop here:
>
> 	while(hx<0x0001000000000000LL) {	/* normalize x */
> 	    hx = hx+hx+(lx>>63); lx = lx+lx;
> 	    iy -= 1;
> 	}
You are correct and I have commit a non-intended older version of the patch that
contains the mistakes lines:

-       int64_t n,hx,hy,hz,ix,iy,sx,i;
-       u_int64_t lx,ly,lz;
+       int64_t n,hx,hy,hz,ix,iy,sx;
+       u_int64_t lx,ly,lz, i;

In my correct patch 'i' was not supposed to unsigned in any way. The patch I was intended to
does not contain the above modifications. How should I proceed in this case?


-- 
Adhemerval Zanella Netto
  Software Engineer
  Linux Technology Center Brazil
  Toolchain / GLIBC on Power Architecture
  azanella@linux.vnet.ibm.com / azanella@br.ibm.com
  +55 61 8642-9890