externals/zexy/src sgn~.c,1.1,1.2 - Pd-cvs

21 Dec 2005


      Update of /cvsroot/pure-data/externals/zexy/src
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv11638
Modified Files:
    sgn~.c 
Log Message:
only do 8x loop unrolling since it seems to be more efficient
(well, rather a joke: 4096 [abs~] need about 24% with 16x unrolling and about 23-24% with 8x unrolling)
Index: sgn~.c
===================================================================
RCS file: /cvsroot/pure-data/externals/zexy/src/sgn~.c,v
retrieving revision 1.1
retrieving revision 1.2
diff -C2 -d -r1.1 -r1.2
*** sgn~.c	21 Dec 2005 17:05:47 -0000	1.1
--- sgn~.c	21 Dec 2005 17:15:41 -0000	1.2
***************
*** 80,84 ****
__m128 val;
!   int n = (int)(w[3])>>4; // yea, we do 16x loop-unrolling
const __m128 sgnmask= _mm_loadu_ps((float*)l_bitmask);
--- 80,84 ----
__m128 val;
!   int n = (int)(w[3])>>3; // we do 8x loop-unrolling
const __m128 sgnmask= _mm_loadu_ps((float*)l_bitmask);
***************
*** 102,119 ****
      out[1]= _mm_or_ps    (xmm1, xmm0);
!     val=in[2];
!     xmm0  = _mm_cmpneq_ps(val , zero);
!     xmm1  = _mm_and_ps   (val, sgnmask);
!     xmm0  = _mm_and_ps   (xmm0, one);
!     out[2]= _mm_or_ps    (xmm1, xmm0);
! 
!     val=in[3];
!     xmm0  = _mm_cmpneq_ps(val , zero);
!     xmm1  = _mm_and_ps   (val, sgnmask);
!     xmm0  = _mm_and_ps   (xmm0, one);
!     out[3]= _mm_or_ps    (xmm1, xmm0);
! 
!     in +=4;
!     out+=4;
}
--- 102,107 ----
      out[1]= _mm_or_ps    (xmm1, xmm0);
!     in +=2;
!     out+=2;
}