SSE intrinsics

The current release of sse/sse2 intrinsics consists of only packed intregral instructions via the use of header sunmedia_intrin.h, located in

.../[intel-S2|intel-Linux]/prod/include/cc/sys

We have implemented a fully icc/gcc compatible set of intrinsics, with compatible header files of mmintrin.h, xmmintrin.h, emmintrin and pmmintrin.h, which will be made available to our compilers in the very near future. These will include all packed integral, floating point, cache and sse3 intrinsics.

For example:

#include <emmintrin.h>

__m128d dm1;

double dp1, dp2;

double *dp3;

void foo()

{

__m128d d1, d2;

d1 = _mm_set_pd(dp1, dp2);

d2 = _mm_load_pd(dp3);

dm1 = _mm_add_pd(d1, d2);

}

A possible generated code will then be:

foo:

movsddp1,%xmm0

unpcklpd dp2,%xmm0

movl dp3,%eax

movapd(%eax),%xmm1

addpd%xmm1,%xmm0

movapd%xmm0,dm1

ret

Alfred Huang

alfredhuang at 2007-7-6 21:28:22 >