最近实现一套纯软件的Power PC speicified intrinsics,包括AltiVec的一些东东,后来发现在x86下执行很慢。于是开始使用SSE/SSE2加速,不过介于PPC和x86指令集巨大的差异,有很多东西需要自己实现,于是有了如下的代码:
#ifdef _SSE_OPTIMIZE_
inline __m128 _mm_ceil_ps( __m128 v )
{
static const __m128 twoTo23 = _mm_set_ps1(1 << 23);
__m128 d = _mm_sub_ps( _mm_add_ps( _mm_add_ps( _mm_sub_ps( v, twoTo23 ), twoTo23 ), twoTo23 ), twoTo23 ); //the meat of ceil
__m128 largeMaskE = _mm_cmpgt_ps( v, twoTo23 ); //-1 if v >= 2**23
__m128 g = _mm_cmpgt_ps( v, d ); //check for possible off by one error
__m128 h = _mm_cvtepi32_ps( *(__m128i*) &g ); //convert positive check result to -1.0, negative to 0.0
__m128 t = _mm_sub_ps( d, h ); //add in the error if there is one
//Select between output result and input value based on v >= 2**23
v = _mm_and_ps( v, largeMaskE );
t = _mm_andnot_ps( largeMaskE, t );
return _mm_or_ps( t, v );
}
#endif