Geek asks and answers

Symbolic Vector Access

I'm new to SIMD programming, and I have some basic questions that I seem to be unable to understand after studying the topic for several days.

the code I want to optimize is basically a simple but big arithmetic formula, it should be pretty simple to parse coude automatically to calculate independent multiplications / additions in parallel, but I read that autovectorization only works for loops.

I read several times when access to individual elements in a vector through a connection or in some other way should be avoided at all costs, instead it should be replaced with _mm_shuffle_pd (I only work with doubles) ...

I know this is a very simple question, but I don't seem to understand how I can store the contents of the __m128d vector as doubles without referring to it as a union. also, does such an operation perform the performance of any gain compared to the scalar code?

union {
  __m128d v;
  double d[2];
} vec;
union {
  __m128d v;
double d[2];
} vec2;

vec.v = index1;
vec2.v = index2;
temp1 = _mm_mul_pd(temp1, _mm_set_pd(bvec[vec.d[1]], bvec[vec2[1]]));

also, the two unions look ridiculously ugly, but when used

union dvec {
  __m128d v;
  double d[2];
} vec;

trying to declare indexX as dvec, the compiler complained that dvec was not declared

+5
source share
2 answers

Unfortunately, if you look at MSDN, it will say the following:

You should not access the __m128d fields directly. However, you can see these types in the debugger. A variable of type __m128 is mapped to XMM registers [0-7].

SIMD, , , , , .

EDIT:

, :

__m128, __m128d __m128i , . , "+" " → ".

:

__m128, __m128d __m128i , (, float) .

, , , . , MSDN.

EDIT2:

, SIMD-

:

#include <math.h>
#include <emmintrin.h>
double in1_min(__m128d x)
{
    return x[0];
}

gcc 4.6 . gcc . , .

+5

_mm_cvtsd_f64 (__m128d a), "emmintrin.h", sse .

Intel Intrinsics:

  • double _mm_cvtsd_f64 (__m128d a)
  • "emmintrin.h"
  • : movsd
  • CPUID: SSE2

: (64-) a dst.

dst [63: 0]: = a [63: 0]

0
source

More articles:


All Articles