Just a snippet to swap red and blue in a BGRA buffer, to get a RGBA buffer (useful when loading TGA).

gistfile1.cpp

#ifdef __SSSE3__
#  include <xmmintrin.h>
#  include <tmmintrin.h>
#endif

posix_memalign((void**)&pixels, 16, sizeof(unsigned int)*PIXEL_COUNT);


/*
 * Mind this will only work for width * height % 16 == 0 textures
 */
void swap_pixels(unsigned int* pixels)
{
#ifdef __SSSE3__
  // Shuffling mask (RGBA -> BGRA) x 4, in reverse byte order
  static const __m128i m = _mm_set_epi8(15,12,13,14,11,8,9,10,7,4,5,6,3,0,1,2);
  // Assert pixels will NOT be aliased here
  __m128i* __restrict__ pix = (__m128i*)pixels;
  // Tile the LHS to match 64B cache line size
  for (; (unsigned int*)pix<pixels+PIXEL_COUNT; pix+=4)
  {
    __m128i p1 = _mm_load_si128(pix);
    __m128i p2 = _mm_load_si128(pix+1);
    __m128i p3 = _mm_load_si128(pix+2);
    __m128i p4 = _mm_load_si128(pix+3);

    p1 = _mm_shuffle_epi8(p1, m);
    p2 = _mm_shuffle_epi8(p2, m);
    p3 = _mm_shuffle_epi8(p3, m);
    p4 = _mm_shuffle_epi8(p4, m);

    _mm_store_si128(pix,   p1);
    _mm_store_si128(pix+1, p2);
    _mm_store_si128(pix+2, p3);
    _mm_store_si128(pix+3, p4);
  }
#else
  for (int i=0; i<PIXEL_COUNT; ++i)
  {
    char* p = (char*)(pixels+i);
    char tmp = p[0];
    p[0] = p[2];
    p[2] = tmp;
  }
#endif
}

free(pixels);

compiler error : static const __m128i m = _mm_set_epi8(15,12,13,14,11,8,9,10,7,4,5,6,3,0,1,2); <--- not const