Method scaleValues

src/librawspeed/common/RawImageDataFloat.cpp:148–257

#if 0 // def WITH_SSE2

  void RawImageDataFloat::scaleValues(int start_y, int end_y) {
    bool WITH_SSE2;
#ifdef _MSC_VER
    int info[4];
    __cpuid(info, 1);
    WITH_SSE2 = !!(info[3]&(1 << 26));
#else
    WITH_SSE2 = true;
#endif

    float app_scale = 65535.0F / (whitePoint - blackLevelSeparate[0]);
    // Check SSE2
    if (WITH_SSE2 && app_scale < 63) {

      __m128i sseround;
      __m128i ssesub2;
      __m128i ssesign;
      auto* sub_mul = alignedMallocArray<uint32_t, 16, __m128i>(4);
	  if (!sub_mul)
		ThrowRDE("Out of memory, failed to allocate 128 bytes");

      uint32_t gw = pitch / 16;
      // 10 bit fraction
      uint32_t mul = (int)(1024.0F * 65535.0F / (float)(whitePoint - blackLevelSeparate[mOffset.x&1]));
      mul |= ((int)(1024.0F * 65535.0F / (float)(whitePoint - blackLevelSeparate[(mOffset.x+1)&1])))<<16;
      uint32_t b = blackLevelSeparate[mOffset.x&1] | (blackLevelSeparate[(mOffset.x+1)&1]<<16);

      for (int i = 0; i< 4; i++) {
        sub_mul[i] = b;     // Subtract even lines
        sub_mul[4+i] = mul;   // Multiply even lines
      }

      mul = (int)(1024.0F * 65535.0F / (float)(whitePoint - blackLevelSeparate[2+(mOffset.x&1)]));
      mul |= ((int)(1024.0F * 65535.0F / (float)(whitePoint - blackLevelSeparate[2+((mOffset.x+1)&1)])))<<16;
      b = blackLevelSeparate[2+(mOffset.x&1)] | (blackLevelSeparate[2+((mOffset.x+1)&1)]<<16);

      for (int i = 0; i< 4; i++) {
        sub_mul[8+i] = b;   // Subtract odd lines
        sub_mul[12+i] = mul;  // Multiply odd lines
      }

      sseround = _mm_set_epi32(512, 512, 512, 512);
      ssesub2 = _mm_set_epi32(32768, 32768, 32768, 32768);
      ssesign = _mm_set_epi32(0x80008000, 0x80008000, 0x80008000, 0x80008000);

      for (int y = start_y; y < end_y; y++) {
        __m128i* pixel = (__m128i*) & data[(mOffset.y+y)*pitch];
        __m128i ssescale, ssesub;
        if (((y+mOffset.y)&1) == 0) {
          ssesub = _mm_load_si128((__m128i*)&sub_mul[0]);
          ssescale = _mm_load_si128((__m128i*)&sub_mul[4]);
        } else {
          ssesub = _mm_load_si128((__m128i*)&sub_mul[8]);
          ssescale = _mm_load_si128((__m128i*)&sub_mul[12]);
        }

        for (uint32_t x = 0 ; x < gw; x++) {
          __m128i pix_high;