mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-25 19:50:38 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

improve Lab2RGB_b()

Browse Source
This commit is contained in:
k-shinotsuka 2016-09-18 14:48:40 +09:00
parent dc3d0cbc22
commit 1f8aa4f7d3
1 changed files with 25 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
modules/imgproc/src/color.cpp
View File

@ -5505,9 +5505,7 @@ struct Lab2RGB_b
v_alpha = vdup_n_u8(ColorChannel<uchar>::max());
v_128 = vdupq_n_f32(128.0f);
#elif CV_SSE2
v_scale_inv = _mm_set1_ps(100.f/255.f);
v_scale = _mm_set1_ps(255.f);
v_128 = _mm_set1_ps(128.0f);
v_zero = _mm_setzero_si128();
haveSIMD = checkHardwareSupport(CV_CPU_SSE2);
#endif
@ -5516,6 +5514,7 @@ struct Lab2RGB_b
#if CV_SSE2
// 16s x 8
void process(__m128i v_r, __m128i v_g, __m128i v_b,
__m128 v_coeffs, __m128 v_res,
float * buf) const
{
__m128 v_r0 = _mm_cvtepi32_ps(_mm_unpacklo_epi16(v_r, v_zero));
@ -5526,15 +5525,20 @@ struct Lab2RGB_b
__m128 v_g1 = _mm_cvtepi32_ps(_mm_unpackhi_epi16(v_g, v_zero));
__m128 v_b1 = _mm_cvtepi32_ps(_mm_unpackhi_epi16(v_b, v_zero));
v_r0 = _mm_mul_ps(v_r0, v_scale_inv);
v_r1 = _mm_mul_ps(v_r1, v_scale_inv);
v_r0 = _mm_sub_ps(_mm_mul_ps(v_r0, v_coeffs), v_res);
v_g1 = _mm_sub_ps(_mm_mul_ps(v_g1, v_coeffs), v_res);
v_g0 = _mm_sub_ps(v_g0, v_128);
v_g1 = _mm_sub_ps(v_g1, v_128);
v_b0 = _mm_sub_ps(v_b0, v_128);
v_b1 = _mm_sub_ps(v_b1, v_128);
v_coeffs = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(v_coeffs), 0x49));
v_res = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(v_res), 0x49));
_mm_interleave_ps(v_r0, v_r1, v_g0, v_g1, v_b0, v_b1);
v_r1 = _mm_sub_ps(_mm_mul_ps(v_r1, v_coeffs), v_res);
v_b0 = _mm_sub_ps(_mm_mul_ps(v_b0, v_coeffs), v_res);
v_coeffs = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(v_coeffs), 0x49));
v_res = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(v_res), 0x49));
v_g0 = _mm_sub_ps(_mm_mul_ps(v_g0, v_coeffs), v_res);
v_b1 = _mm_sub_ps(_mm_mul_ps(v_b1, v_coeffs), v_res);
_mm_store_ps(buf, v_r0);
_mm_store_ps(buf + 4, v_r1);
@ -5550,6 +5554,10 @@ struct Lab2RGB_b
int i, j, dcn = dstcn;
uchar alpha = ColorChannel<uchar>::max();
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
#if CV_SSE2
__m128 v_coeffs = _mm_set_ps(100.f/255.f, 1.f, 1.f, 100.f/255.f);
__m128 v_res = _mm_set_ps(0.f, 128.f, 128.f, 0.f);
#endif
for( i = 0; i < n; i += BLOCK_SIZE, src += BLOCK_SIZE*3 )
{
@ -5578,36 +5586,16 @@ struct Lab2RGB_b
#elif CV_SSE2
if (haveSIMD)
{
for ( ; j <= (dn - 32) * 3; j += 96)
for ( ; j <= (dn - 8) * 3; j += 24)
{
__m128i v_r0 = _mm_loadu_si128((__m128i const *)(src + j));
__m128i v_r1 = _mm_loadu_si128((__m128i const *)(src + j + 16));
__m128i v_g0 = _mm_loadu_si128((__m128i const *)(src + j + 32));
__m128i v_g1 = _mm_loadu_si128((__m128i const *)(src + j + 48));
__m128i v_b0 = _mm_loadu_si128((__m128i const *)(src + j + 64));
__m128i v_b1 = _mm_loadu_si128((__m128i const *)(src + j + 80));
__m128i v_src0 = _mm_loadu_si128((__m128i const *)(src + j));
__m128i v_src1 = _mm_loadl_epi64((__m128i const *)(src + j + 16));
_mm_deinterleave_epi8(v_r0, v_r1, v_g0, v_g1, v_b0, v_b1);
process(_mm_unpacklo_epi8(v_r0, v_zero),
_mm_unpacklo_epi8(v_g0, v_zero),
_mm_unpacklo_epi8(v_b0, v_zero),
process(_mm_unpacklo_epi8(v_src0, v_zero),
_mm_unpackhi_epi8(v_src0, v_zero),
_mm_unpacklo_epi8(v_src1, v_zero),
v_coeffs, v_res,
buf + j);
process(_mm_unpackhi_epi8(v_r0, v_zero),
_mm_unpackhi_epi8(v_g0, v_zero),
_mm_unpackhi_epi8(v_b0, v_zero),
buf + j + 24);
process(_mm_unpacklo_epi8(v_r1, v_zero),
_mm_unpacklo_epi8(v_g1, v_zero),
_mm_unpacklo_epi8(v_b1, v_zero),
buf + j + 48);
process(_mm_unpackhi_epi8(v_r1, v_zero),
_mm_unpackhi_epi8(v_g1, v_zero),
_mm_unpackhi_epi8(v_b1, v_zero),
buf + j + 72);
}
}
#endif
@ -5692,7 +5680,7 @@ struct Lab2RGB_b
float32x4_t v_scale, v_scale_inv, v_128;
uint8x8_t v_alpha;
#elif CV_SSE2
__m128 v_scale, v_scale_inv, v_128;
__m128 v_scale;
__m128i v_zero;
bool haveSIMD;
#endif