generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

cvtColor Luv 2 RGB u8

Browse Source
This commit is contained in:
Ilya Lavrenov
2015-01-12 10:59:30 +03:00
parent e20613a776
commit a340ea872e
2 changed files with 262 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

109
modules/imgproc/src/color.cpp
View File

@@ -4376,7 +4376,7 @@ struct HSV2RGB_b
{
int i, j, dcn = dstcn;
uchar alpha = ColorChannel<uchar>::max();
CV_DECL_ALIGNED(16) float buf[3*BLOCK_SIZE];
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
for( i = 0; i < n; i += BLOCK_SIZE, src += BLOCK_SIZE*3 )
{
@@ -4622,7 +4622,7 @@ struct RGB2HLS_b
void operator()(const uchar* src, uchar* dst, int n) const
{
int i, j, scn = srccn;
CV_DECL_ALIGNED(16) float buf[3*BLOCK_SIZE];
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
for( i = 0; i < n; i += BLOCK_SIZE, dst += BLOCK_SIZE*3 )
{
@@ -4876,7 +4876,7 @@ struct HLS2RGB_b
{
int i, j, dcn = dstcn;
uchar alpha = ColorChannel<uchar>::max();
CV_DECL_ALIGNED(16) float buf[3*BLOCK_SIZE];
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
for( i = 0; i < n; i += BLOCK_SIZE, src += BLOCK_SIZE*3 )
{
@@ -5375,7 +5375,7 @@ struct Lab2RGB_b
{
int i, j, dcn = dstcn;
uchar alpha = ColorChannel<uchar>::max();
CV_DECL_ALIGNED(16) float buf[3*BLOCK_SIZE];
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
for( i = 0; i < n; i += BLOCK_SIZE, src += BLOCK_SIZE*3 )
{
@@ -5731,7 +5731,7 @@ struct RGB2Luv_b
void operator()(const uchar* src, uchar* dst, int n) const
{
int i, j, scn = srccn;
float buf[3*BLOCK_SIZE];
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
for( i = 0; i < n; i += BLOCK_SIZE, dst += BLOCK_SIZE*3 )
{
@@ -5888,14 +5888,54 @@ struct Luv2RGB_b
v_140 = vdupq_n_f32(140.f);
v_scale = vdupq_n_f32(255.f);
v_alpha = vdup_n_u8(ColorChannel<uchar>::max());
#elif CV_SSE2
v_scale_inv = _mm_set1_ps(100.f/255.f);
v_coeff1 = _mm_set1_ps(1.388235294117647f);
v_coeff2 = _mm_set1_ps(1.027450980392157f);
v_134 = _mm_set1_ps(134.f);
v_140 = _mm_set1_ps(140.f);
v_scale = _mm_set1_ps(255.f);
v_zero = _mm_setzero_si128();
#endif
}
#if CV_SSE2
// 16s x 8
void process(__m128i v_l, __m128i v_u, __m128i v_v,
float * buf) const
{
__m128 v_l0 = _mm_cvtepi32_ps(_mm_unpacklo_epi16(v_l, v_zero));
__m128 v_u0 = _mm_cvtepi32_ps(_mm_unpacklo_epi16(v_u, v_zero));
__m128 v_v0 = _mm_cvtepi32_ps(_mm_unpacklo_epi16(v_v, v_zero));
__m128 v_l1 = _mm_cvtepi32_ps(_mm_unpackhi_epi16(v_l, v_zero));
__m128 v_u1 = _mm_cvtepi32_ps(_mm_unpackhi_epi16(v_u, v_zero));
__m128 v_v1 = _mm_cvtepi32_ps(_mm_unpackhi_epi16(v_v, v_zero));
v_l0 = _mm_mul_ps(v_l0, v_scale_inv);
v_l1 = _mm_mul_ps(v_l1, v_scale_inv);
v_u0 = _mm_sub_ps(_mm_mul_ps(v_u0, v_coeff1), v_134);
v_u1 = _mm_sub_ps(_mm_mul_ps(v_u1, v_coeff1), v_134);
v_v0 = _mm_sub_ps(_mm_mul_ps(v_v0, v_coeff2), v_140);
v_v1 = _mm_sub_ps(_mm_mul_ps(v_v1, v_coeff2), v_140);
_MM_INTERLIV_PS(v_l0, v_l1, v_u0, v_u1, v_v0, v_v1)
_mm_store_ps(buf, v_l0);
_mm_store_ps(buf + 4, v_l1);
_mm_store_ps(buf + 8, v_u0);
_mm_store_ps(buf + 12, v_u1);
_mm_store_ps(buf + 16, v_v0);
_mm_store_ps(buf + 20, v_v1);
}
#endif
void operator()(const uchar* src, uchar* dst, int n) const
{
int i, j, dcn = dstcn;
uchar alpha = ColorChannel<uchar>::max();
float buf[3*BLOCK_SIZE];
float CV_DECL_ALIGNED(16) buf[3*BLOCK_SIZE];
for( i = 0; i < n; i += BLOCK_SIZE, src += BLOCK_SIZE*3 )
{
@@ -5921,6 +5961,38 @@ struct Luv2RGB_b
v_dst.val[2] = vsubq_f32(vmulq_f32(vcvtq_f32_u32(vmovl_u16(vget_high_u16(v_t2))), v_coeff2), v_140);
vst3q_f32(buf + j + 12, v_dst);
}
#elif CV_SSE2
for ( ; j <= (dn - 32) * 3; j += 96)
{
__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));
_MM_DEINTERLIV_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),
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
for( ; j < dn*3; j += 3 )
{
@@ -5960,6 +6032,28 @@ struct Luv2RGB_b
vst3_u8(dst, v_dst);
}
}
#elif CV_SSE2
if (dcn == 3)
{
for ( ; j <= (dn * 3 - 16); j += 16, dst += 16)
{
__m128 v_src0 = _mm_mul_ps(_mm_load_ps(buf + j), v_scale);
__m128 v_src1 = _mm_mul_ps(_mm_load_ps(buf + j + 4), v_scale);
__m128 v_src2 = _mm_mul_ps(_mm_load_ps(buf + j + 8), v_scale);
__m128 v_src3 = _mm_mul_ps(_mm_load_ps(buf + j + 12), v_scale);
__m128i v_dst0 = _mm_packs_epi32(_mm_cvtps_epi32(v_src0),
_mm_cvtps_epi32(v_src1));
__m128i v_dst1 = _mm_packs_epi32(_mm_cvtps_epi32(v_src2),
_mm_cvtps_epi32(v_src3));
_mm_storeu_si128((__m128i *)dst, _mm_packus_epi16(v_dst0, v_dst1));
}
int jr = j % 3;
if (jr)
dst -= jr, j -= jr;
}
#endif
for( ; j < dn*3; j += 3, dst += dcn )
@@ -5979,6 +6073,9 @@ struct Luv2RGB_b
#if CV_NEON
float32x4_t v_scale, v_scale_inv, v_coeff1, v_coeff2, v_134, v_140;
uint8x8_t v_alpha;
#elif CV_SSE2
__m128 v_scale, v_scale_inv, v_coeff1, v_coeff2, v_134, v_140;
__m128i v_zero;
#endif
};