mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2025-01-18 22:44:02 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

add SSE code for HSV2RGB_f.

Browse Source
This commit is contained in:
k-shinotsuka 2016-11-12 19:02:43 +09:00
parent 8944875258
commit 614ce7743e
1 changed files with 151 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

154
modules/imgproc/src/color.cpp
View File

@ -4443,16 +4443,161 @@ struct HSV2RGB_f
typedef float channel_type;
HSV2RGB_f(int _dstcn, int _blueIdx, float _hrange)
: dstcn(_dstcn), blueIdx(_blueIdx), hscale(6.f/_hrange) {}
: dstcn(_dstcn), blueIdx(_blueIdx), hscale(6.f/_hrange) {
#if CV_SSE2
haveSIMD = checkHardwareSupport(CV_CPU_SSE2);
#endif
}
#if CV_SSE2
void process(__m128& v_h0, __m128& v_h1, __m128& v_s0,
__m128& v_s1, __m128& v_v0, __m128& v_v1) const
{
v_h0 = _mm_mul_ps(v_h0, _mm_set1_ps(hscale));
v_h1 = _mm_mul_ps(v_h1, _mm_set1_ps(hscale));
__m128 v_pre_sector0 = _mm_cvtepi32_ps(_mm_cvttps_epi32(v_h0));
__m128 v_pre_sector1 = _mm_cvtepi32_ps(_mm_cvttps_epi32(v_h1));
v_h0 = _mm_sub_ps(v_h0, v_pre_sector0);
v_h1 = _mm_sub_ps(v_h1, v_pre_sector1);
__m128 v_tab00 = v_v0;
__m128 v_tab01 = v_v1;
__m128 v_tab10 = _mm_mul_ps(v_v0, _mm_sub_ps(_mm_set1_ps(1.0f), v_s0));
__m128 v_tab11 = _mm_mul_ps(v_v1, _mm_sub_ps(_mm_set1_ps(1.0f), v_s1));
__m128 v_tab20 = _mm_mul_ps(v_v0, _mm_sub_ps(_mm_set1_ps(1.0f), _mm_mul_ps(v_s0, v_h0)));
__m128 v_tab21 = _mm_mul_ps(v_v1, _mm_sub_ps(_mm_set1_ps(1.0f), _mm_mul_ps(v_s1, v_h1)));
__m128 v_tab30 = _mm_mul_ps(v_v0, _mm_sub_ps(_mm_set1_ps(1.0f), _mm_mul_ps(v_s0, _mm_sub_ps(_mm_set1_ps(1.0f), v_h0))));
__m128 v_tab31 = _mm_mul_ps(v_v1, _mm_sub_ps(_mm_set1_ps(1.0f), _mm_mul_ps(v_s1, _mm_sub_ps(_mm_set1_ps(1.0f), v_h1))));
__m128 v_sector0 = _mm_div_ps(v_pre_sector0, _mm_set1_ps(6.0f));
__m128 v_sector1 = _mm_div_ps(v_pre_sector1, _mm_set1_ps(6.0f));
v_sector0 = _mm_cvtepi32_ps(_mm_cvttps_epi32(v_sector0));
v_sector1 = _mm_cvtepi32_ps(_mm_cvttps_epi32(v_sector1));
v_sector0 = _mm_mul_ps(v_sector0, _mm_set1_ps(6.0f));
v_sector1 = _mm_mul_ps(v_sector1, _mm_set1_ps(6.0f));
v_sector0 = _mm_sub_ps(v_pre_sector0, v_sector0);
v_sector1 = _mm_sub_ps(v_pre_sector1, v_sector1);
v_h0 = _mm_and_ps(v_tab10, _mm_cmplt_ps(v_sector0, _mm_set1_ps(2.0f)));
v_h1 = _mm_and_ps(v_tab11, _mm_cmplt_ps(v_sector1, _mm_set1_ps(2.0f)));
v_h0 = _mm_or_ps(v_h0, _mm_and_ps(v_tab30, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(2.0f))));
v_h1 = _mm_or_ps(v_h1, _mm_and_ps(v_tab31, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(2.0f))));
v_h0 = _mm_or_ps(v_h0, _mm_and_ps(v_tab00, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(3.0f))));
v_h1 = _mm_or_ps(v_h1, _mm_and_ps(v_tab01, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(3.0f))));
v_h0 = _mm_or_ps(v_h0, _mm_and_ps(v_tab00, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(4.0f))));
v_h1 = _mm_or_ps(v_h1, _mm_and_ps(v_tab01, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(4.0f))));
v_h0 = _mm_or_ps(v_h0, _mm_and_ps(v_tab20, _mm_cmpgt_ps(v_sector0, _mm_set1_ps(4.0f))));
v_h1 = _mm_or_ps(v_h1, _mm_and_ps(v_tab21, _mm_cmpgt_ps(v_sector1, _mm_set1_ps(4.0f))));
v_s0 = _mm_and_ps(v_tab30, _mm_cmplt_ps(v_sector0, _mm_set1_ps(1.0f)));
v_s1 = _mm_and_ps(v_tab31, _mm_cmplt_ps(v_sector1, _mm_set1_ps(1.0f)));
v_s0 = _mm_or_ps(v_s0, _mm_and_ps(v_tab00, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(1.0f))));
v_s1 = _mm_or_ps(v_s1, _mm_and_ps(v_tab01, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(1.0f))));
v_s0 = _mm_or_ps(v_s0, _mm_and_ps(v_tab00, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(2.0f))));
v_s1 = _mm_or_ps(v_s1, _mm_and_ps(v_tab01, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(2.0f))));
v_s0 = _mm_or_ps(v_s0, _mm_and_ps(v_tab20, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(3.0f))));
v_s1 = _mm_or_ps(v_s1, _mm_and_ps(v_tab21, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(3.0f))));
v_s0 = _mm_or_ps(v_s0, _mm_and_ps(v_tab10, _mm_cmpgt_ps(v_sector0, _mm_set1_ps(3.0f))));
v_s1 = _mm_or_ps(v_s1, _mm_and_ps(v_tab11, _mm_cmpgt_ps(v_sector1, _mm_set1_ps(3.0f))));
v_v0 = _mm_and_ps(v_tab00, _mm_cmplt_ps(v_sector0, _mm_set1_ps(1.0f)));
v_v1 = _mm_and_ps(v_tab01, _mm_cmplt_ps(v_sector1, _mm_set1_ps(1.0f)));
v_v0 = _mm_or_ps(v_v0, _mm_and_ps(v_tab20, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(1.0f))));
v_v1 = _mm_or_ps(v_v1, _mm_and_ps(v_tab21, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(1.0f))));
v_v0 = _mm_or_ps(v_v0, _mm_and_ps(v_tab10, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(2.0f))));
v_v1 = _mm_or_ps(v_v1, _mm_and_ps(v_tab11, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(2.0f))));
v_v0 = _mm_or_ps(v_v0, _mm_and_ps(v_tab10, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(3.0f))));
v_v1 = _mm_or_ps(v_v1, _mm_and_ps(v_tab11, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(3.0f))));
v_v0 = _mm_or_ps(v_v0, _mm_and_ps(v_tab30, _mm_cmpeq_ps(v_sector0, _mm_set1_ps(4.0f))));
v_v1 = _mm_or_ps(v_v1, _mm_and_ps(v_tab31, _mm_cmpeq_ps(v_sector1, _mm_set1_ps(4.0f))));
v_v0 = _mm_or_ps(v_v0, _mm_and_ps(v_tab00, _mm_cmpgt_ps(v_sector0, _mm_set1_ps(4.0f))));
v_v1 = _mm_or_ps(v_v1, _mm_and_ps(v_tab01, _mm_cmpgt_ps(v_sector1, _mm_set1_ps(4.0f))));
}
#endif
void operator()(const float* src, float* dst, int n) const
{
int i, bidx = blueIdx, dcn = dstcn;
int i = 0, bidx = blueIdx, dcn = dstcn;
float _hscale = hscale;
float alpha = ColorChannel<float>::max();
n *= 3;
for( i = 0; i < n; i += 3, dst += dcn )
#if CV_SSE2
if (haveSIMD)
{
for( ; i <= n - 24; i += 24, dst += dcn * 8 )
{
__m128 v_h0 = _mm_loadu_ps(src + i + 0);
__m128 v_h1 = _mm_loadu_ps(src + i + 4);
__m128 v_s0 = _mm_loadu_ps(src + i + 8);
__m128 v_s1 = _mm_loadu_ps(src + i + 12);
__m128 v_v0 = _mm_loadu_ps(src + i + 16);
__m128 v_v1 = _mm_loadu_ps(src + i + 20);
_mm_deinterleave_ps(v_h0, v_h1, v_s0, v_s1, v_v0, v_v1);
process(v_h0, v_h1, v_s0, v_s1, v_v0, v_v1);
if (dcn == 3)
{
if (bidx)
{
_mm_interleave_ps(v_v0, v_v1, v_s0, v_s1, v_h0, v_h1);
_mm_storeu_ps(dst + 0, v_v0);
_mm_storeu_ps(dst + 4, v_v1);
_mm_storeu_ps(dst + 8, v_s0);
_mm_storeu_ps(dst + 12, v_s1);
_mm_storeu_ps(dst + 16, v_h0);
_mm_storeu_ps(dst + 20, v_h1);
}
else
{
_mm_interleave_ps(v_h0, v_h1, v_s0, v_s1, v_v0, v_v1);
_mm_storeu_ps(dst + 0, v_h0);
_mm_storeu_ps(dst + 4, v_h1);
_mm_storeu_ps(dst + 8, v_s0);
_mm_storeu_ps(dst + 12, v_s1);
_mm_storeu_ps(dst + 16, v_v0);
_mm_storeu_ps(dst + 20, v_v1);
}
}
else
{
__m128 v_a0 = _mm_set1_ps(alpha);
__m128 v_a1 = _mm_set1_ps(alpha);
if (bidx)
{
_mm_interleave_ps(v_v0, v_v1, v_s0, v_s1, v_h0, v_h1, v_a0, v_a1);
_mm_storeu_ps(dst + 0, v_v0);
_mm_storeu_ps(dst + 4, v_v1);
_mm_storeu_ps(dst + 8, v_s0);
_mm_storeu_ps(dst + 12, v_s1);
_mm_storeu_ps(dst + 16, v_h0);
_mm_storeu_ps(dst + 20, v_h1);
_mm_storeu_ps(dst + 24, v_a0);
_mm_storeu_ps(dst + 28, v_a1);
}
else
{
_mm_interleave_ps(v_h0, v_h1, v_s0, v_s1, v_v0, v_v1, v_a0, v_a1);
_mm_storeu_ps(dst + 0, v_h0);
_mm_storeu_ps(dst + 4, v_h1);
_mm_storeu_ps(dst + 8, v_s0);
_mm_storeu_ps(dst + 12, v_s1);
_mm_storeu_ps(dst + 16, v_v0);
_mm_storeu_ps(dst + 20, v_v1);
_mm_storeu_ps(dst + 24, v_a0);
_mm_storeu_ps(dst + 28, v_a1);
}
}
}
}
#endif
for( ; i < n; i += 3, dst += dcn )
{
float h = src[i], s = src[i+1], v = src[i+2];
float b, g, r;
@ -4498,6 +4643,9 @@ struct HSV2RGB_f
int dstcn, blueIdx;
float hscale;
#if CV_SSE2
bool haveSIMD;
#endif
};