mirror of
https://github.com/opencv/opencv.git
synced 2025-01-19 06:53:50 +08:00
Merge pull request #9522 from terfendail:resize_move
This commit is contained in:
Alexander Alekhin
commit
42fbbfecc6
@ -55,207 +55,6 @@ namespace cv
|
||||
namespace opt_AVX2
|
||||
{
|
||||
|
||||
class resizeNNInvokerAVX4 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerAVX4(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=AVX
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
int avxWidth = width - (width & 0x7);
|
||||
const __m256i CV_DECL_ALIGNED(64) mask = _mm256_set1_epi32(-1);
|
||||
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 8)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
_mm256_maskstore_epi32((int*)D, mask, pixels);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 8)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
_mm256_storeu_si256((__m256i*)D, pixels);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerAVX4(const resizeNNInvokerAVX4&);
|
||||
resizeNNInvokerAVX4& operator=(const resizeNNInvokerAVX4&);
|
||||
};
|
||||
|
||||
class resizeNNInvokerAVX2 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerAVX2(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=AVX
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
//int avxWidth = (width - 1) - ((width - 1) & 0x7);
|
||||
int avxWidth = width - (width & 0xf);
|
||||
const __m256i CV_DECL_ALIGNED(64) mask = _mm256_set1_epi32(-1);
|
||||
const __m256i CV_DECL_ALIGNED(64) shuffle_mask = _mm256_set_epi8(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
|
||||
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
|
||||
const __m256i CV_DECL_ALIGNED(64) permute_mask = _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
|
||||
//const __m256i CV_DECL_ALIGNED(64) shift_shuffle_mask = _mm256_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2,
|
||||
// 13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
|
||||
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
const uchar* S2 = S - 2;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 16)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
|
||||
__m256i CV_DECL_ALIGNED(64) indices2 = _mm256_lddqu_si256(addr2);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
|
||||
__m256i CV_DECL_ALIGNED(64) unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
|
||||
|
||||
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
|
||||
__m256i CV_DECL_ALIGNED(64) ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
|
||||
_mm256_maskstore_epi32((int*)D, mask, ints_permuted);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
const uchar* S2 = S - 2;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 16)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
|
||||
__m256i CV_DECL_ALIGNED(64) indices2 = _mm256_lddqu_si256(addr2);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
|
||||
__m256i CV_DECL_ALIGNED(64) unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
|
||||
|
||||
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
|
||||
__m256i CV_DECL_ALIGNED(64) ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
|
||||
_mm256_storeu_si256((__m256i*)D, ints_permuted);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerAVX2(const resizeNNInvokerAVX2&);
|
||||
resizeNNInvokerAVX2& operator=(const resizeNNInvokerAVX2&);
|
||||
};
|
||||
|
||||
void resizeNN2_AVX2(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerAVX2 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
void resizeNN4_AVX2(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerAVX4 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
int warpAffineBlockline(int *adelta, int *bdelta, short* xy, short* alpha, int X0, int Y0, int bw)
|
||||
{
|
||||
const int AB_BITS = MAX(10, (int)INTER_BITS);
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
@ -56,8 +56,6 @@ namespace cv
|
||||
namespace opt_AVX2
|
||||
{
|
||||
#if CV_TRY_AVX2
|
||||
void resizeNN2_AVX2(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
void resizeNN4_AVX2(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
int warpAffineBlockline(int *adelta, int *bdelta, short* xy, short* alpha, int X0, int Y0, int bw);
|
||||
#endif
|
||||
}
|
||||
@ -65,10 +63,6 @@ int warpAffineBlockline(int *adelta, int *bdelta, short* xy, short* alpha, int X
|
||||
namespace opt_SSE4_1
|
||||
{
|
||||
#if CV_TRY_SSE4_1
|
||||
void resizeNN2_SSE4_1(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
void resizeNN4_SSE4_1(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
|
||||
int VResizeLanczos4Vec_32f16u_SSE41(const uchar** _src, uchar* _dst, const uchar* _beta, int width);
|
||||
void convertMaps_nninterpolate32f1c16s_SSE41(const float* src1f, const float* src2f, short* dst1, int width);
|
||||
void convertMaps_32f1c16s_SSE41(const float* src1f, const float* src2f, short* dst1, ushort* dst2, int width);
|
||||
void convertMaps_32f2c16s_SSE41(const float* src1f, short* dst1, ushort* dst2, int width);
|
||||
|
||||
@ -55,179 +55,6 @@ namespace cv
|
||||
namespace opt_SSE4_1
|
||||
{
|
||||
|
||||
class resizeNNInvokerSSE2 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerSSE2(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=SSE4.2
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
int sseWidth = width - (width & 0x7);
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
__m128i CV_DECL_ALIGNED(64) pixels = _mm_set1_epi16(0);
|
||||
for(x = 0; x < sseWidth; x += 8)
|
||||
{
|
||||
ushort imm = *(ushort*)(S + x_ofs[x + 0]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 0);
|
||||
imm = *(ushort*)(S + x_ofs[x + 1]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 1);
|
||||
imm = *(ushort*)(S + x_ofs[x + 2]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 2);
|
||||
imm = *(ushort*)(S + x_ofs[x + 3]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 3);
|
||||
imm = *(ushort*)(S + x_ofs[x + 4]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 4);
|
||||
imm = *(ushort*)(S + x_ofs[x + 5]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 5);
|
||||
imm = *(ushort*)(S + x_ofs[x + 6]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 6);
|
||||
imm = *(ushort*)(S + x_ofs[x + 7]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 7);
|
||||
_mm_storeu_si128((__m128i*)D, pixels);
|
||||
D += 16;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerSSE2(const resizeNNInvokerSSE2&);
|
||||
resizeNNInvokerSSE2& operator=(const resizeNNInvokerSSE2&);
|
||||
};
|
||||
|
||||
class resizeNNInvokerSSE4 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerSSE4(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=SSE4.2
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
int sseWidth = width - (width & 0x3);
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
__m128i CV_DECL_ALIGNED(64) pixels = _mm_set1_epi16(0);
|
||||
for(x = 0; x < sseWidth; x += 4)
|
||||
{
|
||||
int imm = *(int*)(S + x_ofs[x + 0]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 0);
|
||||
imm = *(int*)(S + x_ofs[x + 1]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 1);
|
||||
imm = *(int*)(S + x_ofs[x + 2]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 2);
|
||||
imm = *(int*)(S + x_ofs[x + 3]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 3);
|
||||
_mm_storeu_si128((__m128i*)D, pixels);
|
||||
D += 16;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerSSE4(const resizeNNInvokerSSE4&);
|
||||
resizeNNInvokerSSE4& operator=(const resizeNNInvokerSSE4&);
|
||||
};
|
||||
|
||||
void resizeNN2_SSE4_1(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerSSE2 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
void resizeNN4_SSE4_1(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerSSE4 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
int VResizeLanczos4Vec_32f16u_SSE41(const uchar** _src, uchar* _dst, const uchar* _beta, int width)
|
||||
{
|
||||
const float** src = (const float**)_src;
|
||||
const float* beta = (const float*)_beta;
|
||||
const float *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3],
|
||||
*S4 = src[4], *S5 = src[5], *S6 = src[6], *S7 = src[7];
|
||||
short * dst = (short*)_dst;
|
||||
int x = 0;
|
||||
__m128 v_b0 = _mm_set1_ps(beta[0]), v_b1 = _mm_set1_ps(beta[1]),
|
||||
v_b2 = _mm_set1_ps(beta[2]), v_b3 = _mm_set1_ps(beta[3]),
|
||||
v_b4 = _mm_set1_ps(beta[4]), v_b5 = _mm_set1_ps(beta[5]),
|
||||
v_b6 = _mm_set1_ps(beta[6]), v_b7 = _mm_set1_ps(beta[7]);
|
||||
|
||||
for (; x <= width - 8; x += 8)
|
||||
{
|
||||
__m128 v_dst0 = _mm_mul_ps(v_b0, _mm_loadu_ps(S0 + x));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b1, _mm_loadu_ps(S1 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b2, _mm_loadu_ps(S2 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b3, _mm_loadu_ps(S3 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b4, _mm_loadu_ps(S4 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b5, _mm_loadu_ps(S5 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b6, _mm_loadu_ps(S6 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b7, _mm_loadu_ps(S7 + x)));
|
||||
|
||||
__m128 v_dst1 = _mm_mul_ps(v_b0, _mm_loadu_ps(S0 + x + 4));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b1, _mm_loadu_ps(S1 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b2, _mm_loadu_ps(S2 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b3, _mm_loadu_ps(S3 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b4, _mm_loadu_ps(S4 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b5, _mm_loadu_ps(S5 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b6, _mm_loadu_ps(S6 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b7, _mm_loadu_ps(S7 + x + 4)));
|
||||
|
||||
__m128i v_dsti0 = _mm_cvtps_epi32(v_dst0);
|
||||
__m128i v_dsti1 = _mm_cvtps_epi32(v_dst1);
|
||||
|
||||
_mm_storeu_si128((__m128i *)(dst + x), _mm_packus_epi32(v_dsti0, v_dsti1));
|
||||
}
|
||||
|
||||
return x;
|
||||
}
|
||||
|
||||
void convertMaps_nninterpolate32f1c16s_SSE41(const float* src1f, const float* src2f, short* dst1, int width)
|
||||
{
|
||||
int x = 0;
|
||||
|
||||
261
modules/imgproc/src/resize.avx2.cpp
Normal file
261
modules/imgproc/src/resize.avx2.cpp
Normal file
@ -0,0 +1,261 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
/* ////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// Geometrical transforms on images and matrices: rotation, zoom etc.
|
||||
//
|
||||
// */
|
||||
|
||||
#include "precomp.hpp"
|
||||
#include "resize.hpp"
|
||||
|
||||
namespace cv
|
||||
{
|
||||
namespace opt_AVX2
|
||||
{
|
||||
|
||||
class resizeNNInvokerAVX4 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerAVX4(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=AVX
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
int avxWidth = width - (width & 0x7);
|
||||
const __m256i CV_DECL_ALIGNED(64) mask = _mm256_set1_epi32(-1);
|
||||
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 8)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
_mm256_maskstore_epi32((int*)D, mask, pixels);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 8)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
_mm256_storeu_si256((__m256i*)D, pixels);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerAVX4(const resizeNNInvokerAVX4&);
|
||||
resizeNNInvokerAVX4& operator=(const resizeNNInvokerAVX4&);
|
||||
};
|
||||
|
||||
class resizeNNInvokerAVX2 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerAVX2(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=AVX
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
//int avxWidth = (width - 1) - ((width - 1) & 0x7);
|
||||
int avxWidth = width - (width & 0xf);
|
||||
const __m256i CV_DECL_ALIGNED(64) mask = _mm256_set1_epi32(-1);
|
||||
const __m256i CV_DECL_ALIGNED(64) shuffle_mask = _mm256_set_epi8(15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0,
|
||||
15,14,11,10,13,12,9,8,7,6,3,2,5,4,1,0);
|
||||
const __m256i CV_DECL_ALIGNED(64) permute_mask = _mm256_set_epi32(7, 5, 3, 1, 6, 4, 2, 0);
|
||||
//const __m256i CV_DECL_ALIGNED(64) shift_shuffle_mask = _mm256_set_epi8(13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2,
|
||||
// 13,12,15,14,9,8,11,10,5,4,7,6,1,0,3,2);
|
||||
if(((int64)(dst.data + dst.step) & 0x1f) == 0)
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
const uchar* S2 = S - 2;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 16)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
|
||||
__m256i CV_DECL_ALIGNED(64) indices2 = _mm256_lddqu_si256(addr2);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
|
||||
__m256i CV_DECL_ALIGNED(64) unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
|
||||
|
||||
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
|
||||
__m256i CV_DECL_ALIGNED(64) ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
|
||||
_mm256_maskstore_epi32((int*)D, mask, ints_permuted);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
else
|
||||
{
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
const uchar* S2 = S - 2;
|
||||
#ifdef CV_ICC
|
||||
#pragma unroll(4)
|
||||
#endif
|
||||
for(x = 0; x < avxWidth; x += 16)
|
||||
{
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr = (__m256i*)(x_ofs + x);
|
||||
__m256i CV_DECL_ALIGNED(64) indices = _mm256_lddqu_si256(addr);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels1 = _mm256_i32gather_epi32((const int*)S, indices, 1);
|
||||
const __m256i CV_DECL_ALIGNED(64) *addr2 = (__m256i*)(x_ofs + x + 8);
|
||||
__m256i CV_DECL_ALIGNED(64) indices2 = _mm256_lddqu_si256(addr2);
|
||||
__m256i CV_DECL_ALIGNED(64) pixels2 = _mm256_i32gather_epi32((const int*)S2, indices2, 1);
|
||||
__m256i CV_DECL_ALIGNED(64) unpacked = _mm256_blend_epi16(pixels1, pixels2, 0xaa);
|
||||
|
||||
__m256i CV_DECL_ALIGNED(64) bytes_shuffled = _mm256_shuffle_epi8(unpacked, shuffle_mask);
|
||||
__m256i CV_DECL_ALIGNED(64) ints_permuted = _mm256_permutevar8x32_epi32(bytes_shuffled, permute_mask);
|
||||
_mm256_storeu_si256((__m256i*)D, ints_permuted);
|
||||
D += 32;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerAVX2(const resizeNNInvokerAVX2&);
|
||||
resizeNNInvokerAVX2& operator=(const resizeNNInvokerAVX2&);
|
||||
};
|
||||
|
||||
void resizeNN2_AVX2(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerAVX2 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
void resizeNN4_AVX2(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerAVX4 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
/* End of file. */
|
||||
3330
modules/imgproc/src/resize.cpp
Normal file
3330
modules/imgproc/src/resize.cpp
Normal file
File diff suppressed because it is too large
Load Diff
75
modules/imgproc/src/resize.hpp
Normal file
75
modules/imgproc/src/resize.hpp
Normal file
@ -0,0 +1,75 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
/* ////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// Geometrical transforms on images and matrices: rotation, zoom etc.
|
||||
//
|
||||
// */
|
||||
|
||||
#ifndef OPENCV_IMGPROC_RESIZE_HPP
|
||||
#define OPENCV_IMGPROC_RESIZE_HPP
|
||||
#include "precomp.hpp"
|
||||
|
||||
namespace cv
|
||||
{
|
||||
namespace opt_AVX2
|
||||
{
|
||||
#if CV_TRY_AVX2
|
||||
void resizeNN2_AVX2(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
void resizeNN4_AVX2(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
#endif
|
||||
}
|
||||
|
||||
namespace opt_SSE4_1
|
||||
{
|
||||
#if CV_TRY_SSE4_1
|
||||
void resizeNN2_SSE4_1(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
void resizeNN4_SSE4_1(const Range&, const Mat&, Mat&, int*, int, double);
|
||||
|
||||
int VResizeLanczos4Vec_32f16u_SSE41(const uchar** _src, uchar* _dst, const uchar* _beta, int width);
|
||||
#endif
|
||||
}
|
||||
}
|
||||
#endif
|
||||
/* End of file. */
|
||||
233
modules/imgproc/src/resize.sse4_1.cpp
Normal file
233
modules/imgproc/src/resize.sse4_1.cpp
Normal file
@ -0,0 +1,233 @@
|
||||
/*M///////////////////////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
|
||||
//
|
||||
// By downloading, copying, installing or using the software you agree to this license.
|
||||
// If you do not agree to this license, do not download, install,
|
||||
// copy or use the software.
|
||||
//
|
||||
//
|
||||
// License Agreement
|
||||
// For Open Source Computer Vision Library
|
||||
//
|
||||
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
|
||||
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
|
||||
// Copyright (C) 2014-2015, Itseez Inc., all rights reserved.
|
||||
// Third party copyrights are property of their respective owners.
|
||||
//
|
||||
// Redistribution and use in source and binary forms, with or without modification,
|
||||
// are permitted provided that the following conditions are met:
|
||||
//
|
||||
// * Redistribution's of source code must retain the above copyright notice,
|
||||
// this list of conditions and the following disclaimer.
|
||||
//
|
||||
// * Redistribution's in binary form must reproduce the above copyright notice,
|
||||
// this list of conditions and the following disclaimer in the documentation
|
||||
// and/or other materials provided with the distribution.
|
||||
//
|
||||
// * The name of the copyright holders may not be used to endorse or promote products
|
||||
// derived from this software without specific prior written permission.
|
||||
//
|
||||
// This software is provided by the copyright holders and contributors "as is" and
|
||||
// any express or implied warranties, including, but not limited to, the implied
|
||||
// warranties of merchantability and fitness for a particular purpose are disclaimed.
|
||||
// In no event shall the Intel Corporation or contributors be liable for any direct,
|
||||
// indirect, incidental, special, exemplary, or consequential damages
|
||||
// (including, but not limited to, procurement of substitute goods or services;
|
||||
// loss of use, data, or profits; or business interruption) however caused
|
||||
// and on any theory of liability, whether in contract, strict liability,
|
||||
// or tort (including negligence or otherwise) arising in any way out of
|
||||
// the use of this software, even if advised of the possibility of such damage.
|
||||
//
|
||||
//M*/
|
||||
|
||||
/* ////////////////////////////////////////////////////////////////////
|
||||
//
|
||||
// Geometrical transforms on images and matrices: rotation, zoom etc.
|
||||
//
|
||||
// */
|
||||
|
||||
#include "precomp.hpp"
|
||||
#include "resize.hpp"
|
||||
|
||||
namespace cv
|
||||
{
|
||||
namespace opt_SSE4_1
|
||||
{
|
||||
|
||||
class resizeNNInvokerSSE2 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerSSE2(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=SSE4.2
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
int sseWidth = width - (width & 0x7);
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
__m128i CV_DECL_ALIGNED(64) pixels = _mm_set1_epi16(0);
|
||||
for(x = 0; x < sseWidth; x += 8)
|
||||
{
|
||||
ushort imm = *(ushort*)(S + x_ofs[x + 0]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 0);
|
||||
imm = *(ushort*)(S + x_ofs[x + 1]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 1);
|
||||
imm = *(ushort*)(S + x_ofs[x + 2]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 2);
|
||||
imm = *(ushort*)(S + x_ofs[x + 3]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 3);
|
||||
imm = *(ushort*)(S + x_ofs[x + 4]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 4);
|
||||
imm = *(ushort*)(S + x_ofs[x + 5]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 5);
|
||||
imm = *(ushort*)(S + x_ofs[x + 6]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 6);
|
||||
imm = *(ushort*)(S + x_ofs[x + 7]);
|
||||
pixels = _mm_insert_epi16(pixels, imm, 7);
|
||||
_mm_storeu_si128((__m128i*)D, pixels);
|
||||
D += 16;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(ushort*)(Dstart + x*2) = *(ushort*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerSSE2(const resizeNNInvokerSSE2&);
|
||||
resizeNNInvokerSSE2& operator=(const resizeNNInvokerSSE2&);
|
||||
};
|
||||
|
||||
class resizeNNInvokerSSE4 :
|
||||
public ParallelLoopBody
|
||||
{
|
||||
public:
|
||||
resizeNNInvokerSSE4(const Mat& _src, Mat &_dst, int *_x_ofs, int _pix_size4, double _ify) :
|
||||
ParallelLoopBody(), src(_src), dst(_dst), x_ofs(_x_ofs), pix_size4(_pix_size4),
|
||||
ify(_ify)
|
||||
{
|
||||
}
|
||||
#if defined(__INTEL_COMPILER)
|
||||
#pragma optimization_parameter target_arch=SSE4.2
|
||||
#endif
|
||||
virtual void operator() (const Range& range) const
|
||||
{
|
||||
Size ssize = src.size(), dsize = dst.size();
|
||||
int y, x;
|
||||
int width = dsize.width;
|
||||
int sseWidth = width - (width & 0x3);
|
||||
for(y = range.start; y < range.end; y++)
|
||||
{
|
||||
uchar* D = dst.data + dst.step*y;
|
||||
uchar* Dstart = D;
|
||||
int sy = std::min(cvFloor(y*ify), ssize.height-1);
|
||||
const uchar* S = src.data + sy*src.step;
|
||||
__m128i CV_DECL_ALIGNED(64) pixels = _mm_set1_epi16(0);
|
||||
for(x = 0; x < sseWidth; x += 4)
|
||||
{
|
||||
int imm = *(int*)(S + x_ofs[x + 0]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 0);
|
||||
imm = *(int*)(S + x_ofs[x + 1]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 1);
|
||||
imm = *(int*)(S + x_ofs[x + 2]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 2);
|
||||
imm = *(int*)(S + x_ofs[x + 3]);
|
||||
pixels = _mm_insert_epi32(pixels, imm, 3);
|
||||
_mm_storeu_si128((__m128i*)D, pixels);
|
||||
D += 16;
|
||||
}
|
||||
for(; x < width; x++)
|
||||
{
|
||||
*(int*)(Dstart + x*4) = *(int*)(S + x_ofs[x]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private:
|
||||
const Mat src;
|
||||
Mat dst;
|
||||
int* x_ofs, pix_size4;
|
||||
double ify;
|
||||
|
||||
resizeNNInvokerSSE4(const resizeNNInvokerSSE4&);
|
||||
resizeNNInvokerSSE4& operator=(const resizeNNInvokerSSE4&);
|
||||
};
|
||||
|
||||
void resizeNN2_SSE4_1(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerSSE2 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
void resizeNN4_SSE4_1(const Range& range, const Mat& src, Mat &dst, int *x_ofs, int pix_size4, double ify)
|
||||
{
|
||||
resizeNNInvokerSSE4 invoker(src, dst, x_ofs, pix_size4, ify);
|
||||
parallel_for_(range, invoker, dst.total() / (double)(1 << 16));
|
||||
}
|
||||
|
||||
int VResizeLanczos4Vec_32f16u_SSE41(const uchar** _src, uchar* _dst, const uchar* _beta, int width)
|
||||
{
|
||||
const float** src = (const float**)_src;
|
||||
const float* beta = (const float*)_beta;
|
||||
const float *S0 = src[0], *S1 = src[1], *S2 = src[2], *S3 = src[3],
|
||||
*S4 = src[4], *S5 = src[5], *S6 = src[6], *S7 = src[7];
|
||||
short * dst = (short*)_dst;
|
||||
int x = 0;
|
||||
__m128 v_b0 = _mm_set1_ps(beta[0]), v_b1 = _mm_set1_ps(beta[1]),
|
||||
v_b2 = _mm_set1_ps(beta[2]), v_b3 = _mm_set1_ps(beta[3]),
|
||||
v_b4 = _mm_set1_ps(beta[4]), v_b5 = _mm_set1_ps(beta[5]),
|
||||
v_b6 = _mm_set1_ps(beta[6]), v_b7 = _mm_set1_ps(beta[7]);
|
||||
|
||||
for (; x <= width - 8; x += 8)
|
||||
{
|
||||
__m128 v_dst0 = _mm_mul_ps(v_b0, _mm_loadu_ps(S0 + x));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b1, _mm_loadu_ps(S1 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b2, _mm_loadu_ps(S2 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b3, _mm_loadu_ps(S3 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b4, _mm_loadu_ps(S4 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b5, _mm_loadu_ps(S5 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b6, _mm_loadu_ps(S6 + x)));
|
||||
v_dst0 = _mm_add_ps(v_dst0, _mm_mul_ps(v_b7, _mm_loadu_ps(S7 + x)));
|
||||
|
||||
__m128 v_dst1 = _mm_mul_ps(v_b0, _mm_loadu_ps(S0 + x + 4));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b1, _mm_loadu_ps(S1 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b2, _mm_loadu_ps(S2 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b3, _mm_loadu_ps(S3 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b4, _mm_loadu_ps(S4 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b5, _mm_loadu_ps(S5 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b6, _mm_loadu_ps(S6 + x + 4)));
|
||||
v_dst1 = _mm_add_ps(v_dst1, _mm_mul_ps(v_b7, _mm_loadu_ps(S7 + x + 4)));
|
||||
|
||||
__m128i v_dsti0 = _mm_cvtps_epi32(v_dst0);
|
||||
__m128i v_dsti1 = _mm_cvtps_epi32(v_dst1);
|
||||
|
||||
_mm_storeu_si128((__m128i *)(dst + x), _mm_packus_epi32(v_dsti0, v_dsti1));
|
||||
}
|
||||
|
||||
return x;
|
||||
}
|
||||
|
||||
}
|
||||
}
|
||||
/* End of file. */
|
||||
Loading…
Reference in New Issue
Block a user