mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-29 13:47:32 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

cv::countNonZero

Browse Source
This commit is contained in:
Ilya Lavrenov 2014-09-28 06:36:07 -07:00
parent 34a571d37f
commit 44ea50f1c4
1 changed files with 140 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

145
modules/core/src/stat.cpp
View File

@ -181,7 +181,7 @@ struct Sum_SIMD<ushort, int>
for ( ; x <= len - 4; x += 4)
v_sum = vaddq_u32(v_sum, vmovl_u16(vld1_u16(src0 + x)));
unsigned int CV_DECL_ALIGNED(16) ar[4];
vst1q_u32(ar, v_sum);
@ -214,7 +214,7 @@ struct Sum_SIMD<short, int>
for ( ; x <= len - 4; x += 4)
v_sum = vaddq_s32(v_sum, vmovl_s16(vld1_s16(src0 + x)));
int CV_DECL_ALIGNED(16) ar[4];
vst1q_s32(ar, v_sum);
@ -426,6 +426,38 @@ static int countNonZero8u( const uchar* src, int len )
nz += tab[val & 255] + tab[val >> 8];
}
}
#elif CV_NEON
int len0 = len & -16, blockSize1 = (1 << 8) - 16, blockSize0 = blockSize1 << 6;
uint32x4_t v_nz = vdupq_n_u32(0u);
uint8x16_t v_zero = vdupq_n_u8(0), v_1 = vdupq_n_u8(1);
const uchar * src0 = src;
while( i < len0 )
{
int blockSizei = std::min(len0 - i, blockSize0), j = 0;
while (j < blockSizei)
{
int blockSizej = std::min(blockSizei - j, blockSize1), k = 0;
uint8x16_t v_pz = v_zero;
for( ; k <= blockSizej - 16; k += 16 )
v_pz = vaddq_u8(v_pz, vandq_u8(vceqq_u8(vld1q_u8(src0 + k), v_zero), v_1));
uint16x8_t v_p1 = vmovl_u8(vget_low_u8(v_pz)), v_p2 = vmovl_u8(vget_high_u8(v_pz));
v_nz = vaddq_u32(vaddl_u16(vget_low_u16(v_p1), vget_high_u16(v_p1)), v_nz);
v_nz = vaddq_u32(vaddl_u16(vget_low_u16(v_p2), vget_high_u16(v_p2)), v_nz);
src0 += blockSizej;
j += blockSizej;
}
i += blockSizei;
}
CV_DECL_ALIGNED(16) unsigned int buf[4];
vst1q_u32(buf, v_nz);
nz += i - saturate_cast<int>(buf[0] + buf[1] + buf[2] + buf[3]);
#endif
for( ; i < len; i++ )
nz += src[i] != 0;
@ -433,13 +465,116 @@ static int countNonZero8u( const uchar* src, int len )
}
static int countNonZero16u( const ushort* src, int len )
{ return countNonZero_(src, len); }
{
int i = 0, nz = 0;
#if CV_NEON
int len0 = len & -8, blockSize1 = (1 << 15), blockSize0 = blockSize1 << 6;
uint32x4_t v_nz = vdupq_n_u32(0u);
uint16x8_t v_zero = vdupq_n_u16(0), v_1 = vdupq_n_u16(1);
while( i < len0 )
{
int blockSizei = std::min(len0 - i, blockSize0), j = 0;
while (j < blockSizei)
{
int blockSizej = std::min(blockSizei - j, blockSize1), k = 0;
uint16x8_t v_pz = v_zero;
for( ; k <= blockSizej - 8; k += 8 )
v_pz = vaddq_u16(v_pz, vandq_u16(vceqq_u16(vld1q_u16(src + k), v_zero), v_1));
v_nz = vaddq_u32(vaddl_u16(vget_low_u16(v_pz), vget_high_u16(v_pz)), v_nz);
src += blockSizej;
j += blockSizej;
}
i += blockSizei;
}
CV_DECL_ALIGNED(16) unsigned int buf[4];
vst1q_u32(buf, v_nz);
nz += i - saturate_cast<int>(buf[0] + buf[1] + buf[2] + buf[3]);
#endif
return nz + countNonZero_(src, len - i);
}
static int countNonZero32s( const int* src, int len )
{ return countNonZero_(src, len); }
{
int i = 0, nz = 0;
#if CV_NEON
int len0 = len & -8, blockSize1 = (1 << 15), blockSize0 = blockSize1 << 6;
uint32x4_t v_nz = vdupq_n_u32(0u);
int32x4_t v_zero = vdupq_n_s32(0.0f);
uint16x8_t v_1 = vdupq_n_u16(1u), v_zerou = vdupq_n_u16(0u);
while( i < len0 )
{
int blockSizei = std::min(len0 - i, blockSize0), j = 0;
while (j < blockSizei)
{
int blockSizej = std::min(blockSizei - j, blockSize1), k = 0;
uint16x8_t v_pz = v_zerou;
for( ; k <= blockSizej - 8; k += 8 )
v_pz = vaddq_u16(v_pz, vandq_u16(vcombine_u16(vmovn_u32(vceqq_s32(vld1q_s32(src + k), v_zero)),
vmovn_u32(vceqq_s32(vld1q_s32(src + k + 4), v_zero))), v_1));
v_nz = vaddq_u32(vaddl_u16(vget_low_u16(v_pz), vget_high_u16(v_pz)), v_nz);
src += blockSizej;
j += blockSizej;
}
i += blockSizei;
}
CV_DECL_ALIGNED(16) unsigned int buf[4];
vst1q_u32(buf, v_nz);
nz += i - saturate_cast<int>(buf[0] + buf[1] + buf[2] + buf[3]);
#endif
return nz + countNonZero_(src, len - i);
}
static int countNonZero32f( const float* src, int len )
{ return countNonZero_(src, len); }
{
int i = 0, nz = 0;
#if CV_NEON
int len0 = len & -8, blockSize1 = (1 << 15), blockSize0 = blockSize1 << 6;
uint32x4_t v_nz = vdupq_n_u32(0u);
float32x4_t v_zero = vdupq_n_f32(0.0f);
uint16x8_t v_1 = vdupq_n_u16(1u), v_zerou = vdupq_n_u16(0u);
while( i < len0 )
{
int blockSizei = std::min(len0 - i, blockSize0), j = 0;
while (j < blockSizei)
{
int blockSizej = std::min(blockSizei - j, blockSize1), k = 0;
uint16x8_t v_pz = v_zerou;
for( ; k <= blockSizej - 8; k += 8 )
v_pz = vaddq_u16(v_pz, vandq_u16(vcombine_u16(vmovn_u32(vceqq_f32(vld1q_f32(src + k), v_zero)),
vmovn_u32(vceqq_f32(vld1q_f32(src + k + 4), v_zero))), v_1));
v_nz = vaddq_u32(vaddl_u16(vget_low_u16(v_pz), vget_high_u16(v_pz)), v_nz);
src += blockSizej;
j += blockSizej;
}
i += blockSizei;
}
CV_DECL_ALIGNED(16) unsigned int buf[4];
vst1q_u32(buf, v_nz);
nz += i - saturate_cast<int>(buf[0] + buf[1] + buf[2] + buf[3]);
#endif
return nz + countNonZero_(src, len - i);
}
static int countNonZero64f( const double* src, int len )
{ return countNonZero_(src, len); }