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

make NEON more similart to SSE2

Browse Source
This commit is contained in:
Tomoaki Teshima 2016-06-29 11:17:14 +09:00
parent 1b6e02793c
commit 78a5aacb22
1 changed files with 148 additions and 91 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

239
modules/imgproc/src/smooth.cpp
View File

@ -229,6 +229,8 @@ struct ColumnSum<int, uchar> :
#if CV_SSE2
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
#elif CV_NEON
bool haveNEON = checkHardwareSupport(CV_CPU_NEON);
#endif
if( width != (int)sum.size() )
@ -256,8 +258,11 @@ struct ColumnSum<int, uchar> :
}
}
#elif CV_NEON
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
if(haveNEON)
{
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
}
#endif
for( ; i < width; i++ )
SUM[i] += Sp[i];
@ -303,20 +308,23 @@ struct ColumnSum<int, uchar> :
}
}
#elif CV_NEON
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
uint32x4_t v_s0d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
uint32x4_t v_s01d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
uint32x4_t v_s0d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
uint32x4_t v_s01d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
uint16x8_t v_dst = vcombine_u16(vqmovn_u32(v_s0d), vqmovn_u32(v_s01d));
vst1_u8(D + i, vqmovn_u16(v_dst));
uint16x8_t v_dst = vcombine_u16(vqmovn_u32(v_s0d), vqmovn_u32(v_s01d));
vst1_u8(D + i, vqmovn_u16(v_dst));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
for( ; i < width; i++ )
@ -351,16 +359,19 @@ struct ColumnSum<int, uchar> :
}
}
#elif CV_NEON
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
uint16x8_t v_dst = vcombine_u16(vqmovun_s32(v_s0), vqmovun_s32(v_s01));
vst1_u8(D + i, vqmovn_u16(v_dst));
uint16x8_t v_dst = vcombine_u16(vqmovun_s32(v_s0), vqmovun_s32(v_s01));
vst1_u8(D + i, vqmovn_u16(v_dst));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
@ -404,6 +415,8 @@ struct ColumnSum<int, short> :
#if CV_SSE2
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
#elif CV_NEON
bool haveNEON = checkHardwareSupport(CV_CPU_NEON);
#endif
if( width != (int)sum.size() )
@ -411,6 +424,7 @@ struct ColumnSum<int, short> :
sum.resize(width);
sumCount = 0;
}
SUM = &sum[0];
if( sumCount == 0 )
{
@ -430,8 +444,11 @@ struct ColumnSum<int, short> :
}
}
#elif CV_NEON
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
if(haveNEON)
{
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
}
#endif
for( ; i < width; i++ )
SUM[i] += Sp[i];
@ -475,18 +492,21 @@ struct ColumnSum<int, short> :
}
}
#elif CV_NEON
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
int32x4_t v_s0d = cv_vrndq_s32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
int32x4_t v_s01d = cv_vrndq_s32_f32(vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
vst1q_s16(D + i, vcombine_s16(vqmovn_s32(v_s0d), vqmovn_s32(v_s01d)));
int32x4_t v_s0d = cv_vrndq_s32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
int32x4_t v_s01d = cv_vrndq_s32_f32(vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
vst1q_s16(D + i, vcombine_s16(vqmovn_s32(v_s0d), vqmovn_s32(v_s01d)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
for( ; i < width; i++ )
@ -520,15 +540,18 @@ struct ColumnSum<int, short> :
}
}
#elif CV_NEON
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
vst1q_s16(D + i, vcombine_s16(vqmovn_s32(v_s0), vqmovn_s32(v_s01)));
vst1q_s16(D + i, vcombine_s16(vqmovn_s32(v_s0), vqmovn_s32(v_s01)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
@ -570,8 +593,11 @@ struct ColumnSum<int, ushort> :
int* SUM;
bool haveScale = scale != 1;
double _scale = scale;
#if CV_SSE2
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
#elif CV_NEON
bool haveNEON = checkHardwareSupport(CV_CPU_NEON);
#endif
if( width != (int)sum.size() )
@ -579,6 +605,7 @@ struct ColumnSum<int, ushort> :
sum.resize(width);
sumCount = 0;
}
SUM = &sum[0];
if( sumCount == 0 )
{
@ -590,16 +617,19 @@ struct ColumnSum<int, ushort> :
#if CV_SSE2
if(haveSSE2)
{
for( ; i < width-4; i+=4 )
for( ; i <= width-4; i+=4 )
{
__m128i _sum = _mm_loadu_si128((const __m128i*)(SUM+i));
__m128i _sp = _mm_loadu_si128((const __m128i*)(Sp+i));
_mm_storeu_si128((__m128i*)(SUM+i), _mm_add_epi32(_sum, _sp));
_mm_storeu_si128((__m128i*)(SUM+i),_mm_add_epi32(_sum, _sp));
}
}
#elif CV_NEON
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
if(haveNEON)
{
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
}
#endif
for( ; i < width; i++ )
SUM[i] += Sp[i];
@ -642,18 +672,21 @@ struct ColumnSum<int, ushort> :
}
}
#elif CV_NEON
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
uint32x4_t v_s0d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
uint32x4_t v_s01d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
vst1q_u16(D + i, vcombine_u16(vqmovn_u32(v_s0d), vqmovn_u32(v_s01d)));
uint32x4_t v_s0d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
uint32x4_t v_s01d = cv_vrndq_u32_f32(vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
vst1q_u16(D + i, vcombine_u16(vqmovn_u32(v_s0d), vqmovn_u32(v_s01d)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
for( ; i < width; i++ )
@ -686,15 +719,18 @@ struct ColumnSum<int, ushort> :
}
}
#elif CV_NEON
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
vst1q_u16(D + i, vcombine_u16(vqmovun_s32(v_s0), vqmovun_s32(v_s01)));
vst1q_u16(D + i, vcombine_u16(vqmovun_s32(v_s0), vqmovun_s32(v_s01)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
@ -738,6 +774,8 @@ struct ColumnSum<int, int> :
#if CV_SSE2
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
#elif CV_NEON
bool haveNEON = checkHardwareSupport(CV_CPU_NEON);
#endif
if( width != (int)sum.size() )
@ -745,6 +783,7 @@ struct ColumnSum<int, int> :
sum.resize(width);
sumCount = 0;
}
SUM = &sum[0];
if( sumCount == 0 )
{
@ -764,8 +803,11 @@ struct ColumnSum<int, int> :
}
}
#elif CV_NEON
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
if(haveNEON)
{
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
}
#endif
for( ; i < width; i++ )
SUM[i] += Sp[i];
@ -803,15 +845,18 @@ struct ColumnSum<int, int> :
}
}
#elif CV_NEON
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-4; i+=4 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-4; i+=4 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s0d = cv_vrndq_s32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
vst1q_s32(D + i, v_s0d);
int32x4_t v_s0d = cv_vrndq_s32_f32(vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
vst1q_s32(D + i, v_s0d);
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
}
}
#endif
for( ; i < width; i++ )
@ -838,12 +883,15 @@ struct ColumnSum<int, int> :
}
}
#elif CV_NEON
for( ; i <= width-4; i+=4 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
for( ; i <= width-4; i+=4 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
vst1q_s32(D + i, v_s0);
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(D + i, v_s0);
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
}
}
#endif
@ -887,7 +935,9 @@ struct ColumnSum<int, float> :
double _scale = scale;
#if CV_SSE2
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
bool haveSSE2 = checkHardwareSupport(CV_CPU_SSE2);
#elif CV_NEON
bool haveNEON = checkHardwareSupport(CV_CPU_NEON);
#endif
if( width != (int)sum.size() )
@ -899,26 +949,27 @@ struct ColumnSum<int, float> :
SUM = &sum[0];
if( sumCount == 0 )
{
memset((void *)SUM, 0, sizeof(int) * width);
memset((void*)SUM, 0, width*sizeof(int));
for( ; sumCount < ksize - 1; sumCount++, src++ )
{
const int* Sp = (const int*)src[0];
i = 0;
#if CV_SSE2
if(haveSSE2)
{
for( ; i < width-4; i+=4 )
for( ; i <= width-4; i+=4 )
{
__m128i _sum = _mm_loadu_si128((const __m128i*)(SUM+i));
__m128i _sp = _mm_loadu_si128((const __m128i*)(Sp+i));
_mm_storeu_si128((__m128i*)(SUM+i), _mm_add_epi32(_sum, _sp));
_mm_storeu_si128((__m128i*)(SUM+i),_mm_add_epi32(_sum, _sp));
}
}
#elif CV_NEON
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
if(haveNEON)
{
for( ; i <= width - 4; i+=4 )
vst1q_s32(SUM + i, vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i)));
}
#endif
for( ; i < width; i++ )
@ -956,17 +1007,20 @@ struct ColumnSum<int, float> :
}
}
#elif CV_NEON
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
float32x4_t v_scale = vdupq_n_f32((float)_scale);
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
vst1q_f32(D + i, vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
vst1q_f32(D + i + 4, vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
vst1q_f32(D + i, vmulq_f32(vcvtq_f32_s32(v_s0), v_scale));
vst1q_f32(D + i + 4, vmulq_f32(vcvtq_f32_s32(v_s01), v_scale));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif
@ -995,16 +1049,19 @@ struct ColumnSum<int, float> :
}
}
#elif CV_NEON
for( ; i <= width-8; i+=8 )
if(haveNEON)
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
for( ; i <= width-8; i+=8 )
{
int32x4_t v_s0 = vaddq_s32(vld1q_s32(SUM + i), vld1q_s32(Sp + i));
int32x4_t v_s01 = vaddq_s32(vld1q_s32(SUM + i + 4), vld1q_s32(Sp + i + 4));
vst1q_f32(D + i, vcvtq_f32_s32(v_s0));
vst1q_f32(D + i + 4, vcvtq_f32_s32(v_s01));
vst1q_f32(D + i, vcvtq_f32_s32(v_s0));
vst1q_f32(D + i + 4, vcvtq_f32_s32(v_s01));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
vst1q_s32(SUM + i, vsubq_s32(v_s0, vld1q_s32(Sm + i)));
vst1q_s32(SUM + i + 4, vsubq_s32(v_s01, vld1q_s32(Sm + i + 4)));
}
}
#endif