Skip to content

O
opencv
Commit 92a3dbe1 authored by chacha21's avatar chacha21
Browse files
Options

more ICV_HLINE optimization 

added 64b optimization for 3 channels case
not added 64b optimization for 4 channels case since timings did not
show any improvement
split ICV_HLINE cases into inline functions instead of macro for code
size reduction, without significand speed drawback at first sight
parent afbcc071
Hide whitespace changes
Inline Side-by-side
Showing with 226 additions and 9 deletions
modules/imgproc/src/drawing.cpp
View file @ 92a3dbe1
......@@ -1097,6 +1097,7 @@ static const int opencvOne = 1;
# include <intrin.h>
# pragma intrinsic(_byteswap_ushort)
# pragma intrinsic(_byteswap_ulong)
# pragma intrinsic(_byteswap_uint64)
# pragma intrinsic(_ReadWriteBarrier)
# else
# include <cmnintrin.h>
......@@ -1134,20 +1135,20 @@ static inline uint32_t opencvBigToHost32(uint32_t x){
static inline uint32_t opencvLittleToHost32(const uchar* p){
#if OPENCV_BYTEORDER==1234
uint32_t x;
memcpy(&x,p,4);
memcpy(&x,p,sizeof(x));
return x;
#elif OPENCV_BYTEORDER==4321 && defined(__GNUC__)
uint32_t x;
memcpy(&x,p,4);
memcpy(&x,p,sizeof(x));
return __builtin_bswap32(x);
#elif OPENCV_BYTEORDER==4321 && defined(_MSC_VER) && _MSC_VER>=1300
uint32_t x;
memcpy(&x,p,4);
memcpy(&x,p,sizeof(x));
return _byteswap_ulong(x);
#elif OPENCV_LITTLEENDIAN
return x;
#else
return ((unsigned)p[0]<<24) | (p[1]<<16) | (p[2]<<8) | p[3];
return (p[0]<<24) | (p[1]<<16) | (p[2]<<8) | p[3];
#endif
}
......@@ -1159,7 +1160,33 @@ static inline uint32_t opencvLittleToHost32(uint32_t x){
#endif
}
static inline uint64_t opencvLittleToHost64(const uchar* p){
#if OPENCV_BYTEORDER==1234
uint64_t x;
memcpy(&x,p,sizeof(x));
return x;
#elif OPENCV_BYTEORDER==4321 && defined(__GNUC__)
uint64_t x;
memcpy(&x,p,sizeof(x));
return __builtin_bswap64(x);
#elif OPENCV_BYTEORDER==4321 && defined(_MSC_VER) && _MSC_VER>=1300
uint64_t x;
memcpy(&x,p,sizeof(x));
return _byteswap_uint64(x);
#elif OPENCV_LITTLEENDIAN
return x;
#else
return (p[0]<<56) | (p[1]<<40) | (p[2]<<24) | (p[3]<<8) | (p[4]>>8) | (p[5]>>24) | (p[6]>>40) | (p[7]>>56);
#endif
}
static inline uint64_t opencvLittleToHost64(uint64_t x){
#if OPENCV_LITTLEENDIAN
return x;
#else
return opencvLittleToHost64((uchar*)&x);
#endif
}
/* helper macros: filling horizontal row */
#define is_aligned(POINTER, BYTE_COUNT) (((uintptr_t)(const void *)(POINTER)) % (BYTE_COUNT) == 0)
......@@ -1179,6 +1206,38 @@ static inline uint32_t opencvLittleToHost32(uint32_t x){
} \
}*/
/*
template <unsigned pix_size_forced>
static inline void icv_hline_impl(uchar* ptr, size_t xl, size_t xr, const uchar* color, unsigned pix_size_)
{
const unsigned pix_size = pix_size_forced ? pix_size_forced : pix_size_;
uchar* hline_ptr = ptr + xl*pix_size;
uchar* hline_max_ptr = ptr + xr*pix_size;
for ( ; hline_ptr <= hline_max_ptr; hline_ptr += pix_size)
{
for (unsigned c = 0; c < pix_size; c++)
{
hline_ptr[c] = color[c];
}
}
}
#define ICV_HLINE( ptr, xl, xr, color, pix_size ) \
{ \
if (pix_size == 1) \
icv_hline_impl<1>((uchar*)ptr, (xl), (xr), (const uchar*)color,pix_size); \
else if (pix_size == 3) \
icv_hline_impl<3>((uchar*)ptr, (xl), (xr), (const uchar*)color, pix_size); \
else if (pix_size == 4) \
icv_hline_impl<4>((uchar*)ptr, (xl), (xr), (const uchar*)color, pix_size); \
else \
icv_hline_impl<0>((uchar*)ptr, (xl), (xr), (const uchar*)color, pix_size); \
}
*/
/*
#define ICV_HLINE( ptr, xl, xr, color, pix_size ) \
if((pix_size) == 1) \
{ \
......@@ -1192,9 +1251,36 @@ else if((pix_size) == 3) \
{ \
uchar* hline_ptr = (uchar*)(ptr) + (xl)*3; \
uchar* hline_end = (uchar*)(ptr) + (xr+1)*3; \
uchar* hbody24_start = std::min(hline_end, (uchar*)(24*(((uintptr_t)(hline_ptr)+23)/24))); \
uchar* hbody24_end = std::min(hline_end, (uchar*)(24*(((uintptr_t)(hline_end))/24))); \
uchar* hbody12_start = std::min(hline_end, (uchar*)(12*(((uintptr_t)(hline_ptr)+11)/12))); \
uchar* hbody12_end = std::min(hline_end, (uchar*)(12*(((uintptr_t)(hline_end))/12))); \
if ((hbody12_start < hbody12_end)) \
uchar* hbody12_end = std::min(hline_end, (uchar*)(12*(((uintptr_t)(hline_end))/12))); \
if (hbody24_start < hbody24_end) \
{ \
int offset = ((uintptr_t)(hbody24_start-hline_ptr))%3; \
uint64_t c4[3]; \
uchar* ptrC4 = reinterpret_cast<uchar*>(&c4); \
ptrC4[0] = ((uchar*)(color))[(offset++)%3]; \
ptrC4[1] = ((uchar*)(color))[(offset++)%3]; \
ptrC4[2] = ((uchar*)(color))[(offset++)%3]; \
memcpy(&ptrC4[3], &ptrC4[0], 3); \
memcpy(&ptrC4[6], &ptrC4[0], 6); \
memcpy(&ptrC4[12], &ptrC4[0], 12); \
c4[0] = opencvLittleToHost64(c4[0]); \
c4[1] = opencvLittleToHost64(c4[1]); \
c4[2] = opencvLittleToHost64(c4[2]); \
for(offset = 0 ; hline_ptr < hbody24_start; offset = (offset+1)%3)\
*hline_ptr++ = ((uchar*)(color))[offset]; \
for(uint64_t* ptr64 = reinterpret_cast<uint64_t*>(hbody24_start), *ptr64End = reinterpret_cast<uint64_t*>(hbody24_end) ; ptr64<ptr64End ; ) \
{ \
*ptr64++ = c4[0]; \
*ptr64++ = c4[1]; \
*ptr64++ = c4[2]; \
} \
for(offset = ((uintptr_t)(hbody24_end-(uchar*)(ptr)))%3, hline_ptr = hbody24_end ; hline_ptr < hline_end ; offset = (offset+1)%3) \
*hline_ptr++ = ((uchar*)(color))[offset]; \
} \
else if (hbody12_start < hbody12_end) \
{ \
int offset = ((uintptr_t)(hbody12_start-hline_ptr))%3; \
uint32_t c4[3]; \
......@@ -1233,9 +1319,8 @@ else if(((pix_size) == 4) && is_aligned(((uchar*)(ptr) + (xl)*4), 0x4)) \
uint32_t c = opencvLittleToHost32((uchar*)(color)); \
uint32_t* hline_ptr = (uint32_t*)(ptr) + xl; \
uint32_t* hline_max_ptr = (uint32_t*)(ptr) + xr; \
\
for( ; hline_ptr <= hline_max_ptr; ++hline_ptr ) \
*hline_ptr = c; \
for( ; hline_ptr <= hline_max_ptr; ) \
*hline_ptr++ = c; \
} \
else \
{ \
......@@ -1251,6 +1336,138 @@ else \
} \
} \
}
*/
static inline void ICV_HLINE_0(uchar* ptr, int xl, int xr, const uchar* color, int pix_size)
{
uchar* hline_ptr = (uchar*)(ptr) + (xl)*(pix_size);
uchar* hline_max_ptr = (uchar*)(ptr) + (xr)*(pix_size);
for( ; hline_ptr <= hline_max_ptr; hline_ptr += (pix_size))
{
int hline_j;
for( hline_j = 0; hline_j < (4); hline_j++ )
{
hline_ptr[hline_j] = ((uchar*)color)[hline_j];
}
}
}
//end ICV_HLINE_0()
static inline void ICV_HLINE_1(uchar* ptr, int xl, int xr, const uchar* color)
{
uchar* hline_ptr = (uchar*)(ptr) + (xl);
uchar* hline_max_ptr = (uchar*)(ptr) + (xr);
uchar hline_c = *(const uchar*)(color);
memset(hline_ptr, hline_c, (hline_max_ptr - hline_ptr) + 1);
}
static inline void ICV_HLINE_3(uchar* ptr, int xl, int xr, const uchar* color)
{
uchar* hline_ptr = (uchar*)(ptr) + (xl)*3;
uchar* hline_end = (uchar*)(ptr) + (xr+1)*3;
uchar* hbody24_start = std::min(hline_end, (uchar*)(24*(((uintptr_t)(hline_ptr)+23)/24)));
uchar* hbody24_end = std::min(hline_end, (uchar*)(24*(((uintptr_t)(hline_end))/24)));
uchar* hbody12_start = std::min(hline_end, (uchar*)(12*(((uintptr_t)(hline_ptr)+11)/12)));
uchar* hbody12_end = std::min(hline_end, (uchar*)(12*(((uintptr_t)(hline_end))/12)));
if (hbody24_start < hbody24_end)
{
int offset = ((uintptr_t)(hbody24_start-hline_ptr))%3;
uint64_t c4[3];
uchar* ptrC4 = reinterpret_cast<uchar*>(&c4);
ptrC4[0] = ((uchar*)(color))[(offset++)%3];
ptrC4[1] = ((uchar*)(color))[(offset++)%3];
ptrC4[2] = ((uchar*)(color))[(offset++)%3];
memcpy(&ptrC4[3], &ptrC4[0], 3);
memcpy(&ptrC4[6], &ptrC4[0], 6);
memcpy(&ptrC4[12], &ptrC4[0], 12);
c4[0] = opencvLittleToHost64(c4[0]);
c4[1] = opencvLittleToHost64(c4[1]);
c4[2] = opencvLittleToHost64(c4[2]);
for(offset = 0 ; hline_ptr < hbody24_start; offset = (offset+1)%3)
*hline_ptr++ = ((uchar*)(color))[offset];
for(uint64_t* ptr64 = reinterpret_cast<uint64_t*>(hbody24_start), *ptr64End = reinterpret_cast<uint64_t*>(hbody24_end) ; ptr64<ptr64End ; )
{
*ptr64++ = c4[0];
*ptr64++ = c4[1];
*ptr64++ = c4[2];
}
for(offset = ((uintptr_t)(hbody24_end-(uchar*)(ptr)))%3, hline_ptr = hbody24_end ; hline_ptr < hline_end ; offset = (offset+1)%3)
*hline_ptr++ = ((uchar*)(color))[offset];
}
else if (hbody12_start < hbody12_end)
{
int offset = ((uintptr_t)(hbody12_start-hline_ptr))%3;
uint32_t c4[3];
uchar* ptrC4 = reinterpret_cast<uchar*>(&c4);
ptrC4[0] = ((uchar*)(color))[(offset++)%3];
ptrC4[1] = ((uchar*)(color))[(offset++)%3];
ptrC4[2] = ((uchar*)(color))[(offset++)%3];
memcpy(&ptrC4[3], &ptrC4[0], 3);
memcpy(&ptrC4[6], &ptrC4[0], 6);
c4[0] = opencvLittleToHost32(c4[0]);
c4[1] = opencvLittleToHost32(c4[1]);
c4[2] = opencvLittleToHost32(c4[2]);
for(offset = 0 ; hline_ptr < hbody12_start; offset = (offset+1)%3)
*hline_ptr++ = ((uchar*)(color))[offset];
for(uint32_t* ptr32 = reinterpret_cast<uint32_t*>(hbody12_start), *ptr32End = reinterpret_cast<uint32_t*>(hbody12_end) ; ptr32<ptr32End ; )
{
*ptr32++ = c4[0];
*ptr32++ = c4[1];
*ptr32++ = c4[2];
}
for(offset = ((uintptr_t)(hbody12_end-(uchar*)(ptr)))%3, hline_ptr = hbody12_end ; hline_ptr < hline_end ; offset = (offset+1)%3)
*hline_ptr++ = ((uchar*)(color))[offset];
}
else
{
for( ; hline_ptr < hline_end ; )
{
*hline_ptr++ = ((uchar*)(color))[0];
*hline_ptr++ = ((uchar*)(color))[1];
*hline_ptr++ = ((uchar*)(color))[2];
}
}
}
//end ICV_HLINE_3()
static inline void ICV_HLINE_4(uchar* ptr, int xl, int xr, const uchar* color)
{
if (is_aligned(((uchar*)(ptr) + (xl)*4), 0x4))
{
uint32_t c = opencvLittleToHost32((uchar*)(color));
uint32_t* hline_ptr = (uint32_t*)(ptr) + xl;
uint32_t* hline_max_ptr = (uint32_t*)(ptr) + xr;
for( ; hline_ptr <= hline_max_ptr; )
*hline_ptr++ = c;
}
else
{
uchar* hline_ptr = (uchar*)(ptr) + (xl)*(4);
uchar* hline_max_ptr = (uchar*)(ptr) + (xr)*(4);
for( ; hline_ptr <= hline_max_ptr; hline_ptr += (4))
{
int hline_j;
for( hline_j = 0; hline_j < (4); hline_j++ )
{
hline_ptr[hline_j] = ((uchar*)color)[hline_j];
}
}
}
}
//end ICV_HLINE_4()
static inline void ICV_HLINE(uchar* ptr, int xl, int xr, const void* color, int pix_size)
{
if (pix_size == 1)
ICV_HLINE_1(ptr, xl, xr, reinterpret_cast<const uchar*>(color));
else if (pix_size == 3)
ICV_HLINE_3(ptr, xl, xr, reinterpret_cast<const uchar*>(color));
else if (pix_size == 4)
ICV_HLINE_4(ptr, xl, xr, reinterpret_cast<const uchar*>(color));
else
ICV_HLINE_0(ptr, xl, xr, reinterpret_cast<const uchar*>(color), pix_size);
}
//end ICV_HLINE()
/* filling convex polygon. v - array of vertices, ntps - number of points */
static void
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment