opencl gauss filter优化(一)

时间:2022-03-22 11:44:54

Platform: LG G3, Adreno 330 ,img size 3264x2448

C code

neon

GPU

300

60

29

单位:ms

1.
目前按如下行列分解的方式最快29ms,Horizontal
kernel
globalWorksize[1] = {height+256-height%256};Vertical kernel
globalWorksize2[1] = {width+256-width%256};

localWorksize2[]
= {64}; localWorksize2 手动设为64时最快。

Porfile的结果为:Horizontal
kernel 的wait
time 有11ms,实际rum
time 18ms.

这个wait
time是什么呢?注释掉Horizontal
kernel中的
vstore16(convert_uchar16(sum>>(ushort)8),0,pOutLine+j)
; 则wait
time只有0.x
ms.并且
localWorksize
越小wait
time越长,为1时达到200ms,16时20ms.
难道是写内存等待时间,没有足够的ALU指令隐藏访存延时?写内存后进入下一个for循环,马上又读内存,所以没有ALU指令隐藏这个延时。然而Horizontal
kernel的profile结果实际run
time只有0.x
ms,所有时间基本都是在wait.(更正:注释掉vstore16后,sum的计算被优化掉了,0.x
ms是读内存的时间)

__kernel void ImageGaussianFilterHorizontal(__global const uchar* restrict source, // Source image
__global uchar* restrict dest, // Intermediate dest image
const int imgWidth , // Image width
const int imgHeight)
{
const int y = get_global_id();
if(y>=(imgHeight))
return;
const uchar m_nRightShiftNum = ;
const uchar Rounding = ( << (m_nRightShiftNum - ));
const uchar m_nFilter[] = {,,,,,,,,,,}; const int s = ;
const int nStart = ;
const int nWidth = imgWidth; __global const uchar* pInLine = source + y*nWidth;
__global uchar* pOutLine = dest + y*nWidth;
int j;
for(j = ; j < nStart; j ++)
{
ushort sum = ; for (int m = ; m<s / ; m++)
{
int k1 = (j + m - nStart);
k1 = k1< ? -k1 : k1; int k2 = (j + nStart - m );
sum += (pInLine[k1] + pInLine[k2])*m_nFilter[m];
}
sum += pInLine[j] * m_nFilter[s / ];
sum = (sum + Rounding) >> ;
pOutLine[j] = (uchar)clamp(sum,(ushort),(ushort));
} for ( ; (j+)<= (nWidth - nStart); j+=)
{
#define GAUSSIAN_LINE_NEON(m) \
sum += ( convert_ushort16(vload16(,pInLine+j-nStart+m))* m_nFilter[m] ); ushort16 sum = (convert_ushort16(vload16(,pInLine+j-nStart)) * m_nFilter[]);
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON();
GAUSSIAN_LINE_NEON(); sum += (ushort)Rounding;
vstore16(convert_uchar16(sum>>(ushort)),,pOutLine+j) ;
} for( ; j < nWidth; j ++)
{
ushort sum = ; for (int m = ; m<s / ; m++)
{
int k1 = (j + m - nStart); int k2 = (j + nStart - m );
k2 = k2 >= nWidth ? * nWidth - - k2 : k2;
sum += (pInLine[k1] + pInLine[k2])*m_nFilter[m];
}
sum += pInLine[j] * m_nFilter[s / ];
sum = (sum + Rounding) >> m_nRightShiftNum;
pOutLine[j] = (uchar)clamp(sum,(ushort),(ushort));
}
} __kernel void ImageGaussianFilterVertical( __global uchar* restrict source, // Intermediate image processed by ImageGaussianFilterHorizontal()
__global uchar* restrict dest, // Final destination image
const int imgWidth,
const int imgHeight
)
{
const int x = get_global_id();
if(x>=(imgWidth))
return;
const int x_offset = x; const int s = ;
const int nStart = s / ;
const int m_nRightShiftNum = ;
const int Rounding = ( << (m_nRightShiftNum - ));
const uchar m_nFilter[] = {,,,,,,,,,,}; int y;
// mem_fence(CLK_LOCAL_MEM_FENCE); ushort lines[];
lines[nStart] = (ushort)( source[x_offset] );
for(y=;y<=nStart;y++)
{
lines[nStart+y] = (ushort)( source[y*imgWidth+x_offset] );
lines[nStart-y] = lines[nStart+y];
} for(y=;y<(imgHeight-nStart-);)
{ ushort sum = lines[nStart] * m_nFilter[nStart];
#define GaussianTwoLines(m) \
sum += ( (lines[m] + lines[s--m])*m_nFilter[m] );
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines() sum += (ushort)Rounding;
dest[y*imgWidth+x_offset] = (uchar)(sum>>(ushort)); y++;
for(int i = ; i<s-; i++) lines[i] = lines[i+]; lines[s-] = (ushort)( source[(y+nStart)*imgWidth+x_offset] ); } for(y=imgHeight-nStart-;y<(imgHeight-);)
{
ushort sum = lines[nStart] * m_nFilter[nStart];
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines() sum += (ushort)Rounding;
dest[y*imgWidth+x_offset] = (uchar)(sum>>(ushort)); y++;
for(int i = ; i<s-; i++) {
lines[i] = lines[i+];
}
lines[s-] = lines[(imgHeight-y)*-] ; //
}
//last y=imgHeight-1
ushort sum = lines[nStart] * m_nFilter[nStart];
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines()
GaussianTwoLines() sum += (ushort)Rounding;
dest[y*imgWidth+x_offset] = (uchar)(sum>>(ushort));
}

kernel

2.Horizontal kernel改进,预先load 2x16个所需的pixel,计算时从中提取,这样每次循环只需读一次内存。需要26ms,wait time 8ms.

    ushort16 line0 =  convert_ushort16(vload16(,pInLine+j-nStart));
for ( ; (j+)<= (nWidth - nStart); j+=)
{
ushort16 line1 = convert_ushort16(vload16(,pInLine+j-nStart+)); ushort16 temp0;
ushort16 temp1;
temp0 = line0;
temp1.s0123 = line0.sabcd;
temp1.s45 = line0.sef;
temp1.s67 = line1.s01;
temp1.s89abcdef = line1.s23456789;
ushort16 sum = ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s0;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s9;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s1;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s8;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s2;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s7;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s3;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s6;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s4;
sum += ( temp0 ) * m_nFilter[]; sum += (ushort)Rounding;
line0 = line1;
vstore16(convert_uchar16(sum>>(ushort)),,pOutLine+j) ;
}

3.不计算,只读写内存测试。那么wait time 3.2 ms,run time 18.2 ms.说明Horizontal kernel 耗时的极限也需3.2ms. 但是只是注释掉vstore16,还保留了读和计算,反而wait time还只有0.x ms,这又是为何?是读几乎没有wait,3.2ms都是写的wait time? (更正:注释掉vstore16后,sum的计算被优化掉了,0.x ms是读内存的时间)

a.再次测试,只有读wait time 0.xms ,只有写wait time 3.2ms.写比读的周期长.

for ( ; (j+16)<= (nWidth - nStart); j+=16)

{

ushort16 line1 = convert_ushort16(vload16(0,pInLine+j-nStart+16));

vstore16(0,0,pOutLine+j) ;

}

b.另外发现使用*((__global uint4*)(pOutLine+j)) = as_uint4(result);比vstore16快,wait time 2.5ms.高通 80-N8592-1_L_OpenCL_Programming_Guide 中提到:

Vectorized load/store of a larger data type is more optimal than a small data type; e.g., a load of uint2* is more optimal than uchar8* .

For optimal SP to L2 bandwidth performance, align read access to a 32-bit address and write access to a 128-bit address.

c.原来写的内存没有对齐,使用*((__global uint4*)(pOutLine+j-5)) = as_uint4(result);wait time 1.9ms.

d.最后加上sum计算,采用的Horizontal kernel如下,localWorksize[] = {64};时时间最少,需要23ms,wait time 4.7ms , localWorksize = 128时,wait 6ms.

并且使用__attribute__((work_group_size_hint(64,1,1))) ,耗时22ms.

__kernel __attribute__((work_group_size_hint(,,)))
void ImageGaussianFilterHorizontal(__global const uchar* restrict source, // Source image
__global uchar* restrict dest, // Intermediate dest image
const int imgWidth , // Image width
const int imgHeight)
{
const int y = get_global_id();
if(y>=(imgHeight))
return;
const uchar m_nRightShiftNum = ;
const uchar Rounding = ( << (m_nRightShiftNum - ));
const uchar m_nFilter[] = {,,,,,,,,,,}; const int s = ;
const int nStart = ;
const int nWidth = imgWidth; __global const uchar* pInLine = source + y*nWidth;
__global uchar* pOutLine = dest + y*nWidth; int j;
uchar temp[];
for(j = ; j < nStart; j ++)
{
ushort sum = ; for (int m = ; m<s / ; m++)
{
int k1 = (j + m - nStart);
k1 = k1< ? -k1 : k1; int k2 = (j + nStart - m );
sum += (pInLine[k1] + pInLine[k2])*m_nFilter[m];
}
sum += pInLine[j] * m_nFilter[s / ];
sum = (sum + Rounding) >> ;
temp[j] = (uchar)clamp(sum,(ushort),(ushort));
} uchar16 result,pre_result;
pre_result.sbcde = (uchar4)(temp[],temp[],temp[],temp[]);
pre_result.sf = temp[]; ushort16 line0 = convert_ushort16(vload16(,pInLine+j-nStart));
for ( ; (j+)<= (nWidth - nStart); j+=)
{
//prefetch(pInLine+j-nStart,32); //无变化
ushort16 line1 = convert_ushort16(vload16(,pInLine+j-nStart+)); ushort16 temp0;
ushort16 temp1;
temp0 = line0;
temp1.s0123 = line0.sabcd;
temp1.s45 = line0.sef;
temp1.s67 = line1.s01;
temp1.s89abcdef = line1.s23456789;
ushort16 sum = ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s0;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s9;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s1;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s8;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s2;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s7;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s3;
temp1.s0123456789abcdef = temp1.s00123456789abcde;
temp1.s0 = line0.s6;
sum += ( temp0 + temp1 ) * m_nFilter[];
temp0.s0123456789abcdef = temp0.s123456789abcdeff;
temp0.sf = line1.s4;
sum += ( temp0 ) * m_nFilter[]; sum += (ushort)Rounding;
line0 = line1; result.s0123 = pre_result.sbcde;
result.s4 = pre_result.sf;
pre_result = convert_uchar16(sum>>(ushort)) ; result.s5 = pre_result.s0;
result.s67 = pre_result.s12;
result.s89abcdef = pre_result.s3456789a;
*( (__global uint4*)(pOutLine+j-) ) = (as_uint4)(result) ;
} *( (__global uint*)(pOutLine+j-) ) = (as_uint)(pre_result.sbcde);//last 5 bytes
pOutLine[j-] = pre_result.sf; for( ; j < nWidth; j ++)
{
ushort sum = ; for (int m = ; m<s / ; m++)
{
int k1 = (j + m - nStart); int k2 = (j + nStart - m );
k2 = k2 >= nWidth ? * nWidth - - k2 : k2;
sum += (pInLine[k1] + pInLine[k2])*m_nFilter[m];
}
sum += pInLine[j] * m_nFilter[s / ];
sum = (sum + Rounding) >> m_nRightShiftNum;
pOutLine[j] = (uchar)clamp(sum,(ushort),(ushort));
}
}