How can I improve this function under CUDA?
Can I improve the following function under CUDA ?
What the function does is,
Given a min and max , ELM1 and ELM , check if any three numbers of array ans[6] are found in any row, from min to max , in array D1 , D2 , D3 , D4 , D5 , D6 , if found return 1.
I tried using loops , OR -ing, AND -ing, replacing goto with flag etc. etc. But this seems to be the fastest way.
__device__ bool THREEA(unsigned int n0, unsigned int n,unsigned int* ST1,unsigned int* D1, unsigned int* D2,unsigned int* D3,unsigned int* D4,unsigned int* D5,unsigned int* D6,unsigned int* ans)
{
unsigned int ELM, ELM1,flag;
ELM = ST1[n0]+n; //local.37
ELM1 = n; //local.33
while (ELM1 < ELM)
{
flag = 0;
if (D1[ELM1] == ans[0])
{
flag++;
}
if (D2[ELM1] == ans[0])
{
flag++;
}
if (D3[ELM1] == ans[0])
{
flag++;
}
if (D4[ELM1] == ans[0])
{
flag++;
}
if (D5[ELM1] == ans[0])
{
flag++;
}
if (D6[ELM1] == ans[0])
{
flag++;
}
if (flag != 1)
goto onethreefour;
if (D1[ELM1] == ans[1])
{
flag++;
}
if (D2[ELM1] == ans[1])
{
flag++;
}
if (D3[ELM1] == ans[1])
{
flag++;
}
if (D4[ELM1] == ans[1])
{
flag++;
}
if (D5[ELM1] == ans[1])
{
flag++;
}
if (D6[ELM1] == ans[1])
{
flag++;
}
if (flag != 2)
goto onethreefour;
if (D1[ELM1] == ans[2])
{
return 1;
}
if (D2[ELM1] == ans[2])
{
return 1;
}
if (D3[ELM1] == ans[2])
{
return 1;
}
if (D4[ELM1] == ans[2])
{
return 1;
}
if (D5[ELM1] == ans[2])
{
return 1;
}
if (D6[ELM1] == ans[2])
{
return 1;
}
if (D1[ELM1] == ans[3])
{
return 1;
}
if (D2[ELM1] == ans[3])
{
return 1;
}
if (D3[ELM1] == ans[3])
{
return 1;
}
if (D4[ELM1] == ans[3])
{
return 1;
}
if (D5[ELM1] == ans[3])
{
return 1;
}
if (D6[ELM1] == ans[3])
{
return 1;
}
if (D1[ELM1] == ans[4])
{
return 1;
}
if (D2[ELM1] == ans[4])
{
return 1;
}
if (D3[ELM1] == ans[4])
{
return 1;
}
if (D4[ELM1] == ans[4])
{
return 1;
}
if (D5[ELM1] == ans[4])
{
return 1;
}
if (D6[ELM1] == ans[4])
{
return 1;
}
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
onethreefour:
flag = 0;
if (D1[ELM1] == ans[0])
{
flag++;
}
if (D2[ELM1] == ans[0])
{
flag++;
}
if (D3[ELM1] == ans[0])
{
flag++;
}
if (D4[ELM1] == ans[0])
{
flag++;
}
if (D5[ELM1] == ans[0])
{
flag++;
}
if (D6[ELM1] == ans[0])
{
flag++;
}
if (flag != 1)
goto onefourfive;
if (D1[ELM1] == ans[2])
{
flag++;
}
if (D2[ELM1] == ans[2])
{
flag++;
}
if (D3[ELM1] == ans[2])
{
flag++;
}
if (D4[ELM1] == ans[2])
{
flag++;
}
if (D5[ELM1] == ans[2])
{
flag++;
}
if (D6[ELM1] == ans[2])
{
flag++;
}
if (flag != 2)
goto onefourfive;
if (D1[ELM1] == ans[3])
{
return 1;
}
if (D2[ELM1] == ans[3])
{
return 1;
}
if (D3[ELM1] == ans[3])
{
return 1;
}
if (D4[ELM1] == ans[3])
{
return 1;
}
if (D5[ELM1] == ans[3])
{
return 1;
}
if (D6[ELM1] == ans[3])
{
return 1;
}
if (D1[ELM1] == ans[4])
{
return 1;
}
if (D2[ELM1] == ans[4])
{
return 1;
}
if (D3[ELM1] == ans[4])
{
return 1;
}
if (D4[ELM1] == ans[4])
{
return 1;
}
if (D5[ELM1] == ans[4])
{
return 1;
}
if (D6[ELM1] == ans[4])
{
return 1;
}
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
onefourfive:
flag = 0;
if (D1[ELM1] == ans[0])
{
flag++;
}
if (D2[ELM1] == ans[0])
{
flag++;
}
if (D3[ELM1] == ans[0])
{
flag++;
}
if (D4[ELM1] == ans[0])
{
flag++;
}
if (D5[ELM1] == ans[0])
{
flag++;
}
if (D6[ELM1] == ans[0])
{
flag++;
}
if (flag != 1)
goto onefivesix;
if (D1[ELM1] == ans[3])
{
flag++;
}
if (D2[ELM1] == ans[3])
{
flag++;
}
if (D3[ELM1] == ans[3])
{
flag++;
}
if (D4[ELM1] == ans[3])
{
flag++;
}
if (D5[ELM1] == ans[3])
{
flag++;
}
if (D6[ELM1] == ans[3])
{
flag++;
}
if (flag != 2)
goto onefivesix;
if (D1[ELM1] == ans[4])
{
return 1;
}
if (D2[ELM1] == ans[4])
{
return 1;
}
if (D3[ELM1] == ans[4])
{
return 1;
}
if (D4[ELM1] == ans[4])
{
return 1;
}
if (D5[ELM1] == ans[4])
{
return 1;
}
if (D6[ELM1] == ans[4])
{
return 1;
}
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
onefivesix:
flag = 0;
if (D1[ELM1] == ans[0])
{
flag++;
}
if (D2[ELM1] == ans[0])
{
flag++;
}
if (D3[ELM1] == ans[0])
{
flag++;
}
if (D4[ELM1] == ans[0])
{
flag++;
}
if (D5[ELM1] == ans[0])
{
flag++;
}
if (D6[ELM1] == ans[0])
{
flag++;
}
if (flag != 1)
goto twothreefour;
if (D1[ELM1] == ans[4])
{
flag++;
}
if (D2[ELM1] == ans[4])
{
flag++;
}
if (D3[ELM1] == ans[4])
{
flag++;
}
if (D4[ELM1] == ans[4])
{
flag++;
}
if (D5[ELM1] == ans[4])
{
flag++;
}
if (D6[ELM1] == ans[4])
{
flag++;
}
if (flag != 2)
goto twothreefour;
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
twothreefour:
flag = 0;
if (D1[ELM1] == ans[1])
{
flag++;
}
if (D2[ELM1] == ans[1])
{
flag++;
}
if (D3[ELM1] == ans[1])
{
flag++;
}
if (D4[ELM1] == ans[1])
{
flag++;
}
if (D5[ELM1] == ans[1])
{
flag++;
}
if (D6[ELM1] == ans[1])
{
flag++;
}
if (flag != 1)
goto twofourfive;
if (D1[ELM1] == ans[2])
{
flag++;
}
if (D2[ELM1] == ans[2])
{
flag++;
}
if (D3[ELM1] == ans[2])
{
flag++;
}
if (D4[ELM1] == ans[2])
{
flag++;
}
if (D5[ELM1] == ans[2])
{
flag++;
}
if (D6[ELM1] == ans[2])
{
flag++;
}
if (flag != 2)
goto twofourfive;
if (D1[ELM1] == ans[3])
{
return 1;
}
if (D2[ELM1] == ans[3])
{
return 1;
}
if (D3[ELM1] == ans[3])
{
return 1;
}
if (D4[ELM1] == ans[3])
{
return 1;
}
if (D5[ELM1] == ans[3])
{
return 1;
}
if (D6[ELM1] == ans[3])
{
return 1;
}
if (D1[ELM1] == ans[4])
{
return 1;
}
if (D2[ELM1] == ans[4])
{
return 1;
}
if (D3[ELM1] == ans[4])
{
return 1;
}
if (D4[ELM1] == ans[4])
{
return 1;
}
if (D5[ELM1] == ans[4])
{
return 1;
}
if (D6[ELM1] == ans[4])
{
return 1;
}
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
twofourfive:
flag = 0;
if (D1[ELM1] == ans[1])
{
flag++;
}
if (D2[ELM1] == ans[1])
{
flag++;
}
if (D3[ELM1] == ans[1])
{
flag++;
}
if (D4[ELM1] == ans[1])
{
flag++;
}
if (D5[ELM1] == ans[1])
{
flag++;
}
if (D6[ELM1] == ans[1])
{
flag++;
}
if (flag != 1)
goto twofivesix;
if (D1[ELM1] == ans[3])
{
flag++;
}
if (D2[ELM1] == ans[3])
{
flag++;
}
if (D3[ELM1] == ans[3])
{
flag++;
}
if (D4[ELM1] == ans[3])
{
flag++;
}
if (D5[ELM1] == ans[3])
{
flag++;
}
if (D6[ELM1] == ans[3])
{
flag++;
}
if (flag != 2)
goto twofivesix;
if (D1[ELM1] == ans[4])
{
return 1;
}
if (D2[ELM1] == ans[4])
{
return 1;
}
if (D3[ELM1] == ans[4])
{
return 1;
}
if (D4[ELM1] == ans[4])
{
return 1;
}
if (D5[ELM1] == ans[4])
{
return 1;
}
if (D6[ELM1] == ans[4])
{
return 1;
}
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
twofivesix:
flag = 0;
if (D1[ELM1] == ans[1])
{
flag++;
}
if (D2[ELM1] == ans[1])
{
flag++;
}
if (D3[ELM1] == ans[1])
{
flag++;
}
if (D4[ELM1] == ans[1])
{
flag++;
}
if (D5[ELM1] == ans[1])
{
flag++;
}
if (D6[ELM1] == ans[1])
{
flag++;
}
if (flag != 1)
goto threefourfive;
if (D1[ELM1] == ans[4])
{
flag++;
}
if (D2[ELM1] == ans[4])
{
flag++;
}
if (D3[ELM1] == ans[4])
{
flag++;
}
if (D4[ELM1] == ans[4])
{
flag++;
}
if (D5[ELM1] == ans[4])
{
flag++;
}
if (D6[ELM1] == ans[4])
{
flag++;
}
if (flag != 2)
goto threefourfive;
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
threefourfive:
flag = 0;
if (D1[ELM1] == ans[2])
{
flag++;
}
if (D2[ELM1] == ans[2])
{
flag++;
}
if (D3[ELM1] == ans[2])
{
flag++;
}
if (D4[ELM1] == ans[2])
{
flag++;
}
if (D5[ELM1] == ans[2])
{
flag++;
}
if (D6[ELM1] == ans[2])
{
flag++;
}
if (flag != 1)
goto threefivesix;
if (D1[ELM1] == ans[3])
{
flag++;
}
if (D2[ELM1] == ans[3])
{
flag++;
}
if (D3[ELM1] == ans[3])
{
flag++;
}
if (D4[ELM1] == ans[3])
{
flag++;
}
if (D5[ELM1] == ans[3])
{
flag++;
}
if (D6[ELM1] == ans[3])
{
flag++;
}
if (flag != 2)
goto threefivesix;
if (D1[ELM1] == ans[4])
{
return 1;
}
if (D2[ELM1] == ans[4])
{
return 1;
}
if (D3[ELM1] == ans[4])
{
return 1;
}
if (D4[ELM1] == ans[4])
{
return 1;
}
if (D5[ELM1] == ans[4])
{
return 1;
}
if (D6[ELM1] == ans[4])
{
return 1;
}
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
threefivesix:
flag = 0;
if (D1[ELM1] == ans[2])
{
flag++;
}
if (D2[ELM1] == ans[2])
{
flag++;
}
if (D3[ELM1] == ans[2])
{
flag++;
}
if (D4[ELM1] == ans[2])
{
flag++;
}
if (D5[ELM1] == ans[2])
{
flag++;
}
if (D6[ELM1] == ans[2])
{
flag++;
}
if (flag != 1)
goto fourfivesix;
if (D1[ELM1] == ans[4])
{
flag++;
}
if (D2[ELM1] == ans[4])
{
flag++;
}
if (D3[ELM1] == ans[4])
{
flag++;
}
if (D4[ELM1] == ans[4])
{
flag++;
}
if (D5[ELM1] == ans[4])
{
flag++;
}
if (D6[ELM1] == ans[4])
{
flag++;
}
if (flag != 2)
goto fourfivesix;
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
fourfivesix:
flag = 0;
if (D1[ELM1] == ans[3])
{
flag++;
}
if (D2[ELM1] == ans[3])
{
flag++;
}
if (D3[ELM1] == ans[3])
{
flag++;
}
if (D4[ELM1] == ans[3])
{
flag++;
}
if (D5[ELM1] == ans[3])
{
flag++;
}
if (D6[ELM1] == ans[3])
{
flag++;
}
if (flag != 1)
goto increasecounter;
if (D1[ELM1] == ans[4])
{
flag++;
}
if (D2[ELM1] == ans[4])
{
flag++;
}
if (D3[ELM1] == ans[4])
{
flag++;
}
if (D4[ELM1] == ans[4])
{
flag++;
}
if (D5[ELM1] == ans[4])
{
flag++;
}
if (D6[ELM1] == ans[4])
{
flag++;
}
if (flag != 2)
goto increasecounter;
if (D1[ELM1] == ans[5])
{
return 1;
}
if (D2[ELM1] == ans[5])
{
return 1;
}
if (D3[ELM1] == ans[5])
{
return 1;
}
if (D4[ELM1] == ans[5])
{
return 1;
}
if (D5[ELM1] == ans[5])
{
return 1;
}
if (D6[ELM1] == ans[5])
{
return 1;
}
increasecounter:
ELM1++;
}
//If it is Three min
return 0;
}
remove if statements by converting them to Boolean expressions.
flag += (DN[ELM1] == ans[0])
make sure your arrays are in registers or shared memory rather than global
also, on such simple algorithm you got way too complicated logic. change layout of D arrays to be D[N][6] as it will simplify many things
by the way, you may want to crop your post a little bit, way too much to read
3 x3 example
A
|0 0 0| |x x 0
D |0 0 0| -> ... -> |x x 0 -> reduce down -> |x x o| -> reduce across -> 2x
|0 0 0| |x x x
basically you are setting matrix cells to true if match A is in array D. on each iteration you reduce columns to true if entire column is true. then you count number of true.
链接地址: http://www.djcxy.com/p/70930.html上一篇: 如何对List <Point>进行排序
下一篇: 我如何在CUDA下改进此功能？