关于矩阵变化的3种做法（2）

后来想到了一维的方式,除内核与上面那方式不同外，其他的都相同：

内核如下;

const char *KernelSource=

"__kernel void add(__global int *outputB,__global int *inputA)"

"{"

"    int i=get_global_id(0);"

"    if(i<=3||(i>=8&&i<=11)||(i>=16&&i<=19)||(i>=24&&i<=27))"

"      outputB[i]=inputA[i]+1;"

"    else if((i>=4&&i<=7)||(i>=12&&i<=15)||(i>=20&&i<=23)||(i>=28&&i<=31))"

"      outputB[i]=inputA[i]+2;"

"    else if((i>=32&&i<=35)||(i>=40&&i<=43)||(i>=48&&i<=51)||(i>=56&&i<=59))"

"      outputB[i]=inputA[i]+3;"

"    else"

"      outputB[i]=inputA[i]+4;"

"};";

这是通过观察需要变换的外置间的关系写的，但感觉这样做效率就体现不出来，因为感觉判断语句的内容太长了，感觉不是那么好，那能不能从二维的角度来看看呢？

于是我就换了种思路，就有了下面这种：

const char *KernelSource=

"__kernel void add(__global int *outputB,__global int *inputA)"

"{"

"    int i=get_global_id(0);"

"    int j=get_global_id(1);"

"    if(i<4&&j<4)"

"       outputB[i*8+j]=inputA[i*8+j]+1;"

"    else if(i<4&&j>=4)"

"       outputB[i*8+j]=inputA[i*8+j]+2;"

"    else if(i>=4&&j<4)"

"       outputB[i*8+j]=inputA[i*8+j]+3;"

"    else "

"       outputB[i*8+j]=inputA[i*8+j]+4;"

"}";

虽然分支跟上面类似但写法上感觉简练些了，

整个源代码如下：

#include<stdio.h>

#include<stdlib.h>

#include<CL/cl.hpp>

//矩阵变换

const char *KernelSource=

"__kernel void add(__global int *outputB,__global int *inputA)"

"{"

"    int i=get_global_id(0);"

"    int j=get_global_id(1);"

"    if(i<4&&j<4)"

"       outputB[i*8+j]=inputA[i*8+j]+1;"

"    else if(i<4&&j>=4)"

"       outputB[i*8+j]=inputA[i*8+j]+2;"

"    else if(i>=4&&j<4)"

"       outputB[i*8+j]=inputA[i*8+j]+3;"

"    else "

"       outputB[i*8+j]=inputA[i*8+j]+4;"

"}";

int main()

{

const int Row=8;

const int Line=8;

int i,j;

const size_t datasize=sizeof(int)*Row*Line;

int A[Row][Line]=

{

1,1,1,1,2,2,2,2,

1,1,1,1,2,2,2,2,

1,1,1,1,2,2,2,2,

1,1,1,1,2,2,2,2,

3,3,3,3,4,4,4,4,

3,3,3,3,4,4,4,4,

3,3,3,3,4,4,4,4,

3,3,3,3,4,4,4,4

};

int B[Row][Line];

cl_int err;

cl_platform_id platform;

err=clGetPlatformIDs(1,&platform,NULL);

cl_device_id device;

err=clGetDeviceIDs(platform,CL_DEVICE_TYPE_GPU,1,&device,NULL);

cl_context context;

context=clCreateContext(NULL,1,&device,NULL,NULL,&err);

cl_command_queue cmdQueue;

cmdQueue=clCreateCommandQueue(context,device,NULL,&err);

cl_mem BufferA;

cl_mem BufferB;

BufferA=clCreateBuffer(context,CL_MEM_READ_ONLY,datasize,NULL,&err);

BufferB=clCreateBuffer(context,CL_MEM_WRITE_ONLY,datasize,NULL,&err);

err=clEnqueueWriteBuffer(cmdQueue,BufferA,CL_TRUE,0,datasize,A,0,NULL,NULL);

cl_program program;

program=clCreateProgramWithSource(context,1,&KernelSource,NULL,&err);

err=clBuildProgram(program,1,&device,NULL,NULL,NULL);

cl_kernel kernel;

kernel=clCreateKernel(program,"add",&err);

err=clSetKernelArg(kernel,0,sizeof(cl_mem),&BufferB);

err=clSetKernelArg(kernel,1,sizeof(cl_mem),&BufferA);

size_t globalWorkSize[2]={8,8};

err=clEnqueueNDRangeKernel(cmdQueue,kernel,2,NULL,globalWorkSize,NULL,0,NULL,NULL);

err=clEnqueueReadBuffer(cmdQueue,BufferB,CL_TRUE,0,datasize,B,0,NULL,NULL);

for(i=0;i<Row;i++)

{

for(j=0;j<Line;j++)

{

printf("%d ",B[i][j]);

}

printf("\n");

}

system("pause");

}