OpenCL 第8課:旋轉變換(2)

上兩節課都是對一個數組進行處理。這節我們來個有意思的。同樣是旋轉。但我們旋轉的對象是張（２５６*２５６）的圖片。圖片旋轉４５度，旋轉后大小還是２５６*２５６，超出部份進行剪除。

圖片旋轉處理有個特別的地方。buf_A是存儲源圖數據（R,G,B顏色分量）,buf_B是存儲旋轉后數據。我們不能簡單將buf_A中的數據直接計算旋轉后位置。而是遍歷buf_B每個數據，計算那些數據旋轉后存在這個位置的數據他在buf_A的坐標是多少。為什么呢？不是因為不能直接旋轉。而是因為坐標是整數型。進行旋轉計算后會得出浮點數，有小數。計算精度不同。如果還用直接旋轉計算方法。成出的位圖將會出現斑點。大家可試下。

我們來看源碼

rotate.cl源碼

?
__kernel void rotation(__global int* A,
????????????????????__global int* B,
????????????????????int width,
????????????????????int height,
????????????????????float sinangle,
????????????????????float cosangle)
{
????//獲取索引號，這里是二維的，所以可以取兩個
????//否則另一個永遠是0
????int col = get_global_id(0);
????int row = get_global_id(1);
????//計算圖形中心點
????float cx = ((float)width)/2;
????float cy = ((float)height)/2;
????int nx = (int)(cx + cosangle * ((float)col-cx) + sinangle * ((float)row-cy));
????int ny = (int)(cy + (-1*sinangle) * ((float)col-cx) + cosangle * ((float)row-cy));
????//邊界檢測
????if(nx>=0 && nx<width && ny>=0 && ny<height)
????{
????????B[col*3+0+row*width*3] = A[nx*3+0+ny*width*3];
????????B[col*3+1+row*width*3] = A[nx*3+1+ny*width*3];
????????B[col*3+2+row*width*3] = A[nx*3+2+ny*width*3];
????}
}

main.cpp源碼

?
#include <iostream>
#include <stdio.h>
#include <string.h>
#include <string>
#include <conio.h>
#include <math.h>//數學庫
#include <CL/cl.h>//包含CL的頭文件
//調用freeimage
#include <freeimage.h>
using namespace std;
//8x8數組
const int dim_x = 256;
const int dim_y = 256;
//45度的弧度
const float angle = 3.1415926f/4.0f;
static int buf_A[dim_x*dim_y*3];
static int buf_B[dim_x*dim_y*3];
//加載圖片
//以RGBA格式存儲圖片
static bool LoadImg(const char* fname)
{
????//初始化FreeImage
????FreeImage_Initialise(TRUE);
????//定義圖片格式為未知
????FREE_IMAGE_FORMAT fif = FIF_UNKNOWN;
????//獲取圖片格式
????fif = FreeImage_GetFileType(fname,0);
????//根據獲取格式讀取圖片數據
????FIBITMAP* bitmap = FreeImage_Load(fif,fname,0);
????if(!bitmap)
????{
????????printf("load error!\n");
????????return false;
????}
????int x,y;
????RGBQUAD m_rgb;
????//獲取圖片長寬
????int width = (int)FreeImage_GetWidth(bitmap);
????int height = (int)FreeImage_GetHeight(bitmap);
????//獲取圖片數據
????//按RGBA格式保存到數組中
????for(y=0;y<height;y++)
????{
????????for(x=0;x<width;x++)
????????{
????????????//獲取像素值
????????????FreeImage_GetPixelColor(bitmap,x,y,&m_rgb);
????????????//將RGB值存入數組
????????????buf_A[y*width*3+x*3+2] = m_rgb.rgbRed;
????????????buf_A[y*width*3+x*3+1] = m_rgb.rgbGreen;
????????????buf_A[y*width*3+x*3+0] = m_rgb.rgbBlue;
????????}
????}
????FreeImage_Unload(bitmap);
????return true;
}
static bool SaveImg()
{
????//初始化FreeImage
????FreeImage_Initialise(TRUE);
????FIBITMAP* bitmap =FreeImage_Allocate(dim_x,dim_y,32,8,8,8);
????int m,n;
????for(n=0;n<dim_y;n++)
????{
????????BYTE *bits =FreeImage_GetScanLine(bitmap,n);
????????for(m=0;m<dim_x;m++)
????????{
????????????bits[0] = buf_B[dim_x*3*n+m*3+0];
????????????bits[1] = buf_B[dim_x*3*n+m*3+1];
????????????bits[2] = buf_B[dim_x*3*n+m*3+2];
????????????bits[3] = 255;
????????????bits+=4;
????????}
????}
????//保存圖片為PNG格式
????if(false ==FreeImage_Save(FIF_PNG, bitmap,"rotate.png", PNG_DEFAULT))
????{
????????printf("save image error\n");
????}
????FreeImage_Unload(bitmap);
????return true;
}
//從外部文件獲取cl內核代碼
bool GetFileData(const char* fname,string& str)
{
????FILE* fp = fopen(fname,"r");
????if(fp==NULL)
????{
????????printf("no found file\n");
????????return false;
????}
????while(feof(fp)==0)
????{
????????str += fgetc(fp);
????}
????return true;
}
int main()
{
????if(LoadImg("bk.png")==false)
????{
????????printf("error load bk.png!\n");
????????return 0;
????}
????//先讀外部CL核心代碼，如果失敗則退出。
????//代碼存buf_code里面
????string code_file;
????if(false == GetFileData("rotate.cl",code_file))
????{
????????printf("Open rotate.cl error\n");
????????return 0;
????}
????char* buf_code = new char[code_file.size()];
????strcpy(buf_code,code_file.c_str());
????buf_code[code_file.size()-1] = NULL;
????//聲明CL所需變量。
????cl_device_id device;
????cl_platform_id platform_id = NULL;
????cl_context context;
????cl_command_queue cmdQueue;
????cl_mem bufferA,bufferB;
????cl_program program;
????cl_kernel kernel = NULL;
????//我們使用的是二維向量
????//設定向量大小（維數）
????size_t globalWorkSize[2];
????globalWorkSize[0] = dim_x;
????globalWorkSize[1] = dim_y;
????cl_int err;
????/*
????????定義輸入變量和輸出變量，并設定初值
????*/
????size_t datasize = sizeof(int) * dim_x * dim_y * 3;
????//step 1:初始化OpenCL
????err = clGetPlatformIDs(1,&platform_id,NULL);
????if(err!=CL_SUCCESS)
????{
????????cout<<"clGetPlatformIDs error:"<<err<<endl;
????????return 0;
????}
????//這次我們只用CPU來進行并行運算，當然你也可以該成GPU
????clGetDeviceIDs(platform_id,CL_DEVICE_TYPE_CPU,1,&device,NULL);
????//step 2:創建上下文
????context = clCreateContext(NULL,1,&device,NULL,NULL,NULL);
????//step 3:創建命令隊列
????cmdQueue = clCreateCommandQueue(context,device,0,NULL);
????//step 4:創建數據緩沖區
????bufferA = clCreateBuffer(context,
?????????????????????????????CL_MEM_READ_ONLY,
?????????????????????????????datasize,NULL,NULL);
????bufferB = clCreateBuffer(context,
?????????????????????????????CL_MEM_WRITE_ONLY,
?????????????????????????????datasize,NULL,NULL);
????//step 5:將數據上傳到緩沖區
????clEnqueueWriteBuffer(cmdQueue,
?????????????????????????bufferA,CL_FALSE,
?????????????????????????0,datasize,
?????????????????????????buf_A,0,
?????????????????????????NULL,NULL);
????//step 6:加載編譯代碼,創建內核調用函數
????program = clCreateProgramWithSource(context,1,
????????????????????????????????????????(const char**)&buf_code,
????????????????????????????????????????NULL,NULL);
????clBuildProgram(program,1,&device,NULL,NULL,NULL);
????kernel = clCreateKernel(program,"rotation",NULL);
????//step 7:設置參數，執行內核
????float sinangle = sinf(angle);
????float cosangle = cosf(angle);
????clSetKernelArg(kernel,0,sizeof(cl_mem),&bufferA);
????clSetKernelArg(kernel,1,sizeof(cl_mem),&bufferB);
????clSetKernelArg(kernel,2,sizeof(cl_int),&dim_x);
????clSetKernelArg(kernel,3,sizeof(cl_int),&dim_y);
????clSetKernelArg(kernel,4,sizeof(cl_float),&sinangle);
????clSetKernelArg(kernel,5,sizeof(cl_float),&cosangle);
????//注意這里第三個參數已經改成2，表示二維數據。
????clEnqueueNDRangeKernel(cmdQueue,kernel,
???????????????????????????2,NULL,
???????????????????????????globalWorkSize,
???????????????????????????NULL,0,NULL,NULL);
????//step 8:取回計算結果
????clEnqueueReadBuffer(cmdQueue,bufferB,CL_TRUE,0,
????????????????????????datasize,buf_B,0,NULL,NULL);
????SaveImg();
????//釋放所有調用和內存
????clReleaseKernel(kernel);
????clReleaseProgram(program);
????clReleaseCommandQueue(cmdQueue);
????clReleaseMemObject(bufferA);
????clReleaseMemObject(bufferB);
????clReleaseContext(context);
????delete buf_code;
????return 0;
}