perflab實驗報告(同名8817).doc

湖南大學課程實驗報告 課 程 名 稱： 計算機組成與結構 實驗項目名稱： perflab 專 業(yè) 班 級： 姓 名： 學 號： 指 導 教 師： 完 成 時 間： 2015 年 05 月 22 日計算機科學與工程系實驗題目：程序性能調(diào)優(yōu)實驗實驗目的：kernel.c文件中主要有兩個需要進行優(yōu)化的函數(shù)：rotate和smooth，并分別給出了naive_rotate和naive_smooth兩個函數(shù)的基本實現(xiàn)作為baseline作為你改進后的程序的比較對象。你需要讀懂rotate和smooth函數(shù)，并對其進行優(yōu)化。你每寫一個新版本的、優(yōu)化的rotate和smooth函數(shù)，均可在成注冊后使用driver進行測試，并得到對應的CPE和加速比。本次實驗，要求針對每個函數(shù)、每個人均至少寫出3種優(yōu)化版本、并根據(jù)driver報告的結果進行性能分析。實驗環(huán)境：Vmware虛擬機 ubuntu12.04 linux終端實驗步驟和結果分析：函數(shù)源碼：rotate函數(shù)：void naive_rotate(int dim, pixel *src, pixel *dst) int i, j; for (i = 0; i dim; i+)for (j = 0; j dim; j+) dstRIDX(dim-1-j, i, dim) = srcRIDX(i, j, dim);rotate函數(shù)的作用是通過將每個像素進行行列調(diào)位，將一副點陣圖像進行90度旋轉。其中RIDX(i,j,n)即(i)*(n)+(j)。函數(shù)缺點為程序局部性不好，循環(huán)次數(shù)過多?？梢詫ζ溥M行分塊來提高空間局部性，也可以進行循環(huán)展開。smooth函數(shù)：void naive_smooth(int dim, pixel *src, pixel *dst) int i, j; for (i = 0; i dim; i+)for (j = 0; j dim; j+) dstRIDX(i, j, dim) = avg(dim, i, j, src);smooth函數(shù)的作用是通過對圖像每幾點像素求平均值來對圖像進行模糊化處理。函數(shù)缺點是循環(huán)次數(shù)過多和頻繁調(diào)用avg函數(shù)，avg函數(shù)中又包含許多函數(shù)。應該減少avg函數(shù)的調(diào)用次數(shù)，且進行循環(huán)展開。第一種版本：CPE分析：rotate函數(shù)：void rotate(int dim, pixel *src, pixel *dst) int i,j,ii,jj; for(ii=0;iidim;ii+=4) for(jj=0;jjdim;jj+=4) for(i=ii;iii+4;i+) for(j=jj;jjj+4;j+) dstRIDX(dim-1-j,i,dim)=srcRIDX(i,j,dim);多添加了兩個for函數(shù)，將循環(huán)分成了4*4的小塊，在cache存儲體不足夠大的情況下，對循環(huán)分塊能夠提升高速緩存命中率，從高提升了空間局部性。從測試的CPE中也可以看出，在dim是64的時候，原代碼和本代碼CPE相差不大，而隨著dim的增大，本代碼CPE增加不大，而原代碼CPE急劇增加，就是受到了cache存儲的局限性。smooth函數(shù)：void smooth(int dim, pixel *src, pixel *dst)pixel_sum rowsum530530; int i, j, snum; for(i=0; idim; i+) rowsumi0.red = (srcRIDX(i, 0, dim).red+srcRIDX(i, 1, dim).red); rowsumi0.blue = (srcRIDX(i, 0, dim).blue+srcRIDX(i, 1, dim).blue); rowsumi0.green = (srcRIDX(i, 0, dim).green+srcRIDX(i, 1, dim).green); rowsumi0.num = 2; for(j=1; jdim-1; j+) rowsumij.red = (srcRIDX(i, j-1, dim).red+srcRIDX(i, j, dim).red+srcRIDX(i, j+1, dim).red); rowsumij.blue = (srcRIDX(i, j-1, dim).blue+srcRIDX(i, j, dim).blue+srcRIDX(i, j+1, dim).blue); rowsumij.green = (srcRIDX(i, j-1, dim).green+srcRIDX(i, j, dim).green+srcRIDX(i, j+1, dim).green); rowsumij.num = 3; rowsumidim-1.red = (srcRIDX(i, dim-2, dim).red+srcRIDX(i, dim-1, dim).red); rowsumidim-1.blue = (srcRIDX(i, dim-2, dim).blue+srcRIDX(i, dim-1, dim).blue); rowsumidim-1.green = (srcRIDX(i, dim-2, dim).green+srcRIDX(i, dim-1, dim).green); rowsumidim-1.num = 2; for(j=0; jdim; j+) snum = rowsum0j.num+rowsum1j.num; dstRIDX(0, j, dim).red = (unsigned short)(rowsum0j.red+rowsum1j.red)/snum); dstRIDX(0, j, dim).blue = (unsigned short)(rowsum0j.blue+rowsum1j.blue)/snum); dstRIDX(0, j, dim).green = (unsigned short)(rowsum0j.green+rowsum1j.green)/snum); for(i=1; i512時，超出了設置的數(shù)組大小會報錯。第二種版本：CPE分析：rotate函數(shù)：void rotate(int dim, pixel *src, pixel *dst) int i, j; int temp; int it,jt; int im,jm; for(jt=0; jtdim; jt+=32) jm=jt+32; for(it=0; itdim; it+=32) im=it+32; for(j=jt; jjm; j+) temp=dim-1-j; for(i=it; ired=(P1-red+(P1+1)-red+P2-red+(P2+1)-red)2; dst-green=(P1-green+(P1+1)-green+P2-green+(P2+1)-green)2; dst-blue=(P1-blue+(P1+1)-blue+P2-blue+(P2+1)-blue)2; dst+; for(i=1;ired=(P1-red+(P1+1)-red+(P1+2)-red+P2-red+(P2+1)-red+(P2+2)-red)/6; dst-green=(P1-green+(P1+1)-green+(P1+2)-green+P2-green+(P2+1)-green+(P2+2)-green)/6; dst-blue=(P1-blue+(P1+1)-blue+(P1+2)-blue+P2-blue+(P2+1)-blue+(P2+2)-blue)/6; dst+; P1+; P2+; dst-red=(P1-red+(P1+1)-red+P2-red+(P2+1)-red)2; dst-green=(P1-green+(P1+1)-green+P2-green+(P2+1)-green)2; dst-blue=(P1-blue+(P1+1)-blue+P2-blue+(P2+1)-blue)2; dst+; P1=src; P2=P1+dim0; P3=P2+dim0; for(i=1;ired=(P1-red+(P1+1)-red+P2-red+(P2+1)-red+P3-red+(P3+1)-red)/6; dst-green=(P1-green+(P1+1)-green+P2-green+(P2+1)-green+P3-green+(P3+ 1)-green)/6; dst-blue=(P1-blue+(P1+1)-blue+P2-blue+(P2+1)-blue+P3-blue+(P3+1)-blue)/6; dst+; dst1=dst+1; for(j=1;jred=(P1-red+(P1+1)-red+(P1+2)-red+P2-red+(P2+1)-red+(P2+2)-red+P3-red+(P3+1)-red+(P3+2)-red)/9; dst-green=(P1-green+(P1+1)-green+(P1+2)-green+P2-green+(P2+1)-green+(P2+2)-green+P3-green+(P3+1)-green+(P3+2)-green)/9; dst-blue=(P1-blue+(P1+1)-blue+(P1+2)-blue+P2-blue+(P2+1)-blue+(P2+2)-blue+P3-blue+(P3+1)-blue+(P3+2)-blue)/9; dst1-red=(P1+3)-red+(P1+1)-red+(P1+2)-red+(P2+3)-red+(P2+1)-red+(P2+2)-red+(P3+3)-red+(P3+1)-red+(P3+2)-red)/9; dst1-green=(P1+3)-green+(P1+1)-green+(P1+2)-green+(P2+3)-green+(P2+1)-green+(P2+2)-green+(P3+3)-green+(P3+1)-green+(P3+2)-green)/9; dst1-blue=(P1+3)-blue+(P1+1)-blue+(P1+2)-blue+(P2+3)-blue+(P2+1)-blue+(P2+2)-blue+(P3+3)-blue+(P3+1)-blue+(P3+2)-blue)/9; dst+=2; dst1+=2; P1+=2; P2+=2; P3+=2; for(;jred=(P1-red+(P1+1)-red+(P1+2)-red+P2-red+(P2+1)-red+(P2+2)-red+P3-red+(P3+1)-red+(P3+2)-red)/9; dst-green=(P1-green+(P1+1)-green+(P1+2)-green+P2-green+(P2+1)-green+(P2+2)-green+P3-green+(P3+1)-green+(P3+2)-green)/9; dst-blue=(P1-blue+(P1+1)-blue+(P1+2)-blue+P2-blue+(P2+1)-blue+(P2+2)-blue+P3-blue+(P3+1)-blue+(P3+2)-blue)/9; dst+; P1+; P2+; P3+; dst-red=(P1-red+(P1+1)-red+P2-red+(P2+1)-red+P3-red+(P3+1)-red)/6; dst-green=(P1-green+(P1+1)-green+P2-green+(P2+1)-green+P3-green+(P3+1)-green)/6; dst-blue=(P1-blue+(P1+1)-blue+P2-blue+(P2+1)-blue+P3-blue+(P3+1)-blue)/6; dst+; P1+=2; P2+=2; P3+=2; dst-red=(P1-red+(P1+1)-red+P2-red+(P2+1)-red)2; dst-green=(P1-green+(P1+1)-green+P2-green+(P2+1)-green)2; dst-blue=(P1-blue+(P1+1)-blue+P2-blue+(P2+1)-blue)2; dst+; for(i=1;ired=(P1-red+(P1+1)-red+(P1+2)-red+P2-red+(P2+1)-red+(P2+2)-red)/6; dst-green=(P1-green+(P1+1)-green+(P1+2)-green+P2-green+(P2+1)-green+(P2+2)-green)/6; dst-blue=(P1-blue+(P1+1)-blue+(P1+2)-blue+P2-blue+(P2+1)-blue+(P2+2)-blue)/6; dst+; P1+; P2+; dst-red=(P1-red+(P1+1)-red+P2-red+(P2+1)-red)2; dst-green=(P1-green+(P1+1)-green+P2-green+(P2+1)-green)2; dst-blue=(P1-blue+(P1+1)-blue+P2-blue+(P2+1)-blue)2; 這段代碼也是通過不調(diào)用avg函數(shù)來加速程序。將Smooth函數(shù)處理分為4塊，一為主體內(nèi)部，由9點求平均值；二為4個頂點，由4點求平均值；三為四條邊界，由6點求平均值。從圖片的頂部開始處理，再上邊界，順序處理下來，其中在處理左邊界時，for循環(huán)處理一行主體部分，就是以上的代碼。第三種版本：CPE分析：rotate函數(shù)：void rotate(int dim, pixel *src, pixel *dst) int i, j; int dst_base=(dim-1)*dim; dst+=dst_base; for(i=0;idim;i+=32) for(j=0;jdim;j+) *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+=dim; dst+; *dst=*src; src+; src-=(dim5)-dim; dst-=31+dim; dst+=dst_base+dim; dst+=32; src+=(dim2; dst0.blue = (src0.blue+src1.blue+srcdim.blue+srcdim+1.blue)2; dst0.green = (src0.green+src1.green+srcdim.green+srcdim+1.green)2; dstdim-1.red = (srcdim-1.red+srcdim-2.red+srcdim*2-1.red+srcdim*2-2.red)2; dstdim-1.blue = (srcdim-1.blue+srcdim-2.blue+srcdim*2-1.blue+srcdim*2-2.blue)2; dstdim-1.green = (srcdim-1.green+srcdim-2.green+srcdim*2-1.green+srcdim*2-2.green)2; dstdim*(dim-1).red = (srcdim*(dim-1).red+srcdim*(dim-1)+1.red+srcdim*(dim-2).red+srcdim*(dim-2)+1.red)2; dstdim*(dim-1).blue = (srcdim*(dim-1).blue+srcdim*(dim-1)+1.blue+srcdim*(dim-2).blue+srcdim*(dim-2)+1.blue)2; dstdim*(dim-1).green = (srcdim*(dim-1).green+srcdim*(dim-1)+1.green+srcdim*(dim-2).green+srcdim*(dim-2)+1.green)2; dstdim*dim-1.red = (srcdim*dim-1.red+srcdim*dim-2.red+srcdim*(dim-1)-1.red+srcdim*(dim-1)-2.red)2; dstdim*dim-1.blue = (srcdim*dim-1.blue+srcdim*dim-2.blue+srcdim*(dim-1)-1.blue+srcdim*(dim-1)-2.blue)2; dstdim*dim-1.green = (srcdim*dim-1.green+srcdim*dim-2.green+srcdim*(dim-1)-1.green+srcdim*(dim-1)-2.green)2; for (j = 1; j dim-1; j+) dstj.red = (srcj.red+srcj-1.red+srcj+1.red+srcj+dim.red+srcj+1+dim.red+srcj-1+dim.red)/6; dstj.green = (srcj.green+srcj-1.green+srcj+1.green+srcj+dim.green+srcj+1+dim.green+srcj-1+dim.green)/6; dstj.blue = (srcj.blue+srcj-1.blue+srcj+1.blue+srcj+dim.blue+srcj+1+dim.blue+srcj-1+dim.blue)/6; for (j = dim*(dim-1)+1; j dim*dim-1; j+) dstj.red = (srcj.red+srcj-1.red+srcj+1.red+srcj-dim.red+srcj+1-dim.red+srcj-1-dim.red)/6; dstj.green = (srcj.green+srcj-1.green+srcj+1.green+srcj-dim.green+srcj+1-dim.green+srcj-1-dim.green)/6; dstj.blue = (srcj.blue+srcj-1.blue+srcj+1.blue+srcj-dim.blue+srcj+1-dim.blue+srcj-1-dim.blue)/6; for (j = dim; j dim*(dim-1); j+=dim) dstj.red = (srcj.red+srcj-dim.red+srcj+1.red+srcj+dim.red+srcj+1+dim.red+srcj-dim+1.red)/6; dstj.green = (srcj.green+srcj-dim.green+srcj+1.green+srcj+dim.green+srcj+1+dim.green+srcj-dim+1.green)/6; dstj.blue = (srcj.blue+srcj-dim.blue+srcj+1.blue+srcj+dim.blue+srcj+1+dim.blue+srcj-dim+1.blue)/6; for (j = dim+dim-1; j dim*dim-1; j+=d