Changes between Version 10 and Version 11 of adherelinux/installGPU

Timestamp:

Jul 17, 2010, 10:28:17 PM (15 years ago)

Author:

adherelinux

Comment:

--

adherelinux/installGPU

@@ -108 +108 @@
 你將會看到下面的畫面[[br]]
 
-[[Image(nvidia_xserver.jpg)]] 
-
-
-
-
-
-
+[[Image(query_gpu.jpg)]] 
+
+8.安裝CUDA 工具套件和CUDA 軟體開發套件[[br]]
+進入終端機[[br]]
+$sudo sh ./cudatoolkit_2.3_linux_64_ubuntu9.04.run[[br]]
+$sudo sh ./cudasdk_2.3_linux.run[[br]]
+過程中，所有選項按Enter會使用預設的設定。[[br]]
+修改 etc 目錄下的 profile 檔[[br]]
+$cd /etc/[[br]]
+$cudo emacs profile (可用任意的編輯器)[[br]]
+依照CUDA所在目錄，在檔案的最後加上[[br]]
+PATH=/usr/local/cuda/bin:$PATH [[br]]
+LD_LIBRARY_PATH=/usr/local/cuda/lib64[[br]]
+export PATH[[br]]
+export LD_LIBRARY_PATH[[br]]
+這四行。[[br]]
+如果是32 位元的作業系統，第二行改成LD_LIBRARY_PATH=/usr/local/cuda/lib[[br]]
+如果是64 位元的作業系統，第二行改成LD_LIBRARY_PATH=/usr/local/cuda/lib64[[br]]
+登出，再登入, 用CUDA程式測試，如果正確無誤，即完成安裝。[[br]]
+
+若出現以下錯誤訊息[[br]]
+= ./a.out: error while loading shared libraries: libcudart.so.2: cannot open shared object file: No such file or directory =
+
+若是64位元 命令列 [[br]]
+$sudo ln -sf /usr/local/cuda/lib64/libcudart.so.2.3 /lib64/libcudart.so.2[[br]]
+如果是32位元可能需要加此行[[br]]
+$sudo ln -sf /usr/local/cuda/lib/libcudart.so.2.3 /lib/libcudart.so.2[[br]]
+
+可以檢查(option)[[br]]
+$emacs /etc/ld.so.conf[[br]]
+include /usr/local/cuda/lib[[br]]
+
+若32位元 include /usr/local/cuda/lib[[br]]
+若64位元 include /usr/local/cuda/lib64[[br]]
+
+可用以下程式測試。[[br]]
+用nvcc 編譯cuda 程式 number_add_1.cu。[[br]]
+如果CPU的執行結果和GPU相同，代表GPU成功運作。[[br]]
+
+
+$nvcc number_add_1.cu[[br]]
+$./a.out[[br]]
+
+
+number_add_1.cu程式碼:[[br]]
+
+#include <stdio.h> [[br]]
+#include <stdlib.h>[[br]]
+#include <math.h>[[br]]
+#include <time.h>[[br]]
+#include <sys/time.h>[[br]]
+#include <iostream>[[br]]
+#include <iomanip>[[br]]
+using namespace std;[[br]]
+#define DATA_SIZE 1048576[[br]]
+int data[DATA_SIZE];[[br]]
+
+__global__ static void sumOfSquares(int *num, int* result){[[br]]
+    int sum = 0;[[br]]
+    int i;[[br]]
+    for(i = 0; i < DATA_SIZE; i++) {[[br]]
+        sum += num[i] * num[i];[[br]]
+    }[[br]]
+
+    *result = sum;[[br]]
+}[[br]]
+void GenerateNumbers(int *number ,int size){[[br]]
+   int i;[[br]]
+   for(i=0;i<size;i++){ [[br]]
+       number[i]=rand() % 10; [[br]]   //value = 0 to 9 [[br]]
+   }[[br]]
+}[[br]]
+
+double wallclock(void){[[br]]
+  struct timeval tv;[[br]]
+  struct timezone tz;[[br]]
+  double t;[[br]]
+
+  gettimeofday(&tv, &tz);[[br]]
+  t = (double)tv.tv_sec*1000;[[br]]
+  t += ((double)tv.tv_usec)/1000.0;[[br]]
+
+  return t;[[br]]
+}// millisecond[[br]]
+
+int main(){[[br]]
+    cudaSetDevice(0); [[br]]  set device number[[br]]
+    GenerateNumbers(data, DATA_SIZE);[[br]]
+
+    int* gpudata,*result,sum;[[br]]
+    double t1,t2;[[br]]
+
+    cudaMalloc((void**) &gpudata, sizeof(int) * DATA_SIZE);[[br]]
+    cudaMalloc((void**) &result, sizeof(int));[[br]]
+    cudaMemcpy(gpudata, data, sizeof(int) * DATA_SIZE,cudaMemcpyHostToDevice);[[br]]
+
+    t1 = wallclock();[[br]]
+        sumOfSquares<<<1, 1, 0>>>(gpudata, result);[[br]]
+        cudaMemcpy(&sum, result, sizeof(int), cudaMemcpyDeviceToHost);[[br]]
+
+    t2 = wallclock();[[br]]
+  
+    printf("Elapsed time = %f(ms) in GPU\n",t2-t1);[[br]]
+
+    cudaFree(gpudata);[[br]]
+    cudaFree(result);[[br]]
+
+    printf("sum: %d\n", sum);[[br]]
+
+    sum = 0;[[br]]
+    for(int i = 0; i < DATA_SIZE; i++) {[[br]]
+        sum += data[i] * data[i];[[br]]
+    }[[br]]
+    printf("sum (CPU): %d\n", sum);[[br]]
+}[[br]]
+
+
+
+
+
+
+
+