0_Simple__simpleLayeredTexture

二维分层纹理

▶ 源代码。用纹理方法把元素按原顺序从 CUDA3D 数组中取出来，求个相反数再加上层数放入全局内存，输出。

 #include <stdio.h>

 #include "cuda_runtime.h"

 #include "device_launch_parameters.h"

 #include <helper_functions.h>

 #include <helper_cuda.h>

 #define MIN_EPSILON_ERROR 5e-3f

 #define OUTPUT 5

 texture<float, cudaTextureType2DLayered> tex;

 __global__ void transformKernel(float *g_odata, int width, int height, int layer)

 {

     unsigned int x = blockIdx.x*blockDim.x + threadIdx.x;

     unsigned int y = blockIdx.y*blockDim.y + threadIdx.y;

     float u = (x + 0.5f) / (float)width;

     float v = (y + 0.5f) / (float)height;

     g_odata[layer*width*height + y*width + x] = - tex2DLayered(tex, u, v, layer) + layer;

 }

 int main(int argc, char **argv)

 {

     unsigned int width = , height = , num_layers = ;

     unsigned int size = width * height * num_layers * sizeof(float);

     float *h_data = (float *)malloc(size);

     float *h_data_ref = (float *)malloc(size);

     float *d_data = NULL;

     cudaMalloc((void **)&d_data, size);

     for (unsigned int layer = ; layer < num_layers; layer++)

     {

         for (int i = ; i < (int)(width * height); i++)

             h_data[layer*width*height + i] = (float)i;

     }

     for (unsigned int layer = ; layer < num_layers; layer++)

     {

         for (int i = ; i < (int)(width * height); i++)

             h_data_ref[layer*width*height + i] = - h_data[layer*width*height + i] + layer;

     }

     printf("\n\t\Input data\n\t");

     for (int i = ; i < num_layers; i++)

     {

         for (int j = ; j < OUTPUT; j++)

         {

             for(int k=;k<OUTPUT;k++)

                 printf("%2.1f ", h_data[i*width*height+j*width+k]);

             printf("\n\t");

         }

         printf("\n\t");

     }

     printf("\n\t\Ideal output data\n\t");

     for (int i = ; i < num_layers; i++)

     {

         for (int j = ; j < OUTPUT; j++)

         {

             for (int k = ; k<OUTPUT; k++)

                 printf("%2.1f ", h_data_ref[i*width*height + j*width + k]);

             printf("\n\t");

         }

         printf("\n\t");

     }

     // 设置 CUDA 3D 数组参数和数据拷贝

     cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(, , , , cudaChannelFormatKindFloat);

     cudaArray *cu_3darray;

     cudaMalloc3DArray(&cu_3darray, &channelDesc, make_cudaExtent(width, height, num_layers), cudaArrayLayered);

     cudaMemcpy3DParms myparms = {  };

     myparms.srcPos = make_cudaPos(, , );

     myparms.dstPos = make_cudaPos(, , );

     myparms.srcPtr = make_cudaPitchedPtr(h_data, width * sizeof(float), width, height);

     myparms.dstArray = cu_3darray;

     myparms.extent = make_cudaExtent(width, height, num_layers);

     myparms.kind = cudaMemcpyHostToDevice;

     cudaMemcpy3D(&myparms);

     // 设置纹理参数并绑定

     tex.addressMode[] = cudaAddressModeWrap;

     tex.addressMode[] = cudaAddressModeWrap;

     tex.filterMode = cudaFilterModeLinear;

     tex.normalized = true;

     cudaBindTextureToArray(tex, cu_3darray, channelDesc);

     dim3 dimBlock(, , );

     dim3 dimGrid(width / dimBlock.x, height / dimBlock.y, );

     printf("Covering 2D data of %d * %d * %d: Grid size is %d x %d, each block has 8 x 8 threads\n", width, height, num_layers, dimGrid.x, dimGrid.y);

     transformKernel << < dimGrid, dimBlock >> >(d_data, width, height, );// 预跑

     cudaDeviceSynchronize();

     StopWatchInterface *timer = NULL;

     sdkCreateTimer(&timer);

     sdkStartTimer(&timer);

     for (unsigned int layer = ; layer < num_layers; layer++)// 启用多个核，每个核完成一层

         transformKernel << < dimGrid, dimBlock,  >> >(d_data, width, height, layer);

     cudaDeviceSynchronize();

     sdkStopTimer(&timer);

     printf("\n\Time: %.3f msec, %.2f Mtexlookups/sec\n", sdkGetTimerValue(&timer), (width *height *num_layers / (sdkGetTimerValue(&timer) / 1000.0f) / 1e6));

     sdkDeleteTimer(&timer);

     // 返回计算结果并检验

     memset(h_data, , size);

     cudaMemcpy(h_data, d_data, size, cudaMemcpyDeviceToHost);

     if (checkCmdLineFlag(argc, (const char **)argv, "regression"))

         sdkWriteFile<float>("./data/regression.dat", h_data, width * width, 0.0f, false);

     else

         printf("Comparing kernel output to expected data return %d\n", compareData(h_data, h_data_ref, width * height * num_layers, MIN_EPSILON_ERROR, 0.0f));

     printf("\n\tActual output data\n\t");

     for (int i = ; i < num_layers; i++)

     {

         for (int j = ; j < OUTPUT; j++)

         {

             for (int k = ; k<OUTPUT; k++)

                 printf("%2.1f ", h_data[i*width*height + j*width + k]);

             printf("\n\t");

         }

         printf("\n\t");

     }

     free(h_data);

     free(h_data_ref);

     cudaFree(d_data);

     cudaFreeArray(cu_3darray);

     getchar();

     return ;

 }

▶ 输出结果

    Input data

    0.0 1.0 2.0 3.0 4.0

    512.0 513.0 514.0 515.0 516.0

    1024.0 1025.0 1026.0 1027.0 1028.0

    1536.0 1537.0 1538.0 1539.0 1540.0

    2048.0 2049.0 2050.0 2051.0 2052.0

    0.0 1.0 2.0 3.0 4.0

    512.0 513.0 514.0 515.0 516.0

    1024.0 1025.0 1026.0 1027.0 1028.0

    1536.0 1537.0 1538.0 1539.0 1540.0

    2048.0 2049.0 2050.0 2051.0 2052.0

    0.0 1.0 2.0 3.0 4.0

    512.0 513.0 514.0 515.0 516.0

    1024.0 1025.0 1026.0 1027.0 1028.0

    1536.0 1537.0 1538.0 1539.0 1540.0

    2048.0 2049.0 2050.0 2051.0 2052.0

    0.0 1.0 2.0 3.0 4.0

    512.0 513.0 514.0 515.0 516.0

    1024.0 1025.0 1026.0 1027.0 1028.0

    1536.0 1537.0 1538.0 1539.0 1540.0

    2048.0 2049.0 2050.0 2051.0 2052.0

    0.0 1.0 2.0 3.0 4.0

    512.0 513.0 514.0 515.0 516.0

    1024.0 1025.0 1026.0 1027.0 1028.0

    1536.0 1537.0 1538.0 1539.0 1540.0

    2048.0 2049.0 2050.0 2051.0 2052.0

    Ideal output data

    0.0 -1.0 -2.0 -3.0 -4.0

    -512.0 -513.0 -514.0 -515.0 -516.0

    -1024.0 -1025.0 -1026.0 -1027.0 -1028.0

    -1536.0 -1537.0 -1538.0 -1539.0 -1540.0

    -2048.0 -2049.0 -2050.0 -2051.0 -2052.0

    1.0 0.0 -1.0 -2.0 -3.0

    -511.0 -512.0 -513.0 -514.0 -515.0

    -1023.0 -1024.0 -1025.0 -1026.0 -1027.0

    -1535.0 -1536.0 -1537.0 -1538.0 -1539.0

    -2047.0 -2048.0 -2049.0 -2050.0 -2051.0

    2.0 1.0 0.0 -1.0 -2.0

    -510.0 -511.0 -512.0 -513.0 -514.0

    -1022.0 -1023.0 -1024.0 -1025.0 -1026.0

    -1534.0 -1535.0 -1536.0 -1537.0 -1538.0

    -2046.0 -2047.0 -2048.0 -2049.0 -2050.0

    3.0 2.0 1.0 0.0 -1.0

    -509.0 -510.0 -511.0 -512.0 -513.0

    -1021.0 -1022.0 -1023.0 -1024.0 -1025.0

    -1533.0 -1534.0 -1535.0 -1536.0 -1537.0

    -2045.0 -2046.0 -2047.0 -2048.0 -2049.0

    4.0 3.0 2.0 1.0 0.0

    -508.0 -509.0 -510.0 -511.0 -512.0

    -1020.0 -1021.0 -1022.0 -1023.0 -1024.0

    -1532.0 -1533.0 -1534.0 -1535.0 -1536.0

    -2044.0 -2045.0 -2046.0 -2047.0 -2048.0

    Covering 2D data of  *  * : Grid size is  x , each block has  x  threads

Time: 0.995 msec, 1317.00 Mtexlookups/sec

Comparing kernel output to expected data return 

    Actual output data

    0.0 -1.0 -2.0 -3.0 -4.0

    -512.0 -513.0 -514.0 -515.0 -516.0

    -1024.0 -1025.0 -1026.0 -1027.0 -1028.0

    -1536.0 -1537.0 -1538.0 -1539.0 -1540.0

    -2048.0 -2049.0 -2050.0 -2051.0 -2052.0

    1.0 0.0 -1.0 -2.0 -3.0

    -511.0 -512.0 -513.0 -514.0 -515.0

    -1023.0 -1024.0 -1025.0 -1026.0 -1027.0

    -1535.0 -1536.0 -1537.0 -1538.0 -1539.0

    -2047.0 -2048.0 -2049.0 -2050.0 -2051.0

    2.0 1.0 0.0 -1.0 -2.0

    -510.0 -511.0 -512.0 -513.0 -514.0

    -1022.0 -1023.0 -1024.0 -1025.0 -1026.0

    -1534.0 -1535.0 -1536.0 -1537.0 -1538.0

    -2046.0 -2047.0 -2048.0 -2049.0 -2050.0

    3.0 2.0 1.0 0.0 -1.0

    -509.0 -510.0 -511.0 -512.0 -513.0

    -1021.0 -1022.0 -1023.0 -1024.0 -1025.0

    -1533.0 -1534.0 -1535.0 -1536.0 -1537.0

    -2045.0 -2046.0 -2047.0 -2048.0 -2049.0

    4.0 3.0 2.0 1.0 0.0

    -508.0 -509.0 -510.0 -511.0 -512.0

    -1020.0 -1021.0 -1022.0 -1023.0 -1024.0

    -1532.0 -1533.0 -1534.0 -1535.0 -1536.0

    -2044.0 -2045.0 -2046.0 -2047.0 -2048.0

▶ 涨姿势

● 与前面立方体贴图纹理不同的地方：申请 CUDA3D 数组的时候使用标志 cudaArrayLayered 而不是 cudaArrayCubemap，并注意调整相关的维度参数。

0_Simple__simpleLayeredTexture的更多相关文章

随机推荐

Select2 用法
http://www.cnblogs.com/wuhuacong/p/4761637.html 2.这个做详细参考 http://www.jianshu.com/p/c5ab74b91b2e 3.ht ...
oracle 获取星期日期
oracle 中的计算如下: 计算本星期的起始结束日期--得到星期一的日期select trunc(sysdate,''DD'')-to_char(sysdate,''D'')+2 from dual ...
fork 开源项目后如何参与项目
好的开源项目都很想参与到开源活动中,并且会 fork 一份. 经过几个月的学习,大概明白了如果参与开源项目. 当完成 fork 后,就需要在本地 git clone 一份. 有新的功能或需要修复的就开 ...
postman 中 form-data、x-www-form-urlencoded、raw、binary的区别
区别 form-data: 就是http请求中的multipart/form-data,它会将表单的数据处理为一条消息,以标签为单元,用分隔符分开.既可以上传键值对,也可以上传文件.当上传的字段是文件 ...
md5，base64，rsa
MD5功能: 输入任意长度的信息,经过处理,输出为128位的信息(数字指纹): 不同的输入得到的不同的结果(唯一性): 根据128位的输出结果不可能反推出输入的信息(不可逆): 1. ...
yii2 rbac权限管理学习笔记
下面介绍一个 yii2 的 Rbac 权限管理设置,闲话少说,直接上代码, 1.首先我们要在组件里面配置一下 Rbac ,如下所示(common/config/main-local.php或者main ...
二分法查找（Binary Search）
二分法查找适用于排列有序的数据.java实现方法如下: // Find the location of a value in array a // Array a must be sorted // ...
ES(6): access elasticsearch via curl
curl是一个非常实用的.用来与服务器之间传输数据的工具:支持的协议包括 (DICT, FILE, FTP, FTPS, GOPHER, HTTP, HTTPS, IMAP, IMAPS, LDAP, ...
织梦开启PHP 标签
第一步: dedecms出现DedeCMS Error:Tag disabled:php原因解决 --------------------------------------------------- ...
php curl模拟登录（半转载）
参考:http://our2848884.blog.163.com/blog/static/146854834201282039334/ php curl模拟登录参考:http://blog.c ...

0_Simple__simpleLayeredTexture

0_Simple__simpleLayeredTexture的更多相关文章

随机推荐

热门专题