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

Largest Submatrix of All 1’s
Given a m-by-n (0,1)-matrix, of all its submatrices of all 1’s which is the largest? By largest we m ...
hdu2083 简易版之最短距离排序水题
给出数轴n个坐标,求一个点到所有点距离总和最小.排序后最中间一个点或两个点之间就是最优 #include<stdio.h> #include<algorithm> using ...
setsebool命令详解与SELinux管理
setsebool命令是用来修改SElinux策略内各项规则的布尔值.setsebool命令和getsebool命令是SELinux修改和查询布尔值的一套工具组.SELinux的策略与规则管理相关命令 ...
【Quartz】Quartz的数据库表
select * from test.QRTZ_TRIGGERS 触发器表 select * from QRTZ_PAUSED_TRIGGER_GRPS 暂停的分组任务表 select * from ...
lets encrypt 申请nginx 泛域名
1. 安装certbot工具 wget https://dl.eff.org/certbot-auto chmod a+x ./certbot-auto 2. 申请通配符域名 ./certbot-au ...
Linux部署禅道Steps&Q&A
1.查看Linux的位数: getconf LONG_BIT 结果:32/64 2. 禅道开源版安装包下载 Linux 64位下载站点1: http://sourceforge.net/projec ...
Microsoft Dynamics CRM 2011 新建实体需要注意的细节
新建一个实体,需要红色框内的是否勾选的意义,可以进一步加深对CRM的理解.如图: 下面对部分的进行了自我的理解,不对的地方,还请大家指出来.互相学习. 1.CRM2011中,在活动方面加强的新特性包括 ...
golang 自定义类型的排序sort
sort包中提供了很多排序算法,对自定义类型进行排序时,只需要实现sort的Interface即可,包括: func Len() int {... } func Swap(i, j int) {... ...
【jemter】HTTP请求参数化
HTTP请求参数化:就是把URL的参数项做参数化处理我们现在要对子猴博客来进行一番压力测试,压力测试对象为随机的几个网页链接,这几个链接是写在一个文本文件中的,在压力测试的时候会随机读取. 1. ...
netty搭建Tcp服务器实践
在netty基本组件介绍中,我们大致了解了netty的一些基本组件,今天我们来搭建一个基于netty的Tcp服务端程序,通过代码来了解和熟悉这些组件的功能和使用方法. 首先我们自己创建一个Server ...

0_Simple__simpleLayeredTexture

0_Simple__simpleLayeredTexture的更多相关文章

随机推荐

热门专题