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
- The Suspects 并查集
Severe acute respiratory syndrome (SARS), an atypical pneumonia of unknown aetiology, was recognized ...
- Mac无法上网
今天mac突然无法上网了, 家里的大部分设备, 都出现了重启后无法上网的问题, 猜测可能是dns有问题了. 于是乎, 在mac中添加了如下DNS 114.114.114.114 8.8.8.8 1.1 ...
- jQuery--- .hasOwnProperty 用法
☆ obj.hasOwnProperty('prop'): 是用来判断一个对象是否有你给出名称的属性或对象.不过需要注意的是, 此方法无法检查该对象的原型链中是否具有该属性,该属性必须是对象本身的一个 ...
- Django FBV和CBV -
一.FBV和CBV 在Python菜鸟之路:Django 路由.模板.Model(ORM)一节中,已经介绍了几种路由的写法及对应关系,那种写法可以称之为FBV: function base view ...
- 在Win7下新建库并修改图标
win7中在库中添加桌面方法详解 1.在空白处,鼠标右键选择新建——库. 2.命名为桌面,然后选择桌面. 3.鼠标右键选择属性. 4.点击包括文件夹. 5.选择桌面,点击包括文件夹按钮. 6.点击确定 ...
- mac os 里的 JAVA_HOME
google了一下,发现了这篇文章Important Java Directories on Mac OS X(https://developer.apple.com/library/content/ ...
- Vuejs项目的Webpack2构建优化
最近在做的项目因为相对较大(打包有100多个chunk),在build构建的时候速度一直上不去,甚是烦恼.由于用的是vue-cli的webpack2模板,一开始并没有想着要对其进行优化,一直觉得是we ...
- CommonsChunkPlugin知识点
CommonsChunkPlugin 的作用就是提取代码中的公共模块,然后将公共模块打包到一个独立的文件中去,以便在其它的入口和模块中使用. 多个 html共用一个js文件(chunk),可用Comm ...
- netty SimpleChannelInboundHandler<Message>和ChannelInboundHandlerApter
一个兄弟的测试体验:https://blog.csdn.net/linuu/article/details/51307060 比较官方:https://www.imooc.com/article/28 ...
- 一段简单的代码记录如何通过 js 给 HTML 设置自定义属性，并且通过点击事件获取到所设置的自定义属性值
for (var i = 0; i <= 5; i++) { const divJoinDay = document.createElement("div"); divJoi ...