cuda

numba cuda does not produce correct result with += (gpu reduction needed?)

由 白昼怎懂夜的黑 提交于 2021-02-18 19:38:17
问题 I am using numba cuda to calculate a function. The code is simply to add up all the values into one result, but numba cuda gives me a different result from numpy. numba code import math def numba_example(number_of_maximum_loop,gs,ts,bs): from numba import cuda result = cuda.device_array([3,]) @cuda.jit(device=True) def BesselJ0(x): return math.sqrt(2/math.pi/x) @cuda.jit def cuda_kernel(number_of_maximum_loop,result,gs,ts,bs): i = cuda.grid(1) if i < number_of_maximum_loop: result[0] +=

numba cuda does not produce correct result with += (gpu reduction needed?)

由 喜你入骨 提交于 2021-02-18 19:36:40
问题 I am using numba cuda to calculate a function. The code is simply to add up all the values into one result, but numba cuda gives me a different result from numpy. numba code import math def numba_example(number_of_maximum_loop,gs,ts,bs): from numba import cuda result = cuda.device_array([3,]) @cuda.jit(device=True) def BesselJ0(x): return math.sqrt(2/math.pi/x) @cuda.jit def cuda_kernel(number_of_maximum_loop,result,gs,ts,bs): i = cuda.grid(1) if i < number_of_maximum_loop: result[0] +=

Cuda thread scheduling - latency hiding

由 岁酱吖の 提交于 2021-02-18 19:11:09
问题 When is a cuda thread (or a whole warp), that performs a read from global memory, put to sleep by the scheduler? Let's say I do some computations in the kernel, right after the memory read, that do not depend on the read data. Can these be executed while the data from the global read isn't there yet? 回答1: A memory read by itself does not cause a stall (barring the cases where the LD/ST unit is unavailable). The thread stall will occur when the result of that memory read operation needs to be

Cuda thread scheduling - latency hiding

由 拟墨画扇 提交于 2021-02-18 19:09:53
问题 When is a cuda thread (or a whole warp), that performs a read from global memory, put to sleep by the scheduler? Let's say I do some computations in the kernel, right after the memory read, that do not depend on the read data. Can these be executed while the data from the global read isn't there yet? 回答1: A memory read by itself does not cause a stall (barring the cases where the LD/ST unit is unavailable). The thread stall will occur when the result of that memory read operation needs to be

Cuda thread scheduling - latency hiding

由 ℡╲_俬逩灬. 提交于 2021-02-18 19:06:24
问题 When is a cuda thread (or a whole warp), that performs a read from global memory, put to sleep by the scheduler? Let's say I do some computations in the kernel, right after the memory read, that do not depend on the read data. Can these be executed while the data from the global read isn't there yet? 回答1: A memory read by itself does not cause a stall (barring the cases where the LD/ST unit is unavailable). The thread stall will occur when the result of that memory read operation needs to be

ZStack实践汇 | ZStack+Docker支撑GPU业务实践

由 自古美人都是妖i 提交于 2021-02-18 18:30:51
背景 ZStack所聚焦的IaaS，作为云计算里的底座基石，能够更好的实现物理资源隔离，以及服务器等硬件资源的统一管理，为上层大数据、深度学习Tensorflow等业务提供了稳定可靠的基础环境。 近年来，云计算发展探索出了有别于传统虚拟化、更贴近于业务的PaaS型服务，该类型依赖于docker实现，如K8S等典型的容器云，可以直接从镜像商店下载封装好业务软件的镜像，更加快捷地实现业务部署。 此外，GPU场景也是客户业务的典型场景，相比于CPU的运算特点，在数据分析、深度学习有着明显的优势。 ZStack是如何与容器结合，以IaaS+PaaS的组合拳，为上层业务提供支撑的呢？本篇文章带大家了解一下，如何在ZStack 上部署 centos7.6 虚拟机，在虚拟机里部署docker，以及如何使用nvidia-docker实现在容器里调用GPU的业务场景。 环境 虚机系统：Centos 7.6 虚机内核：Linux 172-18-47-133 3.10.0-957.el7.x86_64 #1 SMP Thu Nov 8 23:39:32 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux docker版本：docker-ce 19.03 nvidia-docker版本：nvidia-docker-1.0.11.x86_64 显卡：RTX6000 Cuda版本

使用PyTorch对音频进行分类

由 百般思念 提交于 2021-02-18 05:18:21
作者 | Aakash 来源 | Medium 编辑 | 代码医生团队 什么是分类问题？ 对对象进行分类就是将其分配给特定的类别。这本质上是一个分类问题是什么，即将输入数据从一组这样的类别，也称为类分配到预定义的类别。 机器学习中的分类问题示例包括：识别手写数字，区分垃圾邮件和非垃圾邮件或 识别核中的不同蛋白质 。 https://www.kaggle.com/c/jovian-pytorch-z2g 使用的数据集 为了演示分类问题的工作原理，将使用 UrbanSound8K数据集 。该数据集包括 10 种类别的城市声音：空调，汽车喇叭，儿童游戏，狗吠，钻探， enginge_idling ， gun_shot ，手提钻，警笛和 street_music 。 https://urbansounddataset.weebly.com/urbansound8k.html 目的是将数据提供给模型（目前可以将其视为黑匣子），并确定模型预测的准确性。 数据集的结构 该数据集可以作为压缩包使用，大小约为 5.6GB 。与某些机器学习数据集不同，此特定数据集中的音频数据与元数据文件夹一起存在于 10 个不同的文件夹中，元数据文件夹包含名为“ UrbanSound8K.csv ”的文件。 D:\DL\ZEROTOGANS\06-URBAN8K-CLASSIFICATION\DATA

Sparse matrix-vector multiplication in CUDA

由 十年热恋 提交于 2021-02-17 20:26:04
问题 I'm trying to implement a matrix-vector Multiplication on GPU (using CUDA). In my C++ code (CPU), I load the matrix as a dense matrix, and then I perform the matrix-vector multiplication using CUDA. I'm also using shared memory to improve the performance. How can I load the matrix in an efficient way, knowing that my matrix is a sparse matrix? Below is my C++ function to load the matrix: int readMatrix( char* filename, float* &matrix, unsigned int *dim = NULL, int majority = ROW_MAJOR ) {

Sparse matrix-vector multiplication in CUDA

由 我的梦境 提交于 2021-02-17 20:22:23
问题 I'm trying to implement a matrix-vector Multiplication on GPU (using CUDA). In my C++ code (CPU), I load the matrix as a dense matrix, and then I perform the matrix-vector multiplication using CUDA. I'm also using shared memory to improve the performance. How can I load the matrix in an efficient way, knowing that my matrix is a sparse matrix? Below is my C++ function to load the matrix: int readMatrix( char* filename, float* &matrix, unsigned int *dim = NULL, int majority = ROW_MAJOR ) {

CUDA - How to work with complex numbers?

由 可紊 提交于 2021-02-17 19:31:25
问题 What CUDA headers should I include in my programme if I want to work with complex numbers and do simple maths operations (addition and multiplication) to these complex double numbers within the kernel itself? In C++ I can multiply a constant number with a complex double> as long as they are both double. However in CUDA I get lots of errors when I try to do simple maths operations to complex double>s whenever it isn't with another complex double>. What am I missing? Thank you! 回答1: The header
订阅 cuda