write in front
The content is divided into two parts, the first part is translation "Professional CUDA C Programming" section 2. The timing YOUR KERNEL in CUDA programming model, and the second part is his own experience. Experience is not enough, you are welcome to add greatly.
Cuda, the pursuit of speed ratio, want to get accurate time, the timing function is
CUDA 5, CUDAGPU Architecture
SM (Streaming Multiprocessors) is a very important part of the GPU architecture. The concurrency of GPU hardware is determined by SM.
Taking the Fermi architecture as an example, it includes the following main components:
CUDA cores
Shared Memory/L1Cache
Register File
Load/Store Units
Special Function Units
Warp Scheduler
Each SM in the GPU is designed to support hundred
Use Python to write the CUDA program, and use python to write the cuda Program
There are two ways to write a CUDA program using Python:
* Numba* PyCUDA
Numbapro is no longer recommended. It is split and integrated into accelerate and Numba.
Example
Numba
Numba optimizes Python code through the JIT mechanism. Numba can optimize the hardware environment of the Loca
Cuda Memory Model:
GPU chip: Register, shared memory;
Onboard memory: local memory, constant memory, texture memory, texture memory, global memory;
Host memory: host memory, pinned memory.
Register: extremely low access latency;
Basic Unit: register file (32bit/each)
Computing power 1.0/1.1 hardware: 8192/Sm;
Computing power 1.2/1.3 hardware: 16384/Sm;
The register occupied by each thread is limited. Do not assign too many private variables to it dur
CUDA 6, CUDAWarp
Logically, all threads are parallel. However, from the hardware point of view, not all threads can be executed at the same time. Next we will explain some of the essence of warp.Warps and Thread Blocks
Warp is the basic execution unit of SM. A warp contains 32 parallel threads, which are executed in SMIT mode. That is to say, all threads execute the same command, and each thread uses its own data to execute the command.
A block can be
Document directory
Function qualifier
Variable type qualifier
Execute Configuration
Built-in Variables
Time Functions
Synchronous Functions
1. Parallel Computing
1) Single-core command-level parallel ILP-enables the execution unit of a single processor to execute multiple commands simultaneously
2) multi-core parallel TLP-integrate multiple processor cores on one chip to achieve line-level parallel
3) multi-processor parallelism-Install multiple processors on a single circuit board and i
/cuda-6.5/lib64: $ LD_LIBRARY_PATH
My system is 64-bit, so I use lib64. Of course, I will download it again. I forgot to mention it. I should also pay attention to it when downloading the driver for different systems. Otherwise, I will be white-faced,
Then;
~ $ Source. bashrc
Make environment variable configuration effective
3. Compile the sdk sample
The following is not what I do, but what I do is what I d
Run Devicequery error after installing CUDA8.0.
CUDA Device Query (Runtime API) version (Cudart static linking)Cudagetdevicecount returned 35Cuda driver version is insufficient for CUDA runtime versionResult = FAILThere are a lot of ways to find out, Dpkg-l | grep cuda Discovery
There is libcuda1-304, and the libcuda1-375 version is 375.66, above
Cuda register array resolution, cuda register
About cuda register array
When performing Parallel Optimization on some algorithms based on cuda, in order to improve the running speed of the algorithm as much as possible, sometimes we want to use register arrays to make the algorithm fly fast, but the effect is always u
Win10 with CMake 3.5.2 and vs update1 compiling GPU version (Cuda 8.0, CUDNN v5 for Cuda 8.0) Open compile release and debug version with VS 2015 See the example on the net there are three inside the project Folders include (Include directories containing Mxnet,dmlc,mshadow)Lib (contains Libmxnet.dll, libmxnet.lib, put it in vs. compiled)Python (contains a mxnet,setup.py, and build, but the build contains t
Today we have a few gains, successfully running the array summation code: Just add the number of n sumEnvironment: cuda5.0,vs2010#include "cuda_runtime.h"#include "Device_launch_parameters.h"#include cudaerror_t Addwithcuda (int *c, int *a);#define TOTALN 72120#define Blocks_pergrid 32#define THREADS_PERBLOCK 64//2^8__global__ void Sumarray (int *c, int *a)//, int *b){__shared__ unsigned int mycache[threads_perblock];//sets the shared memory within each block threadsperblock==blockdim.xint i = t
, UltraISO, Chinese cabbage and so on.
Download installation packages and drivers
To download the image file:
(1) Download the corresponding Cuda version on the official website, I choose the 7.0 version here, choose Run on it, official address: [Cuda official DOWNLOAD]Http://developer.nvidia.com/cuda-downloads
(2) Download the co
compiler and language improvements for CUDA9
Increased support for C + + 14 with the Cuda 9,NVCC compiler, including new features
A generic lambda expression that uses the Auto keyword instead of the parameter type;
Auto lambda = [] (auto A,auto b) {return a * b;};
The return type of the feature is deducted (using the Auto keyword as the return type, as shown in the previous example)
The CONSTEXPR function can contain fewer restrictions, including var
Install cuda6.5 + vs2012, the operating system is win8.1 version, first of all the next GPU-Z detected a bit:
It can be seen that this video card is a low-end configuration, the key is to look at two:
Shaders = 384, also known as Sm, or the number of core/stream processors. The larger the number, the more parallel threads are executed, and the larger the computing workload per unit time.
Buswidth = 64bit. The larger the value, the faster the data processing speed.
Next let's take a look at the
One, using the GPU module provided in the OPENCV
At present, many GPU functions have been provided in OpenCV, and the GPU modules provided by OPENCV can be used to accelerate most image processing.
Basic use method, please refer to: http://www.cnblogs.com/dwdxdy/p/3244508.html
The advantage of this method is simple, using Gpumat to manage the data transfer between CPU and GPU, and does not need to pay attention to the setting of kernel function call parameter, only need to pay attention to the l
CUDA 3, CUDAPreface
The thread organization form is crucial to the program performance. This blog post mainly introduces the thread organization form in the following situations:
2D grid 2D block
Thread Index
Generally, a matrix is linearly stored in global memory and linear with rows:
In kernel, the unique index of a thread is very useful. To determine the index of a thread, we take 2D as an example:
Thread and block Indexes
Element coordinates
The following is the configuration of vs2005, and vs2003 and vs2008 are similar.
1. Install the Visual Studio 2005 environment.
2. Install the Development Assistant visual assist X.
3. Download Cuda software from the http://www.nvidia.cn/object/cuda_get_cn.html and install it in order.
The following is the configuration of vs2005, and vs2003 and vs2008 are similar.
1. Install the Visual Studio 2005 environment.2. Install the Development Assistant vis
The content source of this page is from Internet, which doesn't represent Alibaba Cloud's opinion;
products and services mentioned on that page don't have any relationship with Alibaba Cloud. If the
content of the page makes you feel confusing, please write us an email, we will handle the problem
within 5 days after receiving your email.
If you find any instances of plagiarism from the community, please send an email to:
info-contact@alibabacloud.com
and provide relevant evidence. A staff member will contact you within 5 working days.