程序代写代做代考 compiler GPU c/c++ cuda CS402: Lab Session 8

const int N = 16 ;
const int b l o c k s i z e = 16 ;

g l o b a l
void h e l l o (char ∗a , int ∗b) {

a [ threadIdx . x ] += b [ threadIdx . x ] ;
}

int main ( ) {
char a [N] = ”He l lo \0\0\0\0\0\0” ;
int b [N] = {15 , 10 , 6 , 0 , −11, 1 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , 0} ;
char ∗ad ;
int ∗bd ;
const int c s i z e = N∗ s izeof (char ) ;
const int i s i z e = N∗ s izeof ( int ) ;

p r i n t f ( ”%s ” , a ) ;

cudaMalloc ( (void ∗∗)&ad , c s i z e ) ;
cudaMalloc ( (void ∗∗)&bd , i s i z e ) ;
cudaMemcpy( ad , a , c s i z e , cudaMemcpyHostToDevice ) ;

1

http://docs.nvidia.com/cuda/pdf/CUDA_C_Programming_Guide.pdf

cudaMemcpy( bd , b , i s i z e , cudaMemcpyHostToDevice ) ;

dim3 dimBlock ( b l o ck s i z e , 1 ) ;
dim3 dimGrid ( 1 , 1 ) ;
h e l l o<<>>(ad , bd ) ;
cudaMemcpy( a , ad , c s i z e , cudaMemcpyDeviceToHost ) ;
cudaFree ( ad ) ;
cudaFree ( bd ) ;

p r i n t f ( ”%s \n” , a ) ;
return 0 ;

}

Listing 1: CUDA “Hello World” example

In this case we create an array, a, equal to {‘H’,‘e’,‘l’,‘l’,‘o’,‘ ’},
then we use CUDA to transform this into {‘W’,‘o’,‘r’,‘l’,‘d’,‘!’} using
the fact that we can add, for example, 15 to the ASCII value of “H” to create
a “W”. The difference values are stored in the array b. You can find this code
in Listing 1 and the file named helloworldCUDA.cu.

CUDA code is not as easy to compile as the examples in previous labs and
as such requires two steps. The first involves setting up the environment for
the compiler, by using the source command on the file named environment.sh
which is included as part of this lab.

source environment.sh

If you open this file you will see that it appends to the environmental vari-
ables $PATH and $LD LIBRARY PATH with the location of the CUDA compiler
and CUDA libraries respectively. The second step is the compilation process,
which differs from the compilation of plain C in that you must use the nvcc com-
mand. This is a wrapper around gcc which takes care of the CUDA language
extensions.

nvcc -o helloworldCUDA.b helloworldCUDA.cu

Run the compiled binary just like any other, and you will be greeted with
the output below.

Hello World!

The following list contains explanations for the CUDA specific portions of
the code.

• global – This annotation is used to declare a kernel.

• cudaMalloc – This is similar to malloc in C but allocates memory on the
GPU.

• cudaMemcpy – This copies data from hosts memory to device memory.

• hello<<>>(ad,bd) – This invokes the kernel in
much the same was as a function is called in C. The main difference is the
execution configuration within the triple chevrons. The first argument in
the execution configuration specifies the number of thread blocks in the
grid, and the second specifies the number of threads per thread block.

2

• threadIdx.x – The ID of the thread executing the kernel.

As always, if you have any questions at this point, make sure you ask one of
the tutors for help.

3 Using Threads

Task 1 This task will be carried out in the context of a Single-precision A*X
Plus Y (SAXPY) example written in CUDA, which can be found in Listing 2
and the file named saxpy.cu. Your objective is to use the knowledge gained
from the lectures and the CUDA programming guide to complete the kernel; the
key is in understanding the thread hierarchy and how these relate to threadIdx,
blockIdx and blockDim. You will know if you have completed the task correctly
as the reported error value will reduce from 2.0 to 0.0.

#include