Question

6 [15 points] Matrix-vector multiplication using CUDA • Design a CUDA program to perform matrix-vector multiplication $c = A \times b$. The size of matrix A is $1K \times 1K$. The size of vectors $b$ and $c$ is $1K \times 1$. Your program should use 1K threads in total. Assume the shared memory is large enough to hold the entire vector $b$. The input matrix and the vector are initially stored in the host memory. Write a pseudo code for the host function and kernel function. Note that your kernel function must use shared memory to store vector $b$. • Assume each element of A, b, c is 4 bytes; data transfer between CPU and GPU is through PCIe whose bandwidth is 16 GB/s in each direction; the clock rate of GPU is 1 GHz; the access latency to global memory and shared memory is 100 clock cycles and 10 clock cycles, respectively; multiply-add operations are overlapped with memory access operations. What is the execution time of your CUDA program in the best case?

          6 [15 points]
Matrix-vector multiplication using CUDA
• Design a CUDA program to perform matrix-vector multiplication $c = A \times b$. The size
of matrix A is $1K \times 1K$. The size of vectors $b$ and $c$ is $1K \times 1$. Your program should
use 1K threads in total. Assume the shared memory is large enough to hold the entire
vector $b$. The input matrix and the vector are initially stored in the host memory.
Write a pseudo code for the host function and kernel function. Note that your kernel
function must use shared memory to store vector $b$.
• Assume each element of A, b, c is 4 bytes; data transfer between CPU and GPU is
through PCIe whose bandwidth is 16 GB/s in each direction; the clock rate of GPU
is 1 GHz; the access latency to global memory and shared memory is 100 clock cycles
and 10 clock cycles, respectively; multiply-add operations are overlapped with memory
access operations. What is the execution time of your CUDA program in the best case?

Show more…

6 15 points matrix vector multiplication using cuda design a cuda program to perform matrix vector multiplication c a times b the size of matrix a is 1k times 1k the size of vectors b and c 55747

Added by Heather D.

Computer Science and Information Technology

Trishna Knowledge Systems 2018 Edition

Instant Answer

Solved by Expert Supreeta N

08/23/2023

Step 1

Size of matrix A = $1K \times 1K = 1024 \times 1024$ elements. Size of vector b = $1K \times 1 = 1024$ elements. Size of vector c = $1K \times 1 = 1024$ elements. Show more…

Show all steps

lock

Thanks for your feedback!

6 [15 points] Matrix-vector multiplication using CUDA • Design a CUDA program to perform matrix-vector multiplication $c = A \times b$. The size of matrix A is $1K \times 1K$. The size of vectors $b$ and $c$ is $1K \times 1$. Your program should use 1K threads in total. Assume the shared memory is large enough to hold the entire vector $b$. The input matrix and the vector are initially stored in the host memory. Write a pseudo code for the host function and kernel function. Note that your kernel function must use shared memory to store vector $b$. • Assume each element of A, b, c is 4 bytes; data transfer between CPU and GPU is through PCIe whose bandwidth is 16 GB/s in each direction; the clock rate of GPU is 1 GHz; the access latency to global memory and shared memory is 100 clock cycles and 10 clock cycles, respectively; multiply-add operations are overlapped with memory access operations. What is the execution time of your CUDA program in the best case?

Play audio

Feedback

Powered by NumerAI

Supreeta N and 74 other subject AP CS educators are ready to help you.

Ask a new question

Labs

Want to see this concept in action?

NEW

Explore this concept interactively to see how it behaves as you change inputs.

View Labs

Key Concepts

Recommended Videos

Recommended Textbooks

Transcript

00:01 Student a simple matrix program in c+ + that generate random matrix a and b of size n into n multiply them and write the result matrix c to the output file.

00:10 So here random values are taken into the matrices a and b.

00:15 So here i have taken the capital n its type is integer and it's a constant and its value i have initialized to 3.

00:22 We can take any value.

00:23 So here i have taken the function void type generate random matrix because here the matrix having the random number.

00:30 So i am taking as argument matrix 2d matrix n by n.

00:35 So here i am taking the 2 follows because of one for the row and second one for the column.

00:39 And here after this generate random number between 0 and 9.

00:44 So in 2d matrix we will call the we will fill the number by the rand function.

00:50 This will generate the random number between 0 and 9 0 up to 9.

00:54 So in this way we will fill the matrix matrix entries and now call the function multiply matrix.

01:03 So here its type is also void here i am using a 3 for loop int i equal to 0 int j equal to 0 and after this cij initialized to 0 and after this i have taken the third loop k equal to 0 as we know very well matrix multiplication is done with the help of a 3 loop.

01:23 So cij will be the new matrix and its value will be filled by multiplying aij.

01:30 So here cij plus aik into bkj.

01:34 So here rows of the first will be multiplied by the column of the second.

01:39 So in this way matrix multiplication will be performed as we know very well to perform the matrix multiplication.

01:47 So here first matrix and here second matrix.

01:53 So here suppose it's a it's b.

01:55 So here rows will be multiplied by the column of this one and here the result will be saved into the matrix c.

02:04 So in this way the matrix multiplication take place.

02:08 So here i have done with the help of a 3 loop.

02:10 So this will calculate the values of cij and all the entries of cij will be filled.

02:16 Now i am calling the main function in the main function seed the random number generator...

Ace is your personal tutor. It breaks down any question with clear steps so you can learn.

Start Using Ace

Continue solving this problem

Create a free account to:

View full step-by-step solution
Ask follow-up questions with Ace AI
Save progress and study later