Sale!

Assignment 3 Exploiting the Streaming SIMD Extension (SSE)

$30.00

Category:
Rate this product

COSC 530
Programming Assignment 3
Exploiting the Streaming SIMD Extension (SSE)
Your assignment is to make an efficient dot product (sum of products) function. Below is a conventional dot
product implementation in C.
float dotprod(float x[], float y[], int n) {
int i; float sum;
sum = 0.0;
for (i = 0; i < n; i++) {
sum += x[i] * y[i];
}
return sum;
}
You are required to implement other versions of the dotprod function, where each function is placed in a separate
le. The versions are (1) loop is unrolled by a factor of 4 (ur4 dotprod.c) where you can assume n is an integer
multiple of 4, (2) accumulator expansion is applied on the unrolled loop (ae4 dotprod.c), and (3) SSE intrinsics
are used to process 4 elements of the arrays for each operation within each iteration of the loop (sse dotprod.c).
You may find the following SSE intrinsics useful when implementing your version of sse dotprod.c.
m128
A typedef used to declare a variable that can hold four single-precision floating-point values. m128 variables
can be assigned values from other m128 variables, functions that return an m128 value, or from a four element
float array initializer.
m128 mm load ps(float *p);
Is the prototype for a function that loads four single-precision oating-point values from the address specied in
p and returns the four values as a m128 value. p must be aligned on a 16 byte boundary.
void mm store ps(float *p, m128 v);
Is the prototype for a function that stores the m128 value in v at the address specified in p, which must be
aligned on a 16 byte boundary.
m128 mm add ps( m128 a, m128 b);
Is the prototype for a function that returns the sum of the m128 values in a and b.
m128 mm mul ps( m128 a, m128 b);
Is the prototype for a function that returns the product of the m128 values in a and b.
Use the following C preprocessor command to obtain access to the typedef and the prototypes for the intrinsics.
#include <emmintrin.h>
You can precede an array declaration by the following specification to align the array on a 16 byte boundary.
attribute ((aligned(16)))
You should test the efficiency of your solutions on one of the hydra machines using the test.c and run scripts
that are in the starter package. You should test each dot product implementation with the array sizes of 32768,
65536, 131072, and 262144. You should also plot the timing results as a bar graph, including the results with the
regular implementation of dot product solution shown above and included in reg dotprod.c in the starter package.
1
For your submission, you should create a gzipped tar file of your project directory with the completed ur4 dotprod.c,
ae4 dotprod.c, and sse dotprod.c implementations, as well as a pdf file containing the timing results of your
experiments. Upload this tar file to canvas before due date and time specified at the top of this assignment. Partial
credits will be given for incomplete efforts. However, a program that does not compile or run will get 0 points.
Point breakdown is below:
• ur4 dotprod.c (15)
• ae4 dotprod.c (20)
• sse dotprod.c (50)
• results with bar graph (15)
2

Reviews

There are no reviews yet.

Be the first to review “Assignment 3 Exploiting the Streaming SIMD Extension (SSE)”

Your email address will not be published. Required fields are marked *

Scroll to Top