8 - Optimization and Parallelization

Question 1

Describe the program transformation Loop interchange (for a sequential program).
How does it affect performance on modern processor architectures? Give a general
rule for when its application should be beneficial for performance, and a general
rule for when it is safely applicable

Answer 1

a

Question 2

void seqfolr( float *x, float *a, float *b, int N )
{
int i;
x[0] = b[0];
for (i=1; i

Answer 2

a

Question 3

#define N 100000
double x[N]; // array of double precision floats
...
for (i=8; i

Answer 3

a

Question 4

Why is it, in general, so hard for C/C++ compilers to statically analyze a given
sequential legacy program and parallelize it automatically?

Answer 4

a

Question 5

Consider the following sequential loop:
#define N 100000
double x[N]; // array of double precision floats
...
for (i=8; i

Answer 5

a

Question 6

What is auto-tuning, where could it be applied, and what is the main motivation for
it? (2p)

Answer 6

a

Question 7

Name and shortly describe two different loop transformations that can improve the
cache hit rate of a loop, and explain why.

Answer 7

a