Name

GPU Programming
EE 4702-1
Midterm Examination
Wednesday, 18 October 2017 11:30–12:20 CDT

Problem 1 (30 pts)

Problem 2 (20 pts)

Problem 3 (25 pts)

Problem 4 (25 pts)

Alias Exam Total (100 pts)

Good Luck!
 vl
Problem 1: [30 pts] The illustration to the right shows a ring pl
surrounding the point at which a line crosses a plane. When
completed the code fragment below will render the ring. The
line is defined by parametric equation P (t) = pl + t × vl. Point
pp is on the plane and vector np is normal to the plane. The
ring is circular and is on the plane.
np pp
(a) Modify the code below so that t is assigned the value for
which pc is on the plane.
Compute t such that P (t) is on the plane.
(b) Modify the code below to render the ring. Use r1 for the inner radius and r2 for the outer radius. Render
it using slices sets of triangles. Use variable pc even if you haven’t yet solved the previous part.

Include: Code to set up the rendering pass, to set the color (red), to set the correct normal,
and to compute and set vertex coordinates.

World::render_ring( pCoor pl, pVect vl, // Line definition.

pCoor pp, pNorm np, // Plane definition.
float r1, float r2 ){ // Ring radii.

float t = ; // Fill In

pCoor pc = pl + t * vl;

const int slices = 100;

const float delta_theta = 2 * M_PI / slices;
for ( int i=0; i<=slices; i++ ) {
const float theta = i * delta_theta;

}
}

2
Problem 2: [20 pts] Recall that in Homework 2 a sphere was rendered using a spiral slice. The coordinates
of the slice were placed in a buffer object and a sphere was rendered by performing multiple rendering passes
each using the same buffer object but a different transformation. Let c denote the number of coordinates in
the buffer object, and s denote the number of slices per sphere.
(a) What is the size of the buffer object, in bytes? Hint: This is an easy question.

Size of buffer object, in bytes. State any assumptions made.

(b) Appearing below is the code rendering a sphere, adapted from the Homework 2 solution. Based on this
code, estimate the amount of data send from the CPU to the GPU to render the sphere, not counting the
buffer object.
pMatrix_Rotation rot_xz( pVect(0,1,0), delta_theta ), rot_yz( pVect(1,0,0), M_PI );
// Code to render one sphere.
glTranslatefv( ball->position );
glScale1f( ball->radius );
glMultTransposeMatrixf( pMatrix_Rotation(ball->orientation) );

for ( int j=0; j<2; j++ ) {

for ( int i=0; i<s/2; i++ ) {
glColor3fv( i & 1 ? color_lsu_spirit_purple : ball->color );
glDrawArrays(GL_TRIANGLE_STRIP,0,c);
glMultTransposeMatrixf(rot_xz);
}
glMultTransposeMatrixf(rot_yz);
}

In terms of s, amount of data sent to GPU to render sphere.

(c) Buffer object BO2 holds coordinates to render an entire sphere, not just a slice, s × c coordinates total.
Buffer object BO1 holds the slice discussed above. Suppose that initially all data is on the CPU. Let n be
the number of sphere to be rendered. In terms of s and c, what is the smallest value of n for which using
BO2 will require less data than BO1?

Break-even value for n:

3
Problem 3: [25 pts] The diagram below shows a texture on the left-hand side and a scene to be rendered
on the right. The scene consists of three triangles with the texture applied. The text in the texture, written
by Mathy Vanhoef, is from the description of a widespread vulnerability in WiFi WPA2 implementations.
See https://www.krackattacks.com, and please, Fall 2017 people, update your OS and wireless access
point firmware. Grading Note: The description of where the text came from was not in the original exam.
The KRACK vulnerability became public two days before the midterm exam.
y
Texture a c
We discovered serious weaknesses
100 We
in WPA2, a protocol that secures all

modern protected Wi-Fi networks. b

An attacker within range of a victim 50 d f

modern protected Wi-Fi networks.

can exploit these weaknesses using
An attacker within range of a victim
key reinstallation attacks (KRACKs). g
e
0 x
0 50 100 150
Object Space Coordinates

(a) Complete the code below so that it renders the triangles with the texture applied as shown. The color
and normal have already been set, just specify vertex and texture coordinates. Use abbreviation glV for
glVertex2f and glT for glTexCoord2f. Only specify x and y components for the vertex coordinates.

Specify texture and vertex coordinates so that the scene is rendered as shown.

glBegin(GL_TRIANGLES);
glNormal3f(0,0,1);
glColor3f(0.5,0.5,0.5);

glV(30,110); // Vertex a

glEnd();

4
Problem 3, continued:
y
Texture a c
We discovered serious weaknesses
100 We
in WPA2, a protocol that secures all

modern protected Wi-Fi networks. b

An attacker within range of a victim 50 d f

modern protected Wi-Fi networks.

can exploit these weaknesses using
An attacker within range of a victim
key reinstallation attacks (KRACKs). g
e
0 x
0 50 100 150
Object Space Coordinates

(b) Suppose that defg in the diagram above is four pixels across and two pixels down. Suppose that texturing
were set up to use only a single mipmap level and nearest filtering. Explain how defg could appear all white
in that case.

At 4 × 2 pixels defg all white because:

(c) Explain how the 4 × 2 rectangle would appear with linear filtering and mipmap levels. Explain the key
points of linear texture filtering.

Appearance with mipap levels and linear filtering.

Key points of filtering.

5
Problem 4: [25 pts] Answer each question below.
 
7
(a) Show a transformation matrix M that will translate a homogeneous coordinate P to P +  6 . (That
  5
7
is, find M such that M P = P +  6 .) Only show nonzero array elements.
5

M=

(b) Describe the “from” and “to” coordinate spaces for the following OpenGL matrices:

The ModelView matrix maps a coordinate from space to space.

The Projection matrix maps a coordinate from space to space.

6
(c) Show two drawings, each consisting of four triangles. The one should be well-suited for rendering using
a triangle strip, for the other a triangle strip should make no difference.

Show a 4-triangle example well-suited to triangle strips and one in which a triangle strip doesn’t help
at all.

(d) Suppose that buffer object o30 has 30 sphere coordinates and buffer object o1k has 1000 sphere coordi-
nates. In both buffer objects the sphere coordinates are well chosen. They will be used to render a sphere
which is fully visible. Note: In the original exam the sizes were 1000 and 10,000.

Describe a situation in which the sphere would look better using o1k than using o30 and describe a
situation in which the sphere would look the same with o1k as with o30. Explain.

Suppose that it takes much more time to render a sphere using o1k than with o30. The vertex, geometry,
and fragment shader stages are running custom written code which could use some tuning.

Which shaders should be tuned to potentially fix the performance problem? Which shaders be ignored?
Explain.