09_Rasterisation
COMP3421
Rasterisation, Particle Systems
The graphics pipeline
Projection
transformation
Illumination
Clipping
Perspective
division
ViewportRasterisation
Texturing
Frame
buffer
Display
Hidden surface
removal
Model-View Transform
Model
Transform
View
Transform
Model
User
Rasterisation
Rasterisation is the process of converting lines
and polygons represented by their vertices into
fragments.
Fragments are like pixels but include color,
depth, texture coordinate. They may also never
make it to the screen due to hidden surface
removal or culling.
Rasterisation
This operation needs to be accurate and
efficient.
For this reason we prefer to use simple integer
calculations.
All are calculations are now in 2D screen space.
Drawing lines
(x0, y0)
(x1, y1)
(x, y)
Drawing lines – bad
double m = (y1-y0)/(double)(x1-x0);
double b = y0 – m * x0;
for (int x = x0; x <= x1; x++) {
int y = (int) Math.round(m * x + b);
drawPixel(x, y);
}
Problems
• Floating point math is slow and
creates rounding errors
•Floating point multiplication, addition
and round for each pixel
• Code does not consider:
•Points are not connected if m > 1
•Divide by zero if x0 == x1 (vertical
lines)
•Doesn’t work if x0 > x1
Example: y = 2x
Incremental – still bad
// incremental algorithm
double m = (y1-y0)/(double)(x1-x0);
double y = y0;
for (int x = x0; x <= x1; x++) { y += m; //one less multiplication drawPixel(x, (int) Math.round(y)); } Bresenham's algorithm We want to draw lines using only integer calculations and avoid multiplications. Such an algorithm is suitable for fast implementation in hardware. The key idea is that calculations are done incrementally, based on the values for the previous pixel. Bresenham's algorithm We shall assume to begin with that the line is in the first octant. I.e. x1 > x0, y1 > y0 and m <= 1 Bresenham’s Idea For each x we work out which pixel we set next The next pixel with the same y value if the line passes below the midpoint between the two pixels Or the next pixel with an increased y value if the line passes above the midpoint between the two pixels Bresenham's algorithm P (xi, yi) M L (xi+1, yi) U (xi+1, yi+1) M1 M2 Pseudocode int y = y0; for (int x = x0; x <= x1; x++) { setPixel(x,y); M = (x + 1, y + 1/2) if (M is below the line) y++ } Testing above/below We’re on the line when: m = y − y0 x − x0 y − y0 = m(x − x0) 0 = m(x − x0)− (y − y0) Testing above/below We’re above the line when: 0 < m(x − x0)− (y − y0) 0 < (h /w)(x − x0)− (y − y0) 0 < h(x − x0)−w(y − y0) 0 < 2h(x − x0)− 2w(y − y0) Testing above/below We call this value F F(x, y) = 2h(x − x0)− 2w(y − y0) F(x, y) < 0 ⇒ (x, y) is below line F(x, y) > 0 ⇒ (x, y) is above line
Midpoints
P (xi, yi)
M M1
M2
Incrementally
F(M ) = 2h(x0 +1− x0)− 2w(y0 + 12 − y0)
= 2h −w
F(M 1) = 2h(x0 + 2 − x0)− 2w(y0 + 12 − y0)
= F(M )+ 2h
F(M 2) = 2h(x0 + 2 − x0)− 2w(y0 + 32 − y0)
= F(M )+ 2h − 2w
Complete
int y = y0;
int w = x1 – x0; int h = y1 – y0;
int F = 2 * h – w;
for (int x = x0; x <= x1; x++) { drawPixel(x,y); if (F < 0) F += 2*h; else { F += 2*(h-w); y++; } } Example x y F 0 0 2 (0,0) (8,5) w = 8 h = 5 int F = 2 * h - w; Example x y F 0 0 2 (0,0) (8,5) w = 8 h = 5 Example x y F 0 0 2 1 1 -4 (0,0) (8,5) w = 8 h = 5 2 * (h - w) = -6 Example x y F 0 0 2 1 1 -4 2 1 6 (0,0) (8,5) w = 8 h = 5 2 * (h - w) = -6 2 * h = 10 Example x y F 0 0 2 1 1 -4 2 1 6 3 2 0 (0,0) (8,5) w = 8 h = 5 2 * (h - w) = -6 2 * h = 10 Example x y F 0 0 2 1 1 -4 2 1 6 3 2 0 4 3 -6(0,0) (8,5) w = 8 h = 5 2 * (h - w) = -6 2 * h = 10 Example x y F 0 0 2 1 1 -4 2 1 6 3 2 0 4 3 -6 5 3 4 6 4 -2 7 4 8 8 5 2 (0,0) (8,5) w = 8 h = 5 2 * (h - w) = -6 2 * h = 10 Relaxing restrictions Lines in the other quadrants can be drawn by symmetrical versions of the algorithm. We need to be careful that drawing from P to Q and from Q to P set the same pixels. Horizontal and vertical lines are common enough to warrant their own optimised code. Polygon filling Determining which pixels are inside a polygon is a matter of applying the edge-crossing test (from week 3) for each possible pixel. Shared edges Pixels on shared edges between polygons need to be draw consistently regardless of the order the polygons are drawn, with no gaps. We adopt a rule: The edge pixels belong to the rightmost and/or upper polygon ie Do not draw rightmost or uppermost edge pixels Scanline algorithm Testing every pixel is very inefficient. We only need to check where the result changes value, i.e. when we cross an edge. We proceed row by row: Calculate intersections incrementally. Sort by x value. Fill runs of pixels between intersections. Active Edge List We keep a list of active edges that overlap the current scanline. Edges are added to the list as we pass the bottom vertex. Edges are removed from the list as we pass the top vertex. The edge intersection is updated incrementally. Edges For each edge in the AEL we store: The x value of its crossing with the current row (initially the bottom x value) The amount the x value changes from row-to- row (1/gradient) The y value of the top vertex. Edge table The (inactive) edge table is a lookup table index on the y-value of the lower vertex of the edge. This allows for fast addition of new edges. Horizontal edges are not added In this list we store the initial values needed in the active edge list as well as the starting y value for the edge. //For every scanline for (y = minY; y <= maxY; y++){ remove all edges that end at y for (Edge e : active) { e.x = e.x + e.inc; } add all edges that start at y – keep list sorted by x for (int i=0; i < active.size; i+=2){ fillPixels(active[i].x, active[i+1].x,y); } } Example y in x inc y out 0 1 -0.25 4 0 5 1 1 0 9 -3 1 0 9 -0.4 5 3 2 -2 4 3 2 2.5 5(0,0) Edge table Example x inc y out 1 -0.25 4 5 1 1 9 -3 1 9 -0.4 5 Active edge list y=0 Example x inc y out 1 -0.25 4 5 1 1 9 -3 1 9 -0.4 5 Active edge list y=0 Example x inc y out 0.75 -0.25 4 8.6 -0.4 5 Active edge list y=1 Example x inc y out 0.75 -0.25 4 8.6 -0.4 5 Active edge list y=1 Example x inc y out 0.5 -0.25 4 8.2 -0.4 5 Active edge list y=2 Example x inc y out 0.5 -0.25 4 8.2 -0.4 5 Active edge list y=2 Example x inc y out 0.25 -0.25 4 2 -2 4 2 2.5 5 7.8 -0.4 5 Active edge list y=3 Example x inc y out 0.25 -0.25 4 2 -2 4 2 2.5 5 7.8 -0.4 5 Active edge list y=3 Example x inc y out 4.5 2.5 5 7.4 -0.4 5 Active edge list y=4 Example x inc y out 4.5 2.5 5 7.4 -0.4 5 Active edge list y=4 Example x inc y out Active edge list y=5 OpenGL OpenGL is optimised for implementation on hardware. Hardware implementations do not work well with variable length lists. So OpenGL enforces polygons to be convex. This means the active edge list always has 2 entries. More complex polygons need to be tessellated into simple convex pieces. Aliasing Lines and polygons drawn with these algorithms tend to look jagged if the pixel size is too large. This is another form of aliasing. Aliasing Lines and polygons drawn with these algorithms tend to look jagged if the pixel size is too large. This is another form of aliasing. Antialiasing There are two basic approaches to eliminating aliasing (antialiasing). Prefiltering is computing exact pixel values geometrically rather than by sampling. Postfiltering is taking samples at a higher resolution (supersampling) and then averaging. Prefiltering 0 0 0 0.2 0.7 0.5 0.1 0.4 0.8 0.9 0.5 0.1 0.5 0.7 0.3 0 0 0 For each pixel, compute the amount occupied and set pixel value to that percentage. Prefiltering 0.9 For each pixel, compute the amount occupied and set pixel value to that percentage. Postfiltering Draw the line at a higher resolution and average (supersampling). Postfiltering Draw the line at a higher resolution and average (supersampling) Postfiltering Draw the line at a higher resolution and average (supersampling). Comparing Prefiltering Postfiltering Weighted postfiltering It is common to apply weights to the samples to favour values in the center of the pixel. 1/16 1/16 1/16 1/16 1/2 1/16 1/16 1/16 1/16 Stochastic sampling Taking supersamples in a grid still tends to produce noticeably regular aliasing effects. Adding small amounts of jitter to the sampled points makes aliasing effects appear as visual noise. Adaptive Sampling Supersampling in large areas of uniform colour is wasteful. Supersampling is most useful in areas of major colour change. Solution: Sample recursively, at finer levels of detail in areas with more colour variance. Adaptive sampling Samples Adaptive sampling Adaptive sampling Antialiasing Prefiltering is most accurate but requires more computation. Postfiltering can be faster. Accuracy depends on how many samples are taken per pixel. More samples means larger memory usage. OpenGL // implementation dependant may not even do anything ☺ gl.glEnable(GL2.GL_LINE_SMOOTH); gl.glHint(GL2.GL_LINE_SMOOTH_HINT,GL2. GL_NICEST); // also requires alpha blending gl.glEnable(GL2.GL_BLEND); gl.glBlendFunc(GL2.GL_SRC_ALPHA, GL2.GL_ONE_MINUS_SRC_ALPHA); OpenGL // full-screen multi-sampling GLCapabilities capabilities = new GLCapabilities(); capabilities.setNumSamples(4); capabilities.setSampleBuffers(tru e); // ... gl.glEnable(GL.GL_MULTISAMPLE); Particle systems Some visual phenomena are best modelled as collections of small particles. Examples: rain, snow, fire, smoke, dust Particle systems Particles are usually represented as small textured quads or point sprites – single vertices with an image attached. They are billboarded, i.e transformed so that they are always face towards the camera. Billboarding Billboarding An approximate form of billboarding can be achieved by having polygons face a plane perpendicular to the camera Billboarding We can apply this approximation by altering the model-view matrix. ix jx kx φx iy jy ky φy iz jz kz φz 0 0 0 1 ⎛ ⎝ ⎜ ⎜ ⎜ ⎜ ⎞ ⎠ ⎟ ⎟ ⎟ ⎟ ⇒ 1 0 0 φx 0 1 0 φy 0 0 1 φz 0 0 0 1 ⎛ ⎝ ⎜ ⎜ ⎜ ⎜ ⎞ ⎠ ⎟ ⎟ ⎟ ⎟ OpenGL float modelview[16]; gl.glPushMatrix(); // get the current modelview matrix gl.glGetFloatv(GL_MODELVIEW_MATRIX , modelview); // modify the matrix billboard(modelview); gl.glLoadMatrixf(modelview) drawObject(gl) gl.glPopMatrix(); Particle systems Particles are created by an emitter object and evolve over time, usually changing position, size, colour. emitter Particle evolution Usually the rules for particle evolution are simple local equations: interpolate from one colour to another over time move with constant speed or acceleration. To simulate many particles it is important these update steps are kept simple and fast. Particles on the GPU Particle systems are well suited to implementation as vertex shaders. The particles can be represented as individual point vertices. A vertex shader can compute the position of each particle at each moment in time. Particle System uniform vec3 vel; uniform float g, t; void main(){ vec3 pos; pos.x = gl_Vertex.x + vel.x*t; pos.y = gl_Vertex.y + vel.y*t + g*t*t; pos.z = gl_Vertex.z + vel.z*t; gl_Position = ModelViewProjectionMatrix*vec4(pos,1); } Exercise Adapt the fireworks example to create a tornado. Solution See code.