Debian Codegraph @

xgc/Written/FilledRects

raw
1 General concerns
2 ----------------
3
4 Machine type
5 Display type
6 Display on different machine?
7 Visual
8 Clipping
9 Obscured by cursor
10 Background tiling
11 The whole software vs. hardware bit: computational overhead vs. time spent
12 drawing
13 Planemask
14
15 PolyFillRectangle
16 -----------------
17 Use same test as PolyRectangle?
18 Perhaps give FillPoly some of these too, see if it special cases them.
19 Fill Rectangle vs. Fill Poly for same size areas would be an
20 interesting comparison.
21
22 Things to look for:
23 All it's really doing is setting or unsetting planes.
24 When a whole word is filled, it should be quite fast (just
25 setting it to 0 or ~0).
26 Otherwise, the speed of a single rectangle should be about
27 equal to its size.
28
29 Data:
30 Pixels per second.
31 Average size of rectangle.
32
33 GC fields:
34 function:
35 Tiling: likely to be a lot slower, although for whole words
36 you're just setting it to a certain value, rather than 0.
37 (You also have to look at the y value). Strange width
38 tiles (are these allowed?) are likely to be real bad.
39 Stippling: also slower - I'm not sure whether or not this would
40 be slower than tiling. This probably has a greater
41 chance of being done in hardware.
42
43 plane-mask:
44 plane of 0 would be real fast, since it doesn't modify
45 anything. plane of ~0 should be fast, since it doesn't
46 have to think. Anything else would probably be
47 proportional to the number of planes...
48
49 subwindow-mode: I don't believe it matters.
50
51 clip-x-origin
52 clip-y-origin
53 clip-mask:
54 Clipping in general should probably increase speed rather
55 than decrease it, unless the clipping region is strange.
56 Possible increase in speed because less change to the
57 display is being made; possible decrease in speed because
58 more thought has to be given to what is drawn or not.
59 Software vs. hardware, I suppose.
60
61 Possibly
62 foreground: similar to plane-mask in its effect, I would think.
63
64 tile: "nice" width tiles would probably be better. A width which
65 is not a divisor of the length of a word would be
66 significantly slower. A width the size of a word is
67 probably perfect. The actual contents of the tile
68 probably don't matter: Any code to check for tiles
69 which could be "simplified" would probably take too much
70 time to execute to do any good.
71
72 stipple: like tile
73
74 Other:
75 alignment on word boundaries: fastest when one is changing whole
76 words at a time. That way, you're just setting whole words to
77 0 or ~0 and not doing any computation. A 32x1 rectangle should
78 be much faster than a 1x32 rectangle.
79
80 ---------
81 Things to put in test:
82
83 Check all alignments.
84 Perhaps some sort of comparison between thin rectangles oriented
85 horizontally and oriented vertically.
86 Small rectangles, big rectangles: compare pixels/second for each.
87 Good way of computing overhead costs.
88
89 ---------
90 test
91 (1-32)x32 at all 32 bits in a word.
92 (97-128)x32
93 (97-128)x128