1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN" "http://www.w3.org/TR/html4/strict.dtd">
<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=UTF-8">
<title>Mu - 029transforms.cc</title>
<meta name="Generator" content="Vim/8.1">
<meta name="plugin-version" content="vim8.1_v1">
<meta name="syntax" content="cpp">
<meta name="settings" content="number_lines,use_css,pre_wrap,no_foldcolumn,expand_tabs,line_ids,prevent_copy=">
<meta name="colorscheme" content="minimal-light">
<style type="text/css">
<!--
pre { white-space: pre-wrap; font-family: monospace; color: #000000; background-color: #c6c6c6; }
body { font-size:12pt; font-family: monospace; color: #000000; background-color: #c6c6c6; }
a { color:inherit; }
* { font-size:12pt; font-size: 1em; }
.LineNr { }
.Comment { color: #005faf; }
.Delimiter { color: #c000c0; }
-->
</style>
<script type='text/javascript'>
<!--
/* function to open any folds containing a jumped-to line before jumping to it */
function JumpToLine()
{
var lineNum;
lineNum = window.location.hash;
lineNum = lineNum.substr(1); /* strip off '#' */
if (lineNum.indexOf('L') == -1) {
lineNum = 'L'+lineNum;
}
var lineElem = document.getElementById(lineNum);
/* Always jump to new location even if the line was hidden inside a fold, or
* we corrected the raw number to a line ID.
*/
if (lineElem) {
lineElem.scrollIntoView(true);
}
return true;
}
if ('onhashchange' in window) {
window.onhashchange = JumpToLine;
}
-->
</script>
</head>
<body onload='JumpToLine();'>
<a href='https://github.com/akkartik/mu/blob/master/029transforms.cc'>https://github.com/akkartik/mu/blob/master/029transforms.cc</a>
<pre id='vimCodeElement'>
<span id="L1" class="LineNr"> 1 </span><span class="Comment">//: Ordering transforms is a well-known hard problem when building compilers.</span>
<span id="L2" class="LineNr"> 2 </span><span class="Comment">//: In our case we also have the additional notion of layers. The ordering of</span>
<span id="L3" class="LineNr"> 3 </span><span class="Comment">//: layers can have nothing in common with the ordering of transforms when</span>
<span id="L4" class="LineNr"> 4 </span><span class="Comment">//: SubX is tangled and run. This can be confusing for readers, particularly</span>
<span id="L5" class="LineNr"> 5 </span><span class="Comment">//: if later layers start inserting transforms at arbitrary points between</span>
<span id="L6" class="LineNr"> 6 </span><span class="Comment">//: transforms introduced earlier. Over time adding transforms can get harder</span>
<span id="L7" class="LineNr"> 7 </span><span class="Comment">//: and harder, having to meet the constraints of everything that's come</span>
<span id="L8" class="LineNr"> 8 </span><span class="Comment">//: before. It's worth thinking about organization up-front so the ordering is</span>
<span id="L9" class="LineNr"> 9 </span><span class="Comment">//: easy to hold in our heads, and it's obvious where to add a new transform.</span>
<span id="L10" class="LineNr">10 </span><span class="Comment">//: Some constraints:</span>
<span id="L11" class="LineNr">11 </span><span class="Comment">//:</span>
<span id="L12" class="LineNr">12 </span><span class="Comment">//: 1. Layers force us to build SubX bottom-up; since we want to be able to</span>
<span id="L13" class="LineNr">13 </span><span class="Comment">//: build and run SubX after stopping loading at any layer, the overall</span>
<span id="L14" class="LineNr">14 </span><span class="Comment">//: organization has to be to introduce primitives before we start using</span>
<span id="L15" class="LineNr">15 </span><span class="Comment">//: them.</span>
<span id="L16" class="LineNr">16 </span><span class="Comment">//:</span>
<span id="L17" class="LineNr">17 </span><span class="Comment">//: 2. Transforms usually need to be run top-down, converting high-level</span>
<span id="L18" class="LineNr">18 </span><span class="Comment">//: representations to low-level ones so that low-level layers can be</span>
<span id="L19" class="LineNr">19 </span><span class="Comment">//: oblivious to them.</span>
<span id="L20" class="LineNr">20 </span><span class="Comment">//:</span>
<span id="L21" class="LineNr">21 </span><span class="Comment">//: 3. When running we'd often like new representations to be checked before</span>
<span id="L22" class="LineNr">22 </span><span class="Comment">//: they are transformed away. The whole reason for new representations is</span>
<span id="L23" class="LineNr">23 </span><span class="Comment">//: often to add new kinds of automatic checking for our machine code</span>
<span id="L24" class="LineNr">24 </span><span class="Comment">//: programs.</span>
<span id="L25" class="LineNr">25 </span><span class="Comment">//:</span>
<span id="L26" class="LineNr">26 </span><span class="Comment">//: Putting these constraints together, we'll use the following broad</span>
<span id="L27" class="LineNr">27 </span><span class="Comment">//: organization:</span>
<span id="L28" class="LineNr">28 </span><span class="Comment">//:</span>
<span id="L29" class="LineNr">29 </span><span class="Comment">//: a) We'll divide up our transforms into "levels", each level consisting</span>
<span id="L30" class="LineNr">30 </span><span class="Comment">//: of multiple transforms, and dealing in some new set of representational</span>
<span id="L31" class="LineNr">31 </span><span class="Comment">//: ideas. Levels will be added in reverse order to the one their transforms</span>
<span id="L32" class="LineNr">32 </span><span class="Comment">//: will be run in.</span>
<span id="L33" class="LineNr">33 </span><span class="Comment">//:</span>
<span id="L34" class="LineNr">34 </span><span class="Comment">//: To run all transforms:</span>
<span id="L35" class="LineNr">35 </span><span class="Comment">//: Load transforms for level n</span>
<span id="L36" class="LineNr">36 </span><span class="Comment">//: Load transforms for level n-1</span>
<span id="L37" class="LineNr">37 </span><span class="Comment">//: ...</span>
<span id="L38" class="LineNr">38 </span><span class="Comment">//: Load transforms for level 2</span>
<span id="L39" class="LineNr">39 </span><span class="Comment">//: Run code at level 1</span>
<span id="L40" class="LineNr">40 </span><span class="Comment">//:</span>
<span id="L41" class="LineNr">41 </span><span class="Comment">//: b) *Within* a level we'll usually introduce transforms in the order</span>
<span id="L42" class="LineNr">42 </span><span class="Comment">//: they're run in.</span>
<span id="L43" class="LineNr">43 </span><span class="Comment">//:</span>
<span id="L44" class="LineNr">44 </span><span class="Comment">//: To run transforms for level n:</span>
<span id="L45" class="LineNr">45 </span><span class="Comment">//: Perform transform of layer l</span>
<span id="L46" class="LineNr">46 </span><span class="Comment">//: Perform transform of layer l+1</span>
<span id="L47" class="LineNr">47 </span><span class="Comment">//: ...</span>
<span id="L48" class="LineNr">48 </span><span class="Comment">//:</span>
<span id="L49" class="LineNr">49 </span><span class="Comment">//: c) Within a level it's often most natural to introduce a new</span>
<span id="L50" class="LineNr">50 </span><span class="Comment">//: representation by showing how it's transformed to the level below. To</span>
<span id="L51" class="LineNr">51 </span><span class="Comment">//: make such exceptions more obvious checks usually won't be first-class</span>
<span id="L52" class="LineNr">52 </span><span class="Comment">//: transforms; instead code that keeps the program unmodified will run</span>
<span id="L53" class="LineNr">53 </span><span class="Comment">//: within transforms before they mutate the program. As an example:</span>
<span id="L54" class="LineNr">54 </span><span class="Comment">//:</span>
<span id="L55" class="LineNr">55 </span><span class="Comment">//: Layer l introduces a transform</span>
<span id="L56" class="LineNr">56 </span><span class="Comment">//: Layer l+1 adds precondition checks for the transform</span>
<span id="L57" class="LineNr">57 </span><span class="Comment">//:</span>
<span id="L58" class="LineNr">58 </span><span class="Comment">//: This may all seem abstract, but will hopefully make sense over time. The</span>
<span id="L59" class="LineNr">59 </span><span class="Comment">//: goals are basically to always have a working program after any layer, to</span>
<span id="L60" class="LineNr">60 </span><span class="Comment">//: have the order of layers make narrative sense, and to order transforms</span>
<span id="L61" class="LineNr">61 </span><span class="Comment">//: correctly at runtime.</span>
<span id="L62" class="LineNr">62 </span>
<span id="L63" class="LineNr">63 </span><span class="Delimiter">:(before "End One-time Setup")</span>
<span id="L64" class="LineNr">64 </span><span class="Comment">// Begin Transforms</span>
<span id="L65" class="LineNr">65 </span><span class="Comment">// End Transforms</span>
</pre>
</body>
</html>
<!-- vim: set foldmethod=manual : -->
|