]> Joshua Wise's Git repositories - snipe.git/blame - README
Initial import of l2c
[snipe.git] / README
CommitLineData
0a24e44d
JW
1README
2------
3
4This compiler is a big long chain of modules that transform l2 code into
5x86_64 assembly.
6
7These modules include:
8
9 * The parser. The parser was mainly brought in from lab 1, and mainly
10 just a straight-forward extension of the l1 parser. We continued to
11 mark expressions, and pass marking through as needed so that we could
12 produce reasonable error messages up through translation stage. We
13 introduced all needed grammar with no shift/reduce conflicts, but for
14 one in the IF/ELSE stage, with a construct such as:
15 if (x)
16 if (z)
17 a
18 else
19 b
20 (indentation intentionally omitted; there are at least two legitimate
21 ways to parse that!)
22 * The typechecker. This module was completely rewritten since lab1. Three
23 checks are instituted: a check to see if the program has misplaced break
24 or continue statements, a check to see that the program returns in all
25 control paths, and a check that all variables are initialized in all
26 control paths before usage.
27
28 The return and break check is essentially implemented per the rules; the
29 only thing of interest for the variable initialization routine is that
30 there is a helper that computes all assigns to extend contexts from
31 block contents. It was determined that returning 2 accumulators from
32 varcheck would lead to returning 17 accumulators, which would lead to
33 1984193248148132 accumulators; and 238547854478 accumulators leads to
34 the foldl, and foldl leads to anger, anger leads to hate, and hate leads
35 to the Dark Side.
36 * The translator is mainly intact; it was determined that the IR will have
37 basic control flow instructions of labels, jumps, and jump if not
38 conditional, which we deemed sufficient to implement all forms of l2
39 control.
40 * The munch module was fully rewritten; we now munch directly to
41 pseudo-x86_64, in that it has temporaries allowed in it as well. We
42 believe that this allows us to generate much more optimal code than
43 munching into three op, converting from three to two, then converting
44 two to x86_64; in particular, we can run liveness on the x86_64
45 directions directly, which makes translation significantly easier (we do
46 not have to worry about mashing necessary registers).
47 * The liveness analyzer was also fully rewritten; it is now fully
48 def-use-succ, giving us very pretty rules, and a lot of very ugly code
49 to mash them together. Luckily, the ugly code need not be touched ever
50 again.
51 * The grapher had about 4 characters of inconsequential change that had
52 the useful property of speeding it up by two orders of magnitude. You
53 need not worry about it.
54 * The orderer and colorer had no changes.
55 * A new module was introduced -- in particular, the solidifier. The
56 solidifier takes pseudo-x86_64 that is annotated with register locations
57 and emits needed spill and unspill operations to get everything into
58 real registers that the x86_64 chips can access.
59 * The peepholer remains pretty simple; redundant moves are optimized out,
60 and hence the code size drops by a factor of 1.5 or so.
61 * The stringifier is of no interest to you, for it does real things that
62 interact with the real world, and that is not of interest to people who
63 write in ML.
64
65We believe that it's fully functional; we have not had a case in quite some
66time that caused us to generate incorrect code (at least, when we should
67generate code). The internal debug mechanisms are very useful; often a
68line-by-line examination of dumps after each translation phase can narrow
69bugs down into single lines of ML code.
This page took 0.031382 seconds and 4 git commands to generate.