summary refs log tree commit diff
path: root/doc/llvm.txt
blob: cb18b58d05300694324e5753d4f4ee31f63f5718 (plain) (blame)
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
                      ===========
                      QBE vs LLVM
                      ===========


Both QBE and LLVM are compiler backends using an SSA
representation.  This document will explain why LLVM
does not make QBE a redundant project.  Obviously,
everything following is biased, because written by me.

- Scope
-------

QBE is a much smaller scale project with different goals
than LLVM.

  * QBE is for amateur language designers.

    It does not address all the problems faced when
    conceiving an industry-grade language.  If you are
    toying with some language ideas, using LLVM will
    be like hauling your backpack with a truck, but
    using QBE will feel more like riding a bicycle.

  * QBE is about the first 70%, not the last 30%.

    It attempts to pinpoint, in the extremely vast
    compilation literature, the optimizations that get
    you 70% of the performance in 10% of the code of
    full blown compilers.

    For example, copy propagation on SSA form is
    implemented in 160 lines of code in QBE!

  * QBE is extremely hackable.

    First, it is, and will remain, a small project
    (less than 8 kloc).  Second, it is programmed in
    non-fancy C99 without any dependencies.  Third,
    it is able to dump the IL and debug information in
    a uniform format after each pass.

    On my Core 2 Duo machine, QBE compiles in half a
    second (without optimizations).

- Features
----------

LLVM is definitely more packed with features, but there
are a few things provided in QBE to consider.

  * LLVM does NOT provide full C compatibility for you.

    In more technical terms, any language that provides
    good C compatibility and uses LLVM as a backend
    needs to reimplement large chunks of the ABI in
    its frontend!  This well known issue in the LLVM
    community causes a great deal of duplication
    and bugs.

    Implementing a complete C ABI (with struct arguments
    and returns) is incredibly tricky, and not really
    a lot of fun.  QBE provides you with IL operations
    to call in (and be called by) C with no pain.
    Moreover the ABI implementation in QBE has been
    thoroughly tested by fuzzing and manual tests.

  * LLVM IL is more cluttered with memory operations.

    Implementing SSA construction is hard.  To save its
    users from having to implement it, LLVM provides
    stack slots.  This means that one increment of
    a variable `v` will be composed of three LLVM
    instructions: one load, one add, and one store.

    QBE provides simple non-SSA temporaries, so
    incrementing `v` is simply done with one instruction
    `%v =w add %v, 1`.

    This could seem cosmetic, but dividing the size of
    the IL by three makes it easier for the frontend
    writers to spot bugs in the generated code.

  * LLVM IL is more cluttered with type annotations and
    casts.

    For the sake of advanced optimizations and
    correctness, LLVM has complex IL types.  However,
    only a few types are really first class and many
    operations of source languages require casts to be
    compiled.

    Because QBE makes a much lighter use of types, the
    IL is more readable and shorter.  It can of course be
    argued back that the correctness of QBE is jeoparadized,
    but remember that, in practice, the large amount
    of casts necessary in LLVM IL is undermining the
    overall effectiveness of the type system.