For this lab, you are free to choose your own objective(s), selecting those that you think would be most educational and interesting.
I list below some example objectives. You are welcome to choose from among them, or to come up with other ideas of your own. Particularly if you come up with other ideas of your own, it would be wise to discuss them with me in advance, to get a second opinion regarding whether the degree of ambitiousness is appropriate.
If you choose a relatively ambitious objective, a single objective should be sufficient to generate a piece of work worthy of being called an MCS-388 lab report. If you stick with small, incremental objectives, you may need to assemble a few of them together to make a lab's worth. If you are unsure whether the scope of what you are undertaking is appropriate, talk with me. I would also be happy to talk with you as you work through any of these that you choose.
You could do a live-variable analysis to discover those non-parameter local variables that are live at the start of a procedure and report them as being used while potentially uninitialized.
You could report
procedures that can reach their end without passing through
a return
statement.
You could add float
as an alternative to int
and do the appropriate type checking, introduction of conversion
operations, and emission of correct code for the overloaded operators
on each type.
There are lots of possible variations on this, such as:
one-dimensional vs. multi-dimensional
local vs. global
size must be a constant vs. size must be a constant expression vs. size can be any expression
passing by reference vs. by copying vs. not at all
I would suggest you start simple.
The reporting and recovery from syntactic errors is currently rather primitive in most of your compilers. You could improve this.
You could add the boolean operators &&
,
||
, and !
. For the first two, you'll
probably want to do "short-circuit" evaluation, like in the C family
of languages. That is, depending on the value of the left-hand
operand of a &&
or ||
, it may not be
necessary to evaluate the right-hand operand at all.
The LLVM system includes an optimization program, opt
,
which you can use to transform the bitcode produced by your compiler
into a more efficient version. You can use command-line arguments to
select the particular optimization passes your want performed. You can
then use the LLVM disassembler, llvm-dis
, to convert the
optimized program back into human-readable assembly code and write a
report regarding how the optimizations turned out. You could also
measure the performance impact.
Although the LLVM backend does a nice job of optimization, you could also by implement a number of simple optimizations by modifying your AST classes. Many of these would be best done through a transformation pass over the AST prior to the code generation:
constant folding
algebraic identities (addition of zero, multiplication by one)
simplifying control-flow statements with constant tests
avoiding conversions between 1-bit and 32-bit values where possible
If there is anything interesting you didn't get around to in a prior lab, you could do it now.
What is your favorite language feature that is missing? Do you long
for fancier control flow structures? Assignment operators like
+=
in C? Call by reference? Multiple assignment?
Whatever you miss, you are welcome to add!