Code Generation I: Compiler Construction

Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

G22.

2130-001
Compiler Construction
Lecture 12:
Code Generation I

Mohamed Zahran (aka Z)


[email protected]
semantically
equivalent

+
symbol
table

Requirements Challenges
• Preserve semantic • Problem of generating
meaning of source optimal target
program program is
• Make effective use of undecidable
available resources of • Many subprogroblems
target machine encountered in code
• Code generator itself generation are
must run efficiently computationally
intractable
Main Tasks of Code Generator
• Instruction selection: choosing
appropriate target-machine instructions
to implement the IR statements
• Registers allocation and assignment:
deciding what values to keep in which
registers
• Instruction ordering: deciding in what
order to schedule the execution of
instructions
Design Issues of
a Code Generator
Input
– three-address presentations (quadruples,
triples, …)
– Virtual machine presentations (bytecode,
stack-machine, …)
– Linear presentation (postfix, …)
– Graphical presentation (syntax trees,
DAGs,…)
Design Issues of
a Code Generator
Target program
– Instruction set architecture (RISC, CISC)
– Producing absolute machine-language
program
– Producing relocatable machine-language
program
– Producing assembly language programs
Design Issues of
a Code Generator
Instruction Selection
The complexity of mapping IR program
into code-sequence for target machine
depends on:
– Level of IR (high-level or low-level)
– Nature of instruction set (data type
support)
– Desired quality of generated code (speed
and size)
Design Issues of
a Code Generator
Register Allocation
• Selecting the set of variables that will
reside in registers at each point in the
program
Register Assignment
• Picking the specific register that a
variable will reside in
Design Issues of
a Code Generator
Evaluation Order
– Selecting the order in which computations
are performed
– Affects the efficiency of the target code
– Picking a best order is NP-complete
– Some orders require fewer registers than
others
Simple Target-Machine
• Load/store operations
– LD dst, addr
– ST x, r
• Computation operations
– OP dst, src1, src2
• Jump operations
– BR L
• Conditional jumps
– Bcond r, L
• Byte addressable
• n registers: R0, R1, … Rn-1
Simple Target-Machine
• Addressing modes
– variable name
– a(r) means contents(a + contents(r))
– *a(r) means:
contents(contents(a + contents(r)))
– immediate: #constant (e.g. LD R1, #100)
Simple Target-Machine
Cost
• cost of an instruction = 1 + cost of
operands
• cost of register operand = 0
• cost involving memory and constants = 1
• cost of a program = sum of instruction
costs
Examples
X=Y-Z

b = a[i]
(8-byte elements)

x = *p
More Examples
• a[j] = c
• *p = y
• if X<Y goto L
Generating Code for
Handling the Stack
Size and layout of activation records are
determined by the code generator using
information from symbol table
saves return address constants giving address
at beginning of activation of beginning of activation record of callee
record of callee

CALL callee
transfers control to
target code of
procedure callee

RETURN
Basic Blocks and Flow Graphs
• Graph presentation of intermediate code
• Nodes of the graph are called basic blocks
• Edges indicate which block follows which
other block.
• The graph is useful for doing better job in:
– Register allocation
– Instruction selection
Basic Blocks
• Definition: maximal sequence of
consecutive instructions such that
– Flow of control can only enter the basic
block from the first instruction
– Control leaves the block only at the last
instruction
• Each instruction is assigned to exactly
one basic block
Fist we determine leader instructions:
First we determine leader instructions:
Fist we determine leader instructions:

Basic block starts with a leader instruction


and stops before the following leader instruction.
DAG Representation of
Basic Blocks
• Leaves for initial values of variables
(we may not know the values so we use a0,
b0, …)
• Node for each expression
• Node label is the expression operation
• Next to the node we put the variable(s)
for which the node produced last
definition
• Children of a node consist of nodes
producing last definition of operands
Finding Local Common
Subexpressions
Dead Code Elimination
From the basic block DAG:
• Remove any root node that has no live
variables
• Repeat until no nodes can be removed
So
• Skim: 8.3.3, 8.5.4, 8.5.5, 8.5.6, and 8.5.7
• Read: 8.1 -> 8.5

You might also like