Modeling the condition code register in register allocation

=?ISO-8859-1?Q?Roland_Lei=DFa?= <roland.leissa@googlemail.com>
Fri, 9 Jan 2009 17:38:40 -0800 (PST)

          From comp.compilers

Related articles
Modeling the condition code register in register allocation roland.leissa@googlemail.com (=?ISO-8859-1?Q?Roland_Lei=DFa?=) (2009-01-09)
Re: Modeling the condition code register in register allocation nightmie@gmail.com (nightmie) (2009-01-11)
Re: Modeling the condition code register in register allocation cdodd@acm.org (Chris Dodd) (2009-01-11)
Re: Modeling the condition code register in register allocation bartc@freeuk.com (Bartc) (2009-01-13)
Re: Modeling the condition code register in register allocation roland.leissa@googlemail.com (=?ISO-8859-1?Q?Roland_Lei=DFa?=) (2009-01-13)
| List of all articles for this month |

From: =?ISO-8859-1?Q?Roland_Lei=DFa?= <roland.leissa@googlemail.com>
Newsgroups: comp.compilers
Date: Fri, 9 Jan 2009 17:38:40 -0800 (PST)
Organization: Compilers Central
Keywords: code, architecture, question
Posted-Date: 10 Jan 2009 10:32:15 EST

Hello everyone,


I am currently writing a small compiler as a hobby project. My
register allocator is basicly working but I have no idea how I could
model the condition code register (rflags) properly. AMD64 is my
current backend. My intermediate representation (SSA based) handles
comparisons and conditional jumps like this at the moment:


a = b + c
d = a > 10
IF d GOTO label1 ELSE GOTO label2
label2
...
label1
...


a, b, c are longs for instance and d is a bool. My allocator finds
something like this:
a -> %rax
b -> %rbx
c -> %rcx
d -> %dl (because d is a bool)


Somehow I can surely find code for this but this will be far from
optimal. I would like code like this:


addq ...
cmpq $10, %rax
jg label1
label2:
...
label1:
...


So %dl is actually not needed. Things get even worse when there is
something like this:


a = b + c
d = a > 10
e' = e + 1
IF d GOTO label1 ELSE GOTO label2
label2
...
label1
...


with
e -> %rsi
e' -> %rsi


The e' = e + 1 can be compiled to incq %rsi which does not touch the
rflags register. When there is e = e + 2 it could be tranlated to
addq $2, %rsi
or
inc %rsi
inc %rsi
or
the instruction is scheduled even before d = a > 10.


The first variant touches rflags, the second doesn't. As one can
easily see this will be tricky to model especially when the
intermediate representation should be as architecture independent as
possible.


Some ideas or related papers?


Cheers,
Roland Leissa



Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.