gnupic: Re: [gnupic] LLVM - was Re: [gnupic] gputils development
Subject:
Re: [gnupic] LLVM - was Re: [gnupic] gputils development
From:
"Scott Dattalo" ####@####.####
Date:
15 Nov 2008 17:13:27 -0000
Message-Id: <4364.71.139.34.26.1226769202.squirrel@ruckus.brouhaha.com>
Walter Banks wrote:
> LLVM's approach to standardizing IR doesn't solve this problem.
As a matter of principal, one cannot disagree. (And any of you who are
unfamiliar with Walter should know that he is the most experienced PIC
C-compiler writer on this list... And perhaps the most experienced in the
world! - his opinion probably weighs in the most on this subject.)
In my experience in dabbling with SDCC, I found that the IR (intermediate
representation) is a somewhat inadequate paradigm. Even mostly similar
microcontrollers have different/optimal requirements for the IR. I tried
to mitigate this somewhat by creating a second layer of IR unique to PICs
called pCode (prior to this, SDCC only had an IR called iCode). In
hindsight, I learned this approach is somewhat analogous to gcc's backend
RTL. I started my SDCC efforts about the time gputils linker and object
code approach were becoming. Had I stuck with this project, I would have
extended pCode to the linker. This seems to be the new addition that LLVM
brings to the table.
Somewhat off topic, but I've been thinking about the concept of 'process'
lately. Some people seem to think if you have the write 'process' in place
that the chances of success are increased. But ultimately, success hinges
on execution. Successful execution depends on individuals. And successful
individuals seldom need a process for guidance. The LLVM framework defines
a process for writing compilers. Its success ultimately hinges on those
performing the execution...
Scott
