![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
Ages ago, I thought it would be a brilliant idea to write up stuff on the papers I read, much like I do book reviews, but then I promptly... didn't do it. But it's a new year with new papers, and here's the first for this year's seminar.
Photo: small toad by Scott* (Because tiny toads are adorable and compiler papers notes don't lend themselves to obvious illustration)
Superoptimizer -- A Look at the Smallest Program
Henry Massalin
1987
This is a neat little paper about optimizing assembly code. They took a program and then had the computer try to generate the smallest possible functionally equivalent version. The paper is super short and readable and filled with lots of very clever adding of registers and stuff to avoid program jumps and comparisons. They could get it to optimize only fairly small programs (12 lines of assembly), but it still seemed like a lot of these would be useful compiler optimizations and they're probably in use now.
Anyhow, it's three pages of explanation + two pages of cool examples they found, so if you're looking for a fun little bit of computing to read about to fill out some mind-expanding new year's resolution, this is an easy place to start.
Some questions we had in seminar that I don't know the answers to:
- What was the impact of this paper on modern compilers?
- Do we do any of this while compiling, or make use of the things they found in a preset kind of way?
- Has anyone tried to do this using modern computers / other assembly instruction sets?
- It seemed like there was a lot of adding... would it be possible to make reduced assembly instruction sets on the assumption that they will never be programmed by humans and thus can be super-optimal?
Photo: small toad by Scott* (Because tiny toads are adorable and compiler papers notes don't lend themselves to obvious illustration)
Superoptimizer -- A Look at the Smallest Program
Henry Massalin
1987
This is a neat little paper about optimizing assembly code. They took a program and then had the computer try to generate the smallest possible functionally equivalent version. The paper is super short and readable and filled with lots of very clever adding of registers and stuff to avoid program jumps and comparisons. They could get it to optimize only fairly small programs (12 lines of assembly), but it still seemed like a lot of these would be useful compiler optimizations and they're probably in use now.
Anyhow, it's three pages of explanation + two pages of cool examples they found, so if you're looking for a fun little bit of computing to read about to fill out some mind-expanding new year's resolution, this is an easy place to start.
Some questions we had in seminar that I don't know the answers to:
- What was the impact of this paper on modern compilers?
- Do we do any of this while compiling, or make use of the things they found in a preset kind of way?
- Has anyone tried to do this using modern computers / other assembly instruction sets?
- It seemed like there was a lot of adding... would it be possible to make reduced assembly instruction sets on the assumption that they will never be programmed by humans and thus can be super-optimal?