Hacker News Reader

Comment by gwern | Hacker News Reader

These are very early compilers indeed, but it makes me wonder: a HN comment today reminded me of 'Proebsting's law' that compilers double program speed every 18 years. He got this by comparing a -O0 run with a -02 run and then dividing, but the obvious problem with this is that even if a modern compiler is run with -O0 it's probably going to be in effect optimizing a lot just by all the architectural choices and default code generation. It would be interesting to see if one could take a more recent C compiler, say late '70s and compare it to a heavily optimizing GCC run, and answer the question: how much have C compilers improved over the last 30-40 years?

View on HN

I am pleasantly surprised to see that current GCC seems to be able to generate code for a PDP-11 target. I think there may actually even be cycle-accurate PDP-11 simulators around, which would enable you to do the benchmarking. If you do get around to it, I'd love to hear your results!

Aren't most optimisations hardware dependant, and if so: wouldn't it make sense that the work put into these optimisations depends largely on how much the hardware is used? Is the PDP-11 still a useful target in that sense? Wouldn't the original x86 instruction set be more useful for example?

The situation with x86 specifically is a little weirder than that: http://www.emulators.com/docs/nx06_rmw.htm . It's a bit of a long read, but the summary is that on some x86s, single-instruction adds to memory are faster, and on others, load/increment/store in three instructions is faster.

Compiler optimization be a harsh mistress.

Well, if you want to compare a 1970s compiler to a modern compiler, the x86 instruction set will not be a useful target, since no 1970s compilers target it. These particular ancient compilers target the PDP-11.

The larger question of whether optimizations are hardware-dependent is interesting. In some sense, sure. But there are a lot of optimizations in modern compilers that apply equally well to the PDP-11 and to the AMD64; they're just not that different.

My homework assignment for you: write a C function of three or four lines which loops over an array of structs and does some computation for each one, including some conditional and a call to a one-line static function; compile it with -c -O0 -Wa,-adhlns=unoptimized.lst and with -c -O5 -fomit-frame-pointer -Wa,-adhlns=optimized.lst; list the optimizations you notice in the difference between unoptimized.lst, and hazard a guess at which of them are applicable on a PDP-11.

Reply on news.ycombinator.com