00:06 <crc> maw?

08:10 <siraben> phew, finally understood how to compile a small typed functional language to C using closure conversion, lambda lifting, and SSA!

08:12 <siraben> not sure if the techniques needed to target a stack machine would be much different

08:18 <siraben> it's probably better to do register spilling to ensure that every subexpression doesn't have more than n free variables, where n is the number of registers

13:34 <MrMobius> siraben, let us know what you find. I've always thought that was one of the main things that doomed stack machines to being so much slower than register machines

16:23 <siraben> MrMobius: sure

16:25 <siraben> from what I've read, the biggest problem with stack machines is that when the stack gets large, it usually spills over into slower memory. with enough registers this doesn't happen

16:26 <siraben> OTOH if you're register constrained, like the Z80, it may not be as much of a difference (it's still costs more cycles to load from RAM though), given that modern CPUs have dozens of registers the gap is now large, I think

16:41 <MrMobius> siraben, hmm, just thinking about it, the problem wouldnt seem to be large stacks since you need to spill anyway with a register machine when it gets large

16:42 <MrMobius> I mean something like this: FOO @ IF 5 THEN

16:42 <siraben> MrMobius: https://en.wikipedia.org/wiki/Stack_machine#Performance_disadvantages_of_stack_machines , some points aren't as strong as others

16:42 <MrMobius> you're only using one level of stack but you have to spill registers

16:42 <siraben> right

16:43 <siraben> but if you're only using X amount of stack at any moment, might as well be a register machine, no?

16:43 <MrMobius> no

16:44 <MrMobius> if your registers are faster than stack memory, then FOO @ IF 5 THEN will be slower than the equivalent code meant for a register machine since it doesn't have to spill anything

16:44 <MrMobius> since a lot of the speed up or register machines seems to be keeping as much as you can in registers and accessing memory as little as necessary

16:45 <siraben> Right

16:45 <siraben> in FOO @ IF 5 THEN, there's a memory read, though

16:46 <MrMobius> sure but let's assume there's also a memory read in the register machine to make it equal

16:46 <siraben> mov [FOO], reg1; cmp 5, reg1; jeq LBL

16:46 <siraben> something liek that

16:46 <MrMobius> not knowing the size of the stack means you have to spill registers

16:46 <siraben> s/liek/like

16:46 <siraben> Yeah

16:46 <siraben> be back later, got some differential equations homework to do.