<cfbolz>
arigato: the paper *is* a bit different than the old approaches, has some nice ideas imo. however, I agree with your points about cpython
jcea has joined #pypy
Dejan has joined #pypy
derpydoo has quit [Ping timeout: 276 seconds]
itamarst has joined #pypy
<arigato>
skimmed the paper too now. I agree, it's cool how they generate the stencils. I'm pretty sure that the old approaches used hand-written assembly
<arigato>
couldn't the same idea be used to write a JIT backend for PyPy in a platform-independent way?
<arigato>
probably only partially, because there are some tricks we use that are very hard to express as C code
<korvo>
I thought that the platform-independence came from having a C compiler emit stencils. Or from cutting up LLVM in some way.
<cfbolz>
arigato: yes, that's an excellent idea, someone should do that ;-)
<cfbolz>
arigato: I also liked the point of getting a little bit of register allocation by having variants of stencils