Benchmarking temporary Lisp objects [Was: Re: [RFC] temporary Lisp_Strings]

[Dmitry Antipov]

To whom it may concern, I decided to start from conses and vectors
just because this is simpler than strings.  My experimental stuff
is attached, and typical benchmark output may be something like:

Average of 100: stack allocation is 6.879408 times faster for conses
Average of 1000: stack allocation is 7.390145 times faster for conses
Average of 10000: stack allocation is 4.154356 times faster for conses
Average of 100000: stack allocation is 2.417694 times faster for conses

Average of 10: stack allocation is 11.535833 times faster for vectors
Average of 100: stack allocation is 4.731400 times faster for vectors
Average of 1000: stack allocation is 1.443268 times faster for vectors

Note that this is sustained allocation speed; since we're unlikely to
allocate large amounts of Lisp data on stack, real speedup may vary.

Questions:
1) Should alloca_vector fallback to Fmake_vector for large vectors?
2) Is there a way to rewrite alloca_xxx macros to avoid statement
   expressions?  IIUC this is not portable beyond gcc and compilers
   that mimics it (clang, Intel's icc).

Dmitry

cons_vector_benchmark.patch
Description: Text Data

cons-vector-benchmark.el
Description: Text Data