search

Interlisp / medley

The main repo for the Medley Interlisp project. Wiki, Issues are here. Other repositories include maiko (the VM implementation) and Interlisp.github.io (web site sources)
https://Interlisp.org
MIT License
373 stars 19 forks source link

Run benchmarks (with caution ) #1067

Open masinter opened 1 year ago

masinter commented 1 year ago 
          When you get to running Maiko benchmarks, there are a lot of adjustments to make to the compilation options -- I don't turn things up to maximum performance by default.  If you compile for the specific machine you're running on then options like `-mtune=native -march=native` and, if you're using clang, the link-time-optimization and possibly the profile guided optimization may get you a little more.  The native tuning lets it take advantage of the exact instruction set that the processor has rather than generic "x86_64".

Originally posted by @nbriggs in https://github.com/Interlisp/medley/issues/1040#issuecomment-1400783721

I'm not sure what we'll get out of benchmarks to give this a priority but it seemed better to make it a separate issue

masinter commented 11 months ago

There are a couple of issues about benchmarks and I'm going to point them at this one.

How much faster is Medley today than Interlisp-D on a Dorado in 1985?

We should find out.

The book with the data is in the zotero bibliography as Richard P. Gabriel "Performance and Evaluation of LISP Systems". Accessing it from the desktop Zotero app will give you- a searchable PDF, look for "Dorado".

than on a Dorado in 19s85. There should be an issue to re-run the gabriel benchmarks. The book with the data is in the zotero bibliography as Richard P. Gabriel "Performance and Evaluation of LISP Systems" -- a searchable PDF, look for "Dorado".

If you clone the History repository, you can "find . -name gabriel The "1000 times faster" was a ballpark number, so getting some real numbers would be good.

Running the benchmarks before we do releases might be a good idea. I don't know whether BCOMPL/TCOMPL in IL of Interlisp versions of benchmarks vs. CL versions (CL:COMPILE-FILE => DFASL only).

You don't have to type them in. If you clone the History repository:

larry@K4:~/il/history$ find . -iname gabriel
./1990s/1993-envos/users-maiko/lisp/gabriel
./1990s/1993-envos/LISPCORE/gabriel
larry@K4:~/il/history$ diff `find . -iname gabriel`
Only in ./1990s/1993-envos/LISPCORE/gabriel: 00-README.txt
Only in ./1990s/1993-envos/LISPCORE/gabriel: 00-README.txt.~1~
Only in ./1990s/1993-envos/LISPCORE/gabriel: admin
Only in ./1990s/1993-envos/LISPCORE/gabriel: aux
Only in ./1990s/1993-envos/LISPCORE/gabriel: benchmarks
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-OTHER.DFASL
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-OTHER.LCOM
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-TAK.DFASL
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-TAK.LCOM
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-TIMERS
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-TIMERS.dfasl
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: GABRIEL-TIMERS.LCOM
Only in ./1990s/1993-envos/LISPCORE/gabriel: interlisp
Only in ./1990s/1993-envos/users-maiko/lisp/gabriel: README
Only in ./1990s/1993-envos/LISPCORE/gabriel: Results
Only in ./1990s/1993-envos/LISPCORE/gabriel: tools

If you clone the History repository, you can "find . -name gabriel The "1000 times faster" was a ballpark number, so getting some real numbers would be good.

Running the benchmarks before we do releases might be a good idea. I don't know whether BCOMPL/TCOMPL in IL of Interlisp versions of benchmarks vs. CL versions (CL:COMPILE-FILE => DFASL only).

masinter commented 9 months ago

these caveats:

https://github.com/Interlisp/medley/issues/1040#issuecomment-1400783721