surprising result with hipe compilation

[Ulf Wiger (AL/EAB)]

From: Mikael Pettersson [mailto:mikpe@REDACTED]

>To summarise: You compile to native code and save it in files,
>and then benchmark it with timer:tc as the first thing you
>do in a fresh system. This gives much worse results (longer
>time taken) than if the code was only in BEAM.

Yes, obviously, but this is not what I did.
I called timer:tc(...) several times and picked the fastest
one.



>If one repeats the timer:tc call, the runtime for both BEAM and
>native code is reduced to normal levels, and native code is
>consistently (for your code) faster than BEAM.

This is not what happens on my machine (a 400 MHz Ultra 10):


[etxuwig@REDACTED]: erlc +native *.erl
[etxuwig@REDACTED]: which erl
/OTP/LXA_11930_R9C_3/bin/erl
[etxuwig@REDACTED]: erl -boot start_clean -pa .
Erlang (BEAM) emulator version 5.3 [hipe] [threads:0]

Eshell V5.3  (abort with ^G)
1> timer:tc(test,run,[]).
{334894,
 {ok,[{name,"file_server"},
      {vsn,"ubf1.0"},
      {types,[{info,{constant,info},[]},
              {description,{constant,description},[]},
              {services,{constant,services},[]},
              {contract,{constant,contract},[]},
              {file,{prim,string},[]},
              {ls,{constant,ls},[]},
              {files,{tuple,[{constant,files},{list,{prim,file}}]},[]},
              {getFile,{tuple,[{constant,get},{prim,file}]},[]},
              {noSuchFile,{constant,noSuchFile},[]}]},
      {transition,{start,[{input,{prim,ls},[{output,{prim,files},start}]},
                          {input,{prim,getFile},
                                 [{output,{prim,binary},start},
                                  {output,{prim,noSuchFile},stop}]}]}},
      {anystate,[{{prim,info},{prim,string}},
                 {{prim,description},{prim,string}},
                 {{prim,contract},{prim,term}}]}]}}
2> element(1,timer:tc(test,run,[])).
10511
3> element(1,timer:tc(test,run,[])).
27083
4> element(1,timer:tc(test,run,[])).
11149
5> element(1,timer:tc(test,run,[])).
10846
6> element(1,timer:tc(test,run,[])).
11326
7> element(1,timer:tc(test,run,[])).
10887
8> element(1,timer:tc(test,run,[])).
10573
9> 
BREAK: (a)bort (c)ontinue (p)roc info (i)nfo (l)oaded
       (v)ersion (k)ill (D)b-tables (d)istribution
a
[etxuwig@REDACTED]: erlc -W *.erl
[etxuwig@REDACTED]: erl -boot start_clean -pa .
Erlang (BEAM) emulator version 5.3 [hipe] [threads:0]

Eshell V5.3  (abort with ^G)
1> timer:tc(test,run,[]).
{36540,
 {ok,[{name,"file_server"},
      {vsn,"ubf1.0"},
      {types,[{info,{constant,info},[]},
              {description,{constant,description},[]},
              {services,{constant,services},[]},
              {contract,{constant,contract},[]},
              {file,{prim,string},[]},
              {ls,{constant,ls},[]},
              {files,{tuple,[{constant,files},{list,{prim,file}}]},[]},
              {getFile,{tuple,[{constant,get},{prim,file}]},[]},
              {noSuchFile,{constant,noSuchFile},[]}]},
      {transition,{start,[{input,{prim,ls},[{output,{prim,files},start}]},
                          {input,{prim,getFile},
                                 [{output,{prim,binary},start},
                                  {output,{prim,noSuchFile},stop}]}]}},
      {anystate,[{{prim,info},{prim,string}},
                 {{prim,description},{prim,string}},
                 {{prim,contract},{prim,term}}]}]}}
2> element(1,timer:tc(test,run,[])).
6054
3> element(1,timer:tc(test,run,[])).
6451
4> element(1,timer:tc(test,run,[])).
5938
5> element(1,timer:tc(test,run,[])).
6226
6> element(1,timer:tc(test,run,[])).
6407
7> 
BREAK: (a)bort (c)ontinue (p)roc info (i)nfo (l)oaded
       (v)ersion (k)ill (D)b-tables (d)istribution
a


When compiled with hipe, the code runs significantly slower.
Dynamic loading of modules is indeed much (10x) slower when
compiled to native code.

I tried running a call trace to find out which other modules 
were used in the test. It was just prim_file.erl and io_lib.erl,
but compiling those two as well (esp. prim_file.erl) made
the test run slightly slower still when compiled with hipe.

/Uffe