MVCL Benchmark Results 
Author Message
 MVCL Benchmark Results

Benchmark Environment
9121/440
Very quiet time- Lunch time- 1 other program active- all on-line
    system were active.
All results include the instructions required to setup the moves and
the TIME Macro at end of test.
Times are in microseconds, TIME STCK,....

Two tests were executed, first, MVC  followed by MVCL, next, MVCL
followed by MVC.  

The purpose of two test was to distribute the page faults between each
set of moves.   The initial timing for a 1 char move was to calculate
the overhead required to "prep" each set of moves.

Program issued first move test then next move test to the same memory
area, calculated difference between start of loop and completion
of the TIME macro, then formated results.

                 MVC FIRST

BYTES         MVCL            MVC              DIFF

1             68096          83200             15104
512           85760          86272               512 *
1024          97280         109824             12544
2048         121088         156928             35840
4096         168192         244224             76032
8192         268032         420096            152064
100k        7776512        8014848            238336
512k       40658432       60608256          19949824
1Meg       79623168      117659136          38035968    

                MVCL FIRST

BYTES          MVCL           MVC               DIFF

1             85504          67072             18432
256           93952          71168             22784
512           97792          76288             21504
1024         121344          87040             34304
2048         166656         111360             55296
4096         259584         152576            107008
8192         471296         242176            229120
100k        8292608        6558464           1734144
512k       63651840       34668288          28983552
1Meg      109644288       72687104          36957184  

* Now sure if this is right, however, it was about same ratio in 4
  runs.

OK. What the hell does this prove???   There is overhead associated
in the page faults for both MVC and MVCL(We already knew that) that's
the way a paging system works.

My own choice when I'm moving a block over 256 bytes is the move long.
Why?  I always set up registers to use for divide,multiply, mvcl etc.
no big deal(I also use r13 for my program base register and don't
anybody out there tell me r13 is for the save area, it is only 19
words long and leaves a lot at the end for you to use).

I will attempt to re-test on the weekend on a completely empty system
and post results.  (not this weekend)

Happy looping,
Don Eakin
--
From the keyboard of Don Eakin --- Conversion Associates



Sat, 25 Apr 1998 03:00:00 GMT  
 MVCL Benchmark Results
   It is not clear what you are doing in these tests, nor
is it clear what you expect to achieve with them.

   It is not clear what these results are for.  Are you
saying that it took 68096us to move 1 byte using MVCL,
and 83200us to move 1 byte with MVC?

   If so, then what do the results mean?  MVC takes
only a few micsoseconds, not 83 milliseconds!
MVC moving 512 bytes should take around 500us, not
86us.

   What does MVC first/MVCL first mean?  Are you saying that
the time is measured for both MVC and MVCL together,
and that you placed the MVC code before the MVCL code,
and vice versa?  Why?

   What conclusions do you expect to draw from your tests?
_____________________________________________________________________


        >Benchmark Environment
        >9121/440
        >Very quiet time- Lunch time- 1 other program active- all on-line
        >    system were active.
        >All results include the instructions required to setup the moves and
        >the TIME Macro at end of test.
        >Times are in microseconds, TIME STCK,....

        >Two tests were executed, first, MVC  followed by MVCL, next, MVCL
        >followed by MVC.  
        >  
        >The purpose of two test was to distribute the page faults between each
        >set of moves.   The initial timing for a 1 char move was to calculate
        >the overhead required to "prep" each set of moves.

        >Program issued first move test then next move test to the same memory
        >area, calculated difference between start of loop and completion
        >of the TIME macro, then formated results.

        >                 MVC FIRST
        >                      
        >BYTES         MVCL            MVC              DIFF

        >1             68096          83200             15104
        >512           85760          86272               512 *
[ . . . ]



Sun, 26 Apr 1998 03:00:00 GMT  
 
 [ 2 post ] 

 Relevant Pages 

1. Smalltalk vs C vs C++ benchmark results

2. VisualWorks benchmark results

3. RFC on Benchmark results

4. Toy loop/array benchmark yields surprising results.

5. Some benchmark results

6. latest benchmark results

7. Benchmark Results so far .....

8. Benchmark results so far ....

9. Benchmark Results so far .....

10. Dylan vs. C results for simple benchmark

11. (forwarding) Re: Benchmark Results for CL's onWin32

12. ML benchmark results (Long)

 

 
Powered by phpBB® Forum Software