Summation based benchmark for calculators

[Hlib]

max=1000
TI-83plus (ver.1.16) time=105s
TI-85 (ver.9.0) time=131s
internal result=1395.346287744

Code:


:Prompt N:0
:For(X,1,N,1)
:Ans+cube_root(e^(sin(atan(X))))
:End:Ans

[pier4r]

I cannot access the wiki4hp (I know it is up, just the routing somehow is messed up) to update the main version that the gets copied in the first post.

[pier4r]

Updated to post #161

[ijabbott]

CASIO FX-502P - Keystroke program (DSZ loop)

MAX 10: ~18.1s
Result: 13.71183501{70}
(Interesting that it doesn't use the guard digits to round up the 10th digit of the display.)

MAX 100: ~182s
Result: 139.2971870{48}

(It's about half the speed of the FX-602P.)

Code:

001  C6-00   Min 0      ; store max N
002     E0   AC         ; clear accumulator
003  F0-00   LBL 0      ; start of loop
004     E3   +          ; add the following term ...
005  C7-00   MR 0       ; recall N
006  FF-C4   tan^-1
007     C2   sin
008  FF-F9   e^x
009  FF-E2   x^{1/y}
010     03   3          ; this is the end of the term: exp(sin(arctan(N)))^(1/3)
011  FF-01   DSZ        ; decrement N (in M0) and skip if zero
012  F1-00   GOTO 0     ; goto start of loop
013     E5   =          ; final answer

[Jonathan Busby]

Here is my summation code, mostly written in Saturn assembly for the HP48G/GX ( ROM version R ) :

Code:

::
    0LASTOWDOB!
    CK1NOLASTWD
    CK&DISPATCH1
    real    
    ::
        DUP
        % 1000000000
        %>=
        OVER
        % 0 
        %<=
        OR
        IT
        ::
            #203 ERRORSTO AtUserStack ERRJMP
        ;
        COERCE
CODE
speedup     EQU 0
ATANF       EQU #2B6FB
SINF        EQU #2B6E0
EXPF        EQU #2B6AA
GETANGMODE  EQU #2AEF6
!NROOTF     EQU #2B6CE
PUSH%%LOOP  EQU #2A235
ATTNFLG     EQU #807F7
                    

                    
        GOSBVL      =POP#
        GOSBVL      =SAVPTR
        C=A         A
        C=C-1       A
        GONC        no_error
        P=          0
        GOSBVL      =GETPTR
        LA(5)       #203
        GOVLNG      =Errjmp
no_error
        RSTK=C
        IF  speedup
        GOSUB       SPD1
        ENDIF

        GOSBVL  GETANGMODE
        GOSUB   +
        BSS         16
        BSS         16
+       C=RSTK
        D1=C
    
        C=0         W
        DAT1=C      W
        D1=D1+      16
        DAT1=C      W
        D1=D1-      16
        R4=C.F      W
-
        D0=(5)      ATTNFLG
        A=DAT0      A   
        ?A#0        A
        GOYES       onkeyexit1
onkeyexit1
        GONC        +
        GOTO        onkeyexit2
+       A=R4.F      W
        P=          15

        LC(1)       14
        C=C-A       P
        P=C         15
        C=0         W
        C=C+1       P
        SETDEC
        P=          14
        A=A+C       WP
        SETHEX
        GONC        noadjust              
        A=A+1       S
        P=          14
        ASR         WP
        LC(1)       1
        A=A+C       P
noadjust
        C=0         W
        C=A         S
        P=C         15
        C=0         W
        C=P         0



        R4=A.F      W
 
        B=A         W
        B=0         S
        A=C         W    
        P=          0
        SETDEC
        GOSBVL      ATANF
        GOSBVL      SINF
        GOSBVL      EXPF
        SETHEX
        C=0         W
        D=C         W
        P=          14
        LC(1)       3
        D=C         P
        P=          0
        C=0         W
        SETDEC
        GOSBVL      !NROOTF

        SETHEX                  
        C=DAT1      W
        D=C         W
        D1=D1+      16
        C=DAT1      W
        D1=D1-      16
                    
        CDEX        W
        SETDEC
        GOSBVL      =RADDF
        SETHEX
        DAT1=A      W
        C=B         W
        D1=D1+      16
        DAT1=C      W
        D1=D1-      16
        C=RSTK
        C=C-1       A

        GOC         +   
        RSTK=C
        GOTO        -
+


        IF  speedup
        GOSUB       SPD2
        ENDIF
        P=          0
        GOVLNG      PUSH%%LOOP  
onkeyexit2
        IF  speedup
        GOSUB       SPD2
        ENDIF
        P=          0
        GOVLNG      =GETPTRLOOP

SPD1
        GOSBVL      =DispOff
        INTOFF
        RTN
SPD2
        GOSBVL      =DispOn
        INTON
        RTN 
ENDCODE
    ;
;

The results are as follows :

*Display and keyboard scanning turned on*

• 10 : 0.41739 seconds, result = 13.7118350167043
• 100 : 4.31073 seconds, result = 139.297187045924
• 1000 : 39.74079 seconds, result = 1395.34628774339
• 10000 : 376.86231 seconds, result = 13955.8579042908
  

and

*Display and keyboard scanning turned off*

( Same long real numeric results as above )

• 10 : 0.36663 seconds
• 100 : 3.77957 seconds
• 1000 : 34.81678 seconds
• 10000 : 330.17057 seconds
  

I hope this helps.

Jonathan

EDIT : I have attached the HP48G/GX-R binary and the source as well.

[pier4r]

updated until post #165

[ijabbott]

(12-22-2017 09:41 PM)pier4r Wrote:  max = 100
~ 14s - casio fx991 version , 1395.346288 (the slow batch reported by other users) sum function

The reported result is incorrect for that one, it should be 139.2971870.

(N.B. These fx991 listings are for the fx-991EX (classwiz), not for the older fx-991ms, fx-991es, and fx-991es plus.)

[pier4r]

Nice catch. I always hope find some small problems. I'll fix it soon.

[ijabbott]

(02-19-2019 08:20 PM)ijabbott Wrote:  

(12-22-2017 09:41 PM)pier4r Wrote:  max = 100
~ 14s - casio fx991 version , 1395.346288 (the slow batch reported by other users) sum function

The reported result is incorrect for that one, it should be 139.2971870.

(N.B. These fx991 listings are for the fx-991EX (classwiz), not for the older fx-991ms, fx-991es, and fx-991es plus.)

Also, for completeness, the "fast batch" fx-991EX timing for max=100 is ~11s.

[grsbanks]

1000 iterations on a SwissMicros DM10L @ 48MHz:

120s -- result = 1,395.346288

[rprosperi]

(02-21-2019 02:38 PM)grsbanks Wrote:  1000 iterations on a SwissMicros DM10L @ 48MHz:

120s -- result = 1,395.346288

Show off...

[grsbanks]

(02-21-2019 02:46 PM)rprosperi Wrote:  Show off...

Other considerations aside, I'm pleasantly surprised by the speed. It's just marginally faster than a DM11L, itself about 20x the speed of the HP-11C.

[Marc van Lemmen]

Hi,

I am not sure if BIG old calculators are allowed, but I recently bought a HP9821A and I was curious how the HP9821A would perform against the modern calculators. Here are the results :

10 : ~4s result 13.71183502
100 : ~38s result 139.2971870
1000: ~390s result 1395.346288

I'll try my HP9100B later.

Regards,
Marc.

[pier4r]

Everything is allowed that is handheld or it is specifically a calculator rather than a app on a general purpose system.

Or it is an app on a very restricted system mobile without quintillions of apps (like smartphones that were designed to not be tweaked by the users, like nokias, palm, and so on). Although the second category is valid only as loose reference.

So yes the 98x0 and 9100 series are welcomed! I am still impressed by the articles in the hp journal where in practice they say "you can do anything with them" and indeed I guess some did.

edit. wiki 4hp is unreachable for me (but it is up for other sites) so I will try another time. I need to catch up with the last posts.

[pier4r]

updated to post #173 now that I can see http://www.wiki4hp.com (damn routing in my zone!)

Impressive how the 9821A compares with modern day solar powered scientific calculators or the wp 34s . Neat!
How much ram/user storage has it again? I remember the 9100 was around 400 steps (estended to 3000+ with an extension)

[Marc van Lemmen]

My HP9821A has a total of 423 registers ( Option 001 installed, a HP9820 with option 001 has 435 registers according to the manual !). One register is 4 words ( manual ), so that is 3384 bytes. There is also some internal memory, at least 256 words, so the total should be around 4Kb.
I think a HP9820/9821 without option 001 installed has about 2Kb of ram.

By the way, the HP9821A is not RPN but algebraic. Programming method is pre HPL.

My HP9100B is not working fine, he works intermittend. And there is no power or cube-root function, I am working on it but it will take some time. I have a workaround, but less steps, so that would be difficult to compare.

[Marc van Lemmen]

I've got it running on my HP9100B, but with another formula since there is no power function. Math is not my strong point, but i've used exp(ln(exp(sin(atan)))/3) which is al fraction faster than the original formula.
On my HP9821A it gave the following results : 10 ~4 sec. / 100 ~37 sec. / 1000 ~375 sec.


Result for the HP9100B :

10 ~6.5 sec. result 13.71183502
100 ~66 sec. result 139.2971870

Since my Hp9100B stops working after a few minutes I was not able to run it with 1000.

[lrdheat]

I tried the simplest summation, sum of x from x=1 through x=500,000 on my TI 36X Pro, the European edition of the 36X Pro (TI-30X Pro MathPrint), my version A Prime from 5 or 6 years ago?, and the Prime G2 that I got Sunday at BB in Laredo!

36X: ~1 hour 45 minutes
30X: 18 minutes 45 seconds
Prime A: ~8 seconds
Prime G2: ~3.5 seconds (of course, in CAS, the answer appears instantly)

[ijabbott]

(07-09-2019 02:02 AM)lrdheat Wrote:  the European edition of the 36X Pro (TI-30X Pro MathPrint)

Just to be clear, there are two TI-30X Pro models: the older "MultiView" model (which is the same as the TI-36X Pro), and the newer "MathPrint" model (no American version yet).

[Gamo]

Casio fx-991EX
n=1000
t~ 2 min 15 sec
Result=1395.3463

Gamo