Summation based benchmark for calculators

[pier4r]

Update (again after several months) to post #245.

A kind user sent me a PM with those values

Quote:Here are a few additional results for DB48X v0.7.1, where:

HW7 = Hardware accelerated 7 digits precision (32-bit IEEE-754)
HW16 = Hardware accelerated, 16 digits precision (64-bit IEEE-754)
VPnn = Variable precision decimal with nn digits.

...snip... would find it interesting to see that on a single implementation the value you get does not really converge depending on the underlying precision:

7 digits: 1384348,...
16 digits: 1395612,...
24 digits: 1395600,...
36 digits: 1395607,...
48 digits: 1395609,...


1 million loops:

Code:


| Version | Time (ms) | Result |
|----------------|-----------|---------------------------------------------|
| DM32 HW7 | 1748791 | 1'384'348.25 |
| DM32 HW16 | 2188113 | 1'395'612.15872'53834'6 |
| DM42 HW7 | 605102 | 1'384'348.25 |
| DM42 HW16 | 806730 | 1'395'612.15872'53834'6 |
| iPhone 12 VP48 | 1515800 | 1'395'609.35516'01878'23166'99926'37353'94572'23162'39251'9 |
| iPhone 12 VP36 | 1003884 | 1'395'607.17477'7643'16731'54275'27942'243 |
| iPhone 12 VP24 | 583237 | 1'395'600.94680'92149'84730'06 |
| iPhone 12 HW7 | 9462 | 1'384'348.25 |
| iPhone 12 HW16 | 9495 | 1'395'612.15872'53836'9 |

1000 loops on DM32/DM42:

Code:


| DM32 Version | HW7 | HW16 | VP6 | VP12 | VP24 | VP36 |
|--------------|------|------|------|-------|-------|-------|
| 0.6.4 | 1414 | 1719 | 6905 | 13720 | 32346 | 60259 |
| 0.6.2 | | | 7436 | 16017 | 34898 | 62012 |
| 0.6.0 (Note) | | | | | 23773 | |
| 0.5.2 (ID) | 2154 | | | | | |

Code:


| DM42 Version | HW7 | HW16 | VP6 | VP12 | VP24 | VP36 |
|--------------|------|------|------|-------|-------|-------|
| 0.6.4 | 422 | 705 | 5623 | 10548 | 23811 | 42363 |
| 0.6.2 | | | 5842 | 10782 | 23714 | 42269 |
| 0.6.0 (Note) | | | | | 17685 | |
| 0.5.2 (ID) | 1434 | | | | | |

Please post those directly in the thread because if I am practially inactive (as I am more or less since 2019) those (maybe) interesting values could go lost.
Now going to add those too in the main post.

edit: updated to this post (#248)