Re: [Flashforth-devel] C stack - passing the params (pic24/33)

[pito <pi...@vo...>]

I cannot resist to add the others:

777 s>f 1000 s>f f/ fconstant arg   \   = 0.777  ok<#,ram>
 
arg fsin fs. 7.01144E-1  ok<#,ram>
arg fcos fs. 7.13020E-1  ok<#,ram>
arg ftan fs. 9.83343E-1  ok<#,ram>
arg fsqrt fs. 8.81476E-1  ok<#,ram>
arg fexp fs. 2.17494E0  ok<#,ram>
arg flog fs. -2.52315E-1  ok<#,ram>
arg arg fpow fs. 8.21972E-1  ok<#,ram>

pure C functions called from FF:
bsin 2.26716E3 CPU INSTRs per fsin ok<#,ram>
bcos 3.03178E3 CPU INSTRs per fcos ok<#,ram>
btan 2.90925E3 CPU INSTRs per ftan ok<#,ram>
blog 2.53247E3 CPU INSTRs per flog ok<#,ram>
bexp 3.04007E3 CPU INSTRs per fexp ok<#,ram>
bsqrt 525.103E0 CPU INSTRs per fsqrt ok<#,ram>

7-9x faster than the same with 4 Cfloat primitves and forth.. ;)
Very nice!
Pito.

______________________________________________________________
> Od: "pito" <pi...@vo...>
> Komu: Mikael Nordman <mik...@pp...>, "flashforth-devel" <fla...@li...>
> Datum: 16.07.2014 11:51
> Předmět: Re: [Flashforth-devel] C stack - passing the params (pic24/33)
>
>And a final benchmark this week - the fsin fcos ftan fatan written in forth, and in forth but with 4 Cfloat primitives:
>
>forth:
>bsin 315.614E3 CPU INSTRs per fsin ok<#,ram>
>bcos 317.457E3 CPU INSTRs per fcos ok<#,ram>
>btan 645.600E3 CPU INSTRs per ftan ok<#,ram>
>batan 95.9004E3 CPU INSTRs per fatan ok<#,ram>
>
>Cfloat 4 primitives:
>bsin 20.9120E3 CPU INSTRs per fsin ok<#,ram>
>bcos 20.6356E3 CPU INSTRs per fcos ok<#,ram>
>btan 42.2846E3 CPU INSTRs per ftan ok<#,ram>
>batan 2.11884E3 CPU INSTRs per fatan ok<#,ram>
>
>15x faster for sin cos tan, 45x for atan (a simple algorithm only).
>These trigo functions will be even much faster when called from C directly .. :)
>Pito
>
>______________________________________________________________
>> Od: "pito" <pi...@vo...>
>> Komu: Mikael Nordman <mik...@pp...>, "flashforth-devel" <fla...@li...>
>> Datum: 16.07.2014 11:31
>> Předmět: Re: [Flashforth-devel] C stack - passing the params (pic24/33)
>>
>>4 float primitives added, now the benchmarks (empty loop not subtracted):
>>
>>Cfloats:
>>be 35.0069E0 CPU INSTRs per empty loop ok<#,ram>
>>b+ 279.134E0 CPU INSTRs per f+ ok<#,ram>
>>b- 277.291E0 CPU INSTRs per f- ok<#,ram>
>>b* 171.349E0 CPU INSTRs per f* ok<#,ram>
>>b/ 435.743E0 CPU INSTRs per f/ ok<#,ram>
>>
>>
>>forth:
>>be 35.0069E0 CPU INSTRs per empty loop ok<#,ram>
>>b+ 2.49101E3 CPU INSTRs per f+ ok<#,ram>
>>b- 2.72317E3 CPU INSTRs per f- ok<#,ram>
>>b* 26.6107E3 CPU INSTRs per f* ok<#,ram>
>>b/ 13.2768E3 CPU INSTRs per f/ ok<#,ram>
>>
>>Interestingly the forth's version of f/ is faster than f*... :)
>>
>>From 11x to 190x speedup..
>>
>>Nice!
>>
>>The next step is to add the another 30 floating point functions.. :)
>>
>>Pito
>>
>>______________________________________________________________
>>> Od: "pito" <pi...@vo...>
>>> Komu: Mikael Nordman <mik...@pp...>, "flashforth-devel" <fla...@li...>
>>> Datum: 16.07.2014 10:29
>>> Předmět: Re: [Flashforth-devel] C stack - passing the params (pic24/33)
>>>
>>>Ok, reordered:
>>>
>>>100000. d>f 7 s>f Cf/  ok<#,ram> 14043 18015
>>>f. 14285.7  ok<#,ram>
>>>
>>>1000000. d>f 7 s>f Cf/ fs. 1.42857E5  ok<#,ram>
>>>
>>>forth:
>>>bench/ 13.2768E3 CPU INSTRs per f/ ok<#,ram>
>>>
>>>Cf/:
>>>bench/ 435.743E0 CPU INSTRs per f/ ok<#,ram>
>>>
>>>13276 / 436 =  30x faster
>>>Nice..
>>>Pito.
>>>
>>>
>>>______________________________________________________________
>>>> Od: "pito" <pi...@vo...>
>>>> Komu: Mikael Nordman <mik...@pp...>, "flashforth-devel" <fla...@li...>
>>>> Datum: 16.07.2014 09:49
>>>> Předmět: Re: [Flashforth-devel] C stack - passing the params (pic24/33)
>>>>
>>>>It works basically, but the order of the operands and its words is weird, so I have to swap everything before calling Cfsub.
>>>>So the order of the shorts passed needs to be rearranged..
>>>>
>>>>"30000 - 10000"  = 
>>>>
>>>>10000 s>f swap 30000 s>f swap Cf-  f.  20000.0  ok<#,ram>
>>>>
>>>>However, even with 2 addtional swaps I get following result for Cfsub:
>>>>
>>>>forth version:
>>>>bench- 2.72317E3 CPU INSTRs per f- ok<#,ram>
>>>>
>>>>Cf- version:
>>>>bench- 283.740E0 CPU INSTRs per f- ok<#,ram>
>>>>
>>>>So the C version of the f- is ~10x faster than the forth version. Nice...
>>>>
>>>>Pito
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>______________________________________________________________
>>>>> Od: Mikael Nordman <mik...@pp...>
>>>>> Komu: <pi...@vo...>, <fla...@li...>
>>>>> Datum: 16.07.2014 08:53
>>>>> Předmět: Re: [Flashforth-devel] C stack - passing the params (pic24/33)
>>>>>
>>>>>Probably, but check the xc16 help section. I think it describes the parameter passing.
>>>>>
>>>>>Mike
>>>>>
>>>>>Sent from my LG Mobile
>>>>>
>>>>>------ Original message------
>>>>>
>>>>>From: pito
>>>>>
>>>>>Date: Wed, 16/07/2014 02:52
>>>>>
>>>>>To: flashforth-devel;
>>>>>
>>>>>Subject:[Flashforth-devel] C stack - passing the params (pic24/33)
>>>>>
>>>>>In the C example we pass a single 16bit short param: mov [W14], W0 .extern C4add call _C4add mov W0, [W14] So my current understanding is to pass 4 shorts (2 floats) would mean something like this: ( f1l f1h f2l f2h -- fl fh ) mov [W14--], W0 ; f2h mov [W14--], W1 ; f2l mov [W14--], W2 ; f1h mov [W14], W3 ; f1l .extern Cfadd call _Cfadd mov W0, [W14++] ; fl mov W1, [W14] ; fh Am I correct?? Thanks, P. ------------------------------------------------------------------------------ Want fast and easy access to all the code in your enterprise? Index and search up to 200,000 lines of code with a free copy of Black Duck Code Sight - the same software that powers the world's largest code search on Ohloh, the Black Duck Open Hub! Try it now. http://p.sf.net/sfu/bds _______________________________________________ Flashforth-devel mailing list Fla...@li... https://lists.sourceforge.net/lists/listinfo/flashforth-devel
>>>>>
>>>>
>>>
>>
>