使用PIC24F进行浮点冒险

取决于算法有多复杂。我使用16位MPU，你可以用它们做很多事情。UART、I2C和SPI都是标准的。除了寄存器较大之外，编程类似。



      以下为原文

    Depends on how complex the algorithm is.
 
I use 16bit mpus and you can do a lot with them.
uart, i2c and spi are standard.
 
Programming is similar except the registers are bigger.

谢谢您的回复。通用设备配置寄存器和振荡器设置如何？这是从18F到24F的一大步吗？



      以下为原文

   
 
Thanks for your reply.
What about general device CONFIG registers and oscillator setup. Is that a big step from 18F to 24F?
 
 

嗨，如果你看16位PIC/DSPIC，计算速度与CPU速度直接相关，因为浮点计算是由软件完成的。除非您需要低功耗（PIC24F），其他16位内核可以更快地执行浮点运算。目前最快的是DSPIC33 EPXXXX设备，当从（- 40到+85 C）使用时，它可以运行到70MIPS。作为比较，PIC24F只有16MIPS，大约4.5慢。NB：默认情况下，XC16将双浮点视为浮点以加速计算。需要更改参数以强制编译器将双浮点处理为64位（XC16编译器手册的3.4.4.1节）。够了，去PIC32 MZ设备，它有一个硬件FPU，因此有一个更快的计算速度/ MHz。



      以下为原文

    Hi,
If you look at 16 bits PIC / dsPIC, the calculation speed is directly related to the CPU speed as the floating point calculations are done by software. Unless you need low power (PIC24F), other 16 bits cores can perform floating point operations much faster. The fastest one being at this moment the dsPIC33EPxxxx devices which can run up to 70MIPS when used from (-40 to +85C). As a comparison PIC24F is only 16MIPs, about 4.5 slower.
NB : by default, XC16 treats double floats as float to accelerate calculation.
You need to change a parameter to force the compiler to treat double floats as 64bits (section 3.4.4.1 of the XC16 Compiler manual).
If their calculation speed is not enough, go for the PIC32MZ devices which have a hardware FPU and hence a much faster calculation speed / MHz.
Regards
 

谢谢你的详细回答！我实际上注意到，XC8默认双/浮点到24位：-）用XC8和8位MCU，我尝试不使用代码配置器来做“系统设置”（振荡器、配置寄存器……），以便对我正在做的事情有更好的感觉。有一个PIC32 MZ的人，像我一样，只与8位的部分（爱好水平）？



      以下为原文

   


Thanks for the very detailed answer!
I actually noted that XC8 defaults double/float to 24 bits :-)
With XC8 and 8-bit MCUs I try to do the "system setup" (oscillator, CONFIG registers...) myself without using the code configurator in order to get a better feeling for what I'm doing. 
In your opinion, how much more difficult would that be with a PIC32MZ for someone like me that has only worked with 8-bit parts (on a hobby level)?  

好的，抱歉问了一百万个问题（这将是最后一个问题，允诺：-）。我检查了你提到的DSPIC33 EP设备的数据表，在第29.2和第29.5节中，他们谈到了“定点和浮点数据的支持”，并在MaseSube网站上说：离子型：固定/浮点…这意味着什么？它是一个适合浮点的汇编指令吗？HTTPS://www. Mous.SE/DaseHeTe/2/268/7000 00 77H-77982.PDFY提前感谢！



      以下为原文

    Ok, sorry for asking a million questions (this will be the last one, promise :-)).
I checked the datasheet for the dsPIC33EP device you mention and in sections 29.2 and 29.5 they talk about "Support for fixed-point and floating-point data" and on the mouser site under specifications it says: Instruction type: fixed / floating point.... What does this mean? Is it an assembly instruction tailored to floating point?
https://www.mouser.se/datasheet/2/268/70000657H-277982.pdf
 
Thanks in advance!
 

它有数字信号控制器。你可以用Q15号来做浮标。为了详细的扩展，下载16位MCU和DSC程序员。我用了线绘制和步进电机控制加上两个额外的寄存器。数学像犯罪和COS通常使用Q15查找表来完成。



      以下为原文

    It's got the digital signal controllers.
You can do floats using q15 numbers.
For a detailed expanation of that, download the 16bit MCU and DSC Programmers ref.
 
I've used that in line drawing and stepper motor control plus it gives you two additional registers.
 
Trig. maths like sin and cos are usually done using q15 look-up tables.

真的？



      以下为原文

   
Really?

你的“Matlab算法”真的需要浮动，或者这只是你写的方式？你试过改写固定点吗？你有没有使用定点工具箱？



      以下为原文

   


Does your "Matlab algorithm" really need floats, or is that just the way you wrote it? Have you tried rewriting it for fixed point? Do you have access to the fixed point toolbox?

真的？“Q15”是“定点”，而不是“浮点”。http://e.维基百科。org／wiki／qy.（数字格式）



      以下为原文

   
Really?

+1.
"q15" is "fixed point", not "floating point".
https://en.wikipedia.org/wiki/Q_(number_format)

你的“Matlab算法”真的需要浮动，或者这只是你写的方式？你试过改写固定点吗？你有没有使用定点工具箱？首先，感谢大家的回答！老实说，当我说我一直用固定点“工作”时，我的意思是，我只是做了简单的单位转换，比如：3.124米到3124毫米或3124000米，然后做加法、减法、乘法运算。我从来没有尝试过“真正的Q15不动点”。我想使用浮点的一个原因是我需要三角函数，但我猜Q15SiN/COS函数可以用在哪里？你通常在这些情况下做什么？（速度不是我关心的问题）-谢谢你指出了“定点工具箱”，我来看看：



      以下为原文

   


Does your "Matlab algorithm" really need floats, or is that just the way you wrote it? Have you tried rewriting it for fixed point? Do you have access to the fixed point toolbox?




First of all, thanks to everyone for your answers!
To be honest, when I said that I'd been "working" with fixed point, what I meant was that I just done simple unit conversions like: 3.124 m to 3124 mm or 3124000 µm and then done addition, subtraction, multiplication with the integer lengts. I've never tried "real q15 fixed point". 
 
One reason why I want to use floating point is that I need trigonometric functions but I guess that there are q15 sin/cos functions available somewhere? What do you normally do in these situations? (speed is not a concern for me :-) )
 
Thanks for pointing out the "fixed point toolbox", I will have a look at it :-)
 

你需要浮点的唯一原因是动态范围。如果定点有足够的范围（可能在适当的时候有合理的缩放），那么你可以在固定点上做任何事情。你可以定义你自己的定点格式，例如Q15、5Q10、10Q21等等，…并编写适当的函数或宏，给出非标准范围和R解决方案。



      以下为原文

    The only reason you NEED floating point is for dynamic range.
 
If fixed point has sufficient range (maybe with judicious scaling at appropriate times) then you can do everything in fixed point.
 
You can define your own fixed point format, e.g. Q15, 5Q10, 10Q21, etc., ... and write appropriate functions or macros, to give non-standard range and rsolution.

然后你的PIC18可以很好地工作：- 8位CPU已经做了软件浮点几十年。你可能会被困在32位浮点/双倍，有24位的精度。



      以下为原文

   
Then your PIC18 could do the job nicely :-) 
8 bit CPUs have been doing software floating point for decades.  You would probably be stuck at 32bit floats/doubles that have 24 bits of precision. 

然后你的PIC18可以很好地工作：- 8位CPU已经做了软件浮点几十年。你可能会被困在32位的浮标/双打中，这有24位精度。这是一个很好的观点。实际上，我想切换到16位CPU的唯一原因是获得64位双倍：



      以下为原文

   
Then your PIC18 could do the job nicely :-) 
8 bit CPUs have been doing software floating point for decades.  You would probably be stuck at 32bit floats/doubles that have 24 bits of precision. 




That's a good point. Actually the only reason I want to switch to a 16-bit CPU is to get 64-bit doubles :-)

谢谢你的小费！我将试着在将来了解更多这件事，但现在它感觉有点超出我目前的技能水平：



      以下为原文

   


Thanks for a tip!
I will try to learn more about this in the future but right now it feels a bit above my current skill level :-)
 

我知道，我应该说是固定的。



      以下为原文

    I know, I should have said fixed.
 
 
 

在汇编程序中编写自己的FP程序并不困难。我十几岁时就为Z80做过这件事。



      以下为原文

   
It's not difficult to code your own FP routines in assembler. I did it for the Z80 when I was a teenager.

在汇编程序中编写自己的FP程序并不困难。我十几岁时就为Z80做过这件事，也许我应该试一试。实际上，我也发现了一些关于使用CORDIC来计算具有固定点的三角函数的好的参考：-）HTTP://www. DrdBbs.COM/CPP/优化数学增强应用-W/207000 044？PGNO＝2



      以下为原文

   
It's not difficult to code your own FP routines in assembler. I did it for the Z80 when I was a teenager.




Maybe I should give it a try then. I actually also found some nice references on using CORDIC to calculate trigonometric functions with fixed point :-)
http://www.drdobbs.com/cpp/optimizing-math-intensive-applications-w/207000448?pgno=2
 

如果你正在进入超越函数，那么查找表会更好，这取决于所需的精度和可用的存储量。苏珊



      以下为原文

    If you are getting into transcendental functions then perhaps a lookup table would be better depending on the accuracy you need for the range and the amount of storage you have available.
Susan

这是个好建议！我猜你可以用线性插值达到很好的精度。



      以下为原文

   
That's a good suggestion! I guess that you can reach quite good accuracy using linear interpolation?