Bodmer
commented
5 years ago Open scrungy opened 5 years ago
Bodmer
commented
5 years ago 
scrungy
commented
5 years ago Well, I guess that is the problem with the variable. Thank you for pointing that out. Allocating a variable and assigning it a value like that down inside of a function, is really not a good thing. Code should be readable, it is for the programmer. Variables should all be created and grouped together.
I'm old school and absolutely HATE it when programmers initialize and create a variable locally like that. They HAVE to be at the top of the function, so you can see where they are. I also trace most functions at the assembly language level. So I move them. I'm a bare metal type embedded programmer and have been doing it since 1977. My first computer was a 1 mhz 6800 with 256 BYTES of ram and 512 BYTES of rom that held Mikbug. 300 baud serial, baud rate adjusted via two pots, that is how old school I am. :) Currently I work for a medical device manufacturer writing code for a hand held micro current TENS unit...
Porting your library to the PIC has been a real bear. The XC16 compiler is NOT a C++ version of the gnu compiler, even though it is the full gnu compiler with microchip tweaks.
So I have had to make extensive changes all over the place, converting variables that were normally in the class to standalone, taking overloaded functions and creating new names for them etc. Had to change the machine code, but then once you do that and try to run it at 247k, the chip selects and data doesn't line up on the logic analyzer, so I pretty much had to hack it hard.
While I rewrote the spiwait16, etc with the correct assembly code for the PIC24, I wound up getting rid of them and using a simpler macro. Does string pasting. The compiler even on optimzation 1 (which is the max that the free version can do) was generating horrible code. This is 4 cycles tops. and yes, it is much slower than it needs to be, but I had to get it working first.
then go back and optimize. To make things like chip select, and command toggles line up on the Salae logic analyzer, I need all the spi data transmits to hang UNTIL THE SHIFT REGISTER IS EMPTY.... I think a lot of this would not have been necessary, if I had realized that the mcc code generator had used enhanced mode. That really caused lots of grief. Once I figured that out, then it still would not work, and that turned out that the Polarity was set wrong. It took weeks to figure all that out.... anyrate, when I get done, I will have a pure C module, for the PIC24 family, with some routines pulled from other libraries merged into it. Specifically, the round gauge routine from sumtoy.
asm volatile ("1: btss "xstr(addr) ",#" xstr(bit) "\n" \ " bra 1b \n ") hang_until_set (SPISTATL,SPITBF);
Now that I got it running, I am testing the functions, and will write some C wrappers to create something that is more logical. Because this is a non C++ compiler, you can't inherit the properties of the print class, so println does not work.
Once I get this all cleaned up and all the debug lines out, I will send it to you, if you give me your email address..... This is my company email account, my personal one is knapper@realtime.net
-- There are TWO type of people in this world
There is a corner case error. If you pass in a number say 3.1415 and set dp = 5, the floating point number does not print the fractional part. This is because it relies on the variable i, which was set earlier. for (i=0;i<dp;++1) rounding /=10.0; leaves i = the number of characters to the right of the decimal point, in this case 5. Then when it gets to this line: while((i<dp) && (digits <9)) the test fails, because dp and i are equal. the solution is to put i=0; in front of the while. Discovered when porting this to a PIC24.