Linking the C28x FPU fastRTS library

As a side note, I did consider it good news that the FPUmathTables section showed an origin of 0x8000 in the .map file, since that is in fact the start of the L03SARAM section I set it to run from.  I just don't think it's right that the section is length 0!  Are there more include files I needed to provide which actually contain the #pragma CODE_SECTION(function_name, "FPUmathTables") directives to place the fastRTS functions in that section?  Or should that have been taken care of from within the .lib file?  Seems I don't fully understand the linking process.

Thank you for the response Vishal!  I think I may still have some issues with my linking, but I understand a bit better now.

I have tried your suggestion, but do not see the same result in my .map file.  Instead, it looks like this:

.text	0	00302929	0000001c	RUN ADDR = 00008428

		00302929	0000001c	rts2800_fpu32_fast_supplement.lib : isqrt_f32.obj (.text)

I do not actually see FPUmathTables.obj anywhere in the .map file at all.  I have used the sin, cos, atan2, isqrt, and sqrt functions in my code, but the only object which seems to show up there is the isqrt_f32.obj.  Could this be because that particular function is in a separate header file from the rest (C28x_FPU_FastRTS.h)?  I do have math.h included in the project as well - is there some other header I'm missing?

If I look at the section of the .map file it shows each address, I found the following few lines where the supplemental library went:

0030291d	_fastRTS_loadstart

00302939	BIOS_init

00302939	_BIOS_init

00302939	_fastRTS_loadend

Does it strike you as odd that the only things called out between _fastRTS_loadstart and _fastRTS_loadend are called BIOS_init?  Thank you again for your help,

Paul

I have also noticed that within the .text section of the .map file, I can find the following line:

00301af5	00000088	rts2800_fpu32.lib : fs_div.obj (.text)

Shouldn't this have been replaced by the div_f32.obj (.text) function from the fastRTS library instead?  It looks like for some reason, the standard library is still being used.

yeah you're right. It should be calling FS$$DIV from the fastRTS since you have declared the link order correctly. Im not sure what is causing this.

Can you trace which .c file is calling this routine and try calling the isqrt function there just to be sure it sees the fastRTS lib....what you can do is set the properties of the .c file to generate C-interlisted assembly(properties->compiler->assembler options->--keep_asm and under Source interlist choose --c_src_interlist) and then check the assembly to see which routine is being called

Vishal,

In one of my source files, at line 88, in order to show both the division and isqrt functions used, I added the following line:

pwm_deadtime_cnt = (Uint16)(1/isqrt(36));

I would expect it to set pwm_deadtime_cnt to 6 in that line, but unfortunately I don't have my target DSP with me right now so I can't check until next week.  I had it generate and keep an assembly file as suggested, although I am pretty ignorant when it comes to reading assembly.  In the assembly file, I never found any instances of FS$$DIV or any kind of 'div' in general.  I did find several instances of isqrt.  Near the beginning of the file, it seems to declare a global variable named _isqrt with the following:

;**************************************************************
;* UNDEFINED EXTERNAL REFERENCES                              *
;**************************************************************
 .global _isqrt

Later in the assembly file, it references that _isqrt symbol in the following set of code:

$C$DW$4 .dwtag  DW_TAG_subprogram, DW_AT_name("isqrt")
 .dwattr $C$DW$4, DW_AT_TI_symbol_name("_isqrt")
 .dwattr $C$DW$4, DW_AT_type(*$C$DW$T$658)
 .dwattr $C$DW$4, DW_AT_declaration
 .dwattr $C$DW$4, DW_AT_external

Finally, in the section of the code which seems to actually perform what I put on line 88 in my source file, the following bit of assembly was generated:

;----------------------------------------------------------------------
;  88 | pwm_deadtime_cnt = (Uint16)(1/isqrt(36));                              
;----------------------------------------------------------------------
        MOVIZ     R0H,#16912            ; [CPU_] |88| 
$C$DW$15 .dwtag  DW_TAG_TI_branch
 .dwattr $C$DW$15, DW_AT_low_pc(0x00)
 .dwattr $C$DW$15, DW_AT_name("_isqrt")
 .dwattr $C$DW$15, DW_AT_TI_call
        LCR       #_isqrt               ; [CPU_] |88| 
        ; call occurs [#_isqrt] ; [] |88| 
 .dwpsn file "../F2833x_ePWM.c",line 90,column 2,is_stmt

I hope you can make sense of this, because it doesn't mean a lot to me! :)  Thank you again for your help,

Paul

Paul,

All that clutter you are seeing around the assembly is the DWARF debug information. In the properties of the .c file you can turn off the debug symbol generation from :

compiler->basic options->debug model->--symdebug:none

Once you rebuild we should be able to see the C and assembly properly.

Vishal,

After rurning off the DWARF debug info, here is the section of code pertaining to the command I added in line 88:

;----------------------------------------------------------------------
;  88 | pwm_deadtime_cnt = (Uint16)(1/isqrt(36));                              
;----------------------------------------------------------------------
        MOVIZ     R0H,#16912            ; [CPU_] |88| 
        MOVL      @_pwmfullduty,XAR4    ; [CPU_] |79| 
        LCR       #_isqrt               ; [CPU_] |88| 
        ; call occurs [#_isqrt] ; [] |88| 
        MOVW      DP,#_pwm_deadtime_cnt ; [CPU_U] 

It still looks pretty much the same as before, without the extra chatter.  I'm not sure what the MOVIZ command is, but otherwise I don't even see where the divide is taking place.  I was still unable to find anything with the letters 'div' in the .asm file.  The isqrt() function is called by referencing an undefined external variable _isqrt.  At the bottom of the document, the following section defines _isqrt as an external global symbol:

;**************************************************************
;* UNDEFINED EXTERNAL REFERENCES                              *
;**************************************************************
 .global _isqrt

Do you know where I can find the external definition of _isqrt?  I'm still not sure how to verify whether or not it is actually using the fastRTS library.

Thanks for the help,

Paul

Sergey,

At a glance I don't see anything wrong with how you are allocating the function calls, so I'm not really sure what would cause your CPU resets.

Are you using the watchdog timer?  I notice that you are disabling interrupts while the interrupt is being serviced.  I wonder if you have another interrupt that is servicing the watchdog, perhaps that one is missing events because the sin() and cos() functions together are taking enough time and disabling the other interrupts for too long.  If that's the case, you could possibly service the watchdog in that same interrupt, instead of a separate routine.

One other thing worth mentioning is that the fastRTS library does have a sincos() function, which calculates both the sin and cos of an angle simultaneously.  I believe it is faster than calling the sin() and cos() functions separately, so making that change may help.

Hope this is helpful!

Paul