• State Resolved
  • Locked Locked
  • Replies 5 replies
  • Subscribers 81 subscribers
  • Views 1466 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Not using TMU for Floating Point Divide

Adam D. Reynolds
Guru 12800 points


I'm using the F28377D with the compiler options suggested by this post:

http://e2e.ti.com/support/microcontrollers/c2000/f/171/p/333174/1162805.aspx#1162805

But when I take a look at the map file it is still using the divide code from rts2800_fpu32 or rts2800_fpu32_fast_supplement if I include that library.  I've included a piece of my map file.

00082682    00000088     rts2800_fpu32.lib : fs_div.obj (.text)
0008270a    00000068                       : exp.obj (.text)
00082772    00000061                       : log.obj (.text)
000827d3    00000031                       : ldexp.obj (.text)
00082804    00000016                       : frexp.obj (.text)

Shouldn't the compiler use the divide available from the TMU instead of the one provided in the library?

  • 0
    TI__Mastermind 20850 points

    Hi Adam,

    Could you take a look at the disassembly and see what is being used? I tried the following with tmu_support set to tmu0 and fp_mode to relaxed

    float a = 12.0, b = 4.0, c = 0.0, d = 0.0, e = 0.0;

    d = sin(a);
    e = cos(b);
    c = a/b;

    It seems to work, all the TMU instructions are being used instead of the std lib calls.

     

    ___relaxed_sin:
        ADDB      SP,#2                 ; [CPU_U] 
        MOV32     *-SP[2],R0H           ; [CPU_] |258| 
;----------------------------------------------------------------------
; 259 | return __sin(x);                                                       
;----------------------------------------------------------------------
        DIV2PIF32 R0H,R0H               ; [CPU_] |259| 
        NOP       ; [CPU_] 
        SINPUF32  R0H,R0H               ; [CPU_] |259| 
        NOP       ; [CPU_] 
        SUBB      SP,#2                 ; [CPU_U] 
        LRETR     ; [CPU_] 
        ; return occurs ; [] 
	.sect	".text"
	.clink
;----------------------------------------------------------------------
; 267 | __inline double __relaxed_cos(double x)                                
;----------------------------------------------------------------------

;***************************************************************
;* FNAME: ___relaxed_cos                FR SIZE:   2           *
;*                                                             *
;* FUNCTION ENVIRONMENT                                        *
;*                                                             *
;* FUNCTION PROPERTIES                                         *
;*                            0 Parameter,  2 Auto,  0 SOE     *
;***************************************************************

___relaxed_cos:
        ADDB      SP,#2                 ; [CPU_U] 
        MOV32     *-SP[2],R0H           ; [CPU_] |268| 
;----------------------------------------------------------------------
; 269 | return __cos(x);                                                       
;----------------------------------------------------------------------
        DIV2PIF32 R0H,R0H               ; [CPU_] |269| 
        NOP       ; [CPU_] 
        COSPUF32  R0H,R0H               ; [CPU_] |269| 
        NOP       ; [CPU_] 
        SUBB      SP,#2                 ; [CPU_U] 
        LRETR     ; [CPU_] 
        ; return occurs ; [] 
	.sect	".text"
	.clink
	.global	_main
;----------------------------------------------------------------------
; 100 | void main(void)                                                        
; 102 | uint16_t        i;                                                     
;----------------------------------------------------------------------

;***************************************************************
;* FNAME: _main                         FR SIZE:  10           *
;*                                                             *
;* FUNCTION ENVIRONMENT                                        *
;*                                                             *
;* FUNCTION PROPERTIES                                         *
;*                            0 Parameter, 10 Auto,  0 SOE     *
;***************************************************************

_main:
        ADDB      SP,#10                ; [CPU_U] 
;----------------------------------------------------------------------
; 104 | float a = 12.0, b = 4.0, c = 0.0, d = 0.0, e = 0.0;                    
;----------------------------------------------------------------------
        MOVIZ     R0H,#16704            ; [CPU_] |104| 
        MOV32     *-SP[2],R0H           ; [CPU_] |104| 
        MOVIZ     R0H,#16512            ; [CPU_] |104| 
        MOV32     *-SP[4],R0H           ; [CPU_] |104| 
        ZERO      R0H                   ; [CPU_] |104| 
        MOV32     *-SP[6],R0H           ; [CPU_] |104| 
        MOV32     *-SP[8],R0H           ; [CPU_] |104| 
        MOV32     *-SP[10],R0H          ; [CPU_] |104| 
;----------------------------------------------------------------------
; 105 | d = sin(a);                                                            
;----------------------------------------------------------------------
        MOV32     R0H,*-SP[2]           ; [CPU_] |105| 
        LCR       #___relaxed_sin       ; [CPU_] |105| 
        ; call occurs [#___relaxed_sin] ; [] |105| 
        MOV32     *-SP[8],R0H           ; [CPU_] |105| 
;----------------------------------------------------------------------
; 106 | e = cos(b);                                                            
;----------------------------------------------------------------------
        MOV32     R0H,*-SP[4]           ; [CPU_] |106| 
        LCR       #___relaxed_cos       ; [CPU_] |106| 
        ; call occurs [#___relaxed_cos] ; [] |106| 
        MOV32     *-SP[10],R0H          ; [CPU_] |106| 
;----------------------------------------------------------------------
; 107 | c = a/b;                                                               
;----------------------------------------------------------------------
        MOV32     R1H,*-SP[2]           ; [CPU_] |107| 
        MOV32     R0H,*-SP[4]           ; [CPU_] |107| 
        DIVF32    R0H,R1H,R0H           ; [CPU_] |107| 
        NOP       ; [CPU_] 
        NOP       ; [CPU_] 
        NOP       ; [CPU_] 
        NOP       ; [CPU_] 
        MOV32     *-SP[6],R0H           ; [CPU_] |107|

  • 0 in reply to Vishal_Coelho
    Guru 12800 points

    I looked at the disassembly via CCS but it is difficult for me to search the assembly code.  Is there a way that I can put it into a text file so that I can more easily search for the DIVF32 instruction as well as look for the address of the fastRTS div function?

    Also how did you get your disassembly to look like that?

  • 0 in reply to Adam D. Reynolds
    TI__Mastermind 20850 points

    what i normally do is i go to the properties for a particular file, in my case it was 'main.c', and then under

    1. Advanced Options -> Advanced Debug Options -> Debugging Model: choose Suppress all symbolic debug generation (--symdebug:none)

    2. Advanced Options -> Assembler Options: enable --keep_asm

     and for Source Interlist: select --c_src_interlist

    The second option will generate an assembly file 'main.asm' in the output folder; the assembly will be displayed with the corresponding C code in comments along with the line number. you can just search this file for the line number to find its assembly snippet

    The first option gets rid of the DWARF debug information - but if you do this you wont be able to step through your C code while debugging so be sure to set ut back to --symdebug:dwarf when you are debugging.

  • 0 in reply to Vishal_Coelho
    Guru 12800 points

    Vishal,

    Thanks that was really helpful.  In looking at my assembly it does appear to be using the right instructions.  

    MOVB      XAR0,#16              ; [CPU_] 
CMPB      AL,#2                 ; [CPU_] |221| 
DIVF32    R5H,R0H,R2H           ; [CPU_] |220| 
MOV32     R7H,*+XAR2[AR0]       ; [CPU_] 
BF        $C$L7,NEQ             ; [CPU_] |221| 
; branchcc occurs ; [] |221|

    and

    MOVB      XAR0,#8               ; [CPU_] |187| 
ADDF32    R2H,R2H,R2H           ; [CPU_] |184| 
MOV32     R7H,*+XAR2[AR0]       ; [CPU_] |187| 
SQRTF32   R0H,R7H               ; [CPU_] |187|

    Thanks.  I think my confusion came from the div_f32.obj being included in the map file.  But I could not find where it was calling into that object.  

  • 0 in reply to Adam D. Reynolds
    TI__Mastermind 20850 points

    Adam,

    It is probably being called somewhere, else i don't think the linker would have roped it in. I might be wrong but its possible that another standard library function that is making the call to div_f32. The std C library is not compiled with tmu support turned on so it would have to use the std math routines.