topic Re: How can I force a C function to be called using a JSR or BSR instead of a BRA? in CodeWarrior for MCU

How can I force a C function to be called using a JSR or BSR instead of a BRA?

zorrotz — Fri, 01 Apr 2011 14:35:24 GMT

I have a C function which is near (in memory) of its caller (another function also in C).

The compiler generates a BRA instruction to do this call, but I need a JSR or BSR (because I need to save the PC in the stack).

How can I do it?

Thanks in advance.

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

BlackNight — Fri, 01 Apr 2011 14:50:41 GMT

Hello,

it would be good to know which compiler is used.

If it is CodeWarrior for S08, have a look at following options:

-OnP Disable peephole Optimization (-OnP[={<x>}], default: all)
-OnP=m Disable peephole JSR to JMP optimization

-OnB          Disable Branch optimizer (-OnP[={<x>}], default: all)
                   -OnB=t Disable Branch tail optimization
                   -OnB=r Disable Branch to RTS optimization
                   -OnB=b Disable Branch JSR to BSR optimization
                   -OnB=a Disable short BRA optimization

You might try switching off the whole optimization (-OnP or -OnB) and later refine it (e.g. using -OnP=m).

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

zorrotz — Fri, 01 Apr 2011 16:10:55 GMT

It is CodeWarrior 6.3 for S08.

The option -OnP=m works OK. It disables peephole JSR to JMP optimization and BSR to BRA optimization.

Thanks

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

zorrotz — Fri, 01 Apr 2011 17:04:42 GMT

I have another doubt.

In example 1, why the optimizer only changes the JSR to JMP when copy4ram2() function is called?

And in example 2, why the optimizer doesn't do the change if I include a NOP after calling copy4ram2()?

(see also the note at the end of the post).

Thanks

EXAMPLE 1. copy4ram2() called using JMP.

C function:

void March_XA5_allram ()

{

March_XA5_ram4(); // 1/ March X on RAM4 before using as shadow

copy1ram4(); // 2/ copy RAM1 contents into RAM4

March_XA5_ram1(); // 3/ March X on RAM1

copy4ram1(); // 4/ copy RAM 4 contents into RAM1

copy3ram4(); // 8/ copy RAM3 contents into RAM4

March_XA5_ram3(); // 9/ March X on RAM3

copy4ram3(); // 10/ copy RAM1 contents into RAM4

copy2ram4(); // 5/ copy RAM2 contents into RAM4

March_XA5_ram2(); // 6/ March X on RAM2

copy4ram2(); // 7/ copy RAM4 contents into RAM2

}

Compiler generated code:

169:

170: March_XA5_ram4(); // 1/ March X on RAM4 before using as shadow

0000 cd0000 [6] JSR March_XA5_ram4

171:

172: copy1ram4(); // 2/ copy RAM1 contents into RAM4

0003 cd0000 [6] JSR copy1ram4

173: March_XA5_ram1(); // 3/ March X on RAM1

0006 cd0000 [6] JSR March_XA5_ram1

174: copy4ram1(); // 4/ copy RAM 4 contents into RAM1

0009 cd0000 [6] JSR copy4ram1

175: copy3ram4(); // 8/ copy RAM3 contents into RAM4

000c cd0000 [6] JSR copy3ram4

176: March_XA5_ram3(); // 9/ March X on RAM3

000f cd0000 [6] JSR March_XA5_ram3

177: copy4ram3(); // 10/ copy RAM1 contents into RAM4

0012 cd0000 [6] JSR copy4ram3

178: copy2ram4(); // 5/ copy RAM2 contents into RAM4

0015 cd0000 [6] JSR copy2ram4

179: March_XA5_ram2(); // 6/ March X on RAM2

0018 cd0000 [6] JSR March_XA5_ram2

180: copy4ram2(); // 7/ copy RAM4 contents into RAM2

001b cc0000 [4] JMP copy4ram2

181:

182: }

EXAMPLE 2. copy4ram2() called using JSR after including a NOP.

C function:

void March_XA5_allram ()

{

March_XA5_ram4(); // 1/ March X on RAM4 before using as shadow

copy1ram4(); // 2/ copy RAM1 contents into RAM4

March_XA5_ram1(); // 3/ March X on RAM1

copy4ram1(); // 4/ copy RAM 4 contents into RAM1

copy3ram4(); // 8/ copy RAM3 contents into RAM4

March_XA5_ram3(); // 9/ March X on RAM3

copy4ram3(); // 10/ copy RAM1 contents into RAM4

copy2ram4(); // 5/ copy RAM2 contents into RAM4

March_XA5_ram2(); // 6/ March X on RAM2

copy4ram2(); // 7/ copy RAM4 contents into RAM2

asm(nop);

}

Compiler generated code:

169:

170: March_XA5_ram4(); // 1/ March X on RAM4 before using as shadow

0000 cd0000 [6] JSR March_XA5_ram4

171:

172: copy1ram4(); // 2/ copy RAM1 contents into RAM4

0003 cd0000 [6] JSR copy1ram4

173: March_XA5_ram1(); // 3/ March X on RAM1

0006 cd0000 [6] JSR March_XA5_ram1

174: copy4ram1(); // 4/ copy RAM 4 contents into RAM1

0009 cd0000 [6] JSR copy4ram1

175: copy3ram4(); // 8/ copy RAM3 contents into RAM4

000c cd0000 [6] JSR copy3ram4

176: March_XA5_ram3(); // 9/ March X on RAM3

000f cd0000 [6] JSR March_XA5_ram3

177: copy4ram3(); // 10/ copy RAM1 contents into RAM4

0012 cd0000 [6] JSR copy4ram3

178: copy2ram4(); // 5/ copy RAM2 contents into RAM4

0015 cd0000 [6] JSR copy2ram4

179: March_XA5_ram2(); // 6/ March X on RAM2

0018 cd0000 [6] JSR March_XA5_ram2

180: copy4ram2(); // 7/ copy RAM4 contents into RAM2

001b cd0000 [6] JSR copy4ram2

181: asm(nop);

001e 9d [1] NOP

182:

183: }

001f 81 [6] RTS

Note that copy4ram2() function is:

void copy4ram2 ()
{

#asm

          clrh
          clrx
loop1:    lda RAM4START,x
          sta RAM2START,x
          aix #1
          cphx #RAMRANGE
          bne loop1

// verify copy has occurred

        clrh
        clrx
loop2: lda RAM4START,x
        cmp RAM2START,x
        bne error
        aix #1
        cphx #RAMRANGE
        bne loop2
//       STA   _SRS // Refresh WDOG
        rts
error: bra error

#endasm

}

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

kef — Fri, 01 Apr 2011 17:29:32 GMT

In case 1) when you need to do last call in your function and immediately exit and 2) when you don't need to push any parameters to the stack: - intead of jsr to subroutine, executing its body, rts back to your function, and then return from your function; you can just jump to calle saving few CPU cycles on extra jsr-rts. Adding nop in your function makes this optimization not possible, since jmp is not jsr and you can't return to finish your routine (nop).

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

Jim_P — Sat, 09 Apr 2011 01:51:19 GMT

question

is the BRA at the end of the function or procedure that was called with the saving of the PC, Does the BRA jump to a routine that ends with a RTS.

then what is the problem - - -the stack is keep straight, less CPU cycles are needed

Jim P

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

kef — Sat, 09 Apr 2011 17:20:02 GMT

Problem is

but I need a JSR or BSR (because I need to save the PC in the stack).

though copy4ram2() doesn't access return address on the stack or being affected by BSR->BRA optimization.

Re: How can I force a C function to be called using a JSR or BSR instead of a BRA?

Jim_P — Mon, 11 Apr 2011 05:50:21 GMT

Could you give me and example of code that needs the return address on the stack - - - that is effected by the JSR or BSR to BRA