M16C #pragma 32bit register support

[RyanHope]

The M16C compiler supports a #pragma macro that allows the programmer to specify which registers are used for arguments getting passed to a function. The current M16C processor support does not detect these or even seem to support manually editing the function signature to support this. Below is a description of this #pragma macro:

Here is an example of this being used in a binary I am working on.

          0c9a68 0b 12 0e        MOV.B:S    DAT_000e12,R0L
          0c9a6b eb 5f 0d 00     MOVA       0xd,A1
          0c9a6f eb 4f b2 00     MOVA       0xb2,A0
          0c9a73 72 f2 3d 0e     MOV.B:G    DAT_000e3d,R1L
          0c9a77 c3 18           MOV.B:S    #0x18,R0H
          0c9a79 f5 5e 34        JSR.W      FUN_0cced8
          0c9a7c 72 2f 3d 0e     MOV.B:G    R1L,DAT_000e3d
          0c9a80 73 0f 2b 0e     MOV.W:G    R0,DAT_000e2b

                             **************************************************************
                             *                          FUNCTION                          *
                             **************************************************************
                             void FUN_0cced8(void)
             void              <VOID>         <RETURN>
                             FUN_0cced8                                      XREF[4]:     0c9a79(c), 0c9a95(c), 0c9ace(c), 
                                                                                          0c9b07(c)  
          0cced8 8b 02           SUB.B:S    #0x2,R0H
          0cceda c0 21           CMP.B:G    R1L,R0H
          0ccedc 68 03           JGEU       LAB_0ccee0
          0ccede 72 12           MOV.B:G    R0H,R1L
                             LAB_0ccee0                                      XREF[1]:     0ccedc(j)  
          0ccee0 d8 03           MOV.B:Q    #0x0,R1H
          0ccee2 73 12           MOV.W:G    R1,R2
          0ccee4 6a 16           JEQ        LAB_0ccefb
          0ccee6 a1 24           ADD.W:G    R2,A0
          0ccee8 77 e5           ADCF.W     A1
                             LAB_0cceea                                      XREF[1]:     0ccef8(j)  
          0cceea 74 a3           LDE.B      [A1A0],R1H
          0cceec c0 30           CMP.B:G    R1H,R0L
          0cceee 68 15           JGEU       LAB_0ccf04
          0ccef0 77 54 01 00     SUB.W:G    #0x1,A0
          0ccef4 77 75 00 00     SBB.W      #0x0,A1
          0ccef8 f8 f2 f0        ADJNZ.B    #-0x1,R1L

If I try to edit the function signature and with custom storage checked, I can't add A1A0 as a parameter.

How do I fix this?

[RyanHope]

This is what the function editor looks like when I try and add A1A0 as the register source for an argument...

[RyanHope]

If I modify the .cspec and add this prototype:

  <prototype name="__pragmacall" extrapop="0" stackshift="0">
    <input>
        <pentry maxsize="4" minsize="1" align="1">
          <register name="A1A0" />
        </pentry>
        <pentry maxsize="4" minsize="1" align="1">
          <register name="R3R1" />
        </pentry>
        <pentry maxsize="4" minsize="1" align="1">
          <register name="R2R0" />
        </pentry>
    </input>
    <output>
        <pentry maxsize="1" minsize="1">
          <register name="R1L"/>
        </pentry>
    </output>
    <unaffected>
      <register name="SP" />
      <register name="INTB" />
      <register name="FB" />
      <register name="SB" />
      <register name="FLG" />
    </unaffected>
  </prototype>

Then I can modify the signature to look like this. However, as can be seen none of the storage sizes are larger than 1 so there are these filler "register" params that should not actually exist if the storage sizes were correct.

I am stuck here.

[RyanHope]

Hmm, for some reason if I replace this __pragama call with with the default __stdcall, everything seems to be working as better.

[RyanHope]

If I try and edit the datatype for the pointer however things break again because pointers are set to only be 3 bytes.