Jeff F. suggested that I should use the linker's -Map command-line
option to generate a map file. From that I was able to figure out
that the code ended up at address 0x08000000 instead of 0x02000000
(as it should for multiboot). When I initially added the variable
declaration I did it as follows: "volatile const int __gba_multiboot
= 1;" (I had to add the "= 1" for it to compile without errors in C++
mode). Even when I did this the variable still wouldn't show up in
the map file. Eventually I removed the "const" from the declaration
and suddenly the variable appeared in the map file and the code
appeared at address 0x02000000. It seems to me that the C++ compiler
treats a const variable similar to a #define, and doesn't actually
assign a memory address to it (some sort of optimization). When I
removed the "const" from the declaration the "= 1" assignment was
also no longer needed. So simply adding "int __gba_multiboot;"
(*without* the "const") to the main C++ file did the trick. Thanks a
lot, Jeff, for pointing me in the right direction.


--- In gbadev@y..., "gbatag" <Tiaan_Geldenhuys@h...> wrote:
> Hi,
>
> Over the past few days I've been trying to get some C++ code to
> compile and were able to do so. The code runs fine on the Mappy
> emulator but when transferring the multiboot image to my GBA using
> the MBV2 cable I simply get a blank screen once the transfer
> completes. What could be the problem?
>
> After looking through some group posts and browsing the web, I
> eventually found some useful information about compiling C++ code
for
> the GBA in the "Newbie's Guide to GameBoy Advance Development", by
> VerticalE. There's a section providing some details on the CPPTest
> example, written by Jaap Suter if I'm not mistaken, which I used to
> figure out how to get my C++ code to compile. Following that
> example's documentation to the letter, which also includes the
source
> code, I can produce a binary that have this behavior: it runs on
the
> emulator but not on the hardware. The example simply sets up
> graphics Mode 3, paints the screen blue and then has an endless
> loop. It uses Jeff Frohwein's startup code (based on crt0.S v1.25)
> with the ".equ __MultiBootInclude, 1" and ".equ __CPPSupport, 1"
> lines defined. It doesn't define the __gba_multiboot variable in
> main.cpp, but even doing so doesn't make any difference.
>
> When compiling my own application as a pure C application it works
> just fine on both hardware and emulator – it uses graphics Mode 0.
> But when calling this code from a basic C++ class and then
compiling
> it, it just won't work on real hardware. In my application I've
used
> Jeff's latest crt0.S & Linkscript code and am using the crtbegin.o
&
> crtend.o files that come with the latest version of DevKitAdvance.
>
> Could it be a memory alignment issue, or maybe a name mangling
issue
> with the __gba_multiboot variable? What other differences could
> there be? Any help would be appreciated.
>
> Thanks,
> Tiaan.