Assembly code in GCC is not being compiled in Atmel Studio for AVR environment

Question

I am trying to remove the interrupt vector table in my program to reduce the flash memory usage.

I have successfully used the -nostartfiles flag, which removes a lot of assembly code in the output.

To use the program, I have to set the stack pointer. I found this code to do so:

asm volatile ( ".set __stack, %0" :: "i" (RAMEND) );

But when compiled, it does nothing, while other assembly code like

asm volatile ( "clr __zero_reg__" );      
asm volatile ( "rjmp main");

Does work.

This is my current C code:

void __init(void) __attribute__ ((naked)) __attribute__ ((section (".init9")));
void __init(void)
{
    asm volatile(".set __stack, %0" :: "i" (RAMEND) );
    asm volatile ( "clr __zero_reg__" );      
    asm volatile ( "rjmp main");             
}

Which compiles to assembly (the 7892 and 7894 lines):

void __jMain(void) __attribute__ ((naked)) __attribute__ ((section (".init9")));
void __jMain(void)
{
    asm volatile(".set __stack, %0" :: "i" (RAMEND) );
    asm volatile ( "clr __zero_reg__" );        
    7892:   11 24           eor r1, r1
    asm volatile ( "rjmp main");                
    7894:   02 c0           rjmp    .+4         ; 0x789a <main>
}

Why isn't the .set __stack compiled? Am I missing some compiler flag? Have tried a lot of things.. Also reviewed bootloaders with the same piece of code who are doing the same thing, but somehow, mine does not compile correctly.

Answer 1

How about:

asm volatile(
        "ldi  r28, lo8(RAMEND)" "\n\t"
        "ldi  r29, hi8(RAMEND)" "\n\t"
        "out __SP_L__, r28" "\n\t"
        "out __SP_H__, r29" "\n\t"
 );

Main source: https://ucexperiment.wordpress.com/2015/01/02/arduino-stack-painting/

Answer 2

I have found that the correct "c" way of doing this is:

void __init(void) __attribute__ ((naked)) __attribute__ ((section (".init9")));
void __init(void)
{   
    asm volatile ("clr __zero_reg__");
    SP = RAMEND;   
    asm volatile ( "rjmp main");             
}

Where at the end, it jumps to your main function. This produces the following assembly output proposed by yumbrad:

// asm volatile ("clr __zero_reg__");
7800:   11 24           eor r1, r1

// SP = RAMEND;   
7802:   8f ef           ldi r24, 0xFF   ; 255
7804:   98 e0           ldi r25, 0x08   ; 8
7806:   9e bf           out 0x3e, r25   ; 62
7808:   8d bf           out 0x3d, r24   ; 61

// asm volatile ( "rjmp main");             
780a:   00 c0           rjmp    .+0         ; 0x780c <main>