ARM Inline assembly for C code not executing

Question

I am trying to use inline assembly for an ARM C code. My 'main.c' code looks something like this:

...
void jump()
{
 __asm
 {
 B 0x15000
 }
}
INT32 main(void)
{
 ...
 Write val1 into register 1
 jump();
 Write val2 into register 2
 ...
}

When I run this code I expect that val1 to be written in register 1, and then jump to address 0x15000 happens and there is nothing written in register 2. But when I look at register 1 and register 2 both have been written with val1 and val2. So it seems the jump() does not take place. Am I missing something?

Update 1

As suggested in the comments, I created a minimum reproducible example which is:

void jump()
 {
 __asm
 {
 B 0x15000
 }
 }
int main(void)
{ 
 int i, j, k;
 i = 1;
 jump();
 j = 2;
 k = i + j;
 return 128;
}

when I compile it with armcc --asm main.c I get this as main.s:

 ARM
 REQUIRE8
 PRESERVE8
 AREA ||.text||, CODE, READONLY, ALIGN=2
jump PROC
 BX lr
 ENDP
main PROC
 MOV r0,#0x80
 BX lr
 ENDP
 AREA ||.arm_vfe_header||, DATA, READONLY, NOALLOC, ALIGN=2
 DCD 0x00000000

so it looks like the jump() is optimized out? What is that and how can I prevent it?

Update 2

I even tried adding the volatile before the function name as:

 volatile void jump()
 {
 __asm
 {
 B 0x15000
 }
 }

but it didn't help and the jump() function is optimized out.

• Eugene Sh.: Use a debugger, step through assembly instructions. Or for starters at least look at the compiled assembly.
• Eugene Sh.: Then just compile it yourself and post the assembly here. armcc is proprietary and not commonly available.
• Peter Cordes: If you don't have a debugger, that's your number 1 problem. You should fix that. (Even if just in a simulator, not on your real hardware). Many problems in asm become a lot more obvious when you single-step and watch register values change. (Although in this case, looking at the compiler's asm output may be sufficient.)
• Eugene Sh.: FWIW v5.01 is quite buggy. v5.06 is backwards compatible and is fixing many bugs. Also it is using same license, so better move to it..
• Peter Cordes: I'm not surprised volatile void jump() did nothing. If it's not already implicit for an asm {} block instead of GNU C style asm("template" :outputs : inputs : clobbers), you'd need to apply it to the asm keyword, not the function containing it. volatile void foo() doesn't mean "don't optimize this function".