Giovanni,
I received the following reply about the mspgcc behaviour prom Peter Bigot.
On 14/11/12 23:04, Peter Bigot wrote:
> The naked is not being ignored there. gcc doesn't introspect the contents of asm statement templates, so it's obliged to save the incoming registers on the stack before the first asm statement in case they were corrupted. This is not the same as emitting a prolog and epilog, which naked inhibits.
>
> Apparently with -O1 gcc is willing to presume that that r14 and r15 are not clobbered (since the asm statement doesn't mention them outside the template, which is not parsed), so skips the save.
>
> Based on what I can see line 38 is not superfluous: you specify that variable sp is a register input to the asm statement, and gcc chose to put it into r15 for that purpose. If you use the operand in the template, it's not the snippet you've shown.
>
> The documentation for "naked" that you quote is for other back ends, but is fairly descriptive of expectations for mspgcc. The code you show violates its contract in two ways: it interleaves asm statements with C statements, and it uses operands in an asm statement.
>
> In short, while it may not be doing what you expect, I don't see that it's doing anything wrong.
>
> Peter
>
>
> On Wed, Nov 14, 2012 at 4:41 AM, Chris <chrishealy@internode.on.net> wrote:
>
> I am not sure if this is by design or otherwise, but I was trying to
> compile some MSP430 code that used the naked attribute for a function
> and found that if -O0 optimization was set for the compile, the
> attribute was ignored resulting in non functional code. For -O1 or -Os
> the expected code was generated.
>
> The code is like this .
>
> __attribute__((naked, weak))
> void port_switch(Thread *ntp, Thread *otp) {
> register struct intctx *sp asm("r1");
> asm volatile ("push r11 \n\t" \
> "push r10 \n\t" \
> ........
> "push r4" : : : "memory");
> otp->p_ctx.sp = sp;
> sp = ntp->p_ctx.sp;
> asm volatile ("pop r4 \n\t" \
> "pop r5 \n\t" \
> ...........
> "pop r11 \n\t" \
> "ret" : : "r" (sp) : "memory");
> }
>
> And with -O0 the lst looks like this which is definitely not functional.
>
>
> 14 port_switch:
> 15 .LFB0:
> 16 .file 1 "../../os/ports/GCC/MSP430/chcore.c"
> 17 .loc 1 47 0
> 18 0000 814F 0000 mov r15, @r1
> 19 0004 814E 0200 mov r14, 2(r1)
> 20 .loc 1 50 0
> 21 #APP
> 22 ; 50 "../../os/ports/GCC/MSP430/chcore.c" 1
> 23 0008 0B12 push r11
> ............
> 30 0016 0412 push r4
> 31 ; 0 "" 2
> 32 .loc 1 58 0
> 33 #NOAPP
> 34 0018 0E41 mov r1, r14
> 35 001a 1F41 0200 mov 2(r1), r15
> 36 001e 8F4E 0600 mov r14, 6(r15)
> 37 .loc 1 60 0
> 38 0022 2F41 mov @r1, r15
> 39 0024 1F4F 0600 mov 6(r15), r15
> 40 0028 014F mov r15, r1
> 41 .loc 1 62 0
> 42 002a 0F41 mov r1, r15
> 43 #APP
> 44 ; 62 "../../os/ports/GCC/MSP430/chcore.c" 1
> 45 002c 3441 pop r4
> ....................................
>
> If the naked is removed, the compiler produces identical code as -O0 +
> naked.
>
> For -O1 the code lst is like this.
>
> 14 port_switch:
> 15 .LFB0:
> 16 .file 1 "../../os/ports/GCC/MSP430/chcore.c"
> 17 .loc 1 47 0
> 18 .LVL0:
> 19 .loc 1 50 0
> 20 #APP
> 21 ; 50 "../..//os/ports/GCC/MSP430/chcore.c" 1
> 22 0000 0B12 push r11
> ............
> 29 000e 0412 push r4
> 30 ; 0 "" 2
> 31 .loc 1 58 0
> 32 #NOAPP
> 33 0010 8E41 0600 mov r1, 6(r14)
> 34 .loc 1 60 0
> 35 0014 114F 0600 mov 6(r15), r1
> 36 .LVL1:
> 37 .loc 1 62 0
> 38 0018 0F41 mov r1, r15
> 39 .LVL2:
> 40 #APP
> 41 ; 62 "../../os/ports/GCC/MSP430/chcore.c" 1
> 42 001a 3441 pop r4
> .................
> 49 0028 3B41 pop r11
> 50 002a 3041 ret
>
> Which is what is required although line 38 is superfluous.
>
> Information on naked is a bit difficult to come by. The GCC manual states.
>
> |naked|
> Use this attribute on the ARM, AVR, MCORE, RX and SPU ports to
> indicate that the specified function does not need prologue/epilogue
> sequences generated by the compiler. It is up to the programmer to
> provide these sequences. The only statements that can be safely
> included in naked functions are |asm| statements that do not have
> operands. All other statements, including declarations of local
> variables, |if| statements, and so forth, should be avoided. Naked
> functions should be used to implement the body of an assembly
> function, while allowing the compiler to construct the requisite
> function declaration for the assembler.
>
> So perhaps it was expecting a bit much for this code to produce the
> correct machine code.
> But that aside, when the C code is replaced with assembler, there would
> still be unintended writes to the stack.
So first off the use of naked was always intended to exclude the use of C code as stated in the GCC documents.
The
statements that do not have
operands
is not so easy, although the mov assembler instruction produced the correct result in 4.6.3 but whether it always will is another matter.
I'll put in a bug report to keep track of the issue.
Regards saraben.
