The code for portSET_INTERRUPT_MASK() and portCLEAR_INTERRUPT_MASK() of the GCC ARM CM3 port hardcodes the use of the r0 register as the intermediate value to use as "msr basepri, r0". Any general register can be used here, and sometimes letting GCC choose leads to a shorter and faster code.

By using respectively

__asm volatile("msr basepri,%0" : : "r"(configMAX_SYSCALL_INTERRUPT_PRIORITY));

and

_asm volatile("msr basepri,%0" : : "r"(0));

we can only get better code: in the worst case, we get the equivalent two instructions, in the best case, the compiler knows that rconfigMAX_SYSCALL_INTERRUPT_PRIORITY/0 is already loaded into a register at this point and will use it. Also, it decreases the register pressure on r0 by allowing the use of a different register should it lead to a better code.

For example, vPortExitCritical will then get compiled as

1488: 4a03 ldr r2, [pc, #12] (1498 <vPortExitCritical+0x10>)
148a: 6813 ldr r3, [r2, #0]
148c: 3b01 subs r3, #1
148e: 6013 str r3, [r2, #0]
1490: b90b cbnz r3, 1496 <vPortExitCritical+0xe>
1492: f383 8811 msr BASEPRI, r3
1496: 4770 bx lr

Here, at address 1492, the compiler knows that r3 (which contains the uxCriticalNesting value) is necessarily 0 and uses it instead of reloading the constant 0 into another register.

Very good post - thanks for sharing.

Regards.

Richard,

any chance to get this change applied to the SVN repository?

(I'm already using it in my local copy, I just want to make sure it doesn't get forgotten)

Sam

Can you add it to the feature request tracker in SourceForge please. That way it won't get forgotten.

Regards.