Modifying Comipler RT Assembly code to compile for Arm Cortex M3/M4 (CPSR/APSR bit manipulation)

Question

I am trying to get the math routines from Compiler RT working with a GCC toolchain for the ARM Cortex M3/M4F processors (armv7m and armv7em with fpu).

I have everything compiling (with minimal changes) except two lines of code ( msr CPSR_f, ip and msr CPSR_f, #APSR_C ) in the functions below

#define APSR_Z (1 << 30)
#define APSR_C (1 << 29)

DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmple)
    // Per the RTABI, this function must preserve r0-r11.
    // Save lr in the same instruction for compactness
    push {r0-r3, lr}

    bl __aeabi_fcmplt
    cmp r0, #1
    IT(eq)
    moveq ip, #0
    beq 1f

    ldm sp, {r0-r3}
    bl __aeabi_fcmpeq
    cmp r0, #1
    IT(eq)
    moveq ip, #(APSR_C | APSR_Z)
    IT(ne)
    movne ip, #(APSR_C)

1:
    msr CPSR_f, ip
    pop {r0-r3}
    POP_PC()
END_COMPILERRT_FUNCTION(__aeabi_cfcmple)

And the other function:

DEFINE_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)
    push {r0-r3, lr}
    bl __aeabi_cfcmpeq_check_nan
    cmp r0, #1
    pop {r0-r3, lr}

    // NaN has been ruled out, so __aeabi_cfcmple can't trap
    bne __aeabi_cfcmple

    msr CPSR_f, #APSR_C
    JMP(lr)
END_COMPILERRT_FUNCTION(__aeabi_cfcmpeq)

The CPSR_f notation is not available on the armv7m instruction set. How do I convert msr CPSR_f, ip and msr CPSR_f, #APSR_C to armv7m code (should be the same for armv7em)?

Answer 1

You need to use the MOV APSR, Rm instruction. The Cortex-M processors basically don't have a CPSR, and the APSR register acts as its replacement as far as the condition codes go.

The first function is easy to fix, since it uses a register as the source operand. Just replace msr CPSR_f, ip with msr APSR_nzcvq, ip . The second function would require going through a register. Assuming the IP register can be clobbered just like it is in the first function you can use:

mov ip, #APSR_C
msr APSR_nzcvq, ip