[llvm-dev] Where's the optimiser gone (part 8): remainder calculated "by hand", not via call of __divmoddi4

[Stefan Kanthak via llvm-dev]

long long quotient(long long numerator,
                   long long denominator,
                   long long *remainder)
{
    *remainder = numerator % denominator;
    return numerator / denominator;
}

long long remainder(long long numerator,
                    long long denominator,
                    long long *quotient)
{
    *quotient = numerator / denominator;
    return numerator % denominator;
}


Compiled with "-O3 -target i386" this yields the following
BLOATED unoptimised code (<https://godbolt.org/z/2rR2fu>)

JFTR: "-O3 -target i386-win32" too fails to generate a call
      of __alldvrm

quotient: # @quotient
    push  ebp                            |   jmp __divmoddi4
    mov   ebp, esp
    push  ebx
    push  edi
    push  esi
    push  eax
    mov   edi, dword ptr [ebp + 20]
    mov   ebx, dword ptr [ebp + 8]
    mov   esi, dword ptr [ebp + 16]
    push  edi
    push  esi
    push  dword ptr [ebp + 12]
    push  ebx
    call  __divdi3
    add   esp, 16
    mov   ecx, eax
    mov   dword ptr [ebp - 16], edx # 4-byte Spill
    imul  edi, eax
    mul   esi
    add   edx, edi
    mov   edi, dword ptr [ebp - 16] # 4-byte Reload
    imul  esi, edi
    add   esi, edx
    mov   edx, dword ptr [ebp + 12]
    sub   ebx, eax
    mov   eax, dword ptr [ebp + 24]
    sbb   edx, esi
    mov   dword ptr [eax], ebx
    mov   dword ptr [eax + 4], edx
    mov   eax, ecx
    mov   edx, edi
    add   esp, 4
    pop   esi
    pop   edi
    pop   ebx
    pop   ebp
    ret

remainder: # @remainder
    push  ebp
    mov   ebp, esp
    push  ebx
    push  edi
    push  esi
    mov   esi, dword ptr [ebp + 20]
    mov   ebx, dword ptr [ebp + 8]
    mov   edi, dword ptr [ebp + 12]
    push  esi
    push  dword ptr [ebp + 16]
    push  edi
    push  ebx
    call  __divdi3
    add   esp, 16
    mov   ecx, edx
    mov   edx, dword ptr [ebp + 24]
    imul  esi, eax
    mov   dword ptr [edx + 4], ecx
    mov   dword ptr [edx], eax
    mul   dword ptr [ebp + 16]
    imul  ecx, dword ptr [ebp + 16]
    add   edx, esi
    add   ecx, edx
    sub   ebx, eax
    sbb   edi, ecx
    mov   eax, ebx
    mov   edx, edi
    pop   esi
    pop   edi
    pop   ebx
    pop   ebp
    ret