[llvm-bugs] [Bug 38717] New: Suboptimal 128 bit integer division in loop

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=38717

            Bug ID: 38717
           Summary: Suboptimal 128 bit integer division in loop
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Scalar Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: david.bolvansky at gmail.com
                CC: llvm-bugs at lists.llvm.org

int main(void)
{
    const __int128 m = 3329589384618324948;
    __int128 n = m * m;

    enum { base = 10, max_width = 39 };

    char a[max_width + 1] = { '\0' };
    char *p = a + max_width;
    for (; n != 0; n /= base) 
        *--p = (char)('0' + n % base);

    printf("%s\n", p);
}

Clang trunk with O3:
.LBB0_1:                            
        mov     edx, 10
        xor     ecx, ecx
        mov     rdi, r15
        mov     rsi, r14
        call    __divti3
        mov     rbx, rax
        mov     r12, rdx
        lea     eax, [rax + rax]
        lea     eax, [rax + 4*rax]
        mov     ecx, r15d
        sub     ecx, eax
        add     cl, 48
        mov     byte ptr [r13], cl
        mov     edx, 10
        xor     ecx, ecx
        mov     rdi, rbx
        mov     rsi, r12
        call    __modti3
        add     al, 48
        mov     byte ptr [r13 - 1], al
        mov     edx, 100
        xor     ecx, ecx
        mov     rdi, r15
        mov     rsi, r14
        call    __divti3         <---- elide it like GCC did?
        mov     r15, rax
        mov     r14, rdx
        add     rbx, 9
        adc     r12, 0
        add     r13, -2
        cmp     rbp, rbx
        mov     eax, 0
        sbb     rax, r12
        jb      .LBB0_1
        lea     rdi, [rsp + 1]
        call    puts
        xor     eax, eax
        add     rsp, 40
        pop     rbx
        pop     r12
        pop     r13
        pop     r14
        pop     r15
        pop     rbp
        ret

GCC:
.L2:
        mov     rbx, rdx
        mov     rdi, r13
        xor     ecx, ecx
        mov     edx, 10
        mov     rsi, rbx
        sub     rbp, 1
        call    __modti3
        mov     rdi, r13
        mov     edx, 10
        xor     ecx, ecx
        add     eax, 48
        mov     rsi, rbx
        mov     BYTE PTR [rbp+0], al
        call    __divti3
        mov     r13, rax
        cmp     rbp, r12
        jne     .L2
        mov     rdi, rbp
        call    puts
        add     rsp, 56
        xor     eax, eax
        pop     rbx
        pop     rbp
        pop     r12
        pop     r13
        ret

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180827/0c002095/attachment-0001.html>