[Bug tree-optimization/100247] New: x86-64 bad register allocation for unsigned type

[ammarfaizi2 at gmail dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100247

            Bug ID: 100247
           Summary: x86-64 bad register allocation for unsigned type
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ammarfaizi2 at gmail dot com
  Target Milestone: ---

GCC 12.0.0 allocates unnecessary register for unsigned type. Below is the
reproduction code:

Compile with: `gcc -Wall -Wextra -O3 -fno-tree-vectorize -fno-unroll-loops`

--------------------------------------
    #include <stddef.h>

    long add_arrays(long *arr1, long *arr2, size_t num) {
        size_t i = 0;
        long sum = 0;

        for (i = 0; i < num; ++i) {
            sum += arr1[i];
            sum += arr2[i];
        }

        return sum;
    }

    unsigned long unsigned_add_arrays(unsigned long *arr1, unsigned long *arr2,
size_t num) {
        size_t i = 0;
        unsigned long sum = 0;

        for (i = 0; i < num; ++i) {
            sum += arr1[i];
            sum += arr2[i];
        }

        return sum;
    }

    #define PSTR(P) #P
    #define XSTR(P) PSTR(P)

    const char gcc_ver[] = 
    "It is GCC "
    XSTR(__GNUC__) "."
    XSTR(__GNUC_MINOR__) "."
    XSTR(__GNUC_PATCHLEVEL__);

--------------------------------------
GCC 12.0.0 Result

    add_arrays:
        xorl    %eax, %eax
        testq   %rdx, %rdx
        je      .L4
        xorl    %ecx, %ecx
    .L3:
        addq    (%rdi,%rcx,8), %rax
        addq    (%rsi,%rcx,8), %rax
        addq    $1, %rcx
        cmpq    %rcx, %rdx
        jne     .L3
        ret                         # Why do even we need this ret?
    .L4:
        ret

    unsigned_add_arrays:
        xorl    %r8d, %r8d          # Using %r8 is unnecessary
        testq   %rdx, %rdx
        je      .L7
        xorl    %eax, %eax
    .L9:
        movq    (%rsi,%rax,8), %rcx
        addq    (%rdi,%rax,8), %rcx
        addq    $1, %rax
        addq    %rcx, %r8           # %r8 is used as `sum` variable
        cmpq    %rax, %rdx
        jne     .L9
    .L7:
        movq    %r8, %rax           # set return value to %r8
        ret

    gcc_ver:
        .string "It is GCC 12.0.0"

--------------------------------------
GCC 4.6.4 produces better result here:

    add_arrays:
        xorl    %eax, %eax
        testq   %rdx, %rdx
        je      .L2
        xorl    %ecx, %ecx
    .L3:
        addq    (%rdi,%rcx,8), %rax
        addq    (%rsi,%rcx,8), %rax
        addq    $1, %rcx
        cmpq    %rdx, %rcx
        jne     .L3
    .L2:
        rep
        ret

    unsigned_add_arrays:
        xorl    %eax, %eax
        testq   %rdx, %rdx
        je      .L8
        xorl    %ecx, %ecx
    .L9:
        addq    (%rdi,%rcx,8), %rax
        addq    (%rsi,%rcx,8), %rax
        addq    $1, %rcx
        cmpq    %rdx, %rcx
        jne     .L9
    .L8:
        rep
        ret

    gcc_ver:
        .string "It is GCC 4.6.4"
--------------------------------------
Golbolt link: https://godbolt.org/z/9Pj5Ph1Gn