Factor out libc code into a header.

[mjp41]

This pulls the main definitions of the various libc malloc functions into a header for easier use and inclusion in other projects.

[davidchisnall]

Do they still get inlined in the .cc file?

[mjp41]

It looks good:

00000000000054e0 g     F .text  00000000000000b9              free
00000000000054e0 g     F .text  00000000000000b9              cfree
00000000000054e0 g     F .text  00000000000000b9              operator delete(void*)
00000000000054e0 g     F .text  00000000000000b9              operator delete[](void*)
00000000000054e0 g     F .text  00000000000000b9              operator delete(void*, std::align_val_t)
00000000000054e0 g     F .text  00000000000000b9              operator delete[](void*, std::align_val_t)

0000000000005430 g     F .text  00000000000000a8              malloc
0000000000005430 g     F .text  00000000000000a8              operator new(unsigned long)
0000000000005430 g     F .text  00000000000000a8              operator new[](unsigned long)
0000000000005ed0 g     F .text  00000000000000b6              operator new(unsigned long, std::align_val_t)
0000000000005ed0 g     F .text  00000000000000b6              operator new[](unsigned long, std::align_val_t)

and

0000000000005430 <operator new[](unsigned long)>:
    5430:       48 8d 47 ff             lea    -0x1(%rdi),%rax
    5434:       48 3d ff df 00 00       cmp    $0xdfff,%rax
    543a:       77 3a                   ja     5476 <operator new[](unsigned long)+0x46>
    543c:       48 c1 e8 04             shr    $0x4,%rax
    5440:       48 8d 0d 51 cc ff ff    lea    -0x33af(%rip),%rcx        # 2098 <snmalloc::sizeclass_lookup>
    5447:       0f b6 14 08             movzbl (%rax,%rcx,1),%edx
    544b:       48 8b 35 e6 e0 00 00    mov    0xe0e6(%rip),%rsi        # 13538 <_DYNAMIC+0x1e0>
    5452:       64 48 8b 04 25 00 00    mov    %fs:0x0,%rax
    5459:       00 00
    545b:       48 01 f0                add    %rsi,%rax
    545e:       48 8d 0c d0             lea    (%rax,%rdx,8),%rcx
    5462:       64 48 8b 04 d6          mov    %fs:(%rsi,%rdx,8),%rax
    5467:       48 85 c0                test   %rax,%rax
    546a:       74 22                   je     548e <operator new[](unsigned long)+0x5e>
    546c:       48 8b 10                mov    (%rax),%rdx
    546f:       0f 18 0a                prefetcht0 (%rdx)
    5472:       48 89 11                mov    %rdx,(%rcx)
    5475:       c3                      ret
    ....

[nwf-msr]

Two hopefully quick questions, but the bulk of it seems fine.

[nwf-msr]

Does marking this const re-enable the problematic inlining?

[mjp41]

I had a look at the code a bit more, and it no longer seems like it is a problem.

[mjp41]

Also ended up adding some SNMALLOC_FAST_PATH to force some inlining as calloc and realloc had introduced an additional tail call than needed.

[nwf-msr]

LGTM