Re: bignum branch

[Andy Moreton]

On Sat 11 Aug 2018, Eli Zaretskii wrote:
> Interesting: all the other GMP functions are referenced by their name,
> and only __gmpn_popcount is referenced through the DLL import...

Indeed. As building with "-O0" works and building with "-Og" fails, it
seems that inlining could be involved. mpn_popcount is only called from
the inlined mpz_popcount.

> In the dependency walker display, which functions are shown in the
> upper-right window (the "parent import function list"), the one that
> shows "C" icons with green background, when you click on libgmp-10.dll
> in the top-left window, the one that shows the dependency DLLs?

As far as I can tell, all of the GMP exports used in emacs are listed
there, including __gmpn_popcount.

As Tom has completed merging to master, I have switched to the master
branch and rebuilt from a clean tree (after "git clean -Xdf").

Stepping through the code in gdb, I see:

(gdb) stepi
0x000000040016ebcb      1845        __gmp_result =  mpn_popcount 
(__gmp_u->_mp_d, __gmp_usize);
(gdb)
0x000000046ace5dc0 in ?? ()
(gdb)

Thread 1 received signal SIGSEGV, Segmentation fault.
0x000000046ace5dc0 in ?? ()
(gdb) p mpn_popcount
$5 = {<text variable, no debug info>} 0x401e61484 <__imp___gmpn_popcount>
(gdb) x/xg mpn_popcount
0x401e61484 <__imp___gmpn_popcount>:    0x000000006ace5dc0
(gdb) disas 0x000000006ace5dc0,+0x80
Dump of assembler code from 0x6ace5dc0 to 0x6ace5e40:
   0x000000006ace5dc0:  push   %rdi
   0x000000006ace5dc1:  push   %rsi
   0x000000006ace5dc2:  mov    %rcx,%rdi
   0x000000006ace5dc5:  mov    %rdx,%rsi
   0x000000006ace5dc8:  push   %r12
   0x000000006ace5dca:  push   %r13
   0x000000006ace5dcc:  movabs $0x5555555555555555,%r10
   0x000000006ace5dd6:  movabs $0x3333333333333333,%r11
   0x000000006ace5de0:  movabs $0xf0f0f0f0f0f0f0f,%rcx
   0x000000006ace5dea:  movabs $0x101010101010101,%rdx
   0x000000006ace5df4:  lea    (%rdi,%rsi,8),%rdi
   0x000000006ace5df8:  neg    %rsi
   0x000000006ace5dfb:  xor    %eax,%eax
   0x000000006ace5dfd:  bt     $0x0,%esi
   0x000000006ace5e01:  jae    0x6ace5e50
   0x000000006ace5e03:  mov    (%rdi,%rsi,8),%r8
   0x000000006ace5e07:  mov    %r8,%r9
   0x000000006ace5e0a:  shr    %r8
   0x000000006ace5e0d:  and    %r10,%r8
   0x000000006ace5e10:  sub    %r8,%r9
   0x000000006ace5e13:  mov    %r9,%r8
   0x000000006ace5e16:  shr    $0x2,%r9
   0x000000006ace5e1a:  and    %r11,%r8
   0x000000006ace5e1d:  and    %r11,%r9
   0x000000006ace5e20:  add    %r8,%r9
   0x000000006ace5e23:  mov    %r9,%r8
   0x000000006ace5e26:  shr    $0x4,%r9
   0x000000006ace5e2a:  and    %rcx,%r8
   0x000000006ace5e2d:  and    %rcx,%r9
   0x000000006ace5e30:  add    %r8,%r9
   0x000000006ace5e33:  imul   %rdx,%r9
   0x000000006ace5e37:  shr    $0x38,%r9
   0x000000006ace5e3b:  mov    %r9,%rax
   0x000000006ace5e3e:  add    $0x1,%rsi
End of assembler dump.

The disassembly above matches the start of mpn_popcount in the GMP
sources in gmp-6.1.2/mpn/x86_64/popham.asm.

    AndyM