Change subtraction overflow condition in sub_with_carry()

[chfast]

The Clang compiler loves this pattern and will generate optimal subtraction procedure. And the code is also simpler although not symmetric to addition any more.

binop<uint256, uint256, sub>                                        -0.4896         -0.4896             3             2             3             2
binop<uint256, uint256, inline_sub>                                 -0.3636         -0.3635             2             1             2             1
binop<uint512, uint512, sub>                                        -0.4397         -0.4397             7             4             7             4
binop<uint512, uint512, inline_sub>                                 -0.5787         -0.5787             6             3             6             3

Before:

0000000000412390 <sub(intx::uint<256u> const&, intx::uint<256u> const&)>:                                                                                                                                                         
  412390:       53                      push   %rbx                                                                                                                                                                               
  412391:       4c 8b 06                mov    (%rsi),%r8                                                                                                                                                                         
  412394:       48 8b 4e 08             mov    0x8(%rsi),%rcx                                                                                                                                                                     
  412398:       48 2b 4a 08             sub    0x8(%rdx),%rcx                                                                                                                                                                     
  41239c:       41 0f 92 c1             setb   %r9b                                                                                                                                                                               
  4123a0:       4c 2b 02                sub    (%rdx),%r8                                                                                                                                                                         
  4123a3:       49 89 ca                mov    %rcx,%r10                                                                                                                                                                          
  4123a6:       49 83 da 00             sbb    $0x0,%r10                                                                                                                                                                          
  4123aa:       48 89 f8                mov    %rdi,%rax                                                                                                                                                                          
  4123ad:       49 39 ca                cmp    %rcx,%r10                                                                                                                                                                          
  4123b0:       0f 97 c1                seta   %cl                                                                                                                                                                                
  4123b3:       44 08 c9                or     %r9b,%cl                                                                                                                                                                           
  4123b6:       48 8b 7e 10             mov    0x10(%rsi),%rdi                                                                                                                                                                    
  4123ba:       48 2b 7a 10             sub    0x10(%rdx),%rdi                                                                                                                                                                    
  4123be:       41 0f 92 c1             setb   %r9b                                                                                                                                                                               
  4123c2:       44 0f b6 d9             movzbl %cl,%r11d                                                                                                                                                                          
  4123c6:       48 89 fb                mov    %rdi,%rbx                                                                                                                                                                          
  4123c9:       4c 29 db                sub    %r11,%rbx                                                                                                                                                                          
  4123cc:       48 39 fb                cmp    %rdi,%rbx                                                                                                                                                                          
  4123cf:       0f 97 c1                seta   %cl                                                                                                                                                                                
  4123d2:       44 08 c9                or     %r9b,%cl                                                                                                                                                                           
  4123d5:       48 8b 76 18             mov    0x18(%rsi),%rsi                                                                                                                                                                    
  4123d9:       48 2b 72 18             sub    0x18(%rdx),%rsi                                                                                                                                                                    
  4123dd:       0f b6 c9                movzbl %cl,%ecx                                                                                                                                                                           
  4123e0:       48 29 ce                sub    %rcx,%rsi                                                                                                                                                                          
  4123e3:       4c 89 00                mov    %r8,(%rax)
  4123e6:       4c 89 50 08             mov    %r10,0x8(%rax)
  4123ea:       48 89 58 10             mov    %rbx,0x10(%rax)
  4123ee:       48 89 70 18             mov    %rsi,0x18(%rax)
  4123f2:       5b                      pop    %rbx
  4123f3:       c3                      retq

After:

00000000004122d0 <sub(intx::uint<256u> const&, intx::uint<256u> const&)>:
  4122d0:       48 89 f8                mov    %rdi,%rax
  4122d3:       4c 8b 06                mov    (%rsi),%r8
  4122d6:       4c 2b 02                sub    (%rdx),%r8
  4122d9:       48 8b 7e 08             mov    0x8(%rsi),%rdi
  4122dd:       48 1b 7a 08             sbb    0x8(%rdx),%rdi
  4122e1:       48 8b 4e 10             mov    0x10(%rsi),%rcx
  4122e5:       48 1b 4a 10             sbb    0x10(%rdx),%rcx
  4122e9:       48 8b 76 18             mov    0x18(%rsi),%rsi
  4122ed:       48 1b 72 18             sbb    0x18(%rdx),%rsi
  4122f1:       4c 89 00                mov    %r8,(%rax)
  4122f4:       48 89 78 08             mov    %rdi,0x8(%rax)
  4122f8:       48 89 48 10             mov    %rcx,0x10(%rax)
  4122fc:       48 89 70 18             mov    %rsi,0x18(%rax)
  412300:       c3                      retq

[axic]

Nice!

[axic]

Sidenote: There's such an implicit conversion for carry in intx.hpp:sub_with_carry too. Clion complained about it.

[chfast]

Not sure where.

[axic]

I think all these are suspect:

• https://github.com/chfast/intx/blob/master/include/intx/intx.hpp#L853 (this is what clion showed)
• https://github.com/chfast/intx/blob/master/include/intx/intx.hpp#L801
• https://github.com/chfast/intx/blob/master/include/intx/intx.hpp#L1109