Define sizeOf Char# to always be 4 byte.

Motivation

Smaller code, which without any other changes is better code.

For example we start with this function.

elemCString# :: Char# -> Addr# -> Bool
elemCString# c addr =
    check 0#
  where
    check :: Int# -> Bool
    check nh
      | isTrue# (ch `eqChar#` '\0'#) = False
      | isTrue# (ch `eqChar#` c    ) = True
      | True                         = check (nh +# 1#)
      where
        !ch = indexCharOffAddr# addr nh

It compiles as expected to a tight loop:

[M.elemCString2#_entry() { //  [R3, R2]
         { info_tbls: [(c1Ra,
                        label: M.elemCString2#_info
                        rep: HeapRep static { Fun {arity: 2 fun_type: ArgSpec 12} }
                        srt: Nothing)]
           stack_info: arg_space: 8 updfr_space: Just 8
         }
     {offset
       c1Ra: // global
           _s1Q6::I64 = R3;
           _s1Q5::I64 = R2;
           _s1Q8::I64 = 0;
           goto c1Re;
       c1Re: // global
           _s1Q9::I64 = %MO_UU_Conv_W8_W64(I8[_s1Q6::I64 + _s1Q8::I64]);
           if (_s1Q9::I64 != 0) goto c1Rm; else goto c1Rx;
       c1Rm: // global
           if (_s1Q9::I64 != _s1Q5::I64) goto c1Rt; else goto c1Ru;
       c1Rt: // global
           _s1Q8::I64 = _s1Q8::I64 + 1;
           goto c1Re;
       c1Ru: // global
           R1 = GHC.Types.True_closure+2;
           call (P64[Sp])(R1) args: 8, res: 0, upd: 8;
       c1Rx: // global
           R1 = GHC.Types.False_closure+1;
           call (P64[Sp])(R1) args: 8, res: 0, upd: 8;
     }
 },

However we expand all Char# values to 64 bit! This gives us this assembly:

  a0:   31 c0                   xor    %eax,%eax
  a2:   eb 03                   jmp    a7 <M_elemCString2zh_info+0x7>
  a4:   48 ff c0                inc    %rax
  a7:   0f b6 1c 06             movzbl (%rsi,%rax,1),%ebx
  ab:   48 85 db                test   %rbx,%rbx
  ae:   74 0d                   je     bd <M_elemCString2zh_info+0x1d>
  b0:   4c 39 f3                cmp    %r14,%rbx
  b3:   75 ef                   jne    a4 <M_elemCString2zh_info+0x4>
  b5:   bb 02 00 00 00          mov    $0x2,%ebx
  ba:   ff 65 00                jmpq   *0x0(%rbp)
  bd:   bb 01 00 00 00          mov    $0x1,%ebx
  c2:   ff 65 00                jmpq   *0x0(%rbp)

There are three instructions which currently require the 64bit prefix: mov, test and cmp

If we interpret Char# as 32bits the code is 3 bytes smaller, which means it would fit perfectly in 32bytes.

Proposal

Define the size of Char# as min (4 byte, wordSize)