Re: Filed PR jit/66812 for the code generation issue

[Dibyendu Majumdar <mobile@majumdar.org.uk>]

On 8 July 2015 at 21:46, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
> On 8 July 2015 at 21:32, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
>> On 8 July 2015 at 21:22, Dibyendu Majumdar <mobile@majumdar.org.uk> wrote:
>>> Here is what I think the flow is.
>>>
>>> part 1
>>> --------
>>>
>>> 1 [2] LOADNIL   0 0
>>>
>>> entry:
>>>   cl = (struct ravi_LClosure *)L->ci->func->value_.gc;
>>>   (void)raviV_op_loadnil (L->ci, (int)0, (int)0);
>>>
>>>
>>> This sets register 0 (local variable IX) to NIL
>>>
>>> 2 [3] LOADK     1 -1 ; 10
>>>
>>>   (&L->ci->u.l.base[(int)1])->value_.i = (&cl->p->k[(int)0])->value_.i;
>>>   (&L->ci->u.l.base[(int)1])->tt_ = (&cl->p->k[(int)0])->tt_;
>>>
>>> This sets the register at 1 (temporary) to 10.
>>>
>>> 3 [3] TEST     1 1
>>> 4 [3] JMP       0 0 ; to 5
>>>
>>> The above two bytecodes go together. This says that if register 1 is
>>> true then skip the JMP else JMP.
>>> So we get:
>>>
>>>   comparison_0_5 = (&L->ci->u.l.base[(int)1])->tt_ == (int)0;
>>>   comparison_0_6 = (&L->ci->u.l.base[(int)1])->tt_ == (int)1;
>>>   comparison_0_7 = (&L->ci->u.l.base[(int)1])->value_.b == (int)0;
>>>   isfalse_0_4 = comparison_0_5 || comparison_0_6 && comparison_0_7;
>>>   if (!(isfalse_0_4)) goto OP_TEST_do_jmp_2_8; else goto OP_TEST_do_skip_2_9;
>>>
>>> The test is checking that register 1 is true (which it is as it has
>>> the constant 10) - then goto skip_2_9.
>>
>> Since we skipped the jump the next instruction is
>>
>> 5 [3] LOADBOOL 1 0 0
>>
>> This is setting register 1 to false so we get:
>>
>> OP_TEST_do_skip_2_9:
>>   goto jmp_5_1;
>>
>> jmp_5_1:
>>   (&L->ci->u.l.base[(int)1])->value_.b = (int)0;
>>   (&L->ci->u.l.base[(int)1])->tt_ = (int)1;
>>
>> Next instruction is:
>>
>> 6 [3] TEST     1 0
>> 7 [3] JMP       0 1 ; to 9
>>
>> Now this is saying that if register is _false_ then skip JMP else JMP.
>> Since we know that register 1 is false as above then the code should
>> skip the jump.
>>
>>   comparison_0_11 = (&L->ci->u.l.base[(int)1])->tt_ == (int)0;
>>   comparison_0_12 = (&L->ci->u.l.base[(int)1])->tt_ == (int)1;
>>   comparison_0_13 = (&L->ci->u.l.base[(int)1])->value_.b == (int)0;
>>   isfalse_0_10 = comparison_0_11 || comparison_0_12 && comparison_0_13;
>>   if (!(!(isfalse_0_10))) goto OP_TEST_do_jmp_5_14; else goto
>> OP_TEST_do_skip_5_15;
>>
>> So we go to skip_5_15:
>>
>> 8 [3] LOADBOOL 0 1 0
>>
>> OP_TEST_do_skip_5_15:
>>   (&L->ci->u.l.base[(int)0])->value_.b = (int)1;
>>   (&L->ci->u.l.base[(int)0])->tt_ = (int)1;
>>   goto jmp_9_2;
>>
>> So you see that the local IX is set to false above which is what we
>> want. Next we jump to 9_2 which is the start of the return statement.
>
> part 3
> -------
>
> Here we essentially have a repeat of what was in entry block:
>
> 9 [3] LOADK     1 -1 ; 10
> 10 [3] TEST     1 1
> 11 [3] JMP       0 0 ; to 12
>
> jmp_9_2:
>   (&L->ci->u.l.base[(int)1])->value_.i = (&cl->p->k[(int)0])->value_.i;
>   (&L->ci->u.l.base[(int)1])->tt_ = (&cl->p->k[(int)0])->tt_;
>   comparison_0_17 = (&L->ci->u.l.base[(int)1])->tt_ == (int)0;
>   comparison_0_18 = (&L->ci->u.l.base[(int)1])->tt_ == (int)1;
>   comparison_0_19 = (&L->ci->u.l.base[(int)1])->value_.b == (int)0;
>   isfalse_0_16 = comparison_0_17 || comparison_0_18 && comparison_0_19;
>   if (!(isfalse_0_16)) goto OP_TEST_do_jmp_9_20; else goto OP_TEST_do_skip_9_21;
>
> So the JMP is not taken as register 1 is _true_ and we go to 9_21:
>
> 12 [3] LOADBOOL 1 0 0
>
> OP_TEST_do_skip_9_21:
>   goto jmp_12_3;
>
> jmp_12_3:
>   (&L->ci->u.l.base[(int)1])->value_.b = (int)0;
>   (&L->ci->u.l.base[(int)1])->tt_ = (int)1;
>
> Above sets the register 1 to false and this is the return value.
>
> Next the return instruction begins:
>
>   L->top = &L->ci->u.l.base[(int)2];
>   comparison_0_22 = cl->p->sizep > (int)0;
>   if (comparison_0_22) goto OP_RETURN_if_sizep_gt_0_12_23; else goto
> OP_RETURN_else_sizep_gt_0_12_24;
>
>
> So as far as I can see the generated code is correct and matches the
> Lua bytecode.
>


Having said that ... I have doubts now. Let me check this again.

Regards