Re: [Qemu-devel] Trouble with adding TCG branch instructions

[Soumyadeep Ghosh]

Thanks a lot! The problem was with using dead temporaries. The ts->val_type 
assertion failed because the value of val_type for these instructions was 
TEMP_VAL_DEAD.

Deep

On Mar 16, 2013, at 5:44 AM, Peter Maydell <address@hidden> wrote:

> On 15 March 2013 18:59, Soumyadeep Ghosh <address@hidden> wrote:
>>        int label_rn = gen_new_label();
>>        int label_done = gen_new_label();
>>        tcg_gen_brcond_tl (TCG_cond_LT, tmp2, tmp, label_rn);      // If tmp2 
>> <  tmp, branch to label_rn
>>        store_reg (s, 20, tmp2);                                              
>>                       // Store tmp2 in one of the newly added registers
>>        tcg_gen_br (label_done);                                              
>>                       // Branch to end
>>        gen_set_label (label_rn);                                             
>>                       // Set label_rn to the current code location
>>        store_reg (s, 20, tmp);                                               
>>                       // Store tmp in one of the newly added registers
>>        gen_set_label (label_done);                                           
>>               // Set label_done to the current code location
>>        // Rest of the program
> 
> The TCG README says that a TCG temporary is live only in a basic
> block, and that a basic block is terminated by a branch instruction.
> You're trying to use a temporary across a branch instruction,
> which isn't going to work (the temporary is dead at the end
> of the block).
> 
>> When I try to emulate my ARM program using the compiled version of
>> the modified emulator, I get an error (Uncaught signal 11). Debugging
>> with gdb tells me that the following assertion on line 1885 of
>> tcg/tcg.c (the function tcg_reg_alloc_op) raises the signal:
>> 
>>        assert (ts->val_type == TEMP_VAL_REG);
>> 
>> Does anyone have any insight into what is causing the problem?
> 
> I haven't checked, but this is probably the register allocator
> complaining because it's trying to handle the operand to an
> op and it's dead/doesn't exist (because of the problem above).
> 
> -- PMM