Re: [PATCH V2] RISC-V: Rework Phase 5 && Phase 6 of VSETVL PASS

[Richard Sandiford <richard.sandiford@arm.com>]

Jeff Law via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
> On 6/9/23 04:41, juzhe.zhong@rivai.ai wrote:
>> @@ -4342,135 +4510,81 @@ pass_vsetvl::cleanup_insns (void) const
>>       }
>>   }
>>   
>> +/* Return true if the SET result is not used by any instructions.  */
>> +static bool
>> +has_no_uses (basic_block cfg_bb, rtx_insn *rinsn, int regno)
>> +{
>> +  /* Handle the following case that can not be detected in RTL_SSA.  */
>> +  /* E.g.
>> +	  li a5, 100
>> +	  vsetvli a6, a5...
>> +	  ...
>> +	  vadd (use a6)
>> +
>> +	The use of "a6" is removed from "vadd" but the information is
>> +	not updated in RTL_SSA framework. We don't want to re-new
>> +	a new RTL_SSA which is expensive, instead, we use data-flow
>> +	analysis to check whether "a6" has no uses.  */
> I'm a bit surprised there wasn't a reasonable way to update the RTL SSA 
> framework for this case.  If we were to remove the entire vadd, then we 
> would have to update the uses of a6.  If we have that capability, then I 
> would expect we could refactor the updating code so that we had an API 
> to remove an operand from an instruction.
>
> In fact, if we have a constant propagator in the RTL SSA framework, 
> wouldn't it have to have this capability?

RTL-SSA isn't supposed to be feature-complete in its current state.
So yeah, if something is missing, it's better to add it to RTL-SSA
rather than work around it in consumers.

(Responding without fully understanding the context though, sorry.)

Thanks,
Richard