Re: [PATCH] Loop unswitching: support gswitch statements.

[Andrew MacLeod <amacleod@redhat.com>]

On 12/9/21 07:59, Martin Liška wrote:
> On 12/3/21 15:09, Andrew MacLeod wrote:
>> On 12/2/21 11:02, Martin Liška wrote:
>>> On 12/2/21 15:27, Andrew MacLeod wrote:
>>>> ranger->gori().outgoing_edge_range_p (irange &r, edge e, tree name, 
>>>> *get_global_range_query ());
>>>
>>> Thank you! It works for me!
>>>
>>> Martin
>>>
>> btw, this applies to names not  on the stmt as well.   The function 
>> returns TRUE if there is an outgoing range calculable, and false if 
>> not.  so:
>>
>> a = b + 20
>> if (a > 40)        <- returns TRUE for outgoing range of 'a' or 'b'
>>      {
>>         c = foo()
>>         if (c > 60)   <- returns false for 'a' or 'b'
>>             {
>>                 if (b < 100)   <- Also returns TRUE for 'a' or 'b'
>
> Hello.
>
> That's quite interesting support. However, for the bug mentioned 
> earlier in this email conversion,
> I only want local range for a gcond/gswitch, ignoring all path leading 
> to the statement.
>
> Martin 
In which case just directly invoking gori with the global range query 
should be fine.   You could go directly to the lower level routines like 
gimple_range_calc_op1, but this saves you from doing a little footwork 
to call them.  Ultimately, it works out the same for you on the true 
side of
if (a > 40)
as  setting up and invoking:

    if (gimple_range_calc_op1  (r, stmt, int_range<2> 
(boolean_true_node, boolean_true_node), [40,40])

Its just less work to do it thru gori()->outgoing_edge_range_p ()   :-)

Andrew