This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Question about comment in vect_prune_runtime_alias_test_list


Hi,
In that function, we have below comments:

  /* Basically, for each pair of dependent data refs store_ptr_0
     and load_ptr_0, we create an expression:

     ((store_ptr_0 + store_segment_length_0) <= load_ptr_0)
     || (load_ptr_0 + load_segment_length_0) <= store_ptr_0))

     for aliasing checks.  However, in some cases we can decrease
     the number of checks by combining two checks into one.  For
     example, suppose we have another pair of data refs store_ptr_0
     and load_ptr_1, and if the following condition is satisfied:

     load_ptr_0 < load_ptr_1  &&
     load_ptr_1 - load_ptr_0 - load_segment_length_0 < store_segment_length_0

     (this condition means, in each iteration of vectorized loop,
     the accessed memory of store_ptr_0 cannot be between the memory
     of load_ptr_0 and load_ptr_1.)

     we then can use only the following expression to finish the
     alising checks between store_ptr_0 & load_ptr_0 and
     store_ptr_0 & load_ptr_1:

     ((store_ptr_0 + store_segment_length_0) <= load_ptr_0)
     || (load_ptr_1 + load_segment_length_1 <= store_ptr_0))

     Note that we only consider that load_ptr_0 and load_ptr_1 have the
     same basic address.  */

I don't quite follow the example.  If (load_ptr_0 < load_ptr_1) is
true, IIUC store_ptr_0 & load_ptr_0 and store_ptr_0 & load_ptr_1 can
be simplified to:

     ((store_ptr_0 + store_segment_length_0) <= load_ptr_0)
     || (load_ptr_1 + load_segment_length_1 <= store_ptr_0))

under conditions:
     load_ptr_0 + load_segment_length_0 < load_ptr_1 + load_segment_length_1

But I can't get this from:

     load_ptr_0 < load_ptr_1  &&
     load_ptr_1 - load_ptr_0 - load_segment_length_0 < store_segment_length_0


Also it looks like the comment doesn't match code either.  The code
merges checks for store_ptr_0 & load_ptr_0 and store_ptr_0 &
load_ptr_1, but the comment is more like mixed checks between the two
pairs.

Thanks,
bin


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]