I noticed some duplicate functions in the datatype creation code. E.g., there's a DLOOP_Type_blockindexed_count_contig() function in .../datatype/dataloop/dataloop_create_blockindexed.c and a MPIDI_Type_blockindexed_count_contig() function in .../datatype/mpid_type_blockindexed.c. These look to be identical after macro expansion. Is there a reason that I shouldn't remove one? My plan is to make the functions like DLOOP_Type_blockindexed_count_contig into PREPEND_PREFIX(DLOOP_Type_blockindexed_count_contig) and remove the MPID_ versions. Does this make sense? -d
Hi Darius, I do not recall exactly why this is the case. I believe that the calls were originally in the mpid_ files, but I found that I needed them in the dataloop files when I went to pull that code out (an exercise that was never quite completed). So I made extra versions in the dataloop subdirectory. It's possible that the versions might end up with different type sizes under certain circumstances? That's all I can think of. If you do need/want to remove one, be sure to leave the ones in the dataloop subdirectory. Thanks, Rob On Jul 22, 2008, at 5:01 PM, Darius Buntinas wrote:
I noticed some duplicate functions in the datatype creation code. E.g., there's a DLOOP_Type_blockindexed_count_contig() function in .../datatype/dataloop/dataloop_create_blockindexed.c and a MPIDI_Type_blockindexed_count_contig() function in .../datatype/ mpid_type_blockindexed.c. These look to be identical after macro expansion.
Is there a reason that I shouldn't remove one? My plan is to make the functions like DLOOP_Type_blockindexed_count_contig into PREPEND_PREFIX(DLOOP_Type_blockindexed_count_contig) and remove the MPID_ versions.
Does this make sense?
-d
Thanks, Rob. I fixed this in r1150 just as you suggested. The ones in datatype/ were removed and the ones in datatype/dataloop were exported. -d On 07/22/2008 09:28 PM, Rob Ross wrote:
Hi Darius,
I do not recall exactly why this is the case. I believe that the calls were originally in the mpid_ files, but I found that I needed them in the dataloop files when I went to pull that code out (an exercise that was never quite completed). So I made extra versions in the dataloop subdirectory.
It's possible that the versions might end up with different type sizes under certain circumstances?
That's all I can think of. If you do need/want to remove one, be sure to leave the ones in the dataloop subdirectory.
Thanks,
Rob
On Jul 22, 2008, at 5:01 PM, Darius Buntinas wrote:
I noticed some duplicate functions in the datatype creation code. E.g., there's a DLOOP_Type_blockindexed_count_contig() function in .../datatype/dataloop/dataloop_create_blockindexed.c and a MPIDI_Type_blockindexed_count_contig() function in .../datatype/mpid_type_blockindexed.c. These look to be identical after macro expansion.
Is there a reason that I shouldn't remove one? My plan is to make the functions like DLOOP_Type_blockindexed_count_contig into PREPEND_PREFIX(DLOOP_Type_blockindexed_count_contig) and remove the MPID_ versions.
Does this make sense?
-d
participants (2)
-
Darius Buntinas -
Rob Ross