Hi, I've found that commit b531973419a056696e6f88d813769aa4f1f1aee6 doesn't work Author: Jeff Squyres <[email protected]> List-Post: [email protected] Date: Tue Apr 22 19:48:56 2014 +0000
caused new failures with derived datatypes. Collectives return incorrect results. But it doesn't reproduce on a regular manner. Need many iterations to reproduce. Examples are: - gather with short_int datatype in amount of 20000 and in-place is on, - allgather with a random generated derived datatype in amount of 20000. It leads to the following output (this is a little fragment): ../../../opal/datatype/opal_datatype_position.c:72 Pointer 0x7fff8119c11d size 16 is outside [0x7fff80cd6040,0x7fff80f4702d] for base ptr 0x7fff80cd6040 count 40000 and data Datatype 0xa72080[] size 38 align 16 id 0 length 7 used 6 true_lb 0 true_ub 45 (true_extent 45) lb 0 ub 64 (extent 64) nbElems 6 loops 0 flags 104 (commited )-c-----GD--[---][---] contain OPAL_INT2 OPAL_INT4 OPAL_UINT8 OPAL_FLOAT4 OPAL_FLOAT16 --C---P-D--[---][---] OPAL_UINT8 count 1 disp 0x0 (0) extent 8 (size 8) --C---P-D--[---][---] OPAL_FLOAT4 count 1 disp 0xa (10) extent 4 (size 4) --C---P-D--[---][---] OPAL_INT4 count 1 disp 0xe (14) extent 4 (size 4) --C---P-D--[---][---] OPAL_INT2 count 1 disp 0x17 (23) extent 2 (size 2) --C---P-D--[---][---] OPAL_FLOAT16 count 1 disp 0x19 (25) extent 16 (size 16) --C---P-D--[---][---] OPAL_INT4 count 1 disp 0x29 (41) extent 4 (size 4) -------G---[---][---] OPAL_END_LOOP prev 6 elements first elem displacement 0 size of data 38 Optimized description -cC---P-DB-[---][---] OPAL_UINT8 count 1 disp 0x0 (0) extent 8 (size 8) -cC---P-DB-[---][---] OPAL_UINT1 count 8 disp 0xa (10) extent 1 (size 8) -cC---P-DB-[---][---] OPAL_UINT1 count 22 disp 0x17 (23) extent 1 (size 22) -------G---[---][---] OPAL_END_LOOP prev 3 elements first elem displacement 0 size of data 38 Best regards, Elena
