HPCG crash when nx=440 ny=440 nz=424 about hpcg

[luszczek]

Please take a look at Geometry.hpp line 37. Make sure that HPCG_NO_LONG_LONG is not defined so that sizeof(global_int_t) is at least 8. It seems that your global indices use 32-bit integers. More information about your build would help indicate the specific flag that caused this.

[NguyenDacLiem]

Thanks for your reply, I have a small mistake in my status, I updated it. I sure the HPCG_NO_LONG_LONG is not defined and sizeof(global_int_t) is 8
When I set nx = 440, ny=424, nz=424, the program run well. But when I increase ny to 440, it crashes. It also happens when I set nx=ny=nz=512. I add debug and below are the out put:
// If this assert fails, it most likely means that the global_int_t is set to int and should be set to long long
// This assert is usually the first to fail as problem size increases beyond the 32-bit integer range.
printf("LIEM %d %lld\n", sizeof(totalNumberOfNonzeros), totalNumberOfNonzeros);
assert(totalNumberOfNonzeros>0); // Throw an exception of the number of nonzeros is less than zero (can happen if int overflow)

Output:
nx = 440 ny = 424 nz =424
LIEM 8 34105190392
LIEM 8 34105190392
LIEM 8 34105190392
LIEM 8 34105190392
nx = 440 ny = 440 nz = 424
LIEM 8 -33326285448
LIEM 8 -33326285448
LIEM 8 -33326285448
LIEM 8 -33326285448
LIEM 8 -33326285448
LIEM 8 -33326285448

[NguyenDacLiem]

#@Header

----------------------------------------------------------------------

- shell --------------------------------------------------------------

----------------------------------------------------------------------

SHELL = /bin/sh

CD = cd
CP = cp
LN_S = ln -s -f
MKDIR = mkdir -p
RM = /bin/rm -f
TOUCH = touch

----------------------------------------------------------------------

- HPCG Directory Structure / HPCG library ------------------------------

----------------------------------------------------------------------

TOPdir = .
SRCdir = $(TOPdir)/src
INCdir = $(TOPdir)/src
BINdir = $(TOPdir)/bin

----------------------------------------------------------------------

- Message Passing library (MPI) --------------------------------------

----------------------------------------------------------------------

MPinc tells the C compiler where to find the Message Passing library

header files, MPlib is defined to be the name of the library to be

used. The variable MPdir is only used for defining MPinc and MPlib.

MPdir = /root/opt/openmpi
MPinc = -I$(MPdir)/include
MPlib = $(MPdir)/lib/libmpi.so

----------------------------------------------------------------------

- HPCG includes / libraries / specifics -------------------------------

----------------------------------------------------------------------

HPCG_INCLUDES = -I$(INCdir) -I$(INCdir)/$(arch) $(MPinc)
HPCG_LIBS =

- Compile time options -----------------------------------------------

-DHPCG_NO_MPI Define to disable MPI

-DHPCG_NO_OPENMP Define to disable OPENMP

-DHPCG_CONTIGUOUS_ARRAYS Define to have sparse matrix arrays long and contiguous

-DHPCG_DEBUG Define to enable debugging output

-DHPCG_DETAILED_DEBUG Define to enable very detailed debugging output

By default HPCG will:

*) Build with MPI enabled.

*) Build with OpenMP enabled.

*) Not generate debugging output.

HPCG_OPTS = -DHPCG_ENABLE_DETAILED_DEBUG

----------------------------------------------------------------------

HPCG_DEFS = $(HPCG_OPTS) $(HPCG_INCLUDES)

----------------------------------------------------------------------

- Compilers / linkers - Optimization flags ---------------------------

----------------------------------------------------------------------

CXX = mpicxx
CXXFLAGS = $(HPCG_DEFS) -v -O3 -ffast-math -ftree-vectorize -ftree-vectorizer-verbose=0 -fopenmp

LINKER = $(CXX)
LINKFLAGS = $(CXXFLAGS)

ARCHIVER = ar
ARFLAGS = r
RANLIB = echo

----------------------------------------------------------------------

[NguyenDacLiem]

The issue is local_int_t is set to int not long long type. How can I run test by 64/64 - global and local are 64-bit without change source code?