On 20210916@08:54, Paul Gevers wrote:
> Hi Michele,
> 
> On 16-09-2021 02:40, Michele Martone wrote:
> > Dear Paul,
> > 
> > The problem seems librsb-sided.
> > 
> > Assuming librsb fails at detecting cache memory size, maybe can you
> > first export e.g.
> >   RSB_USER_SET_MEM_HIERARCHY_INFO="L2:4/64/512K,L1:8/64/32K" 
> > and see if this avoids that?
> > Same with 
> >   OMP_NUM_THREADS=1
> > explicitly?
> > 
> > Maybe you send me outputs of  `rsbench -C` and `rsbench -I` ?
> 
> See below. By the way, I'm surprised to see negative free memory
> reported there. Is that an integer overflow bug? We have 255GB memory on
> this machine and nearly all of it is free.
> 
> > If I had a guest account to such an armhf machine it would be a bit
> > easier.
> 
> If I'm correct you're not a DD,
Correct.

> but if there's a DD that is willing to
> support your request, I can temporarily give you access if really needed.
Let's see what can be done without that first.

> Paul
> 
> root@autopkgtest-lxc-bexxro:/tmp/autopkgtest-lxc.ghhdpc6_/downtmp/build.J9f/src#
> lsmem
> RANGE                                  SIZE  STATE REMOVABLE       BLOCK
> 0x0000000080000000-0x00000000bfffffff    1G online       yes           2
> 0x0000080000000000-0x000008007fffffff    2G online       yes   8192-8193
> 0x00000800c0000000-0x0000081fffffffff  125G online       yes   8195-8319
> 0x0000400000000000-0x00004000bfffffff    3G online       yes 65536-65538
> 0x0000400100000000-0x0000401fffffffff  124G online       yes 65540-65663
> 
> Memory block size:         1G
> Total online memory:     255G
> Total offline memory:      0B
> 
> 
> root@autopkgtest-lxc-bexxro:/tmp/autopkgtest-lxc.ghhdpc6_/downtmp/build.J9f/src#
> rsbench -CERROR 0xfffffc00 : The requested feature (e.g.:blocking) is
> not available because it was opted out or not configured at built time.
You have up to 160 threads.
The hardcoded default (overridable ./configure --with-max-threads N) is likely 
128.

Can you export
 OMP_NUM_THREADS=3 # or anything below the hardcoded limit of 128
and rerun rsbench -C
and/or redo the octave-sparsersb experiments ?
If that was the problem, 
 $ rsbench -B
shall run fine (a few s will suffice); and
 $ ./rsbench  -oa -Ob --dense 100 --verbose --verbose 
--write-performance-record foo | grep ^pr.*1
shall lead to something similar to
 pr:    1:r_r  dense-100x100-10000nz 100 100 10000 1 D G N  4  0  0 4.1664 
4.1664 16 16 1525.20 1.311e-05 0.000e+00 1.311e-05 0.000e+00 2.420e-02 9.46e+00 
6.16e+00 0 2.00e-02

If that's the problem, maybe I can devise a fix to make the error easier
to spot.

Michele

> root@autopkgtest-lxc-bexxro:/tmp/autopkgtest-lxc.ghhdpc6_/downtmp/build.J9f/src#
> rsbench -I
> hwloc size of cache level 1: 65536
> hwloc size of cache level 2: 1048576
> detected max available cores/threads : 160
> ...
> detected max OpenMP procs : 160
> detected 2 levels of cache
> L1 size: 65536
> L2 size: 1048576
> sysconf() : 4096 bytes per pagesize
> sysconf() : 65553496 physical pages
> sysconf() : -2075820032 bytes (2116 MB) of physical memory
> sysconf() : 38395759 available (free) physical pages
> sysconf() : -1644761088 available (free) physical memory
> sysconf() , processors : 160
> sysconf() , processors online : 160
> sysconf() : no level 1 cache
> sysconf() : no level 2 cache
> sysconf() : no level 3 cache
> sysconf() : no level 4 cache
> 8 bits per byte. Good.
> cache block size                : 16384
> SHRT_MAX : 32767
> SHRT_MIN : -32768
> USHRT_MAX : 65535
> INT_MIN : -2147483648
> INT_MAX : 2147483647
> UINT_MAX : 4294967295
> LONG_MAX : 2147483647
> LONG_MIN : -2147483648
> ULONG_MAX : 4294967295
> LLONG_MAX : 9223372036854775807
> LLONG_MIN : -9223372036854775808
> ULLONG_MAX : 18446744073709551615
> RSB_MARKER_COO_VALUE : 2147483138
> RSB_MARKER_NNZ_VALUE : 2147483393
> RSB_SUBM_IDX_MARKER : 2147483647
> RSB_MAX_ALLOCATABLE_MEMORY_CHUNK: 4294967295
> timing min delta (if negative, don't complain with us)   : 0 s
> timing granularity : 2.09045e-07 s
> CFLAGS   : -g -O2
> -fdebug-prefix-map=/build/librsb-SgJkYm/librsb-1.2.0.9+real+dfsg=.
> -fstack-protector-strong -Wformat -Werror=format-security -O3 -fopenmp
> -std=c99
> CC       : gcc
> memhinfo :
> detected free  memory : -537382912
> detected total memory : -2075820032
> ...
> for array sized 1048576 elems, took 0 s for linear search and 0 s for
> binary search for element 7, in 222139 tries, for a total of 0.100001 s
> (ignore this:-1200778098)

Attachment: signature.asc
Description: PGP signature

Reply via email to