Initial ADIOS2 implementation #338
Conversation
| @@ -1038,6 +1044,137 @@ int Geometry::Draw(pugi::xml_node & node) | |||
| Dot(x, y, z); | |||
| } | |||
| fclose(f); | |||
| } else if (strcmp(n.name(), "ADIOS2") == 0) { | |||
| lbRegion crop = getRegion(n.parent()); | |||
There was a problem hiding this comment.
This also has to be in #ifdef
| @@ -379,6 +389,24 @@ if test "x${with_hdf5}" != "xno"; then | |||
| fi | |||
| fi | |||
|
|
|||
| if test "x${with_adios}" != "xno"; then | |||
| adios2_config="${ADIOS_DIR}/bin/adios2-config" | |||
There was a problem hiding this comment.
I don't see where ADIOS2_USE_MPI is set to true.
It has to be:
- Add
ADIOS2_USE_MPI=1in theifinconfigure.ac - Add
AC_SUBST(ADIOS2_USE_MPI)at the bottom ofconfigure.ac - Add
#undef ADIOS2_USE_MPIto insrc/config.h.in
There was a problem hiding this comment.
ADIOS2_USE_MPI is an ADIOS config flag (the adios2-config returns it in --flags string) - I'm abusing it a little now
This is to be changed I think:
- Option 1: If TCLB is MPI then ADIOS must be MPI and vice versa, nice but it doesn't make sense even for testing imho?
- Option 2: Enable adios in MPI version of TCLB and require ADIOS2_USE_MPI.
There was a problem hiding this comment.
I think we need our own flag for enabling ADIOS then. Remember that if an flag is in the compilation (like -DFLAG) and not in the source (like #define FLAG 1) then make cannot see the change and will not recompile properly. Please make a flag for the ADIOS support as described above (maybe WITH_ADIOS)
TCLB is always MPI, we don't have an option for that.
|
@llaniewski TCLB is only X-decomposed? or X-Z? |
|
@mdzik TCLB is MPI decomposed in Y and Z. It is not divided in X (it's aligned with warps of 32 threads in X). |
|
@mdzik Please do the |
|
On my TODO - I've got submission deadline tomorrow
sob., 12 gru 2020 o 03:50 Łukasz Łaniewski-Wołłk <[email protected]>
napisał(a):
… @mdzik <https://github.com/mdzik> Please do the #ifdefs, as currently the
tests don't go through and this cannot be compiled without ADIOS.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#338 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AA3A5R2ZRFHUD2IKDN2SAA3SULK6VANCNFSM4TWNN7HQ>
.
|
|
@mdzik thanks for the update |
|
@mdzik Any update on this? Are you using ADIOS in your setups? (or TCLB in general now?) |
Reference:
ADIOS2
dataReader
Digital rock data
Idea (TCLB d2q9 karman example:
TCLB output
Reader output