Exascale FastForward
  1. Exascale FastForward

ParallelIO

Public

Network

 
AuthorCommitMessageCommit Date
Y. PoliusY. Polius
a063aa79154Added conditional use piovdc to pio; Altered piodarray.F90.in to only build real functions; Improved read/write vdc2 modularity
Y. PoliusY. Polius
70848fcf089renamed libpiovdc.F90 -> piovdc.F90; updated Makefile with the new name
Y. PoliusY. Polius
39f17ff81a5modified libpiovdc.F90 to have automatic arrays; added special ifdef for mpi; cleaned up pio use statements
Y. PoliusY. Polius
580a906d56aCreated VDC2_var_desc_t type; Added ifdef Compression to enable optional VDC2 var type
Y. PoliusY. Polius
3b659fb82cfAdded refinement level option for writing/reading VDC2 data
Y. PoliusY. Polius
48f31ad9521fixed _COMPRESSION defines inside of piodarray.F90.in for VDC2
Y. PoliusY. Polius
8d09d25913aMakefile for conditional build
Y. PoliusY. Polius
c8d110f4d18added --enable-compression option to optionally build libpiovdc
Jim EdwardsJim Edwards
2b78e644e1evdc branch
Jim EdwardsJim Edwards
fc6b0e0f0e1vdc branch
Jim EdwardsJim Edwards
b3e2d4b9709new branch tag
John DennisJohn Dennis
c630f93f4cdTest scripts used for the PIO paper.
John DennisJohn Dennis
1616833f1efModifications due to work on PIOpaper.
John DennisJohn Dennis
1569187a95cUpdates for Hopper II.
John DennisJohn Dennis
d21fe5a58d7Added some higher core count configurations.
John DennisJohn Dennis
54a53f46d62
John DennisJohn Dennis
eb651a2321f
John DennisJohn Dennis
bbf662c70e6Added some more core configurations for POPB and WRFB benchmarks
John DennisJohn Dennis
f72ffeaecdaFixed a number of issues related to the Lustre optimization work including: * Added new PIO_getnum_ost and PIO_setnum_ost that get and set the default number of OST over which to strip. * Fixed minor issue with the name of the test files that are created.
John DennisJohn Dennis
b41da432e1eA couple of changes from compiling on Hopper
John DennisJohn Dennis
5122e906305Forgot my new file for calculating decompositions.
John DennisJohn Dennis
e95ed5b7c3cAdded the new decomposition algorithm
John DennisJohn Dennis
83c20c109efBranch for lustre optimization work.
Jim EdwardsJim Edwards
3fcb3fc7484new trunk tag
Jim EdwardsJim Edwards
3d59ca76134initialize holegridsize
Jim EdwardsJim Edwards
ae07fba9467fix nightly build issues
Jim EdwardsJim Edwards
4556501e72eport to bluewaters
Jim EdwardsJim Edwards
b3a2fa1b40ftry yellowstone again ;
Jim EdwardsJim Edwards
03e5dd27993nominally working on cetus
Jim EdwardsJim Edwards
216df9a4934onesided mpiio flags for mira
Jim EdwardsJim Edwards
a4a94e9257btry new macros file
Jim EdwardsJim Edwards
406daf95e60mail results of cetus testing
Jim EdwardsJim Edwards
2c79831f0aefine tuning cetus testing
Jim EdwardsJim Edwards
6b8f14f6d4ctests are working on cetus now
Jim EdwardsJim Edwards
dd7ad9f73datry again on cetus
Jim EdwardsJim Edwards
d80aa03c6a6port to cce compiler
Jim EdwardsJim Edwards
bd384b67465update externals
Jim EdwardsJim Edwards
6c61a1b67c1more cmake work
Jim EdwardsJim Edwards
0778e5bc87efix typo
Jim EdwardsJim Edwards
e937cdeefe7policy was a bad idea
Jim EdwardsJim Edwards
e010df33698update build flags for mira
Jayesh KrishnaJayesh Krishna
0143589728cAdding some fillvalue tests for reads. These are similar to the write tests added in r1485 and r1488
Jayesh KrishnaJayesh Krishna
6194e3fa5a5Remove debug prints and re-introduce deletion of test output files - added in r1489
Jim EdwardsJim Edwards
42aad2abbdffixed fillval test
Jayesh KrishnaJayesh Krishna
f31b63e8c42Fixing bug in the fillvalue tests. Making sure that we read the complete data before checking the results. The previous versions used the same iodesc for read and write, so the explicit or implicit holes in the write iodesc (write compdof) were also present in the read iodesc (read compdof). Hence the reads were not reading the complete array
Jim EdwardsJim Edwards
e337fb31527 make wbuf allocatable
Jayesh KrishnaJayesh Krishna
4f7119b7697Adding a test for writing data with fillvalues where the data holes are implicitly specified (rather than filling the compdof with 0s for the invalid data values, the compdof for invalid values is not specified). The old test where data holes were explicitly specified (by filling corresponding compdofs with 0s) is now renamed appropriately
Jim EdwardsJim Edwards
0ce01f75716fix issue in netcdf4p interface
Jim EdwardsJim Edwards
9daa924c337fix for openmpi: MPI_Datatype is not an integer and cannot be used in switch statements
Jayesh KrishnaJayesh Krishna
518547a1d83Adding a write test with fillvalues. The compdof is defined with holes (0s) and the user specifies the fillvalue (-2) explicitly during the write.