Tags

follow dev16, clear a serious bug with grid_index_1dim in lattice_set.h, but still run result wrony, all the test run in snsc with na100-ins_gpu025, very emergency!!!

follow dev15, add #pragma optimize(5) in all src code, not clear improve or not.

follow dev14, 1. replace MPI_Isend with MPI_SendRcev; 2. new comm vec by malloc instead of cudaMallocHost in clover dslash (lattice_set.h) to cut down pinned memory use; not clear improve or not.

follow dev13, finish clover dslash (mpi version) and tun right in snsc with 2  NV-A100, good jod!!!

follow dev12, cut half of comm time (x wards, caused by "fake edge") in wilson dslash, very useful in x ward grid!!!

follow dev11, replace some give with get in define.h.

follow dev10, replace malloc with cudaMallocHost in lattice_set.h, big improve!!!

follow bug6, bug has been cleaned:"The compiler initializer is not necessarily zero (there may be no initializer at all), so host params[ DAGGER ] = NO USE ; // needed!!!"

follow dev9, find a bug in wilson dslash, not in cg or bicg, may caused by dag......

follow dev8, done dev5::task:1.3. cg function optimize done!!! now waiting for multi threading test in snsc. :)

follow dev7, done dev5::task:1.2. cg function run right!!!

follow dev6, done dev5::task:1.1. cg function build (first version);

follow stab15, done dev5::task:1.0. cg function define;

follow stab14, done dev5::task:4. namespace;

follow dev5, done task:2. multi i;

follow stab13, give up dev2 - bug5, task: 1. cg; 2. multi i; 3. mpi version; 4. namespace;

follow dev4, find io bug in bistabcg: the inputed b_e's norm==0 !!!

follow dev3 and bug4, debug in docker...

change define.h, redefine some vals, then a bug show...

follow bug3, fix '_wilson_dslash_all', but bug just change and still here...