| phony screws it up |
|
More...
|
almost 15 years ago
|
| finally done with this i hope |
|
More...
|
almost 15 years ago
|
| man i am stupid tonight |
|
More...
|
almost 15 years ago
|
| the second half of this commit, oops. |
|
More...
|
almost 15 years ago
|
| abort is void since rc can never be used |
|
More...
|
almost 15 years ago
|
| need the type headers in a1u.h to compile |
|
More...
|
almost 15 years ago
|
| adding the A1U_Read_params + A1U_Settings/a1u_settings input |
|
More...
|
almost 15 years ago
|
| splitting up settings into: 1. A1D_Settings_t/a1d_settings for device-dependent settings 2. A1U_Settings_t/a1u_settings for device-independent settings |
|
More...
|
almost 15 years ago
|
| adding "if(target == my_rank && a1_settings.network_bypass)" to memcpy routines so we can turn it off and on for benchmarks and i suspect some application codes will benefit from this. maybe should be threshold-based instead. |
|
More...
|
almost 15 years ago
|
| fixed trivial bug that caused this test to crash |
|
More...
|
almost 15 years ago
|
| adding a global variable that will ensure handoff does not go into an infinite loop. when we are inside handoff function, it does not make sense to call handoff again |
|
More...
|
almost 15 years ago
|
| adding localcpy implementation for getv |
|
More...
|
almost 15 years ago
|
| removing debug prints, my bad!! |
|
More...
|
almost 15 years ago
|
| 1) fixing dispalcement cacluation in pack/unpack functions 2) implementing local cpy operations |
|
More...
|
almost 15 years ago
|
| oops |
|
More...
|
almost 15 years ago
|
| now one can do the whole testing from command line with make minimal-effort-wise in some sense |
|
More...
|
almost 15 years ago
|
| makefile target now submits all tests, but note that they will blow away old test outputs |
|
More...
|
almost 15 years ago
|
| (1) fix an minor issue wherein e.g. A1D_Get was called after A1D_Get_memcpy for local ops (2) added all A1U_*_memcpy ops in the appropriate ADL functions |
|
More...
|
almost 15 years ago
|
| adding localcopy implementations for put,get and acc - inprogress exposing locking through A1D functions adding armci error function |
|
More...
|
almost 15 years ago
|
| add a few comments and started working on the new memregion-based API |
|
More...
|
almost 15 years ago
|
| restored the memory segments part of the API for now but it will be deprecated as soon as we move put/get/acc to use memregions+offsets instead of pointers |
|
More...
|
almost 15 years ago
|
| adding DMAPP test from the Cray GNU/DMAPP manual |
|
More...
|
almost 15 years ago
|
| meh |
|
More...
|
almost 15 years ago
|
| this has the no-scaling test in it which shows that FMA makes this optimization pointless |
|
More...
|
almost 15 years ago
|
| change unrolling |
|
More...
|
almost 15 years ago
|
| negligible change |
|
More...
|
almost 15 years ago
|
| so this is the data from the accumulate_unaligned.x test as of this date |
|
More...
|
almost 15 years ago
|
| it's working now. still thinking about results. |
|
More...
|
almost 15 years ago
|
| fix all syntax errors and incorrect use of assert :-( |
|
More...
|
almost 15 years ago
|
| ready to test on BG/P |
|
More...
|
almost 15 years ago
|
