OSPRI

Settings | Report Duplicate

1

I Use This!

Inactive

Commits : Listings

Analyzed about 13 hours ago. based on code collected about 13 hours ago.

Commit Message	Contributor	Files Modified	Lines Added	Lines Removed	Code Location	Date
Sep 04, 2024 — Sep 04, 2025 Showing page 66 of 84 Search / Filter on:
minor correction in check for single process jobs	potluri	More...				almost 15 years ago
minor correction in the check for single process tests in collectives	potluri	More...				almost 15 years ago
adding measurements for remote completion	potluri	More...				almost 15 years ago
adding active variable to avoid multiple initializations. other cosmetic changes	potluri	More...				almost 15 years ago
error macro not defined at this level. should fix when reorganizing to OSPRI	Jeff Hammond as jhammond	More...				almost 15 years ago
memregion changed slightly	Jeff Hammond as jhammond	More...				almost 15 years ago
add TAU profiling for RMW and Flush. not sure how to do multicast profiling (one-to-many as in Flush_group)	Jeff Hammond as jhammond	More...				almost 15 years ago
add TAU profiling interface in p2p calls at ADL	Jeff Hammond as jhammond	More...				almost 15 years ago
fixing dumbness	Jeff Hammond as jhammond	More...				almost 15 years ago
size->sizeof oops	Jeff Hammond as jhammond	More...				almost 15 years ago
network bypass on all collectives except sync since flush still requires us to do something within some implementations sreeram - there was a slight bug in returning success instead of goto fn_exit since that does not close the enter/exit pair	Jeff Hammond as jhammond	More...				almost 15 years ago
there is some issue with DCMF collectives with 1 rank. i am going to skip init here and bypass all DCMF collectives when size=1	Jeff Hammond as jhammond	More...				almost 15 years ago
A1_ should not call A1_ rather A1D_ (Process_rank)	Jeff Hammond as jhammond	More...				almost 15 years ago
i think that the benefits of DMA in 2D are quite limited since DMA > memcpy only for very large patches and we know that DMA > memcpy only when NIC is quiet. NWChem is going to hammer network so 2D optimizations are likely not useful in this case. We can explain the tail of the benchmark in the text of the paper and why it is irrelevant that ARMCI beats OSPRI.	Jeff Hammond as jhammond	More...				almost 15 years ago
making buffer pool sizes a runtime option	potluri	More...				almost 15 years ago
fixed my errors	Jeff Hammond as jhammond	More...				almost 15 years ago
1. remove likely_if in dcmfd_collectives 2. added nic bypass thresholding to ADL	Jeff Hammond as jhammond	More...				almost 15 years ago
some initialization	potluri	More...				almost 15 years ago
adding ignore rule	Jeff Hammond as jhammond	More...				almost 15 years ago
performance bench adapted from ARMCI	potluri	More...				almost 15 years ago
initializing local variables properly	potluri	More...				almost 15 years ago
adding error alerts when memory allocation fails	potluri	More...				almost 15 years ago
finally	Jeff Hammond as jhammond	More...				almost 15 years ago
oops	Jeff Hammond as jhammond	More...				almost 15 years ago
need a makefile	Jeff Hammond as jhammond	More...				almost 15 years ago
sum and prod tests	Jeff Hammond as jhammond	More...				almost 15 years ago
forgot this file in the last checkin	potluri	More...				almost 15 years ago
have make some of the objects available globally for allreduce	potluri	More...				almost 15 years ago
Modifying allreduce implementation to use ccmi collectives instead of global collectives. Using the torus_binomial protocol instead of tree. Non-blocking version of allreduce has been temporarily removed. Will add it back after modifying for use of ccmi collectives	potluri	More...				almost 15 years ago
adding Allreduce test for SUM (tree based) and PROD (torus based)	potluri	More...				almost 15 years ago

←
1
2
…
62
63
64
65
66
67
68
69
70
…
83
84
→