13 Commits (1446d1575d560d4e6a5e3cb8c34f198fdeb71786)

Author SHA1 Message Date
Constantin Fürst fccc255aae rewrite the benchmark to meassure timings for the entire run of all threads, doing multiple sync-steps with the launch barrier as done in the qdp bench 11 months ago
Constantin Fürst 405166cbe8 add peak perf benchmark descriptors 1 year ago
Constantin Fürst b44b52b600 meassure nanoseconds instead of microseconds 1 year ago
Constantin Fürst e6df656845 use vector for timing - this update got lost somehow 1 year ago
Constantin Fürst bc8c4f8ab3 restructure of directory layout 1 year ago
Constantin Fürst 8065dd4345 collect entire duration vector and dont condense the information down 1 year ago
Constantin Fürst 8c5a061343 remove buggy option for multiple sizes 1 year ago
Constantin Fürst 1bfb1f316c ignore first five runs to reduce influence of warmup, add non-batch-descriptor batch loop for testing, calculate standard deviation for all three messurements 1 year ago
Constantin Fürst 5fa12feb7d add required explicit block_on_fault-option to the task submission so that the DSA can handle page faults 1 year ago
Constantin Fürst 4e9688224b create a custom barrier structure that allows synchronization of each iteration of the meassurement loop 1 year ago
Constantin Fürst 80d1b5f543 remove fractional average calculation in favour of std::accumulate and remove option to set thread affinity to core while keeping support for node affinity assignment 1 year ago
Constantin Fürst 659883a765 check the status of the batch operation too - forgot this 1 year ago
Constantin Fürst 22f3ed8956 clean up naming of structs, functions and files 1 year ago
Constantin Fürst cbdf9b3dcf rename execute_move to execute_dml_memcpy and rename the file that contains this function to benchmark-dml-memcpy.hpp 1 year ago
Constantin Fürst 7e8c9acbc3 implement batch operation and add control parameters to the ThreadArgs struct, also add more timing information: now submission and completion will be timed separately 1 year ago
Constantin Fürst 9083ba834f small changes to execute move; remove unused repetition-options, pass args as ref to allow for writing results, dont use numa-node-setting dml submit call 1 year ago
Constantin Fürst b14ca88e03 start implementation of benchmarks code, begin with state from test project, execute-move.hpp contains numa-aware task submit routine which is WIP 1 year ago