Lines Matching refs:threads

22 STREAMS measures the total memory bandwidth achievable when running `n` independent threads or proc…
25 …read or process works on its own data, completely independently of other threads or processes data.
29 As more threads or processes are added, the bandwidth achieved begins to saturate at some `n`, gene…
34 …connect between the cores and their various levels of caches and how the threads or processes are …
45 …ware subsets of the shared memory node. The Unix operating system allows threads and processes to …
49 - Thread or process **mapping** (assignment) to hardware subsets when more threads or processes are…
58 …ce only one process (on a particular core) will ever touch its data. For threads, care must be tak…
59 …ill suffer if the first thread initializes an entire array that multiple threads will later access.
86 `ex69f` with four OpenMP threads without `mpiexec` and see almost perfect scaling.
87 The CPU time of the process, which is summed over the four threads in process, is the same as the w…
96 Running under `mpiexec` gives a very different wall clock time, indicating that all four threads ra…
114 …implementation is, by default, binding the process (including all of its threads) to a single core.
127 …_time()` can sometimes produce misleading results when run with multiple threads. Consider again t…
128 … example. For an OpenMP parallel loop with enough available cores and the proper binding of threads
129 to cores, one expects the CPU time for the process to be roughly the number of threads times the wa…
131 …bserved the recorded CPU for the second loop to be roughly the number of threads times the wall cl…
161 These are a result of a bug in the gcc `spread` option, placing more threads in one NUMA domain tha…
162 For example, with gcc, the `OMP_DISPLAY_AFFINITY` shows that for 28 threads, 12 are placed on NUMA …
170 using `OMP_DISPLAY_AFFINITY` confirms, for example, that 14 threads (out of 28) are assigned to eac…