测试运行

for ((i = 0; i < 608; i += 38)); do
    OMP_NUM_THREADS=38 OMP_PROC_BIND=close taskset -c ${i}-$((${i}+37)) numactl -m $((${i}/38)) ./stream_c.exe
done

参考测试结果中的Triad值，每个NUMA测试结果在120000~150000MB/s：

-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 141648512 (elements), Offset = 0 (elements)
Memory per array = 1080.7 MiB (= 1.1 GiB).
Total memory required = 3242.1 MiB (= 3.2 GiB).
Each kernel will be executed 20 times.
 The *best* time for each kernel (excluding the first iteration)
 will be used to compute the reported bandwidth.
-------------------------------------------------------------
Number of Threads requested = 38
Number of Threads counted = 38
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 17826 microseconds.
   (= 17826 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function    Best Rate MB/s  Avg time     Min time     Max time
Copy:          125994.0     0.018811     0.017988     0.020024
Scale:         135105.3     0.017003     0.016775     0.017418
Add:           144631.7     0.023637     0.023505     0.023769
Triad:         144521.8     0.023850     0.023523     0.024183
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------

for ((i = 0; i < 608; i += 38)); do
    OMP_NUM_THREADS=38 OMP_PROC_BIND=close taskset -c ${i}-$((${i}+37)) numactl -m $((${i}/38+16)) ./stream_c.exe
done

参考测试结果中的Triad值，每个NUMA的带宽约在380000~400000MB/s：

-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 141648512 (elements), Offset = 0 (elements)
Memory per array = 1080.7 MiB (= 1.1 GiB).
Total memory required = 3242.1 MiB (= 3.2 GiB).
Each kernel will be executed 20 times.
 The *best* time for each kernel (excluding the first iteration)
 will be used to compute the reported bandwidth.
-------------------------------------------------------------
Number of Threads requested = 38
Number of Threads counted = 38
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 5151 microseconds.
   (= 5151 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function    Best Rate MB/s  Avg time     Min time     Max time
Copy:          403423.6     0.005716     0.005618     0.005873
Scale:         399288.9     0.005857     0.005676     0.006221
Add:           432229.1     0.007951     0.007865     0.008594
Triad:         406941.0     0.008642     0.008354     0.009097
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------