|
14 | 14 | * with the following limitations:
|
15 | 15 | * 3a. In order to be referred to as "BabelStream benchmark results",
|
16 | 16 | * published results must be in conformance to the BabelStream
|
17 |
| -* Run Rules published at |
18 |
| -* http://github.com/UoB-HPC/BabelStream/wiki/Run-Rules |
19 |
| -* and incorporated herein by reference. |
| 17 | +* Run Rules incorporated herein by reference. |
20 | 18 | * The copyright holders retain the
|
21 | 19 | * right to determine conformity with the Run Rules.
|
22 | 20 | * 3b. Results based on modified source code or on runs not in
|
|
33 | 31 | * program constitutes acceptance of these licensing restrictions.
|
34 | 32 | * 5. Absolutely no warranty is expressed or implied.
|
35 | 33 | *———————————————————————————————————-------------------------------------------
|
| 34 | +* Run Rules: |
| 35 | +* |
| 36 | +* In order to generate a valid BabelStream result measuring |
| 37 | +* *main memory bandwidth*, you should obey the following rules: |
| 38 | +* |
| 39 | +* 1. The array size must be large enough that increasing the array size does not |
| 40 | +* drastically change the reported bandwidth. If you satisfy this you will no |
| 41 | +* longer be affected by any cache effects. |
| 42 | +* |
| 43 | +* 2. Each kernel should take a few milliseconds. The resolution of the host timer |
| 44 | +* is probably in milliseconds so each kernel execution should be longer than the |
| 45 | +* timer resolution. Increase the array size if the kernels are too fast. You are |
| 46 | +* unlikely to break this rule if you follow rule 1. |
| 47 | +* |
| 48 | +* If you see a single BabelStream result, expect this to be the largest bandwidth |
| 49 | +* reported by the benchmark unless stated otherwise. |
| 50 | +*———————————————————————————————————------------------------------------------- |
0 commit comments