Enable ZenDNN in WeGO TensorFlow 2
ZenDNN is disabled by default. Set export
TF_ENABLE_ZENDNN_OPTS=1 to enable it.
Environment Variables
You must export the following environment variables explicitly to enable
ZenDNN to work correctly in WeGO TensorFlow 2.
| Name | Description |
|---|---|
| OMP_DYNAMIC | Set it to FALSE explicitly when ZenDNN is enabled. |
| OMP_NUM_THREADS | Set it explicitly to achieve better performance. See tuning guidelines for more details. |
| ZENDNN_GEMM_ALGO | The default value is 3. You can set [0, 1, 2, 3, 4] to tune different GEMM ALGO paths. |
| ZENDNN_TENSOR_POOL_LIMIT | The default value is 32. See tuning guidelines for more details. |
| ZENDNN_TENSOR_BUF_MAXSIZE_ENABLE | Default is 0.
|
| TF_ENABLE_ZENDNN_OPTS | The default value is 0. Set it to 1 to enable ZenDNN. |
Tuning Guidelines
Set OMP_NUM_THREADS per the core number of
the user system. AMD recommends setting a small
number like 1 or 2.
In some cases, set ZENDNN_TENSOR_POOL_LIMIT
to a small number like 1, so some layers use default memory allocation instead of
the tensor pool once it hits the pool limit with ZENDNN_TENSOR_POOL_LIMIT.