| DynamicDispatch Offload - not offloaded | |
| +-------------------+-------+------------------------+------------------------+ | |
| | Op Type | Count | Inputs | Outputs | | |
| +===================+=======+========================+========================+ | |
| | CastAvx | 1 | [1,64,64,4] - FLOAT | [1,64,64,4] - BFLOAT16 | | |
| | CastAvx | 1 | [1,512,512,3] - | [1,512,512,3] - FLOAT | | |
| | | | BFLOAT16 | | | |
| | Transpose | 1 | [1,4,64,64] - FLOAT | [1,64,64,4] - FLOAT | | |
| | Transpose | 1 | [1,512,512,3] - FLOAT | [1,3,512,512] - FLOAT | | |
| +-------------------+-------+------------------------+------------------------+ | |
| | Not offloaded sum | 4 | | | | |
| +-------------------+-------+------------------------+------------------------+ | |
| DynamicDispatch Offload - offloaded | |
| +--------------------+-------------+--------------------+---------------------+ | |
| | Op Type | Count | Inputs | Outputs | | |
| +====================+=============+====================+=====================+ | |
| | DynamicDispatch | 1 | [1,64,64,4] - | [1,512,512,3] - | | |
| | | | BFLOAT16 | BFLOAT16 | | |
| +--------------------+-------------+--------------------+---------------------+ | |
| | Offloaded sum | 1 | | | | |
| +--------------------+-------------+--------------------+---------------------+ | |
| | Offloaded Op Types | SDConv | | | | |
| | | SDAdd | | | | |
| | | SDResize | | | | |
| | | SDGemm | | | | |
| | | SDMHA_VAE | | | | |
| | | SDGroupNorm | | | | |
| +--------------------+-------------+--------------------+---------------------+ | |
| | Offloaded sum (dd | 89 | | | | |
| | fusion) | | | | | |
| | Offload Ratio (dd | 95.7% | | | | |
| | fusion) | | | | | |
| +--------------------+-------------+--------------------+---------------------+ | |