AOCL-DLP handles some implicit type conversions: Quantized to float: Automatic dequantization in mixed-precision operations BFloat16 to float32: Automatic promotion for accumulation Integer widening: Automatic promotion to prevent overflow