Ticket #3 (new defect)
Unhelpful error when using Double on pre-1.3 CUDA devices
|Reported by:||chak||Owned by:|
Only CUDA devices with compute capability 1.3 and up contain hardware support for Double. Currently, Accelerate will nevertheless generate code for Doubles for these devices, which leads to a failure further down the pipeline. This is not very user-friendly.
This raises the general questions of how to handle device capabilities elegantly.