[ARROW-7860] [C++] Support cast to/from halffloat

• Type: Improvement
• Status: Closed
• Priority: Major
• Resolution: Duplicate
• Fix Version/s: 0.17.0
• Component/s: C++
• External issue URL: https://github.com/apache/arrow/issues/24088
• Assignee: Unassigned
• Reporter: Neal Richardson
• Created: 14/Feb/20 17:40
• Updated: 11/Jan/23 07:56
• Resolved: 14/Feb/20 17:50

Description

In trying to do ARROW-7753 I realized I couldn't make a halffloat. I tried creating a float64 (as R does naturally) and casting to float16, but it's not implemented. Looking at compute/kernels/cast.cc, and the associated source in compute/kernels/generated/codegen.py, FLOATING_TYPES = ['Float', 'Double']. Maybe halffloat just needs to be added there?

Aside: searching through the code, it seems that this limitation of float types to float32 and float64 is the norm.