On Mar 30, 2011, at 11:37 AM, Eli Stevens (Gmail) wrote:
Numpy 1.6.0 adds support for a half-float (16-bit) data type, but cannot currently export a buffer interface to the data, since the closest type that PEP 3118 supports is an unsigned short ('H'). This makes working with the data from outside numpy (for example, from Cython) difficult, since even if numpy were to expose a buffer interface to the data, it's unclear that the data needs special treatment to interpret correctly (numpy does this with bit shifting functions to convert it to a float32, but it has access to the array dtype which isn't available through the buffer interface, per my understanding).
What would be required to get a float16 data type added to PEP 3118 (either implicitly via inclusion of the struct module, or explicitly in the PEP itself)?
+1 I would support adding float16 to the struct module. It's a well defined format so we might as well provide an accessor. Just open a feature request for it. Any issues surrounding its use (i.e. double-rounding) are no different that the usual float/double conversion issues. Raymond