We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.
Hi all,
I want to use half-precision arithmetic on Raspberry pi 3. I installed clang compiler to use _Float16 data type (which is for 16-bit arithmetic purpose).
I was wondering is there any library for half-precision mathematic functions? (like <quadmath.h> for quadruple-precision numbers)
DuckDuckGo is your friend ;-) : http://half.sourceforge.net/
Thank you.
I already saw this link but I think this library is for software support of half-precision number? isn't it?
Right. Raspi 3 (Cortex-A53) has no HW support for half-precision floating point. FP16 was added in Armv8.2 and can be found in Cortex-A76 AFAIK.
Ok. so is there any half-precision math library that should be used for Cortex-A76?
What do you expect? FP16 version of sinf/cosf? Doesn't clang provide it?
I want to use all mathematic functions for FP16 numbers? the functions which are included in <math.h> are in single and double formats so clang doesn't provide them for FP16.
according to my research when ARMv8.2-FP16 is implemented, ARMV8 (Cortex-A53) supports half-precision data type for data processing in hardware. Am I right?
According the TRM CA53 supports only conversion:
[23:20] SIMDHP Indicates whether the Advanced SIMD and Floating-point extension supports half-precision floating-point conversion operations:0x1 Implemented.