• State Accepted Answer
  • +2 person also asked this people also asked this
  • Replies 3 replies
  • Subscribers 144 subscribers
  • Views 9612 views
  • Users 0 members are here
2025 survey

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.

Options
Related
How was your experience today?

How to implement saturation instructions in ARMv8 architecture?

Natesh Raina

Is there a way to write an equivalent of 32-bit saturation instruction such  as QADD or QSUB without using neon saturation instructions for Armv8 architecture?  

  • 0 Arm Employee Badge
    It should be do-able in C. QADD saturates to the signed integer range -2^31 to (2^31) - 1.

    You can do a regular add and then check that the result. If the result is less than -2^31 or more than (2^31) - 1 you know saturation has occurred (and you can force the result to that value.)

    Stackoverflow has a few posts on the various flavours of singed/unsigned saturating maths you can do, usually with portable C examples.
  • 0 in reply to Peter Rielly
    Thank you Peter for replying. I know it is pretty straightforward in C but I want to implement it in assembly and that too in ARMv8 architecture. The thing is that in ARMv7 we have a single instruction for that i.e. QADD. But we don't have such instruction in v8 so we need to use neon instruction such as SQADD. Do we have a single instruction to perform such operation in v8?
  • +1 Arm Employee Badge in reply to Natesh Raina
    Ah, sorry, I missed the intent of your question.

    The only way to perform saturation in a single instruction is using the SIMD&FP instructions. They do have scalar variants (effectively ignoring all but a single lane - including for flag setting,) but they still operate on the FP register bank and set flags in FPSR rather than PSTATE.