• Locked Locked
  • Replies 4 replies
  • Subscribers 119 subscribers
  • Views 3290 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?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Neon suitable for max() on char[]?

John Papandriopoulos
Parents
  • Note: This was originally posted on 9th September 2011 at http://forums.arm.com

    It turns out you can simply use vcgt with vbit to do the "branching part":


    // Compare the 16-byte maximum with the global maximum
    vcgt.u8 d3, d0, d2 // d3[:] = (d0[:] > d2[:]) ?0xff :0x00
    // Update the global maximum if the 16-byte maximum is bigger
    vbit d2, d0, d3  // d2[:] = (d3[:] == 0xff) ?d0[:] :d2[:]

    where:
    • d0 holds the maximum value in all lanes (from the vpmax earlier),
    • d3 holds a "greater than" flag set to all-one when d0 > d2,
    • d2 holds the "current maximum" value (in all lanes).
    Turns out to be a whole lot faster than the reference C code, even with a small section of arm code following the 16-byte/iteration neon loop that determines the precise position of the maximum in the 16-byte read buffer.

    jpap
Reply
  • Note: This was originally posted on 9th September 2011 at http://forums.arm.com

    It turns out you can simply use vcgt with vbit to do the "branching part":


    // Compare the 16-byte maximum with the global maximum
    vcgt.u8 d3, d0, d2 // d3[:] = (d0[:] > d2[:]) ?0xff :0x00
    // Update the global maximum if the 16-byte maximum is bigger
    vbit d2, d0, d3  // d2[:] = (d3[:] == 0xff) ?d0[:] :d2[:]

    where:
    • d0 holds the maximum value in all lanes (from the vpmax earlier),
    • d3 holds a "greater than" flag set to all-one when d0 > d2,
    • d2 holds the "current maximum" value (in all lanes).
    Turns out to be a whole lot faster than the reference C code, even with a small section of arm code following the 16-byte/iteration neon loop that determines the precise position of the maximum in the 16-byte read buffer.

    jpap
Children
No data