<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-cite-prefix">On 4/7/19 12:16 AM, Ben Grasset wrote:<br>
    </div>
    <blockquote type="cite"
cite="mid:CAL4d7Fi0_Jo0Pf5GiU_R+J3mLLW8ppY6ArGkgbessr3_utMyrQ@mail.gmail.com">
      <meta http-equiv="content-type" content="text/html; charset=UTF-8">
      <div dir="ltr">
        <div dir="ltr">On Sat, Apr 6, 2019 at 5:27 PM Jeppe Johansen
          <<a href="mailto:jeppe@j-software.dk"
            moz-do-not-send="true">jeppe@j-software.dk</a>> wrote:<br>
        </div>
        <div class="gmail_quote">
          <blockquote class="gmail_quote" style="margin:0px 0px 0px
            0.8ex;border-left:1px solid
            rgb(204,204,204);padding-left:1ex">
            <div bgcolor="#FFFFFF">
              <p>That can't happen or won't benefit much, before the
                compiler supports super-natual alignments. So there's a
                deeper level of support needed.</p>
            </div>
          </blockquote>
          <div>I'm fairly confident the existing functionality could be
            improved upon in a useful way. Gareth has already done so in
            the past.</div>
        </div>
      </div>
    </blockquote>
    Perhaps but it's a nightmare in so many places to satisfy
    requirements for potential uses. Ideally the programmer would
    specify minimum alignment requirements per type<br>
    <blockquote type="cite"
cite="mid:CAL4d7Fi0_Jo0Pf5GiU_R+J3mLLW8ppY6ArGkgbessr3_utMyrQ@mail.gmail.com">
      <div dir="ltr">
        <div class="gmail_quote">
          <div> On Sat, Apr 6, 2019 at 5:27 PM Jeppe Johansen <<a
              href="mailto:jeppe@j-software.dk" moz-do-not-send="true">jeppe@j-software.dk</a>>
            wrote:</div>
          <div class="gmail_quote">
            <blockquote class="gmail_quote" style="margin:0px 0px 0px
              0.8ex;border-left:1px solid
              rgb(204,204,204);padding-left:1ex">
              <div bgcolor="#FFFFFF">
                <p>And personally I don't think that's the right long
                  term direction. It takes a long time to develop and
                  maintain this stuff and you never know what the market
                  will look like in 10 years.<br>
                  ARM has SVE, and RISC-V has the upcoming vector
                  extension which will move far away from the
                  traditional SIMD stuff.</p>
              </div>
            </blockquote>
            <div>I sincerely doubt I will be using a RISC-V based
              desktop PC in 10 years. Also nobody said -Sv could not
              also be extended to stuff such as ARM Neon at some point
              down the line if someone was up to it. One thing at a time
              though.</div>
          </div>
        </div>
      </div>
    </blockquote>
    <p>It's too hard to say, and the RISC-V foundation has made really
      bad decisions that will prevent them from getting a good start.</p>
    <p>But you will not find anyone who don't think vector extensions
      are better for software than SIMD. ARM SVE and RISC-V Vector
      extensions are true vector extensions, whereas the  SSE, AVX,
      NEON, Advanced SIMD, etc are packed SIMD extensions requiring a
      ton of hoops and jumps to use in real block vectorized code<br>
    </p>
    <blockquote type="cite"
cite="mid:CAL4d7Fi0_Jo0Pf5GiU_R+J3mLLW8ppY6ArGkgbessr3_utMyrQ@mail.gmail.com">
      <div dir="ltr">
        <div class="gmail_quote">
          <div class="gmail_quote">
            <div>
              <div dir="ltr">On Sat, Apr 6, 2019 at 5:27 PM Jeppe
                Johansen <<a href="mailto:jeppe@j-software.dk"
                  moz-do-not-send="true">jeppe@j-software.dk</a>>
                wrote:<br>
              </div>
              <div class="gmail_quote">
                <blockquote class="gmail_quote" style="margin:0px 0px
                  0px 0.8ex;border-left:1px solid
                  rgb(204,204,204);padding-left:1ex">
                  <div bgcolor="#FFFFFF">
                    <p>Compiler support for block vectorization has
                      rarely paid off really well given the amount of
                      work that needs to go into it. So maybe it's
                      better to wait for the next iteration :) </p>
                  </div>
                </blockquote>
                <div>It very objectively pays off in every compiler that
                  has it. I don't think "it's too hard" is a good
                  excuse. I made my previous comment because it seems
                  like Gareth if anyone specifically does *not* think it
                  is too hard.<br>
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    Do you have any quantifiable data to point at here?<br>
  </body>
</html>