<div dir="auto">Yes, I came across that issue with hashed collections of consecutive integers some years ago, and that’s why we changed SmallInteger>>hash in Cuis. However, I think it might be better to change Set>>scanFor: instead (and in the appropriate subclasses like Dictionary, etc), put the call to hashMultiply there (or “shuffle” the hash in some other way), and make integers be their own hash again.</div><div dir="auto"><br></div><div><div class="gmail_quote gmail_quote_container"><div dir="ltr" class="gmail_attr">On Wed, Jan 15, 2025 at 06:14 Nicolas Cellier <<a href="mailto:nicolas.cellier.aka.nice@gmail.com">nicolas.cellier.aka.nice@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;padding-left:1ex;border-left-color:rgb(204,204,204)"><div dir="ltr"><div><div>This time it seems a bit different though, the bench is not dominated by GC, but by hash collisions.<br><div><div><br>Cuis Set uses the old fashioned capacity:</div><div>sizeFor: nElements<br> "Large enough size to hold nElements with some slop (see fullCheck)"<br> nElements <= 0 ifTrue: [^ 1].<br> ^ nElements+1*4//3 <br></div></div><div>Squeak Set uses more advanced #<b>goodPrimeAtLeast:</b> for determining the capacity<b>.<br></b>So we would expect Squeak to be better... But no, it isn't.<b><br></b></div><div><b><br></b></div><div>First observe that we have:<br></div><div><br></div><div> (hashes sorted = (1 to: hashes size) asArray).</div><div><br></div><div>Hence, we try to put consecutives SmallInteger in a Set.<br></div><div><br></div><div>If I bench Eliot's test, but with gradually growing size (1<<18-1), (1<<19-1),(1<<20-1), (1<<21-1)<br></div><div>I get:<br> '6.87 per second. 146 milliseconds per run. 13.48116 % GC time.'<br> '3.23 per second. 310 milliseconds per run. 14.60311 % GC time.'<br> '1.56 per second. 641 milliseconds per run. 15.1462 % GC time.'</div><div> '0.087 per second. 11.5 seconds per run. 0.04346 % GC time.'</div><div><br></div><div>So the cost grows linearly with size up to 2^20,<br>but 2^21 costs us a factor x20 instead of expected x2,<br>that's 10 times more than expected,<br></div><div>and GC cost drops in percentage...<br></div><div><div><br></div><div>I won't even try (1<<22-1) on my slow machine...</div></div><div>What happens ?<br><br></div><div>Simple : the answer is in SmallInteger>>hash ^self<br></div><div>The identity hashes are varying from 1 to 1<<22-1.</div><div>Up to 1<<20-1, we do have not so many consecutives hashes statistically.<br></div><div>But we have way more consecutive with 1<<21-1 elements (amonf 1<<22-1 possible hashes).<br></div><div>And since candidate index in scanFor: is simply object hash\\array size+1,</div><div>once we have a collision (and we will have until the array size grows above 1<<22-2), <br>since we have already filled consecutive slots with consecutives hashes,</div><div>we are getting long collision chains.<br><br></div><div>So why is Cuis better behaved?<br></div><div>Because Integer>>hash will return self hashMultiply...<br></div><div>Note that above 1<<54, consecutive Integer hashes will show quite some collisions because first convert asFloat, and many consecutive LargeInteger convert to the same Float.<br></div><div><br></div><div>Just redefine SmallInteger hash>>^self hashMultiply, and you'll get more decent benchmark.<br></div><div>Except, that you can't do it interactively, because you have to rehash all hashed collection before Morphic takes the hand back.</div></div></div></div><div dir="ltr"><div><div><div><br><br></div><div>Nicolas<br></div></div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">Le mar. 14 janv. 2025 à 21:55, Nicolas Cellier <<a href="mailto:nicolas.cellier.aka.nice@gmail.com" target="_blank">nicolas.cellier.aka.nice@gmail.com</a>> a écrit :<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;padding-left:1ex;border-left-color:rgb(204,204,204)"><div dir="ltr">That remind me this thread:<br><br><a href="https://lists.cuis.st/mailman/archives/cuis-dev/2024-March/008555.html" target="_blank">https://lists.cuis.st/mailman/archives/cuis-dev/2024-March/008555.html</a></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">Le mar. 14 janv. 2025 à 20:38, David T Lewis via Squeak-dev <<a href="mailto:squeak-dev@lists.squeakfoundation.org" target="_blank">squeak-dev@lists.squeakfoundation.org</a>> a écrit :<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:solid;padding-left:1ex;border-left-color:rgb(204,204,204)"><p></p>
<p dir="auto">Reported by Eliot on squeak-dev:<br>
<a href="https://lists.squeakfoundation.org/archives/list/squeak-dev@lists.squeakfoundation.org/thread/ABGLWILW6YJ2NPAHICWSLCOUIAKXKLFL/" rel="nofollow" target="_blank">https://lists.squeakfoundation.org/archives/list/squeak-dev@lists.squeakfoundation.org/thread/ABGLWILW6YJ2NPAHICWSLCOUIAKXKLFL/</a></p>
<p dir="auto">Copied from the original report:</p>
<p dir="auto">Hi All,</p>
<p dir="auto"> recently the identityHash implementation in the VM has been upgraded. </p>
<p dir="auto">Open Smalltalk Cog[Spur] VM [CoInterpreterPrimitives VMMaker.oscog-eem.3500] 5.20250103.0325<br>
platform sources revision VM: 202501030325 <a href="mailto:eliot@Machado.local" target="_blank">eliot@Machado.local</a>:oscogvm Date: Thu Jan 2 19:25:19 2025 CommitHash: 5e05e38</p>
<p dir="auto">It now uses an xorshift register PNG that cycles through all possible 2^22 - 1 valid identity hashes in 2^22 - 1 iterations. One way to test this is to create a set, and loop creating 2^22-1 objects, adding their hashes to the set, and answering the size of the set.</p>
<p dir="auto">So let's do this and see how long it takes at the same time. e.g. this is on my 2021 M1 Max MBP</p>
<p dir="auto">| size |<br>
{ [| hashes |<br>
hashes := Set new: (2 raisedTo: 22).<br>
1 to: (2 raisedTo: 22) - 1 do:<br>
[:ign| hashes add: Object new identityHash].<br>
size := hashes size] timeToRun.<br>
size.<br>
(2 raisedTo: 22) - 1 }. #(450 4194303 4194303)</p>
<p dir="auto">also #(483 4194303 4194303)</p>
<p dir="auto">So avoiding grows this takes less than half a second.</p>
<p dir="auto">However, if we just use Set new and rely on it growing we get approximately a 1,500 to 1,800 fold slowdown:</p>
<p dir="auto">| size |<br>
{ [| hashes |<br>
hashes := Set new.<br>
1 to: (2 raisedTo: 22) - 1 do:<br>
[:ign| hashes add: Object new identityHash].<br>
size := hashes size] timeToRun.<br>
size.<br>
(2 raisedTo: 22) - 1 }. #(768992 4194303 4194303)</p>
<p dir="auto">also #(800874 4194303 4194303).</p>
<p dir="auto">768992 / 483.0 = 1592<br>
800874 / 450.0 1780</p>
<p dir="auto">Now Cuis has moved away from allowing stretchy collections to have their capacity initialized with new:. One has to use newWithRoomForMoreThan:. So</p>
<p dir="auto">| size |<br>
{ [| hashes |<br>
hashes := Set newWithRoomForMoreThan: (2 raisedTo: 22).<br>
1 to: (2 raisedTo: 22) - 1 do:<br>
[:ign| hashes add: Object new identityHash].<br>
size := hashes size] timeToRun.<br>
size.<br>
(2 raisedTo: 22) - 1 }. #(506 4194303 4194303)</p>
<p dir="auto">BUT!!!! If we just use new and allow Cuis to grow the set we get e.g.</p>
<p dir="auto">| size |<br>
{ [| hashes |<br>
hashes := Set new.<br>
1 to: (2 raisedTo: 22) - 1 do:<br>
[:ign| hashes add: Object new identityHash].<br>
size := hashes size] timeToRun.<br>
size.<br>
(2 raisedTo: 22) - 1 }. #(725 4194303 4194303) .</p>
<p dir="auto">This is at least 1,000 times faster than Squeak. Something is seriously wrong with the Squeak code.</p>
<p dir="auto">768992 / 725.0 = 1061</p>
<p dir="auto"><strong>,,,^..^,,,</strong><br>
Happy New Year! Eliot</p>
<p style="font-size:small;color:rgb(102,102,102)">—<br>Reply to this email directly, <a href="https://github.com/squeak-smalltalk/squeak-object-memory/issues/129" target="_blank">view it on GitHub</a>, or <a href="https://github.com/notifications/unsubscribe-auth/BFYAK6LM7BP4BHBELJMTK4L2KVRQPAVCNFSM6AAAAABVFUFTZWVHI2DSMVQWIX3LMV43ASLTON2WKOZSG44DQMJTG42DCMA" target="_blank">unsubscribe</a>.<br>You are receiving this because you are subscribed to this thread.<img height="1" width="1" alt=""><span style="font-size:0px;display:none;overflow:hidden;opacity:0;width:0px;height:0px;max-width:0px;max-height:0px;color:transparent">Message ID: <span><squeak-smalltalk/squeak-object-memory/issues/129</span><span>@</span><span>github</span><span>.</span><span>com></span></span></p>
Squeak-dev mailing list -- <a href="mailto:squeak-dev@lists.squeakfoundation.org" target="_blank">squeak-dev@lists.squeakfoundation.org</a><br>
To unsubscribe send an email to <a href="mailto:squeak-dev-leave@lists.squeakfoundation.org" target="_blank">squeak-dev-leave@lists.squeakfoundation.org</a></blockquote></div>
</blockquote></div>
Squeak-dev mailing list -- <a href="mailto:squeak-dev@lists.squeakfoundation.org" target="_blank">squeak-dev@lists.squeakfoundation.org</a><br>
To unsubscribe send an email to <a href="mailto:squeak-dev-leave@lists.squeakfoundation.org" target="_blank">squeak-dev-leave@lists.squeakfoundation.org</a></blockquote></div></div>