<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style>
</head>
<body dir="ltr">
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
>I would tend to disagree that the first two bullets are from the orginal paper, unless there is a different </div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
>revision of the Llosa et al. paper: In my copy, the paper’s first condition is ‘if (Pred_L(O) ∩ S) ≠ ∅’, </div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
>which is not equivalent to the statement ‘if (Pred_L(O) ⊆ S) in the implementation
<div><br>
</div>
<div>An aside:</div>
<div>There are indeed two SMS papers. The first is the conference version:</div>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
J. Llosa, A. Gonzalez, E. Ayguade and M. Valero, "Swing module scheduling: a lifetime-sensitive approach,"</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
Proceedings of the 1996 Conference on Parallel Architectures and Compilation Technique, 1996, pp. 80-86, </div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
doi: 10.1109/PACT.1996.554030.<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
The second is the journal version:</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
J. Llosa, E. Ayguade, A. Gonzalez, M. Valero and J. Eckhardt, "Lifetime-sensitive modulo scheduling in a </div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
production environment," in IEEE Transactions on Computers, vol. 50, no. 3, pp. 234-249, March 2001, </div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
doi: 10.1109/12.910814.<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
The journal version likely should be the followed for the ordering algorithm ('if (Pred_L(O) ∩ S) ≠ ∅'). </div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
Additionally (although <span style="color: rgb(0, 0, 0); font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt;">I don't remember the details 20+ years later), there were other changes to avoid some</span></div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
of the arbitrary choices made in the node ordering to tune or otherwise avoid heuristic noise.</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
</re-lurk></div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div id="appendonsend"></div>
<hr style="display:inline-block;width:98%" tabindex="-1">
<div id="divRplyFwdMsg" dir="ltr"><font face="Calibri, sans-serif" style="font-size:11pt" color="#000000"><b>From:</b> llvm-dev <llvm-dev-bounces@lists.llvm.org> on behalf of Nagurne, James via llvm-dev <llvm-dev@lists.llvm.org><br>
<b>Sent:</b> Friday, September 10, 2021 3:04 PM<br>
<b>To:</b> Cahoon, Brendon <Brendon.Cahoon@amd.com><br>
<b>Cc:</b> llvm-dev@lists.llvm.org <llvm-dev@lists.llvm.org>; Jayaraj, Ajay <ajayj@ti.com><br>
<b>Subject:</b> Re: [llvm-dev] Looking for some history on MachinePipeliner</font>
<div> </div>
</div>
<style>
<!--
@font-face
{font-family:Wingdings}
@font-face
{font-family:"Cambria Math"}
@font-face
{font-family:Calibri}
p.x_MsoNormal, li.x_MsoNormal, div.x_MsoNormal
{margin:0in;
margin-bottom:.0001pt;
font-size:11.0pt;
font-family:"Calibri",sans-serif}
a:link, span.x_MsoHyperlink
{color:blue;
text-decoration:underline}
a:visited, span.x_MsoHyperlinkFollowed
{color:purple;
text-decoration:underline}
p.x_MsoListParagraph, li.x_MsoListParagraph, div.x_MsoListParagraph
{margin-top:0in;
margin-right:0in;
margin-bottom:0in;
margin-left:.5in;
margin-bottom:.0001pt;
font-size:11.0pt;
font-family:"Calibri",sans-serif}
p.x_msonormal0, li.x_msonormal0, div.x_msonormal0
{margin-right:0in;
margin-left:0in;
font-size:11.0pt;
font-family:"Calibri",sans-serif}
p.x_msipheaderc10f11a2, li.x_msipheaderc10f11a2, div.x_msipheaderc10f11a2
{margin-right:0in;
margin-left:0in;
font-size:11.0pt;
font-family:"Calibri",sans-serif}
span.x_EmailStyle20
{font-family:"Calibri",sans-serif;
color:windowtext}
span.x_EmailStyle21
{font-family:"Calibri",sans-serif;
color:windowtext}
.x_MsoChpDefault
{font-size:10.0pt}
@page WordSection1
{margin:1.0in 1.0in 1.0in 1.0in}
div.x_WordSection1
{}
ol
{margin-bottom:0in}
ul
{margin-bottom:0in}
-->
</style>
<div lang="EN-US" link="blue" vlink="purple">
<table bgcolor="#FFEB9C" border="1">
<tbody>
<tr>
<td><font face="verdana" color="black" size="1"><b>External email: Use caution opening links or attachments</b>
</font></td>
</tr>
</tbody>
</table>
<br>
<div>
<div class="x_WordSection1">
<p class="x_MsoNormal">Hi Brendon, glad to see the original committer is still around
<span style="font-family:"Segoe UI Emoji",sans-serif">😊</span></p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">“Three of the cases are from Figure 6 in the original paper by Llosa and others.</p>
<ul type="disc" style="margin-top:0in">
<li class="x_MsoListParagraph" style="margin-left:0in">If pred_L is a subset of the NodeSet, BottomUp and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If succ_L is a subset of the NodeSet, TopDown and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">Otherwise, BottomUp and add node with the highest ASAP to work list</li></ul>
<p class="x_MsoNormal">“</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">I would tend to disagree that the first two bullets are from the orginal paper, unless there is a different revision of the Llosa et al. paper:</p>
<p class="x_MsoNormal" style="text-indent:.5in">In my copy, the paper’s first condition is ‘if (Pred_L(O) ∩ S) ≠
<span style="font-family:"Cambria Math",serif">∅</span>’, which is not equivalent to the statement ‘if (Pred_L(O)
<span style="font-family:"Cambria Math",serif">⊆</span> S) in the implementation</p>
<p class="x_MsoNormal">If your intent was to say that the cases are modified versions of those base cases, then I understand.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">“The SMS algorithm tries to limit the cases when a node has both successors and predecessors scheduled already.”</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">This is the problem my particular loop runs into.</p>
<p class="x_MsoNormal">With NodeSets:</p>
<p class="x_MsoNormal" style="margin-left:.5in">S0 = {1, 2, 3, 4}, maxASAP is node 4</p>
<p class="x_MsoNormal" style="margin-left:.5in">S1 = {5, 6, 7, 8}, maxASAP is node 8 (tied with 7, tie-broken because id 8 > 7)</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">With a skeletal set of DAG edges</p>
<p class="x_MsoNormal" style="margin-left:.5in">2 -> 6 -> 7</p>
<p class="x_MsoNormal" style="margin-left:.5in">6 -> 8</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">computeNodeOrder looks something like:</p>
<ol start="1" type="1" style="margin-top:0in">
<li class="x_MsoListParagraph" style="margin-left:0in">S0 is added to the node order bottom-up (default), say NodeOrder = {4, 3, 2, 1}</li><li class="x_MsoListParagraph" style="margin-left:0in">Starting on S1: pred_L(NodeOrder) is non-empty, but not a subset of S1, so execution eventually drops into bottom-up (default), with node 8 added to the work list as maxASAP</li><li class="x_MsoListParagraph" style="margin-left:0in">8 gets added to NodeOrder ({4, 3, 2, 1, 8}), adds its predecessors (6) and back-edges (<span style="font-family:"Cambria Math",serif">∅</span>) to the work list, and switches to top-down</li><li class="x_MsoListParagraph" style="margin-left:0in">6 gets added to NodeOrder ({4, 3, 2, 1, 8, 6}, and we’ve now ordered the predecessor and successor of 7 before 7, leading to the infeasibility of a schedule at any ii.</li></ol>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">“In the second case, with only 1 node-set, all the nodes in the basic block belong to a single node-set. With the fall back, the algorithm goes bottom-up with a single node, and it may be difficult to choose the best one to start with.
This heuristic chooses multiple nodes to start with so that bottom-up traversal occurs as a wave over a set. That is, rather than going up, and then down, etc. It helped to prime the traversal with multiple nodes rather than a single node. (Unfortunately,
there is no test case that fails if this heuristic is removed).</p>
<p class="x_MsoNormal">“</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">This case makes intuitive sense, as the ordering is either similar (or identical?) to what you’d get from a bottom-up, performance-focused list scheduler, and that’s just what a loop with a singular NodeSet would boil down to should there
be no overlapped iterations.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">“In your case, are the two new heuristics the reason why your example doesn’t schedule?”</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">The single NodeSet and succ_L intersect cases never evaluate true in my example; it’s the subset check that causes the invalid order, as detailed roughly above.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">In testing, I found no performance degradations in our embedded and dsp focused benchmarks using the paper-precise ordering algorithm. While the node ordering for my particular loop is legal with these changes, our compiler still fails
to pipeline. For this failure, I blame the contrived nature of the particular loop (volatile address to replicate a real-time data input).</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Perhaps I should try this loop out on Hexagon and see what happens. Probably another good place to look.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Regards,</p>
<p class="x_MsoNormal">JB</p>
<p class="x_MsoNormal"> </p>
<div>
<div style="border:none; border-top:solid #E1E1E1 1.0pt; padding:3.0pt 0in 0in 0in">
<p class="x_MsoNormal"><b>From:</b> Cahoon, Brendon <Brendon.Cahoon@amd.com> <br>
<b>Sent:</b> Thursday, September 9, 2021 4:15 PM<br>
<b>To:</b> Nagurne, James <j-nagurne@ti.com><br>
<b>Cc:</b> Jayaraj, Ajay <ajayj@ti.com>; llvm-dev@lists.llvm.org<br>
<b>Subject:</b> [EXTERNAL] RE: Looking for some history on MachinePipeliner</p>
</div>
</div>
<p class="x_MsoNormal"> </p>
<p class="x_msipheaderc10f11a2" style="margin:0in; margin-bottom:.0001pt"><span style="font-size:10.0pt; font-family:"Arial",sans-serif; color:green">[Public]</span></p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">It’s been a while since I’ve looked at the pipeliner… The lack of discussion and history is because the pipeliner was developed downsteam for a long period of time before it was upstreamed.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Three of the cases are from Figure 6 in the original paper by Llosa and others.</p>
<ul type="disc" style="margin-top:0in">
<li class="x_MsoListParagraph" style="margin-left:0in">If pred_L is a subset of the NodeSet, BottomUp and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If succ_L is s subset of the NodeSet, TopDown and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">Otherwise, BottomUp and add node with the highest ASAP to work list</li></ul>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">I’m pretty sure, at various points, we tried intersection for pred_L and succ_L, but I don’t think that gave us very good results.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">The two additional cases,</p>
<ul type="disc" style="margin-top:0in">
<li class="x_MsoListParagraph" style="margin-left:0in">If intersect(succ_L, NodeSet) is non-empty, TopDown and add intersect to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If there is only 1 NodeSet, BottomUp and add nodes at the bottom of the schedule (Succs.size() == 0) to work list</li></ul>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">These were added to improve node ordering and the generated schedule, which were identified when analyzing benchmarks. For the first case, if that heuristic is removed, then one of the tests, swp-conv3x3-nested.ll fails. That was an
important kernel to pipeline, and it doesn’t pipeline without that heuristic.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">The SMS algorithm tries to limit the cases when a node has both successors and predecessors scheduled already. This is a concern for node-sets with recurrences (because there is one node that will have successors and predecessors scheduled
already). In loops that contain many node-sets that have recurrences, finding a valid ordering may be difficult. The challenge with the original three cases is that the fallback is to do a bottom-up schedule, and it starts with a single node only. We noticed
that it’s important to choose the correct direction and it’s important to identify the initial set of nodes used (to start going top-down or bottom-up).</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">For the first case, with the intersection, we don’t want to go bottom-up with a single node. Instead, starting top-down is better because the current node-set contains successors of nodes scheduled already. Furthermore, we want to start
with the set of nodes. If we start with a single node going top-down, then the algorithm ends up switching eventually to bottom-up and creating a case where a node has a successor and predecessor scheduled already.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">In the second case, with only 1 node-set, all the nodes in the basic block belong to a single node-set. With the fall back, the algorithm goes bottom-up with a single node, and it may be difficult to choose the best one to start with.
This heuristic chooses multiple nodes to start with so that bottom-up traversal occurs as a wave over a set. That is, rather than going up, and then down, etc. It helped to prime the traversal with multiple nodes rather than a single node. (Unfortunately,
there is no test case that fails if this heuristic is removed).</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Let me know if you have any questions. In your case, are the two new heuristics the reason why your example doesn’t schedule?</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Thanks,</p>
<p class="x_MsoNormal">Brendon</p>
<p class="x_MsoNormal"> </p>
<div>
<div style="border:none; border-top:solid #E1E1E1 1.0pt; padding:3.0pt 0in 0in 0in">
<p class="x_MsoNormal"><b>From:</b> llvm-dev <<a href="mailto:llvm-dev-bounces@lists.llvm.org">llvm-dev-bounces@lists.llvm.org</a>>
<b>On Behalf Of </b>Nagurne, James via llvm-dev<br>
<b>Sent:</b> Thursday, September 9, 2021 1:12 PM<br>
<b>To:</b> 'llvm-dev@lists.llvm.org' <<a href="mailto:llvm-dev@lists.llvm.org">llvm-dev@lists.llvm.org</a>><br>
<b>Cc:</b> Jayaraj, Ajay <<a href="mailto:ajayj@ti.com">ajayj@ti.com</a>><br>
<b>Subject:</b> [llvm-dev] Looking for some history on MachinePipeliner</p>
</div>
</div>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">[CAUTION: External Email] </p>
<div>
<p class="x_MsoNormal">I’ve been tinkering with the MachinePipeliner for my team’s downstream target, and have run into a particular issue with node ordering in the SMS algorithm. In particular, I’ve noticed that in some cases, I’m receiving an ‘Invalid node
order’ warning in the debug. I’ve tracked this down to how ‘computeNodeOrder’ is implemented.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">computeNodeOrder’s algorithm seems to be relatively untouched from the initial commit:</p>
<p class="x_MsoNormal" style="text-indent:.5in"><a href="https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Freviews.llvm.org%2FD16829&data=04%7C01%7Cjeckhardt%40nvidia.com%7Ceac9bbd49d404c9dc88c08d97496344f%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637669010749367397%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=g468V9nqJX0%2B38CBuHnK9VlkIP2B0FaYMTJhEjjVEXQ%3D&reserved=0" originalsrc="https://reviews.llvm.org/D16829" shash="DrbYfhn3Tf3u79WtSuZzu3Ntx7rApiD1694l2h6WtLFzlXfKZwYYxPhT++YRulk7w4KLNx15unzvsSlAeCjaaaWcO2NWmW3niuP6BFlLVhyIHEPcETaF6xCj6U+RyxmRTsY+rJldGPHdgfcqj6uMbAvCQe0H/2LcmEKBJiIV8tM=">https://reviews.llvm.org/D16829</a></p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">However, the algorithm is very different from those found in the 3 papers upon which the implementation is based:</p>
<p class="x_MsoNormal"> <a href="https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fllvm%2Fllvm-project%2Fblob%2Ff40bba48a593d4297a6d0b4d1534082858b1d0ac%2Fllvm%2Finclude%2Fllvm%2FCodeGen%2FMachinePipeliner.h%23L18&data=04%7C01%7Cjeckhardt%40nvidia.com%7Ceac9bbd49d404c9dc88c08d97496344f%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637669010749377391%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=SUNS6l%2BhAoPa1jvcXb2HUhtdIWng7KcjNKjJqj3JW9I%3D&reserved=0" originalsrc="https://github.com/llvm/llvm-project/blob/f40bba48a593d4297a6d0b4d1534082858b1d0ac/llvm/include/llvm/CodeGen/MachinePipeliner.h#L18" shash="oodf3EaGzU/fa+O4OjUt20gHuxDfnoKWlw9NTi4yA5V1uGhFTEA//nL+czkqWgn9Ye4NeYPVjaHDmH8VIInAFKEbUwt72ykqVx8FJnQfgs3NJyvWUXXFCepJR1pggjk2hZeQRPzjGSTew9AwtwFReZX2VLXVdLGwnBelWc9SmjY=">
https://github.com/llvm/llvm-project/blob/f40bba48a593d4297a6d0b4d1534082858b1d0ac/llvm/include/llvm/CodeGen/MachinePipeliner.h#L18</a></p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">In the implementation:</p>
<p class="x_MsoNormal"> <a href="https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fllvm%2Fllvm-project%2Fblob%2Ff40bba48a593d4297a6d0b4d1534082858b1d0ac%2Fllvm%2Flib%2FCodeGen%2FMachinePipeliner.cpp%23L1877&data=04%7C01%7Cjeckhardt%40nvidia.com%7Ceac9bbd49d404c9dc88c08d97496344f%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637669010749387382%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=lXUO4uxHyEqDKwTOS2azOOAPSggMP22xXqeINCneahw%3D&reserved=0" originalsrc="https://github.com/llvm/llvm-project/blob/f40bba48a593d4297a6d0b4d1534082858b1d0ac/llvm/lib/CodeGen/MachinePipeliner.cpp#L1877" shash="PCPj7VSAOkTmJ4nhVxp6fV2QtIWDO6n7veohNDhuGmTE56ASa9ARXE6mmNBKL872jQYyibB6eZ6ZY+u3fRKFeANeVX4rqDY3Dd+2OI05rcnHRDgWK6t7AlLugaTMeIgmGoA6cVHkHNbgVBydgpUfKzGoDx8/xkW+wNhAVYzKYy4=">
https://github.com/llvm/llvm-project/blob/f40bba48a593d4297a6d0b4d1534082858b1d0ac/llvm/lib/CodeGen/MachinePipeliner.cpp#L1877</a></p>
<p class="x_MsoNormal">The order seems to be:</p>
<ul type="disc" style="margin-top:0in">
<li class="x_MsoListParagraph" style="margin-left:0in">If pred_L is a subset of the NodeSet, BottomUp and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If succ_L is s subset of the NodeSet, TopDown and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If intersect(succ_L, NodeSet) is non-empty, TopDown and add intersect to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If there is only 1 NodeSet, BottomUp and add nodes at the bottom of the schedule (Succs.size() == 0) to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">Otherwise, BottomUp and add node with the highest ASAP to work list</li></ul>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">In the papers (Using <a href="https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fllvm.org%2Fpubs%2F2005-06-17-LattnerMSThesis.html&data=04%7C01%7Cjeckhardt%40nvidia.com%7Ceac9bbd49d404c9dc88c08d97496344f%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C637669010749387382%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=79q9a3srnIPhCG8EmjSUVX6TPKFQsSkLgekH8n8fkhg%3D&reserved=0" originalsrc="https://llvm.org/pubs/2005-06-17-LattnerMSThesis.html" shash="D7TyKDByx4hbvpbOtHEm6VNnD7d6Jr9FtEC++/aSkyCq77e1VgozBhrV+hRWsFP+uh1c09p1JjCMNLUBOLPzQ48BEQePE61W1rZToHORxD+jGUc55xQaPHjiMPAu50wNzNJzyM0D6anj7YjMSH0/eVHDh+0zZmCt1G/2d/8k8Tc=">
Tanya Lattner’s master’s thesis</a> as the example), the order is:</p>
<ul type="disc" style="margin-top:0in">
<li class="x_MsoListParagraph" style="margin-left:0in">If intersect(pred_L, NodeSet) is non-empty, BottomUp and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">If intersect(succ_L, NodeSet) is non-empty, TopDown and add subset to work list</li><li class="x_MsoListParagraph" style="margin-left:0in">Otherwise, BottomUp and add node with the highest ASAP to work list</li></ul>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Interestingly, when I implement the ordering from the paper, the node order for my loop is valid.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">This difference in implementation is not mentioned or commented on in the review, so I’m wondering if there are any documents or emails I may have missed in my search that would shed some light. Certainly, this ordering is just a heuristic,
so I can understand that the current incarnation just resulted in better performance for the targets utilizing it, but I’m curious if there was something more to the decision.</p>
<p class="x_MsoNormal"> </p>
<p class="x_MsoNormal">Regards,</p>
<p class="x_MsoNormal">J.B. Nagurne</p>
<p class="x_MsoNormal">Code Generation</p>
<p class="x_MsoNormal">Texas Instruments</p>
</div>
</div>
</div>
</div>
</body>
</html>