<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"><head><meta http-equiv=Content-Type content="text/html; charset=utf-8"><meta name=Generator content="Microsoft Word 14 (filtered medium)"><style><!--
/* Font Definitions */
@font-face
{font-family:Wingdings;
panose-1:5 0 0 0 0 0 0 0 0 0;}
@font-face
{font-family:Wingdings;
panose-1:5 0 0 0 0 0 0 0 0 0;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
@font-face
{font-family:Tahoma;
panose-1:2 11 6 4 3 5 4 4 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0cm;
margin-bottom:.0001pt;
font-size:12.0pt;
font-family:"Times New Roman","serif";}
a:link, span.MsoHyperlink
{mso-style-priority:99;
color:blue;
text-decoration:underline;}
a:visited, span.MsoHyperlinkFollowed
{mso-style-priority:99;
color:purple;
text-decoration:underline;}
p.MsoAcetate, li.MsoAcetate, div.MsoAcetate
{mso-style-priority:99;
mso-style-link:"Balloon Text Char";
margin:0cm;
margin-bottom:.0001pt;
font-size:8.0pt;
font-family:"Tahoma","sans-serif";}
span.BalloonTextChar
{mso-style-name:"Balloon Text Char";
mso-style-priority:99;
mso-style-link:"Balloon Text";
font-family:"Tahoma","sans-serif";}
span.hoenzb
{mso-style-name:hoenzb;}
span.EmailStyle20
{mso-style-type:personal;
font-family:"Calibri","sans-serif";
color:#1F497D;}
span.EmailStyle21
{mso-style-type:personal;
font-family:"Calibri","sans-serif";
color:#1F497D;}
span.EmailStyle22
{mso-style-type:personal-reply;
font-family:"Calibri","sans-serif";
color:#1F497D;}
.MsoChpDefault
{mso-style-type:export-only;
font-size:10.0pt;}
@page WordSection1
{size:612.0pt 792.0pt;
margin:72.0pt 72.0pt 72.0pt 72.0pt;}
div.WordSection1
{page:WordSection1;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]--></head><body lang=EN-GB link=blue vlink=purple><div class=WordSection1><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>The dump of the graph can certainly be improved. I’ll look into it.<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>The problem you had with an ARM backend is due to the fact that the DEBUG_TYPE macro is supposed to be used cautiously (if at all) in a header, as it will probably collide when it is included in .cpp files: as it did with the AArch64 target: see the comment at the top of include/llvm/Support/Debug.h.<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>Cheers,<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>--<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>Arnaud<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><div style='border:none;border-left:solid blue 1.5pt;padding:0cm 0cm 0cm 4.0pt'><div><div style='border:none;border-top:solid #B5C4DF 1.0pt;padding:3.0pt 0cm 0cm 0cm'><p class=MsoNormal><b><span lang=EN-US style='font-size:10.0pt;font-family:"Tahoma","sans-serif"'>From:</span></b><span lang=EN-US style='font-size:10.0pt;font-family:"Tahoma","sans-serif"'> Jonas Paulsson [mailto:jonas.paulsson@ericsson.com] <br><b>Sent:</b> 29 January 2015 18:19<br><b>To:</b> Arnaud De Grandmaison; 'Lang Hames'<br><b>Cc:</b> llvmdev@cs.uiuc.edu<br><b>Subject:</b> RE: [LLVMdev] PBQP crash<o:p></o:p></span></p></div></div><p class=MsoNormal><o:p> </o:p></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Hi,<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Sorry for the delay, it has taken some extra time as more than one bug showed up </span><span lang=EN-US style='font-size:11.0pt;font-family:Wingdings;color:black'>J</span><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>I continued to look into this with your viewpoint that a node that is conservatively allocatable should never be spilled. The first thing I did was therefore to add some extra code with an assert for this.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>I believe I then found three bugs and fixed the two:<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Bug 1: Incorrect transpositions in handleAddEdge() / hanldeRemoveEdge(). For the heuristic of DeniedOpts, if N runs along the columns (Transpose == true), the interesting dimension is along the rows, and vice versa. In other words, the question is what the neighbour could maximally deny N, and the answer is found in WorstRow, not WorstCol (as it was). This makes a difference if sub/super registers overlap.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Bug 2: When it happens that there are no available physical registers for a node (e.g because of regmasks or calls), the number of rows / columns is just 1, for the spill option. This case must be detected in MatrixMetadata(), or WorstColCountForCurRow will get an undefined value, as std::max_element() is called with an empty vector.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Bug 3: Again a conservatively allocatable node was spilled, and the assert triggered. This is a description of what happened in my out-of-tree test case:<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>applyR2() called on node N, which is overlapping Y and Z. The edges (N,Y) and (N,Z) are all-zeroes. Y and Z are overlapping and already have an interference edge. Z is just on the limit of not being conservatively allocatable: NumOpts is 8 and DeniedOpts is also 8. It is contained in NotProvablyAllocatableNodes. G.setEdgeCosts() is called and then the call stack grows with Solver->handleSetEdgeCosts(), handleRemoveEdge() into handleDisconnectEdge(), where NMd.handleRemoveEdge() is called, which decreases the DeniedOpts by one. After this, it looks like a bug to me that Z is moved to ConservativelyAllocatableNodes, because eventually handleSetEdgeCosts() will complete, and the edge between Y and Z will have been added again in handleAddEdge(), and Z:DeniedOpts is again 8! <o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>I think this also shows up in a test case for arm. It was found by using the assert mentioned above, and running 'bin/llvm-stress -size 200 -seed 17761'. The attached file (pbqp_reduced.ll) is the failing test case found reduced with bugpoint. Apply patch 2 (the assert), and then run 'llc pbqp_reduced.ll -mtriple=aarch64-none-linux-gnu -mcpu=cortex-a57 -mattr=+neon -optimize-regalloc -regalloc=pbqp'. The assert shows a node that is spilled that was conservatively allocatable. The test case itself will not fail without this assert. Add the Verbose-PBQP-dump patch and use '-debug-only=regalloc -pbqp-dump-graphs' to see the more verbose output which shows the progress of the algorithm leading to the assert:<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>* Applied R2(NId 18)handleDisconnectEdge(9, 2) : DeniedOpts 10 -> 9<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>NId 9(%vreg15, GPR64common) moved to conservatively-allocatables.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>handleDisconnectEdge(2, 9) : DeniedOpts 10 -> 9<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>NId 2(%vreg4, GPR64common) moved to conservatively-allocatables.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>...<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Popped NId 2(%vreg4, GPR64common) , all edge costs added:<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>2.002748e+01 inf inf inf inf inf inf inf inf inf inf ** selection: 0<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>llc: ../include/llvm/CodeGen/PBQP/ReductionRules.h:214: llvm::PBQP::Solution llvm::PBQP::backpropagate(GraphT&, StackT, llvm::PBQP::NodeSet&) [with GraphT = llvm::PBQP::Gra<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>ph<llvm::PBQP::RegAlloc::RegAllocSolverImpl>; StackT = std::vector<unsigned int>; llvm::PBQP::NodeSet = std::set<unsigned int>]: Assertion `PushedAsConservativelyAllocatabl<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>eNodes.find(NId) == PushedAsConservativelyAllocatableNodes.end() && "Node in conservatively allocatable set spilled!"' failed.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>I am not that familiar with the arm architecture, but it looks like NId 2 is incorrectly moved from not-provens to conservatively-allocables, after R2(NId 18). This looks to me like the same bug, but I could be wrong. Note that it *<b>does not show up if the bugfix for the transpositions is applied</b>* (for random reasons I beleive).<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>I am not sure how to fix it myself, as it seems non-trivial relating to temporary edge removal. Does anyone have any idea?<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>In addition to the reduced test case, I attach four patches, where 1 and 4 are not ready to commit, but I would be happy to modify and commit them if wanted:<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>1. 0001-Assert… The assert (with some extra code related to it) that checks that a node that was conservatively allocatable never spills. My quick way to achieve this was to use an extra set of nodes as a means of remembering their origin. This is a very good idea, considering that this should never happen, and that there is some worry that the code might be incorrect.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>2. 0001-Bugfix-in-PBQP-matrix-transpositions… A bugfix for the transposition bug. (Bug 1). Small patch and also sent to llvm-commits for review.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>3. 0001-Bugfix-in-PBQP-regarding-Matrixes-and… A bugfix for the 1-column case (Bug 2). Small patch and also sent to llvm-commits for review.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>4. 0001-Verbose… Increased debug-dump and pbqp-graph output. Quite crude code, because I had a conflict in the ARM backend when I tried to #define DEBUG_TYPE "regalloc". I think this could help PBQP get a wider audience, perhaps, even if it is still far from perfect. Perhaps this could be cleaned up and a target-independent name for all pbqp-debug-dumps could be found and then commited.<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Looking forward to your reply,<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'>Jonas Paulsson<o:p></o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:black'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span lang=EN-US style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><div><div style='border:none;border-top:solid #B5C4DF 1.0pt;padding:3.0pt 0cm 0cm 0cm'><p class=MsoNormal><b><span lang=EN-US style='font-size:10.0pt;font-family:"Tahoma","sans-serif"'>From:</span></b><span lang=EN-US style='font-size:10.0pt;font-family:"Tahoma","sans-serif"'> Arnaud A. de Grandmaison [<a href="mailto:arnaud.degrandmaison@arm.com">mailto:arnaud.degrandmaison@arm.com</a>] <br><b>Sent:</b> den 27 januari 2015 08:43<br><b>To:</b> 'Lang Hames'; Jonas Paulsson<br><b>Cc:</b> <a href="mailto:llvmdev@cs.uiuc.edu">llvmdev@cs.uiuc.edu</a><br><b>Subject:</b> RE: [LLVMdev] PBQP crash<o:p></o:p></span></p></div></div><p class=MsoNormal><span lang=EN-US><o:p> </o:p></span></p><p class=MsoNormal><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>> A node should never be put into the conservatively allocatable list if there is a chance of it spilling.</span><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>I can understand why the logic of NodeMetadata::isConservativelyAllocatable is necessary for the node to be allocatable, but I have not been able to convince myself this is sufficient, especially when the node degree > available registers.<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>Cheers,<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'>Arnaud<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal style='margin-left:36.0pt'><b><span lang=EN-US style='font-size:10.0pt;font-family:"Tahoma","sans-serif"'>From:</span></b><span lang=EN-US style='font-size:10.0pt;font-family:"Tahoma","sans-serif"'> <a href="mailto:llvmdev-bounces@cs.uiuc.edu">llvmdev-bounces@cs.uiuc.edu</a> [<a href="mailto:llvmdev-bounces@cs.uiuc.edu">mailto:llvmdev-bounces@cs.uiuc.edu</a>] <b>On Behalf Of </b>Lang Hames<br><b>Sent:</b> 27 January 2015 06:06<br><b>To:</b> Jonas Paulsson<br><b>Cc:</b> <a href="mailto:llvmdev@cs.uiuc.edu">llvmdev@cs.uiuc.edu</a><br><b>Subject:</b> Re: [LLVMdev] PBQP crash<o:p></o:p></span></p><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p><div><p class=MsoNormal style='margin-left:36.0pt'>Hi Jonas,<o:p></o:p></p><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>> * The problematic node that was spilled again, was in the ConservativelyAllocatableNodes set during reduce(). The comment in reduce() “Conservatively allocatable nodes will never spill…” indicates that perhaps this is an incorrect insertion, as the regs did in fact run out in this case.</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>Arnaud is correct: A node should never be put into the conservatively allocatable list if there is a chance of it spilling. Off the top of my head I can imagine 2 things going wrong here:</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>(1) Conservative allocability is mostly-precomputed, and updated with deltas as nodes are removed. It is possible that there is some subtle bug in this code.</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>(2) I think the current conservative allocability test bakes in the assumption that all register options have non-infinite cost. If you assign infinite costs to any physical register I would expect this to blow up.</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>Are you able to share a test case at all? If so that would be great. If not, I can add an option to the allocator to dump abstract PBQP graphs, and I could use these to test the problem on my end.</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>Regards,</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><span style='font-size:11.5pt;font-family:"Calibri","sans-serif"'>Lang.</span><o:p></o:p></p></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div></div><div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p><div><p class=MsoNormal style='margin-left:36.0pt'>On Mon, Jan 26, 2015 at 7:55 AM, Jonas Paulsson <<a href="mailto:jonas.paulsson@ericsson.com" target="_blank">jonas.paulsson@ericsson.com</a>> wrote:<o:p></o:p></p><div><div><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=SV>Hi,</span><span lang=EN-US><o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=SV> </span><span lang=EN-US><o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>I have run into a test case on an out-of-tree target where PBQP fails to complete register allocation after “Attempting to spill already spilled value” (the triggered assert in InlineSpiller::spill().<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US> <o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>First, the original LiveInterval is spilled. It is a load of a symbol into a narrow register class, i.e. a subset of the class of address registers. InlineSpiller decides to rematerialize the load of the symbol to lie right before its only user, which makes good sense. The original def is removed.<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US> <o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>The new LiveInterval pushed is thus much smaller in the next PBQP round. The spill cost is marked as ‘inf’ during graph building. This small interval has also a lot of overlapping intervals and thus edges in the PBQP graph. It gets pushed on the node stack to later be popped after 17 others.<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>Those 17 nodes use up all registers of the narrow reg-class, and the cost vector has become all infinities. Spill option is selected again, and thus the error is a fact of spilling an already spilled value.<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US> <o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>I wonder what has gone wrong here, and have some initial thoughts:<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US> <o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>* The problematic node that was spilled again, was in the ConservativelyAllocatableNodes set during reduce(). The comment in reduce() “Conservatively allocatable nodes will never spill…” indicates that perhaps this is an incorrect insertion, as the regs did in fact run out in this case.<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US> In setup(), the node is first put into not-provably-allocatables. However, one of it’s neigbhour invoked handleDisconnectEdge(), and moves it into conservatively-allocatables, because DeniedOpts had become less than NumOpts (in isConservativelyAllocatable().<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>* There are lots of spillable nodes being popped before the one that can’t be spilled. This seems intuitively wrong, as they are intervals that actually could be spilled.<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US> <o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US>I would really appreciate some help and pointers on what might be going wrong here,<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US style='color:#888888'> <o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US style='color:#888888'>Jonas Paulsson<o:p></o:p></span></p><p class=MsoNormal style='mso-margin-top-alt:auto;mso-margin-bottom-alt:auto;margin-left:36.0pt'><span lang=EN-US style='color:#888888'> <o:p></o:p></span></p></div></div></div><p class=MsoNormal style='margin-left:36.0pt'><o:p> </o:p></p></div></div></div></body></html>