<html><head><style type='text/css'>p { margin: 0; }</style></head><body><div style='font-family: arial,helvetica,sans-serif; font-size: 10pt; color: #000000'><br><hr id="zwchr"><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; color: rgb(0, 0, 0); font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><b>From: </b>"vivek pandya" <vivekvpandya@gmail.com><br><b>To: </b>"Quentin Colombet" <qcolombet@apple.com><br><b>Cc: </b>"Hal Finkel" <hfinkel@anl.gov>, "llvm-dev" <llvm-dev@lists.llvm.org>, "Matthias Braun" <matze@braunis.de>, "Mehdi Amini" <mehdi.amini@apple.com><br><b>Sent: </b>Tuesday, May 24, 2016 1:00:58 PM<br><b>Subject: </b>Re: [GSoC 2016] Interprocedural Register Allocation - Introduction and Feedback<br><br><div dir="ltr">Hello,<div><br></div><div>I have written following code to check each register if it is used by machineFunction or not :</div><div><br></div><div><div><font face="monospace, monospace">MachineRegisterInfo *MRI = &MF.getRegInfo();</font></div><div><font id="DWT13737" face="monospace, monospace"><span class="" style="white-space: pre;"> </span>TargetRegisterInfo *TRI = (TargetRegisterInfo *)MF.getSubtarget().getRegisterInfo();</font></div></div></div></blockquote>Some reason you can't use a const pointer here?<br><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; color: rgb(0, 0, 0); font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><div><div><font face="monospace, monospace"></font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>const TargetMachine &TM = MF.getTarget();</font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>const MCRegisterInfo *MCRI = TM.getMCRegisterInfo();</font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>DEBUG(dbgs() << "Function Name : " << MF.getName() << "\n");</font></div><div><font face="monospace, monospace"><br></font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>for(TargetRegisterInfo::regclass_iterator i = (*TRI).regclass_begin(), e = (*TRI).regclass_end(); i != e; i++ ) {</font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>for(TargetRegisterClass::iterator pregi = (*i)->begin(), prege = (*i)->end(); pregi != prege; pregi++ ) {</font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>DEBUG( dbgs() << "Physical Register : " << MCRI->getName(*pregi) << " is modified "<< MRI->isPhysRegModified(*pregi) << " \n");<span id="DWT14329" class="" style="white-space: pre;"> </span></font></div></div></div></blockquote>Try isPhysRegUsed.<br><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; color: rgb(0, 0, 0); font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"></span></font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>}</font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>}</font></div><div><font face="monospace, monospace"><span class="" style="white-space: pre;"> </span>DEBUG(dbgs() << "\n");</font></div></div><div><br></div><div id="DWT14330">The pass which is executing this code is schedule POST-RA stage but this gives me true for all registers i.e in each function all registers are being used except EBP and some other similar, Is this a correct way to get register usage information ? I think I have made some mistake please help.</div></div></blockquote><br>You might look at the implementation of these functions in lib/CodeGen/MachineRegisterInfo.cpp and figure out if they're returning true because UsedPhysRegMask.test(PhysReg) is true or because reg_nodbg_empty(*AliasReg) is true.<br><br> -Hal<br><br><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; color: rgb(0, 0, 0); font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><div></div><div><br></div><div>Vivek</div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, May 18, 2016 at 11:42 PM, Quentin Colombet <span dir="ltr"><<a href="mailto:qcolombet@apple.com" target="_blank">qcolombet@apple.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin: 0pt 0pt 0pt 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><br><div><div><div class="h5"><blockquote><div>On May 18, 2016, at 11:00 AM, vivek pandya <<a href="mailto:vivekvpandya@gmail.com" target="_blank">vivekvpandya@gmail.com</a>> wrote:</div><br><div><br><br style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;" clear="all"><div style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;"><div><div dir="ltr"><div><div dir="ltr"><i><font face="monospace, monospace" size="2"><b>Vivek Pandya</b></font></i><div><br></div></div></div></div></div></div><br style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;"><div class="gmail_quote" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;">On Wed, May 18, 2016 at 11:25 PM, Quentin Colombet<span> </span><span dir="ltr"><<a href="mailto:qcolombet@apple.com" target="_blank">qcolombet@apple.com</a>></span><span> </span>wrote:<br><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><br><div><div><div><blockquote><div>On May 18, 2016, at 10:46 AM, vivek pandya <<a href="mailto:vivekvpandya@gmail.com" target="_blank">vivekvpandya@gmail.com</a>> wrote:</div><br><div><br><br style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;" clear="all"><div style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;"><div><div dir="ltr"><div><div dir="ltr"><i><font face="monospace, monospace" size="2"><b>Vivek Pandya</b></font></i><div><br></div></div></div></div></div></div><br style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;"><div class="gmail_quote" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;">On Wed, May 11, 2016 at 4:01 PM, Hal Finkel<span> </span><span dir="ltr"><<a href="mailto:hfinkel@anl.gov" target="_blank">hfinkel@anl.gov</a>></span><span> </span>wrote:<br><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div><div style="font-family: arial,helvetica,sans-serif; font-size: 10pt;"><br><hr><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><b>From:<span> </span></b>"vivek pandya" <<a href="mailto:vivekvpandya@gmail.com" target="_blank">vivekvpandya@gmail.com</a>><br><b>To:<span> </span></b>"Mehdi Amini" <<a href="mailto:mehdi.amini@apple.com" target="_blank">mehdi.amini@apple.com</a>><br><b>Cc:<span> </span></b>"Hal Finkel" <<a href="mailto:hfinkel@anl.gov" target="_blank">hfinkel@anl.gov</a>>, "Quentin Colombet" <<a href="mailto:qcolombet@apple.com" target="_blank">qcolombet@apple.com</a>>, "llvm-dev" <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>>, "Matthias Braun" <<a href="mailto:matze@braunis.de" target="_blank">matze@braunis.de</a>><br><b>Sent:<span> </span></b>Wednesday, May 11, 2016 3:15:03 AM<br><b>Subject:<span> </span></b>Re: [GSoC 2016] Interprocedural Register Allocation - Introduction and Feedback<div><div><br><br><div dir="ltr"><br><div class="gmail_extra"><br clear="all"><div><div><div dir="ltr"><div><div dir="ltr"><i><font face="monospace, monospace" size="2"><b>Vivek Pandya</b></font></i><div><br></div></div></div></div></div></div><br><div class="gmail_quote">On Wed, May 11, 2016 at 10:02 AM, vivek pandya<span> </span><span dir="ltr"><<a href="mailto:vivekvpandya@gmail.com" target="_blank">vivekvpandya@gmail.com</a>></span><span> </span>wrote:<br><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div dir="ltr"><br><div class="gmail_extra"><br clear="all"><div><div><div dir="ltr"><div><div dir="ltr"><i><font face="monospace, monospace" size="2"><b>Vivek Pandya</b></font></i><div><br></div></div></div></div></div></div><br><div class="gmail_quote"><div><div>On Wed, May 11, 2016 at 9:43 AM, Mehdi Amini<span> </span><span dir="ltr"><<a href="mailto:mehdi.amini@apple.com" target="_blank">mehdi.amini@apple.com</a>></span><span> </span>wrote:<br><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><br><div><span><blockquote><div>On May 10, 2016, at 6:06 PM, Hal Finkel <<a href="mailto:hfinkel@anl.gov" target="_blank">hfinkel@anl.gov</a>> wrote:</div><br><div><div style="font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; font-family: arial,helvetica,sans-serif; font-size: 10pt;"><br><br><hr><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><b>From:<span> </span></b>"vivek pandya" <<a href="mailto:vivekvpandya@gmail.com" target="_blank">vivekvpandya@gmail.com</a>><br><b>To:<span> </span></b>"llvm-dev" <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>>, "Tim Amini Golling" <<a href="mailto:mehdi.amini@apple.com" target="_blank">mehdi.amini@apple.com</a>>, "Hal Finkel" <<a href="mailto:hfinkel@anl.gov" target="_blank">hfinkel@anl.gov</a>><br><b>Cc:<span> </span></b>"Quentin Colombet" <<a href="mailto:qcolombet@apple.com" target="_blank">qcolombet@apple.com</a>><br><b>Sent:<span> </span></b>Tuesday, May 10, 2016 2:59:16 PM<br><b>Subject:<span> </span></b>[GSoC 2016] Interprocedural Register Allocation - Introduction and Feedback<br><br><div dir="ltr"><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">Hello LLVM Community,<br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">Sorry for delay as I was busy in final exams.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">I am Vivek from India. Thanks for choosing my proposal for Interprocedural Register Allocation (IPRA) in LLVM. Mehdi Amini and Hal Finkel will be mentoring me for this project.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">IPRA can reduce code size and runtime of programs by allocating register across the module and procedure boundaries.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">I have identified some old but effective research work on this area.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">I want community's feedback for feasibility of these approach and I am targeting to implement two of them during this project.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">Here is list of the papers, I have read first two papers and I would like to discuss those approach first, I will read other two paper then initiate discussion for them as well. All I want is to find out a concrete implementation plan before 23 May, 2016 and for that I need community's help.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">1) Compile time ----- Minimizing register usage penalty at procedure calls -<span> </span><a href="http://dl.acm.org/citation.cfm?id=53999" target="_blank">http://dl.acm.org/citation.cfm?id=53999</a></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">====================================================================In this approach intra-procedural register allocation is used as base but machine code generation order is bottom up traversal of call graph and inter-procedural effect is achieved by propagating register usage information of callee function to caller (i.e child to parent in CallGraph) so that caller can use different registers than callee and can save load store cost at procedure call, this is not trivial as it seems due to recursive calls, library function usage etc. Also for upper region of the graph in this technique available number of registers might become zero in that case it should fall back to normal load store at procedure call. Apart from these difficulties other difficulties have been identified please follow this mail-chain<span> </span><a href="https://groups.google.com/d/topic/llvm-dev/HOYAXv3m1LY/discussion" target="_blank">https://groups.google.com/d/topic/llvm-dev/HOYAXv3m1LY/discussion</a></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">My mentor has already provided me a patch that alters code generation order as per bottom up call graph traversal, I am working from that point now. Any other help/suggestion is always welcomed.</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica; min-height: 14px;"><br></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">2) Link time ----- Global register allocation at link time -<span> </span><a href="http://dl.acm.org/citation.cfm?id=989415" target="_blank">http://dl.acm.org/citation.cfm?id=989415</a></div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">====================================================================In this particular approach (sort of true IPRA) registers will be reallocated (this optimization will be optional if turned off still code will be compiled as per intra-procedural allocation) at link time. Here modules are first complied as per normal compilation but the object code is annotated with details so that linker can build call graph and also calculate usage information at link time. Compiler also write hints in object code that if particular variable is allocated in some other register ( due to new allocation) then how the code should be changed? Thus linker can use these information to decide which variables (global) need to be in same register through out the program execution and also according to register usage information in call graph which procedure will not be active simultaneously so that locals for that procedures can be in same registers with out load store at procedure calls. </div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">For these particular method help me to analyze feasibility: </div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">1) Can llvm collects following information at module level in MachineIR? list of procedures in module, list of locals in procedures, list of procedures that a particular procedure can call, and a list of the variables this procedure references. Each entry in the last two lists includes an estimate of the number of times the procedure is called or the variable is referenced in each execution of this procedure </div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">2) Can llvm write informative commands to object files?</div><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">3) Can LTO is capable of leveraging those commands?<span> </span><br></div></div></blockquote>In terms of scoping the project for the summer, I definitely recommend that you focus on (1) first. If you finish that, we can certainly move on to other things.<span> </span></div></div></blockquote><div><br></div></span><div>I'll add +1 here, but I already wrote the same thing on IRC when discussing with Vivek. True IPRA without a proper MachineModule infrastructure won't be doable in my opinion (even with such infrastructure, it may not be trivial in LLVM in general).</div><span><br><blockquote><div><div style="font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; font-family: arial,helvetica,sans-serif; font-size: 10pt;">Regarding link time, note that any such a design would likely look much different than in David Wall's paper however, because our LTO re-codegens everything anyway. The paper says, "Finally, it keeps us honest as designers of the system; once we postpone anything until link time, the temptation is great to postpone everything, ..." - Well, we've long-since succumb to that temptation when we LTO. C'est la vie.<br></div></div></blockquote><div><br></div></span><div>+1 as well, our LTO will benefit naturally from the leaf-to-root information propagation. ThinLTO will be more challenging/interesting though!</div><span><blockquote><div><div style="font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; font-family: arial,helvetica,sans-serif; font-size: 10pt;"><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><p style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;"></p><div style="margin: 0px; font-size: 12px; line-height: normal; font-family: Helvetica;">For the first part a mechanism similar to MachineModulePass would be desirable but that may not be possible during this project, but if we can make some sort of smaller version of that to suit our purpose.</div></div></blockquote>I don't think we need to make any kind of MachineModulePass to make this work. Once we alter the visitation order based on the CGSCC iteration scheme, we can keep state in-between functions in the pre-existing hacky way (using static members of the relevant function passes).<br></div></div></blockquote><div><span style="font-size: 13px;"></span></div></span></div></div></blockquote></div></div><span><div> <span style="font-size: 13px;">Sorry my mistake here by first part I mean 1) requirement in the link time approach.</span></div><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><div><div></div></div></div></blockquote><div> </div></span><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><div><div>I also don't see where/why we need a MachineModule(Pass) for the CGSCC scheme, that said I'd rather avoid using a function pass with static members, if we can have a ModuleAnalysis that is bookkeeping the results for functions in the module and queries by the register allocator somehow.</div><span><div>Matthias/Quentin may have other inputs on this aspect.</div></span></div></div></blockquote></div></div></div></blockquote><div> </div><div>@Hal do you mean to add a simple MachineFunction pass that will just operate on register allocated function and prepare a BitVector to indicate which register is being used by MachineFunction, and then use this pass as analysis pass (i.e just simply return static BitVector for clobbered register when register allocation for next function begins. This part is not much clear to me) this thing can be done by scheduling a pass post register allocation in lib/CodeGen/Passes.cpp</div><div><div><br></div><div>void TargetPassConfig::addMachinePasses() {</div><div>. </div><div>.</div><div>.</div><div> <span> </span>// Run pre-ra passes.</div><div> <span> </span>addPreRegAlloc();</div><div><br></div><div> <span> </span>// Run register allocation and passes that are tightly coupled with it,</div><div> <span> </span>// including phi elimination and scheduling.</div><div> <span> </span>if (getOptimizeRegAlloc())</div><div> <span> </span>addOptimizedRegAlloc(createRegAllocPass(true));</div><div> <span> </span>else</div><div> <span> </span>addFastRegAlloc(createRegAllocPass(false));</div><div><br></div><div> <span> </span>// Run post-ra passes.</div><div> <span> </span>addPostRegAlloc();</div></div><div>// Adding a new pass here which keeps register mask information across function calls.</div><div>.</div><div>.</div><div>.</div><div>}</div><div><br></div><div>But this also requires current register allocators to use this information in someway because RegMaskBits in LiveIntervalAnalysis.cpp is not static across calls. I mean I am not clear for how to propagate static info to Intra-procedural Register allocators (if possible without disturbing their code )</div></div></div></div></div></div></blockquote>First, my hope is that we won't need to change the register allocators, as such, in order to make use of this information. Instead, we'll simply be able to alter the register masks generated for the call instructions. These masks will indicate fewer clobbers than might otherwise be present based on the ABI because of information gathered during the codegen of the callee. These masks are generally constructed by target based on the calling convention. The PowerPC backend, for example, looks like this:<br><br> <span> </span>// Add a register mask operand representing the call-preserved registers.<br> <span> </span>const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();<br> <span> </span>const uint32_t *Mask =<br> <span> </span>TRI->getCallPreservedMask(DAG.getMachineFunction(), CallConv);<br> <span> </span>assert(Mask && "Missing call preserved mask for calling convention");<br> <span> </span>Ops.push_back(DAG.getRegisterMask(Mask));<br><br>but it can be more complicated. If you look for uses of 'getRegisterMask' in Target/*/*ISelLowering.cpp, you'll see what I mean. Regardless, the code ends up calling some method is the targets TargetRegisterInfo subclass. These methods generally look something like this:<br><br>const uint32_t *<br>PPCRegisterInfo::getCallPreservedMask(const MachineFunction &MF,<br> <span> </span>CallingConv::ID CC) const {<br> <span> </span>const PPCSubtarget &Subtarget = MF.getSubtarget<PPCSubtarget>();<br> <span> </span>...<br> <span> </span>return TM.isPPC64() ? (Subtarget.hasAltivec() ? CSR_SVR464_Altivec_RegMask<br> <span> </span>: CSR_SVR464_RegMask)<br> <span> </span>: (Subtarget.hasAltivec() ? CSR_SVR432_Altivec_RegMask<br> <span> </span>: CSR_SVR432_RegMask);<br>}<br><br>In any case, the fundamental idea here is that, when someone calls getCallPreservedMask in order to set the regmask on a call, we might not have to use the CC at all. Instead, if we've already codegened the function, we might use a cache of 'exact' register masks computed during codegen of the potential callees instead.<br></div></div></blockquote><div>I am thinking to add a simple Immutable pass MachineRegisterUsageInfo similar to MachineBranchProbabilityInfo that can maintain RegisterUsageInformation per function. Can it be simply done by using UsedPhysRegMask from MachineRegisterInfo ??</div></div></div></blockquote><div><br></div></div></div><div>No, like the comment said, UsedPhysRegMask gives only the registers clobbered by calls:</div><div><span style="font-family: Menlo; font-size: 10px;">// This bit vector represents all the registers clobbered by function calls.</span> </div><div><br></div>You want to build this information yourself on top of MachineRegisterInfo::<span style="font-family: Menlo; font-size: 10px;">isPhysRegModified</span></div></div></blockquote><div>Ok but then the time complexity will be O(n) n = number of physical register on the target. Am I going correct?</div></div></div></blockquote><div><br></div></div></div><div>Yes, this is correct.</div><span class=""><br><blockquote><div><div class="gmail_quote" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;"><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><div><span><blockquote><div><div class="gmail_quote" style="font-family: Helvetica; font-size: 12px; font-style: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px;"><div> </div><div>Here getCallPreservedMask will call API provided by MachineRegisterUsageInfo to avail the exact register mask but how it can know that the function is already codegen or it will query each time when getCallPreservedMask is called and of available MachineRegisterUsageInfo will return the details otherwise simply return NULL.</div><div>So changes will be now in TargetRegisterInfo implementation for each target right ??</div><div><br></div><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div><div style="font-family: arial,helvetica,sans-serif; font-size: 10pt;"><br>In order to do this, I think we'll need to provide a function callable from the target's getCallPreservedMask implementation, which can return such an 'exact' regmask when available. I think we need to do it this way for two reasons:<br><br> 1. Not all of the target code calls getCallPreservedMask, but sometimes calls other similar target-specific functions (e.g. getTLSCallPreservedMask).<br> 2. The targets need to opt-in to this behavior because only the target can know that all register uses are really tagged correctly post "pre-emit".<br><br>Because the target is free to introduce uses of registers at essentially any time, we need to do the scanning for used registers after the "pre-emit" passes run. This can be done by scheduling some simple register-use scanning pass after the call to addPreEmitPass in lib/CodeGen/Passes.cpp.<span><br><br><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div></div><div><br></div><div>I think this also applies in someway to Mehdi Amini's idea to keep a ModulePass for bookkeeping but then existing register allocators will be required to change so that the code can query the ModulePass for RegMaskBits for particular function.</div></div></div></div></blockquote></span>I think that the simplest way to do this is to create an immutable analysis pass (e.g. BasicAA) that keeps the cache of the computed register masks. This is somewhat similar in spirit to how the 'AssumptionCache' analysis works at the IR level. This analysis can then be created by lib/CodeGen/Passes.cpp early, and then queried and passed around later by the CodeGen/Target code. Because it is an immutable analysis, it won't get destroyed until the very end, which is also important because, I imagine, it will need to own the memory associated with the generated register masks.<span><font color="#888888"><br><br> -Hal</font></span><span><br><blockquote style="border-left: 2px solid rgb(16, 16, 255); margin-left: 5px; padding-left: 5px; font-weight: normal; font-style: normal; text-decoration: none; font-family: Helvetica,Arial,sans-serif; font-size: 12pt;"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div></div><div><br></div><div>Vivek</div><div><br></div><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div dir="ltr"><div class="gmail_extra"><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 style="word-wrap: break-word;"><div><span><span><font color="#888888"><div><br></div></font></span></span></div></div></blockquote><div>Yes for propagating register usage approach we don't need MachineModulePass</div><span><font color="#888888"><div> </div><div>Vivek</div><blockquote class="gmail_quote" style="margin: 0px 0px 0px 0.8ex; border-left: 1px solid rgb(204, 204, 204); padding-left: 1ex;"><div style="word-wrap: break-word;"><div><span><font color="#888888"><div></div><div>-- </div><div>Mehdi</div><div><br></div></font></span></div></div></blockquote></font></span></div><br></div></div></blockquote></div><br></div></div></blockquote><br><br><br></span><span>--<span> </span><br><div><span></span>Hal Finkel<br>Assistant Computational Scientist<br>Leadership Computing Facility<br>Argonne National Laboratory</div></span></div></div></blockquote></div></div></blockquote></span></div></div></blockquote></div></div></blockquote></span></div><br></div></blockquote></div><br></div>
</blockquote><br><br><br>-- <br><div><span name="x"></span>Hal Finkel<br>Assistant Computational Scientist<br>Leadership Computing Facility<br>Argonne National Laboratory<span name="x"></span><br></div></div></body></html>