<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">2015-08-05 14:17 GMT-07:00 Duncan P. N. Exon Smith <span dir="ltr"><<a href="mailto:dexonsmith@apple.com" target="_blank">dexonsmith@apple.com</a>></span>:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5"><br>
> On 2015-Jul-30, at 16:47, Alex L <<a href="mailto:arphaman@gmail.com">arphaman@gmail.com</a>> wrote:<br>
><br>
> Hi,<br>
><br>
> This patch serializes the UsesPhysRegMask register mask from the machine register<br>
> information class. The mask is serialized as an inverted 'calleeSavedRegisters' mask<br>
> to keep the output minimal. It uses the following syntax:<br>
><br>
> calleeSavedRegisters: [ '%bh', '%bl', '%bp', '%bpl', '%bx', '%ebp', '%ebx',<br>
> '%rbp', '%rbx', '%r12', '%r13', '%r14', '%r15',<br>
> '%r12b', '%r13b', '%r14b', '%r15b', '%r12d', '%r13d',<br>
> '%r14d', '%r15d', '%r12w', '%r13w', '%r14w', '%r15w' ]<br>
><br>
> As you can see, the sub registers are included in this mask as well.<br>
><br>
> This patch also allows the MIR parser to infer this mask from the register mask<br>
> operands if the machine function doesn't specify it.<br>
><br>
> Cheers,<br>
> Alex<br>
</div></div>> <0001-MIR-Serialization-serialize-the-UsesPhysRegMask-from.patch><br>
<br>
> From ca446d1f7a90691ffcc0c1ac16855e5b0eb7a9d0 Mon Sep 17 00:00:00 2001<br>
> From: Alex Lorenz <<a href="mailto:arphaman@gmail.com">arphaman@gmail.com</a>><br>
> Date: Thu, 30 Jul 2015 16:37:25 -0700<br>
> Subject: [PATCH] MIR Serialization: serialize the UsesPhysRegMask from<br>
> register info<br>
><br>
> ---<br>
> include/llvm/CodeGen/MIRYamlMapping.h | 2 +<br>
> include/llvm/CodeGen/MachineRegisterInfo.h | 4 +<br>
> lib/CodeGen/MIRParser/MIRParser.cpp | 31 ++++++++<br>
> lib/CodeGen/MIRPrinter.cpp | 12 +++<br>
> .../MIR/X86/used-physical-register-info.mir | 89 ++++++++++++++++++++++<br>
> 5 files changed, 138 insertions(+)<br>
> create mode 100644 test/CodeGen/MIR/X86/used-physical-register-info.mir<br>
><br>
> diff --git a/lib/CodeGen/MIRPrinter.cpp b/lib/CodeGen/MIRPrinter.cpp<br>
> index f34cef7..9ce7e19 100644<br>
> --- a/lib/CodeGen/MIRPrinter.cpp<br>
> +++ b/lib/CodeGen/MIRPrinter.cpp<br>
> @@ -213,6 +213,18 @@ void MIRPrinter::convert(yaml::MachineFunction &MF,<br>
> printReg(I->second, LiveIn.VirtualRegister, TRI);<br>
> MF.LiveIns.push_back(LiveIn);<br>
> }<br>
> + // The used physical register mask is printed as an inverted callee saved<br>
> + // register mask.<br>
> + const BitVector &UsedPhysRegMask = RegInfo.getUsedPhysRegsMask();<br>
> + if (UsedPhysRegMask.none())<br>
> + return;<br>
<br>
Will this cause a problem on the parsing side? In particular, won't the<br>
reader infer "the default set" here? To put it another way, how do you<br>
express in MIR that no registers should be marked callee-saved?<br></blockquote><div><br></div><div>So, this none condition is triggered only when the machine function</div><div>doesn't have any register mask operands. Because of that, we don't</div><div>really want to serialize it as this mask isn't really active in this case.</div><div>And the parser will try to infer it when parsing it back in this case,</div><div>but I think it's better if it will do that - if the MIR file wasn't modified</div><div>the result mask will stay exactly the same as it was before it was</div><div>serialized, and otherwise, if the user modified the MIR file and added</div><div>some instructions with register mask operands then the parser will</div><div>compute the mask correctly.</div><div><br></div><div>But I have <span style="line-height:normal">updated the patch anyway with a related feature - so now the</span></div><div><span style="line-height:normal">parser won't infer the mask </span><span style="line-height:normal">when </span><span style="line-height:normal">you specify an empty array of</span></div><div><span style="line-height:normal">callee-saved registers. Although I'm not </span><span style="line-height:normal">sure if it's really that useful,</span></div><div><span style="line-height:normal">you wouldn't normally want to have an empty </span><span style="line-height:normal">array here anyway.</span></div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<br>
> + for (unsigned I = 0, E = UsedPhysRegMask.size(); I != E; ++I) {<br>
> + if (!UsedPhysRegMask[I]) {<br>
> + yaml::FlowStringValue Reg;<br>
> + printReg(I, Reg, TRI);<br>
> + MF.CalleeSavedRegisters.push_back(Reg);<br>
> + }<br>
> + }<br>
> }<br>
><br>
> void MIRPrinter::convert(ModuleSlotTracker &MST,<br>
<br>
</blockquote></div><br></div></div>