[llvm] [x64][win] Add compiler support for x64 import call optimization (equivalent to MSVC /d2guardretpoline) (PR #126631)

[Eli Friedman via llvm-commits]

================
@@ -2490,20 +2598,47 @@ void X86AsmPrinter::emitInstruction(const MachineInstr *MI) {
   MCInst TmpInst;
   MCInstLowering.Lower(MI, TmpInst);
 
-  // Stackmap shadows cannot include branch targets, so we can count the bytes
-  // in a call towards the shadow, but must ensure that the no thread returns
-  // in to the stackmap shadow.  The only way to achieve this is if the call
-  // is at the end of the shadow.
   if (MI->isCall()) {
-    // Count then size of the call towards the shadow
-    SMShadowTracker.count(TmpInst, getSubtargetInfo(), CodeEmitter.get());
-    // Then flush the shadow so that we fill with nops before the call, not
-    // after it.
-    SMShadowTracker.emitShadowPadding(*OutStreamer, getSubtargetInfo());
-    // Then emit the call
-    OutStreamer->emitInstruction(TmpInst, getSubtargetInfo());
+    emitCallInstruction(TmpInst);
     return;
   }
 
   EmitAndCountInstruction(TmpInst);
 }
+
+void X86AsmPrinter::emitCallInstruction(const llvm::MCInst &MCI) {
+  // Stackmap shadows cannot include branch targets, so we can count the bytes
+  // in a call towards the shadow, but must ensure that the no thread returns
+  // in to the stackmap shadow.  The only way to achieve this is if the call
+  // is at the end of the shadow.
+
+  // Count then size of the call towards the shadow
+  SMShadowTracker.count(MCI, getSubtargetInfo(), CodeEmitter.get());
+  // Then flush the shadow so that we fill with nops before the call, not
+  // after it.
+  SMShadowTracker.emitShadowPadding(*OutStreamer, getSubtargetInfo());
+  // Then emit the call
+  OutStreamer->emitInstruction(MCI, getSubtargetInfo());
+}
+
+void X86AsmPrinter::emitLabelAndRecordForImportCallOptimization(
+    ImportCallKind Kind) {
+  assert(EnableImportCallOptimization);
+
+  MCSymbol *CallSiteSymbol = MMI->getContext().createNamedTempSymbol("impcall");
+  OutStreamer->emitLabel(CallSiteSymbol);
+
+  SectionToImportedFunctionCalls[OutStreamer->getCurrentSectionOnly()]
+      .push_back({CallSiteSymbol, Kind});
+}
+
+void X86AsmPrinter::ensureRaxUsedForOperand(MCInst &TmpInst) {
----------------
efriedma-quic wrote:

This seems like a weird way to handle this... why not use a call pseudo-instruction that constrains the input register, so the rest of the compiler is aware of what's happening here?

What happens if a parameter needs to be in RAX?  That shouldn't happen for the C calling convention, but it can happen for alternative calling conventions.

https://github.com/llvm/llvm-project/pull/126631