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

[Daniel Paoliello via llvm-commits]

================
@@ -2271,20 +2293,64 @@ void X86AsmPrinter::emitInstruction(const MachineInstr *MI) {
   case X86::TAILJMPd64:
     if (IndCSPrefix && MI->hasRegisterImplicitUseOperand(X86::R11))
       EmitAndCountInstruction(MCInstBuilder(X86::CS_PREFIX));
-    [[fallthrough]];
-  case X86::TAILJMPr:
+
+    if (EnableImportCallOptimization && isImportedFunction(MI->getOperand(0))) {
+      emitLabelAndRecordForImportCallOptimization(
+          IMAGE_RETPOLINE_AMD64_IMPORT_BR);
+    }
+
+    // Lower these as normal, but add some comments.
+    OutStreamer->AddComment("TAILCALL");
+    break;
+  case X86::TAILJMPm64_REX:
+    if (EnableImportCallOptimization && isCallToCFGuardFunction(MI)) {
+      emitLabelAndRecordForImportCallOptimization(
+          IMAGE_RETPOLINE_AMD64_CFG_BR_REX);
+    }
+
+    // Lower these as normal, but add some comments.
+    OutStreamer->AddComment("TAILCALL");
+    break;
   case X86::TAILJMPm:
   case X86::TAILJMPd:
   case X86::TAILJMPd_CC:
-  case X86::TAILJMPr64:
   case X86::TAILJMPm64:
   case X86::TAILJMPd64_CC:
-  case X86::TAILJMPr64_REX:
-  case X86::TAILJMPm64_REX:
     // Lower these as normal, but add some comments.
     OutStreamer->AddComment("TAILCALL");
     break;
 
+  case X86::TAILJMPr:
+  case X86::TAILJMPr64:
+  case X86::TAILJMPr64_REX: {
+    MCInst TmpInst;
+    MCInstLowering.Lower(MI, TmpInst);
+
+    if (EnableImportCallOptimization) {
+      // Import call optimization requires all indirect calls go via RAX.
+      ensureRaxUsedForOperand(TmpInst);
+      emitLabelAndRecordForImportCallOptimization(
+          IMAGE_RETPOLINE_AMD64_INDIR_BR);
+    }
+
+    // Lower these as normal, but add some comments.
+    OutStreamer->AddComment("TAILCALL");
+    EmitAndCountInstruction(TmpInst);
+    return;
+  }
+
+  case X86::JMP64r:
+  case X86::JMP64m:
+    if (EnableImportCallOptimization && hasJumpTableInfoInBlock(MI)) {
+      uint16_t EncodedReg =
+          this->getSubtarget().getRegisterInfo()->getEncodingValue(
+              MI->getOperand(0).getReg().asMCReg());
----------------
dpaoliello wrote:

Good call: looks like x64 Windows will never use `JMP64m` for jump tables that's used only for non-PIC targets, so I can remove this.

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