[LTO] Override TargetABI from module flags if present when creating TargetMachine (#126497)

…argetMachine

RISC-V's data layout is determined by the ABI, not just the target
triple. However, the TargetMachine is created using the data layout from
the target triple, which is not always correct. This patch uses the
target ABI from the module and passes it to the TargetMachine, ensuring
that the data layout is set correctly according to the ABI.

The same problem will happen with other targets like MIPS, but
unfortunately, MIPS didn't emit the target-abi into the module flags, so
this patch only fixes the issue for RISC-V.

NOTE: MIPS with -mabi=n32 can trigger the same issue.

Another possible solution is add new parameter to the TargetMachine
constructor, but that would require changes in all the targets.

Author: kito-cheng

llvm/include/llvm/IR/Module.h

@@ -1067,6 +1067,9 @@ class LLVM_ABI Module {
 
   /// Set the target variant version build SDK version metadata.
   void setDarwinTargetVariantSDKVersion(VersionTuple Version);
+
+  /// Returns target-abi from MDString, null if target-abi is absent.
+  StringRef getTargetABIFromMD();
 };
 
 /// Given "llvm.used" or "llvm.compiler.used" as a global name, collect the

llvm/lib/IR/Module.cpp

@@ -915,3 +915,11 @@ VersionTuple Module::getDarwinTargetVariantSDKVersion() const {
 void Module::setDarwinTargetVariantSDKVersion(VersionTuple Version) {
   addSDKVersionMD(Version, *this, "darwin.target_variant.SDK Version");
 }
+
+StringRef Module::getTargetABIFromMD() {
+  StringRef TargetABI;
+  if (auto *TargetABIMD =
+          dyn_cast_or_null<MDString>(getModuleFlag("target-abi")))
+    TargetABI = TargetABIMD->getString();
+  return TargetABI;
+}

llvm/lib/LTO/LTOBackend.cpp

@@ -221,8 +221,13 @@ createTargetMachine(const Config &Conf, const Target *TheTarget, Module &M) {
   else
     CodeModel = M.getCodeModel();
 
+  TargetOptions TargetOpts = Conf.Options;
+  if (TargetOpts.MCOptions.ABIName.empty()) {
+    TargetOpts.MCOptions.ABIName = M.getTargetABIFromMD();
+  }
+
   std::unique_ptr<TargetMachine> TM(TheTarget->createTargetMachine(
-      TheTriple.str(), Conf.CPU, Features.getString(), Conf.Options, RelocModel,
+      TheTriple.str(), Conf.CPU, Features.getString(), TargetOpts, RelocModel,
       CodeModel, Conf.CGOptLevel));
 
   assert(TM && "Failed to create target machine");

llvm/test/LTO/RISCV/lit.local.cfg

@@ -0,0 +1,2 @@
+if "RISCV" not in config.root.targets:
+    config.unsupported = True

llvm/test/LTO/RISCV/riscv-ilp32e.ll

@@ -0,0 +1,21 @@
+; Check that we don't crash on DataLayout incompatibility issue.
+; RUN: llvm-as %s -o %t.o
+; RUN: llvm-lto2 run -r %t.o,_start %t.o -o %t.elf
+; RUN: llvm-readobj -h %t.elf.0 | FileCheck %s --check-prefixes=CHECK
+; CHECK:  Machine: EM_RISCV (0xF3)
+; CHECK:  EF_RISCV_RVE (0x8)
+
+
+target datalayout = "e-m:e-p:32:32-i64:64-n32-S32"
+target triple = "riscv32-unknown-unknown-elf"
+
+define dso_local i32 @_start() #0 {
+entry:
+  ret i32 0
+}
+
+attributes #0 = { "target-cpu"="generic-rv32" "target-features"="+32bit,+e" }
+
+!llvm.module.flags = !{!0}
+
+!0 = !{i32 1, !"target-abi", !"ilp32e"}