[X86] Align i128 to 16 bytes in x86 datalayouts

This is an attempt at rebooting https://reviews.llvm.org/D28990

I've included AutoUpgrade changes to modify the data layout to satisfy the compatible layout check. But this does mean alloca, loads, stores, etc in old IR will automatically get this new alignment.

This should fix PR46320.

Reviewed By: echristo, rnk, tmgross

Differential Revision: https://reviews.llvm.org/D86310

1 files changed, 18 insertions, 2 deletions

diff --git a/llvm/lib/IR/AutoUpgrade.cpp b/llvm/lib/IR/AutoUpgrade.cpp
index 304edc2..e102aae 100644
--- a/llvm/lib/IR/AutoUpgrade.cpp
+++ b/llvm/lib/IR/AutoUpgrade.cpp
@@ -5201,13 +5201,29 @@ std::string llvm::UpgradeDataLayoutString(StringRef DL, StringRef TT) {
   // If the datalayout matches the expected format, add pointer size address
   // spaces to the datalayout.
   std::string AddrSpaces = "-p270:32:32-p271:32:32-p272:64:64";
-  if (!DL.contains(AddrSpaces)) {
+  if (StringRef Ref = Res; !Ref.contains(AddrSpaces)) {
     SmallVector<StringRef, 4> Groups;
     Regex R("(e-m:[a-z](-p:32:32)?)(-[if]64:.*$)");
-    if (R.match(DL, &Groups))
+    if (R.match(Res, &Groups))
       Res = (Groups[1] + AddrSpaces + Groups[3]).str();
   }
 
+  // i128 values need to be 16-byte-aligned. LLVM already called into libgcc
+  // for i128 operations prior to this being reflected in the data layout, and
+  // clang mostly produced LLVM IR that already aligned i128 to 16 byte
+  // boundaries, so although this is a breaking change, the upgrade is expected
+  // to fix more IR than it breaks.
+  // Intel MCU is an exception and uses 4-byte-alignment.
+  if (!T.isOSIAMCU()) {
+    std::string I128 = "-i128:128";
+    if (StringRef Ref = Res; !Ref.contains(I128)) {
+      SmallVector<StringRef, 4> Groups;
+      Regex R("^(e(-[mpi][^-]*)*)((-[^mpi][^-]*)*)$");
+      if (R.match(Res, &Groups))
+        Res = (Groups[1] + I128 + Groups[3]).str();
+    }
+  }
+
   // For 32-bit MSVC targets, raise the alignment of f80 values to 16 bytes.
   // Raising the alignment is safe because Clang did not produce f80 values in
   // the MSVC environment before this upgrade was added.