[M68k] Update pointer data layout

[ricky26]

Summary

Fixes PR51626.

The M68k requires that all instruction, word and long word reads are
aligned to word boundaries. From the 68020 onwards, there is a
performance benefit from aligning long words to long word boundaries.

The M68k uses the same data layout for pointers and integers.

In line with this, this commit updates the pointer data layout to
match the layout already set for 32-bit integers: 32:16:32.

[jrtc27]

Comments could perhaps be clearer about highlighting that 32-bit integers/pointers are only 16-bit aligned in the ABI as that's the weird bit, not the 32-bit preferred alignment

[glaubitz]

@ricky26 Could you also get this fix backported in the LLVM fork of Rust?

clang/lib/Basic/Targets/M68k.cpp

M68kTargetInfo::M68kTargetInfo(const llvm::Triple &Triple,
   std::string Layout = "";
 
   // M68k is Big Endian
   Layout += "E";
 
   // FIXME how to wire it with the used object format?
   Layout += "-m:e";
 
-  // M68k pointers are always 32 bit wide even for 16 bit cpus
-  Layout += "-p:32:32";
+  // M68k pointers are always 32 bit wide even for 16-bit CPUs
+  Layout += "-p:32:16:32";
 
   // M68k integer data types
   Layout += "-i8:8:8-i16:16:16-i32:16:32";
 
   // FIXME no floats at the moment
 
   // The registers can hold 8, 16, 32 bits
   Layout += "-n8:16:32";

llvm/lib/Target/M68k/M68kTargetMachine.cpp

std::string computeDataLayout(const Triple &TT, StringRef CPU,
                               const TargetOptions &Options) {
   std::string Ret = "";
   // M68k is Big Endian
   Ret += "E";
 
   // FIXME how to wire it with the used object format?
   Ret += "-m:e";
 
-  // M68k pointers are always 32 bit wide even for 16 bit cpus
-  Ret += "-p:32:32";
+  // M68k pointers are always 32 bit wide even for 16-bit CPUs.
+  // The ABI only specifies 16-bit alignment.
+  // On at least the 68020+ with a 32-bit bus, there is a performance benefit
+  // to having 32-bit alignment.
+  Ret += "-p:32:16:32";
 
-  // M68k requires i8 to align on 2 byte boundry
+  // Bytes do not require special alignment, words are word aligned and
+  // long words are word aligned at minimum.
   Ret += "-i8:8:8-i16:16:16-i32:16:32";
 
   // FIXME no floats at the moment
 
   // The registers can hold 8, 16, 32 bits
   Ret += "-n8:16:32";
 
   Ret += "-a:0:16-S16";