[mlir][AMDGPU] Set ABI version constant when linking device libs

[krzysz00]

Summary

Currently, linking the device libraries requires setting a constant
that indicates the code object ABI version the compilation is
targeting.

This fixes the MLIR linking process by setting this constant to 400,
which is the value corresponding to the current code object ABI
default, version 4.

[Mogball]

We don't have tests for this stuff?

[krzysz00]

The only test I can think of that could catch this is an integration test, and our buildbot isn't testing against pre-release versions of the backend

mlir/lib/Dialect/GPU/Transforms/SerializeToHsaco.cpp

if (needOcml || needOckl) {
     if (chipSet.startswith("gfx"))
       chipSet = chipSet.substr(3);
     uint32_t minor =
         llvm::APInt(32, chipSet.substr(chipSet.size() - 2), 16).getZExtValue();
     uint32_t major = llvm::APInt(32, chipSet.substr(0, chipSet.size() - 2), 10)
                          .getZExtValue();
     uint32_t isaNumber = minor + 1000 * major;
     addControlConstant("__oclc_ISA_version", isaNumber, 32);
+
+    // This constant must always match the default code object ABI version
+    // of the AMDGPU backend.
+    addControlConstant("__oclc_ABI_version", 400, 32);
   }
 
   // Determine libraries we need to link - order matters due to dependencies
   llvm::SmallVector<StringRef, 4> libraries;
   if (needOpenCl)
     libraries.push_back("opencl.bc");
   if (needOcml)
     libraries.push_back("ocml.bc");