[AMDGPU] Fix codegen for inline assembly

Need to override convertConstraint to recognise amdgpu specific register names.

Differential Revision: https://reviews.llvm.org/D44533

llvm-svn: 328359
diff --git a/clang/lib/CodeGen/CGStmt.cpp b/clang/lib/CodeGen/CGStmt.cpp
index b1d1bf8..b89d9f0 100644
--- a/clang/lib/CodeGen/CGStmt.cpp
+++ b/clang/lib/CodeGen/CGStmt.cpp
@@ -1926,7 +1926,7 @@
     // Simplify the output constraint.
     std::string OutputConstraint(S.getOutputConstraint(i));
     OutputConstraint = SimplifyConstraint(OutputConstraint.c_str() + 1,
-                                          getTarget());
+                                          getTarget(), &OutputConstraintInfos);
 
     const Expr *OutExpr = S.getOutputExpr(i);
     OutExpr = OutExpr->IgnoreParenNoopCasts(getContext());