diff --git a/src/cmd/cgo/out.go b/src/cmd/cgo/out.go
index d6740028da9..6ad601493c3 100644
--- a/src/cmd/cgo/out.go
+++ b/src/cmd/cgo/out.go
@@ -1003,7 +1003,7 @@ func (p *Package) writeExports(fgo2, fm, fgcc, fgcch io.Writer) {
 		// string.h for memset, and is also robust to C++
 		// types with constructors. Both GCC and LLVM optimize
 		// this into just zeroing _cgo_a.
-		fmt.Fprintf(fgcc, "\ttypedef %s %v _cgo_argtype;\n", ctype, p.packedAttribute())
+		fmt.Fprintf(fgcc, "\ttypedef %s %v __attribute__((aligned(8))) _cgo_argtype;\n", ctype, p.packedAttribute())
 		fmt.Fprintf(fgcc, "\tstatic _cgo_argtype _cgo_zero;\n")
 		fmt.Fprintf(fgcc, "\t_cgo_argtype _cgo_a = _cgo_zero;\n")
 		if gccResult != "void" && (len(fntype.Results.List) > 1 || len(fntype.Results.List[0].Names) > 1) {