[clang] [FMV] Emit the resolver along with the default version definition. (PR #84405)

[Alexandros Lamprineas via cfe-commits]

================
@@ -84,9 +84,33 @@ int hoo(void) {
   return fp1() + fp2();
 }
 
+// This should generate one target version but no resolver.
+__attribute__((target_version("default"))) int unused_with_forward_default_decl(void);
+__attribute__((target_version("mops"))) int unused_with_forward_default_decl(void) { return 0; }
 
+// This should also generate one target version but no resolver.
+extern int unused_with_implicit_extern_forward_default_decl(void);
+__attribute__((target_version("dotprod")))
+int unused_with_implicit_extern_forward_default_decl(void) { return 0; }
 
+// This should also generate one target version but no resolver.
+__attribute__((target_version("aes"))) int unused_with_default_decl(void) { return 0; }
+__attribute__((target_version("default"))) int unused_with_default_decl(void);
 
+// This should generate two target versions and the resolver.
+__attribute__((target_version("sve"))) int unused_with_default_def(void) { return 0; }
+__attribute__((target_version("default"))) int unused_with_default_def(void) { return 1; }
+
+// This should also generate two target versions and the resolver.
+__attribute__((target_version("fp16"))) int unused_with_implicit_default_def(void) { return 0; }
+int unused_with_implicit_default_def(void) { return 1; }
+
+// This should also generate two target versions and the resolver.
+int unused_with_implicit_forward_default_def(void) { return 0; }
+__attribute__((target_version("lse"))) int unused_with_implicit_forward_default_def(void) { return 1; }
+
+// This should generate a normal function.
+__attribute__((target_version("rdm"))) int unused_without_default(void) { return 0; }
----------------
labrinea wrote:

I do remember this use case from another review (https://godbolt.org/z/3zaM9MzPb) so I wanted to make sure we support it, which is the case! Now if we would further like to support multi versioning in the absense of default declaration (the header not included) that's a different story. I am not opposing, but will have to check how feasible it is to implement and what do the rest of the stakeholders think about it.

https://github.com/llvm/llvm-project/pull/84405