[llvm] [PowerPC][NFC] Allow different orders of .extern in some test cases (PR #89714)

Chen Zheng via llvm-commits llvm-commits at lists.llvm.org
Tue Apr 23 17:45:17 PDT 2024


================
@@ -612,14 +612,14 @@ entry:
 
 ; External symbol reference checks for .__tls_get_addr/.__tls_get_mod
 
-; SMALL32: .extern .__tls_get_addr[PR]
-; SMALL32: .extern .__tls_get_mod[PR]
-; SMALL64: .extern .__tls_get_addr[PR]
-; SMALL64: .extern .__tls_get_mod[PR]
-; LARGE32: .extern .__tls_get_addr[PR]
-; LARGE32: .extern .__tls_get_mod[PR]
-; LARGE64: .extern .__tls_get_addr[PR]
-; LARGE64: .extern .__tls_get_mod[PR]
+; SMALL32: .extern .{{__tls_get_addr|__tls_get_mod}}[PR]
----------------
chenzheng1030 wrote:

OK. One more question is do we know why TLS external function calls are impacted by the iteration order, while for below cases:
```
int foo();
int bar();

int test() {
  return foo() + bar();
}
```

The external `foo()` and `bar()` are the same with both iteration order.

Could you please help to figure that out? And yes, like @efriedma-quic , we may need to fix this in the compiler, not the test case.

Thanks a lot.

https://github.com/llvm/llvm-project/pull/89714


More information about the llvm-commits mailing list