[ubsan][test] Add test to check interaction between inlining and attribute no_sanitize("undefined")

[vhscampos]

Summary

A function with attribute no_sanitize("undefined") should not have UBSAN
checks generated for its body. This requirement must stay true even if
said function is inlined.

This patch adds a tests to check this behaviour. Similar tests already
exist for ASAN and MSAN.

[eugenis]

Similar tests already exist for ASAN and MSAN.

Is that really true? Any difference in sanitize/no_sanitize IR attributes suppresses normal inlining, but always_inline trumps that.

This test should work for ubsan because ubsan instrumentation is done in the frontend before any inlining, but I believe this is not true for asan and msan.

compiler-rt/test/ubsan/TestCases/Misc/inline.c

+// RUN: %clang -fsanitize=undefined -O3 %s -o %t0
+// RUN: %run %t0 2> %t1
+// RUN: FileCheck --allow-empty %s < %t1
+
+// Test that no_sanitize("undefined") attribute applies even when the
+// function is inlined.
+
+// CHECK-NOT: runtime error: signed integer overflow
+
+__attribute__((always_inline, no_sanitize("undefined")))
+int f(int p) {
+    return p+1;
+}
+
+int main() {
+    volatile int x = 0x7FFFFFFF;
+    volatile int res = f(x);
+    return 0;
+}