[llvm-bugs] [Bug 46681] New: Missed optimization: Inlining behavior can lead to larger code with -Oz than other optimization levels

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=46681

            Bug ID: 46681
           Summary: Missed optimization: Inlining behavior can lead to
                    larger code with -Oz than other optimization levels
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Interprocedural Optimizations
          Assignee: unassignedbugs at nondot.org
          Reporter: vlad at tsyrklevich.net
                CC: llvm-bugs at lists.llvm.org

The following example demonstrates this problem. Note that the call to f(a)
with an uninitialized value is somewhat nonsensical--I blame creduce for that:
__attribute__((__noreturn__)) void assert();
void* d(void *e) { return e; }
void f(void *g) {
  if (g)
    assert();
}
static void compiles_to_ret() {
  void *a;
  f(a);
}
int example() {
  typedef void (*i)();
  i j = d(compiles_to_ret);
  j();

  return 2;
}

When compiled with clang -arch arm64 -Oz, example is compiled to:
0000000000000018 _example:
      18: fd 7b bf a9                   stp     x29, x30, [sp, #-16]!
      1c: fd 03 00 91                   mov     x29, sp
      20: 00 00 00 94                   bl      _compiles_to_ret
      24: 40 00 80 52                   mov     w0, #2
      28: fd 7b c1 a8                   ldp     x29, x30, [sp], #16
      2c: c0 03 5f d6                   ret

0000000000000030 _compiles_to_ret:
      30: c0 03 5f d6                   ret

With -Os, it compiles to:
0000000000000018 _example:
      18: 40 00 80 52                   mov     w0, #2
      1c: c0 03 5f d6                   ret

Using -mllvm -inline-threshold=10 fixes this trivial example; however, in the
original source code, which I can't share, it takes an inline threshold of over
100 to cause compiles_to_ret to be inlined. The following fixes the issue by
running the optimization pipeline twice:
$ clang -arch arm64 ex.c -Oz -S -emit-llvm -o ex.ll
$ clang -arch arm64 ex.ll -Oz -c ex.ll -o ex.o

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20200711/c410537c/attachment-0001.html>