Compilation time of coroutine grows non-linearly

[dyzsr]

In clang 15.0.3, the compilation time of coroutine seems to grow non-linearly against code size (I have tested on linux and macOS).

Here are the simplified examples I generated.

• task.h which defines the coroutine data structures:

#include <coroutine>

struct promise;

struct task : std::coroutine_handle<promise> {
  using promise_type = struct promise;
};

struct promise {
  task get_return_object() { return {task::from_promise(*this)}; }
  std::suspend_always initial_suspend() noexcept { return {}; }
  std::suspend_always final_suspend() noexcept { return {}; }
  void return_void() noexcept {}
  void unhandled_exception() {}
};

• gen.cpp which generates the coroutine use cases:

#include <cassert>
#include <cstdio>
#include <cstdlib>

int main(int argc, char **argv) {
  assert(argc == 2);
  int n = atoi(argv[1]);
  printf("#include <cstdio>\n");
  printf("#include \"task.h\"\n");
  printf("task t() {\n");
  for (int i = 1; i <= n; i++) printf("int cond_%d = 1;\n", i);
  printf("int val = 0;\n");
  for (int i = 1; i <= n; i++) printf("if (cond_%d) val++;\n", i);
  printf("printf(\"%%d\\n\", val);\n");
  printf("co_return;\n");
  printf("}\n");
  return 0;
}

• test.sh used to run the test:

#!/bin/sh

clang++ -o gen gen.cpp
for i in 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000
do
  printf "n: %d\n" $i
  ./gen $i > a.cpp
  time clang++ -c a.cpp -std=c++20 -stdlib=libc++
  printf "\n"
done

The generated a.cpp is like:

#include <cstdio>
#include "task.h"
task t() {
int cond_1 = 1;
int cond_2 = 1;
int cond_3 = 1;
...
int val = 0;
if (cond_1) val++;
if (cond_2) val++;
if (cond_3) val++;
...
printf("%d\n", val);
co_return;
}

My test result:

n: 2000

real	0m2.697s
user	0m2.583s
sys	0m0.105s

n: 4000

real	0m8.662s
user	0m8.314s
sys	0m0.332s

n: 6000

real	0m19.107s
user	0m18.330s
sys	0m0.734s

n: 8000

real	0m35.622s
user	0m34.018s
sys	0m1.436s

n: 10000

real	0m50.344s
user	0m48.158s
sys	0m1.959s

n: 12000

real	1m12.844s
user	1m9.896s
sys	0m2.728s

n: 14000

real	1m57.271s
user	1m52.938s
sys	0m4.076s

n: 16000

real	2m42.044s
user	2m35.904s
sys	0m5.635s

n: 18000

real	3m28.608s
user	3m20.904s
sys	0m7.057s

n: 20000

real	4m6.561s
user	3m56.836s
sys	0m9.012s

The compilation speed is really slow for very large coroutines (which can be generated by program in our use cases).

[llvmbot]

@llvm/issue-subscribers-coroutines

[ChuanqiXu9]

Thanks for the reporting! It is concise and interesting. I'll take a look.

[yurai007]

Proposed fix: https://reviews.llvm.org/D140818

[ChuanqiXu9]

This one should be fixed by the above patch.

[microcai]

I was hited by this bug!