How to make C++ global compile to Wasm global?

[GirkovArpa]

This code:

typedef long int i32;

i32 myGlobal = 1337;

extern "C" {
  i32 myFunc() {
    return myGlobal;
  }
}

emcc -O3 -Wl,--no-entry -s INITIAL_MEMORY=64kb -s MAXIMUM_MEMORY=64kb -s ALLOW_MEMORY_GROWTH=0 -s TOTAL_STACK=0kb -s STANDALONE_WASM -s EXPORTED_FUNCTIONS="['_myFunc']" "index.cpp" -o "index.wasm" -s GLOBAL_BASE=0

Compiles to this:

(module
  (type $t0 (func (result i32)))
  (func $myFunc (type $t0) (result i32)
    i32.const 0
    i32.load)
  (memory $memory 1 1)
  (export "memory" (memory 0))
  (export "myFunc" (func $myFunc))
  (data $d0 (i32.const 0) "9\05") ;; 9\05 I guess is 1337 in some base?
  (data $d1 (i32.const 512) "\a0\02")) ;; bug, separate issue

It appears that the value of myGlobal is stored in linear memory, as opposed to an actual WebAssembly global. I would like to make it compile to an actual Wasm global, maybe like this:

  (global $g0 (mut i32) (i32.const 1337))

I know that Emscripten supports this because a separate project I have had variables compiled to globals, but it's a lot of code and I figured I'd ask here first before trying to go Sherlock-Holmes on that. 😄

[sbc100]

Clang doesn't currently support wasm globals in C/C++. Global variables in C/C++ live in linear memory and have address. This is fairly fundamental to the language. If you want mess with actual wasm globals would have use assembly in some way. I'll come back to you with an example of this if I can.

Having said that, with the wasm backend you can export the address of the global directly:

For example:

$ cat test.c 
#include <emscripten.h>

EMSCRIPTEN_KEEPALIVE int foo = 3;

EMSCRIPTEN_KEEPALIVE int bar() {
  return foo;
}

int main() {
  return 0;
}
$ ./emcc -O2 test.c
$ wasm-objdump -x a.out.wasm
...
Global[3]:
 - global[0] i32 mutable=1 - init i32=5244416
 - global[1] i32 mutable=0 <foo> - init i32=1024
 - global[2] i32 mutable=0 <__data_end> - init i32=1528
Export[12]:
 - func[1] <__wasm_call_ctors> -> "__wasm_call_ctors"
 - func[2] <bar> -> "bar"
 - global[1] -> "foo"
 - func[3] <main> -> "main"
...

So here you can see that its possible export the address of foo, which is 1024 (that start the static data section). You can then gets its value in JS by doing HEAPU32[Module['_foo']>>2].

Note that this did not work in the old fastcomp backend. Exporting a data symbol never worked there. Its also not yet a well known, used, or tested feature.

[sbc100]

It looks like we currently can't define new wasm globals, even in the assembly format. I opened a bug for that: https://bugs.llvm.org/show_bug.cgi?id=45742

[sbc100]

You can now define new wasm globals using the assembly format if you need to.

See: https://github.com/emscripten-core/emscripten/blob/master/system/lib/compiler-rt/stack_ops.s

[TerrorJack]

Late to the party, but the URL above is outdated, now it's https://github.com/emscripten-core/emscripten/blob/main/system/lib/compiler-rt/stack_ops.S