Function alignment #22369
Replies: 1 comment 12 replies
-
|
Yes, this looks like a clang bug. Testcase: void foo() {}
void bar() {}
int main(int argc, char **argv) {
if (argc == 17) return (int)&foo;
else return (int)&bar;
}Compiling with or without @sbc100 is this a bug in wasm-ld specifically perhaps, I think the table is laid out there? |
Beta Was this translation helpful? Give feedback.
-
|
Ah, that does seem relatively simple. I've modified the function to the below, assuming my logic is correct, it should align all functions by 2: void ElemSection::addEntry(FunctionSymbol *sym) {
if (sym->hasTableIndex() || sym->isStub)
return;
uint64_t alignment = 2;
uint64_t index = config->tableBase + (indirectFunctions.size() * alignment);
sym->setTableIndex(index);
indirectFunctions.emplace_back(sym);
}However, linking with wasm-opt: /build/emscripten/src/binaryen/src/mixed_arena.h:188: T& ArenaVectorBase<SubType, T>::operator[](size_t) const [with SubType = ArenaVector<wasm::Expression*>; T = wasm::Expression*; size_t = long unsigned int]: Assertion `index < usedElements' failed.
emcc: error: '/usr/bin/wasm-opt --strip-target-features --post-emscripten -O3 --low-memory-unused --asyncify --pass-arg=asyncify-propagate-addlist --pass-arg=asyncify-imports@env.invoke_*,env.__asyncjs__*,*.fd_sync,*.emscripten_promise_await,*.emscripten_idb_load,*.emscripten_idb_store,*.emscripten_idb_delete,*.emscripten_idb_exists,*.emscripten_idb_clear,*.emscripten_idb_load_blob,*.emscripten_idb_store_blob,*.emscripten_sleep,*.emscripten_wget_data,*.emscripten_scan_registers,*.emscripten_lazy_load_code,*._load_secondary_module,*.emscripten_fiber_swap,*.SDL_Delay --zero-filled-memory --pass-arg=directize-initial-contents-immutable picodrive_libretro.wasm -o picodrive_libretro.wasm --mvp-features --enable-multivalue --enable-mutable-globals --enable-reference-types --enable-sign-ext' failed (received SIGABRT (-6))
make: *** [Makefile.emscripten:236: picodrive_libretro.js] Error 1Building with Perhaps it may not be as simple as modifying that function? Or my math is wrong |
Beta Was this translation helpful? Give feedback.
-
|
I think likely the setTableIndex argument has to match the index into indirectFunctions. Do you would likely need to push a one of more |
Beta Was this translation helpful? Give feedback.
-
|
Ah that would make sense. It's pretty unclean but this works. void ElemSection::addEntry(FunctionSymbol *sym) {
// Don't add stub functions to the wasm table. The address of all stub
// functions should be zero and they should they don't appear in the table.
// They only exist so that the calls to missing functions can validate.
uint32_t padding = 1;
uint64_t alignment = 2;
if (indirectFunctions.size() == 0 && padding > 0) {
for (uint32_t i=0; i<padding; i++) {
indirectFunctions.push_back(nullptr);
}
}
uint64_t index = config->tableBase + indirectFunctions.size();
if (sym->hasTableIndex() || sym->isStub)
return;
sym->setTableIndex(index);
indirectFunctions.emplace_back(sym);
if (alignment > 1) {
for (uint32_t i=0; i<alignment-1; i++) {
indirectFunctions.push_back(nullptr);
}
}
}I also had mo modify if (sym == nullptr) {
(void) tableIndex;
writeUleb128(os, 0, "function index");
++tableIndex;
continue;
} |
Beta Was this translation helpful? Give feedback.
-
|
Great! Would you like to send a PR and we can discuss further there? |
Beta Was this translation helpful? Give feedback.
-
|
Opened here: llvm/llvm-project#105532 Thanks! |
Beta Was this translation helpful? Give feedback.
-
I'm working on compiling a certain emulator via emscripten, where the lowest bit of every code pointer needs to be zero.
Setting
-falign-functions=2. Though this appears to be ignored (with no warnings from the compiler)https://github.com/libretro/picodrive/blob/d6f625a1251c78caf6f2dc81c1ffdb724587bb24/Makefile#L26
Is there a hacky way to get this done with a modern build of emscripten?
(P.S. The "EMULATED_FUNCTION_POINTERS" option seems to have disappeared without a trace)
Beta Was this translation helpful? Give feedback.
All reactions