wasm-opt missing support for table.fill

[kg]

While wasm-opt appears to support table.grow correctly, any compiled module containing a table.fill instruction will fail to parse. This is true even if you pass the appropriate --enable-reference-types flag.

Sample code:

void
mono_jiterp_grow_function_table (int count, int fill_value) {
	__asm(
		// table.grow expects stack layout [fill_ref] [count]
		// start by loading fill_value, then fetching that reference from the indirect fn table
		"local.get 1;"
		"table.get __indirect_function_table;"
		// now fill_ref is on the stack, load count and perform the actual table grow
		"local.get 0;"
		"table.grow __indirect_function_table;"
		// table.grow returns either the new size or an error code. i'm too lazy to deal with it
		"drop;"
	);
}

#if 0
void
mono_jiterp_fill_function_table (int offset, int count, int fill_value) {
	__asm(
		// table.fill expects stack layout [offset] [fill_ref] [count]
		// load offset
		"local.get 0;"
		// load fill_value, then fetch that reference from the indirect fn table
		"local.get 2;"
		"table.get __indirect_function_table;"
		// now fill_ref is on the stack, load count and perform the actual table grow
		"local.get 1;"
		"table.fill __indirect_function_table;"
	);
}
#endif

change the #if 0 to #if 1 and the resulting built module will cause wasm-opt to fail with something like:

[parse exception: invalid code after misc prefix: 17 (at 0:266808)]
  Fatal: error parsing wasm (try --debug for more info)
emcc : error : '/home/kate/Projects/dotnet-runtime-wasm/src/mono/wasm/emsdk/upstream/bin/wasm-opt --strip-dwarf --post-emscripten -O2 --low-memory-unused --zero-filled-memory --pass-arg=directize-initial-contents-immutable --strip-debug --strip-producers dotnet.native.wasm -o dotnet.native.wasm -g --mvp-features --enable-mutable-globals --enable-reference-types --enable-sign-ext --enable-simd' failed (returned 1) [/home/kate/Projects/dotnet-runtime-wasm/src/mono/wasm/wasm.proj]

From checking the spec, 17 is table.fill. Looking through the repo it seems like support for table.init and table.copy might be missing too, but I'm not planning to use them...

[tlively]

You're the first person to ever ask for table.fill support! I can add it today.

[csjh]

Any chance table.init is being worked on? Seems like it's the lonely last table instruction left

[tlively]

There's no active work at the moment, but we might implement table.init soon as part of #6612.