Disable lsr in the wasm backend [ci skip] (emscripten-core#7386)

I did another round of tests now to follow up on WebAssembly/binaryen#1054 , comparing v8 and sm, on the wasm backend with and without lsr. Disabling lsr reduces code size in all the benchmark suite, up to 1% (but usually less). It's also somewhat faster (5-10%) in a few microbenchmarks, on both VMs (so this is not a VM-specific issue). So seems worthwhile.

LLVM bug: https://bugs.llvm.org/show_bug.cgi?id=39488

Author: kripken

tools/shared.py

@@ -1876,6 +1876,11 @@ def llvm_backend_args():
 
     # asm.js-style setjmp/longjmp handling
     args += ['-enable-emscripten-sjlj']
+
+    # better (smaller, sometimes faster) codegen, see binaryen#1054
+    # and https://bugs.llvm.org/show_bug.cgi?id=39488
+    args += ['-disable-lsr']
+
     return args
 
   @staticmethod