Use emscripten_stack_set_limits in wasm_worker_initialize.S (#22755)

Followup to #22683. This change avoids making __stack_base and
__stack_end into public/external symbols and avoids reimplementing
`emscripten_stack_set_limits`.

This is a minor code size win too.

Author: sbc100

system/lib/compiler-rt/stack_limits.S

@@ -15,17 +15,14 @@
 #endif
 
 .globaltype __stack_pointer, PTR
-.globl __stack_pointer
 
 .section .globals,"",@
 
 # TODO(sbc): It would be nice if these we initialized directly
 # using PTR.const rather than using the `emscripten_stack_init`
 .globaltype __stack_end, PTR
 __stack_end:
-.globl __stack_end
 .globaltype __stack_base, PTR
-.globl __stack_base
 __stack_base:
 
 .section .text,"",@

system/lib/wasm_worker/wasm_worker_initialize.S

@@ -1,6 +1,4 @@
 .extern __stack_pointer
-.extern __stack_base
-.extern __stack_end
 
 #ifdef __wasm64__
 #define PTR i64
@@ -13,32 +11,37 @@
 #endif
 
 .globaltype __stack_pointer, PTR
-.globaltype __stack_end, PTR
-.globaltype __stack_base, PTR
+.globaltype __tls_size, PTR, immutable
+.functype emscripten_stack_set_limits (PTR /*base*/, PTR /*end*/) -> ()
 
 .globl _emscripten_wasm_worker_initialize
 _emscripten_wasm_worker_initialize:
   .functype _emscripten_wasm_worker_initialize (PTR /*stackLowestAddress*/, i32 /*stackSize*/) -> ()
+  .local PTR, PTR
 
-  // __stack_end = stackLowestAddress + (__builtin_wasm_tls_size() + 15) & -16;
+  // stack_end = stackLowestAddress + (__builtin_wasm_tls_size() + 15) & -16;
   local.get 0
-  .globaltype __tls_size, PTR, immutable
   global.get __tls_size
   PTR.add
   PTR.const 0xf
   PTR.add
   PTR.const -0x10
   PTR.and
-  global.set __stack_end
+  local.set 2
 
-  // __stack_base = stackLowestAddress + stackSize;
+  // stack_base = stackLowestAddress + stackSize;
   local.get 0
   local.get 1
 #ifdef __wasm64__
   i64.extend_i32_u
 #endif
   PTR.add
-  global.set __stack_base
+  local.set 3
+
+  local.get 3
+  local.get 2
+  // emscripten_stack_set_limits(stack_base, stack_end);
+  call emscripten_stack_set_limits
 
 // TODO: We'd like to do this here to avoid JS side calls to __set_stack_limits.
 //       (or even better, we'd like to avoid duplicate versions of the stack variables)
@@ -48,6 +51,7 @@ _emscripten_wasm_worker_initialize:
 //  .functype __set_stack_limits (PTR, PTR) -> ()
 //  call __set_stack_limits
 
+
   // __wasm_init_tls(stackLowestAddress);
   local.get 0
   .functype __wasm_init_tls (PTR) -> ()
@@ -57,8 +61,8 @@ _emscripten_wasm_worker_initialize:
   // __stack_pointer initialized, and it destroys the value it was set to.
   // So we must initialize __stack_pointer only *after* completing __wasm_init_tls:
 
-  // __stack_pointer = __stack_base;
-  global.get __stack_base
+  // __stack_pointer = stack_base;
+  local.get 3
   global.set __stack_pointer
 
   end_function

test/code_size/hello_wasm_worker_wasm.json

@@ -5,8 +5,8 @@
   "a.js.gz": 455,
   "a.ww.js": 115,
   "a.ww.js.gz": 127,
-  "a.wasm": 1894,
-  "a.wasm.gz": 1077,
-  "total": 3292,
-  "total_gz": 2043
+  "a.wasm": 1879,
+  "a.wasm.gz": 1066,
+  "total": 3277,
+  "total_gz": 2032
 }

test/core/test_dlfcn_self.exports

@@ -9,9 +9,7 @@ __progname_full
 __sig_actions
 __sig_pending
 __signgam
-__stack_base
 __stack_chk_guard
-__stack_end
 __threwValue
 __timezone
 __tzname