py: Provide mp_decode_uint_skip() to help reduce stack usage.

dpgeorge · dpgeorge · commit a8a5d1e8c8db · 2017-06-09T13:36:33.000+10:00
Taking the address of a local variable leads to increased stack usage, so
the mp_decode_uint_skip() function is added to reduce the need for taking
addresses.  The changes in this patch reduce stack usage of a Python call
by 8 bytes on ARM Thumb, by 16 bytes on non-windowing Xtensa archs, and by
16 bytes on x86-64.  Code size is also slightly reduced on most archs by
around 32 bytes.
diff --git a/py/bc.c b/py/bc.c
@@ -64,6 +64,14 @@ mp_uint_t mp_decode_uint_value(const byte *ptr) {
     return mp_decode_uint(&ptr);
 }
 
+// This function is used to help reduce stack usage at the caller, for the case when
+// the caller doesn't need the actual value and just wants to skip over it.
+const byte *mp_decode_uint_skip(const byte *ptr) {
+    while ((*ptr++) & 0x80) {
+    }
+    return ptr;
+}
+
 STATIC NORETURN void fun_pos_args_mismatch(mp_obj_fun_bc_t *f, size_t expected, size_t given) {
 #if MICROPY_ERROR_REPORTING == MICROPY_ERROR_REPORTING_TERSE
     // generic message, used also for other argument issues
@@ -115,7 +123,7 @@ void mp_setup_code_state(mp_code_state_t *code_state, size_t n_args, size_t n_kw
 
     // get params
     size_t n_state = mp_decode_uint(&code_state->ip);
-    mp_decode_uint(&code_state->ip); // skip n_exc_stack
+    code_state->ip = mp_decode_uint_skip(code_state->ip); // skip n_exc_stack
     size_t scope_flags = *code_state->ip++;
     size_t n_pos_args = *code_state->ip++;
     size_t n_kwonly_args = *code_state->ip++;
diff --git a/py/bc.h b/py/bc.h
@@ -92,6 +92,7 @@ typedef struct _mp_code_state_t {
 
 mp_uint_t mp_decode_uint(const byte **ptr);
 mp_uint_t mp_decode_uint_value(const byte *ptr);
+const byte *mp_decode_uint_skip(const byte *ptr);
 
 mp_vm_return_kind_t mp_execute_bytecode(mp_code_state_t *code_state, volatile mp_obj_t inject_exc);
 mp_code_state_t *mp_obj_fun_bc_prepare_codestate(mp_obj_t func, size_t n_args, size_t n_kw, const mp_obj_t *args);
diff --git a/py/objfun.c b/py/objfun.c
@@ -141,11 +141,11 @@ const mp_obj_type_t mp_type_fun_builtin_var = {
 /* byte code functions                                                        */
 
 qstr mp_obj_code_get_name(const byte *code_info) {
-    mp_decode_uint(&code_info); // skip code_info_size entry
+    code_info = mp_decode_uint_skip(code_info); // skip code_info_size entry
     #if MICROPY_PERSISTENT_CODE
     return code_info[0] | (code_info[1] << 8);
     #else
-    return mp_decode_uint(&code_info);
+    return mp_decode_uint_value(code_info);
     #endif
 }
 
@@ -163,8 +163,8 @@ qstr mp_obj_fun_get_name(mp_const_obj_t fun_in) {
     #endif
 
     const byte *bc = fun->bytecode;
-    mp_decode_uint(&bc); // skip n_state
-    mp_decode_uint(&bc); // skip n_exc_stack
+    bc = mp_decode_uint_skip(bc); // skip n_state
+    bc = mp_decode_uint_skip(bc); // skip n_exc_stack
     bc++; // skip scope_params
     bc++; // skip n_pos_args
     bc++; // skip n_kwonly_args
@@ -205,12 +205,9 @@ mp_code_state_t *mp_obj_fun_bc_prepare_codestate(mp_obj_t self_in, size_t n_args
     MP_STACK_CHECK();
     mp_obj_fun_bc_t *self = MP_OBJ_TO_PTR(self_in);
 
-    // get start of bytecode
-    const byte *ip = self->bytecode;
-
     // bytecode prelude: state size and exception stack size
-    size_t n_state = mp_decode_uint(&ip);
-    size_t n_exc_stack = mp_decode_uint(&ip);
+    size_t n_state = mp_decode_uint_value(self->bytecode);
+    size_t n_exc_stack = mp_decode_uint_value(mp_decode_uint_skip(self->bytecode));
 
     // allocate state for locals and stack
     size_t state_size = n_state * sizeof(mp_obj_t) + n_exc_stack * sizeof(mp_exc_stack_t);
@@ -243,12 +240,9 @@ STATIC mp_obj_t fun_bc_call(mp_obj_t self_in, size_t n_args, size_t n_kw, const
     mp_obj_fun_bc_t *self = MP_OBJ_TO_PTR(self_in);
     DEBUG_printf("Func n_def_args: %d\n", self->n_def_args);
 
-    // get start of bytecode
-    const byte *ip = self->bytecode;
-
     // bytecode prelude: state size and exception stack size
-    size_t n_state = mp_decode_uint(&ip);
-    size_t n_exc_stack = mp_decode_uint(&ip);
+    size_t n_state = mp_decode_uint_value(self->bytecode);
+    size_t n_exc_stack = mp_decode_uint_value(mp_decode_uint_skip(self->bytecode));
 
 #if VM_DETECT_STACK_OVERFLOW
     n_state += 1;
diff --git a/py/objgenerator.c b/py/objgenerator.c
@@ -54,12 +54,9 @@ STATIC mp_obj_t gen_wrap_call(mp_obj_t self_in, size_t n_args, size_t n_kw, cons
     mp_obj_fun_bc_t *self_fun = (mp_obj_fun_bc_t*)self->fun;
     assert(self_fun->base.type == &mp_type_fun_bc);
 
-    // get start of bytecode
-    const byte *ip = self_fun->bytecode;
-
     // bytecode prelude: get state size and exception stack size
-    mp_uint_t n_state = mp_decode_uint(&ip);
-    mp_uint_t n_exc_stack = mp_decode_uint(&ip);
+    size_t n_state = mp_decode_uint_value(self_fun->bytecode);
+    size_t n_exc_stack = mp_decode_uint_value(mp_decode_uint_skip(self_fun->bytecode));
 
     // allocate the generator object, with room for local stack and exception stack
     mp_obj_gen_instance_t *o = m_new_obj_var(mp_obj_gen_instance_t, byte,
diff --git a/py/vm.c b/py/vm.c
@@ -1363,22 +1363,25 @@ unwind_jump:;
             // TODO need a better way of not adding traceback to constant objects (right now, just GeneratorExit_obj and MemoryError_obj)
             if (nlr.ret_val != &mp_const_GeneratorExit_obj && nlr.ret_val != &mp_const_MemoryError_obj) {
                 const byte *ip = code_state->fun_bc->bytecode;
-                mp_decode_uint(&ip); // skip n_state
-                mp_decode_uint(&ip); // skip n_exc_stack
+                ip = mp_decode_uint_skip(ip); // skip n_state
+                ip = mp_decode_uint_skip(ip); // skip n_exc_stack
                 ip++; // skip scope_params
                 ip++; // skip n_pos_args
                 ip++; // skip n_kwonly_args
                 ip++; // skip n_def_pos_args
                 size_t bc = code_state->ip - ip;
-                size_t code_info_size = mp_decode_uint(&ip);
+                size_t code_info_size = mp_decode_uint_value(ip);
+                ip = mp_decode_uint_skip(ip); // skip code_info_size
                 bc -= code_info_size;
                 #if MICROPY_PERSISTENT_CODE
                 qstr block_name = ip[0] | (ip[1] << 8);
                 qstr source_file = ip[2] | (ip[3] << 8);
                 ip += 4;
                 #else
-                qstr block_name = mp_decode_uint(&ip);
-                qstr source_file = mp_decode_uint(&ip);
+                qstr block_name = mp_decode_uint_value(ip);
+                ip = mp_decode_uint_skip(ip);
+                qstr source_file = mp_decode_uint_value(ip);
+                ip = mp_decode_uint_skip(ip);
                 #endif
                 size_t source_line = 1;
                 size_t c;
