Hulon's and mine changes

Co-Authored-By: Hulon Jenkins <[email protected]>

Author: Fidget-Spinner

Include/pyport.h

@@ -385,6 +385,24 @@ extern "C" {
 #  define Py_NO_INLINE
 #endif
 
+// Any function annotated with this MUST not modify global state.
+// It can only modify state referenced by its parameters.
+// This is useful for optimizations on certain compilers.
+// Please see https://learn.microsoft.com/en-us/cpp/cpp/noalias
+#if defined(__GNUC__) || defined(__clang__) || defined(__INTEL_COMPILER)
+#  define Py_NOALIAS
+#elif defined(_MSC_VER)
+#  define Py_NOALIAS __declspec(noalias)
+#else
+#  define Py_NOALIAS
+#endif
+
+// A no-op at compile time. Hints to the programmer
+// That any local variable defined within this block MUST
+// not escape from the current definition.
+# define Py_BEGIN_LOCALS_MUST_NOT_ESCAPE() {
+# define Py_END_LOCALS_MUST_NOT_ESCAPE() }
+
 #include "exports.h"
 
 #ifdef Py_LIMITED_API

PCbuild/pythoncore.vcxproj

@@ -108,6 +108,7 @@
       <PreprocessorDefinitions Condition="'$(UseJIT)' == 'true'">_Py_JIT;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <PreprocessorDefinitions Condition="'$(UseTIER2)' != '' and '$(UseTIER2)' != '0'">_Py_TIER2=$(UseTIER2);%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <PreprocessorDefinitions Condition="'$(UseTailCallInterp)' == 'true'">Py_TAIL_CALL_INTERP=1;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <AdditionalOptions Condition="'$(UseTailCallInterp)' == 'true'">/std:clatest %(AdditionalOptions)</AdditionalOptions>
       <PreprocessorDefinitions Condition="'$(WITH_COMPUTED_GOTOS)' != ''">HAVE_COMPUTED_GOTOS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <PreprocessorDefinitions Condition="'$(DisableRemoteDebug)' != 'true'">Py_REMOTE_DEBUG;%(PreprocessorDefinitions)</PreprocessorDefinitions>
     </ClCompile>

Python/bytecodes.c

@@ -2226,8 +2226,11 @@ dummy_func(
             }
             // we make no attempt to optimize here; specializations should
             // handle any case whose performance we care about
-            PyObject *stack[] = {class, self};
-            PyObject *super = PyObject_Vectorcall(global_super, stack, oparg & 2, NULL);
+            PyObject *super;
+            Py_BEGIN_LOCALS_MUST_NOT_ESCAPE();
+                PyObject *restrict stack[] = {class, self};
+                super = PyObject_Vectorcall(global_super, stack, oparg & 2, NULL);
+            Py_END_LOCALS_MUST_NOT_ESCAPE();
             if (opcode == INSTRUMENTED_LOAD_SUPER_ATTR) {
                 PyObject *arg = oparg & 2 ? class : &_PyInstrumentation_MISSING;
                 if (super == NULL) {
@@ -3509,10 +3512,13 @@ dummy_func(
             }
             assert(PyStackRef_IsTaggedInt(lasti));
             (void)lasti; // Shut up compiler warning if asserts are off
-            PyObject *stack[5] = {NULL, PyStackRef_AsPyObjectBorrow(exit_self), exc, val_o, tb};
+            PyObject* res_o;
+            Py_BEGIN_LOCALS_MUST_NOT_ESCAPE();
+            PyObject *restrict stack[5] = {NULL, PyStackRef_AsPyObjectBorrow(exit_self), exc, val_o, tb};
             int has_self = !PyStackRef_IsNull(exit_self);
-            PyObject *res_o = PyObject_Vectorcall(exit_func_o, stack + 2 - has_self,
+            res_o = PyObject_Vectorcall(exit_func_o, stack + 2 - has_self,
                     (3 + has_self) | PY_VECTORCALL_ARGUMENTS_OFFSET, NULL);
+            Py_END_LOCALS_MUST_NOT_ESCAPE();
             Py_XDECREF(original_tb);
             ERROR_IF(res_o == NULL);
             res = PyStackRef_FromPyObjectSteal(res_o);

Python/ceval_macros.h

@@ -85,9 +85,9 @@
 #   else
 #       define Py_MUSTTAIL [[clang::musttail]]
 #       define Py_PRESERVE_NONE_CC __attribute__((preserve_none))
-    Py_PRESERVE_NONE_CC typedef PyObject* (*py_tail_call_funcptr)(TAIL_CALL_PARAMS);
 #   endif
-#   define TARGET(op) Py_PRESERVE_NONE_CC PyObject *_TAIL_CALL_##op(TAIL_CALL_PARAMS)
+    typedef PyObject *(Py_PRESERVE_NONE_CC *py_tail_call_funcptr)(TAIL_CALL_PARAMS);
+#   define TARGET(op) Py_NO_INLINE PyObject *Py_PRESERVE_NONE_CC _TAIL_CALL_##op(TAIL_CALL_PARAMS)
 #   define DISPATCH_GOTO() \
         do { \
             Py_MUSTTAIL return (INSTRUCTION_TABLE[opcode])(TAIL_CALL_ARGS); \

Python/executor_cases.c.h

@@ -692,6 +692,8 @@ def has_error_without_pop(op: parser.CodeDef) -> bool:
     "PyStackRef_Wrap",
     "PyStackRef_Unwrap",
     "_PyLong_CheckExactAndCompact",
+    "Py_BEGIN_LOCALS_MUST_NOT_ESCAPE",
+    "Py_END_LOCALS_MUST_NOT_ESCAPE",
 )