pythongh-117224: Move some large uops to external functions

mdboom · Apr 5, 2024 · 6d74608 · 6d74608
1 parent d9cfe7e
commit 6d74608
Show file tree

Hide file tree

Showing 16 changed files with 451 additions and 189 deletions.
diff --git a/.gitattributes b/.gitattributes
@@ -73,6 +73,7 @@ Doc/data/stable_abi.dat                             generated
 Doc/library/token-list.inc                          generated
 Include/internal/pycore_ast.h                       generated
 Include/internal/pycore_ast_state.h                 generated
+Include/internal/pycore_executor_externals.h        generated
 Include/internal/pycore_opcode.h                    generated
 Include/internal/pycore_opcode_metadata.h           generated
 Include/internal/pycore_*_generated.h               generated
@@ -96,6 +97,7 @@ Parser/token.c                                      generated
 Programs/test_frozenmain.h                          generated
 Python/Python-ast.c                                 generated
 Python/executor_cases.c.h                           generated
+Python/executor_externals.c                         generated
 Python/generated_cases.c.h                          generated
 Python/optimizer_cases.c.h                          generated
 Python/opcode_targets.h                             generated

diff --git a/Include/internal/pycore_executor_externals.h b/Include/internal/pycore_executor_externals.h
diff --git a/Makefile.pre.in b/Makefile.pre.in
@@ -422,6 +422,7 @@ PYTHON_OBJS=	\
 		Python/crossinterp.o \
 		Python/dynamic_annotations.o \
 		Python/errors.o \
+		Python/executor_externals.o \
 		Python/flowgraph.o \
 		Python/frame.o \
 		Python/frozenmain.o \
@@ -1145,6 +1146,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_dict_state.h \
 		$(srcdir)/Include/internal/pycore_dtoa.h \
 		$(srcdir)/Include/internal/pycore_exceptions.h \
+		$(srcdir)/Include/internal/pycore_executor_externals.h \
 		$(srcdir)/Include/internal/pycore_faulthandler.h \
 		$(srcdir)/Include/internal/pycore_fileutils.h \
 		$(srcdir)/Include/internal/pycore_floatobject.h \
@@ -1905,8 +1907,10 @@ regen-generated-cases:
 .PHONY: regen-executor-cases
 regen-executor-cases:
 	$(PYTHON_FOR_REGEN) $(srcdir)/Tools/cases_generator/tier2_generator.py \
-	    -o $(srcdir)/Python/executor_cases.c.h.new $(srcdir)/Python/bytecodes.c
+	    -o $(srcdir)/Python/executor_cases.c.h.new $(srcdir)/Python/executor_externals.c.new $(srcdir)/Include/internal/pycore_executor_externals.h.new $(srcdir)/Python/bytecodes.c
 	$(UPDATE_FILE) $(srcdir)/Python/executor_cases.c.h $(srcdir)/Python/executor_cases.c.h.new
+	$(UPDATE_FILE) $(srcdir)/Python/executor_externals.c $(srcdir)/Python/executor_externals.c.new
+	$(UPDATE_FILE) $(srcdir)/Include/internal/pycore_executor_externals.h $(srcdir)/Include/internal/pycore_executor_externals.h.new
 
 .PHONY: regen-optimizer-cases
 regen-optimizer-cases:

diff --git a/PCbuild/_freeze_module.vcxproj b/PCbuild/_freeze_module.vcxproj
@@ -202,6 +202,7 @@
     <ClCompile Include="..\Python\dynamic_annotations.c" />
     <ClCompile Include="..\Python\dynload_win.c" />
     <ClCompile Include="..\Python\errors.c" />
+    <ClCompile Include="..\Python\executor_externals.c" />
     <ClCompile Include="..\Python\fileutils.c" />
     <ClCompile Include="..\Python\flowgraph.c" />
     <ClCompile Include="..\Python\formatter_unicode.c" />

diff --git a/PCbuild/_freeze_module.vcxproj.filters b/PCbuild/_freeze_module.vcxproj.filters
@@ -136,6 +136,7 @@
     <ClCompile Include="..\Objects\exceptions.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="..\Python\executor_externals.c" />
     <ClCompile Include="..\Modules\faulthandler.c">
       <Filter>Source Files</Filter>
     </ClCompile>

diff --git a/PCbuild/pythoncore.vcxproj b/PCbuild/pythoncore.vcxproj
@@ -226,6 +226,7 @@
     <ClInclude Include="..\Include\internal\pycore_dict_state.h" />
     <ClInclude Include="..\Include\internal\pycore_dtoa.h" />
     <ClInclude Include="..\Include\internal\pycore_exceptions.h" />
+    <ClInclude Include="..\Include\internal\pycore_executor_externals.h" />
     <ClInclude Include="..\Include\internal\pycore_faulthandler.h" />
     <ClInclude Include="..\Include\internal\pycore_fileutils.h" />
     <ClInclude Include="..\Include\internal\pycore_fileutils_windows.h" />
@@ -563,6 +564,7 @@
     <ClCompile Include="..\Python\dynamic_annotations.c" />
     <ClCompile Include="..\Python\dynload_win.c" />
     <ClCompile Include="..\Python\errors.c" />
+    <ClCompile Include="..\Python\executor_externals.c" />
     <ClCompile Include="..\Python\fileutils.c" />
     <ClCompile Include="..\Python\flowgraph.c" />
     <ClCompile Include="..\Python\formatter_unicode.c" />

diff --git a/PCbuild/pythoncore.vcxproj.filters b/PCbuild/pythoncore.vcxproj.filters
@@ -510,6 +510,9 @@
     <ClInclude Include="..\Include\cpython\initconfig.h">
       <Filter>Include\cpython</Filter>
     </ClInclude>
+    <ClInclude Include="..\Include\internal\pycore_executor_externals.h">
+      <Filter>Include\internal</Filter>
+    </ClInclude>
     <ClInclude Include="..\Include\internal\pycore_unicodeobject.h">
       <Filter>Include\internal</Filter>
     </ClInclude>
@@ -1283,6 +1286,9 @@
     <ClCompile Include="..\Python\errors.c">
       <Filter>Python</Filter>
     </ClCompile>
+    <ClCompile Include="..\Python\executor_externals.c">
+      <Filter>Python</Filter>
+    </ClCompile>
     <ClCompile Include="..\Python\fileutils.c">
       <Filter>Python</Filter>
     </ClCompile>

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
@@ -1571,7 +1571,7 @@ dummy_func(
             Py_XDECREF(oldobj);
         }
 
-        inst(COPY_FREE_VARS, (--)) {
+        externalize inst(COPY_FREE_VARS, (--)) {
             /* Copy closure variables to free variables */
             PyCodeObject *co = _PyFrame_GetCode(frame);
             assert(PyFunction_Check(frame->f_funcobj));
@@ -3129,7 +3129,7 @@ dummy_func(
             DEOPT_IF(Py_TYPE(callable) != &PyMethod_Type);
         }
 
-        op(_INIT_CALL_BOUND_METHOD_EXACT_ARGS, (callable, unused, unused[oparg] -- func, self, unused[oparg])) {
+        externalize op(_INIT_CALL_BOUND_METHOD_EXACT_ARGS, (callable, unused, unused[oparg] -- func, self, unused[oparg])) {
             STAT_INC(CALL, hit);
             self = Py_NewRef(((PyMethodObject *)callable)->im_self);
             stack_pointer[-1 - oparg] = self;  // Patch stack as it is used by _INIT_CALL_PY_EXACT_ARGS
@@ -3157,7 +3157,7 @@ dummy_func(
             DEOPT_IF(tstate->py_recursion_remaining <= 1);
         }
 
-        replicate(5) pure op(_INIT_CALL_PY_EXACT_ARGS, (callable, self_or_null, args[oparg] -- new_frame: _PyInterpreterFrame*)) {
+        externalize replicate(5) pure op(_INIT_CALL_PY_EXACT_ARGS, (callable, self_or_null, args[oparg] -- new_frame: _PyInterpreterFrame*)) {
             int has_self = (self_or_null != NULL);
             STAT_INC(CALL, hit);
             PyFunctionObject *func = (PyFunctionObject *)callable;
@@ -3538,7 +3538,7 @@ dummy_func(
             DISPATCH();
         }
 
-         op(_CALL_METHOD_DESCRIPTOR_O, (callable, self_or_null, args[oparg] -- res)) {
+        op(_CALL_METHOD_DESCRIPTOR_O, (callable, self_or_null, args[oparg] -- res)) {
             int total_args = oparg;
             if (self_or_null != NULL) {
                 args--;
@@ -3847,7 +3847,7 @@ dummy_func(
             func = (PyObject *)func_obj;
         }
 
-        inst(SET_FUNCTION_ATTRIBUTE, (attr, func -- func)) {
+        externalize inst(SET_FUNCTION_ATTRIBUTE, (attr, func -- func)) {
             assert(PyFunction_Check(func));
             PyFunctionObject *func_obj = (PyFunctionObject *)func;
             switch(oparg) {

diff --git a/Python/ceval.c b/Python/ceval.c
@@ -8,6 +8,7 @@
 #include "pycore_ceval.h"
 #include "pycore_code.h"
 #include "pycore_emscripten_signal.h"  // _Py_CHECK_EMSCRIPTEN_SIGNALS
+#include "pycore_executor_externals.h"
 #include "pycore_function.h"
 #include "pycore_instruments.h"
 #include "pycore_intrinsics.h"