Address review partially

Author: Fidget-Spinner

Misc/NEWS.d/next/Core_and_Builtins/2025-01-10-18-56-20.gh-issue-128563.baDvls.rst

@@ -1 +1 @@
-A new type of interpreter has been added to CPython. This interpreter uses tail calls for its instruction handlers. Preliminary benchmark results suggest 7-11% geometric mean faster on pyperformance (depending on platform), and up to 30% faster on Python-intensive workloads. This interpreter currently only works on newer compilers, such as ``clang-19``. Other compilers will continue using the old interpreter. Patch by Ken Jin, with ideas by Mark Shannon, Garret Gu, Haoran Xu, and Josh Haberman.
+A new type of interpreter has been added to CPython. This interpreter uses tail calls for its instruction handlers. Preliminary benchmark results suggest 7-11% geometric mean faster on pyperformance (depending on platform), and up to 30% faster on Python-intensive workloads. This interpreter currently only works on newer compilers, such as ``clang-19``. Other compilers will continue using the old interpreter. Patch by Ken Jin, with ideas on how to implement this in CPython by Mark Shannon, Garret Gu, Haoran Xu, and Josh Haberman.

Python/ceval_macros.h

@@ -282,24 +282,6 @@ GETITEM(PyObject *v, Py_ssize_t i) {
 #define UPDATE_MISS_STATS(INSTNAME) ((void)0)
 #endif
 
-#ifdef Py_TAIL_CALL_INTERP
-#   define GO_TO_INSTRUCTION_IF(COND, INSTNAME, SIZE)           \
-        if ((COND)) {                                           \
-            /* This is only a single jump on release builds! */ \
-            UPDATE_MISS_STATS((INSTNAME));                      \
-            assert(_PyOpcode_Deopt[opcode] == (INSTNAME));      \
-            Py_MUSTTAIL                                         \
-            return (INSTRUCTION_TABLE[INSTNAME])(frame, stack_pointer, tstate, next_instr - 1 - SIZE, opcode, oparg); \
-        }
-#else
-#   define GO_TO_INSTRUCTION_IF(COND, INSTNAME, SIZE)           \
-        if ((COND)) {                                           \
-            /* This is only a single jump on release builds! */ \
-            UPDATE_MISS_STATS((INSTNAME));                      \
-            assert(_PyOpcode_Deopt[opcode] == (INSTNAME));      \
-            goto PREDICTED_##INSTNAME;                          \
-        }
-#endif
 
 // Try to lock an object in the free threading build, if it's not already
 // locked. Use with a DEOPT_IF() to deopt if the object is already locked.

Python/generated_cases.c.h

@@ -150,20 +150,21 @@ def deopt_if(
         storage: Storage,
         inst: Instruction | None,
     ) -> bool:
-        self.out.emit_at("GO_TO_INSTRUCTION_IF", tkn)
+        self.out.start_line()
+        self.out.emit("if (")
         lparen = next(tkn_iter)
-        self.emit(lparen)
         assert lparen.kind == "LPAREN"
         first_tkn = tkn_iter.peek()
         emit_to(self.out, tkn_iter, "RPAREN")
+        self.emit(") {\n")
         next(tkn_iter)  # Semi colon
-        self.out.emit(", ")
         assert inst is not None
         assert inst.family is not None
-        self.out.emit(inst.family.name)
-        self.out.emit(", ")
-        self.out.emit(inst.family.size)
-        self.out.emit(");\n")
+        family_name = inst.family.name
+        self.emit(f"UPDATE_MISS_STATS({family_name});\n")
+        self.emit(f"assert(_PyOpcode_Deopt[opcode] == ({family_name}));\n")
+        self.emit(f"goto PREDICTED_{family_name};\n")
+        self.emit("}\n")
         return not always_true(first_tkn)
 
     exit_if = deopt_if

Python/generated_tail_call_handlers.c.h

@@ -129,6 +129,9 @@ def uses_this(inst: Instruction) -> bool:
         for cache in uop.caches:
             if cache.name != "unused":
                 return True
+        for tkn in uop.body:
+            if tkn.kind == "IDENTIFIER" and (tkn.text == "DEOPT_IF" or tkn.text == "EXIT_IF"):
+                return True
     return False
 
 

Tools/cases_generator/generators_common.py

@@ -14,6 +14,8 @@
     CWriter,
     Emitter,
     TokenIterator,
+    emit_to,
+    always_true,
 )
 
 from analyzer import (
@@ -67,6 +69,34 @@ def go_to_instruction(
         self.emit(f"Py_MUSTTAIL return (INSTRUCTION_TABLE[{name.text}])(frame, stack_pointer, tstate, next_instr - 1 - {size}, opcode, oparg);\n")
         return True
 
+    def deopt_if(
+        self,
+        tkn: Token,
+        tkn_iter: TokenIterator,
+        uop: Uop,
+        storage: Storage,
+        inst: Instruction | None,
+    ) -> bool:
+        self.out.start_line()
+        self.out.emit("if (")
+        lparen = next(tkn_iter)
+        assert lparen.kind == "LPAREN"
+        first_tkn = tkn_iter.peek()
+        emit_to(self.out, tkn_iter, "RPAREN")
+        self.emit(") {\n")
+        next(tkn_iter)  # Semi colon
+        assert inst is not None
+        assert inst.family is not None
+        family_name = inst.family.name
+        self.emit(f"UPDATE_MISS_STATS({family_name});\n")
+        self.emit(f"assert(_PyOpcode_Deopt[opcode] == ({family_name}));\n")
+        self.emit(f"Py_MUSTTAIL return _TAIL_CALL_{family_name}(frame, stack_pointer, tstate, this_instr, opcode, oparg);\n")
+        self.emit("}\n")
+        return not always_true(first_tkn)
+
+
+    exit_if = deopt_if
+
 class TailCallLabelsEmitter(Emitter):
     def __init__(self, out: CWriter):
         super().__init__(out)