python · markshannon · Dec 8, 2023 · Nov 26, 2023 · Nov 26, 2023 · Nov 26, 2023
diff --git a/Include/opcode_ids.h b/Include/opcode_ids.h
@@ -1587,13 +1587,14 @@ regen-cases:
 	$(PYTHON_FOR_REGEN) \
 	    $(srcdir)/Tools/cases_generator/generate_cases.py \
 		$(CASESFLAG) \
-		-n $(srcdir)/Include/opcode_ids.h.new \
 		-t $(srcdir)/Python/opcode_targets.h.new \
 		-m $(srcdir)/Include/internal/pycore_opcode_metadata.h.new \
 		-e $(srcdir)/Python/executor_cases.c.h.new \
 		-p $(srcdir)/Lib/_opcode_metadata.py.new \
 		-a $(srcdir)/Python/abstract_interp_cases.c.h.new \
 		$(srcdir)/Python/bytecodes.c
+	$(PYTHON_FOR_REGEN) \
+	    $(srcdir)/Tools/cases_generator/opcode_id_generator.py -o $(srcdir)/Include/opcode_ids.h.new $(srcdir)/Python/bytecodes.c
 	$(PYTHON_FOR_REGEN) \
 	    $(srcdir)/Tools/cases_generator/tier1_generator.py -o $(srcdir)/Python/generated_cases.c.h.new $(srcdir)/Python/bytecodes.c
 	$(UPDATE_FILE) $(srcdir)/Python/generated_cases.c.h $(srcdir)/Python/generated_cases.c.h.new

@@ -48,8 +48,13 @@ def maybe_indent(self, txt: str) -> None:
             if offset <= self.indents[-1] or offset > 40:
                 offset = self.indents[-1] + 4
             self.indents.append(offset)
-        elif "{" in txt or is_label(txt):
+        if is_label(txt):
             self.indents.append(self.indents[-1] + 4)
+        elif "{" in txt:
+            if 'extern "C"' in txt:
+                self.indents.append(self.indents[-1])
+            else:
+                self.indents.append(self.indents[-1] + 4)
 
     def emit_text(self, txt: str) -> None:
         self.out.write(txt)

@@ -101,13 +101,6 @@
 arg_parser.add_argument(
     "-o", "--output", type=str, help="Generated code", default=DEFAULT_OUTPUT
 )
-arg_parser.add_argument(
-    "-n",
-    "--opcode_ids_h",
-    type=str,
-    help="Header file with opcode number definitions",
-    default=DEFAULT_OPCODE_IDS_H_OUTPUT,
-)
 arg_parser.add_argument(
     "-t",
     "--opcode_targets_h",
@@ -334,42 +327,8 @@ def map_op(op: int, name: str) -> None:
         self.opmap = opmap
         self.markers = markers
 
-    def write_opcode_ids(
-        self, opcode_ids_h_filename: str, opcode_targets_filename: str
-    ) -> None:
-        """Write header file that defined the opcode IDs"""
-
-        with open(opcode_ids_h_filename, "w") as f:
-            # Create formatter
-            self.out = Formatter(f, 0)
-
-            self.write_provenance_header()
-
-            self.out.emit("")
-            self.out.emit("#ifndef Py_OPCODE_IDS_H")
-            self.out.emit("#define Py_OPCODE_IDS_H")
-            self.out.emit("#ifdef __cplusplus")
-            self.out.emit('extern "C" {')
-            self.out.emit("#endif")
-            self.out.emit("")
-            self.out.emit("/* Instruction opcodes for compiled code */")
-
-            def define(name: str, opcode: int) -> None:
-                self.out.emit(f"#define {name:<38} {opcode:>3}")
-
-            all_pairs: list[tuple[int, int, str]] = []
-            # the second item in the tuple sorts the markers before the ops
-            all_pairs.extend((i, 1, name) for (name, i) in self.markers.items())
-            all_pairs.extend((i, 2, name) for (name, i) in self.opmap.items())
-            for i, _, name in sorted(all_pairs):
-                assert name is not None
-                define(name, i)
-
-            self.out.emit("")
-            self.out.emit("#ifdef __cplusplus")
-            self.out.emit("}")
-            self.out.emit("#endif")
-            self.out.emit("#endif /* !Py_OPCODE_IDS_H */")
+    def write_opcode_targets(self, opcode_targets_filename: str) -> None:
+        """Write header file that defines the jump target table"""
 
         with open(opcode_targets_filename, "w") as f:
             # Create formatter
@@ -885,7 +844,7 @@ def main() -> None:
     # These raise OSError if output can't be written
 
     a.assign_opcode_ids()
-    a.write_opcode_ids(args.opcode_ids_h, args.opcode_targets_h)
+    a.write_opcode_targets(args.opcode_targets_h)
     a.write_metadata(args.metadata, args.pymetadata)
     a.write_executor_instructions(args.executor_cases, args.emit_line_directives)
     a.write_abstract_interpreter_instructions(

@@ -0,0 +1,19 @@
+from pathlib import Path
+from typing import TextIO
+
+ROOT = Path(__file__).parent.parent.parent
+DEFAULT_INPUT = (ROOT / "Python/bytecodes.c").absolute()
+
+
+def root_relative_path(filename: str) -> str:
+    return Path(filename).relative_to(ROOT).as_posix()
+
+
+def write_header(generator: str, source: str, outfile: TextIO) -> None:
+    outfile.write(
+        f"""// This file is generated by {root_relative_path(generator)}
+// from:
+//   {source}
+// Do not edit!
+"""
+    )
@@ -0,0 +1,153 @@
+"""Generate the list of opcode IDs.
+Reads the instruction definitions from bytecodes.c.
+Writes the IDs to opcode._ids.h by default.
+"""
+
+import argparse
+import os.path
+import sys
+
+from analyzer import (
+    Analysis,
+    Instruction,
+    analyze_files,
+)
+from generators_common import (
+    DEFAULT_INPUT,
+    ROOT,
+    write_header,
+)
+from cwriter import CWriter
+from typing import TextIO
+
+
+DEFAULT_OUTPUT = ROOT / "Include/opcode_ids.h"
+
+
+def generate_opcode_header(filenames: str, analysis: Analysis, outfile: TextIO) -> None:
+    write_header(__file__, filenames, outfile)
+    out = CWriter(outfile, 0, False)
+    out.emit("\n")
+    instmap: dict[str, int] = {}
+
+    # 0 is reserved for cache entries. This helps debugging.
+    instmap["CACHE"] = 0
+
+    # 17 is reserved as it is the initial value for the specializing counter.
+    # This helps catch cases where we attempt to execute a cache.
+    instmap["RESERVED"] = 17
+
+    # 149 is RESUME - it is hard coded as such in Tools/build/deepfreeze.py
+    instmap["RESUME"] = 149
+    instmap["INSTRUMENTED_LINE"] = 254
+
+    instrumented = [
+        name for name in analysis.instructions if name.startswith("INSTRUMENTED")
+    ]
+
+    # Special case: this instruction is implemented in ceval.c
+    # rather than bytecodes.c, so we need to add it explicitly
+    # here (at least until we add something to bytecodes.c to
+    # declare external instructions).
+    instrumented.append("INSTRUMENTED_LINE")
+
+    specialized: set[str] = set()
+    no_arg: list[str] = []
+    has_arg: list[str] = []
+
+    for family in analysis.families.values():
+        specialized.update(inst.name for inst in family.members)
+
+    for inst in analysis.instructions.values():
+        name = inst.name
+        if name in specialized:
+            continue
+        if name in instrumented:
+            continue
+        if inst.properties.oparg:
+            has_arg.append(name)
+        else:
+            no_arg.append(name)
+
+    # Specialized ops appear in their own section
+    # Instrumented opcodes are at the end of the valid range
+    min_internal = 150
+    min_instrumented = 254 - (len(instrumented) - 1)
+    assert min_internal + len(specialized) < min_instrumented
+
+    next_opcode = 1
+
+    def add_instruction(name: str) -> None:
+        nonlocal next_opcode
+        if name in instmap:
+            return  # Pre-defined name
+        while next_opcode in instmap.values():
+            next_opcode += 1
+        instmap[name] = next_opcode
+        next_opcode += 1
+
+    for name in sorted(no_arg):
+        add_instruction(name)
+    for name in sorted(has_arg):
+        add_instruction(name)
+    # For compatibility
+    next_opcode = min_internal
+    for name in sorted(specialized):
+        add_instruction(name)
+    next_opcode = min_instrumented
+    for name in instrumented:
+        add_instruction(name)
+
+    for op, name in enumerate(sorted(analysis.pseudos), 256):
+        instmap[name] = op
+
+    assert 255 not in instmap.values()
+
+    out.emit(
+        """#ifndef Py_OPCODE_IDS_H
+#define Py_OPCODE_IDS_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Instruction opcodes for compiled code */
+"""
+    )
+
+    def write_define(name: str, op: int) -> None:
+        out.emit(f"#define {name:<38} {op:>3}\n")
+
+    for op, name in sorted([(op, name) for (name, op) in instmap.items()]):
+        write_define(name, op)
+
+    out.emit("\n")
+    write_define("HAVE_ARGUMENT", len(no_arg))
+    write_define("MIN_INSTRUMENTED_OPCODE", min_instrumented)
+
+    out.emit("\n")
+    out.emit("#ifdef __cplusplus\n")
+    out.emit("}\n")
+    out.emit("#endif\n")
+    out.emit("#endif /* !Py_OPCODE_IDS_H */\n")
+
+
+arg_parser = argparse.ArgumentParser(
+    description="Generate the header file with all opcode IDs.",
+    formatter_class=argparse.ArgumentDefaultsHelpFormatter,
+)
+
+arg_parser.add_argument(
+    "-o", "--output", type=str, help="Generated code", default=DEFAULT_OUTPUT
+)
+
+arg_parser.add_argument(
+    "input", nargs=argparse.REMAINDER, help="Instruction definition file(s)"
+)
+
+if __name__ == "__main__":
+    args = arg_parser.parse_args()
+    if len(args.input) == 0:
+        args.input.append(DEFAULT_INPUT)
+    data = analyze_files(args.input)
+    with open(args.output, "w") as outfile:
+        generate_opcode_header(args.input, data, outfile)
@@ -17,33 +17,18 @@
     StackItem,
     analysis_error,
 )
+from generators_common import (
+    DEFAULT_INPUT,
+    ROOT,
+    write_header,
+)
 from cwriter import CWriter
 from typing import TextIO, Iterator
 from lexer import Token
 from stack import StackOffset
 
 
-HERE = os.path.dirname(__file__)
-ROOT = os.path.join(HERE, "../..")
-THIS = os.path.relpath(__file__, ROOT).replace(os.path.sep, "/")
-
-DEFAULT_INPUT = os.path.relpath(os.path.join(ROOT, "Python/bytecodes.c"))
-DEFAULT_OUTPUT = os.path.relpath(os.path.join(ROOT, "Python/generated_cases.c.h"))
-
-
-def write_header(filename: str, outfile: TextIO) -> None:
-    outfile.write(
-        f"""// This file is generated by {THIS}
-// from:
-//   {filename}
-// Do not edit!
-
-#ifdef TIER_TWO
-    #error "This file is for Tier 1 only"
-#endif
-#define TIER_ONE 1
-"""
-    )
+DEFAULT_OUTPUT = ROOT / "Python/generated_cases.c.h"
 
 
 FOOTER = "#undef TIER_ONE\n"
@@ -351,7 +336,15 @@ def uses_this(inst: Instruction) -> bool:
 def generate_tier1(
     filenames: str, analysis: Analysis, outfile: TextIO, lines: bool
 ) -> None:
-    write_header(filenames, outfile)
+    write_header(__file__, filenames, outfile)
+    outfile.write(
+        """
+#ifdef TIER_TWO
+    #error "This file is for Tier 1 only"
+#endif
+#define TIER_ONE 1
+"""
+    )
     out = CWriter(outfile, 2, lines)
     out.emit("\n")
     for name, inst in sorted(analysis.instructions.items()):