gh-107557: Setup abstract interpretation

Include/internal/pycore_uops.h

@@ -8,7 +8,7 @@ extern "C" {
 #  error "this header requires Py_BUILD_CORE define"
 #endif
 
-#define _Py_UOP_MAX_TRACE_LENGTH 256
+#define _Py_UOP_MAX_TRACE_LENGTH 64
 
 typedef struct {
     int32_t opcode;

Lib/test/test_capi/test_misc.py

@@ -2618,7 +2618,7 @@ def testfunc(it):
             with self.assertRaises(StopIteration):
                 next(it)
 
-
+@unittest.skipIf(os.getenv("PYTHONUOPSOPTIMIZE") is None, "UOps optimization isn't enabled")
 class TestUopsOptimization(unittest.TestCase):
 
     def test_int_constant_propagation(self):
@@ -2643,5 +2643,6 @@ def testfunc(loops):
         binop_count = [opname for opname, _, _ in ex if opname == "_BINARY_OP_ADD_INT"]
         self.assertEqual(len(binop_count), 1)
 
+
 if __name__ == "__main__":
     unittest.main()

Makefile.pre.in

@@ -1553,10 +1553,12 @@ regen-cases:
 		-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
 	$(UPDATE_FILE) $(srcdir)/Python/generated_cases.c.h $(srcdir)/Python/generated_cases.c.h.new
 	$(UPDATE_FILE) $(srcdir)/Include/internal/pycore_opcode_metadata.h $(srcdir)/Include/internal/pycore_opcode_metadata.h.new
 	$(UPDATE_FILE) $(srcdir)/Python/executor_cases.c.h $(srcdir)/Python/executor_cases.c.h.new
+	$(UPDATE_FILE) $(srcdir)/Python/abstract_interp_cases.c.h $(srcdir)/Python/abstract_interp_cases.c.h.new
 	$(UPDATE_FILE) $(srcdir)/Lib/_opcode_metadata.py $(srcdir)/Lib/_opcode_metadata.py.new
 
 Python/compile.o: $(srcdir)/Include/internal/pycore_opcode_metadata.h

Python/bytecodes.c

@@ -3743,13 +3743,12 @@ dummy_func(
             return frame;
         }
 
-        op(INSERT, (--)) {
+        op(INSERT, (stuff[oparg], top -- top, stuff[oparg])) {
             // Inserts TOS at position specified by oparg
             PyObject *tos = TOP();
             for (int i = 1; i < oparg + 1; i++) {
                 stack_pointer[i] = stack_pointer[i - 1];
             }
-            POKE(oparg, tos);
         }
 
 

Python/optimizer.c

@@ -698,7 +698,10 @@ uop_optimize(
         return trace_length;
     }
     OBJECT_STAT_INC(optimization_traces_created);
-    trace_length = _Py_uop_analyze_and_optimize(code, trace, trace_length, curr_stackentries);
+    char *uop_optimize = Py_GETENV("PYTHONUOPSOPTIMIZE");
+    if (uop_optimize != NULL && *uop_optimize >= '0') {
+        trace_length = _Py_uop_analyze_and_optimize(code, trace, trace_length, curr_stackentries);
+    }
     _PyUOpExecutorObject *executor = PyObject_NewVar(_PyUOpExecutorObject, &UOpExecutor_Type, trace_length);
     if (executor == NULL) {
         return -1;

Python/optimizer_analysis.c

@@ -173,7 +173,7 @@ _Py_UOpsAbstractInterpContext_New(int stack_len, int locals_len, int curr_stackl
         return NULL;
     }
 
-    for (int i = 0; i < (locals_len + stack_len) * 2; i++) {
+    for (int i = 0; i < (locals_len + stack_len); i++) {
         locals_with_stack[i] = PARTITIONNODE_NULLROOT;
     }
 

Tools/cases_generator/generate_cases.py

@@ -61,7 +61,7 @@
 
 INSTR_FMT_PREFIX = "INSTR_FMT_"
 
-# @TODO generate all these after updating the DSL
+# TODO: generate all these after updating the DSL
 SPECIALLY_HANDLED_ABSTRACT_INSTR = {
     "LOAD_FAST",
     "LOAD_FAST_CHECK",
@@ -135,7 +135,7 @@ def effect_str(effects: list[StackEffect]) -> str:
         pushed: str | None
         match thing:
             case parsing.InstDef():
-                if thing.kind != "op" or (thing.kind != "inst" and self.instrs[thing.name].is_viable_uop()):
+                if thing.kind != "op" or self.instrs[thing.name].is_viable_uop():
                     instr = self.instrs[thing.name]
                     popped = effect_str(instr.input_effects)
                     pushed = effect_str(instr.output_effects)
@@ -641,17 +641,14 @@ def write_abstract_interpreter_instructions(
             for thing in self.everything:
                 match thing:
                     case OverriddenInstructionPlaceHolder():
-                        # TODO: Is this helpful?
-                        self.write_overridden_instr_place_holder(thing)
+                        pass
                     case parsing.InstDef():
                         instr = AbstractInstruction(self.instrs[thing.name].inst)
                         if instr.is_viable_uop() and instr.name not in SPECIALLY_HANDLED_ABSTRACT_INSTR:
                             self.out.emit("")
                             with self.out.block(f"case {thing.name}:"):
                                 instr.write(self.out, tier=TIER_TWO)
                                 self.out.emit("break;")
-                            # elif instr.kind != "op":
-                            #     print(f"NOTE: {thing.name} is not a viable uop")
                     case parsing.Macro():
                         pass
                     case parsing.Pseudo():

Tools/cases_generator/instructions.py

@@ -255,15 +255,6 @@ def __init__(self, inst: parsing.InstDef):
 
     def write(self, out: Formatter, tier: Tiers = TIER_ONE) -> None:
         """Write one abstract instruction, sans prologue and epilogue."""
-        # Write a static assertion that a family's cache size is correct
-        if family := self.family:
-            if self.name == family.name:
-                if cache_size := family.size:
-                    out.emit(
-                        f"static_assert({cache_size} == "
-                        f'{self.cache_offset}, "incorrect cache size");'
-                    )
-
         stacking.write_single_instr_for_abstract_interp(self, out)
 
     def write_body(