GH-131798: Narrow types more aggressively in the JIT (GH-134373)

Author: brandtbucher

Include/internal/pycore_uop_ids.h

@@ -2137,6 +2137,25 @@ def testfunc(n):
         self.assertNotIn("_TO_BOOL_BOOL", uops)
         self.assertIn("_GUARD_IS_TRUE_POP", uops)
 
+    def test_set_type_version_sets_type(self):
+        class C:
+            A = 1
+
+        def testfunc(n):
+            x = 0
+            c = C()
+            for _ in range(n):
+                x += c.A  # Guarded.
+                x += type(c).A  # Unguarded!
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, 2 * TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_GUARD_TYPE_VERSION", uops)
+        self.assertNotIn("_CHECK_ATTR_CLASS", uops)
+
 
 def global_identity(x):
     return x

Include/internal/pycore_uop_metadata.h

@@ -0,0 +1 @@
+Improve the JIT's ability to narrow unknown classes to constant values.

Lib/test/test_capi/test_opt.py

@@ -344,6 +344,11 @@ dummy_func(
             PyStackRef_CLOSE(value);
         }
 
+        tier2 op(_POP_TWO, (nos, tos --)) {
+            PyStackRef_CLOSE(tos);
+            PyStackRef_CLOSE(nos);
+        }
+
         pure inst(PUSH_NULL, (-- res)) {
             res = PyStackRef_NULL;
         }

Misc/NEWS.d/next/Core_and_Builtins/2025-05-20-13-58-18.gh-issue-131798.hG8xBw.rst

@@ -523,6 +523,25 @@ optimize_uops(
 
 }
 
+const uint16_t op_without_push[MAX_UOP_ID + 1] = {
+    [_COPY] = _NOP,
+    [_LOAD_CONST_INLINE] = _NOP,
+    [_LOAD_CONST_INLINE_BORROW] = _NOP,
+    [_LOAD_FAST] = _NOP,
+    [_LOAD_FAST_BORROW] = _NOP,
+    [_LOAD_SMALL_INT] = _NOP,
+    [_POP_TOP_LOAD_CONST_INLINE] = _POP_TOP,
+    [_POP_TOP_LOAD_CONST_INLINE_BORROW] = _POP_TOP,
+    [_POP_TWO_LOAD_CONST_INLINE_BORROW] = _POP_TWO,
+};
+
+const uint16_t op_without_pop[MAX_UOP_ID + 1] = {
+    [_POP_TOP] = _NOP,
+    [_POP_TOP_LOAD_CONST_INLINE] = _LOAD_CONST_INLINE,
+    [_POP_TOP_LOAD_CONST_INLINE_BORROW] = _LOAD_CONST_INLINE_BORROW,
+    [_POP_TWO_LOAD_CONST_INLINE_BORROW] = _POP_TOP_LOAD_CONST_INLINE_BORROW,
+};
+
 
 static int
 remove_unneeded_uops(_PyUOpInstruction *buffer, int buffer_size)
@@ -551,50 +570,23 @@ remove_unneeded_uops(_PyUOpInstruction *buffer, int buffer_size)
                     buffer[pc].opcode = _NOP;
                 }
                 break;
-            case _POP_TOP:
-            case _POP_TOP_LOAD_CONST_INLINE:
-            case _POP_TOP_LOAD_CONST_INLINE_BORROW:
-            case _POP_TWO_LOAD_CONST_INLINE_BORROW:
-            optimize_pop_top_again:
+            default:
             {
-                _PyUOpInstruction *last = &buffer[pc-1];
-                while (last->opcode == _NOP) {
-                    last--;
-                }
-                switch (last->opcode) {
-                    case _POP_TWO_LOAD_CONST_INLINE_BORROW:
-                        last->opcode = _POP_TOP;
+                // Cancel out pushes and pops, repeatedly. So:
+                //     _LOAD_FAST + _POP_TWO_LOAD_CONST_INLINE_BORROW + _POP_TOP
+                // ...becomes:
+                //     _NOP + _POP_TOP + _NOP
+                while (op_without_pop[opcode]) {
+                    _PyUOpInstruction *last = &buffer[pc - 1];
+                    while (last->opcode == _NOP) {
+                        last--;
+                    }
+                    if (!op_without_push[last->opcode]) {
                         break;
-                    case _POP_TOP_LOAD_CONST_INLINE:
-                    case _POP_TOP_LOAD_CONST_INLINE_BORROW:
-                        last->opcode = _NOP;
-                        goto optimize_pop_top_again;
-                    case _COPY:
-                    case _LOAD_CONST_INLINE:
-                    case _LOAD_CONST_INLINE_BORROW:
-                    case _LOAD_FAST:
-                    case _LOAD_FAST_BORROW:
-                    case _LOAD_SMALL_INT:
-                        last->opcode = _NOP;
-                        if (opcode == _POP_TOP) {
-                            opcode = buffer[pc].opcode = _NOP;
-                        }
-                        else if (opcode == _POP_TOP_LOAD_CONST_INLINE) {
-                            opcode = buffer[pc].opcode = _LOAD_CONST_INLINE;
-                        }
-                        else if (opcode == _POP_TOP_LOAD_CONST_INLINE_BORROW) {
-                            opcode = buffer[pc].opcode = _LOAD_CONST_INLINE_BORROW;
-                        }
-                        else {
-                            assert(opcode == _POP_TWO_LOAD_CONST_INLINE_BORROW);
-                            opcode = buffer[pc].opcode = _POP_TOP_LOAD_CONST_INLINE_BORROW;
-                            goto optimize_pop_top_again;
-                        }
+                    }
+                    last->opcode = op_without_push[last->opcode];
+                    opcode = buffer[pc].opcode = op_without_pop[opcode];
                 }
-                _Py_FALLTHROUGH;
-            }
-            default:
-            {
                 /* _PUSH_FRAME doesn't escape or error, but it
                  * does need the IP for the return address */
                 bool needs_ip = opcode == _PUSH_FRAME;

Python/bytecodes.c

@@ -118,6 +118,18 @@ dummy_func(void) {
         sym_set_type(left, &PyLong_Type);
     }
 
+    op(_CHECK_ATTR_CLASS, (type_version/2, owner -- owner)) {
+        PyObject *type = (PyObject *)_PyType_LookupByVersion(type_version);
+        if (type) {
+            if (type == sym_get_const(ctx, owner)) {
+                REPLACE_OP(this_instr, _NOP, 0, 0);
+            }
+            else {
+                sym_set_const(owner, type);
+            }
+        }
+    }
+
     op(_GUARD_TYPE_VERSION, (type_version/2, owner -- owner)) {
         assert(type_version);
         if (sym_matches_type_version(owner, type_version)) {