python · Fidget-Spinner · Feb 13, 2024 · Feb 6, 2024 · Feb 6, 2024 · Feb 6, 2024
diff --git a/.gitattributes b/.gitattributes
@@ -94,6 +94,7 @@ Programs/test_frozenmain.h                          generated
 Python/Python-ast.c                                 generated
 Python/executor_cases.c.h                           generated
 Python/generated_cases.c.h                          generated
+Python/tier2_redundancy_eliminator_bytecodes.c.h    generated
 Python/opcode_targets.h                             generated
 Python/stdlib_module_names.h                        generated
 Tools/peg_generator/pegen/grammar_parser.py         generated

diff --git a/Include/cpython/pystats.h b/Include/cpython/pystats.h
@@ -120,6 +120,9 @@ typedef struct _optimization_stats {
     uint64_t trace_length_hist[_Py_UOP_HIST_SIZE];
     uint64_t trace_run_length_hist[_Py_UOP_HIST_SIZE];
     uint64_t optimized_trace_length_hist[_Py_UOP_HIST_SIZE];
+    uint64_t optimizer_attempts;
+    uint64_t optimizer_successes;
+    uint64_t optimizer_failure_reason_no_memory;
 } OptimizationStats;
 
 typedef struct _rare_event_stats {

diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h
diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h
@@ -8,6 +8,13 @@ extern "C" {
 #  error "this header requires Py_BUILD_CORE define"
 #endif
 
+#include "pycore_uop_ids.h"
+
+// This is the length of the trace we project initially.
+#define UOP_MAX_TRACE_LENGTH 512
+
+#define TRACE_STACK_SIZE 5
+
 int _Py_uop_analyze_and_optimize(_PyInterpreterFrame *frame,
     _PyUOpInstruction *trace, int trace_len, int curr_stackentries,
     _PyBloomFilter *dependencies);

diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h
@@ -16,7 +16,7 @@ extern const char * const _PyOpcode_uop_name[MAX_UOP_ID+1];
 
 #ifdef NEED_OPCODE_METADATA
 const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = {
-    [_NOP] = 0,
+    [_NOP] = HAS_PURE_FLAG,
     [_RESUME_CHECK] = HAS_DEOPT_FLAG,
     [_LOAD_FAST_CHECK] = HAS_ARG_FLAG | HAS_LOCAL_FLAG | HAS_ERROR_FLAG,
     [_LOAD_FAST] = HAS_ARG_FLAG | HAS_LOCAL_FLAG | HAS_PURE_FLAG,
@@ -202,10 +202,10 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = {
     [_SAVE_RETURN_OFFSET] = HAS_ARG_FLAG,
     [_EXIT_TRACE] = HAS_DEOPT_FLAG,
     [_CHECK_VALIDITY] = HAS_DEOPT_FLAG,
-    [_LOAD_CONST_INLINE] = 0,
-    [_LOAD_CONST_INLINE_BORROW] = 0,
-    [_LOAD_CONST_INLINE_WITH_NULL] = 0,
-    [_LOAD_CONST_INLINE_BORROW_WITH_NULL] = 0,
+    [_LOAD_CONST_INLINE] = HAS_PURE_FLAG,
+    [_LOAD_CONST_INLINE_BORROW] = HAS_PURE_FLAG,
+    [_LOAD_CONST_INLINE_WITH_NULL] = HAS_PURE_FLAG,
+    [_LOAD_CONST_INLINE_BORROW_WITH_NULL] = HAS_PURE_FLAG,
     [_CHECK_GLOBALS] = HAS_DEOPT_FLAG,
     [_CHECK_BUILTINS] = HAS_DEOPT_FLAG,
     [_INTERNAL_INCREMENT_OPT_COUNTER] = 0,

diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
@@ -3,6 +3,7 @@
 import sys
 import textwrap
 import unittest
+import gc
 
 import _testinternalcapi
 
@@ -556,6 +557,214 @@ def testfunc(n):
         # too much already.
         self.assertEqual(count, 1)
 
+class TestUopsOptimization(unittest.TestCase):
+
+    def test_int_type_propagation(self):
+        def testfunc(loops):
+            num = 0
+            while num < loops:
+                x = num + num
+                a = x + 1
+                num += 1
+            return a
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        res = None
+        with temporary_optimizer(opt):
+            res = testfunc(32)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        self.assertEqual(res, 63)
+        binop_count = [opname for opname, _, _ in ex if opname == "_BINARY_OP_ADD_INT"]
+        guard_both_int_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_INT"]
+        self.assertGreaterEqual(len(binop_count), 3)
+        self.assertLessEqual(len(guard_both_int_count), 1)
+
+    def test_int_type_propagation_through_frame(self):
+        def double(x):
+            return x + x
+        def testfunc(loops):
+            num = 0
+            while num < loops:
+                x = num + num
+                a = double(x)
+                num += 1
+            return a
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        res = None
+        with temporary_optimizer(opt):
+            res = testfunc(32)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        self.assertEqual(res, 124)
+        binop_count = [opname for opname, _, _ in ex if opname == "_BINARY_OP_ADD_INT"]
+        guard_both_int_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_INT"]
+        self.assertGreaterEqual(len(binop_count), 3)
+        self.assertLessEqual(len(guard_both_int_count), 1)
+
+    def test_int_type_propagation_from_frame(self):
+        def double(x):
+            return x + x
+        def testfunc(loops):
+            num = 0
+            while num < loops:
+                a = double(num)
+                x = a + a
+                num += 1
+            return x
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        res = None
+        with temporary_optimizer(opt):
+            res = testfunc(32)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        self.assertEqual(res, 124)
+        binop_count = [opname for opname, _, _ in ex if opname == "_BINARY_OP_ADD_INT"]
+        guard_both_int_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_INT"]
+        self.assertGreaterEqual(len(binop_count), 3)
+        self.assertLessEqual(len(guard_both_int_count), 1)
+
+    def test_int_impure_region(self):
+        def testfunc(loops):
+            num = 0
+            while num < loops:
+                x = num + num
+                y = 1
+                x // 2
+                a = x + y
+                num += 1
+            return a
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        res = None
+        with temporary_optimizer(opt):
+            res = testfunc(64)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        binop_count = [opname for opname, _, _ in ex if opname == "_BINARY_OP_ADD_INT"]
+        self.assertGreaterEqual(len(binop_count), 3)
+
+    def test_call_py_exact_args(self):
+        def testfunc(n):
+            def dummy(x):
+                return x+1
+            for i in range(n):
+                dummy(i)
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        with temporary_optimizer(opt):
+            testfunc(20)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        uops = {opname for opname, _, _ in ex}
+        self.assertIn("_PUSH_FRAME", uops)
+        self.assertIn("_BINARY_OP_ADD_INT", uops)
+        self.assertNotIn("_CHECK_PEP_523", uops)
+
+    def test_int_type_propagate_through_range(self):
+        def testfunc(n):
+
+            for i in range(n):
+                x = i + i
+            return x
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        with temporary_optimizer(opt):
+            res = testfunc(20)
+
+        ex = get_first_executor(testfunc)
+        self.assertEqual(res, 19 * 2)
+        self.assertIsNotNone(ex)
+        uops = {opname for opname, _, _ in ex}
+        self.assertNotIn("_GUARD_BOTH_INT", uops)
+
+    def test_int_value_numbering(self):
+        def testfunc(n):
+
+            y = 1
+            for i in range(n):
+                x = y
+                z = x
+                a = z
+                b = a
+                res = x + z + a + b
+            return res
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        with temporary_optimizer(opt):
+            res = testfunc(20)
+
+        ex = get_first_executor(testfunc)
+        self.assertEqual(res, 4)
+        self.assertIsNotNone(ex)
+        uops = {opname for opname, _, _ in ex}
+        self.assertIn("_GUARD_BOTH_INT", uops)
+        guard_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_INT"]
+        self.assertEqual(len(guard_count), 1)
+
+    def test_comprehension(self):
+        def testfunc(n):
+            for _ in range(n):
+                return [i for i in range(n)]
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        with temporary_optimizer(opt):
+            testfunc(20)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        uops = {opname for opname, _, _ in ex}
+        self.assertNotIn("_BINARY_OP_ADD_INT", uops)
+
+    def test_call_py_exact_args_disappearing(self):
+        def dummy(x):
+            return x+1
+
+        def testfunc(n):
+            for i in range(n):
+                dummy(i)
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        # Trigger specialization
+        testfunc(8)
+        with temporary_optimizer(opt):
+            del dummy
+            gc.collect()
+
+            def dummy(x):
+                return x + 2
+            testfunc(10)
+
+        ex = get_first_executor(testfunc)
+        # Honestly as long as it doesn't crash it's fine.
+        # Whether we get an executor or not is non-deterministic,
+        # because it's decided by when the function is freed.
+        # This test is a little implementation specific.
+
+    def test_promote_globals_to_constants(self):
+        def testfunc(n):
+            for i in range(n):
+                x = range(i)
+            return x
+
+        opt = _testinternalcapi.get_uop_optimizer()
+        with temporary_optimizer(opt):
+            testfunc(20)
+
+        ex = get_first_executor(testfunc)
+        self.assertIsNotNone(ex)
+        uops = {opname for opname, _, _ in ex}
+        self.assertNotIn("_LOAD_GLOBAL_BUILTIN", uops)
+        self.assertIn("_LOAD_CONST_INLINE_BORROW_WITH_NULL", uops)
+
+
 
 if __name__ == "__main__":
     unittest.main()