From 7d83f7bcc484145596bae1ff015fed0762da345d Mon Sep 17 00:00:00 2001 From: Guido van Rossum Date: Tue, 30 Apr 2024 18:26:34 -0700 Subject: gh-118335: Configure Tier 2 interpreter at build time (#118339) The code for Tier 2 is now only compiled when configured with `--enable-experimental-jit[=yes|interpreter]`. We drop support for `PYTHON_UOPS` and -`Xuops`, but you can disable the interpreter or JIT at runtime by setting `PYTHON_JIT=0`. You can also build it without enabling it by default using `--enable-experimental-jit=yes-off`; enable with `PYTHON_JIT=1`. On Windows, the `build.bat` script supports `--experimental-jit`, `--experimental-jit-off`, `--experimental-interpreter`. In the C code, `_Py_JIT` is defined as before when the JIT is enabled; the new variable `_Py_TIER2` is defined when the JIT *or* the interpreter is enabled. It is actually a bitmask: 1: JIT; 2: default-off; 4: interpreter. --- Doc/whatsnew/3.13.rst | 30 +++++++++++++++++----- Include/internal/pycore_opcode_metadata.h | 4 +-- Include/internal/pycore_uop_metadata.h | 2 +- Lib/dis.py | 2 +- Lib/test/support/__init__.py | 8 +++--- Lib/test/test_capi/test_opt.py | 8 ++++++ Lib/test/test_monitoring.py | 10 +++++--- Lib/test/test_opcache.py | 2 ++ Lib/test/test_optimizer.py | 2 ++ Lib/test/test_weakref.py | 2 ++ .../2024-04-26-14-06-18.gh-issue-118335.SRFsxO.rst | 4 +++ Modules/_opcode.c | 6 +++++ Modules/_testinternalcapi.c | 13 +++++++++- Objects/codeobject.c | 6 +++++ Objects/object.c | 4 +++ PCbuild/_testinternalcapi.vcxproj | 6 +++++ PCbuild/build.bat | 11 ++++++-- PCbuild/pythoncore.vcxproj | 1 + Python/bytecodes.c | 6 +++++ Python/ceval.c | 5 +++- Python/generated_cases.c.h | 6 +++++ Python/instrumentation.c | 6 +++++ Python/optimizer.c | 4 +++ Python/optimizer_analysis.c | 4 +++ Python/optimizer_symbols.c | 3 +++ Python/pylifecycle.c | 28 +++++++++++--------- Python/pystate.c | 6 +++++ Python/sysmodule.c | 2 ++ Tools/cases_generator/analyzer.py | 1 + Tools/jit/README.md | 6 ++--- configure | 12 +++++++-- configure.ac | 13 ++++++++-- 32 files changed, 181 insertions(+), 42 deletions(-) create mode 100644 Misc/NEWS.d/next/Core and Builtins/2024-04-26-14-06-18.gh-issue-118335.SRFsxO.rst diff --git a/Doc/whatsnew/3.13.rst b/Doc/whatsnew/3.13.rst index a9a7e48..ee50eff 100644 --- a/Doc/whatsnew/3.13.rst +++ b/Doc/whatsnew/3.13.rst @@ -888,7 +888,7 @@ Experimental JIT Compiler ========================= When CPython is configured using the ``--enable-experimental-jit`` option, -a just-in-time compiler is added which can speed up some Python programs. +a just-in-time compiler is added which may speed up some Python programs. The internal architecture is roughly as follows. @@ -905,19 +905,35 @@ The internal architecture is roughly as follows. before it is interpreted or translated to machine code. * There is a Tier 2 interpreter, but it is mostly intended for debugging - the earlier stages of the optimization pipeline. If the JIT is not - enabled, the Tier 2 interpreter can be invoked by passing Python the - ``-X uops`` option or by setting the ``PYTHON_UOPS`` environment - variable to ``1``. + the earlier stages of the optimization pipeline. + The Tier 2 interpreter can be enabled by configuring Python + with ``--enable-experimental-jit=interpreter``. -* When the ``--enable-experimental-jit`` option is used, the optimized +* When the JIT is enabled, the optimized Tier 2 IR is translated to machine code, which is then executed. - This does not require additional runtime options. * The machine code translation process uses an architecture called *copy-and-patch*. It has no runtime dependencies, but there is a new build-time dependency on LLVM. +The ``--enable-experimental-jit`` flag has the following optional values: + +* ``no`` (default) -- Disable the entire Tier 2 and JIT pipeline. + +* ``yes`` (default if the flag is present without optional value) + -- Enable the JIT. To disable the JIT at runtime, + pass the environment variable ``PYTHON_JIT=0``. + +* ``yes-off`` -- Build the JIT but disable it by default. + To enable the JIT at runtime, pass the environment variable + ``PYTHON_JIT=1``. + +* ``interpreter`` -- Enable the Tier 2 interpreter but disable the JIT. + The interpreter can be disabled by running with + ``PYTHON_JIT=0``. + +(On Windows, use ``PCbuild/build.bat --enable-jit`` to enable the JIT.) + See :pep:`744` for more details. (JIT by Brandt Bucher, inspired by a paper by Haoran Xu and Fredrik Kjolstad. diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h index 2ccc548..808bade 100644 --- a/Include/internal/pycore_opcode_metadata.h +++ b/Include/internal/pycore_opcode_metadata.h @@ -981,7 +981,7 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[268] = { [BUILD_SLICE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG }, [BUILD_STRING] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG }, [BUILD_TUPLE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG }, - [CACHE] = { true, INSTR_FMT_IX, HAS_ESCAPES_FLAG }, + [CACHE] = { true, INSTR_FMT_IX, 0 }, [CALL] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG }, [CALL_ALLOC_AND_ENTER_INIT] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG }, [CALL_BOUND_METHOD_EXACT_ARGS] = { true, INSTR_FMT_IBC00, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_EXIT_FLAG }, @@ -1121,7 +1121,7 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[268] = { [PUSH_NULL] = { true, INSTR_FMT_IX, HAS_PURE_FLAG }, [RAISE_VARARGS] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG }, [RERAISE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG }, - [RESERVED] = { true, INSTR_FMT_IX, HAS_ESCAPES_FLAG }, + [RESERVED] = { true, INSTR_FMT_IX, 0 }, [RESUME] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_EVAL_BREAK_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG }, [RESUME_CHECK] = { true, INSTR_FMT_IX, HAS_DEOPT_FLAG }, [RETURN_CONST] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_CONST_FLAG }, diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h index a84212c..5697b13 100644 --- a/Include/internal/pycore_uop_metadata.h +++ b/Include/internal/pycore_uop_metadata.h @@ -249,7 +249,7 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = { [_COLD_EXIT] = HAS_ARG_FLAG | HAS_ESCAPES_FLAG, [_DYNAMIC_EXIT] = HAS_ARG_FLAG | HAS_ESCAPES_FLAG, [_START_EXECUTOR] = HAS_DEOPT_FLAG, - [_FATAL_ERROR] = HAS_ESCAPES_FLAG, + [_FATAL_ERROR] = 0, [_CHECK_VALIDITY_AND_SET_IP] = HAS_DEOPT_FLAG, [_DEOPT] = 0, [_SIDE_EXIT] = 0, diff --git a/Lib/dis.py b/Lib/dis.py index 111d624..76934eb 100644 --- a/Lib/dis.py +++ b/Lib/dis.py @@ -216,7 +216,7 @@ def _get_code_array(co, adaptive): if op == ENTER_EXECUTOR: try: ex = get_executor(co, i) - except ValueError: + except (ValueError, RuntimeError): ex = None if ex: diff --git a/Lib/test/support/__init__.py b/Lib/test/support/__init__.py index 15f6543..52573e6 100644 --- a/Lib/test/support/__init__.py +++ b/Lib/test/support/__init__.py @@ -2539,17 +2539,17 @@ Py_TRACE_REFS = hasattr(sys, 'getobjects') # Decorator to disable optimizer while a function run def without_optimizer(func): try: - import _testinternalcapi + from _testinternalcapi import get_optimizer, set_optimizer except ImportError: return func @functools.wraps(func) def wrapper(*args, **kwargs): - save_opt = _testinternalcapi.get_optimizer() + save_opt = get_optimizer() try: - _testinternalcapi.set_optimizer(None) + set_optimizer(None) return func(*args, **kwargs) finally: - _testinternalcapi.set_optimizer(save_opt) + set_optimizer(save_opt) return wrapper diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py index c798b34..6e5b626 100644 --- a/Lib/test/test_capi/test_opt.py +++ b/Lib/test/test_capi/test_opt.py @@ -34,6 +34,8 @@ def clear_executors(func): @requires_specialization +@unittest.skipUnless(hasattr(_testinternalcapi, "get_optimizer"), + "Requires optimizer infrastructure") class TestOptimizerAPI(unittest.TestCase): def test_new_counter_optimizer_dealloc(self): @@ -136,6 +138,8 @@ def get_opnames(ex): @requires_specialization +@unittest.skipUnless(hasattr(_testinternalcapi, "get_optimizer"), + "Requires optimizer infrastructure") class TestExecutorInvalidation(unittest.TestCase): def setUp(self): @@ -215,6 +219,8 @@ class TestExecutorInvalidation(unittest.TestCase): @requires_specialization +@unittest.skipUnless(hasattr(_testinternalcapi, "get_optimizer"), + "Requires optimizer infrastructure") @unittest.skipIf(os.getenv("PYTHON_UOPS_OPTIMIZE") == "0", "Needs uop optimizer to run.") class TestUops(unittest.TestCase): @@ -579,6 +585,8 @@ class TestUops(unittest.TestCase): @requires_specialization +@unittest.skipUnless(hasattr(_testinternalcapi, "get_optimizer"), + "Requires optimizer infrastructure") @unittest.skipIf(os.getenv("PYTHON_UOPS_OPTIMIZE") == "0", "Needs uop optimizer to run.") class TestUopsOptimization(unittest.TestCase): diff --git a/Lib/test/test_monitoring.py b/Lib/test/test_monitoring.py index 11c61bc..a9140d4 100644 --- a/Lib/test/test_monitoring.py +++ b/Lib/test/test_monitoring.py @@ -1831,15 +1831,17 @@ class TestOptimizer(MonitoringTestBase, unittest.TestCase): def setUp(self): _testinternalcapi = import_module("_testinternalcapi") - self.old_opt = _testinternalcapi.get_optimizer() - opt = _testinternalcapi.new_counter_optimizer() - _testinternalcapi.set_optimizer(opt) + if hasattr(_testinternalcapi, "get_optimizer"): + self.old_opt = _testinternalcapi.get_optimizer() + opt = _testinternalcapi.new_counter_optimizer() + _testinternalcapi.set_optimizer(opt) super(TestOptimizer, self).setUp() def tearDown(self): super(TestOptimizer, self).tearDown() import _testinternalcapi - _testinternalcapi.set_optimizer(self.old_opt) + if hasattr(_testinternalcapi, "get_optimizer"): + _testinternalcapi.set_optimizer(self.old_opt) def test_for_loop(self): def test_func(x): diff --git a/Lib/test/test_opcache.py b/Lib/test/test_opcache.py index f4e954f..92a3411 100644 --- a/Lib/test/test_opcache.py +++ b/Lib/test/test_opcache.py @@ -16,6 +16,8 @@ _testinternalcapi = import_module("_testinternalcapi") def disabling_optimizer(func): def wrapper(*args, **kwargs): + if not hasattr(_testinternalcapi, "get_optimizer"): + return func(*args, **kwargs) old_opt = _testinternalcapi.get_optimizer() _testinternalcapi.set_optimizer(None) try: diff --git a/Lib/test/test_optimizer.py b/Lib/test/test_optimizer.py index 899a450..fac4d1a 100644 --- a/Lib/test/test_optimizer.py +++ b/Lib/test/test_optimizer.py @@ -80,6 +80,8 @@ class TestRareEventCounters(unittest.TestCase): class TestOptimizerSymbols(unittest.TestCase): + @unittest.skipUnless(hasattr(_testinternalcapi, "uop_symbols_test"), + "requires _testinternalcapi.uop_symbols_test") def test_optimizer_symbols(self): _testinternalcapi.uop_symbols_test() diff --git a/Lib/test/test_weakref.py b/Lib/test/test_weakref.py index df90647..16da24d 100644 --- a/Lib/test/test_weakref.py +++ b/Lib/test/test_weakref.py @@ -17,6 +17,7 @@ from test.support import script_helper, ALWAYS_EQ, suppress_immortalization from test.support import gc_collect from test.support import import_helper from test.support import threading_helper +from test.support import is_wasi, Py_DEBUG # Used in ReferencesTestCase.test_ref_created_during_del() . ref_from_del = None @@ -960,6 +961,7 @@ class ReferencesTestCase(TestBase): self.assertEqual(hash(a), hash(42)) self.assertRaises(TypeError, hash, b) + @unittest.skipIf(is_wasi and Py_DEBUG, "requires deep stack") def test_trashcan_16602(self): # Issue #16602: when a weakref's target was part of a long # deallocation chain, the trashcan mechanism could delay clearing diff --git a/Misc/NEWS.d/next/Core and Builtins/2024-04-26-14-06-18.gh-issue-118335.SRFsxO.rst b/Misc/NEWS.d/next/Core and Builtins/2024-04-26-14-06-18.gh-issue-118335.SRFsxO.rst new file mode 100644 index 0000000..e13edbb --- /dev/null +++ b/Misc/NEWS.d/next/Core and Builtins/2024-04-26-14-06-18.gh-issue-118335.SRFsxO.rst @@ -0,0 +1,4 @@ +Change how to use the tier 2 interpreter. Instead of running Python with +``-X uops`` or setting the environment variable ``PYTHON_UOPS=1``, this +choice is now made at build time by configuring with +``--enable-experimental-jit=interpreter``. diff --git a/Modules/_opcode.c b/Modules/_opcode.c index 5350adb..85e0ffe 100644 --- a/Modules/_opcode.c +++ b/Modules/_opcode.c @@ -367,7 +367,13 @@ _opcode_get_executor_impl(PyObject *module, PyObject *code, int offset) Py_TYPE(code)->tp_name); return NULL; } +#ifdef _Py_TIER2 return (PyObject *)PyUnstable_GetExecutor((PyCodeObject *)code, offset); +#else + PyErr_Format(PyExc_RuntimeError, + "Executors are not available in this build"); + return NULL; +#endif } static PyMethodDef diff --git a/Modules/_testinternalcapi.c b/Modules/_testinternalcapi.c index 99e80ba..f7952a1 100644 --- a/Modules/_testinternalcapi.c +++ b/Modules/_testinternalcapi.c @@ -985,6 +985,8 @@ get_co_framesize(PyObject *self, PyObject *arg) return PyLong_FromLong(code->co_framesize); } +#ifdef _Py_TIER2 + static PyObject * new_counter_optimizer(PyObject *self, PyObject *arg) { @@ -1012,7 +1014,10 @@ set_optimizer(PyObject *self, PyObject *opt) static PyObject * get_optimizer(PyObject *self, PyObject *Py_UNUSED(ignored)) { - PyObject *opt = (PyObject *)PyUnstable_GetOptimizer(); + PyObject *opt = NULL; +#ifdef _Py_TIER2 + opt = (PyObject *)PyUnstable_GetOptimizer(); +#endif if (opt == NULL) { Py_RETURN_NONE; } @@ -1045,6 +1050,8 @@ invalidate_executors(PyObject *self, PyObject *obj) Py_RETURN_NONE; } +#endif + static int _pending_callback(void *arg) { /* we assume the argument is callable object to which we own a reference */ @@ -2020,12 +2027,14 @@ static PyMethodDef module_functions[] = { {"iframe_getline", iframe_getline, METH_O, NULL}, {"iframe_getlasti", iframe_getlasti, METH_O, NULL}, {"get_co_framesize", get_co_framesize, METH_O, NULL}, +#ifdef _Py_TIER2 {"get_optimizer", get_optimizer, METH_NOARGS, NULL}, {"set_optimizer", set_optimizer, METH_O, NULL}, {"new_counter_optimizer", new_counter_optimizer, METH_NOARGS, NULL}, {"new_uop_optimizer", new_uop_optimizer, METH_NOARGS, NULL}, {"add_executor_dependency", add_executor_dependency, METH_VARARGS, NULL}, {"invalidate_executors", invalidate_executors, METH_O, NULL}, +#endif {"pending_threadfunc", _PyCFunction_CAST(pending_threadfunc), METH_VARARGS | METH_KEYWORDS}, {"pending_identify", pending_identify, METH_VARARGS, NULL}, @@ -2072,7 +2081,9 @@ static PyMethodDef module_functions[] = { {"py_thread_id", get_py_thread_id, METH_NOARGS}, #endif {"set_immortalize_deferred", set_immortalize_deferred, METH_VARARGS}, +#ifdef _Py_TIER2 {"uop_symbols_test", _Py_uop_symbols_test, METH_NOARGS}, +#endif {NULL, NULL} /* sentinel */ }; diff --git a/Objects/codeobject.c b/Objects/codeobject.c index 0146329..605167c 100644 --- a/Objects/codeobject.c +++ b/Objects/codeobject.c @@ -1496,6 +1496,8 @@ PyCode_GetFreevars(PyCodeObject *code) return _PyCode_GetFreevars(code); } +#ifdef _Py_TIER2 + static void clear_executors(PyCodeObject *co) { @@ -1515,6 +1517,8 @@ _PyCode_Clear_Executors(PyCodeObject *code) clear_executors(code); } +#endif + static void deopt_code(PyCodeObject *code, _Py_CODEUNIT *instructions) { @@ -1739,9 +1743,11 @@ code_dealloc(PyCodeObject *co) PyMem_Free(co_extra); } +#ifdef _Py_TIER2 if (co->co_executors != NULL) { clear_executors(co); } +#endif Py_XDECREF(co->co_consts); Py_XDECREF(co->co_names); diff --git a/Objects/object.c b/Objects/object.c index 8d85693..45310a6 100644 --- a/Objects/object.c +++ b/Objects/object.c @@ -2281,9 +2281,11 @@ static PyTypeObject* static_types[] = { &_PyBufferWrapper_Type, &_PyContextTokenMissing_Type, &_PyCoroWrapper_Type, +#ifdef _Py_TIER2 &_PyCounterExecutor_Type, &_PyCounterOptimizer_Type, &_PyDefaultOptimizer_Type, +#endif &_Py_GenericAliasIterType, &_PyHamtItems_Type, &_PyHamtKeys_Type, @@ -2304,8 +2306,10 @@ static PyTypeObject* static_types[] = { &_PyPositionsIterator, &_PyUnicodeASCIIIter_Type, &_PyUnion_Type, +#ifdef _Py_TIER2 &_PyUOpExecutor_Type, &_PyUOpOptimizer_Type, +#endif &_PyWeakref_CallableProxyType, &_PyWeakref_ProxyType, &_PyWeakref_RefType, diff --git a/PCbuild/_testinternalcapi.vcxproj b/PCbuild/_testinternalcapi.vcxproj index a825cac..d4cd8ad 100644 --- a/PCbuild/_testinternalcapi.vcxproj +++ b/PCbuild/_testinternalcapi.vcxproj @@ -108,6 +108,12 @@ false + + + _Py_JIT;%(PreprocessorDefinitions) + _Py_TIER2=$(UseTIER2);%(PreprocessorDefinitions) + + diff --git a/PCbuild/build.bat b/PCbuild/build.bat index 83b50db..43a99aa 100644 --- a/PCbuild/build.bat +++ b/PCbuild/build.bat @@ -36,7 +36,9 @@ echo. overrides -c and -d echo. --disable-gil Enable experimental support for running without the GIL. echo. --test-marker Enable the test marker within the build. echo. --regen Regenerate all opcodes, grammar and tokens. -echo. --experimental-jit Enable the experimental just-in-time compiler. +echo. --experimental-jit Enable the experimental just-in-time compiler. +echo. --experimental-jit-off Ditto but off by default (PYTHON_JIT=1 enables). +echo. --experimental-interpreter Enable the experimental Tier 2 interpreter. echo. echo.Available flags to avoid building certain modules. echo.These flags have no effect if '-e' is not given: @@ -66,6 +68,7 @@ set verbose=/nologo /v:m /clp:summary set kill= set do_pgo= set pgo_job=-m test --pgo +set UseTIER2=0 :CheckOpts if "%~1"=="-h" goto Usage @@ -86,7 +89,10 @@ if "%~1"=="--disable-gil" (set UseDisableGil=true) & shift & goto CheckOpts if "%~1"=="--test-marker" (set UseTestMarker=true) & shift & goto CheckOpts if "%~1"=="-V" shift & goto Version if "%~1"=="--regen" (set Regen=true) & shift & goto CheckOpts -if "%~1"=="--experimental-jit" (set UseJIT=true) & shift & goto CheckOpts +if "%~1"=="--experimental-jit" (set UseJIT=true) & (set UseTIER2=1) & shift & goto CheckOpts +if "%~1"=="--experimental-jit-off" (set UseJIT=true) & (set UseTIER2=3) & shift & goto CheckOpts +if "%~1"=="--experimental-interpreter" (set UseTIER2=4) & shift & goto CheckOpts +if "%~1"=="--experimental-interpreter-off" (set UseTIER2=6) & shift & goto CheckOpts rem These use the actual property names used by MSBuild. We could just let rem them in through the environment, but we specify them on the command line rem anyway for visibility so set defaults after this @@ -179,6 +185,7 @@ echo on /p:DisableGil=%UseDisableGil%^ /p:UseTestMarker=%UseTestMarker% %GITProperty%^ /p:UseJIT=%UseJIT%^ + /p:UseTIER2=%UseTIER2%^ %1 %2 %3 %4 %5 %6 %7 %8 %9 @echo off diff --git a/PCbuild/pythoncore.vcxproj b/PCbuild/pythoncore.vcxproj index 4cb3e0d..a24667d 100644 --- a/PCbuild/pythoncore.vcxproj +++ b/PCbuild/pythoncore.vcxproj @@ -105,6 +105,7 @@ _USRDLL;Py_BUILD_CORE;Py_BUILD_CORE_BUILTIN;Py_ENABLE_SHARED;MS_DLL_ID="$(SysWinVer)";%(PreprocessorDefinitions) _Py_HAVE_ZLIB;%(PreprocessorDefinitions) _Py_JIT;%(PreprocessorDefinitions) + _Py_TIER2=$(UseTIER2);%(PreprocessorDefinitions) version.lib;ws2_32.lib;pathcch.lib;bcrypt.lib;%(AdditionalDependencies) diff --git a/Python/bytecodes.c b/Python/bytecodes.c index 18837ae..28766d6 100644 --- a/Python/bytecodes.c +++ b/Python/bytecodes.c @@ -2363,6 +2363,7 @@ dummy_func( CHECK_EVAL_BREAKER(); assert(oparg <= INSTR_OFFSET()); JUMPBY(-oparg); + #ifdef _Py_TIER2 #if ENABLE_SPECIALIZATION _Py_BackoffCounter counter = this_instr[1].counter; if (backoff_counter_triggers(counter) && this_instr->op.code == JUMP_BACKWARD) { @@ -2388,6 +2389,7 @@ dummy_func( ADVANCE_ADAPTIVE_COUNTER(this_instr[1].counter); } #endif /* ENABLE_SPECIALIZATION */ + #endif /* _Py_TIER2 */ } pseudo(JUMP) = { @@ -2401,6 +2403,7 @@ dummy_func( }; tier1 inst(ENTER_EXECUTOR, (--)) { + #ifdef _Py_TIER2 int prevoparg = oparg; CHECK_EVAL_BREAKER(); if (this_instr->op.code != ENTER_EXECUTOR || @@ -2418,6 +2421,9 @@ dummy_func( tstate->previous_executor = Py_None; Py_INCREF(executor); GOTO_TIER_TWO(executor); + #else + Py_FatalError("ENTER_EXECUTOR is not supported in this build"); + #endif /* _Py_TIER2 */ } replaced op(_POP_JUMP_IF_FALSE, (cond -- )) { diff --git a/Python/ceval.c b/Python/ceval.c index 8b23bc6..59498bc 100644 --- a/Python/ceval.c +++ b/Python/ceval.c @@ -755,7 +755,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int _Py_CODEUNIT *next_instr; PyObject **stack_pointer; -#ifndef _Py_JIT +#if defined(_Py_TIER2) && !defined(_Py_JIT) /* Tier 2 interpreter state */ _PyExecutorObject *current_executor = NULL; const _PyUOpInstruction *next_uop = NULL; @@ -959,6 +959,7 @@ resume_with_error: goto error; +#ifdef _Py_TIER2 // Tier 2 is also here! enter_tier_two: @@ -1113,6 +1114,8 @@ exit_to_trace: #endif // _Py_JIT +#endif // _Py_TIER2 + } #if defined(__GNUC__) diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h index 1444f5c..602cce4 100644 --- a/Python/generated_cases.c.h +++ b/Python/generated_cases.c.h @@ -2492,6 +2492,7 @@ (void)this_instr; next_instr += 1; INSTRUCTION_STATS(ENTER_EXECUTOR); + #ifdef _Py_TIER2 int prevoparg = oparg; CHECK_EVAL_BREAKER(); if (this_instr->op.code != ENTER_EXECUTOR || @@ -2508,6 +2509,9 @@ tstate->previous_executor = Py_None; Py_INCREF(executor); GOTO_TIER_TWO(executor); + #else + Py_FatalError("ENTER_EXECUTOR is not supported in this build"); + #endif /* _Py_TIER2 */ DISPATCH(); } @@ -3432,6 +3436,7 @@ CHECK_EVAL_BREAKER(); assert(oparg <= INSTR_OFFSET()); JUMPBY(-oparg); + #ifdef _Py_TIER2 #if ENABLE_SPECIALIZATION _Py_BackoffCounter counter = this_instr[1].counter; if (backoff_counter_triggers(counter) && this_instr->op.code == JUMP_BACKWARD) { @@ -3457,6 +3462,7 @@ ADVANCE_ADAPTIVE_COUNTER(this_instr[1].counter); } #endif /* ENABLE_SPECIALIZATION */ + #endif /* _Py_TIER2 */ DISPATCH(); } diff --git a/Python/instrumentation.c b/Python/instrumentation.c index ce97c3a..5614f48 100644 --- a/Python/instrumentation.c +++ b/Python/instrumentation.c @@ -1705,10 +1705,12 @@ instrument_lock_held(PyCodeObject *code, PyInterpreterState *interp) ); return 0; } +#ifdef _Py_TIER2 if (code->co_executors != NULL) { _PyCode_Clear_Executors(code); } _Py_Executors_InvalidateDependency(interp, code, 1); +#endif int code_len = (int)Py_SIZE(code); /* Exit early to avoid creating instrumentation * data for potential statically allocated code @@ -1946,7 +1948,9 @@ _PyMonitoring_SetEvents(int tool_id, _PyMonitoringEventSet events) goto done; } set_global_version(tstate, new_version); +#ifdef _Py_TIER2 _Py_Executors_InvalidateAll(interp, 1); +#endif res = instrument_all_executing_code_objects(interp); done: _PyEval_StartTheWorld(interp); @@ -1986,7 +1990,9 @@ _PyMonitoring_SetLocalEvents(PyCodeObject *code, int tool_id, _PyMonitoringEvent code->_co_instrumentation_version -= MONITORING_VERSION_INCREMENT; } +#ifdef _Py_TIER2 _Py_Executors_InvalidateDependency(interp, code, 1); +#endif res = instrument_lock_held(code, interp); diff --git a/Python/optimizer.c b/Python/optimizer.c index a9a35fc..a5e7430 100644 --- a/Python/optimizer.c +++ b/Python/optimizer.c @@ -1,3 +1,5 @@ +#ifdef _Py_TIER2 + #include "Python.h" #include "opcode.h" #include "pycore_interp.h" @@ -1622,3 +1624,5 @@ _Py_Executors_InvalidateAll(PyInterpreterState *interp, int is_invalidation) } } } + +#endif /* _Py_TIER2 */ diff --git a/Python/optimizer_analysis.c b/Python/optimizer_analysis.c index 9315d72..842b2e4 100644 --- a/Python/optimizer_analysis.c +++ b/Python/optimizer_analysis.c @@ -1,3 +1,5 @@ +#ifdef _Py_TIER2 + /* * This file contains the support code for CPython's uops optimizer. * It also performs some simple optimizations. @@ -603,3 +605,5 @@ _Py_uop_analyze_and_optimize( OPT_STAT_INC(optimizer_successes); return length; } + +#endif /* _Py_TIER2 */ diff --git a/Python/optimizer_symbols.c b/Python/optimizer_symbols.c index 204599b..d52f490 100644 --- a/Python/optimizer_symbols.c +++ b/Python/optimizer_symbols.c @@ -1,3 +1,4 @@ +#ifdef _Py_TIER2 #include "Python.h" @@ -506,3 +507,5 @@ fail: Py_XDECREF(val_43); return NULL; } + +#endif /* _Py_TIER2 */ diff --git a/Python/pylifecycle.c b/Python/pylifecycle.c index 790398e..7726ccc 100644 --- a/Python/pylifecycle.c +++ b/Python/pylifecycle.c @@ -624,9 +624,11 @@ static int builtins_dict_watcher(PyDict_WatchEvent event, PyObject *dict, PyObject *key, PyObject *new_value) { PyInterpreterState *interp = _PyInterpreterState_GET(); +#ifdef _Py_TIER2 if (interp->rare_events.builtin_dict < _Py_MAX_ALLOWED_BUILTINS_MODIFICATIONS) { _Py_Executors_InvalidateAll(interp, 1); } +#endif RARE_EVENT_INTERP_INC(interp, builtin_dict); return 0; } @@ -1272,30 +1274,30 @@ init_interp_main(PyThreadState *tstate) } // Turn on experimental tier 2 (uops-based) optimizer + // This is also needed when the JIT is enabled +#ifdef _Py_TIER2 if (is_main_interp) { -#ifndef _Py_JIT - // No JIT, maybe use the tier two interpreter: - char *envvar = Py_GETENV("PYTHON_UOPS"); - int enabled = envvar != NULL && *envvar > '0'; - if (_Py_get_xoption(&config->xoptions, L"uops") != NULL) { - enabled = 1; + int enabled = 1; +#if _Py_TIER2 & 2 + enabled = 0; +#endif + char *env = Py_GETENV("PYTHON_JIT"); + if (env && *env != '\0') { + // PYTHON_JIT=0|1 overrides the default + enabled = *env != '0'; } if (enabled) { -#else - // Always enable tier two for JIT builds (ignoring the environment - // variable and command-line option above): - if (true) { -#endif PyObject *opt = PyUnstable_Optimizer_NewUOpOptimizer(); if (opt == NULL) { return _PyStatus_ERR("can't initialize optimizer"); } if (PyUnstable_SetOptimizer((_PyOptimizerObject *)opt)) { - return _PyStatus_ERR("can't initialize optimizer"); + return _PyStatus_ERR("can't install optimizer"); } Py_DECREF(opt); } } +#endif if (!is_main_interp) { // The main interpreter is handled in Py_Main(), for now. @@ -1655,10 +1657,12 @@ finalize_modules(PyThreadState *tstate) { PyInterpreterState *interp = tstate->interp; +#ifdef _Py_TIER2 // Invalidate all executors and turn off tier 2 optimizer _Py_Executors_InvalidateAll(interp, 0); _PyOptimizerObject *old = _Py_SetOptimizer(interp, NULL); Py_XDECREF(old); +#endif // Stop watching __builtin__ modifications PyDict_Unwatch(0, interp->builtins); diff --git a/Python/pystate.c b/Python/pystate.c index 9d7b73b..3d6e76e 100644 --- a/Python/pystate.c +++ b/Python/pystate.c @@ -653,8 +653,10 @@ init_interpreter(PyInterpreterState *interp, } interp->sys_profile_initialized = false; interp->sys_trace_initialized = false; +#ifdef _Py_TIER2 (void)_Py_SetOptimizer(interp, NULL); interp->executor_list_head = NULL; +#endif if (interp != &runtime->_main_interpreter) { /* Fix the self-referential, statically initialized fields. */ interp->dtoa = (struct _dtoa_state)_dtoa_state_INIT(interp); @@ -806,9 +808,11 @@ interpreter_clear(PyInterpreterState *interp, PyThreadState *tstate) tstate->_status.cleared = 0; } +#ifdef _Py_TIER2 _PyOptimizerObject *old = _Py_SetOptimizer(interp, NULL); assert(old != NULL); Py_DECREF(old); +#endif /* It is possible that any of the objects below have a finalizer that runs Python code or otherwise relies on a thread state @@ -2821,9 +2825,11 @@ _PyInterpreterState_SetEvalFrameFunc(PyInterpreterState *interp, if (eval_frame == interp->eval_frame) { return; } +#ifdef _Py_TIER2 if (eval_frame != NULL) { _Py_Executors_InvalidateAll(interp, 1); } +#endif RARE_EVENT_INC(set_eval_frame_func); interp->eval_frame = eval_frame; } diff --git a/Python/sysmodule.c b/Python/sysmodule.c index 7af3636..7260515 100644 --- a/Python/sysmodule.c +++ b/Python/sysmodule.c @@ -2165,8 +2165,10 @@ static PyObject * sys__clear_internal_caches_impl(PyObject *module) /*[clinic end generated code: output=0ee128670a4966d6 input=253e741ca744f6e8]*/ { +#ifdef _Py_TIER2 PyInterpreterState *interp = _PyInterpreterState_GET(); _Py_Executors_InvalidateAll(interp, 0); +#endif PyType_ClearCache(); Py_RETURN_NONE; } diff --git a/Tools/cases_generator/analyzer.py b/Tools/cases_generator/analyzer.py index 18cefa0..fdb6354 100644 --- a/Tools/cases_generator/analyzer.py +++ b/Tools/cases_generator/analyzer.py @@ -411,6 +411,7 @@ NON_ESCAPING_FUNCTIONS = ( "PyCell_New", "PyFloat_AS_DOUBLE", "_PyFrame_PushUnchecked", + "Py_FatalError", ) ESCAPING_FUNCTIONS = ( diff --git a/Tools/jit/README.md b/Tools/jit/README.md index 7b33f99..0f5aa9c 100644 --- a/Tools/jit/README.md +++ b/Tools/jit/README.md @@ -7,18 +7,18 @@ This version of CPython can be built with an experimental just-in-time compiler. The JIT compiler does not require end users to install any third-party dependencies, but part of it must be *built* using LLVM[^why-llvm]. You are *not* required to build the rest of CPython using LLVM, or even the same version of LLVM (in fact, this is uncommon). -LLVM version 16 is required. Both `clang` and `llvm-readobj` need to be installed and discoverable (version suffixes, like `clang-16`, are okay). It's highly recommended that you also have `llvm-objdump` available, since this allows the build script to dump human-readable assembly for the generated code. +LLVM version 18 is required. Both `clang` and `llvm-readobj` need to be installed and discoverable (version suffixes, like `clang-18`, are okay). It's highly recommended that you also have `llvm-objdump` available, since this allows the build script to dump human-readable assembly for the generated code. It's easy to install all of the required tools: ### Linux -Install LLVM 16 on Ubuntu/Debian: +Install LLVM 18 on Ubuntu/Debian: ```sh wget https://apt.llvm.org/llvm.sh chmod +x llvm.sh -sudo ./llvm.sh 16 +sudo ./llvm.sh 18 ``` ### macOS diff --git a/configure b/configure index e1d0f36..01c00d2 100755 --- a/configure +++ b/configure @@ -1818,7 +1818,7 @@ Optional Features: --disable-gil enable experimental support for running without the GIL (default is no) --enable-pystats enable internal statistics gathering (default is no) - --enable-experimental-jit + --enable-experimental-jit[=no|yes|yes-off|interpreter] build the experimental just-in-time compiler (default is no) --enable-optimizations enable expensive, stable optimizations (PGO, etc.) @@ -8229,11 +8229,19 @@ else $as_nop enable_experimental_jit=no fi +case $enable_experimental_jit in + no) enable_experimental_jit=no ;; + yes) enable_experimental_jit="-D_Py_JIT -D_Py_TIER2=1" ;; + yes-off) enable_experimental_jit="-D_Py_JIT -D_Py_TIER2=3" ;; + interpreter) enable_experimental_jit="-D_Py_TIER2=4" ;; + interpreter-off) enable_experimental_jit="-D_Py_TIER2=6" ;; # Secret option + *) as_fn_error $? "invalid argument: --enable-experimental-jit=$enable_experimental_jit; expected no|yes|yes-off|interpreter" "$LINENO" 5 ;; +esac if test "x$enable_experimental_jit" = xno then : else $as_nop - as_fn_append CFLAGS_NODIST " -D_Py_JIT" + as_fn_append CFLAGS_NODIST " $enable_experimental_jit" REGEN_JIT_COMMAND="\$(PYTHON_FOR_REGEN) \$(srcdir)/Tools/jit/build.py $host" JIT_STENCILS_H="jit_stencils.h" if test "x$Py_DEBUG" = xtrue diff --git a/configure.ac b/configure.ac index 7681ea3..ae65efe 100644 --- a/configure.ac +++ b/configure.ac @@ -1768,14 +1768,23 @@ fi # Check for --enable-experimental-jit: AC_MSG_CHECKING([for --enable-experimental-jit]) AC_ARG_ENABLE([experimental-jit], - [AS_HELP_STRING([--enable-experimental-jit], + [AS_HELP_STRING([--enable-experimental-jit@<:@=no|yes|yes-off|interpreter@:>@], [build the experimental just-in-time compiler (default is no)])], [], [enable_experimental_jit=no]) +case $enable_experimental_jit in + no) enable_experimental_jit=no ;; + yes) enable_experimental_jit="-D_Py_JIT -D_Py_TIER2=1" ;; + yes-off) enable_experimental_jit="-D_Py_JIT -D_Py_TIER2=3" ;; + interpreter) enable_experimental_jit="-D_Py_TIER2=4" ;; + interpreter-off) enable_experimental_jit="-D_Py_TIER2=6" ;; # Secret option + *) AC_MSG_ERROR( + [invalid argument: --enable-experimental-jit=$enable_experimental_jit; expected no|yes|yes-off|interpreter]) ;; +esac AS_VAR_IF([enable_experimental_jit], [no], [], - [AS_VAR_APPEND([CFLAGS_NODIST], [" -D_Py_JIT"]) + [AS_VAR_APPEND([CFLAGS_NODIST], [" $enable_experimental_jit"]) AS_VAR_SET([REGEN_JIT_COMMAND], ["\$(PYTHON_FOR_REGEN) \$(srcdir)/Tools/jit/build.py $host"]) AS_VAR_SET([JIT_STENCILS_H], ["jit_stencils.h"]) -- cgit v0.12