Feat: Merge Brandt's Copy and Patch compiler (python#52)

Co-authored-by: Ken Jin <[email protected]>
Co-authored-by: Brandt Bucher <[email protected]>

.github/workflows/jit.yml

@@ -0,0 +1,121 @@
+name: JIT
+on: push
+jobs:
+  jit:
+    name: ${{ matrix.target }} (${{ matrix.debug && 'Debug' || 'Release' }}, LLVM ${{ matrix.llvm }})
+    runs-on: ${{ matrix.runner }}
+    strategy:
+      fail-fast: false
+      matrix:
+        target:
+          # - i686-pc-windows-msvc/msvc
+          - x86_64-pc-windows-msvc/msvc
+          # - x86_64-apple-darwin/clang
+          # - x86_64-unknown-linux-gnu/gcc
+          # - aarch64-apple-darwin/clang
+          # - aarch64-unknown-linux-gnu/gcc
+          # - aarch64-unknown-linux-gnu/clang
+          # - powerpc64le-unknown-linux-gnu/gcc
+          # - x86_64-unknown-linux-gnu/clang
+        debug:
+          - true
+          # - false
+        llvm:
+          #- 14
+          #- 15
+          - 16
+        include:
+          # - target: i686-pc-windows-msvc/msvc
+          #   architecture: i686
+          #   runner: windows-latest
+          #   compiler: msvc
+          #   tier: 1
+          #   windows_platform: Win32
+          - target: x86_64-pc-windows-msvc/msvc
+            architecture: x86_64
+            runner: windows-latest
+            compiler: msvc
+            tier: 1
+            windows_platform: x64
+          # - target: x86_64-apple-darwin/clang
+          #   architecture: x86_64
+          #   runner: macos-latest
+          #   compiler: clang
+          #   tier: 1
+          # - target: x86_64-unknown-linux-gnu/gcc
+          #   architecture: x86_64
+          #   runner: ubuntu-latest
+          #   compiler: gcc
+          #   tier: 1
+          # - target: aarch64-apple-darwin/clang
+          #   architecture: aarch64
+          #   runner: macos-latest
+          #   compiler: clang
+          #   tier: 2
+          # - target: aarch64-unknown-linux-gnu/gcc
+          #   architecture: aarch64
+          #   runner: ubuntu-latest
+          #   compiler: gcc
+          #   tier: 2
+          # - target: aarch64-unknown-linux-gnu/clang
+          #   architecture: aarch64
+          #   runner: ubuntu-latest
+          #   compiler: clang
+          #   tier: 2
+          # - target: powerpc64le-unknown-linux-gnu/gcc
+          #   architecture: ppc64le
+          #   runner: ubuntu-latest
+          #   compiler: gcc
+          #   tier: 2
+          # - target: x86_64-unknown-linux-gnu/clang
+          #   architecture: x86_64
+          #   runner: ubuntu-latest
+          #   compiler: clang
+          #   tier: 2
+    env:
+      CC: ${{ matrix.compiler }}
+      PYTHON_LLVM_VERSION: ${{ matrix.llvm }}
+    steps:
+      - uses: actions/checkout@v3
+      - uses: actions/setup-python@v4
+        with:
+          python-version: '3.10'
+      # Only support Window x64 ~ Jules
+      # - name: macOS
+      #   if: runner.os == 'macOS'
+      #   run: |
+      #     brew install llvm@${{ matrix.llvm }}
+      #     export SDKROOT="$(xcrun --show-sdk-path)"
+      #     ./configure ${{ matrix.debug && '--with-pydebug' || '--enable-optimizations --with-lto' }}
+      #     make
+      #     ./python.exe -m test -j0 -wW
+      # - name: Native Linux
+      #   if: runner.os == 'Linux' && matrix.architecture == 'x86_64'
+      #   run: |
+      #     sudo bash -c "$(wget -O - https://apt.llvm.org/llvm.sh)" ./llvm.sh ${{ matrix.llvm }}
+      #     ${{ matrix.llvm == 14 && 'sudo apt install libclang-rt-14-dev' || '' }}
+      #     export PATH="$(llvm-config-${{ matrix.llvm }} --bindir):$PATH"
+      #     ./configure ${{ matrix.debug && '--with-pydebug' || '--enable-optimizations --with-lto' }}
+      #     make
+      #     ./python -m test -j0 -wW
+      # - name: Emulated Linux
+      #   if: runner.os == 'Linux' && matrix.architecture != 'x86_64'
+      #   uses: uraimo/run-on-arch-action@v2
+      #   with:
+      #     arch: ${{ matrix.architecture }}
+      #     distro: ubuntu_latest
+      #     run: |
+      #       sudo bash -c "$(wget -O - https://apt.llvm.org/llvm.sh)" ./llvm.sh ${{ matrix.llvm }}
+      #       ${{ matrix.llvm == 14 && 'sudo apt install libclang-rt-14-dev' || '' }}
+      #       export PATH="$(llvm-config-${{ matrix.llvm }} --bindir):$PATH"
+      #       ./configure ${{ matrix.debug && '--with-pydebug' || '--enable-optimizations --with-lto' }}
+      #       make
+      #       ./python -m test -j0 -wW
+      - name: Windows
+        if: runner.os == 'Windows'
+        run: |
+          choco install llvm --allow-downgrade --version ${{ matrix.llvm }}
+          ./PCbuild/build ${{ matrix.debug && '-d' || '--pgo' }} -p ${{ matrix.windows_platform }}
+      # ./PCbuild/rt${{ matrix.debug && ' -d' || '' }} -p ${{ matrix.windows_platform }} -q -j0 -wW          
+      - name: Tests
+        run: .\python.bat tier2_test.py

.gitignore

@@ -140,6 +140,7 @@ Tools/msi/obj
 Tools/ssl/amd64
 Tools/ssl/win32
 Tools/freeze/test/outdir
+Python/jit_stencils.h
 
 # The frozen modules are always generated by the build so we don't
 # keep them in the repo.  Also see Tools/build/freeze_modules.py.

Include/internal/pycore_abstract.h

@@ -16,8 +16,8 @@ _PyIndex_Check(PyObject *obj)
     return (tp_as_number != NULL && tp_as_number->nb_index != NULL);
 }
 
-PyObject *_PyNumber_PowerNoMod(PyObject *lhs, PyObject *rhs);
-PyObject *_PyNumber_InPlacePowerNoMod(PyObject *lhs, PyObject *rhs);
+PyAPI_FUNC(PyObject *)_PyNumber_PowerNoMod(PyObject *lhs, PyObject *rhs);
+PyAPI_FUNC(PyObject *)_PyNumber_InPlacePowerNoMod(PyObject *lhs, PyObject *rhs);
 
 #ifdef __cplusplus
 }

Include/internal/pycore_ceval.h

@@ -18,6 +18,7 @@ struct _ceval_runtime_state;
 
 #include "pycore_interp.h"        // PyInterpreterState.eval_frame
 #include "pycore_pystate.h"       // _PyThreadState_GET()
+#include "pycore_frame.h"
 
 
 extern void _Py_FinishPendingCalls(PyThreadState *tstate);
@@ -124,9 +125,18 @@ PyAPI_FUNC(int) _Py_CheckRecursiveCall(
     PyThreadState *tstate,
     const char *where);
 
-int _Py_CheckRecursiveCallPy(
+PyAPI_FUNC(int) _Py_CheckRecursiveCallPy(
     PyThreadState *tstate);
 
+static inline int _Py_EnterRecursivePy(PyThreadState *tstate) {
+    return (tstate->py_recursion_remaining-- <= 0) &&
+        _Py_CheckRecursiveCallPy(tstate);
+}
+
+static inline void _Py_LeaveRecursiveCallPy(PyThreadState *tstate)  {
+    tstate->py_recursion_remaining++;
+}
+
 static inline int _Py_EnterRecursiveCallTstate(PyThreadState *tstate,
                                                const char *where) {
     return (_Py_MakeRecCheck(tstate) && _Py_CheckRecursiveCall(tstate, where));
@@ -148,10 +158,14 @@ static inline void _Py_LeaveRecursiveCall(void)  {
 
 extern struct _PyInterpreterFrame* _PyEval_GetFrame(void);
 
-extern PyObject* _Py_MakeCoro(PyFunctionObject *func);
+PyAPI_FUNC(PyObject *)_Py_MakeCoro(PyFunctionObject *func);
+
+PyAPI_FUNC(int) _Py_HandlePending(PyThreadState *tstate);
 
-extern int _Py_HandlePending(PyThreadState *tstate);
+PyAPI_FUNC(void) _PyEvalFrameClearAndPop(PyThreadState *tstate, _PyInterpreterFrame *frame);
+PyAPI_FUNC(_PyInterpreterFrame *)_PyEvalFramePushAndInit(PyThreadState *tstate, PyFunctionObject *func, PyObject *locals, PyObject* const* args, size_t argcount, PyObject *kwnames);
 
+PyAPI_FUNC(PyObject *)trace_call_function(PyThreadState *tstate, PyObject *func, PyObject **args, Py_ssize_t nargs, PyObject *kwnames);
 
 
 #ifdef __cplusplus

Include/internal/pycore_code.h

@@ -238,26 +238,26 @@ extern int _PyLineTable_PreviousAddressRange(PyCodeAddressRange *range);
 
 /* Specialization functions */
 
-extern void _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr,
+PyAPI_FUNC(void) _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr,
                                     PyObject *name);
-extern void _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr,
+PyAPI_FUNC(void) _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr,
                                      PyObject *name);
-extern void _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins,
+PyAPI_FUNC(void) _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins,
                                       _Py_CODEUNIT *instr, PyObject *name);
-extern void _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container,
+PyAPI_FUNC(void) _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container,
                                         _Py_CODEUNIT *instr);
-extern void _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub,
+PyAPI_FUNC(void) _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub,
                                        _Py_CODEUNIT *instr);
-extern void _Py_Specialize_Call(PyObject *callable, _Py_CODEUNIT *instr,
+PyAPI_FUNC(void) _Py_Specialize_Call(PyObject *callable, _Py_CODEUNIT *instr,
                                 int nargs, PyObject *kwnames);
-extern void _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
+PyAPI_FUNC(void) _Py_Specialize_BinaryOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
                                     int oparg, PyObject **locals);
-extern void _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs,
+PyAPI_FUNC(void) _Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs,
                                      _Py_CODEUNIT *instr, int oparg);
-extern void _Py_Specialize_UnpackSequence(PyObject *seq, _Py_CODEUNIT *instr,
+PyAPI_FUNC(void) _Py_Specialize_UnpackSequence(PyObject *seq, _Py_CODEUNIT *instr,
                                           int oparg);
-extern void _Py_Specialize_ForIter(PyObject *iter, _Py_CODEUNIT *instr, int oparg, char is_bb);
-extern void _Py_Specialize_Send(PyObject *receiver, _Py_CODEUNIT *instr);
+PyAPI_FUNC(void) _Py_Specialize_ForIter(PyObject *iter, _Py_CODEUNIT *instr, int oparg, char is_bb);
+PyAPI_FUNC(void) _Py_Specialize_Send(PyObject *receiver, _Py_CODEUNIT *instr);
 
 /* Finalizer function for static codeobjects used in deepfreeze.py */
 extern void _PyStaticCode_Fini(PyCodeObject *co);

Include/internal/pycore_dict.h

@@ -52,7 +52,7 @@ extern Py_ssize_t _PyDictKeys_StringLookup(PyDictKeysObject* dictkeys, PyObject
 extern PyObject *_PyDict_LoadGlobal(PyDictObject *, PyDictObject *, PyObject *);
 
 /* Consumes references to key and value */
-extern int _PyDict_SetItem_Take2(PyDictObject *op, PyObject *key, PyObject *value);
+PyAPI_FUNC(int) _PyDict_SetItem_Take2(PyDictObject *op, PyObject *key, PyObject *value);
 extern int _PyObjectDict_SetItem(PyTypeObject *tp, PyObject **dictptr, PyObject *name, PyObject *value);
 
 extern PyObject *_PyDict_Pop_KnownHash(PyObject *, PyObject *, Py_hash_t, PyObject *);
@@ -176,7 +176,7 @@ _PyDict_NotifyEvent(PyInterpreterState *interp,
 }
 
 extern PyObject *_PyObject_MakeDictFromInstanceAttributes(PyObject *obj, PyDictValues *values);
-extern PyObject *_PyDict_FromItems(
+PyAPI_FUNC(PyObject *)_PyDict_FromItems(
         PyObject *const *keys, Py_ssize_t keys_offset,
         PyObject *const *values, Py_ssize_t values_offset,
         Py_ssize_t length);

Include/internal/pycore_floatobject.h

@@ -50,7 +50,7 @@ struct _Py_float_state {
 #endif
 };
 
-void _PyFloat_ExactDealloc(PyObject *op);
+PyAPI_FUNC(void) _PyFloat_ExactDealloc(PyObject *op);
 
 
 PyAPI_FUNC(void) _PyFloat_DebugMallocStats(FILE* out);

Include/internal/pycore_frame.h

@@ -114,7 +114,7 @@ _PyFrame_NumSlotsForCodeObject(PyCodeObject *code)
     return res;
 }
 
-void _PyFrame_Copy(_PyInterpreterFrame *src, _PyInterpreterFrame *dest);
+PyAPI_FUNC(void) _PyFrame_Copy(_PyInterpreterFrame *src, _PyInterpreterFrame *dest);
 
 /* Consumes reference to func and locals.
    Does not initialize frame->previous, which happens
@@ -265,7 +265,7 @@ _PyThreadState_HasStackSpace(PyThreadState *tstate, int size)
 extern _PyInterpreterFrame *
 _PyThreadState_PushFrame(PyThreadState *tstate, size_t size);
 
-void _PyThreadState_PopFrame(PyThreadState *tstate, _PyInterpreterFrame *frame);
+PyAPI_FUNC(void) _PyThreadState_PopFrame(PyThreadState *tstate, _PyInterpreterFrame *frame);
 
 /* Pushes a frame without checking for space.
  * Must be guarded by _PyThreadState_HasStackSpace()

Include/internal/pycore_genobject.h

@@ -9,7 +9,7 @@ extern "C" {
 #endif
 
 extern PyObject *_PyGen_yf(PyGenObject *);
-extern PyObject *_PyCoro_GetAwaitableIter(PyObject *o);
+PyAPI_FUNC(PyObject *)_PyCoro_GetAwaitableIter(PyObject *o);
 extern PyObject *_PyAsyncGenValueWrapperNew(PyThreadState *state, PyObject *);
 
 /* runtime lifecycle */

Include/internal/pycore_intrinsics.h

@@ -21,6 +21,6 @@
 typedef PyObject *(*instrinsic_func1)(PyThreadState* tstate, PyObject *value);
 typedef PyObject *(*instrinsic_func2)(PyThreadState* tstate, PyObject *value1, PyObject *value2);
 
-extern const instrinsic_func1 _PyIntrinsics_UnaryFunctions[];
-extern const instrinsic_func2 _PyIntrinsics_BinaryFunctions[];
+PyAPI_DATA(const instrinsic_func1) _PyIntrinsics_UnaryFunctions[];
+PyAPI_DATA(const instrinsic_func2) _PyIntrinsics_BinaryFunctions[];
 

Include/internal/pycore_jit.h

@@ -0,0 +1,10 @@
+typedef enum {
+    _JUSTIN_RETURN_DEOPT = -1,
+    _JUSTIN_RETURN_OK = 0,
+    _JUSTIN_RETURN_GOTO_ERROR = 1,
+} _PyJITReturnCode;
+
+typedef _PyJITReturnCode (*_PyJITFunction)(PyThreadState *tstate, _PyInterpreterFrame *frame, PyObject **stack_pointer, _Py_CODEUNIT *next_instr);
+
+PyAPI_FUNC(_PyJITFunction)_PyJIT_CompileTrace(int size, _Py_CODEUNIT **trace);
+PyAPI_FUNC(void)_PyJIT_Free(_PyJITFunction trace);

Include/internal/pycore_list.h

@@ -37,7 +37,7 @@ struct _Py_list_state {
 
 #define _PyList_ITEMS(op) _Py_RVALUE(_PyList_CAST(op)->ob_item)
 
-extern int
+PyAPI_FUNC(int)
 _PyList_AppendTakeRefListResize(PyListObject *self, PyObject *newitem);
 
 static inline int
@@ -75,7 +75,7 @@ typedef struct {
     PyListObject *it_seq; /* Set to NULL when iterator is exhausted */
 } _PyListIterObject;
 
-extern PyObject *_PyList_FromArraySteal(PyObject *const *src, Py_ssize_t n);
+PyAPI_FUNC(PyObject *)_PyList_FromArraySteal(PyObject *const *src, Py_ssize_t n);
 
 #ifdef __cplusplus
 }

Include/internal/pycore_long.h

@@ -78,9 +78,9 @@ static inline PyObject* _PyLong_FromUnsignedChar(unsigned char i)
     return Py_NewRef((PyObject *)&_PyLong_SMALL_INTS[_PY_NSMALLNEGINTS+i]);
 }
 
-PyObject *_PyLong_Add(PyLongObject *left, PyLongObject *right);
-PyObject *_PyLong_Multiply(PyLongObject *left, PyLongObject *right);
-PyObject *_PyLong_Subtract(PyLongObject *left, PyLongObject *right);
+PyAPI_FUNC(PyObject *)_PyLong_Add(PyLongObject *left, PyLongObject *right);
+PyAPI_FUNC(PyObject *)_PyLong_Multiply(PyLongObject *left, PyLongObject *right);
+PyAPI_FUNC(PyObject *)_PyLong_Subtract(PyLongObject *left, PyLongObject *right);
 
 /* Used by Python/mystrtoul.c, _PyBytes_FromHex(),
    _PyBytes_DecodeEscape(), etc. */

Include/internal/pycore_sliceobject.h

@@ -13,7 +13,7 @@ extern "C" {
 
 extern void _PySlice_Fini(PyInterpreterState *);
 
-extern PyObject *
+PyAPI_FUNC(PyObject *)
 _PyBuildSlice_ConsumeRefs(PyObject *start, PyObject *stop);
 
 #ifdef __cplusplus

Include/internal/pycore_tuple.h

@@ -65,7 +65,7 @@ struct _Py_tuple_state {
 #define _PyTuple_ITEMS(op) _Py_RVALUE(_PyTuple_CAST(op)->ob_item)
 
 extern PyObject *_PyTuple_FromArray(PyObject *const *, Py_ssize_t);
-extern PyObject *_PyTuple_FromArraySteal(PyObject *const *, Py_ssize_t);
+PyAPI_FUNC(PyObject *)_PyTuple_FromArraySteal(PyObject *const *, Py_ssize_t);
 
 
 typedef struct {

Include/internal/pycore_unicodeobject.h

@@ -11,7 +11,7 @@ extern "C" {
 #include "pycore_fileutils.h"     // _Py_error_handler
 #include "pycore_ucnhash.h"       // _PyUnicode_Name_CAPI
 
-void _PyUnicode_ExactDealloc(PyObject *op);
+PyAPI_FUNC(void) _PyUnicode_ExactDealloc(PyObject *op);
 
 /* runtime lifecycle */
 

Makefile.pre.in

@@ -394,6 +394,7 @@ PYTHON_OBJS=	\
 		Python/importdl.o \
 		Python/initconfig.o \
 		Python/intrinsics.o \
+		Python/jit.o \
 		Python/marshal.o \
 		Python/modsupport.o \
 		Python/mysnprintf.o \
@@ -1260,7 +1261,7 @@ regen-limited-abi: all
 regen-all: regen-cases regen-opcode regen-opcode-targets regen-typeslots \
 	regen-token regen-ast regen-keyword regen-sre regen-frozen clinic \
 	regen-pegen-metaparser regen-pegen regen-test-frozenmain \
-	regen-test-levenshtein regen-global-objects
+	regen-test-levenshtein regen-global-objects regen-jit-stencils
 	@echo
 	@echo "Note: make regen-stdlib-module-names and make regen-configure should be run manually"
 
@@ -1695,6 +1696,7 @@ PYTHON_HEADERS= \
 		$(srcdir)/Include/internal/pycore_interp.h \
 		$(srcdir)/Include/internal/pycore_interpreteridobject.h \
 		$(srcdir)/Include/internal/pycore_intrinsics.h \
+		$(srcdir)/Include/internal/pycore_jit.h \
 		$(srcdir)/Include/internal/pycore_list.h \
 		$(srcdir)/Include/internal/pycore_long.h \
 		$(srcdir)/Include/internal/pycore_moduleobject.h \
@@ -2423,6 +2425,16 @@ config.status:	$(srcdir)/configure
 Python/asm_trampoline.o: $(srcdir)/Python/asm_trampoline.S
 	$(CC) -c $(PY_CORE_CFLAGS) -o $@ $<
 
+Python/jit.o: $(srcdir)/Python/jit_stencils.h
+
+Python/jit_stencils.h: regen-jit-stencils
+
+.PHONY: regen-jit-stencils
+regen-jit-stencils: regen-cases
+	$(PYTHON_FOR_REGEN) $(srcdir)/Tools/justin/build.py \
+		$(srcdir)/Python/generated_cases.c.h $(srcdir)/Python/jit_stencils.h
+
+
 # Some make's put the object file in the current directory
 .c.o:
 	$(CC) -c $(PY_CORE_CFLAGS) -o $@ $<