Jan 13, 2024 · Jan 13, 2024 · Jan 13, 2024 · Jan 13, 2024 · Jan 14, 2024 · Jan 14, 2024
diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h

 #include "pycore_uop_ids.h"

 // This is the length of the trace we project initially.
 #define UOP_MAX_TRACE_LENGTH 512
 // This the above + additional working space we need.
 #define UOP_MAX_TRACE_WORKING_LENGTH (UOP_MAX_TRACE_LENGTH * 2)

 #define TRACE_STACK_SIZE 5

 int _Py_uop_analyze_and_optimize(_PyInterpreterFrame *frame,
    _PyUOpInstruction *trace, int trace_len, int curr_stackentries,
    _PyBloomFilter *dependencies);


 static void
 clear_strong_refs_in_uops(_PyUOpInstruction *trace, Py_ssize_t uop_len)
 {
    for (Py_ssize_t i = 0; i < uop_len; i++) {
        if (trace[i].opcode == _LOAD_CONST_INLINE ||
            trace[i].opcode == _LOAD_CONST_INLINE_WITH_NULL) {
            PyObject *c = (PyObject*)trace[i].operand;
            Py_CLEAR(c);
        }
        if (trace[i].opcode == _JUMP_ABSOLUTE ||
            trace[i].opcode == _JUMP_TO_TOP ||
            trace[i].opcode == _EXIT_TRACE) {
            return;
        }
    }
 }


 extern PyTypeObject _PyCounterExecutor_Type;
 extern PyTypeObject _PyCounterOptimizer_Type;
diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py
        # 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.
        self.assertIsNone(ex)

    def test_promote_globals_to_constants(self):
        def dummy(x):
diff --git a/Python/optimizer.c b/Python/optimizer.c
 #include "pycore_uop_metadata.h" // Uop tables
 #undef NEED_OPCODE_METADATA

 // This is the length of the trace we project initially.
 #define UOP_MAX_TRACE_LENGTH 512
 // This the above + additional working space we need.
 #define UOP_MAX_TRACE_WORKING_LENGTH (UOP_MAX_TRACE_LENGTH * 2)

 #define MAX_EXECUTORS_SIZE 256


 ///////////////////// Experimental UOp Optimizer /////////////////////


 static void
 clear_strong_refs_in_uops(_PyUOpInstruction *trace, Py_ssize_t uop_len)
 {
    for (Py_ssize_t i = 0; i < uop_len; i++) {
        if (trace[i].opcode == _LOAD_CONST_INLINE ||
            trace[i].opcode == _LOAD_CONST_INLINE_WITH_NULL) {
            PyObject *c = (PyObject*)trace[i].operand;
            Py_CLEAR(c);
        }
        if (trace[i].opcode == _JUMP_ABSOLUTE ||
            trace[i].opcode == _JUMP_TO_TOP ||
            trace[i].opcode == _EXIT_TRACE) {
            return;
        }
    }
 }

 static void
 uop_dealloc(_PyExecutorObject *self) {
    clear_strong_refs_in_uops(&self->trace[0], Py_SIZE(self));
diff --git a/Python/optimizer_analysis.c b/Python/optimizer_analysis.c
    }
    else {
        inst->opcode = (inst->oparg & 1) ? _LOAD_CONST_INLINE_WITH_NULL : _LOAD_CONST_INLINE;
        Py_INCREF(res);
    }
    inst->operand = (uint64_t)res;
 }
 )
 {
    OPT_STAT_INC(optimizer_attempts);
    _PyUOpInstruction *temp_writebuffer = NULL;
    _PyUOpInstruction temp_writebuffer[UOP_MAX_TRACE_WORKING_LENGTH];
    _PyUOpInstruction temp_readbuffer[UOP_MAX_TRACE_WORKING_LENGTH];

    temp_writebuffer = PyMem_New(_PyUOpInstruction, buffer_size);
    if (temp_writebuffer == NULL) {
        goto error;
    }
    memcpy(temp_readbuffer, buffer, buffer_size * sizeof(_PyUOpInstruction));

    int err = remove_globals(frame,buffer, buffer_size, dependencies);
    int err = remove_globals(frame,temp_readbuffer, buffer_size, dependencies);
    if (err <= 0) {
 return err;
 goto error;
    }

    // Pass: Abstract interpretation and symbolic analysis
    int new_trace_len = uop_abstract_interpret(
        (PyCodeObject *)frame->f_executable,buffer, temp_writebuffer,
        (PyCodeObject *)frame->f_executable,temp_readbuffer, temp_writebuffer,
        buffer_size, curr_stacklen);

    if (new_trace_len < 0) {
        goto error;
    }

    clear_strong_refs_in_uops(buffer, buffer_size);

    remove_unneeded_uops(temp_writebuffer, new_trace_len);

    // Fill in our new trace!
    memcpy(buffer, temp_writebuffer, new_trace_len * sizeof(_PyUOpInstruction));

    PyMem_Free(temp_writebuffer);

    OPT_STAT_INC(optimizer_successes);
    return 1;
 error:

    // The only valid error we can raise is MemoryError.
    // Other times it's not really errors but things like not being able
    // to fetch a function version because the function got deleted.
    PyMem_Free(temp_writebuffer);
    return PyErr_Occurred() ? -1 : 0;
 }
Original file line number	Diff line number	Diff line change
Expand Up		@@ -10,30 +10,18 @@ extern "C" {

		#include "pycore_uop_ids.h"

		// This is the length of the trace we project initially.
		#define UOP_MAX_TRACE_LENGTH 512
		// This the above + additional working space we need.
		#define UOP_MAX_TRACE_WORKING_LENGTH (UOP_MAX_TRACE_LENGTH * 2)
Comment on lines +15 to +16 Copy link Member gvanrossumFeb 5, 2024 Choose a reason for hiding this comment The reason will be displayed to describe this comment to others.Learn more. I don't see this used in the code any more.

		#define TRACE_STACK_SIZE 5

		int _Py_uop_analyze_and_optimize(_PyInterpreterFrame *frame,
		_PyUOpInstruction *trace, int trace_len, int curr_stackentries,
		_PyBloomFilter *dependencies);


		static void
		clear_strong_refs_in_uops(_PyUOpInstruction *trace, Py_ssize_t uop_len)
		{
		for (Py_ssize_t i = 0; i < uop_len; i++) {
		if (trace[i].opcode == _LOAD_CONST_INLINE \|\|
		trace[i].opcode == _LOAD_CONST_INLINE_WITH_NULL) {
		PyObject c = (PyObject)trace[i].operand;
		Py_CLEAR(c);
		}
		if (trace[i].opcode == _JUMP_ABSOLUTE \|\|
		trace[i].opcode == _JUMP_TO_TOP \|\|
		trace[i].opcode == _EXIT_TRACE) {
		return;
		}
		}
		}


		extern PyTypeObject _PyCounterExecutor_Type;
		extern PyTypeObject _PyCounterOptimizer_Type;
Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -895,7 +895,6 @@ def dummy(x):
		# 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.
		self.assertIsNone(ex)

		def test_promote_globals_to_constants(self):
		def dummy(x):
Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -17,11 +17,6 @@
		#include "pycore_uop_metadata.h" // Uop tables
		#undef NEED_OPCODE_METADATA

		// This is the length of the trace we project initially.
		#define UOP_MAX_TRACE_LENGTH 512
		// This the above + additional working space we need.
		#define UOP_MAX_TRACE_WORKING_LENGTH (UOP_MAX_TRACE_LENGTH * 2)

		#define MAX_EXECUTORS_SIZE 256


Expand DownExpand Up		@@ -227,6 +222,23 @@ static PyMethodDef executor_methods[] = {
		///////////////////// Experimental UOp Optimizer /////////////////////


		static void
		clear_strong_refs_in_uops(_PyUOpInstruction *trace, Py_ssize_t uop_len)
		{
		for (Py_ssize_t i = 0; i < uop_len; i++) {
		if (trace[i].opcode == _LOAD_CONST_INLINE \|\|
		trace[i].opcode == _LOAD_CONST_INLINE_WITH_NULL) {
		PyObject c = (PyObject)trace[i].operand;
		Py_CLEAR(c);
		}
		if (trace[i].opcode == _JUMP_ABSOLUTE \|\|
		trace[i].opcode == _JUMP_TO_TOP \|\|
		trace[i].opcode == _EXIT_TRACE) {
		return;
		}
		}
		}

		static void
		uop_dealloc(_PyExecutorObject *self) {
		clear_strong_refs_in_uops(&self->trace[0], Py_SIZE(self));
Expand Down
Original file line number	Diff line number	Diff line change
Expand Up		@@ -1082,7 +1082,6 @@ global_to_const(_PyUOpInstruction inst, PyObject obj)
		}
		else {
		inst->opcode = (inst->oparg & 1) ? _LOAD_CONST_INLINE_WITH_NULL : _LOAD_CONST_INLINE;
		Py_INCREF(res);
		}
		inst->operand = (uint64_t)res;
		}
Expand DownExpand Up		@@ -1397,42 +1396,38 @@ _Py_uop_analyze_and_optimize(
		)
		{
		OPT_STAT_INC(optimizer_attempts);
		_PyUOpInstruction *temp_writebuffer = NULL;
		_PyUOpInstruction temp_writebuffer[UOP_MAX_TRACE_WORKING_LENGTH];
		_PyUOpInstruction temp_readbuffer[UOP_MAX_TRACE_WORKING_LENGTH];

		temp_writebuffer = PyMem_New(_PyUOpInstruction, buffer_size);
		if (temp_writebuffer == NULL) {
		goto error;
		}
		memcpy(temp_readbuffer, buffer, buffer_size * sizeof(_PyUOpInstruction));

		int err = remove_globals(frame,buffer, buffer_size, dependencies);
		int err = remove_globals(frame,temp_readbuffer, buffer_size, dependencies);
		if (err <= 0) {
		return err;
		goto error;
		}

		// Pass: Abstract interpretation and symbolic analysis
		int new_trace_len = uop_abstract_interpret(
		(PyCodeObject *)frame->f_executable,buffer, temp_writebuffer,
		(PyCodeObject *)frame->f_executable,temp_readbuffer, temp_writebuffer,
		buffer_size, curr_stacklen);

		if (new_trace_len < 0) {
		goto error;
		}

		clear_strong_refs_in_uops(buffer, buffer_size);

		remove_unneeded_uops(temp_writebuffer, new_trace_len);

		// Fill in our new trace!
		memcpy(buffer, temp_writebuffer, new_trace_len * sizeof(_PyUOpInstruction));

		PyMem_Free(temp_writebuffer);

		OPT_STAT_INC(optimizer_successes);
		return 1;
		error:

		// The only valid error we can raise is MemoryError.
		// Other times it's not really errors but things like not being able
		// to fetch a function version because the function got deleted.
		PyMem_Free(temp_writebuffer);
		return PyErr_Occurred() ? -1 : 0;
		}