diff --git a/Misc/NEWS.d/next/Core and Builtins/2024-02-04-07-45-29.gh-issue-107674.q8mCmi.rst b/Misc/NEWS.d/next/Core and Builtins/2024-02-04-07-45-29.gh-issue-107674.q8mCmi.rst new file mode 100644 index 00000000000..f9b96788bfa --- /dev/null +++ b/Misc/NEWS.d/next/Core and Builtins/2024-02-04-07-45-29.gh-issue-107674.q8mCmi.rst @@ -0,0 +1 @@ +Improved the performance of :func:`sys.settrace` significantly diff --git a/Python/bytecodes.c b/Python/bytecodes.c index e9e9425f826..565379afc4b 100644 --- a/Python/bytecodes.c +++ b/Python/bytecodes.c @@ -143,22 +143,23 @@ dummy_func( tier1 inst(RESUME, (--)) { assert(frame == tstate->current_frame); - uintptr_t global_version = - _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker) & - ~_PY_EVAL_EVENTS_MASK; - uintptr_t code_version = _PyFrame_GetCode(frame)->_co_instrumentation_version; - assert((code_version & 255) == 0); - if (code_version != global_version) { - int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp); - ERROR_IF(err, error); - next_instr = this_instr; - } - else { - if ((oparg & RESUME_OPARG_LOCATION_MASK) < RESUME_AFTER_YIELD_FROM) { - CHECK_EVAL_BREAKER(); + if (tstate->tracing == 0) { + uintptr_t global_version = + _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker) & + ~_PY_EVAL_EVENTS_MASK; + uintptr_t code_version = _PyFrame_GetCode(frame)->_co_instrumentation_version; + assert((code_version & 255) == 0); + if (code_version != global_version) { + int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp); + ERROR_IF(err, error); + next_instr = this_instr; + DISPATCH(); } - this_instr->op.code = RESUME_CHECK; } + if ((oparg & RESUME_OPARG_LOCATION_MASK) < RESUME_AFTER_YIELD_FROM) { + CHECK_EVAL_BREAKER(); + } + this_instr->op.code = RESUME_CHECK; } inst(RESUME_CHECK, (--)) { @@ -169,13 +170,13 @@ dummy_func( uintptr_t eval_breaker = _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker); uintptr_t version = _PyFrame_GetCode(frame)->_co_instrumentation_version; assert((version & _PY_EVAL_EVENTS_MASK) == 0); - DEOPT_IF(eval_breaker != version); + DEOPT_IF(eval_breaker != version && tstate->tracing == 0); } inst(INSTRUMENTED_RESUME, (--)) { uintptr_t global_version = _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker) & ~_PY_EVAL_EVENTS_MASK; uintptr_t code_version = _PyFrame_GetCode(frame)->_co_instrumentation_version; - if (code_version != global_version) { + if (code_version != global_version && tstate->tracing == 0) { if (_Py_Instrument(_PyFrame_GetCode(frame), tstate->interp)) { GOTO_ERROR(error); } diff --git a/Python/ceval.c b/Python/ceval.c index 06c136aeb25..41e9310938d 100644 --- a/Python/ceval.c +++ b/Python/ceval.c @@ -800,17 +800,23 @@ resume_frame: { _Py_CODEUNIT *prev = frame->instr_ptr; _Py_CODEUNIT *here = frame->instr_ptr = next_instr; - _PyFrame_SetStackPointer(frame, stack_pointer); - int original_opcode = _Py_call_instrumentation_line( - tstate, frame, here, prev); - stack_pointer = _PyFrame_GetStackPointer(frame); - if (original_opcode < 0) { - next_instr = here+1; - goto error; - } - next_instr = frame->instr_ptr; - if (next_instr != here) { - DISPATCH(); + int original_opcode = 0; + if (tstate->tracing) { + PyCodeObject *code = _PyFrame_GetCode(frame); + original_opcode = code->_co_monitoring->lines[(int)(here - _PyCode_CODE(code))].original_opcode; + } else { + _PyFrame_SetStackPointer(frame, stack_pointer); + original_opcode = _Py_call_instrumentation_line( + tstate, frame, here, prev); + stack_pointer = _PyFrame_GetStackPointer(frame); + if (original_opcode < 0) { + next_instr = here+1; + goto error; + } + next_instr = frame->instr_ptr; + if (next_instr != here) { + DISPATCH(); + } } if (_PyOpcode_Caches[original_opcode]) { _PyBinaryOpCache *cache = (_PyBinaryOpCache *)(next_instr+1); diff --git a/Python/ceval_macros.h b/Python/ceval_macros.h index 085199416c1..6ad41950ea3 100644 --- a/Python/ceval_macros.h +++ b/Python/ceval_macros.h @@ -339,12 +339,16 @@ do { \ // for an exception handler, displaying the traceback, and so on #define INSTRUMENTED_JUMP(src, dest, event) \ do { \ - _PyFrame_SetStackPointer(frame, stack_pointer); \ - next_instr = _Py_call_instrumentation_jump(tstate, event, frame, src, dest); \ - stack_pointer = _PyFrame_GetStackPointer(frame); \ - if (next_instr == NULL) { \ - next_instr = (dest)+1; \ - goto error; \ + if (tstate->tracing) {\ + next_instr = dest; \ + } else { \ + _PyFrame_SetStackPointer(frame, stack_pointer); \ + next_instr = _Py_call_instrumentation_jump(tstate, event, frame, src, dest); \ + stack_pointer = _PyFrame_GetStackPointer(frame); \ + if (next_instr == NULL) { \ + next_instr = (dest)+1; \ + goto error; \ + } \ } \ } while (0); diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h index 0b1d75985e1..20fab8f4c61 100644 --- a/Python/executor_cases.c.h +++ b/Python/executor_cases.c.h @@ -20,7 +20,7 @@ uintptr_t eval_breaker = _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker); uintptr_t version = _PyFrame_GetCode(frame)->_co_instrumentation_version; assert((version & _PY_EVAL_EVENTS_MASK) == 0); - if (eval_breaker != version) goto deoptimize; + if (eval_breaker != version && tstate->tracing == 0) goto deoptimize; break; } diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h index 8b90e448a7b..bb26dac0e2b 100644 --- a/Python/generated_cases.c.h +++ b/Python/generated_cases.c.h @@ -3125,7 +3125,7 @@ INSTRUCTION_STATS(INSTRUMENTED_RESUME); uintptr_t global_version = _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker) & ~_PY_EVAL_EVENTS_MASK; uintptr_t code_version = _PyFrame_GetCode(frame)->_co_instrumentation_version; - if (code_version != global_version) { + if (code_version != global_version && tstate->tracing == 0) { if (_Py_Instrument(_PyFrame_GetCode(frame), tstate->interp)) { GOTO_ERROR(error); } @@ -4780,22 +4780,23 @@ PREDICTED(RESUME); _Py_CODEUNIT *this_instr = next_instr - 1; assert(frame == tstate->current_frame); - uintptr_t global_version = - _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker) & - ~_PY_EVAL_EVENTS_MASK; - uintptr_t code_version = _PyFrame_GetCode(frame)->_co_instrumentation_version; - assert((code_version & 255) == 0); - if (code_version != global_version) { - int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp); - if (err) goto error; - next_instr = this_instr; - } - else { - if ((oparg & RESUME_OPARG_LOCATION_MASK) < RESUME_AFTER_YIELD_FROM) { - CHECK_EVAL_BREAKER(); + if (tstate->tracing == 0) { + uintptr_t global_version = + _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker) & + ~_PY_EVAL_EVENTS_MASK; + uintptr_t code_version = _PyFrame_GetCode(frame)->_co_instrumentation_version; + assert((code_version & 255) == 0); + if (code_version != global_version) { + int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp); + if (err) goto error; + next_instr = this_instr; + DISPATCH(); } - this_instr->op.code = RESUME_CHECK; } + if ((oparg & RESUME_OPARG_LOCATION_MASK) < RESUME_AFTER_YIELD_FROM) { + CHECK_EVAL_BREAKER(); + } + this_instr->op.code = RESUME_CHECK; DISPATCH(); } @@ -4811,7 +4812,7 @@ uintptr_t eval_breaker = _Py_atomic_load_uintptr_relaxed(&tstate->eval_breaker); uintptr_t version = _PyFrame_GetCode(frame)->_co_instrumentation_version; assert((version & _PY_EVAL_EVENTS_MASK) == 0); - DEOPT_IF(eval_breaker != version, RESUME); + DEOPT_IF(eval_breaker != version && tstate->tracing == 0, RESUME); DISPATCH(); } diff --git a/Python/instrumentation.c b/Python/instrumentation.c index 6f1bc2e0a10..4b7d8b5a587 100644 --- a/Python/instrumentation.c +++ b/Python/instrumentation.c @@ -1156,15 +1156,13 @@ int _Py_call_instrumentation_line(PyThreadState *tstate, _PyInterpreterFrame* frame, _Py_CODEUNIT *instr, _Py_CODEUNIT *prev) { PyCodeObject *code = _PyFrame_GetCode(frame); + assert(tstate->tracing == 0); assert(is_version_up_to_date(code, tstate->interp)); assert(instrumentation_cross_checks(tstate->interp, code)); int i = (int)(instr - _PyCode_CODE(code)); _PyCoMonitoringData *monitoring = code->_co_monitoring; _PyCoLineInstrumentationData *line_data = &monitoring->lines[i]; - if (tstate->tracing) { - goto done; - } PyInterpreterState *interp = tstate->interp; int8_t line_delta = line_data->line_delta; int line = compute_line(code, i, line_delta);