JIT: Streamline MAKE_WARM - move coldness check to executor creation

Author: alonme

Include/internal/pycore_optimizer.h

@@ -116,7 +116,8 @@ PyAPI_FUNC(void) _Py_Executors_InvalidateCold(PyInterpreterState *interp);
 
 // Used as the threshold to trigger executor invalidation when
 // trace_run_counter is greater than this value.
-#define JIT_CLEANUP_THRESHOLD 100000
+// TODO: Test what should be the optimal value for this.
+#define JIT_CLEANUP_THRESHOLD 1000
 
 // This is the length of the trace we project initially.
 #define UOP_MAX_TRACE_LENGTH 800

Python/bytecodes.c

@@ -11,7 +11,6 @@
 #include "pycore_audit.h"         // _PySys_Audit()
 #include "pycore_backoff.h"
 #include "pycore_cell.h"          // PyCell_GetRef()
-#include "pycore_ceval.h"
 #include "pycore_code.h"
 #include "pycore_emscripten_signal.h"  // _Py_CHECK_EMSCRIPTEN_SIGNALS
 #include "pycore_function.h"
@@ -5367,10 +5366,6 @@ dummy_func(
 
         tier2 op(_MAKE_WARM, (--)) {
             current_executor->vm_data.warm = true;
-            // It's okay if this ends up going negative.
-            if (--tstate->interp->trace_run_counter == 0) {
-                _Py_set_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
-            }
         }
 
         tier2 op(_FATAL_ERROR, (--)) {

Python/executor_cases.c.h

@@ -6,6 +6,7 @@
 #include "pycore_interp.h"
 #include "pycore_backoff.h"
 #include "pycore_bitutils.h"        // _Py_popcount32()
+#include "pycore_ceval.h"       // _Py_set_eval_breaker_bit
 #include "pycore_code.h"            // _Py_GetBaseCodeUnit
 #include "pycore_function.h"        // _PyFunction_LookupByVersion()
 #include "pycore_interpframe.h"
@@ -1322,6 +1323,14 @@ uop_optimize(
         return -1;
     }
     assert(length <= UOP_MAX_TRACE_LENGTH);
+
+    // Check executor coldness
+    PyThreadState *tstate = PyThreadState_Get();
+    // It's okay if this ends up going negative.
+    if (--tstate->interp->trace_run_counter == 0) {
+        _Py_set_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
+    }
+
     *exec_ptr = executor;
     return 1;
 }