GH-108614: Add `RESUME_CHECK` instruction (GH-108630)

7 files changed, 58 insertions, 40 deletions

diff --git a/Python/abstract_interp_cases.c.h b/Python/abstract_interp_cases.c.h
index 07e8a71..11fbf44 100644
--- a/Python/abstract_interp_cases.c.h
+++ b/Python/abstract_interp_cases.c.h
@@ -7,7 +7,7 @@
             break;
         }
 
-        case RESUME: {
+        case RESUME_CHECK: {
             break;
         }
 
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index fae1da3..8820b52 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -132,24 +132,37 @@ dummy_func(
         inst(NOP, (--)) {
         }
 
+        family(RESUME, 0) = {
+            RESUME_CHECK,
+        };
+
         inst(RESUME, (--)) {
+            TIER_ONE_ONLY
             assert(frame == tstate->current_frame);
-            /* Possibly combine this with eval breaker */
             if (_PyFrame_GetCode(frame)->_co_instrumentation_version != tstate->interp->monitoring_version) {
                 int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp);
                 ERROR_IF(err, error);
-                #if TIER_ONE
                 next_instr--;
-                #endif
-                #if TIER_TWO
-                goto deoptimize;
-                #endif
             }
-            else if (oparg < 2) {
-                CHECK_EVAL_BREAKER();
+            else {
+                if (oparg < 2) {
+                    CHECK_EVAL_BREAKER();
+                }
+                next_instr[-1].op.code = RESUME_CHECK;
             }
         }
 
+        inst(RESUME_CHECK, (--)) {
+#if defined(__EMSCRIPTEN__)
+            DEOPT_IF(emscripten_signal_clock == 0, RESUME);
+            emscripten_signal_clock -= Py_EMSCRIPTEN_SIGNAL_HANDLING;
+#endif
+            /* Possibly combine these two checks */
+            DEOPT_IF(_PyFrame_GetCode(frame)->_co_instrumentation_version
+                != tstate->interp->monitoring_version, RESUME);
+            DEOPT_IF(_Py_atomic_load_relaxed_int32(&tstate->interp->ceval.eval_breaker), RESUME);
+        }
+
         inst(INSTRUMENTED_RESUME, (--)) {
             /* Possible performance enhancement:
              *   We need to check the eval breaker anyway, can we
diff --git a/Python/ceval_macros.h b/Python/ceval_macros.h
index 81fbb79..f5d9155 100644
--- a/Python/ceval_macros.h
+++ b/Python/ceval_macros.h
@@ -374,6 +374,8 @@ static inline void _Py_LeaveRecursiveCallPy(PyThreadState *tstate)  {
     tstate->py_recursion_remaining++;
 }
 
+/* Marker to specify tier 1 only instructions */
+#define TIER_ONE_ONLY
 
 /* Implementation of "macros" that modify the instruction pointer,
  * stack pointer, or frame pointer.
diff --git a/Python/emscripten_signal.c b/Python/emscripten_signal.c
index d617ddf..1a19638 100644
--- a/Python/emscripten_signal.c
+++ b/Python/emscripten_signal.c
@@ -38,19 +38,17 @@ _Py_CheckEmscriptenSignals(void)
     }
 }
 
-
 #define PY_EMSCRIPTEN_SIGNAL_INTERVAL 50
 static int emscripten_signal_clock = PY_EMSCRIPTEN_SIGNAL_INTERVAL;
 
 void
 _Py_CheckEmscriptenSignalsPeriodically(void)
 {
-    if (!Py_EMSCRIPTEN_SIGNAL_HANDLING) {
-        return;
-    }
-    emscripten_signal_clock--;
     if (emscripten_signal_clock == 0) {
         emscripten_signal_clock = PY_EMSCRIPTEN_SIGNAL_INTERVAL;
         _Py_CheckEmscriptenSignals();
     }
+    else if (Py_EMSCRIPTEN_SIGNAL_HANDLING) {
+        emscripten_signal_clock--;
+    }
 }
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 0d5606f..f4c526a 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -7,22 +7,15 @@
             break;
         }
 
-        case RESUME: {
-            assert(frame == tstate->current_frame);
-            /* Possibly combine this with eval breaker */
-            if (_PyFrame_GetCode(frame)->_co_instrumentation_version != tstate->interp->monitoring_version) {
-                int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp);
-                if (err) goto error;
-                #if TIER_ONE
-                next_instr--;
-                #endif
-                #if TIER_TWO
-                goto deoptimize;
-                #endif
-            }
-            else if (oparg < 2) {
-                CHECK_EVAL_BREAKER();
-            }
+        case RESUME_CHECK: {
+#if defined(__EMSCRIPTEN__)
+            DEOPT_IF(emscripten_signal_clock == 0, RESUME);
+            emscripten_signal_clock -= Py_EMSCRIPTEN_SIGNAL_HANDLING;
+#endif
+            /* Possibly combine these two checks */
+            DEOPT_IF(_PyFrame_GetCode(frame)->_co_instrumentation_version
+                != tstate->interp->monitoring_version, RESUME);
+            DEOPT_IF(_Py_atomic_load_relaxed_int32(&tstate->interp->ceval.eval_breaker), RESUME);
             break;
         }
 
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 34cea84..84f83db 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -8,24 +8,36 @@
         }
 
         TARGET(RESUME) {
+            PREDICTED(RESUME);
+            static_assert(0 == 0, "incorrect cache size");
+            TIER_ONE_ONLY
             assert(frame == tstate->current_frame);
-            /* Possibly combine this with eval breaker */
             if (_PyFrame_GetCode(frame)->_co_instrumentation_version != tstate->interp->monitoring_version) {
                 int err = _Py_Instrument(_PyFrame_GetCode(frame), tstate->interp);
                 if (err) goto error;
-                #if TIER_ONE
                 next_instr--;
-                #endif
-                #if TIER_TWO
-                goto deoptimize;
-                #endif
             }
-            else if (oparg < 2) {
-                CHECK_EVAL_BREAKER();
+            else {
+                if (oparg < 2) {
+                    CHECK_EVAL_BREAKER();
+                }
+                next_instr[-1].op.code = RESUME_CHECK;
             }
             DISPATCH();
         }
 
+        TARGET(RESUME_CHECK) {
+#if defined(__EMSCRIPTEN__)
+            DEOPT_IF(emscripten_signal_clock == 0, RESUME);
+            emscripten_signal_clock -= Py_EMSCRIPTEN_SIGNAL_HANDLING;
+#endif
+            /* Possibly combine these two checks */
+            DEOPT_IF(_PyFrame_GetCode(frame)->_co_instrumentation_version
+                != tstate->interp->monitoring_version, RESUME);
+            DEOPT_IF(_Py_atomic_load_relaxed_int32(&tstate->interp->ceval.eval_breaker), RESUME);
+            DISPATCH();
+        }
+
         TARGET(INSTRUMENTED_RESUME) {
             /* Possible performance enhancement:
              *   We need to check the eval breaker anyway, can we
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index 305eb0b..413bb88 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -46,6 +46,7 @@ static void *opcode_targets[256] = {
     &&TARGET_POP_TOP,
     &&TARGET_PUSH_EXC_INFO,
     &&TARGET_PUSH_NULL,
+    &&TARGET_RESUME_CHECK,
     &&TARGET_RETURN_GENERATOR,
     &&TARGET_RETURN_VALUE,
     &&TARGET_SETUP_ANNOTATIONS,
@@ -164,8 +165,8 @@ static void *opcode_targets[256] = {
     &&TARGET_POP_JUMP_IF_NOT_NONE,
     &&TARGET_POP_JUMP_IF_TRUE,
     &&TARGET_RAISE_VARARGS,
-    &&TARGET_RERAISE,
     &&TARGET_RESUME,
+    &&TARGET_RERAISE,
     &&TARGET_RETURN_CONST,
     &&TARGET_SEND,
     &&TARGET_SEND_GEN,
@@ -235,7 +236,6 @@ static void *opcode_targets[256] = {
     &&_unknown_opcode,
     &&_unknown_opcode,
     &&_unknown_opcode,
-    &&_unknown_opcode,
     &&TARGET_INSTRUMENTED_RESUME,
     &&TARGET_INSTRUMENTED_END_FOR,
     &&TARGET_INSTRUMENTED_END_SEND,