gh-115687: Split up guards from COMPARE_OP (GH-115688)

5 files changed, 88 insertions, 55 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 27c439b..10bb152 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -2200,9 +2200,16 @@ dummy_func(
 
         macro(COMPARE_OP) = _SPECIALIZE_COMPARE_OP + _COMPARE_OP;
 
-        inst(COMPARE_OP_FLOAT, (unused/1, left, right -- res)) {
-            DEOPT_IF(!PyFloat_CheckExact(left));
-            DEOPT_IF(!PyFloat_CheckExact(right));
+        macro(COMPARE_OP_FLOAT) =
+            _GUARD_BOTH_FLOAT + unused/1 + _COMPARE_OP_FLOAT;
+
+        macro(COMPARE_OP_INT) =
+            _GUARD_BOTH_INT + unused/1 + _COMPARE_OP_INT;
+
+        macro(COMPARE_OP_STR) =
+            _GUARD_BOTH_UNICODE + unused/1 + _COMPARE_OP_STR;
+
+        op(_COMPARE_OP_FLOAT, (left, right -- res)) {
             STAT_INC(COMPARE_OP, hit);
             double dleft = PyFloat_AS_DOUBLE(left);
             double dright = PyFloat_AS_DOUBLE(right);
@@ -2215,9 +2222,7 @@ dummy_func(
         }
 
         // Similar to COMPARE_OP_FLOAT
-        inst(COMPARE_OP_INT, (unused/1, left, right -- res)) {
-            DEOPT_IF(!PyLong_CheckExact(left));
-            DEOPT_IF(!PyLong_CheckExact(right));
+        op(_COMPARE_OP_INT, (left, right -- res)) {
             DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)left));
             DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)right));
             STAT_INC(COMPARE_OP, hit);
@@ -2234,9 +2239,7 @@ dummy_func(
         }
 
         // Similar to COMPARE_OP_FLOAT, but for ==, != only
-        inst(COMPARE_OP_STR, (unused/1, left, right -- res)) {
-            DEOPT_IF(!PyUnicode_CheckExact(left));
-            DEOPT_IF(!PyUnicode_CheckExact(right));
+        op(_COMPARE_OP_STR, (left, right -- res)) {
             STAT_INC(COMPARE_OP, hit);
             int eq = _PyUnicode_Equal(left, right);
             assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE);
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index b46885e..445f98b 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -2100,8 +2100,6 @@
             oparg = CURRENT_OPARG();
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            if (!PyFloat_CheckExact(left)) goto deoptimize;
-            if (!PyFloat_CheckExact(right)) goto deoptimize;
             STAT_INC(COMPARE_OP, hit);
             double dleft = PyFloat_AS_DOUBLE(left);
             double dright = PyFloat_AS_DOUBLE(right);
@@ -2123,8 +2121,6 @@
             oparg = CURRENT_OPARG();
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            if (!PyLong_CheckExact(left)) goto deoptimize;
-            if (!PyLong_CheckExact(right)) goto deoptimize;
             if (!_PyLong_IsCompact((PyLongObject *)left)) goto deoptimize;
             if (!_PyLong_IsCompact((PyLongObject *)right)) goto deoptimize;
             STAT_INC(COMPARE_OP, hit);
@@ -2150,8 +2146,6 @@
             oparg = CURRENT_OPARG();
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            if (!PyUnicode_CheckExact(left)) goto deoptimize;
-            if (!PyUnicode_CheckExact(right)) goto deoptimize;
             STAT_INC(COMPARE_OP, hit);
             int eq = _PyUnicode_Equal(left, right);
             assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE);
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 324e53d..7899106 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -2026,20 +2026,26 @@
             PyObject *right;
             PyObject *left;
             PyObject *res;
-            /* Skip 1 cache entry */
+            // _GUARD_BOTH_FLOAT
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_OP);
-            DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_OP);
-            STAT_INC(COMPARE_OP, hit);
-            double dleft = PyFloat_AS_DOUBLE(left);
-            double dright = PyFloat_AS_DOUBLE(right);
-            // 1 if NaN, 2 if <, 4 if >, 8 if ==; this matches low four bits of the oparg
-            int sign_ish = COMPARISON_BIT(dleft, dright);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            res = (sign_ish & oparg) ? Py_True : Py_False;
-            // It's always a bool, so we don't care about oparg & 16.
+            {
+                DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_OP);
+                DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_OP);
+            }
+            /* Skip 1 cache entry */
+            // _COMPARE_OP_FLOAT
+            {
+                STAT_INC(COMPARE_OP, hit);
+                double dleft = PyFloat_AS_DOUBLE(left);
+                double dright = PyFloat_AS_DOUBLE(right);
+                // 1 if NaN, 2 if <, 4 if >, 8 if ==; this matches low four bits of the oparg
+                int sign_ish = COMPARISON_BIT(dleft, dright);
+                _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
+                _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
+                res = (sign_ish & oparg) ? Py_True : Py_False;
+                // It's always a bool, so we don't care about oparg & 16.
+            }
             stack_pointer[-2] = res;
             stack_pointer += -1;
             DISPATCH();
@@ -2053,24 +2059,30 @@
             PyObject *right;
             PyObject *left;
             PyObject *res;
-            /* Skip 1 cache entry */
+            // _GUARD_BOTH_INT
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            DEOPT_IF(!PyLong_CheckExact(left), COMPARE_OP);
-            DEOPT_IF(!PyLong_CheckExact(right), COMPARE_OP);
-            DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)left), COMPARE_OP);
-            DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)right), COMPARE_OP);
-            STAT_INC(COMPARE_OP, hit);
-            assert(_PyLong_DigitCount((PyLongObject *)left) <= 1 &&
+            {
+                DEOPT_IF(!PyLong_CheckExact(left), COMPARE_OP);
+                DEOPT_IF(!PyLong_CheckExact(right), COMPARE_OP);
+            }
+            /* Skip 1 cache entry */
+            // _COMPARE_OP_INT
+            {
+                DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)left), COMPARE_OP);
+                DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)right), COMPARE_OP);
+                STAT_INC(COMPARE_OP, hit);
+                assert(_PyLong_DigitCount((PyLongObject *)left) <= 1 &&
                    _PyLong_DigitCount((PyLongObject *)right) <= 1);
-            Py_ssize_t ileft = _PyLong_CompactValue((PyLongObject *)left);
-            Py_ssize_t iright = _PyLong_CompactValue((PyLongObject *)right);
-            // 2 if <, 4 if >, 8 if ==; this matches the low 4 bits of the oparg
-            int sign_ish = COMPARISON_BIT(ileft, iright);
-            _Py_DECREF_SPECIALIZED(left, (destructor)PyObject_Free);
-            _Py_DECREF_SPECIALIZED(right, (destructor)PyObject_Free);
-            res = (sign_ish & oparg) ? Py_True : Py_False;
-            // It's always a bool, so we don't care about oparg & 16.
+                Py_ssize_t ileft = _PyLong_CompactValue((PyLongObject *)left);
+                Py_ssize_t iright = _PyLong_CompactValue((PyLongObject *)right);
+                // 2 if <, 4 if >, 8 if ==; this matches the low 4 bits of the oparg
+                int sign_ish = COMPARISON_BIT(ileft, iright);
+                _Py_DECREF_SPECIALIZED(left, (destructor)PyObject_Free);
+                _Py_DECREF_SPECIALIZED(right, (destructor)PyObject_Free);
+                res = (sign_ish & oparg) ? Py_True : Py_False;
+                // It's always a bool, so we don't care about oparg & 16.
+            }
             stack_pointer[-2] = res;
             stack_pointer += -1;
             DISPATCH();
@@ -2084,21 +2096,27 @@
             PyObject *right;
             PyObject *left;
             PyObject *res;
-            /* Skip 1 cache entry */
+            // _GUARD_BOTH_UNICODE
             right = stack_pointer[-1];
             left = stack_pointer[-2];
-            DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_OP);
-            DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_OP);
-            STAT_INC(COMPARE_OP, hit);
-            int eq = _PyUnicode_Equal(left, right);
-            assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE);
-            _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(right, _PyUnicode_ExactDealloc);
-            assert(eq == 0 || eq == 1);
-            assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS);
-            assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS);
-            res = ((COMPARISON_NOT_EQUALS + eq) & oparg) ? Py_True : Py_False;
-            // It's always a bool, so we don't care about oparg & 16.
+            {
+                DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_OP);
+                DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_OP);
+            }
+            /* Skip 1 cache entry */
+            // _COMPARE_OP_STR
+            {
+                STAT_INC(COMPARE_OP, hit);
+                int eq = _PyUnicode_Equal(left, right);
+                assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE);
+                _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc);
+                _Py_DECREF_SPECIALIZED(right, _PyUnicode_ExactDealloc);
+                assert(eq == 0 || eq == 1);
+                assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS);
+                assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS);
+                res = ((COMPARISON_NOT_EQUALS + eq) & oparg) ? Py_True : Py_False;
+                // It's always a bool, so we don't care about oparg & 16.
+            }
             stack_pointer[-2] = res;
             stack_pointer += -1;
             DISPATCH();
diff --git a/Python/tier2_redundancy_eliminator_bytecodes.c b/Python/tier2_redundancy_eliminator_bytecodes.c
index 3f6e8ce..e9b556d 100644
--- a/Python/tier2_redundancy_eliminator_bytecodes.c
+++ b/Python/tier2_redundancy_eliminator_bytecodes.c
@@ -77,6 +77,14 @@ dummy_func(void) {
         sym_set_type(right, &PyFloat_Type);
     }
 
+    op(_GUARD_BOTH_UNICODE, (left, right -- left, right)) {
+        if (sym_matches_type(left, &PyUnicode_Type) &&
+            sym_matches_type(right, &PyUnicode_Type)) {
+            REPLACE_OP(this_instr, _NOP, 0 ,0);
+        }
+        sym_set_type(left, &PyUnicode_Type);
+        sym_set_type(right, &PyUnicode_Type);
+    }
 
     op(_BINARY_OP_ADD_INT, (left, right -- res)) {
         if (is_const(left) && is_const(right)) {
diff --git a/Python/tier2_redundancy_eliminator_cases.c.h b/Python/tier2_redundancy_eliminator_cases.c.h
index 904700a..f41fe32 100644
--- a/Python/tier2_redundancy_eliminator_cases.c.h
+++ b/Python/tier2_redundancy_eliminator_cases.c.h
@@ -351,6 +351,16 @@
         }
 
         case _GUARD_BOTH_UNICODE: {
+            _Py_UOpsSymType *right;
+            _Py_UOpsSymType *left;
+            right = stack_pointer[-1];
+            left = stack_pointer[-2];
+            if (sym_matches_type(left, &PyUnicode_Type) &&
+                sym_matches_type(right, &PyUnicode_Type)) {
+                REPLACE_OP(this_instr, _NOP, 0 ,0);
+            }
+            sym_set_type(left, &PyUnicode_Type);
+            sym_set_type(right, &PyUnicode_Type);
             break;
         }