GH-100982: Add `COMPARE_AND_BRANCH` instruction (GH-100983)

6 files changed, 199 insertions, 202 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index faa6df6..18f9eab 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -90,11 +90,6 @@ static size_t jump;
 // Dummy variables for cache effects
 static uint16_t invert, counter, index, hint;
 static uint32_t type_version;
-// Dummy opcode names for 'op' opcodes
-#define _COMPARE_OP_FLOAT 1003
-#define _COMPARE_OP_INT 1004
-#define _COMPARE_OP_STR 1005
-#define _JUMP_IF 1006
 
 static PyObject *
 dummy_func(
@@ -1829,64 +1824,76 @@ dummy_func(
             Py_DECREF(owner);
         }
 
-        family(compare_op) = {
-            COMPARE_OP,
-            _COMPARE_OP_FLOAT,
-            _COMPARE_OP_INT,
-            _COMPARE_OP_STR,
+        inst(COMPARE_OP, (unused/1, left, right -- res)) {
+            STAT_INC(COMPARE_OP, deferred);
+            assert((oparg >> 4) <= Py_GE);
+            res = PyObject_RichCompare(left, right, oparg>>4);
+            Py_DECREF(left);
+            Py_DECREF(right);
+            ERROR_IF(res == NULL, error);
+        }
+
+        family(compare_and_branch) = {
+            COMPARE_AND_BRANCH,
+            COMPARE_AND_BRANCH_FLOAT,
+            COMPARE_AND_BRANCH_INT,
+            COMPARE_AND_BRANCH_STR,
         };
 
-        inst(COMPARE_OP, (unused/1, left, right -- res)) {
+        inst(COMPARE_AND_BRANCH, (unused/2, left, right -- )) {
             _PyCompareOpCache *cache = (_PyCompareOpCache *)next_instr;
             if (ADAPTIVE_COUNTER_IS_ZERO(cache->counter)) {
                 assert(cframe.use_tracing == 0);
                 next_instr--;
-                _Py_Specialize_CompareOp(left, right, next_instr, oparg);
+                _Py_Specialize_CompareAndBranch(left, right, next_instr, oparg);
                 DISPATCH_SAME_OPARG();
             }
-            STAT_INC(COMPARE_OP, deferred);
+            STAT_INC(COMPARE_AND_BRANCH, deferred);
             DECREMENT_ADAPTIVE_COUNTER(cache->counter);
             assert((oparg >> 4) <= Py_GE);
-            res = PyObject_RichCompare(left, right, oparg>>4);
+            PyObject *cond = PyObject_RichCompare(left, right, oparg>>4);
             Py_DECREF(left);
             Py_DECREF(right);
-            ERROR_IF(res == NULL, error);
+            ERROR_IF(cond == NULL, error);
+            assert(_Py_OPCODE(next_instr[1]) == POP_JUMP_IF_FALSE ||
+                   _Py_OPCODE(next_instr[1]) == POP_JUMP_IF_TRUE);
+            bool jump_on_true = _Py_OPCODE(next_instr[1]) == POP_JUMP_IF_TRUE;
+            int offset = _Py_OPARG(next_instr[1]);
+            int err = PyObject_IsTrue(cond);
+            Py_DECREF(cond);
+            if (err < 0) {
+                goto error;
+            }
+            if (jump_on_true == (err != 0)) {
+                JUMPBY(offset);
+            }
         }
 
-        // The result is an int disguised as an object pointer.
-        op(_COMPARE_OP_FLOAT, (unused/1, left, right -- jump: size_t)) {
+        inst(COMPARE_AND_BRANCH_FLOAT, (unused/2, left, right -- )) {
             assert(cframe.use_tracing == 0);
-            // Combined: COMPARE_OP (float ? float) + POP_JUMP_IF_(true/false)
-            DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_OP);
-            DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_OP);
-            STAT_INC(COMPARE_OP, hit);
+            DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_AND_BRANCH);
+            DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_AND_BRANCH);
+            STAT_INC(COMPARE_AND_BRANCH, 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);
-            jump = sign_ish & oparg;
-        }
-        // The input is an int disguised as an object pointer!
-        op(_JUMP_IF, (jump: size_t --)) {
-            assert(opcode == POP_JUMP_IF_FALSE || opcode == POP_JUMP_IF_TRUE);
-            if (jump) {
-                JUMPBY(oparg);
+            if (sign_ish & oparg) {
+                int offset = _Py_OPARG(next_instr[1]);
+                JUMPBY(offset);
             }
         }
-        // We're praying that the compiler optimizes the flags manipuations.
-        super(COMPARE_OP_FLOAT_JUMP) = _COMPARE_OP_FLOAT + _JUMP_IF;
 
-        // Similar to COMPARE_OP_FLOAT
-        op(_COMPARE_OP_INT, (unused/1, left, right -- jump: size_t)) {
+        // Similar to COMPARE_AND_BRANCH_FLOAT
+        inst(COMPARE_AND_BRANCH_INT, (unused/2, left, right -- )) {
             assert(cframe.use_tracing == 0);
-            // Combined: COMPARE_OP (int ? int) + POP_JUMP_IF_(true/false)
-            DEOPT_IF(!PyLong_CheckExact(left), COMPARE_OP);
-            DEOPT_IF(!PyLong_CheckExact(right), COMPARE_OP);
-            DEOPT_IF((size_t)(Py_SIZE(left) + 1) > 2, COMPARE_OP);
-            DEOPT_IF((size_t)(Py_SIZE(right) + 1) > 2, COMPARE_OP);
-            STAT_INC(COMPARE_OP, hit);
+            DEOPT_IF(!PyLong_CheckExact(left), COMPARE_AND_BRANCH);
+            DEOPT_IF(!PyLong_CheckExact(right), COMPARE_AND_BRANCH);
+            DEOPT_IF((size_t)(Py_SIZE(left) + 1) > 2, COMPARE_AND_BRANCH);
+            DEOPT_IF((size_t)(Py_SIZE(right) + 1) > 2, COMPARE_AND_BRANCH);
+            STAT_INC(COMPARE_AND_BRANCH, hit);
             assert(Py_ABS(Py_SIZE(left)) <= 1 && Py_ABS(Py_SIZE(right)) <= 1);
             Py_ssize_t ileft = Py_SIZE(left) * ((PyLongObject *)left)->ob_digit[0];
             Py_ssize_t iright = Py_SIZE(right) * ((PyLongObject *)right)->ob_digit[0];
@@ -1894,17 +1901,18 @@ dummy_func(
             int sign_ish = COMPARISON_BIT(ileft, iright);
             _Py_DECREF_SPECIALIZED(left, (destructor)PyObject_Free);
             _Py_DECREF_SPECIALIZED(right, (destructor)PyObject_Free);
-            jump = sign_ish & oparg;
+            if (sign_ish & oparg) {
+                int offset = _Py_OPARG(next_instr[1]);
+                JUMPBY(offset);
+            }
         }
-        super(COMPARE_OP_INT_JUMP) = _COMPARE_OP_INT + _JUMP_IF;
 
-        // Similar to COMPARE_OP_FLOAT, but for ==, != only
-        op(_COMPARE_OP_STR, (unused/1, left, right -- jump: size_t)) {
+        // Similar to COMPARE_AND_BRANCH_FLOAT, but for ==, != only
+        inst(COMPARE_AND_BRANCH_STR, (unused/2, left, right -- )) {
             assert(cframe.use_tracing == 0);
-            // Combined: COMPARE_OP (str == str or str != str) + POP_JUMP_IF_(true/false)
-            DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_OP);
-            DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_OP);
-            STAT_INC(COMPARE_OP, hit);
+            DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_AND_BRANCH);
+            DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_AND_BRANCH);
+            STAT_INC(COMPARE_AND_BRANCH, hit);
             int res = _PyUnicode_Equal(left, right);
             assert((oparg >>4) == Py_EQ || (oparg >>4) == Py_NE);
             _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc);
@@ -1912,11 +1920,12 @@ dummy_func(
             assert(res == 0 || res == 1);
             assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS);
             assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS);
-            jump = (res + COMPARISON_NOT_EQUALS) & oparg;
+            if ((res + COMPARISON_NOT_EQUALS) & oparg) {
+                int offset = _Py_OPARG(next_instr[1]);
+                JUMPBY(offset);
+            }
         }
 
-        super(COMPARE_OP_STR_JUMP) = _COMPARE_OP_STR + _JUMP_IF;
-
         inst(IS_OP, (left, right -- b)) {
             int res = Py_Is(left, right) ^ oparg;
             DECREF_INPUTS();
diff --git a/Python/compile.c b/Python/compile.c
index c0177fb..f98892e 100644
--- a/Python/compile.c
+++ b/Python/compile.c
@@ -1195,6 +1195,9 @@ stack_effect(int opcode, int oparg, int jump)
         case POP_JUMP_IF_TRUE:
             return -1;
 
+        case COMPARE_AND_BRANCH:
+            return -2;
+
         case LOAD_GLOBAL:
             return (oparg & 1) + 1;
 
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 9874ddf..e9819df 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -2077,19 +2077,10 @@
         }
 
         TARGET(COMPARE_OP) {
-            PREDICTED(COMPARE_OP);
             PyObject *right = PEEK(1);
             PyObject *left = PEEK(2);
             PyObject *res;
-            _PyCompareOpCache *cache = (_PyCompareOpCache *)next_instr;
-            if (ADAPTIVE_COUNTER_IS_ZERO(cache->counter)) {
-                assert(cframe.use_tracing == 0);
-                next_instr--;
-                _Py_Specialize_CompareOp(left, right, next_instr, oparg);
-                DISPATCH_SAME_OPARG();
-            }
             STAT_INC(COMPARE_OP, deferred);
-            DECREMENT_ADAPTIVE_COUNTER(cache->counter);
             assert((oparg >> 4) <= Py_GE);
             res = PyObject_RichCompare(left, right, oparg>>4);
             Py_DECREF(left);
@@ -2101,112 +2092,108 @@
             DISPATCH();
         }
 
-        TARGET(COMPARE_OP_FLOAT_JUMP) {
-            PyObject *_tmp_1 = PEEK(1);
-            PyObject *_tmp_2 = PEEK(2);
-            {
-                PyObject *right = _tmp_1;
-                PyObject *left = _tmp_2;
-                size_t jump;
+        TARGET(COMPARE_AND_BRANCH) {
+            PREDICTED(COMPARE_AND_BRANCH);
+            PyObject *right = PEEK(1);
+            PyObject *left = PEEK(2);
+            _PyCompareOpCache *cache = (_PyCompareOpCache *)next_instr;
+            if (ADAPTIVE_COUNTER_IS_ZERO(cache->counter)) {
                 assert(cframe.use_tracing == 0);
-                // Combined: COMPARE_OP (float ? float) + POP_JUMP_IF_(true/false)
-                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);
-                jump = sign_ish & oparg;
-                _tmp_2 = (PyObject *)jump;
+                next_instr--;
+                _Py_Specialize_CompareAndBranch(left, right, next_instr, oparg);
+                DISPATCH_SAME_OPARG();
             }
-            JUMPBY(1);
-            NEXTOPARG();
-            JUMPBY(1);
-            {
-                size_t jump = (size_t)_tmp_2;
-                assert(opcode == POP_JUMP_IF_FALSE || opcode == POP_JUMP_IF_TRUE);
-                if (jump) {
-                    JUMPBY(oparg);
-                }
+            STAT_INC(COMPARE_AND_BRANCH, deferred);
+            DECREMENT_ADAPTIVE_COUNTER(cache->counter);
+            assert((oparg >> 4) <= Py_GE);
+            PyObject *cond = PyObject_RichCompare(left, right, oparg>>4);
+            Py_DECREF(left);
+            Py_DECREF(right);
+            if (cond == NULL) goto pop_2_error;
+            assert(_Py_OPCODE(next_instr[1]) == POP_JUMP_IF_FALSE ||
+                   _Py_OPCODE(next_instr[1]) == POP_JUMP_IF_TRUE);
+            bool jump_on_true = _Py_OPCODE(next_instr[1]) == POP_JUMP_IF_TRUE;
+            int offset = _Py_OPARG(next_instr[1]);
+            int err = PyObject_IsTrue(cond);
+            Py_DECREF(cond);
+            if (err < 0) {
+                goto error;
+            }
+            if (jump_on_true == (err != 0)) {
+                JUMPBY(offset);
             }
             STACK_SHRINK(2);
+            JUMPBY(2);
             DISPATCH();
         }
 
-        TARGET(COMPARE_OP_INT_JUMP) {
-            PyObject *_tmp_1 = PEEK(1);
-            PyObject *_tmp_2 = PEEK(2);
-            {
-                PyObject *right = _tmp_1;
-                PyObject *left = _tmp_2;
-                size_t jump;
-                assert(cframe.use_tracing == 0);
-                // Combined: COMPARE_OP (int ? int) + POP_JUMP_IF_(true/false)
-                DEOPT_IF(!PyLong_CheckExact(left), COMPARE_OP);
-                DEOPT_IF(!PyLong_CheckExact(right), COMPARE_OP);
-                DEOPT_IF((size_t)(Py_SIZE(left) + 1) > 2, COMPARE_OP);
-                DEOPT_IF((size_t)(Py_SIZE(right) + 1) > 2, COMPARE_OP);
-                STAT_INC(COMPARE_OP, hit);
-                assert(Py_ABS(Py_SIZE(left)) <= 1 && Py_ABS(Py_SIZE(right)) <= 1);
-                Py_ssize_t ileft = Py_SIZE(left) * ((PyLongObject *)left)->ob_digit[0];
-                Py_ssize_t iright = Py_SIZE(right) * ((PyLongObject *)right)->ob_digit[0];
-                // 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);
-                jump = sign_ish & oparg;
-                _tmp_2 = (PyObject *)jump;
-            }
-            JUMPBY(1);
-            NEXTOPARG();
-            JUMPBY(1);
-            {
-                size_t jump = (size_t)_tmp_2;
-                assert(opcode == POP_JUMP_IF_FALSE || opcode == POP_JUMP_IF_TRUE);
-                if (jump) {
-                    JUMPBY(oparg);
-                }
+        TARGET(COMPARE_AND_BRANCH_FLOAT) {
+            PyObject *right = PEEK(1);
+            PyObject *left = PEEK(2);
+            assert(cframe.use_tracing == 0);
+            DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_AND_BRANCH);
+            DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_AND_BRANCH);
+            STAT_INC(COMPARE_AND_BRANCH, 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);
+            if (sign_ish & oparg) {
+                int offset = _Py_OPARG(next_instr[1]);
+                JUMPBY(offset);
             }
             STACK_SHRINK(2);
+            JUMPBY(2);
             DISPATCH();
         }
 
-        TARGET(COMPARE_OP_STR_JUMP) {
-            PyObject *_tmp_1 = PEEK(1);
-            PyObject *_tmp_2 = PEEK(2);
-            {
-                PyObject *right = _tmp_1;
-                PyObject *left = _tmp_2;
-                size_t jump;
-                assert(cframe.use_tracing == 0);
-                // Combined: COMPARE_OP (str == str or str != str) + POP_JUMP_IF_(true/false)
-                DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_OP);
-                DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_OP);
-                STAT_INC(COMPARE_OP, hit);
-                int res = _PyUnicode_Equal(left, right);
-                assert((oparg >>4) == Py_EQ || (oparg >>4) == Py_NE);
-                _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc);
-                _Py_DECREF_SPECIALIZED(right, _PyUnicode_ExactDealloc);
-                assert(res == 0 || res == 1);
-                assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS);
-                assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS);
-                jump = (res + COMPARISON_NOT_EQUALS) & oparg;
-                _tmp_2 = (PyObject *)jump;
+        TARGET(COMPARE_AND_BRANCH_INT) {
+            PyObject *right = PEEK(1);
+            PyObject *left = PEEK(2);
+            assert(cframe.use_tracing == 0);
+            DEOPT_IF(!PyLong_CheckExact(left), COMPARE_AND_BRANCH);
+            DEOPT_IF(!PyLong_CheckExact(right), COMPARE_AND_BRANCH);
+            DEOPT_IF((size_t)(Py_SIZE(left) + 1) > 2, COMPARE_AND_BRANCH);
+            DEOPT_IF((size_t)(Py_SIZE(right) + 1) > 2, COMPARE_AND_BRANCH);
+            STAT_INC(COMPARE_AND_BRANCH, hit);
+            assert(Py_ABS(Py_SIZE(left)) <= 1 && Py_ABS(Py_SIZE(right)) <= 1);
+            Py_ssize_t ileft = Py_SIZE(left) * ((PyLongObject *)left)->ob_digit[0];
+            Py_ssize_t iright = Py_SIZE(right) * ((PyLongObject *)right)->ob_digit[0];
+            // 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);
+            if (sign_ish & oparg) {
+                int offset = _Py_OPARG(next_instr[1]);
+                JUMPBY(offset);
             }
-            JUMPBY(1);
-            NEXTOPARG();
-            JUMPBY(1);
-            {
-                size_t jump = (size_t)_tmp_2;
-                assert(opcode == POP_JUMP_IF_FALSE || opcode == POP_JUMP_IF_TRUE);
-                if (jump) {
-                    JUMPBY(oparg);
-                }
+            STACK_SHRINK(2);
+            JUMPBY(2);
+            DISPATCH();
+        }
+
+        TARGET(COMPARE_AND_BRANCH_STR) {
+            PyObject *right = PEEK(1);
+            PyObject *left = PEEK(2);
+            assert(cframe.use_tracing == 0);
+            DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_AND_BRANCH);
+            DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_AND_BRANCH);
+            STAT_INC(COMPARE_AND_BRANCH, hit);
+            int res = _PyUnicode_Equal(left, right);
+            assert((oparg >>4) == Py_EQ || (oparg >>4) == Py_NE);
+            _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc);
+            _Py_DECREF_SPECIALIZED(right, _PyUnicode_ExactDealloc);
+            assert(res == 0 || res == 1);
+            assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS);
+            assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS);
+            if ((res + COMPARISON_NOT_EQUALS) & oparg) {
+                int offset = _Py_OPARG(next_instr[1]);
+                JUMPBY(offset);
             }
             STACK_SHRINK(2);
+            JUMPBY(2);
             DISPATCH();
         }
 
diff --git a/Python/opcode_metadata.h b/Python/opcode_metadata.h
index 4c8b9dd..d293ada 100644
--- a/Python/opcode_metadata.h
+++ b/Python/opcode_metadata.h
@@ -2,7 +2,7 @@
 // from Python/bytecodes.c
 // Do not edit!
 enum Direction { DIR_NONE, DIR_READ, DIR_WRITE };
-enum InstructionFormat { INSTR_FMT_IB, INSTR_FMT_IBC, INSTR_FMT_IBC000, INSTR_FMT_IBCIB, INSTR_FMT_IBIB, INSTR_FMT_IX, INSTR_FMT_IXC, INSTR_FMT_IXC000 };
+enum InstructionFormat { INSTR_FMT_IB, INSTR_FMT_IBC, INSTR_FMT_IBC0, INSTR_FMT_IBC000, INSTR_FMT_IBIB, INSTR_FMT_IX, INSTR_FMT_IXC, INSTR_FMT_IXC000 };
 static const struct {
     short n_popped;
     short n_pushed;
@@ -113,9 +113,10 @@ static const struct {
     [STORE_ATTR_WITH_HINT] = { 2, 0, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC000 },
     [STORE_ATTR_SLOT] = { 2, 0, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IXC000 },
     [COMPARE_OP] = { 2, 1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC },
-    [COMPARE_OP_FLOAT_JUMP] = { 3, 1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBCIB },
-    [COMPARE_OP_INT_JUMP] = { 3, 1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBCIB },
-    [COMPARE_OP_STR_JUMP] = { 3, 1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBCIB },
+    [COMPARE_AND_BRANCH] = { 2, 0, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0 },
+    [COMPARE_AND_BRANCH_FLOAT] = { 2, 0, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0 },
+    [COMPARE_AND_BRANCH_INT] = { 2, 0, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0 },
+    [COMPARE_AND_BRANCH_STR] = { 2, 0, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC0 },
     [IS_OP] = { 2, 1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [CONTAINS_OP] = { 2, 1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [CHECK_EG_MATCH] = { -1, -1, DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IX },
diff --git a/Python/opcode_targets.h b/Python/opcode_targets.h
index 83b3af7..f1c3f3e 100644
--- a/Python/opcode_targets.h
+++ b/Python/opcode_targets.h
@@ -47,7 +47,7 @@ static void *opcode_targets[256] = {
     &&TARGET_CALL_NO_KW_STR_1,
     &&TARGET_CALL_NO_KW_TUPLE_1,
     &&TARGET_CALL_NO_KW_TYPE_1,
-    &&TARGET_COMPARE_OP_FLOAT_JUMP,
+    &&TARGET_COMPARE_AND_BRANCH_FLOAT,
     &&TARGET_WITH_EXCEPT_START,
     &&TARGET_GET_AITER,
     &&TARGET_GET_ANEXT,
@@ -55,8 +55,8 @@ static void *opcode_targets[256] = {
     &&TARGET_BEFORE_WITH,
     &&TARGET_END_ASYNC_FOR,
     &&TARGET_CLEANUP_THROW,
-    &&TARGET_COMPARE_OP_INT_JUMP,
-    &&TARGET_COMPARE_OP_STR_JUMP,
+    &&TARGET_COMPARE_AND_BRANCH_INT,
+    &&TARGET_COMPARE_AND_BRANCH_STR,
     &&TARGET_FOR_ITER_LIST,
     &&TARGET_FOR_ITER_TUPLE,
     &&TARGET_STORE_SUBSCR,
@@ -140,9 +140,9 @@ static void *opcode_targets[256] = {
     &&TARGET_STORE_DEREF,
     &&TARGET_DELETE_DEREF,
     &&TARGET_JUMP_BACKWARD,
-    &&TARGET_STORE_FAST__STORE_FAST,
+    &&TARGET_COMPARE_AND_BRANCH,
     &&TARGET_CALL_FUNCTION_EX,
-    &&TARGET_STORE_SUBSCR_DICT,
+    &&TARGET_STORE_FAST__STORE_FAST,
     &&TARGET_EXTENDED_ARG,
     &&TARGET_LIST_APPEND,
     &&TARGET_SET_ADD,
@@ -152,15 +152,15 @@ static void *opcode_targets[256] = {
     &&TARGET_YIELD_VALUE,
     &&TARGET_RESUME,
     &&TARGET_MATCH_CLASS,
+    &&TARGET_STORE_SUBSCR_DICT,
     &&TARGET_STORE_SUBSCR_LIST_INT,
-    &&TARGET_UNPACK_SEQUENCE_LIST,
     &&TARGET_FORMAT_VALUE,
     &&TARGET_BUILD_CONST_KEY_MAP,
     &&TARGET_BUILD_STRING,
+    &&TARGET_UNPACK_SEQUENCE_LIST,
     &&TARGET_UNPACK_SEQUENCE_TUPLE,
     &&TARGET_UNPACK_SEQUENCE_TWO_TUPLE,
     &&_unknown_opcode,
-    &&_unknown_opcode,
     &&TARGET_LIST_EXTEND,
     &&TARGET_SET_UPDATE,
     &&TARGET_DICT_MERGE,
diff --git a/Python/specialize.c b/Python/specialize.c
index 38bb151..500fd71 100644
--- a/Python/specialize.c
+++ b/Python/specialize.c
@@ -126,6 +126,7 @@ print_spec_stats(FILE *out, OpcodeStats *stats)
     /* Mark some opcodes as specializable for stats,
      * even though we don't specialize them yet. */
     fprintf(out, "opcode[%d].specializable : 1\n", BINARY_SLICE);
+    fprintf(out, "opcode[%d].specializable : 1\n", COMPARE_OP);
     fprintf(out, "opcode[%d].specializable : 1\n", STORE_SLICE);
     for (int i = 0; i < 256; i++) {
         if (_PyOpcode_Caches[i]) {
@@ -311,6 +312,7 @@ _PyCode_Quicken(PyCodeObject *code)
                 if (opcode == POP_JUMP_IF_FALSE) {
                     mask = mask ^ 0xf;
                 }
+                instructions[i - 1 - INLINE_CACHE_ENTRIES_COMPARE_OP].opcode = COMPARE_AND_BRANCH;
                 instructions[i - 1 - INLINE_CACHE_ENTRIES_COMPARE_OP].oparg = (oparg & 0xf0) | mask;
                 break;
             }
@@ -431,19 +433,19 @@ _PyCode_Quicken(PyCodeObject *code)
 #define SPEC_FAIL_CALL_OPERATOR_WRAPPER 29
 
 /* COMPARE_OP */
-#define SPEC_FAIL_COMPARE_OP_DIFFERENT_TYPES 12
-#define SPEC_FAIL_COMPARE_OP_STRING 13
-#define SPEC_FAIL_COMPARE_OP_NOT_FOLLOWED_BY_COND_JUMP 14
-#define SPEC_FAIL_COMPARE_OP_BIG_INT 15
-#define SPEC_FAIL_COMPARE_OP_BYTES 16
-#define SPEC_FAIL_COMPARE_OP_TUPLE 17
-#define SPEC_FAIL_COMPARE_OP_LIST 18
-#define SPEC_FAIL_COMPARE_OP_SET 19
-#define SPEC_FAIL_COMPARE_OP_BOOL 20
-#define SPEC_FAIL_COMPARE_OP_BASEOBJECT 21
-#define SPEC_FAIL_COMPARE_OP_FLOAT_LONG 22
-#define SPEC_FAIL_COMPARE_OP_LONG_FLOAT 23
-#define SPEC_FAIL_COMPARE_OP_EXTENDED_ARG 24
+#define SPEC_FAIL_COMPARE_DIFFERENT_TYPES 12
+#define SPEC_FAIL_COMPARE_STRING 13
+#define SPEC_FAIL_COMPARE_NOT_FOLLOWED_BY_COND_JUMP 14
+#define SPEC_FAIL_COMPARE_BIG_INT 15
+#define SPEC_FAIL_COMPARE_BYTES 16
+#define SPEC_FAIL_COMPARE_TUPLE 17
+#define SPEC_FAIL_COMPARE_LIST 18
+#define SPEC_FAIL_COMPARE_SET 19
+#define SPEC_FAIL_COMPARE_BOOL 20
+#define SPEC_FAIL_COMPARE_BASEOBJECT 21
+#define SPEC_FAIL_COMPARE_FLOAT_LONG 22
+#define SPEC_FAIL_COMPARE_LONG_FLOAT 23
+#define SPEC_FAIL_COMPARE_EXTENDED_ARG 24
 
 /* FOR_ITER */
 #define SPEC_FAIL_FOR_ITER_GENERATOR 10
@@ -1968,87 +1970,82 @@ compare_op_fail_kind(PyObject *lhs, PyObject *rhs)
 {
     if (Py_TYPE(lhs) != Py_TYPE(rhs)) {
         if (PyFloat_CheckExact(lhs) && PyLong_CheckExact(rhs)) {
-            return SPEC_FAIL_COMPARE_OP_FLOAT_LONG;
+            return SPEC_FAIL_COMPARE_FLOAT_LONG;
         }
         if (PyLong_CheckExact(lhs) && PyFloat_CheckExact(rhs)) {
-            return SPEC_FAIL_COMPARE_OP_LONG_FLOAT;
+            return SPEC_FAIL_COMPARE_LONG_FLOAT;
         }
-        return SPEC_FAIL_COMPARE_OP_DIFFERENT_TYPES;
+        return SPEC_FAIL_COMPARE_DIFFERENT_TYPES;
     }
     if (PyBytes_CheckExact(lhs)) {
-        return SPEC_FAIL_COMPARE_OP_BYTES;
+        return SPEC_FAIL_COMPARE_BYTES;
     }
     if (PyTuple_CheckExact(lhs)) {
-        return SPEC_FAIL_COMPARE_OP_TUPLE;
+        return SPEC_FAIL_COMPARE_TUPLE;
     }
     if (PyList_CheckExact(lhs)) {
-        return SPEC_FAIL_COMPARE_OP_LIST;
+        return SPEC_FAIL_COMPARE_LIST;
     }
     if (PySet_CheckExact(lhs) || PyFrozenSet_CheckExact(lhs)) {
-        return SPEC_FAIL_COMPARE_OP_SET;
+        return SPEC_FAIL_COMPARE_SET;
     }
     if (PyBool_Check(lhs)) {
-        return SPEC_FAIL_COMPARE_OP_BOOL;
+        return SPEC_FAIL_COMPARE_BOOL;
     }
     if (Py_TYPE(lhs)->tp_richcompare == PyBaseObject_Type.tp_richcompare) {
-        return SPEC_FAIL_COMPARE_OP_BASEOBJECT;
+        return SPEC_FAIL_COMPARE_BASEOBJECT;
     }
     return SPEC_FAIL_OTHER;
 }
 #endif
 
 void
-_Py_Specialize_CompareOp(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
+_Py_Specialize_CompareAndBranch(PyObject *lhs, PyObject *rhs, _Py_CODEUNIT *instr,
                          int oparg)
 {
-    assert(_PyOpcode_Caches[COMPARE_OP] == INLINE_CACHE_ENTRIES_COMPARE_OP);
+    assert(_PyOpcode_Caches[COMPARE_AND_BRANCH] == INLINE_CACHE_ENTRIES_COMPARE_OP);
     _PyCompareOpCache *cache = (_PyCompareOpCache *)(instr + 1);
+#ifndef NDEBUG
     int next_opcode = _Py_OPCODE(instr[INLINE_CACHE_ENTRIES_COMPARE_OP + 1]);
-    if (next_opcode != POP_JUMP_IF_FALSE && next_opcode != POP_JUMP_IF_TRUE) {
-        if (next_opcode == EXTENDED_ARG) {
-            SPECIALIZATION_FAIL(COMPARE_OP, SPEC_FAIL_COMPARE_OP_EXTENDED_ARG);
-            goto failure;
-        }
-        SPECIALIZATION_FAIL(COMPARE_OP, SPEC_FAIL_COMPARE_OP_NOT_FOLLOWED_BY_COND_JUMP);
-        goto failure;
-    }
+    assert(next_opcode == POP_JUMP_IF_FALSE || next_opcode == POP_JUMP_IF_TRUE);
+#endif
     if (Py_TYPE(lhs) != Py_TYPE(rhs)) {
-        SPECIALIZATION_FAIL(COMPARE_OP, compare_op_fail_kind(lhs, rhs));
+        SPECIALIZATION_FAIL(COMPARE_AND_BRANCH, compare_op_fail_kind(lhs, rhs));
         goto failure;
     }
     if (PyFloat_CheckExact(lhs)) {
-        _py_set_opcode(instr, COMPARE_OP_FLOAT_JUMP);
+        _py_set_opcode(instr, COMPARE_AND_BRANCH_FLOAT);
         goto success;
     }
     if (PyLong_CheckExact(lhs)) {
         if (Py_ABS(Py_SIZE(lhs)) <= 1 && Py_ABS(Py_SIZE(rhs)) <= 1) {
-            _py_set_opcode(instr, COMPARE_OP_INT_JUMP);
+            _py_set_opcode(instr, COMPARE_AND_BRANCH_INT);
             goto success;
         }
         else {
-            SPECIALIZATION_FAIL(COMPARE_OP, SPEC_FAIL_COMPARE_OP_BIG_INT);
+            SPECIALIZATION_FAIL(COMPARE_AND_BRANCH, SPEC_FAIL_COMPARE_BIG_INT);
             goto failure;
         }
     }
     if (PyUnicode_CheckExact(lhs)) {
         int cmp = oparg >> 4;
         if (cmp != Py_EQ && cmp != Py_NE) {
-            SPECIALIZATION_FAIL(COMPARE_OP, SPEC_FAIL_COMPARE_OP_STRING);
+            SPECIALIZATION_FAIL(COMPARE_AND_BRANCH, SPEC_FAIL_COMPARE_STRING);
             goto failure;
         }
         else {
-            _py_set_opcode(instr, COMPARE_OP_STR_JUMP);
+            _py_set_opcode(instr, COMPARE_AND_BRANCH_STR);
             goto success;
         }
     }
-    SPECIALIZATION_FAIL(COMPARE_OP, compare_op_fail_kind(lhs, rhs));
+    SPECIALIZATION_FAIL(COMPARE_AND_BRANCH, compare_op_fail_kind(lhs, rhs));
 failure:
-    STAT_INC(COMPARE_OP, failure);
-    _py_set_opcode(instr, COMPARE_OP);
+    STAT_INC(COMPARE_AND_BRANCH, failure);
+    _py_set_opcode(instr, COMPARE_AND_BRANCH);
     cache->counter = adaptive_counter_backoff(cache->counter);
     return;
 success:
-    STAT_INC(COMPARE_OP, success);
+    STAT_INC(COMPARE_AND_BRANCH, success);
     cache->counter = adaptive_counter_cooldown();
 }