gh-98831: Finish the UNPACK_SEQUENCE family (#101666)

New generator feature: Generate useful glue for output arrays, so you can just write values to the output array (no bounds checking). Rewrote UNPACK_SEQUENCE_TWO_TUPLE to use this, and also UNPACK_SEQUENCE_{TUPLE,LIST}.

3 files changed, 46 insertions, 45 deletions

diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index 0d7d922..c6c00a7 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -876,6 +876,13 @@ dummy_func(
             }
         }
 
+        family(unpack_sequence, INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE) = {
+            UNPACK_SEQUENCE,
+            UNPACK_SEQUENCE_TWO_TUPLE,
+            UNPACK_SEQUENCE_TUPLE,
+            UNPACK_SEQUENCE_LIST,
+        };
+
         inst(UNPACK_SEQUENCE, (unused/1, seq -- unused[oparg])) {
             #if ENABLE_SPECIALIZATION
             _PyUnpackSequenceCache *cache = (_PyUnpackSequenceCache *)next_instr;
@@ -894,43 +901,36 @@ dummy_func(
             ERROR_IF(res == 0, error);
         }
 
-        inst(UNPACK_SEQUENCE_TWO_TUPLE, (unused/1, seq -- v1, v0)) {
+        inst(UNPACK_SEQUENCE_TWO_TUPLE, (unused/1, seq -- values[oparg])) {
             DEOPT_IF(!PyTuple_CheckExact(seq), UNPACK_SEQUENCE);
             DEOPT_IF(PyTuple_GET_SIZE(seq) != 2, UNPACK_SEQUENCE);
+            assert(oparg == 2);
             STAT_INC(UNPACK_SEQUENCE, hit);
-            v1 = Py_NewRef(PyTuple_GET_ITEM(seq, 1));
-            v0 = Py_NewRef(PyTuple_GET_ITEM(seq, 0));
+            values[0] = Py_NewRef(PyTuple_GET_ITEM(seq, 1));
+            values[1] = Py_NewRef(PyTuple_GET_ITEM(seq, 0));
             Py_DECREF(seq);
         }
 
-        // stack effect: (__0 -- __array[oparg])
-        inst(UNPACK_SEQUENCE_TUPLE) {
-            PyObject *seq = TOP();
+        inst(UNPACK_SEQUENCE_TUPLE, (unused/1, seq -- values[oparg])) {
             DEOPT_IF(!PyTuple_CheckExact(seq), UNPACK_SEQUENCE);
             DEOPT_IF(PyTuple_GET_SIZE(seq) != oparg, UNPACK_SEQUENCE);
             STAT_INC(UNPACK_SEQUENCE, hit);
-            STACK_SHRINK(1);
             PyObject **items = _PyTuple_ITEMS(seq);
-            while (oparg--) {
-                PUSH(Py_NewRef(items[oparg]));
+            for (int i = oparg; --i >= 0; ) {
+                *values++ = Py_NewRef(items[i]);
             }
             Py_DECREF(seq);
-            JUMPBY(INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE);
         }
 
-        // stack effect: (__0 -- __array[oparg])
-        inst(UNPACK_SEQUENCE_LIST) {
-            PyObject *seq = TOP();
+        inst(UNPACK_SEQUENCE_LIST, (unused/1, seq -- values[oparg])) {
             DEOPT_IF(!PyList_CheckExact(seq), UNPACK_SEQUENCE);
             DEOPT_IF(PyList_GET_SIZE(seq) != oparg, UNPACK_SEQUENCE);
             STAT_INC(UNPACK_SEQUENCE, hit);
-            STACK_SHRINK(1);
             PyObject **items = _PyList_ITEMS(seq);
-            while (oparg--) {
-                PUSH(Py_NewRef(items[oparg]));
+            for (int i = oparg; --i >= 0; ) {
+                *values++ = Py_NewRef(items[i]);
             }
             Py_DECREF(seq);
-            JUMPBY(INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE);
         }
 
         inst(UNPACK_EX, (seq -- unused[oparg & 0xFF], unused, unused[oparg >> 8])) {
@@ -3185,6 +3185,3 @@ family(call, INLINE_CACHE_ENTRIES_CALL) = {
     CALL_NO_KW_METHOD_DESCRIPTOR_O, CALL_NO_KW_STR_1, CALL_NO_KW_TUPLE_1,
     CALL_NO_KW_TYPE_1 };
 family(store_fast) = { STORE_FAST, STORE_FAST__LOAD_FAST, STORE_FAST__STORE_FAST };
-family(unpack_sequence, INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE) = {
-    UNPACK_SEQUENCE, UNPACK_SEQUENCE_LIST,
-    UNPACK_SEQUENCE_TUPLE, UNPACK_SEQUENCE_TWO_TUPLE };
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index de98b1a..ded68d0 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -1125,6 +1125,7 @@
 
         TARGET(UNPACK_SEQUENCE) {
             PREDICTED(UNPACK_SEQUENCE);
+            static_assert(INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE == 1, "incorrect cache size");
             PyObject *seq = PEEK(1);
             #if ENABLE_SPECIALIZATION
             _PyUnpackSequenceCache *cache = (_PyUnpackSequenceCache *)next_instr;
@@ -1149,48 +1150,51 @@
 
         TARGET(UNPACK_SEQUENCE_TWO_TUPLE) {
             PyObject *seq = PEEK(1);
-            PyObject *v1;
-            PyObject *v0;
+            PyObject **values = stack_pointer - (1);
             DEOPT_IF(!PyTuple_CheckExact(seq), UNPACK_SEQUENCE);
             DEOPT_IF(PyTuple_GET_SIZE(seq) != 2, UNPACK_SEQUENCE);
+            assert(oparg == 2);
             STAT_INC(UNPACK_SEQUENCE, hit);
-            v1 = Py_NewRef(PyTuple_GET_ITEM(seq, 1));
-            v0 = Py_NewRef(PyTuple_GET_ITEM(seq, 0));
+            values[0] = Py_NewRef(PyTuple_GET_ITEM(seq, 1));
+            values[1] = Py_NewRef(PyTuple_GET_ITEM(seq, 0));
             Py_DECREF(seq);
-            STACK_GROW(1);
-            POKE(1, v0);
-            POKE(2, v1);
+            STACK_SHRINK(1);
+            STACK_GROW(oparg);
             JUMPBY(1);
             DISPATCH();
         }
 
         TARGET(UNPACK_SEQUENCE_TUPLE) {
-            PyObject *seq = TOP();
+            PyObject *seq = PEEK(1);
+            PyObject **values = stack_pointer - (1);
             DEOPT_IF(!PyTuple_CheckExact(seq), UNPACK_SEQUENCE);
             DEOPT_IF(PyTuple_GET_SIZE(seq) != oparg, UNPACK_SEQUENCE);
             STAT_INC(UNPACK_SEQUENCE, hit);
-            STACK_SHRINK(1);
             PyObject **items = _PyTuple_ITEMS(seq);
-            while (oparg--) {
-                PUSH(Py_NewRef(items[oparg]));
+            for (int i = oparg; --i >= 0; ) {
+                *values++ = Py_NewRef(items[i]);
             }
             Py_DECREF(seq);
-            JUMPBY(INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE);
+            STACK_SHRINK(1);
+            STACK_GROW(oparg);
+            JUMPBY(1);
             DISPATCH();
         }
 
         TARGET(UNPACK_SEQUENCE_LIST) {
-            PyObject *seq = TOP();
+            PyObject *seq = PEEK(1);
+            PyObject **values = stack_pointer - (1);
             DEOPT_IF(!PyList_CheckExact(seq), UNPACK_SEQUENCE);
             DEOPT_IF(PyList_GET_SIZE(seq) != oparg, UNPACK_SEQUENCE);
             STAT_INC(UNPACK_SEQUENCE, hit);
-            STACK_SHRINK(1);
             PyObject **items = _PyList_ITEMS(seq);
-            while (oparg--) {
-                PUSH(Py_NewRef(items[oparg]));
+            for (int i = oparg; --i >= 0; ) {
+                *values++ = Py_NewRef(items[i]);
             }
             Py_DECREF(seq);
-            JUMPBY(INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE);
+            STACK_SHRINK(1);
+            STACK_GROW(oparg);
+            JUMPBY(1);
             DISPATCH();
         }
 
diff --git a/Python/opcode_metadata.h b/Python/opcode_metadata.h
index bae5492..c1e12a4 100644
--- a/Python/opcode_metadata.h
+++ b/Python/opcode_metadata.h
@@ -127,9 +127,9 @@ _PyOpcode_num_popped(int opcode, int oparg, bool jump) {
         case UNPACK_SEQUENCE_TWO_TUPLE:
             return 1;
         case UNPACK_SEQUENCE_TUPLE:
-            return -1;
+            return 1;
         case UNPACK_SEQUENCE_LIST:
-            return -1;
+            return 1;
         case UNPACK_EX:
             return 1;
         case STORE_ATTR:
@@ -473,11 +473,11 @@ _PyOpcode_num_pushed(int opcode, int oparg, bool jump) {
         case UNPACK_SEQUENCE:
             return oparg;
         case UNPACK_SEQUENCE_TWO_TUPLE:
-            return 2;
+            return oparg;
         case UNPACK_SEQUENCE_TUPLE:
-            return -1;
+            return oparg;
         case UNPACK_SEQUENCE_LIST:
-            return -1;
+            return oparg;
         case UNPACK_EX:
             return (oparg & 0xFF) + (oparg >> 8) + 1;
         case STORE_ATTR:
@@ -765,9 +765,9 @@ struct opcode_metadata {
     [STORE_NAME] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [DELETE_NAME] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [UNPACK_SEQUENCE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC },
-    [UNPACK_SEQUENCE_TWO_TUPLE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IXC },
-    [UNPACK_SEQUENCE_TUPLE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
-    [UNPACK_SEQUENCE_LIST] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
+    [UNPACK_SEQUENCE_TWO_TUPLE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC },
+    [UNPACK_SEQUENCE_TUPLE] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC },
+    [UNPACK_SEQUENCE_LIST] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC },
     [UNPACK_EX] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },
     [STORE_ATTR] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IBC000 },
     [DELETE_ATTR] = { DIR_NONE, DIR_NONE, DIR_NONE, true, INSTR_FMT_IB },