GH-102300: Reuse objects with refcount == 1 in float specialized binary ops. (GH-102301)

4 files changed, 27 insertions, 24 deletions

diff --git a/Misc/NEWS.d/next/Core and Builtins/2023-02-27-15-48-31.gh-issue-102300.8o-_Mt.rst b/Misc/NEWS.d/next/Core and Builtins/2023-02-27-15-48-31.gh-issue-102300.8o-_Mt.rst
new file mode 100644
index 0000000000..4227014582
--- /dev/null
+++ b/Misc/NEWS.d/next/Core and Builtins/2023-02-27-15-48-31.gh-issue-102300.8o-_Mt.rst
@@ -0,0 +1 @@
+Reuse operands with refcount of 1 in float specializations of BINARY_OP.
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index dc2a20f22b..178519fcb6 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -195,10 +195,7 @@ dummy_func(
             STAT_INC(BINARY_OP, hit);
             double dprod = ((PyFloatObject *)left)->ob_fval *
                 ((PyFloatObject *)right)->ob_fval;
-            prod = PyFloat_FromDouble(dprod);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            ERROR_IF(prod == NULL, error);
+            DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dprod, prod);
         }
 
         inst(BINARY_OP_SUBTRACT_INT, (unused/1, left, right -- sub)) {
@@ -218,10 +215,7 @@ dummy_func(
             DEOPT_IF(!PyFloat_CheckExact(right), BINARY_OP);
             STAT_INC(BINARY_OP, hit);
             double dsub = ((PyFloatObject *)left)->ob_fval - ((PyFloatObject *)right)->ob_fval;
-            sub = PyFloat_FromDouble(dsub);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            ERROR_IF(sub == NULL, error);
+            DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dsub, sub);
         }
 
         inst(BINARY_OP_ADD_UNICODE, (unused/1, left, right -- res)) {
@@ -278,10 +272,7 @@ dummy_func(
             STAT_INC(BINARY_OP, hit);
             double dsum = ((PyFloatObject *)left)->ob_fval +
                 ((PyFloatObject *)right)->ob_fval;
-            sum = PyFloat_FromDouble(dsum);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            ERROR_IF(sum == NULL, error);
+            DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dsum, sum);
         }
 
         inst(BINARY_OP_ADD_INT, (unused/1, left, right -- sum)) {
diff --git a/Python/ceval_macros.h b/Python/ceval_macros.h
index ac1fec77da..98b72ec1b3 100644
--- a/Python/ceval_macros.h
+++ b/Python/ceval_macros.h
@@ -350,3 +350,23 @@ GETITEM(PyObject *v, Py_ssize_t i) {
 
 #define KWNAMES_LEN() \
     (kwnames == NULL ? 0 : ((int)PyTuple_GET_SIZE(kwnames)))
+
+#define DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dval, result) \
+do { \
+    if (Py_REFCNT(left) == 1) { \
+        ((PyFloatObject *)left)->ob_fval = (dval); \
+        _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);\
+        result = (left); \
+    } \
+    else if (Py_REFCNT(right) == 1)  {\
+        ((PyFloatObject *)right)->ob_fval = (dval); \
+        _Py_DECREF_NO_DEALLOC(left); \
+        result = (right); \
+    }\
+    else { \
+        result = PyFloat_FromDouble(dval); \
+        if ((result) == NULL) goto error; \
+        _Py_DECREF_NO_DEALLOC(left); \
+        _Py_DECREF_NO_DEALLOC(right); \
+    } \
+} while (0)
diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h
index 3839aeecb9..21073cbf97 100644
--- a/Python/generated_cases.c.h
+++ b/Python/generated_cases.c.h
@@ -263,10 +263,7 @@
             STAT_INC(BINARY_OP, hit);
             double dprod = ((PyFloatObject *)left)->ob_fval *
                 ((PyFloatObject *)right)->ob_fval;
-            prod = PyFloat_FromDouble(dprod);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            if (prod == NULL) goto pop_2_error;
+            DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dprod, prod);
             STACK_SHRINK(1);
             stack_pointer[-1] = prod;
             next_instr += 1;
@@ -300,10 +297,7 @@
             DEOPT_IF(!PyFloat_CheckExact(right), BINARY_OP);
             STAT_INC(BINARY_OP, hit);
             double dsub = ((PyFloatObject *)left)->ob_fval - ((PyFloatObject *)right)->ob_fval;
-            sub = PyFloat_FromDouble(dsub);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            if (sub == NULL) goto pop_2_error;
+            DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dsub, sub);
             STACK_SHRINK(1);
             stack_pointer[-1] = sub;
             next_instr += 1;
@@ -372,10 +366,7 @@
             STAT_INC(BINARY_OP, hit);
             double dsum = ((PyFloatObject *)left)->ob_fval +
                 ((PyFloatObject *)right)->ob_fval;
-            sum = PyFloat_FromDouble(dsum);
-            _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc);
-            _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc);
-            if (sum == NULL) goto pop_2_error;
+            DECREF_INPUTS_AND_REUSE_FLOAT(left, right, dsum, sum);
             STACK_SHRINK(1);
             stack_pointer[-1] = sum;
             next_instr += 1;