mirrors/cpython
1
0
Fork 0
mirror of https://github.com/python/cpython.git synced 2025-12-23 09:19:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

gh-134584: Remove custom float decref ops (GH-142576)

Browse source
This commit is contained in:
Ken Jin 2025-12-16 03:38:58 +08:00 • committed by GitHub
parent b3b644b6b8
commit 0ac4e6c6cd
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
15 changed files with 2606 additions and 3131 deletions
Download patch file Download diff file Expand all files Collapse all files

2
Include/internal/pycore_ceval.h
View file

@ -377,8 +377,6 @@ _Py_eval_breaker_bit_is_set(PyThreadState *tstate, uintptr_t bit)
void _Py_set_eval_breaker_bit_all(PyInterpreterState *interp, uintptr_t bit);
void _Py_unset_eval_breaker_bit_all(PyInterpreterState *interp, uintptr_t bit);
PyAPI_FUNC(_PyStackRef) _PyFloat_FromDouble_ConsumeInputs(_PyStackRef left, _PyStackRef right, double value);
#ifndef Py_SUPPORTS_REMOTE_DEBUG
#if defined(__APPLE__)
#include <TargetConditionals.h>

12
Include/internal/pycore_opcode_metadata.h generated
View file

@ -1079,12 +1079,12 @@ extern const struct opcode_metadata _PyOpcode_opcode_metadata[267];
#ifdef NEED_OPCODE_METADATA
const struct opcode_metadata _PyOpcode_opcode_metadata[267] = {
[BINARY_OP] = { true, INSTR_FMT_IBC0000, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_ADD_FLOAT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG },
[BINARY_OP_ADD_FLOAT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG },
[BINARY_OP_ADD_INT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG },
[BINARY_OP_ADD_UNICODE] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG },
[BINARY_OP_EXTEND] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_INPLACE_ADD_UNICODE] = { true, INSTR_FMT_IXC0000, HAS_LOCAL_FLAG | HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_MULTIPLY_FLOAT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG },
[BINARY_OP_MULTIPLY_FLOAT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG },
[BINARY_OP_MULTIPLY_INT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG },
[BINARY_OP_SUBSCR_DICT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_GETITEM] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_SYNC_SP_FLAG | HAS_NEEDS_GUARD_IP_FLAG },
@ -1092,7 +1092,7 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[267] = {
[BINARY_OP_SUBSCR_LIST_SLICE] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_STR_INT] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_TUPLE_INT] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBTRACT_FLOAT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG },
[BINARY_OP_SUBTRACT_FLOAT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG },
[BINARY_OP_SUBTRACT_INT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG },
[BINARY_SLICE] = { true, INSTR_FMT_IX, HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[BUILD_INTERPOLATION] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
@ -1329,12 +1329,12 @@ extern const struct opcode_macro_expansion _PyOpcode_macro_expansion[256];
const struct opcode_macro_expansion
_PyOpcode_macro_expansion[256] = {
[BINARY_OP] = { .nuops = 1, .uops = { { _BINARY_OP, OPARG_SIMPLE, 4 } } },
[BINARY_OP_ADD_FLOAT] = { .nuops = 3, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_FLOAT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_ADD_FLOAT] = { .nuops = 5, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_ADD_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_INT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_ADD_UNICODE] = { .nuops = 3, .uops = { { _GUARD_TOS_UNICODE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_ADD_UNICODE, OPARG_SIMPLE, 5 } } },
[BINARY_OP_EXTEND] = { .nuops = 2, .uops = { { _GUARD_BINARY_OP_EXTEND, 4, 1 }, { _BINARY_OP_EXTEND, 4, 1 } } },
[BINARY_OP_INPLACE_ADD_UNICODE] = { .nuops = 3, .uops = { { _GUARD_TOS_UNICODE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_INPLACE_ADD_UNICODE, OPARG_SIMPLE, 5 } } },
[BINARY_OP_MULTIPLY_FLOAT] = { .nuops = 3, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_MULTIPLY_FLOAT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_MULTIPLY_FLOAT] = { .nuops = 5, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_MULTIPLY_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_MULTIPLY_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_MULTIPLY_INT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_DICT] = { .nuops = 2, .uops = { { _GUARD_NOS_DICT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_DICT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_GETITEM] = { .nuops = 4, .uops = { { _CHECK_PEP_523, OPARG_SIMPLE, 5 }, { _BINARY_OP_SUBSCR_CHECK_FUNC, OPARG_SIMPLE, 5 }, { _BINARY_OP_SUBSCR_INIT_CALL, OPARG_SIMPLE, 5 }, { _PUSH_FRAME, OPARG_SIMPLE, 5 } } },
@ -1342,7 +1342,7 @@ _PyOpcode_macro_expansion[256] = {
[BINARY_OP_SUBSCR_LIST_SLICE] = { .nuops = 3, .uops = { { _GUARD_TOS_SLICE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_SLICE, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_STR_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_STR_INT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_TUPLE_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_TUPLE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_TUPLE_INT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBTRACT_FLOAT] = { .nuops = 3, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBTRACT_FLOAT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBTRACT_FLOAT] = { .nuops = 5, .uops = { { _GUARD_TOS_FLOAT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_FLOAT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBTRACT_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 }, { _POP_TOP_FLOAT, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBTRACT_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBTRACT_INT, OPARG_SIMPLE, 5 } } },
[BINARY_SLICE] = { .nuops = 1, .uops = { { _BINARY_SLICE, OPARG_SIMPLE, 0 } } },
[BUILD_INTERPOLATION] = { .nuops = 1, .uops = { { _BUILD_INTERPOLATION, OPARG_SIMPLE, 0 } } },

2000
Include/internal/pycore_uop_ids.h generated
View file

File diff suppressed because it is too large Load diff

141
Include/internal/pycore_uop_metadata.h generated
View file

@ -108,12 +108,9 @@ const uint32_t _PyUop_Flags[MAX_UOP_ID+1] = {
[_BINARY_OP_SUBTRACT_INT] = HAS_EXIT_FLAG | HAS_PURE_FLAG,
[_GUARD_NOS_FLOAT] = HAS_EXIT_FLAG,
[_GUARD_TOS_FLOAT] = HAS_EXIT_FLAG,
[_BINARY_OP_MULTIPLY_FLOAT] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_ADD_FLOAT] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_SUBTRACT_FLOAT] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_MULTIPLY_FLOAT] = HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_ADD_FLOAT] = HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_SUBTRACT_FLOAT] = HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_ADD_UNICODE] = HAS_ERROR_FLAG | HAS_PURE_FLAG,
[_BINARY_OP_INPLACE_ADD_UNICODE] = HAS_LOCAL_FLAG | HAS_DEOPT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_GUARD_BINARY_OP_EXTEND] = HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG,
@ -1026,57 +1023,30 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
},
},
[_BINARY_OP_MULTIPLY_FLOAT] = {
.best = { 0, 1, 2, 3 },
.best = { 0, 1, 2, 2 },
.entries = {
{ 1, 0, _BINARY_OP_MULTIPLY_FLOAT_r01 },
{ 1, 1, _BINARY_OP_MULTIPLY_FLOAT_r11 },
{ 1, 2, _BINARY_OP_MULTIPLY_FLOAT_r21 },
{ 2, 3, _BINARY_OP_MULTIPLY_FLOAT_r32 },
{ 3, 0, _BINARY_OP_MULTIPLY_FLOAT_r03 },
{ 3, 1, _BINARY_OP_MULTIPLY_FLOAT_r13 },
{ 3, 2, _BINARY_OP_MULTIPLY_FLOAT_r23 },
{ -1, -1, -1 },
},
},
[_BINARY_OP_ADD_FLOAT] = {
.best = { 0, 1, 2, 3 },
.best = { 0, 1, 2, 2 },
.entries = {
{ 1, 0, _BINARY_OP_ADD_FLOAT_r01 },
{ 1, 1, _BINARY_OP_ADD_FLOAT_r11 },
{ 1, 2, _BINARY_OP_ADD_FLOAT_r21 },
{ 2, 3, _BINARY_OP_ADD_FLOAT_r32 },
{ 3, 0, _BINARY_OP_ADD_FLOAT_r03 },
{ 3, 1, _BINARY_OP_ADD_FLOAT_r13 },
{ 3, 2, _BINARY_OP_ADD_FLOAT_r23 },
{ -1, -1, -1 },
},
},
[_BINARY_OP_SUBTRACT_FLOAT] = {
.best = { 0, 1, 2, 3 },
.best = { 0, 1, 2, 2 },
.entries = {
{ 1, 0, _BINARY_OP_SUBTRACT_FLOAT_r01 },
{ 1, 1, _BINARY_OP_SUBTRACT_FLOAT_r11 },
{ 1, 2, _BINARY_OP_SUBTRACT_FLOAT_r21 },
{ 2, 3, _BINARY_OP_SUBTRACT_FLOAT_r32 },
},
},
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS] = {
.best = { 0, 1, 2, 3 },
.entries = {
{ 1, 0, _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r01 },
{ 1, 1, _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r11 },
{ 1, 2, _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r21 },
{ 2, 3, _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r32 },
},
},
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS] = {
.best = { 0, 1, 2, 3 },
.entries = {
{ 1, 0, _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r01 },
{ 1, 1, _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r11 },
{ 1, 2, _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r21 },
{ 2, 3, _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r32 },
},
},
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS] = {
.best = { 0, 1, 2, 3 },
.entries = {
{ 1, 0, _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r01 },
{ 1, 1, _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r11 },
{ 1, 2, _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r21 },
{ 2, 3, _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r32 },
{ 3, 0, _BINARY_OP_SUBTRACT_FLOAT_r03 },
{ 3, 1, _BINARY_OP_SUBTRACT_FLOAT_r13 },
{ 3, 2, _BINARY_OP_SUBTRACT_FLOAT_r23 },
{ -1, -1, -1 },
},
},
[_BINARY_OP_ADD_UNICODE] = {
@ -3438,30 +3408,15 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
[_GUARD_TOS_FLOAT_r11] = _GUARD_TOS_FLOAT,
[_GUARD_TOS_FLOAT_r22] = _GUARD_TOS_FLOAT,
[_GUARD_TOS_FLOAT_r33] = _GUARD_TOS_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT_r01] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT_r11] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT_r21] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT_r32] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_ADD_FLOAT_r01] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_ADD_FLOAT_r11] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_ADD_FLOAT_r21] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_ADD_FLOAT_r32] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r01] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r11] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r21] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r32] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r01] = _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r11] = _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r21] = _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r32] = _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r01] = _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r11] = _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r21] = _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r32] = _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r01] = _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r11] = _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r21] = _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r32] = _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_MULTIPLY_FLOAT_r03] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT_r13] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_MULTIPLY_FLOAT_r23] = _BINARY_OP_MULTIPLY_FLOAT,
[_BINARY_OP_ADD_FLOAT_r03] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_ADD_FLOAT_r13] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_ADD_FLOAT_r23] = _BINARY_OP_ADD_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r03] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r13] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_SUBTRACT_FLOAT_r23] = _BINARY_OP_SUBTRACT_FLOAT,
[_BINARY_OP_ADD_UNICODE_r01] = _BINARY_OP_ADD_UNICODE,
[_BINARY_OP_ADD_UNICODE_r11] = _BINARY_OP_ADD_UNICODE,
[_BINARY_OP_ADD_UNICODE_r21] = _BINARY_OP_ADD_UNICODE,
@ -3944,15 +3899,9 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
[_BINARY_OP] = "_BINARY_OP",
[_BINARY_OP_r21] = "_BINARY_OP_r21",
[_BINARY_OP_ADD_FLOAT] = "_BINARY_OP_ADD_FLOAT",
[_BINARY_OP_ADD_FLOAT_r01] = "_BINARY_OP_ADD_FLOAT_r01",
[_BINARY_OP_ADD_FLOAT_r11] = "_BINARY_OP_ADD_FLOAT_r11",
[_BINARY_OP_ADD_FLOAT_r21] = "_BINARY_OP_ADD_FLOAT_r21",
[_BINARY_OP_ADD_FLOAT_r32] = "_BINARY_OP_ADD_FLOAT_r32",
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS] = "_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS",
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r01] = "_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r01",
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r11] = "_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r11",
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r21] = "_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r21",
[_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r32] = "_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS_r32",
[_BINARY_OP_ADD_FLOAT_r03] = "_BINARY_OP_ADD_FLOAT_r03",
[_BINARY_OP_ADD_FLOAT_r13] = "_BINARY_OP_ADD_FLOAT_r13",
[_BINARY_OP_ADD_FLOAT_r23] = "_BINARY_OP_ADD_FLOAT_r23",
[_BINARY_OP_ADD_INT] = "_BINARY_OP_ADD_INT",
[_BINARY_OP_ADD_INT_r01] = "_BINARY_OP_ADD_INT_r01",
[_BINARY_OP_ADD_INT_r11] = "_BINARY_OP_ADD_INT_r11",
@ -3968,15 +3917,9 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
[_BINARY_OP_INPLACE_ADD_UNICODE] = "_BINARY_OP_INPLACE_ADD_UNICODE",
[_BINARY_OP_INPLACE_ADD_UNICODE_r20] = "_BINARY_OP_INPLACE_ADD_UNICODE_r20",
[_BINARY_OP_MULTIPLY_FLOAT] = "_BINARY_OP_MULTIPLY_FLOAT",
[_BINARY_OP_MULTIPLY_FLOAT_r01] = "_BINARY_OP_MULTIPLY_FLOAT_r01",
[_BINARY_OP_MULTIPLY_FLOAT_r11] = "_BINARY_OP_MULTIPLY_FLOAT_r11",
[_BINARY_OP_MULTIPLY_FLOAT_r21] = "_BINARY_OP_MULTIPLY_FLOAT_r21",
[_BINARY_OP_MULTIPLY_FLOAT_r32] = "_BINARY_OP_MULTIPLY_FLOAT_r32",
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS] = "_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS",
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r01] = "_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r01",
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r11] = "_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r11",
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r21] = "_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r21",
[_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r32] = "_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS_r32",
[_BINARY_OP_MULTIPLY_FLOAT_r03] = "_BINARY_OP_MULTIPLY_FLOAT_r03",
[_BINARY_OP_MULTIPLY_FLOAT_r13] = "_BINARY_OP_MULTIPLY_FLOAT_r13",
[_BINARY_OP_MULTIPLY_FLOAT_r23] = "_BINARY_OP_MULTIPLY_FLOAT_r23",
[_BINARY_OP_MULTIPLY_INT] = "_BINARY_OP_MULTIPLY_INT",
[_BINARY_OP_MULTIPLY_INT_r01] = "_BINARY_OP_MULTIPLY_INT_r01",
[_BINARY_OP_MULTIPLY_INT_r11] = "_BINARY_OP_MULTIPLY_INT_r11",
@ -4000,15 +3943,9 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
[_BINARY_OP_SUBSCR_TUPLE_INT] = "_BINARY_OP_SUBSCR_TUPLE_INT",
[_BINARY_OP_SUBSCR_TUPLE_INT_r21] = "_BINARY_OP_SUBSCR_TUPLE_INT_r21",
[_BINARY_OP_SUBTRACT_FLOAT] = "_BINARY_OP_SUBTRACT_FLOAT",
[_BINARY_OP_SUBTRACT_FLOAT_r01] = "_BINARY_OP_SUBTRACT_FLOAT_r01",
[_BINARY_OP_SUBTRACT_FLOAT_r11] = "_BINARY_OP_SUBTRACT_FLOAT_r11",
[_BINARY_OP_SUBTRACT_FLOAT_r21] = "_BINARY_OP_SUBTRACT_FLOAT_r21",
[_BINARY_OP_SUBTRACT_FLOAT_r32] = "_BINARY_OP_SUBTRACT_FLOAT_r32",
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS] = "_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS",
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r01] = "_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r01",
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r11] = "_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r11",
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r21] = "_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r21",
[_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r32] = "_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS_r32",
[_BINARY_OP_SUBTRACT_FLOAT_r03] = "_BINARY_OP_SUBTRACT_FLOAT_r03",
[_BINARY_OP_SUBTRACT_FLOAT_r13] = "_BINARY_OP_SUBTRACT_FLOAT_r13",
[_BINARY_OP_SUBTRACT_FLOAT_r23] = "_BINARY_OP_SUBTRACT_FLOAT_r23",
[_BINARY_OP_SUBTRACT_INT] = "_BINARY_OP_SUBTRACT_INT",
[_BINARY_OP_SUBTRACT_INT_r01] = "_BINARY_OP_SUBTRACT_INT_r01",
[_BINARY_OP_SUBTRACT_INT_r11] = "_BINARY_OP_SUBTRACT_INT_r11",
@ -5098,12 +5035,6 @@ int _PyUop_num_popped(int opcode, int oparg)
return 2;
case _BINARY_OP_SUBTRACT_FLOAT:
return 2;
case _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS:
return 2;
case _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS:
return 2;
case _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS:
return 2;
case _BINARY_OP_ADD_UNICODE:
return 2;
case _BINARY_OP_INPLACE_ADD_UNICODE:

8
Lib/test/test_capi/test_opt.py
View file

@ -666,7 +666,7 @@ class TestUopsOptimization(unittest.TestCase):
self.assertLessEqual(len(guard_nos_float_count), 1)
# TODO gh-115506: this assertion may change after propagating constants.
# We'll also need to verify that propagation actually occurs.
self.assertIn("_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS", uops)
self.assertIn("_POP_TOP_NOP", uops)
def test_float_subtract_constant_propagation(self):
def testfunc(n):
@ -688,7 +688,7 @@ class TestUopsOptimization(unittest.TestCase):
self.assertLessEqual(len(guard_nos_float_count), 1)
# TODO gh-115506: this assertion may change after propagating constants.
# We'll also need to verify that propagation actually occurs.
self.assertIn("_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS", uops)
self.assertIn("_POP_TOP_NOP", uops)
def test_float_multiply_constant_propagation(self):
def testfunc(n):
@ -710,7 +710,7 @@ class TestUopsOptimization(unittest.TestCase):
self.assertLessEqual(len(guard_nos_float_count), 1)
# TODO gh-115506: this assertion may change after propagating constants.
# We'll also need to verify that propagation actually occurs.
self.assertIn("_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS", uops)
self.assertIn("_POP_TOP_NOP", uops)
def test_add_unicode_propagation(self):
def testfunc(n):
@ -2466,7 +2466,7 @@ class TestUopsOptimization(unittest.TestCase):
self.assertAlmostEqual(res, TIER2_THRESHOLD * (0.1 + 0.1))
self.assertIsNotNone(ex)
uops = get_opnames(ex)
self.assertIn("_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS", uops)
self.assertIn("_POP_TOP_NOP", uops)
def test_remove_guard_for_slice_list(self):
def f(n):

7
Objects/floatobject.c
View file

@ -135,13 +135,6 @@ PyFloat_FromDouble(double fval)
return (PyObject *) op;
}
_PyStackRef _PyFloat_FromDouble_ConsumeInputs(_PyStackRef left, _PyStackRef right, double value)
{
PyStackRef_CLOSE_SPECIALIZED(left, _PyFloat_ExactDealloc);
PyStackRef_CLOSE_SPECIALIZED(right, _PyFloat_ExactDealloc);
return PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(value));
}
static PyObject *
float_from_string_inner(const char *s, Py_ssize_t len, void *obj)
{

76
Python/bytecodes.c
View file

@ -678,53 +678,7 @@ dummy_func(
EXIT_IF(!PyFloat_CheckExact(value_o));
}
pure op(_BINARY_OP_MULTIPLY_FLOAT, (left, right -- res)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
assert(PyFloat_CheckExact(left_o));
assert(PyFloat_CheckExact(right_o));
STAT_INC(BINARY_OP, hit);
double dres =
((PyFloatObject *)left_o)->ob_fval *
((PyFloatObject *)right_o)->ob_fval;
res = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
INPUTS_DEAD();
ERROR_IF(PyStackRef_IsNull(res));
}
pure op(_BINARY_OP_ADD_FLOAT, (left, right -- res)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
assert(PyFloat_CheckExact(left_o));
assert(PyFloat_CheckExact(right_o));
STAT_INC(BINARY_OP, hit);
double dres =
((PyFloatObject *)left_o)->ob_fval +
((PyFloatObject *)right_o)->ob_fval;
res = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
INPUTS_DEAD();
ERROR_IF(PyStackRef_IsNull(res));
}
pure op(_BINARY_OP_SUBTRACT_FLOAT, (left, right -- res)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
assert(PyFloat_CheckExact(left_o));
assert(PyFloat_CheckExact(right_o));
STAT_INC(BINARY_OP, hit);
double dres =
((PyFloatObject *)left_o)->ob_fval -
((PyFloatObject *)right_o)->ob_fval;
res = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
INPUTS_DEAD();
ERROR_IF(PyStackRef_IsNull(res));
}
pure op(_BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS, (left, right -- res)) {
pure op(_BINARY_OP_MULTIPLY_FLOAT, (left, right -- res, l, r)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
assert(PyFloat_CheckExact(left_o));
@ -735,11 +689,15 @@ dummy_func(
((PyFloatObject *)left_o)->ob_fval *
((PyFloatObject *)right_o)->ob_fval;
res = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
ERROR_NO_POP();
}
l = left;
r = right;
INPUTS_DEAD();
ERROR_IF(PyStackRef_IsNull(res));
}
pure op(_BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS, (left, right -- res)) {
pure op(_BINARY_OP_ADD_FLOAT, (left, right -- res, l, r)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
assert(PyFloat_CheckExact(left_o));
@ -750,11 +708,15 @@ dummy_func(
((PyFloatObject *)left_o)->ob_fval +
((PyFloatObject *)right_o)->ob_fval;
res = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
ERROR_NO_POP();
}
l = left;
r = right;
INPUTS_DEAD();
ERROR_IF(PyStackRef_IsNull(res));
}
pure op(_BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS, (left, right -- res)) {
pure op(_BINARY_OP_SUBTRACT_FLOAT, (left, right -- res, l, r)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
assert(PyFloat_CheckExact(left_o));
@ -765,16 +727,20 @@ dummy_func(
((PyFloatObject *)left_o)->ob_fval -
((PyFloatObject *)right_o)->ob_fval;
res = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
ERROR_NO_POP();
}
l = left;
r = right;
INPUTS_DEAD();
ERROR_IF(PyStackRef_IsNull(res));
}
macro(BINARY_OP_MULTIPLY_FLOAT) =
_GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_MULTIPLY_FLOAT;
_GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_MULTIPLY_FLOAT + _POP_TOP_FLOAT + _POP_TOP_FLOAT;
macro(BINARY_OP_ADD_FLOAT) =
_GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_ADD_FLOAT;
_GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_ADD_FLOAT + _POP_TOP_FLOAT + _POP_TOP_FLOAT;
macro(BINARY_OP_SUBTRACT_FLOAT) =
_GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_SUBTRACT_FLOAT;
_GUARD_TOS_FLOAT + _GUARD_NOS_FLOAT + unused/5 + _BINARY_OP_SUBTRACT_FLOAT + _POP_TOP_FLOAT + _POP_TOP_FLOAT;
pure op(_BINARY_OP_ADD_UNICODE, (left, right -- res)) {
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);

7
Python/ceval_macros.h
View file

@ -252,10 +252,11 @@ GETITEM(PyObject *v, Py_ssize_t i) {
(frame->owner == FRAME_OWNED_BY_INTERPRETER || (STACK_LEVEL() >= 0 && STACK_LEVEL() <= STACK_SIZE()))
#if defined(Py_DEBUG) && !defined(_Py_JIT)
#define WITHIN_STACK_BOUNDS_WITH_CACHE() \
(frame->owner == FRAME_OWNED_BY_INTERPRETER || (STACK_LEVEL() >= 0 && (STACK_LEVEL() + current_cached_values) <= STACK_SIZE()))
// This allows temporary stack "overflows", provided it's all in the cache at any point of time.
#define WITHIN_STACK_BOUNDS_IGNORING_CACHE() \
(frame->owner == FRAME_OWNED_BY_INTERPRETER || (STACK_LEVEL() >= 0 && (STACK_LEVEL()) <= STACK_SIZE()))
#else
#define WITHIN_STACK_BOUNDS_WITH_CACHE WITHIN_STACK_BOUNDS
#define WITHIN_STACK_BOUNDS_IGNORING_CACHE WITHIN_STACK_BOUNDS
#endif
/* Data access macros */

3229
Python/executor_cases.c.h generated
View file

File diff suppressed because it is too large Load diff

60
Python/generated_cases.c.h generated
View file

@ -96,6 +96,8 @@
_PyStackRef left;
_PyStackRef right;
_PyStackRef res;
_PyStackRef l;
_PyStackRef r;
// _GUARD_TOS_FLOAT
{
value = stack_pointer[-1];
@ -128,10 +130,24 @@
double dres =
((PyFloatObject *)left_o)->ob_fval +
((PyFloatObject *)right_o)->ob_fval;
res = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
res = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
JUMP_TO_LABEL(pop_2_error);
JUMP_TO_LABEL(error);
}
l = left;
r = right;
}
// _POP_TOP_FLOAT
{
value = r;
assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
}
// _POP_TOP_FLOAT
{
value = l;
assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
}
stack_pointer[-2] = res;
stack_pointer += -1;
@ -421,6 +437,8 @@
_PyStackRef left;
_PyStackRef right;
_PyStackRef res;
_PyStackRef l;
_PyStackRef r;
// _GUARD_TOS_FLOAT
{
value = stack_pointer[-1];
@ -453,10 +471,24 @@
double dres =
((PyFloatObject *)left_o)->ob_fval *
((PyFloatObject *)right_o)->ob_fval;
res = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
res = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
JUMP_TO_LABEL(pop_2_error);
JUMP_TO_LABEL(error);
}
l = left;
r = right;
}
// _POP_TOP_FLOAT
{
value = r;
assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
}
// _POP_TOP_FLOAT
{
value = l;
assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
}
stack_pointer[-2] = res;
stack_pointer += -1;
@ -1003,6 +1035,8 @@
_PyStackRef left;
_PyStackRef right;
_PyStackRef res;
_PyStackRef l;
_PyStackRef r;
// _GUARD_TOS_FLOAT
{
value = stack_pointer[-1];
@ -1035,10 +1069,24 @@
double dres =
((PyFloatObject *)left_o)->ob_fval -
((PyFloatObject *)right_o)->ob_fval;
res = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
res = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
JUMP_TO_LABEL(pop_2_error);
JUMP_TO_LABEL(error);
}
l = left;
r = right;
}
// _POP_TOP_FLOAT
{
value = r;
assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
}
// _POP_TOP_FLOAT
{
value = l;
assert(PyFloat_CheckExact(PyStackRef_AsPyObjectBorrow(value)));
PyStackRef_CLOSE_SPECIALIZED(value, _PyFloat_ExactDealloc);
}
stack_pointer[-2] = res;
stack_pointer += -1;

11
Python/optimizer_analysis.c
View file

@ -196,7 +196,7 @@ check_stack_bounds(JitOptContext *ctx, JitOptRef *stack_pointer, int offset, int
(opcode == _RETURN_VALUE) ||
(opcode == _RETURN_GENERATOR) ||
(opcode == _YIELD_VALUE);
if (should_check && (stack_level < 0 || stack_level > STACK_SIZE())) {
if (should_check && (stack_level < 0 || stack_level > STACK_SIZE() + MAX_CACHED_REGISTER)) {
ctx->contradiction = true;
ctx->done = true;
return 1;
@ -312,7 +312,7 @@ _Py_opt_assert_within_stack_bounds(
fflush(stdout);
abort();
}
int size = (int)(frame->stack_len);
int size = (int)(frame->stack_len) + MAX_CACHED_REGISTER;
if (level > size) {
printf("Stack overflow (depth = %d) at %s:%d\n", level, filename, lineno);
fflush(stdout);
@ -327,13 +327,6 @@ _Py_opt_assert_within_stack_bounds(
#define ASSERT_WITHIN_STACK_BOUNDS(F, L) (void)0
#endif
// TODO (gh-134584) generate most of this table automatically
const uint16_t op_without_decref_inputs[MAX_UOP_ID + 1] = {
[_BINARY_OP_MULTIPLY_FLOAT] = _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_ADD_FLOAT] = _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS,
[_BINARY_OP_SUBTRACT_FLOAT] = _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS,
};
/* >0 (length) for success, 0 for not ready, clears all possible errors. */
static int
optimize_uops(

30
Python/optimizer_bytecodes.c
View file

@ -261,31 +261,25 @@ dummy_func(void) {
res = sym_new_compact_int(ctx);
}
op(_BINARY_OP_ADD_FLOAT, (left, right -- res)) {
op(_BINARY_OP_ADD_FLOAT, (left, right -- res, l, r)) {
REPLACE_OPCODE_IF_EVALUATES_PURE(left, right);
res = sym_new_type(ctx, &PyFloat_Type);
// TODO (gh-134584): Refactor this to use another uop
if (PyJitRef_IsBorrowed(left) && PyJitRef_IsBorrowed(right)) {
REPLACE_OP(this_instr, op_without_decref_inputs[opcode], oparg, 0);
}
l = left;
r = right;
}
op(_BINARY_OP_SUBTRACT_FLOAT, (left, right -- res)) {
op(_BINARY_OP_SUBTRACT_FLOAT, (left, right -- res, l, r)) {
REPLACE_OPCODE_IF_EVALUATES_PURE(left, right);
res = sym_new_type(ctx, &PyFloat_Type);
// TODO (gh-134584): Refactor this to use another uop
if (PyJitRef_IsBorrowed(left) && PyJitRef_IsBorrowed(right)) {
REPLACE_OP(this_instr, op_without_decref_inputs[opcode], oparg, 0);
}
l = left;
r = right;
}
op(_BINARY_OP_MULTIPLY_FLOAT, (left, right -- res)) {
op(_BINARY_OP_MULTIPLY_FLOAT, (left, right -- res, l, r)) {
REPLACE_OPCODE_IF_EVALUATES_PURE(left, right);
res = sym_new_type(ctx, &PyFloat_Type);
// TODO (gh-134584): Refactor this to use another uop
if (PyJitRef_IsBorrowed(left) && PyJitRef_IsBorrowed(right)) {
REPLACE_OP(this_instr, op_without_decref_inputs[opcode], oparg, 0);
}
l = left;
r = right;
}
op(_BINARY_OP_ADD_UNICODE, (left, right -- res)) {
@ -546,6 +540,12 @@ dummy_func(void) {
}
}
op(_POP_TOP_FLOAT, (value --)) {
if (PyJitRef_IsBorrowed(value)) {
REPLACE_OP(this_instr, _POP_TOP_NOP, 0, 0);
}
}
op(_COPY, (bottom, unused[oparg-1] -- bottom, unused[oparg-1], top)) {
assert(oparg > 0);
top = bottom;

149
Python/optimizer_cases.c.h generated
View file

@ -153,6 +153,11 @@
}
case _POP_TOP_FLOAT: {
JitOptRef value;
value = stack_pointer[-1];
if (PyJitRef_IsBorrowed(value)) {
REPLACE_OP(this_instr, _POP_TOP_NOP, 0, 0);
}
CHECK_STACK_BOUNDS(-1);
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
@ -668,6 +673,8 @@
JitOptRef right;
JitOptRef left;
JitOptRef res;
JitOptRef l;
JitOptRef r;
right = stack_pointer[-1];
left = stack_pointer[-2];
if (
@ -679,6 +686,8 @@
_PyStackRef left = sym_get_const_as_stackref(ctx, left_sym);
_PyStackRef right = sym_get_const_as_stackref(ctx, right_sym);
_PyStackRef res_stackref;
_PyStackRef l_stackref;
_PyStackRef r_stackref;
/* Start of uop copied from bytecodes for constant evaluation */
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
@ -688,32 +697,32 @@
double dres =
((PyFloatObject *)left_o)->ob_fval *
((PyFloatObject *)right_o)->ob_fval;
res_stackref = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
if (PyStackRef_IsNull(res_stackref )) {
goto error;
res_stackref = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
JUMP_TO_LABEL(error);
}
l_stackref = left;
r_stackref = right;
/* End of uop copied from bytecodes for constant evaluation */
res = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(res_stackref));
if (sym_is_const(ctx, res)) {
PyObject *result = sym_get_const(ctx, res);
if (_Py_IsImmortal(result)) {
// Replace with _POP_TWO_LOAD_CONST_INLINE_BORROW since we have two inputs and an immortal result
REPLACE_OP(this_instr, _POP_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)result);
}
}
CHECK_STACK_BOUNDS(-1);
l = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(l_stackref));
r = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(r_stackref));
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = l;
stack_pointer[0] = r;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
res = sym_new_type(ctx, &PyFloat_Type);
if (PyJitRef_IsBorrowed(left) && PyJitRef_IsBorrowed(right)) {
REPLACE_OP(this_instr, op_without_decref_inputs[opcode], oparg, 0);
}
CHECK_STACK_BOUNDS(-1);
l = left;
r = right;
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = l;
stack_pointer[0] = r;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
@ -722,6 +731,8 @@
JitOptRef right;
JitOptRef left;
JitOptRef res;
JitOptRef l;
JitOptRef r;
right = stack_pointer[-1];
left = stack_pointer[-2];
if (
@ -733,6 +744,8 @@
_PyStackRef left = sym_get_const_as_stackref(ctx, left_sym);
_PyStackRef right = sym_get_const_as_stackref(ctx, right_sym);
_PyStackRef res_stackref;
_PyStackRef l_stackref;
_PyStackRef r_stackref;
/* Start of uop copied from bytecodes for constant evaluation */
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
@ -742,32 +755,32 @@
double dres =
((PyFloatObject *)left_o)->ob_fval +
((PyFloatObject *)right_o)->ob_fval;
res_stackref = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
if (PyStackRef_IsNull(res_stackref )) {
goto error;
res_stackref = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
JUMP_TO_LABEL(error);
}
l_stackref = left;
r_stackref = right;
/* End of uop copied from bytecodes for constant evaluation */
res = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(res_stackref));
if (sym_is_const(ctx, res)) {
PyObject *result = sym_get_const(ctx, res);
if (_Py_IsImmortal(result)) {
// Replace with _POP_TWO_LOAD_CONST_INLINE_BORROW since we have two inputs and an immortal result
REPLACE_OP(this_instr, _POP_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)result);
}
}
CHECK_STACK_BOUNDS(-1);
l = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(l_stackref));
r = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(r_stackref));
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = l;
stack_pointer[0] = r;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
res = sym_new_type(ctx, &PyFloat_Type);
if (PyJitRef_IsBorrowed(left) && PyJitRef_IsBorrowed(right)) {
REPLACE_OP(this_instr, op_without_decref_inputs[opcode], oparg, 0);
}
CHECK_STACK_BOUNDS(-1);
l = left;
r = right;
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = l;
stack_pointer[0] = r;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
@ -776,6 +789,8 @@
JitOptRef right;
JitOptRef left;
JitOptRef res;
JitOptRef l;
JitOptRef r;
right = stack_pointer[-1];
left = stack_pointer[-2];
if (
@ -787,6 +802,8 @@
_PyStackRef left = sym_get_const_as_stackref(ctx, left_sym);
_PyStackRef right = sym_get_const_as_stackref(ctx, right_sym);
_PyStackRef res_stackref;
_PyStackRef l_stackref;
_PyStackRef r_stackref;
/* Start of uop copied from bytecodes for constant evaluation */
PyObject *left_o = PyStackRef_AsPyObjectBorrow(left);
PyObject *right_o = PyStackRef_AsPyObjectBorrow(right);
@ -796,62 +813,32 @@
double dres =
((PyFloatObject *)left_o)->ob_fval -
((PyFloatObject *)right_o)->ob_fval;
res_stackref = _PyFloat_FromDouble_ConsumeInputs(left, right, dres);
if (PyStackRef_IsNull(res_stackref )) {
goto error;
res_stackref = PyStackRef_FromPyObjectSteal(PyFloat_FromDouble(dres));
if (PyStackRef_IsNull(res)) {
JUMP_TO_LABEL(error);
}
l_stackref = left;
r_stackref = right;
/* End of uop copied from bytecodes for constant evaluation */
res = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(res_stackref));
if (sym_is_const(ctx, res)) {
PyObject *result = sym_get_const(ctx, res);
if (_Py_IsImmortal(result)) {
// Replace with _POP_TWO_LOAD_CONST_INLINE_BORROW since we have two inputs and an immortal result
REPLACE_OP(this_instr, _POP_TWO_LOAD_CONST_INLINE_BORROW, 0, (uintptr_t)result);
}
}
CHECK_STACK_BOUNDS(-1);
l = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(l_stackref));
r = sym_new_const_steal(ctx, PyStackRef_AsPyObjectSteal(r_stackref));
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = l;
stack_pointer[0] = r;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
res = sym_new_type(ctx, &PyFloat_Type);
if (PyJitRef_IsBorrowed(left) && PyJitRef_IsBorrowed(right)) {
REPLACE_OP(this_instr, op_without_decref_inputs[opcode], oparg, 0);
}
CHECK_STACK_BOUNDS(-1);
l = left;
r = right;
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
case _BINARY_OP_MULTIPLY_FLOAT__NO_DECREF_INPUTS: {
JitOptRef res;
res = sym_new_not_null(ctx);
CHECK_STACK_BOUNDS(-1);
stack_pointer[-2] = res;
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
case _BINARY_OP_ADD_FLOAT__NO_DECREF_INPUTS: {
JitOptRef res;
res = sym_new_not_null(ctx);
CHECK_STACK_BOUNDS(-1);
stack_pointer[-2] = res;
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}
case _BINARY_OP_SUBTRACT_FLOAT__NO_DECREF_INPUTS: {
JitOptRef res;
res = sym_new_not_null(ctx);
CHECK_STACK_BOUNDS(-1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = l;
stack_pointer[0] = r;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}

1
Tools/cases_generator/analyzer.py
View file

@ -634,7 +634,6 @@ NON_ESCAPING_FUNCTIONS = (
"_PyCode_CODE",
"_PyDictValues_AddToInsertionOrder",
"_PyErr_Occurred",
"_PyFloat_FromDouble_ConsumeInputs",
"_PyFrame_GetBytecode",
"_PyFrame_GetCode",
"_PyFrame_IsIncomplete",

4
Tools/cases_generator/tier2_generator.py
View file

@ -305,7 +305,7 @@ def generate_tier2(
emitter = Tier2Emitter(out, analysis.labels, exit_depth)
out.emit(f"case {uop.name}_r{inputs}{outputs}: {{\n")
out.emit(f"CHECK_CURRENT_CACHED_VALUES({inputs});\n")
out.emit("assert(WITHIN_STACK_BOUNDS_WITH_CACHE());\n")
out.emit("assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());\n")
declare_variables(uop, out)
stack = Stack()
stack.push_cache([f"_tos_cache{i}" for i in range(inputs)], out)
@ -313,7 +313,7 @@ def generate_tier2(
reachable, stack = write_uop(uop, emitter, stack, offset_strs, outputs)
out.start_line()
if reachable:
out.emit("assert(WITHIN_STACK_BOUNDS_WITH_CACHE());\n")
out.emit("assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());\n")
if not uop.properties.always_exits:
out.emit("break;\n")
out.start_line()