Stowage/cpython
2
0
Fork 0
mirror of https://github.com/python/cpython.git synced 2026-01-03 05:54:03 +00:00
Code Issues Projects Releases Packages Wiki Activity

GH-139757: JIT: Remove redundant branches to jumps in the assembly optimizer (GH-140800)

Browse source 
JIT: Remove redundant branches to jump in the assembly optimizer

* Refactor JIT assembly optimizer making instructions instances not just strings
* Remove redundant jumps and branches where legal to do so
* Modifies _BINARY_OP_SUBSCR_STR_INT to avoid excessive inlining depth
This commit is contained in:
Mark Shannon 2025-12-08 17:57:11 +00:00 • committed by GitHub
parent 37988c57ea
commit e0451ceef8
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
8 changed files with 220 additions and 74 deletions
Download patch file Download diff file Expand all files Collapse all files

7
Python/bytecodes.c
View file

@ -982,9 +982,10 @@ dummy_func(
DEOPT_IF(!_PyLong_IsNonNegativeCompact((PyLongObject *)sub));
Py_ssize_t index = ((PyLongObject*)sub)->long_value.ob_digit[0];
DEOPT_IF(PyUnicode_GET_LENGTH(str) <= index);
// Specialize for reading an ASCII character from any string:
Py_UCS4 c = PyUnicode_READ_CHAR(str, index);
DEOPT_IF(Py_ARRAY_LENGTH(_Py_SINGLETON(strings).ascii) <= c);
// Specialize for reading an ASCII character from an ASCII string:
DEOPT_IF(!PyUnicode_IS_COMPACT_ASCII(str));
uint8_t c = PyUnicode_1BYTE_DATA(str)[index];
assert(c < 128);
STAT_INC(BINARY_OP, hit);
PyObject *res_o = (PyObject*)&_Py_SINGLETON(strings).ascii[c];
PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);

5
Python/executor_cases.c.h generated
View file

@ -1502,11 +1502,12 @@
UOP_STAT_INC(uopcode, miss);
JUMP_TO_JUMP_TARGET();
}
Py_UCS4 c = PyUnicode_READ_CHAR(str, index);
if (Py_ARRAY_LENGTH(_Py_SINGLETON(strings).ascii) <= c) {
if (!PyUnicode_IS_COMPACT_ASCII(str)) {
UOP_STAT_INC(uopcode, miss);
JUMP_TO_JUMP_TARGET();
}
uint8_t c = PyUnicode_1BYTE_DATA(str)[index];
assert(c < 128);
STAT_INC(BINARY_OP, hit);
PyObject *res_o = (PyObject*)&_Py_SINGLETON(strings).ascii[c];
PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);

5
Python/generated_cases.c.h generated
View file

@ -892,12 +892,13 @@
assert(_PyOpcode_Deopt[opcode] == (BINARY_OP));
JUMP_TO_PREDICTED(BINARY_OP);
}
Py_UCS4 c = PyUnicode_READ_CHAR(str, index);
if (Py_ARRAY_LENGTH(_Py_SINGLETON(strings).ascii) <= c) {
if (!PyUnicode_IS_COMPACT_ASCII(str)) {
UPDATE_MISS_STATS(BINARY_OP);
assert(_PyOpcode_Deopt[opcode] == (BINARY_OP));
JUMP_TO_PREDICTED(BINARY_OP);
}
uint8_t c = PyUnicode_1BYTE_DATA(str)[index];
assert(c < 128);
STAT_INC(BINARY_OP, hit);
PyObject *res_o = (PyObject*)&_Py_SINGLETON(strings).ascii[c];
PyStackRef_CLOSE_SPECIALIZED(sub_st, _PyLong_ExactDealloc);

3
Python/jit.c
View file

@ -185,6 +185,7 @@ set_bits(uint32_t *loc, uint8_t loc_start, uint64_t value, uint8_t value_start,
#define IS_AARCH64_ADRP(I) (((I) & 0x9F000000) == 0x90000000)
#define IS_AARCH64_BRANCH(I) (((I) & 0x7C000000) == 0x14000000)
#define IS_AARCH64_BRANCH_COND(I) (((I) & 0x7C000000) == 0x54000000)
#define IS_AARCH64_BRANCH_ZERO(I) (((I) & 0x7E000000) == 0x34000000)
#define IS_AARCH64_TEST_AND_BRANCH(I) (((I) & 0x7E000000) == 0x36000000)
#define IS_AARCH64_LDR_OR_STR(I) (((I) & 0x3B000000) == 0x39000000)
#define IS_AARCH64_MOV(I) (((I) & 0x9F800000) == 0x92800000)
@ -352,7 +353,7 @@ void
patch_aarch64_19r(unsigned char *location, uint64_t value)
{
uint32_t *loc32 = (uint32_t *)location;
assert(IS_AARCH64_BRANCH_COND(*loc32));
assert(IS_AARCH64_BRANCH_COND(*loc32) || IS_AARCH64_BRANCH_ZERO(*loc32));
value -= (uintptr_t)location;
// Check that we're not out of range of 21 signed bits:
assert((int64_t)value >= -(1 << 20));