Mirror
/
cpython
mirror of https://github.com/python/cpython
4
1
Fork 0
Code Issues Releases Wiki Activity

GH-126795: Increase the JIT threshold from 16 to 4096 (GH-126816)

This commit is contained in:
Brandt Bucher 2024-11-18 11:11:23 -08:00 committed by GitHub
parent 933f21c3c9
commit 4cd10762b0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 104 additions and 101 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
Include/internal/pycore_backoff.h
View File

@ -102,10 +102,9 @@ backoff_counter_triggers(_Py_BackoffCounter counter)
} }
/* Initial JUMP_BACKWARD counter. /* Initial JUMP_BACKWARD counter.
* This determines when we create a trace for a loop. * This determines when we create a trace for a loop. */
* Backoff sequence 16, 32, 64, 128, 256, 512, 1024, 2048, 4096. */ #define JUMP_BACKWARD_INITIAL_VALUE 4095
#define JUMP_BACKWARD_INITIAL_VALUE 15 #define JUMP_BACKWARD_INITIAL_BACKOFF 12
#define JUMP_BACKWARD_INITIAL_BACKOFF 4
static inline _Py_BackoffCounter static inline _Py_BackoffCounter
initial_jump_backoff_counter(void) initial_jump_backoff_counter(void)
{ {

194
Lib/test/test_capi/test_opt.py
View File

@ -62,9 +62,9 @@ class TestOptimizerAPI(unittest.TestCase):
def test_counter_optimizer(self): def test_counter_optimizer(self):
# Generate a new function at each call # Generate a new function at each call
ns = {} ns = {}
exec(textwrap.dedent(""" exec(textwrap.dedent(f"""
def loop(): def loop():
for _ in range(1000): for _ in range({TIER2_THRESHOLD + 1000}):
pass pass
"""), ns, ns) """), ns, ns)
loop = ns['loop'] loop = ns['loop']
@ -75,20 +75,19 @@ class TestOptimizerAPI(unittest.TestCase):
self.assertEqual(opt.get_count(), 0) self.assertEqual(opt.get_count(), 0)
with clear_executors(loop): with clear_executors(loop):
loop() loop()
# Subtract because optimizer doesn't kick in sooner self.assertEqual(opt.get_count(), 1001)
self.assertEqual(opt.get_count(), 1000 - TIER2_THRESHOLD)
def test_long_loop(self): def test_long_loop(self):
"Check that we aren't confused by EXTENDED_ARG" "Check that we aren't confused by EXTENDED_ARG"
# Generate a new function at each call # Generate a new function at each call
ns = {} ns = {}
exec(textwrap.dedent(""" exec(textwrap.dedent(f"""
def nop(): def nop():
pass pass
def long_loop(): def long_loop():
for _ in range(20): for _ in range({TIER2_THRESHOLD + 20}):
nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop();
nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop();
nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop(); nop();
@ -103,7 +102,7 @@ class TestOptimizerAPI(unittest.TestCase):
with temporary_optimizer(opt): with temporary_optimizer(opt):
self.assertEqual(opt.get_count(), 0) self.assertEqual(opt.get_count(), 0)
long_loop() long_loop()
self.assertEqual(opt.get_count(), 20 - TIER2_THRESHOLD) # Need iterations to warm up self.assertEqual(opt.get_count(), 21) # Need iterations to warm up
def test_code_restore_for_ENTER_EXECUTOR(self): def test_code_restore_for_ENTER_EXECUTOR(self):
def testfunc(x): def testfunc(x):
@ -159,7 +158,7 @@ class TestExecutorInvalidation(unittest.TestCase):
func_src = "\n".join( func_src = "\n".join(
f""" f"""
def f{n}(): def f{n}():
for _ in range(1000): for _ in range({TIER2_THRESHOLD}):
pass pass
""" for n in range(5) """ for n in range(5)
) )
@ -190,9 +189,9 @@ class TestExecutorInvalidation(unittest.TestCase):
def test_uop_optimizer_invalidation(self): def test_uop_optimizer_invalidation(self):
# Generate a new function at each call # Generate a new function at each call
ns = {} ns = {}
exec(textwrap.dedent(""" exec(textwrap.dedent(f"""
def f(): def f():
for i in range(1000): for i in range({TIER2_THRESHOLD}):
pass pass
"""), ns, ns) """), ns, ns)
f = ns['f'] f = ns['f']
@ -207,7 +206,7 @@ class TestExecutorInvalidation(unittest.TestCase):
def test_sys__clear_internal_caches(self): def test_sys__clear_internal_caches(self):
def f(): def f():
for _ in range(1000): for _ in range(TIER2_THRESHOLD):
pass pass
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
@ -236,7 +235,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(1000) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -247,7 +246,7 @@ class TestUops(unittest.TestCase):
def test_extended_arg(self): def test_extended_arg(self):
"Check EXTENDED_ARG handling in superblock creation" "Check EXTENDED_ARG handling in superblock creation"
ns = {} ns = {}
exec(textwrap.dedent(""" exec(textwrap.dedent(f"""
def many_vars(): def many_vars():
# 260 vars, so z9 should have index 259 # 260 vars, so z9 should have index 259
a0 = a1 = a2 = a3 = a4 = a5 = a6 = a7 = a8 = a9 = 42 a0 = a1 = a2 = a3 = a4 = a5 = a6 = a7 = a8 = a9 = 42
@ -275,7 +274,7 @@ class TestUops(unittest.TestCase):
w0 = w1 = w2 = w3 = w4 = w5 = w6 = w7 = w8 = w9 = 42 w0 = w1 = w2 = w3 = w4 = w5 = w6 = w7 = w8 = w9 = 42
x0 = x1 = x2 = x3 = x4 = x5 = x6 = x7 = x8 = x9 = 42 x0 = x1 = x2 = x3 = x4 = x5 = x6 = x7 = x8 = x9 = 42
y0 = y1 = y2 = y3 = y4 = y5 = y6 = y7 = y8 = y9 = 42 y0 = y1 = y2 = y3 = y4 = y5 = y6 = y7 = y8 = y9 = 42
z0 = z1 = z2 = z3 = z4 = z5 = z6 = z7 = z8 = z9 = 42 z0 = z1 = z2 = z3 = z4 = z5 = z6 = z7 = z8 = z9 = {TIER2_THRESHOLD}
while z9 > 0: while z9 > 0:
z9 = z9 - 1 z9 = z9 - 1
+z9 +z9
@ -308,7 +307,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -323,7 +322,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -338,7 +337,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(range(20)) testfunc(range(TIER2_THRESHOLD))
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -355,7 +354,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(range(20)) testfunc(range(TIER2_THRESHOLD))
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -371,7 +370,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -386,7 +385,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -406,7 +405,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -424,8 +423,8 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
total = testfunc(20) total = testfunc(TIER2_THRESHOLD)
self.assertEqual(total, 190) self.assertEqual(total, sum(range(TIER2_THRESHOLD)))
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -445,9 +444,9 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
a = list(range(20)) a = list(range(TIER2_THRESHOLD))
total = testfunc(a) total = testfunc(a)
self.assertEqual(total, 190) self.assertEqual(total, sum(a))
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -467,9 +466,9 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
a = tuple(range(20)) a = tuple(range(TIER2_THRESHOLD))
total = testfunc(a) total = testfunc(a)
self.assertEqual(total, 190) self.assertEqual(total, sum(a))
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -503,7 +502,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -521,7 +520,7 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -549,9 +548,9 @@ class TestUops(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
x = testfunc(10, 10) x = testfunc(TIER2_THRESHOLD, TIER2_THRESHOLD)
self.assertEqual(x, sum(range(10)) * 10010) self.assertEqual(x, sum(range(TIER2_THRESHOLD)) * TIER2_THRESHOLD * 1001)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
@ -572,15 +571,13 @@ class TestUops(unittest.TestCase):
bits += 1 bits += 1
if i&0x10: if i&0x10:
bits += 1 bits += 1
if i&0x20:
bits += 1
return bits return bits
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
x = testfunc(20) x = testfunc(TIER2_THRESHOLD * 2)
self.assertEqual(x, 40) self.assertEqual(x, TIER2_THRESHOLD * 5)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
ops = list(iter_opnames(ex)) ops = list(iter_opnames(ex))
@ -615,9 +612,9 @@ class TestUopsOptimization(unittest.TestCase):
num += 1 num += 1
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(res, 63) self.assertEqual(res, (TIER2_THRESHOLD - 1) * 2 + 1)
binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"] binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"]
guard_both_int_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_INT"] guard_both_int_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_INT"]
self.assertGreaterEqual(len(binop_count), 3) self.assertGreaterEqual(len(binop_count), 3)
@ -637,11 +634,11 @@ class TestUopsOptimization(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
res = None res = None
with temporary_optimizer(opt): with temporary_optimizer(opt):
res = testfunc(32) res = testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(res, 124) self.assertEqual(res, (TIER2_THRESHOLD - 1) * 4)
binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"] binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"]
guard_both_int_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_INT"] guard_both_int_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_INT"]
self.assertGreaterEqual(len(binop_count), 3) self.assertGreaterEqual(len(binop_count), 3)
@ -661,11 +658,11 @@ class TestUopsOptimization(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
res = None res = None
with temporary_optimizer(opt): with temporary_optimizer(opt):
res = testfunc(32) res = testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(res, 124) self.assertEqual(res, (TIER2_THRESHOLD - 1) * 4)
binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"] binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"]
guard_both_int_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_INT"] guard_both_int_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_INT"]
self.assertGreaterEqual(len(binop_count), 3) self.assertGreaterEqual(len(binop_count), 3)
@ -682,7 +679,7 @@ class TestUopsOptimization(unittest.TestCase):
num += 1 num += 1
return a return a
res, ex = self._run_with_optimizer(testfunc, 64) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"] binop_count = [opname for opname in iter_opnames(ex) if opname == "_BINARY_OP_ADD_INT"]
self.assertGreaterEqual(len(binop_count), 3) self.assertGreaterEqual(len(binop_count), 3)
@ -694,7 +691,7 @@ class TestUopsOptimization(unittest.TestCase):
for i in range(n): for i in range(n):
dummy(i) dummy(i)
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
self.assertIn("_PUSH_FRAME", uops) self.assertIn("_PUSH_FRAME", uops)
@ -708,8 +705,8 @@ class TestUopsOptimization(unittest.TestCase):
x = i + i x = i + i
return x return x
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 62) self.assertEqual(res, (TIER2_THRESHOLD - 1) * 2)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
self.assertNotIn("_GUARD_BOTH_INT", uops) self.assertNotIn("_GUARD_BOTH_INT", uops)
@ -726,7 +723,7 @@ class TestUopsOptimization(unittest.TestCase):
res = x + z + a + b res = x + z + a + b
return res return res
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 4) self.assertEqual(res, 4)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -739,8 +736,8 @@ class TestUopsOptimization(unittest.TestCase):
for _ in range(n): for _ in range(n):
return [i for i in range(n)] return [i for i in range(n)]
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, list(range(32))) self.assertEqual(res, list(range(TIER2_THRESHOLD)))
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
self.assertNotIn("_BINARY_OP_ADD_INT", uops) self.assertNotIn("_BINARY_OP_ADD_INT", uops)
@ -797,7 +794,7 @@ class TestUopsOptimization(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
_testinternalcapi.set_optimizer(opt) _testinternalcapi.set_optimizer(opt)
testfunc(64) testfunc(_testinternalcapi.TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
assert ex is not None assert ex is not None
@ -817,8 +814,8 @@ class TestUopsOptimization(unittest.TestCase):
a = a + 0.25 a = a + 0.25
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertAlmostEqual(res, 33.0) self.assertAlmostEqual(res, TIER2_THRESHOLD + 1)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
guard_both_float_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_FLOAT"] guard_both_float_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_FLOAT"]
@ -837,8 +834,8 @@ class TestUopsOptimization(unittest.TestCase):
a = a - 0.25 a = a - 0.25
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertAlmostEqual(res, -31.0) self.assertAlmostEqual(res, -TIER2_THRESHOLD + 1)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
guard_both_float_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_FLOAT"] guard_both_float_count = [opname for opname in iter_opnames(ex) if opname == "_GUARD_BOTH_FLOAT"]
@ -857,7 +854,7 @@ class TestUopsOptimization(unittest.TestCase):
a = a * 1.0 a = a * 1.0
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertAlmostEqual(res, 1.0) self.assertAlmostEqual(res, 1.0)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -877,7 +874,7 @@ class TestUopsOptimization(unittest.TestCase):
a + a a + a
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, "") self.assertEqual(res, "")
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -895,7 +892,7 @@ class TestUopsOptimization(unittest.TestCase):
x = a == a x = a == a
return x return x
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertTrue(res) self.assertTrue(res)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -913,7 +910,7 @@ class TestUopsOptimization(unittest.TestCase):
x = a == a x = a == a
return x return x
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertTrue(res) self.assertTrue(res)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -931,7 +928,7 @@ class TestUopsOptimization(unittest.TestCase):
x = 1 x = 1
return x return x
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 1) self.assertEqual(res, 1)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -951,7 +948,7 @@ class TestUopsOptimization(unittest.TestCase):
x = 1 x = 1
return x return x
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 1) self.assertEqual(res, 1)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -971,7 +968,7 @@ class TestUopsOptimization(unittest.TestCase):
x = a == a x = a == a
return x return x
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertTrue(res) self.assertTrue(res)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
@ -989,10 +986,10 @@ class TestUopsOptimization(unittest.TestCase):
exec(src, ns, ns) exec(src, ns, ns)
testfunc = ns['testfunc'] testfunc = ns['testfunc']
ns['_test_global'] = 0 ns['_test_global'] = 0
_, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD) _, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD - 1)
self.assertIsNone(ex) self.assertIsNone(ex)
ns['_test_global'] = 1 ns['_test_global'] = 1
_, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD) _, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD - 1)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops = get_opnames(ex) uops = get_opnames(ex)
self.assertNotIn("_GUARD_BOTH_INT", uops) self.assertNotIn("_GUARD_BOTH_INT", uops)
@ -1003,10 +1000,10 @@ class TestUopsOptimization(unittest.TestCase):
exec(src, ns, ns) exec(src, ns, ns)
testfunc = ns['testfunc'] testfunc = ns['testfunc']
ns['_test_global'] = 0 ns['_test_global'] = 0
_, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD) _, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD - 1)
self.assertIsNone(ex) self.assertIsNone(ex)
ns['_test_global'] = 3.14 ns['_test_global'] = 3.14
_, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD) _, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD - 1)
self.assertIsNone(ex) self.assertIsNone(ex)
def test_combine_stack_space_checks_sequential(self): def test_combine_stack_space_checks_sequential(self):
@ -1023,8 +1020,8 @@ class TestUopsOptimization(unittest.TestCase):
a += b + c + d a += b + c + d
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 832) self.assertEqual(res, TIER2_THRESHOLD * 26)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1050,8 +1047,8 @@ class TestUopsOptimization(unittest.TestCase):
a += b + c a += b + c
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 224) self.assertEqual(res, TIER2_THRESHOLD * 7)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1085,8 +1082,8 @@ class TestUopsOptimization(unittest.TestCase):
a += b + c + d + e a += b + c + d + e
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 800) self.assertEqual(res, TIER2_THRESHOLD * 25)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1121,8 +1118,8 @@ class TestUopsOptimization(unittest.TestCase):
a += b + c + d + e a += b + c + d + e
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 800) self.assertEqual(res, TIER2_THRESHOLD * 25)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1157,16 +1154,16 @@ class TestUopsOptimization(unittest.TestCase):
z = dummy0(y) z = dummy0(y)
return dummy4(z) return dummy4(z)
def testfunc(n): def testfunc(n):
a = 0; a = 0
for _ in range(32): for _ in range(n):
b = dummy5(1) b = dummy5(1)
c = dummy0(1) c = dummy0(1)
d = dummy6(1) d = dummy6(1)
a += b + c + d a += b + c + d
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 96) self.assertEqual(res, TIER2_THRESHOLD * 3)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1225,8 +1222,8 @@ class TestUopsOptimization(unittest.TestCase):
b += dummy15(7) b += dummy15(7)
return b return b
res, ex = self._run_with_optimizer(testfunc, 32) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 32 * (repetitions + 9)) self.assertEqual(res, TIER2_THRESHOLD * (repetitions + 9))
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1259,8 +1256,13 @@ class TestUopsOptimization(unittest.TestCase):
a += dummy15(n) a += dummy15(n)
return a return a
res, ex = self._run_with_optimizer(testfunc, 32) recursion_limit = sys.getrecursionlimit()
self.assertEqual(res, 42 * 32) try:
sys.setrecursionlimit(TIER2_THRESHOLD + recursion_limit)
res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
finally:
sys.setrecursionlimit(recursion_limit)
self.assertEqual(res, TIER2_THRESHOLD * 42)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex] uops_and_operands = [(opcode, operand) for opcode, _, _, operand in ex]
@ -1307,8 +1309,8 @@ class TestUopsOptimization(unittest.TestCase):
for i in range(n): for i in range(n):
gen() gen()
return i return i
res, ex = self._run_with_optimizer(testfunc, 20) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 19) self.assertEqual(res, TIER2_THRESHOLD - 1)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertIn("_RETURN_GENERATOR", get_opnames(ex)) self.assertIn("_RETURN_GENERATOR", get_opnames(ex))
@ -1322,8 +1324,8 @@ class TestUopsOptimization(unittest.TestCase):
for i in g: for i in g:
s += i s += i
return s return s
res, ex = self._run_with_optimizer(testfunc, 20) res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, 190) self.assertEqual(res, sum(range(TIER2_THRESHOLD)))
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertIn("_FOR_ITER_GEN_FRAME", get_opnames(ex)) self.assertIn("_FOR_ITER_GEN_FRAME", get_opnames(ex))
@ -1342,7 +1344,7 @@ class TestUopsOptimization(unittest.TestCase):
def test_guard_type_version_removed(self): def test_guard_type_version_removed(self):
def thing(a): def thing(a):
x = 0 x = 0
for _ in range(100): for _ in range(TIER2_THRESHOLD):
x += a.attr x += a.attr
x += a.attr x += a.attr
return x return x
@ -1353,7 +1355,7 @@ class TestUopsOptimization(unittest.TestCase):
res, ex = self._run_with_optimizer(thing, Foo()) res, ex = self._run_with_optimizer(thing, Foo())
opnames = list(iter_opnames(ex)) opnames = list(iter_opnames(ex))
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(res, 200) self.assertEqual(res, TIER2_THRESHOLD * 2)
guard_type_version_count = opnames.count("_GUARD_TYPE_VERSION") guard_type_version_count = opnames.count("_GUARD_TYPE_VERSION")
self.assertEqual(guard_type_version_count, 1) self.assertEqual(guard_type_version_count, 1)
@ -1367,7 +1369,7 @@ class TestUopsOptimization(unittest.TestCase):
def thing(a): def thing(a):
x = 0 x = 0
for _ in range(100): for _ in range(TIER2_THRESHOLD):
x += a.attr x += a.attr
fn() fn()
x += a.attr x += a.attr
@ -1379,7 +1381,7 @@ class TestUopsOptimization(unittest.TestCase):
res, ex = self._run_with_optimizer(thing, Foo()) res, ex = self._run_with_optimizer(thing, Foo())
opnames = list(iter_opnames(ex)) opnames = list(iter_opnames(ex))
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(res, 200) self.assertEqual(res, TIER2_THRESHOLD * 2)
guard_type_version_count = opnames.count("_GUARD_TYPE_VERSION") guard_type_version_count = opnames.count("_GUARD_TYPE_VERSION")
self.assertEqual(guard_type_version_count, 1) self.assertEqual(guard_type_version_count, 1)
@ -1390,13 +1392,13 @@ class TestUopsOptimization(unittest.TestCase):
def thing(a): def thing(a):
x = 0 x = 0
for i in range(100): for i in range(TIER2_THRESHOLD + 100):
x += a.attr x += a.attr
# for the first 90 iterations we set the attribute on this dummy function which shouldn't # for the first (TIER2_THRESHOLD + 90) iterations we set the attribute on this dummy function which shouldn't
# trigger the type watcher # trigger the type watcher
# then after 90 it should trigger it and stop optimizing # then after for the next 10 it should trigger it and stop optimizing
# Note that the code needs to be in this weird form so it's optimized inline without any control flow # Note that the code needs to be in this weird form so it's optimized inline without any control flow
setattr((Foo, Bar)[i < 90], "attr", 2) setattr((Foo, Bar)[i < TIER2_THRESHOLD + 90], "attr", 2)
x += a.attr x += a.attr
return x return x
@ -1410,7 +1412,7 @@ class TestUopsOptimization(unittest.TestCase):
opnames = list(iter_opnames(ex)) opnames = list(iter_opnames(ex))
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(res, 219) self.assertEqual(res, (TIER2_THRESHOLD * 2) + 219)
guard_type_version_count = opnames.count("_GUARD_TYPE_VERSION") guard_type_version_count = opnames.count("_GUARD_TYPE_VERSION")
self.assertEqual(guard_type_version_count, 2) self.assertEqual(guard_type_version_count, 2)
@ -1449,7 +1451,7 @@ class TestUopsOptimization(unittest.TestCase):
def thing(a): def thing(a):
x = 0 x = 0
for i in range(100): for i in range(TIER2_THRESHOLD):
x += a.attr x += a.attr
x += a.attr x += a.attr
return x return x
@ -1458,7 +1460,7 @@ class TestUopsOptimization(unittest.TestCase):
attr = 1 attr = 1
res, ex = self._run_with_optimizer(thing, Foo()) res, ex = self._run_with_optimizer(thing, Foo())
self.assertEqual(res, 200) self.assertEqual(res, TIER2_THRESHOLD * 2)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)
self.assertEqual(list(iter_opnames(ex)).count("_GUARD_TYPE_VERSION"), 1) self.assertEqual(list(iter_opnames(ex)).count("_GUARD_TYPE_VERSION"), 1)
self.assertTrue(ex.is_valid()) self.assertTrue(ex.is_valid())
@ -1494,7 +1496,7 @@ class TestUopsOptimization(unittest.TestCase):
opt = _testinternalcapi.new_uop_optimizer() opt = _testinternalcapi.new_uop_optimizer()
with temporary_optimizer(opt): with temporary_optimizer(opt):
testfunc(20) testfunc(TIER2_THRESHOLD)
ex = get_first_executor(testfunc) ex = get_first_executor(testfunc)
self.assertIsNotNone(ex) self.assertIsNotNone(ex)

2
Misc/NEWS.d/next/Core_and_Builtins/2024-11-13-17-18-13.gh-issue-126795._JBX9e.rst Normal file
View File

@ -0,0 +1,2 @@
Increase the threshold for JIT code warmup. Depending on platform and workload,
this can result in performance gains of 1-9% and memory savings of 3-5%.

2
Modules/_testinternalcapi.c
View File

@ -2229,7 +2229,7 @@ module_exec(PyObject *module)
} }
if (PyModule_Add(module, "TIER2_THRESHOLD", if (PyModule_Add(module, "TIER2_THRESHOLD",
PyLong_FromLong(JUMP_BACKWARD_INITIAL_VALUE)) < 0) { PyLong_FromLong(JUMP_BACKWARD_INITIAL_VALUE + 1)) < 0) {
return 1; return 1;
} }