Replace if xxx is None: with POP_JUMP_IF_NOT_NONE ?

[penguin-wwy]

if xxx is None:

These codes will be compiled to

LOAD_FAST                xxx
LOAD_CONST               NONE
IS_OP                    0
POP_JUMP_IF_FALSE        target

I add new instr POP_JUMP_IF_NOT_NONE and POP_JUMP_IF_NONE to replace these:

LOAD_FAST               xxx
POP_JUMP_IF_NOT_NONE    target

This is very effective for such a sample:

def is_none_test(data):
    if data is None:
        return 1
    return 0

Call this function 1, 000, 000 times

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
1000000    0.091    0.000    0.091    0.000 is_none_test.py:4(is_none_test)

After optimization, 10%+ speedup.

ncalls  tottime  percall  cumtime  percall filename:lineno(function)
1000000    0.080    0.000    0.080    0.000 is_none_test.py:4(is_none_test)

Result on pyperformance:

+-------------------------+---------+-----------------------+
| Benchmark               | main    | opt                   |
+=========================+=========+=======================+
| richards                | 82.9 ms | 79.7 ms: 1.04x faster |
+-------------------------+---------+-----------------------+
| json_loads              | 41.1 us | 39.8 us: 1.03x faster |
+-------------------------+---------+-----------------------+
| meteor_contest          | 138 ms  | 134 ms: 1.03x faster  |
+-------------------------+---------+-----------------------+
| raytrace                | 523 ms  | 507 ms: 1.03x faster  |
+-------------------------+---------+-----------------------+
| crypto_pyaes            | 131 ms  | 127 ms: 1.03x faster  |
+-------------------------+---------+-----------------------+
| logging_silent          | 171 ns  | 167 ns: 1.03x faster  |
+-------------------------+---------+-----------------------+
| sympy_str               | 476 ms  | 466 ms: 1.02x faster  |
+-------------------------+---------+-----------------------+
| telco                   | 12.1 ms | 11.9 ms: 1.02x faster |
+-------------------------+---------+-----------------------+
| sympy_sum               | 250 ms  | 245 ms: 1.02x faster  |
+-------------------------+---------+-----------------------+
| logging_format          | 11.0 us | 10.8 us: 1.02x faster |
+-------------------------+---------+-----------------------+
| sympy_integrate         | 31.1 ms | 30.6 ms: 1.02x faster |
+-------------------------+---------+-----------------------+
| mako                    | 17.5 ms | 17.2 ms: 1.02x faster |
+-------------------------+---------+-----------------------+
| sympy_expand            | 795 ms  | 782 ms: 1.02x faster  |
+-------------------------+---------+-----------------------+
| logging_simple          | 9.95 us | 9.80 us: 1.02x faster |
+-------------------------+---------+-----------------------+
| sqlite_synth            | 5.49 us | 5.41 us: 1.01x faster |
+-------------------------+---------+-----------------------+
| pathlib                 | 27.9 ms | 27.5 ms: 1.01x faster |
+-------------------------+---------+-----------------------+
| 2to3                    | 377 ms  | 372 ms: 1.01x faster  |
+-------------------------+---------+-----------------------+
| deltablue               | 6.70 ms | 6.61 ms: 1.01x faster |
+-------------------------+---------+-----------------------+
| nbody                   | 153 ms  | 151 ms: 1.01x faster  |
+-------------------------+---------+-----------------------+
| dulwich_log             | 97.4 ms | 96.2 ms: 1.01x faster |
+-------------------------+---------+-----------------------+
| pyflate                 | 697 ms  | 690 ms: 1.01x faster  |
+-------------------------+---------+-----------------------+
| chaos                   | 125 ms  | 123 ms: 1.01x faster  |
+-------------------------+---------+-----------------------+
| tornado_http            | 142 ms  | 141 ms: 1.01x faster  |
+-------------------------+---------+-----------------------+
| pickle_pure_python      | 544 us  | 540 us: 1.01x faster  |
+-------------------------+---------+-----------------------+
| chameleon               | 12.0 ms | 11.9 ms: 1.01x faster |
+-------------------------+---------+-----------------------+
| django_template         | 63.6 ms | 63.1 ms: 1.01x faster |
+-------------------------+---------+-----------------------+
| go                      | 220 ms  | 219 ms: 1.01x faster  |
+-------------------------+---------+-----------------------+
| pidigits                | 219 ms  | 218 ms: 1.00x faster  |
+-------------------------+---------+-----------------------+
| nqueens                 | 141 ms  | 142 ms: 1.00x slower  |
+-------------------------+---------+-----------------------+
| python_startup_no_site  | 6.00 ms | 6.02 ms: 1.00x slower |
+-------------------------+---------+-----------------------+
| xml_etree_process       | 95.0 ms | 95.6 ms: 1.01x slower |
+-------------------------+---------+-----------------------+
| regex_compile           | 212 ms  | 213 ms: 1.01x slower  |
+-------------------------+---------+-----------------------+
| unpickle                | 22.6 us | 22.8 us: 1.01x slower |
+-------------------------+---------+-----------------------+
| scimark_sor             | 230 ms  | 233 ms: 1.01x slower  |
+-------------------------+---------+-----------------------+
| unpickle_list           | 6.55 us | 6.67 us: 1.02x slower |
+-------------------------+---------+-----------------------+
| pickle_list             | 4.71 us | 4.81 us: 1.02x slower |
+-------------------------+---------+-----------------------+
| xml_etree_parse         | 203 ms  | 209 ms: 1.03x slower  |
+-------------------------+---------+-----------------------+
| regex_v8                | 33.6 ms | 34.7 ms: 1.03x slower |
+-------------------------+---------+-----------------------+
| scimark_sparse_mat_mult | 7.73 ms | 8.01 ms: 1.04x slower |
+-------------------------+---------+-----------------------+
| regex_effbot            | 3.89 ms | 4.09 ms: 1.05x slower |
+-------------------------+---------+-----------------------+
| spectral_norm           | 166 ms  | 175 ms: 1.05x slower  |
+-------------------------+---------+-----------------------+
| scimark_lu              | 217 ms  | 229 ms: 1.06x slower  |
+-------------------------+---------+-----------------------+
| hexiom                  | 10.2 ms | 10.7 ms: 1.06x slower |
+-------------------------+---------+-----------------------+
| pickle_dict             | 31.2 us | 33.8 us: 1.09x slower |
+-------------------------+---------+-----------------------+
| Geometric mean          | (ref)   | 1.00x faster          |
+-------------------------+---------+-----------------------+

However, I am confused about, why pickle_dict is so slow ? This case is just calling the CFunction in the loop !

[markshannon]

pickle_dict, like many microbenchmarks, is very sensitive to code placement, so unrelated changes can impact it significantly.

[markshannon]

See also #36

[Fidget-Spinner]

Thanks for writing the code and conducting the benchmarks! Unfortunately, 10% speedup on a microbenchmark usually means 0 real-world speedup, so depending on how complex the code is, I don't know if it's worth it :(.

[markshannon]

@penguin-wwy Is the code on a publicly visible branch?

[gvanrossum]

Moving to a discussion. @penguin-wwy if you want to contribute your code please open a PR and reference this issue or the resulting GitHub Discussion.