[HP aCC 6.10] native library faster than stdcxx 4.2.1

• Key: STDCXX-928
• Type: Improvement
• Status: Open
• Priority: Critical
• Assignee: Unassigned
• Reporter: Martin Sebor
• Component/s: 23. Containers, 25. Algorithms
• Affects Version/s: 4.2.1
• Fix Version/s: 4.3.0

Description

When compiled with HP aCC 6.10 on HP-UX 11.23/IPF and stdcxx 4.2.1 the Stepanov and Stroustup's Standard Container Benchmark runs with the following output.

size	array	vector<T*>	vector<iterator>	deque	list	set	multiset
10	1.11	1.35	1.34	3.79	3.76	2.41	3.24
100	0.47	0.50	0.50	1.38	1.92	1.07	1.62
1000	0.45	0.48	0.47	1.25	1.56	0.82	1.17
10000	0.45	0.47	0.47	1.23	1.63	0.78	1.09
100000	0.50	0.51	0.51	1.41	3.65	1.29	2.36
1000000	0.56	0.58	0.58	1.51	7.08	3.60	4.62

Total runtimes:

real    1m6.370s
user    1m5.920s
sys     0m0.130s

When compiled with the same compiler but the native C++ Standard Library the same benchmarks runs with the output below. Note that in most cases the times in this table are better than in the one above.

size	array	vector<T*>	vector<iterator>	deque	list	set	multiset
10	1.06	1.11	1.12	4.00	3.49	2.35	3.05
100	0.46	0.47	0.47	1.38	1.80	0.98	1.51
1000	0.44	0.45	0.44	1.09	1.46	0.75	1.11
10000	0.42	0.43	0.43	0.97	1.54	0.73	1.02
100000	0.47	0.45	0.46	0.99	3.57	1.25	2.32
1000000	0.51	0.52	0.50	1.02	6.83	3.54	4.70

Total runtimes:

real    1m3.320s
user    1m2.800s
sys     0m0.140s

Comments

[Martin Sebor]

Similar results with aCC 6.14:

aCC 6.14 +O2, HP-UX 11.31, 1.6GHz Itanium II, stdcxx 4.1.2 (12d build type)

size	array	vector<T*>	vector<iterator>	deque	list	set	multiset
10	0.74	0.86	0.86	2.58	2.71	1.78	2.46
100	0.39	0.42	0.43	1.12	1.58	0.92	1.40
1000	0.39	0.42	0.41	1.05	1.34	0.74	1.06
10000	0.39	0.41	0.41	1.05	1.40	0.71	0.98
100000	0.41	0.42	0.42	1.14	1.88	0.92	1.33
1000000	0.45	0.46	0.45	1.23	4.01	2.01	2.88

Runtimes:

real    0m47.524s
user    0m47.210s
sys     0m0.140s

and

aCC 6.14 -AA +O2 -mt, HP-UX 11.31, 1.6GHz Itanium II, libstd_v2 2.2.1

size	array	vector<T*>	vector<iterator>	deque	list	set	multiset
10	0.70	0.72	0.72	2.52	2.47	1.48	2.08
100	0.42	0.43	0.42	1.07	1.45	0.78	1.20
1000	0.38	0.39	0.38	0.88	1.24	0.64	0.92
10000	0.37	0.38	0.38	0.78	1.33	0.63	0.86
100000	0.38	0.38	0.38	0.74	1.87	0.79	1.19
1000000	0.41	0.40	0.40	0.80	4.14	1.89	2.77

Runtimes:

real    0m43.061s
user    0m42.920s
sys     0m0.120s

[Martin Sebor]

See also this remotely relevant post.

[Martin Sebor]

Let's try to look into why we're slower in the 4.3 timeframe.
Taking a SWAG at the effort involved...