|
From: Julian S. <js...@ac...> - 2015-01-19 10:53:34
|
On 14/01/15 11:43, Julian Seward wrote: > I believe the implementation is correct. Unfortunately it doesn't seem > to give much of a speedup :-/. Is under investigation. As of revs 14874/3070, the branch now runs perf/bz2 in 10% less time and perf/tinycc in 12% less time, compared to trunk. At least on Intel Haswell. J -- bigcode1 -- bigcode1 trunk :0.06s no: 1.2s (19.3x, -----) me: 2.4s (39.5x, -----) bigcode1 ncode :0.06s no: 1.1s (18.8x, 2.6%) me: 2.3s (38.7x, 2.1%) -- bigcode2 -- bigcode2 trunk :0.06s no: 2.6s (43.5x, -----) me: 5.9s (98.2x, -----) bigcode2 ncode :0.06s no: 2.5s (42.2x, 3.1%) me: 5.9s (97.7x, 0.5%) -- bz2 -- bz2 trunk :0.42s no: 1.4s ( 3.2x, -----) me: 4.5s (10.8x, -----) bz2 ncode :0.42s no: 1.3s ( 3.1x, 2.2%) me: 4.1s ( 9.8x, 9.9%) -- fbench -- fbench trunk :0.17s no: 0.8s ( 4.9x, -----) me: 2.7s (16.0x, -----) fbench ncode :0.17s no: 0.8s ( 4.7x, 3.6%) me: 2.5s (14.9x, 7.0%) -- ffbench -- ffbench trunk :0.18s no: 0.8s ( 4.3x, -----) me: 2.4s (13.3x, -----) ffbench ncode :0.18s no: 0.8s ( 4.2x, 1.3%) me: 2.3s (12.8x, 3.3%) -- heap -- heap trunk :0.05s no: 0.5s (10.2x, -----) me: 4.1s (81.6x, -----) heap ncode :0.05s no: 0.5s ( 9.4x, 7.8%) me: 4.0s (79.8x, 2.2%) -- heap_pdb4 -- heap_pdb4 trunk :0.06s no: 0.5s ( 9.0x, -----) me: 6.2s (103.3x, -----) heap_pdb4 ncode :0.06s no: 0.5s ( 8.5x, 5.6%) me: 6.0s (100.8x, 2.4%) -- many-loss-records -- many-loss-records trunk :0.01s no: 0.2s (19.0x, -----) me: 1.0s (102.0x, -----) many-loss-records ncode :0.01s no: 0.2s (17.0x, 10.5%) me: 1.0s (97.0x, 4.9%) -- many-xpts -- many-xpts trunk :0.03s no: 0.2s ( 7.7x, -----) me: 1.1s (37.7x, -----) many-xpts ncode :0.03s no: 0.2s ( 7.0x, 8.7%) me: 1.1s (36.0x, 4.4%) -- sarp -- sarp trunk :0.01s no: 0.2s (21.0x, -----) me: 1.7s (169.0x, -----) sarp ncode :0.01s no: 0.2s (20.0x, 4.8%) me: 1.6s (160.0x, 5.3%) -- tinycc -- tinycc trunk :0.12s no: 0.8s ( 7.0x, -----) me: 6.2s (51.8x, -----) tinycc ncode :0.12s no: 0.8s ( 6.8x, 2.4%) me: 5.4s (45.3x, 12.5%) |