|
From: Florian K. <br...@ac...> - 2012-04-21 03:15:44
|
Here are the performance numbers for s390 (z10-EC). Measured with --perf=5 Before: without tchaining -- Running tests in perf ---------------------------------------------- bigcode1 trunk :0.50s no: 5.9s (11.9x, -----) me:11.0s (22.1x, ----) bigcode2 trunk :0.50s no:14.2s (28.4x, -----) me:28.1s (56.1x, ----) bz2 trunk :1.06s no: 6.6s ( 6.3x, -----) me:26.7s (25.2x, ----) fbench trunk :0.80s no: 3.0s ( 3.8x, -----) me:11.8s (14.7x, ----) ffbench trunk :0.52s no: 1.6s ( 3.2x, -----) me: 6.4s (12.3x, ----) heap trunk :0.34s no: 2.1s ( 6.1x, -----) me:15.2s (44.6x, ----) heap_pdb4 trunk :0.33s no: 2.2s ( 6.6x, -----) me:22.0s (66.6x, ----) many-loss trunk :0.03s no: 0.6s (19.3x, -----) me: 3.9s (128.3x, ---) many-xpts trunk :0.08s no: 0.7s ( 9.2x, -----) me: 5.4s (67.5x, ----) sarp trunk :0.04s no: 0.7s (16.2x, -----) me: 6.1s (153.2x, ---) tinycc trunk :0.36s no: 5.2s (14.6x, -----) me:29.7s (82.5x, ----) -- Finished tests in perf ---------------------------------------------- After: with tchaining -- Running tests in perf ---------------------------------------------- bigcode1 trunk :0.50s no: 4.6s ( 9.3x, -----) me: 8.1s (16.2x, ----) bigcode2 trunk :0.51s no: 9.0s (17.6x, -----) me:17.0s (33.4x, ----) bz2 trunk :1.07s no: 5.5s ( 5.1x, -----) me:24.0s (22.4x, ----) fbench trunk :0.80s no: 2.8s ( 3.5x, -----) me:10.9s (13.6x, ----) ffbench trunk :0.51s no: 1.5s ( 2.9x, -----) me: 5.8s (11.3x, ----) heap trunk :0.35s no: 1.9s ( 5.3x, -----) me:14.8s (42.3x, ----) heap_pdb4 trunk :0.33s no: 2.0s ( 6.2x, -----) me:21.4s (64.8x, ----) many-loss trunk :0.03s no: 0.4s (14.3x, -----) me: 3.5s (116.3x, ---) many-xpts trunk :0.08s no: 0.7s ( 8.5x, -----) me: 5.1s (63.5x, ----) sarp trunk :0.04s no: 0.5s (12.5x, -----) me: 5.8s (145.8x, ---) tinycc trunk :0.35s no: 4.1s (11.8x, -----) me:26.7s (76.4x, ----) -- Finished tests in perf ---------------------------------------------- Well worth the effort. Florian |
|
From: Nicholas N. <n.n...@gm...> - 2012-10-14 13:19:33
|
On Sat, Apr 21, 2012 at 1:15 PM, Florian Krohm <br...@ac...> wrote: > Here are the performance numbers for s390 (z10-EC). Measured with --perf=5 > > Before: without tchaining > > -- Running tests in perf ---------------------------------------------- > bigcode1 trunk :0.50s no: 5.9s (11.9x, -----) me:11.0s (22.1x, ----) > bigcode2 trunk :0.50s no:14.2s (28.4x, -----) me:28.1s (56.1x, ----) > bz2 trunk :1.06s no: 6.6s ( 6.3x, -----) me:26.7s (25.2x, ----) > fbench trunk :0.80s no: 3.0s ( 3.8x, -----) me:11.8s (14.7x, ----) > ffbench trunk :0.52s no: 1.6s ( 3.2x, -----) me: 6.4s (12.3x, ----) > heap trunk :0.34s no: 2.1s ( 6.1x, -----) me:15.2s (44.6x, ----) > heap_pdb4 trunk :0.33s no: 2.2s ( 6.6x, -----) me:22.0s (66.6x, ----) > many-loss trunk :0.03s no: 0.6s (19.3x, -----) me: 3.9s (128.3x, ---) > many-xpts trunk :0.08s no: 0.7s ( 9.2x, -----) me: 5.4s (67.5x, ----) > sarp trunk :0.04s no: 0.7s (16.2x, -----) me: 6.1s (153.2x, ---) > tinycc trunk :0.36s no: 5.2s (14.6x, -----) me:29.7s (82.5x, ----) > -- Finished tests in perf ---------------------------------------------- > > After: with tchaining > > -- Running tests in perf ---------------------------------------------- > bigcode1 trunk :0.50s no: 4.6s ( 9.3x, -----) me: 8.1s (16.2x, ----) > bigcode2 trunk :0.51s no: 9.0s (17.6x, -----) me:17.0s (33.4x, ----) > bz2 trunk :1.07s no: 5.5s ( 5.1x, -----) me:24.0s (22.4x, ----) > fbench trunk :0.80s no: 2.8s ( 3.5x, -----) me:10.9s (13.6x, ----) > ffbench trunk :0.51s no: 1.5s ( 2.9x, -----) me: 5.8s (11.3x, ----) > heap trunk :0.35s no: 1.9s ( 5.3x, -----) me:14.8s (42.3x, ----) > heap_pdb4 trunk :0.33s no: 2.0s ( 6.2x, -----) me:21.4s (64.8x, ----) > many-loss trunk :0.03s no: 0.4s (14.3x, -----) me: 3.5s (116.3x, ---) > many-xpts trunk :0.08s no: 0.7s ( 8.5x, -----) me: 5.1s (63.5x, ----) > sarp trunk :0.04s no: 0.5s (12.5x, -----) me: 5.8s (145.8x, ---) > tinycc trunk :0.35s no: 4.1s (11.8x, -----) me:26.7s (76.4x, ----) > -- Finished tests in perf ---------------------------------------------- > > Well worth the effort. I realize this is 6 months old, but if you use the --vg option to vg_perf you can directly compare two different Valgrinds and it computes the improvements, instead of having to run it twice and eyeball the differences. Nick |