Here's some pretty pictures of the relative CoreMark performance on a Cortex-A9. Note that in accordance with the CoreMark rules, all results are estimates.
Linaro GCC 4.5 and mainline GCC vs 4.5.2: http://people.linaro.org/~michaelh/incoming/all-o3-vs-gcc-4.5.2.png
Linaro GCC 4.5 releases vs our first: http://people.linaro.org/~michaelh/incoming/releases-o3-vs-gcc-linaro-4.5-20...
Linaro GCC 4.5 builds between 2010.08 and 2010.10: http://people.linaro.org/~michaelh/incoming/releases-o3-vs-gcc-linaro-4.5-20...
The 2.5 % drop in performance for 2010.09 is probably due to r99379. The further 2 % drop in 2010.10 is due to re-syncing with mainline 4.5.2.
Performance has been moving solidly upwards but it has to start from the -4.5 % from these two updates. Chung Lin's unsigned extension changes should be interesting...
-- Michael
On Mon, Apr 04, 2011 at 04:31:45PM +1200, Michael Hope wrote:
Linaro GCC 4.5 and mainline GCC vs 4.5.2: http://people.linaro.org/~michaelh/incoming/all-o3-vs-gcc-4.5.2.png
Interesting and nice to see the positive progress, but it does seem from the picture it's hard to get a grasp of exactly what makes that particular benchmark faster. Can we/are we setting something up so we can always benchmark daily and pick up regressions easily, or is the issue not so much finding what causes the regressions as it is solving them? IOW, are most of the slowdowns happening because people are replumbing the code and the performance regression is fallout, or is it because they're tuning looking at something else?
Linaro GCC 4.5 releases vs our first: http://people.linaro.org/~michaelh/incoming/releases-o3-vs-gcc-linaro-4.5-20...
Linaro GCC 4.5 builds between 2010.08 and 2010.10: http://people.linaro.org/~michaelh/incoming/releases-o3-vs-gcc-linaro-4.5-20...
Trying to fool us with the old same URL trick eh?
On Tue, Apr 5, 2011 at 9:20 AM, Christian Robottom Reis kiko@linaro.org wrote:
On Mon, Apr 04, 2011 at 04:31:45PM +1200, Michael Hope wrote:
Linaro GCC 4.5 releases vs our first: http://people.linaro.org/~michaelh/incoming/releases-o3-vs-gcc-linaro-4.5-20...
Linaro GCC 4.5 builds between 2010.08 and 2010.10: http://people.linaro.org/~michaelh/incoming/releases-o3-vs-gcc-linaro-4.5-20...
Trying to fool us with the old same URL trick eh?
Pfft. I meant: http://people.linaro.org/~michaelh/incoming/builds-o3-vs-gcc-linaro-4.5-2010...
which answers your earlier question as it shows all the revisions I have builds for and narrows down what introduced the regressions.
-- Michael
linaro-toolchain@lists.linaro.org
-
Christian Robottom Reis -
Michael Hope