| 19:40 | jnettlet | cbxbiker61, what distro are you using? |
| 20:23 | cbxbiker61 | jnettlet, i roll my own - xilka |
| 20:23 | jnettlet | cbxbiker61, which version of glibc? |
| 20:23 | cbxbiker61 | gives me the freedom to do whatever strikes my fancy |
| 20:24 | cbxbiker61 | 2.22 |
| 20:24 | jnettlet | K, I think my patch should work although I am still thinking about tweaking it more |
| 20:25 | cbxbiker61 | yeah, i'll massage it if it doesn't apply cleanly |
| 20:27 | jnettlet | yeah let me just make a note of where I got the memcpy algorithm from |
| 20:27 | jnettlet | don't want anyone thinking I am stealing it. |
| 20:59 | jnettlet | cbxbiker61, Here you go. https://dl.dropboxusercontent.com/u/736509/0001-Test-optimization-for-Cortex-A9-cpus.patch |
| 21:00 | jnettlet | fyi, I am seeing this bump in scp. 100% 1685MB 22.2MB/s 01:16 |
| 21:00 | jnettlet | normally I may see 12MB/s usually around 9-10MB/s |
| 21:12 | cbxbiker61 | looks good, where does MEMCPY_NEON get set? |
| 21:14 | jnettlet | cbxbiker61, I am pretty sure the build script picks it up |
| 21:15 | jnettlet | cbxbiker61, also this is highly experimental. do not test this on something that you value the data. |
| 21:15 | jnettlet | I have done some testing and haven't seen any problems, but there are always corner cases |
| 21:16 | cbxbiker61 | i was just thinking that i may want to break that out into a separate func, and decide at runtime that neon is supported |
| 21:16 | cbxbiker61 | my current glibc, is used on cubox1 without neon also |
| 21:17 | cbxbiker61 | it would keep library management more simple, rather than having two versions of glibc |
| 21:18 | jnettlet | cbxbiker61, of course you can just build and install libc.so.6 and then install it in another directory and LD_PRELOAD it |
| 21:18 | cbxbiker61 | yeah, i'll test it out properly before i trust it thoroughly |