# 00:19:53 |
Tick_ |
quits : Ping timeout: 246 seconds |
# 03:57:59 |
nandub |
joins #crosstool-ng |
# 04:44:45 |
nandub |
quits : Quit: Get MacIrssi - http://www.sysctl.co.uk/projects/macirssi/ |
# 04:52:48 |
nandub |
joins #crosstool-ng |
# 04:53:00 |
nandub |
parts #crosstool-ng |
# 04:55:55 |
nandub |
joins #crosstool-ng |
# 06:36:49 |
nandub |
quits : Quit: Get MacIrssi - http://www.sysctl.co.uk/projects/macirssi/ |
# 07:36:53 |
smartin |
joins #crosstool-ng |
# 13:16:09 |
impulze |
joins #crosstool-ng |
# 13:21:20 |
impulze |
hi there, after manually adding gcc-4.7 i received the following during the final compiler compilation stage: "checking dynamic linker characteristics... configure: error: Link tests are not allowed after GCC_NO_EXECUTABLES." |
# 13:21:33 |
impulze |
same setup works with gcc-4.6.3, any ideas? |
# 13:31:34 |
impulze |
that's with hg tip btw |
# 13:33:41 |
sh4rm4 |
sed -i 's,gcc_no_link=yes,gcc_no_link=no,' ./libstdc++-v3/configure |
# 13:34:27 |
impulze |
and it won't break? |
# 13:34:36 |
impulze |
sh4rm4: i think another library is affected aswell though |
# 13:34:37 |
sh4rm4 |
you have to try |
# 13:34:57 |
impulze |
well, i mean the runtime |
# 13:34:59 |
impulze |
not compilation :) |
# 13:35:02 |
sh4rm4 |
apply same patch to its configure script |
# 13:35:11 |
impulze |
does that have any affections on the runtime component? |
# 13:35:43 |
sh4rm4 |
i dunno, test what i did when i run into that error and it fixed things for me |
# 13:35:55 |
impulze |
ok cool, thanks will try it now |
# 13:35:58 |
sh4rm4 |
*i tested |
# 13:36:45 |
sh4rm4 |
*that's |
# 13:36:49 |
sh4rm4 |
:) |
# 14:25:58 |
impulze |
sh4rm4: now it fails a bit later |
# 14:26:09 |
impulze |
configure:3989: error: C compiler cannot create executables |
# 14:26:18 |
impulze |
because of: "xgcc: error: unrecognized command line option '-EL'" |
# 14:27:24 |
sh4rm4 |
where does that switch come from ? |
# 14:27:56 |
impulze |
from crosstool-ng |
# 14:28:01 |
impulze |
if you select little endian |
# 15:11:44 |
impulze |
i should probably wait till someone figured out what's wrong :) |
# 15:12:09 |
impulze |
gcc and friends have the worst build system hacks |
# 16:57:53 |
impulze |
sh4rm4: here's a thread regarding that aswell http://www.digipedia.pl/usenet/thread/12142/14610/ |
# 17:56:43 |
smartin |
quits |
# 18:09:55 |
sh4rm4 |
impulze, so apparently configure line 3989 should not invoke xgcc but $(LD_FOR_TARGET) |
# 18:11:29 |
sh4rm4 |
oh, without the braces, ofc |
# 18:12:20 |
impulze |
dunno |
# 18:12:26 |
sh4rm4 |
but i do |
# 18:12:37 |
impulze |
will you provide a patch upstream? |
# 18:12:53 |
sh4rm4 |
why me ? you're the one having the problem |
# 18:13:06 |
impulze |
obviously because i don't know what i'm doing there |
# 18:13:09 |
sh4rm4 |
change configure and try if it works, then file a bug on gcc bugtracker |
# 18:13:22 |
sh4rm4 |
and link the ct-ng thread |
# 18:53:06 |
smartin |
joins #crosstool-ng |
# 19:11:07 |
nandub |
joins #crosstool-ng |
# 19:30:43 |
nandub |
quits : Quit: Get MacIrssi - http://www.sysctl.co.uk/projects/macirssi/ |
# 19:56:14 |
y_morin |
joins #crosstool-ng |
# 21:30:50 |
y_morin |
quits : Quit: Back in a sec... |
# 21:31:36 |
y_morin |
joins #crosstool-ng |
# 21:32:12 |
y_morin |
quits : Client Quit |
# 21:37:57 |
y_morin |
joins #crosstool-ng |
# 22:16:41 |
alan_o |
joins #crosstool-ng |
# 22:18:22 |
alan_o |
I regularly build native and cross side-by-side, and I've noticed that my cross builds are about half as fast as my native ones. It doesn't really bother me, but it is interesting. Have any of you observed this? |
# 22:19:54 |
y_morin |
alan_o: what do you mean by "cross builds" vs "native builds"? |
# 22:20:22 |
y_morin |
alan_o: do you mean "when I cross build a package" vs "when I natively build a package" ? |
# 22:20:43 |
alan_o |
y_morin: native would be just building my code with the GCC that came with ubuntu (64-bit), and cross builds are building from my ct-ng generated compiler for ARM |
# 22:20:50 |
alan_o |
I know there's a lot of variables there |
# 22:21:01 |
alan_o |
But it's surprisingly slower with the ct-ng compiler |
# 22:21:16 |
alan_o |
versions are: native compiler: 4.4.3 ct-ng: 4.4.5 |
# 22:21:19 |
alan_o |
so there's that too :) |
# 22:21:25 |
alan_o |
crosstool-ng 1.9.3 |
# 22:21:37 |
alan_o |
Maybe building for ARM is just slower |
# 22:22:08 |
y_morin |
alan_o: OK, so your experience is that your native toolchain is faster then the cross toolchain. |
# 22:22:21 |
y_morin |
OK, got it |
# 22:22:43 |
alan_o |
yeah, I guess it helps if I use the proper terminology :) |
# 22:22:44 |
y_morin |
What order of magnitude slower? |
# 22:23:01 |
alan_o |
cross toolchain is about half as fast as native toolchain |
# 22:23:16 |
y_morin |
OK, 2 (it would help if I read the question to the end). |
# 22:23:21 |
alan_o |
but again, the native one is amd64 building for amd64 |
# 22:23:53 |
y_morin |
alan_o: I would not say that I am surprised. Disapointed, maybe; surprised, no. |
# 22:24:42 |
y_morin |
An order of 2 is quite OK in my opinion. Should you have said '10', then there wuold have been a problem. 2 is reasonable. We're speaking cc, here! ;-) |
# 22:24:57 |
alan_o |
sure |
# 22:25:04 |
y_morin |
(Yes, '10' is 2, too... Don't confuse me, please! ;-) ) |
# 22:25:26 |
alan_o |
like I said, I'm not bothered by it, and I don't blame crosstool-ng. Really just a curiosity |
# 22:25:45 |
y_morin |
alan_o: that being aid, maybe there *is* an issue. |
# 22:25:53 |
y_morin |
*said. |
# 22:38:05 |
alan_o |
I just did another test, and it's not twice as slow, it's more like 1.5x. So for the native build it's 5.119s real and for the cross build it's 8.401s |
# 22:38:14 |
alan_o |
Maybe it depends on the project |
# 22:38:30 |
alan_o |
I seem to remember seeing 1.5x and 3.x on another project though |
# 22:38:53 |
y_morin |
alan_o: I would think that the target architecture is also a big culprit. |
# 22:39:01 |
alan_o |
yeah, that's what I'd expect |
# 22:39:10 |
alan_o |
I mean it's a completely different back-end |
# 22:39:26 |
y_morin |
alan_o: because ARM is RSIC, it takes more time for the compiler to generate the assembly than it takes for x86_64 than is CISC. |
# 22:39:42 |
y_morin |
At least, such an explanation would make sense. |
# 22:39:45 |
alan_o |
That's true, more instructions |
# 22:39:46 |
alan_o |
yeah |
# 22:40:05 |
y_morin |
So 1.5-2 is OK, I think. |
# 22:40:49 |
y_morin |
alan_o: did you include ./configure-time, or just make-time ? |
# 22:41:06 |
alan_o |
just make time |
# 22:41:33 |
y_morin |
alan_o: OK. I would not be surprised that ./configure would suffer too. |
# 23:31:21 |
y_morin |
quits : Quit: Night! |
# 23:34:52 |
smartin |
quits |