Project's future plans related to new aarch64 architecture
Hello, I'd like to start a thread talking about the new support for aarch64. This could be the next step in CRUX-ARM where we can start developing a newer toolchain with 64bit support for this architecture. It can be researched to add this support in the next 3.1 release (or at least to develop until we can put hands on devices with this support) Gcc seems to give support to it [1] and this could be very interesting for the next year, when probably we'll start seeing first devices with aarch64 support (armv8) with backwards compatibility and not relay only in ARM emulator (Foundationv8). Could be interesting to start a new git repository (toolchain-aarch64) to start playing with the new armv8 arch? Any thoughts about this? Regards, Victor. [1] http://gcc.gnu.org/gcc-4.8/changes.html -- Learning bit by bit Victor Martinez | http://lokalix.dyndns.org
Hey, On 12/20/2013 12:29 PM, Victor Martinez wrote:
Hello, I'd like to start a thread talking about the new support for aarch64.
+1 Well, first I reorganized your questions in the order that I think more appropriate.
Gcc seems to give support to it [1] and this could be very interesting for the next year, when probably we'll start seeing first devices with aarch64 support (armv8) with backwards compatibility and not relay only in ARM emulator (Foundationv8).
Generally I like the idea and also I think we will soon see someone with an ARMv8 device wondering if CRUX-ARM can run on it.
Could be interesting to start a new git repository (toolchain-aarch64) to start playing with the new armv8 arch?
Sure, anticipate to this sounds good to me.
This could be the next step in CRUX-ARM where we can start developing a newer toolchain with 64bit support for this architecture. It can be researched to add this support in the next 3.1 release (or at least to develop until we can put hands on devices with this support) Any thoughts about this?
I agree with start playing for aarch64 but I honestly do not think we should include it in our future plans for 3.1 for now. As you say, that should happen when we have enough ARMv8 devices. On the other hand, I not want to be negative but realistic. We are a very small team and we have to concentrate our efforts on something concrete but very well done and not try to cover many things we could leave incomplete or failing to maintain properly. Besides, ,I think that start developing for toolchain-aarch64 means that we will again have 2 main releases for our product. In 2.7 we had the softfp main branch which was widely supported by all devices, and introduced a development hardfp branch. In 2.8 we keep de same branches: softp as the main one and hardfp as development branch. In 3.0 we switch our mainline to hardfp which becomes stable and decided to leave the softfp branch. So far so good, but the incorporation of aarch64 means more when you consider the steps taken in CRUX to move from i686 to x86_64, or what is the same, from x32 to x64. You thought we would have to take that product? Purelib or multilib? This things will decide our for first steps fro toolchain-aarch64. I would bet for a version that had its purelib overlays, but also think about the state of the software that can run aarch64. I'm not talking about the kernel or gcc toolchain and all, but for example firefox, chromium, etc. A repo compat32 will also be needed? I think we need to plan this well before anything else. Since the difficulty is not to create a Makefile for toolchain-aarch64 and see that is functional, but is usable, and we can keep all that good work. Regards, Victor.
xmas!! and best regards, -- Jose V Beneyto | http://sepen.it.cx/
On 12/24/13 12:46, Jose V Beneyto wrote:
Hey,
Hello,
On 12/20/2013 12:29 PM, Victor Martinez wrote:
Hello, I'd like to start a thread talking about the new support for aarch64.
+1
Well, first I reorganized your questions in the order that I think more appropriate.
Gcc seems to give support to it [1] and this could be very interesting for the next year, when probably we'll start seeing first devices with aarch64 support (armv8) with backwards compatibility and not relay only in ARM emulator (Foundationv8).
Generally I like the idea and also I think we will soon see someone with an ARMv8 device wondering if CRUX-ARM can run on it.
Probably along the next year... Currently it's in development status and there isn't much information about when those devices (armv8) will be released and sold to people.
Could be interesting to start a new git repository (toolchain-aarch64) to start playing with the new armv8 arch?
Sure, anticipate to this sounds good to me.
Great.
This could be the next step in CRUX-ARM where we can start developing a newer toolchain with 64bit support for this architecture. It can be researched to add this support in the next 3.1 release (or at least to develop until we can put hands on devices with this support) Any thoughts about this?
I agree with start playing for aarch64 but I honestly do not think we should include it in our future plans for 3.1 for now. As you say, that should happen when we have enough ARMv8 devices.
You are right. This was my intention. It's in development status upstream... then we only can start researching about it and being patient waiting for devices to be deployed. Time to sort all the information and to plan how to act.
On the other hand, I not want to be negative but realistic. We are a very small team and we have to concentrate our efforts on something concrete but very well done and not try to cover many things we could leave incomplete or failing to maintain properly.
Yes, this is the same history as always... too many fronts to take care. I think we can start the work flow related to aarch64 and without any hurry. May be someone else see this interesting and want to give a hand.
Besides, ,I think that start developing for toolchain-aarch64 means that we will again have 2 main releases for our product. In 2.7 we had the softfp main branch which was widely supported by all devices, and introduced a development hardfp branch. In 2.8 we keep de same branches: softp as the main one and hardfp as development branch. In 3.0 we switch our mainline to hardfp which becomes stable and decided to leave the softfp branch. So far so good, but the incorporation of aarch64 means more when you consider the steps taken in CRUX to move from i686 to x86_64, or what is the same, from x32 to x64.
We follow CRUX as we did always... and we can use/maintain a development branch until things are stable. As you said, we did it in the past and IMHO it's something we can face. All the time things are there, are improved and fixed with time. Curently without a real device to start playing with it will make development process a bit hard. We need to wait until qemu adds support for it and this will be ready probably when we start seeing armv8 devices.
You thought we would have to take that product? Purelib or multilib? This things will decide our for first steps fro toolchain-aarch64.
I think we can put hands on it and see what happens. All things like this give some knowledge, as always, an excuse to learn a bit more.
I would bet for a version that had its purelib overlays, but also think about the state of the software that can run aarch64. I'm not talking about the kernel or gcc toolchain and all, but for example firefox, chromium, etc. A repo compat32 will also be needed?
Let's see things bit by bit... I think multilib will be the way to go, as CRUX. But we don't know too much about aarch64 and its aarch32 backwards support.
I think we need to plan this well before anything else. Since the difficulty is not to create a Makefile for toolchain-aarch64 and see that is functional, but is usable, and we can keep all that good work. Regards, Victor. xmas!! and best regards,
Yeah, this is the start point of this discussion and for future plans... We have time and resources to sort all the info and make the best from our side. Probably things can change along the time when we put hands on it. Merry christmas for all people!!!! -- Learning bit by bit Victor Martinez | http://lokalix.dyndns.org
On 12/24/2013 02:31 PM, Victor Martinez wrote:
[...]
I would bet for a version that had its purelib overlays, but also think about the state of the software that can run aarch64. I'm not talking about the kernel or gcc toolchain and all, but for example firefox, chromium, etc. A repo compat32 will also be needed?
Let's see things bit by bit... I think multilib will be the way to go, as CRUX. But we don't know too much about aarch64 and its aarch32 backwards support.
I think we need to plan this well before anything else. Since the difficulty is not to create a Makefile for toolchain-aarch64 and see that is functional, but is usable, and we can keep all that good work. Regards, Victor. xmas!! and best regards,
Yeah, this is the start point of this discussion and for future plans... We have time and resources to sort all the info and make the best from our side. Probably things can change along the time when we put hands on it.
Well seems we'll have more conversations about that but we have discussed some things at IRC channel that could be a good idea to have on this thread for now. http://crux-arm.nu/irc/?d=24&m=12&y=2013 (around 13:57) I have ordered some ARMv8 devices to Santa ;D Best wishes, -- Jose V Beneyto | http://sepen.it.cx/
participants (2)
-
Jose V Beneyto -
Victor Martinez