From mboxrd@z Thu Jan  1 00:00:00 1970
From: Marek Vasut <marex@denx.de>
Date: Thu, 1 Nov 2012 00:45:40 +0100
Subject: [U-Boot] [PATCH] common: Link with GCC instead of LD
In-Reply-To: <20121031233257.5F3F32005BD@gemini.denx.de>
References: <1351701168-7293-1-git-send-email-marex@denx.de>
	<201210312351.35977.marex@denx.de>
	<20121031233257.5F3F32005BD@gemini.denx.de>
Message-ID: <201211010045.40817.marex@denx.de>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

Dear Wolfgang Denk,

> Dear Marek,
> 
> In message <201210312351.35977.marex@denx.de> you wrote:
> > > Could you please explain these supposed benefits in a bit more detail?
> > 
> > Yes, thanks for the review (I really needed one from you on this)!
> > 
> > So, here comes the list:
> > 1) It's cool to link with gcc, duh ;-)
> 
> Is it - and why exactly?

It's all the rage now (better ignore this point) ;-)

> > 2) You can enable LTO eventually -- long term plan -- which proved can
> > shrink the code size a bit
> 
> Do you have a workign example?  You know, we don;t like adding dead
> code...

We've been digging with Albert, but nothing we can release just yet. This is a 
groundwork patch.

> > 3) Usage of LLVM/CLANG is now much more straightforward
> 
> Please elucidate?

Now you can simply replace gcc with clang and that _should_ work. Sure, I expect 
more work will be needed here, but this patch gives you all the ground to do 
that.

> > 4) Binutils usually lag behind the GCC development a bit, by using GCC to
> > do the link, we can remedy this problem
> 
> Grrrrghhhg.  I lost you here.  By not calling ld directly (where we
> know exactly what we are doing), but through a (somewhat unknown) GCC
> framework, we can magicly adjust any (claimed - by you) version
> differences?   I cannot parse this.

GCC can do all the optimization magic (LTO ...) and then just let linker do the 
linking.

> > But actually, the 2) is the most important part. LTO will hopefully help
> > us optimize U-Boot for size even better than it is now. But that will
> > still need much more research. So this is only a patch that shall
> > prepare the groundwork for this further research.
> 
> As is, it's just unused or dead code that adds a lot of complexity and
> buys nothing.  Please submit as part of a patch series that makes
> successfull (and tested) use of LTO.

No, I won't, I don't have that part ready yet.

Besides, the sandbox target already uses gcc linking. Shall we also convert it 
to use LD instead then ?

Best regards,
Marek Vasut