[llvm-commits] Please review: Support for Hexagon new value jump

[Tony Linthicum]

On 5/1/2012 12:17 AM, Andrew Trick wrote:
> On Apr 25, 2012, at 10:13 PM, Sirish Pande<spande at codeaurora.org>  wrote:
>> Hi,
>>
>> This is the support hexagon new value jump architectural feature.
>>
>> This patch does not give warnings and passes all test cases with make check-all.
> Sirish,
>
> Given that this patch didn't build from trunk at the time you sent it, I wouldn't expect reviewers to pay attention to it. It does build for me after applying a few other patches in the right order.
>
> Since this change is entirely in the Hexagon target, I'm ok with it going in unless someone else objects. One thing you should keep in mind though as you upsteam gobs of messy target code: this code must be maintained by contributors who care about Hexagon. The more assumptions you make about other codegen passes, the heavier the burden will be. If the target isn't kept up to speed, it will gradually be disabled and eventually removed. So you should seriously think about *not* upstreaming experimental or optional passes that other Hexagon users don't need. In fact, are there other Hexagon users? Do you really need to upstream everything?
>
>
Hi Andy,

Thanks for your comments and taking the time to look over these 
patches.  We will be keeping the target up to speed going forward from 
here.  Unfortunately, we allowed ourselves to get badly out of synch and 
have been trying to correct that.  We have stumbled in that process and 
we do apologize for the trouble that it has caused.  As we complete the 
task of updating our backend, we will be sure to have each patch be 
independent of the others as much as is possible and be explicit when we 
cannot.  In the future, we will be immediately upstreaming all of our 
local changes when appropriate to do so and you will not see such large 
target dependent patches from us.

Thanks for your assistance and patience.

Tony