Adds armv5 build and test target.

Adds armv5 build and test target.

R=regis@google.com

Committed: https://code.google.com/p/dart/source/detail?r=44828

[zra, 2015-03-31 20:06:27]

zra@google.com changed reviewers:
+ iposva@google.com, regis@google.com

[regis, 2015-03-31 21:03:43]

LGTM, except for the string "armv5".

If I understand correctly, what is relevant here is that we compile for arm
thumb, not that the arm version is armv5.

armv5 can run both 32-bit or 16-bit (thumb) instructions and I can imagine an
armv5 board requiring 32-bit instructions (target arch would still be "arm", not
"armv5").

I think that "armthumb" (or just "thumb") would be better, even if we require
armv5te code for our only application requiring thumb so far. We can always
rename it to "thumbarmv5te" when we need a "thumbarmv7".

[zra, 2015-03-31 21:38:26]

On 2015/03/31 21:03:43, regis wrote:
> LGTM, except for the string "armv5".
> 
> If I understand correctly, what is relevant here is that we compile for arm
> thumb, not that the arm version is armv5.
> 
> armv5 can run both 32-bit or 16-bit (thumb) instructions and I can imagine an
> armv5 board requiring 32-bit instructions (target arch would still be "arm",
not
> "armv5").
> 
> I think that "armthumb" (or just "thumb") would be better, even if we require
> armv5te code for our only application requiring thumb so far. We can always
> rename it to "thumbarmv5te" when we need a "thumbarmv7".

I think we are going to move all arm builds to --mthumb since it has a better
size/performance trade-off on the platforms we're running on. So that would make
the salient difference between the arm and armv5 targets the --march=armv5te
flag (which necessitates --mlong-calls for our VM code). I'd prefer to keep
armv5, or change it to armv5te, and wait to bifurcate arm vs. thumb in the
target names until we are actually supporting a platform that doesn't have
thumb. In a subsequent change, I'll add an armv6 target and rename arm to armv7,
passing appropriate --march= flags for both targets. WDYT?

[regis, 2015-03-31 21:58:03]

On 2015/03/31 21:38:26, zra wrote:
> On 2015/03/31 21:03:43, regis wrote:
> > LGTM, except for the string "armv5".
> > 
> > If I understand correctly, what is relevant here is that we compile for arm
> > thumb, not that the arm version is armv5.
> > 
> > armv5 can run both 32-bit or 16-bit (thumb) instructions and I can imagine
an
> > armv5 board requiring 32-bit instructions (target arch would still be "arm",
> not
> > "armv5").
> > 
> > I think that "armthumb" (or just "thumb") would be better, even if we
require
> > armv5te code for our only application requiring thumb so far. We can always
> > rename it to "thumbarmv5te" when we need a "thumbarmv7".
> 
> I think we are going to move all arm builds to --mthumb since it has a better
> size/performance trade-off on the platforms we're running on. So that would
make
> the salient difference between the arm and armv5 targets the --march=armv5te
> flag (which necessitates --mlong-calls for our VM code). I'd prefer to keep
> armv5, or change it to armv5te, and wait to bifurcate arm vs. thumb in the
> target names until we are actually supporting a platform that doesn't have
> thumb. In a subsequent change, I'll add an armv6 target and rename arm to
armv7,
> passing appropriate --march= flags for both targets. WDYT?

I was not aware we were going to generate thumb code for all arm builds. In that
case, sure, "armv5" is fine.

Still LGTM.

I'm actually surprised that we have no issue with arm thumb interworking. I do
not remember ever testing it (since my ARM V8 days). 

Thanks!

[zra, 2015-03-31 23:11:47]

Committed patchset #1 (id:1) manually as r44828 (presubmit successful).