ppl: update to 0.12

[ryandesign@…]

A new version of ppl is available since February:

$ port livecheck ppl
ppl seems to have been updated (port version: 0.11.2, new version: 0.12)

If this gcc46 and other ports depending on ppl still work with this new version, perhaps the port should be updated.

The homepage URL has changed too.

[adfernandes@…]

ppl and cloog (which has also been updated) are integral to the loop-transform and vectorizer for gcc46. I was just surfing the latest gcc-4.6.3 build instructions, and it seems we should be good to upgrade ppl, cloog, and make the cloog variant default on gcc46 (or really, just remove the variant and make it a strict dependency).

I'll build and test when I have time; this could take a while... :-)

[adfernandes@…]

Just had a great round of fun with this. Although cloog-0.17.0 is out, gcc46 and gcc47 both require cloog-0.16.1, specifically. However, there are "serious" bugs fixed in cloog-0.16.1 that has been version bumped to cloog-0.16.3, and cloog-0.16.3 reports itself as cloog-0.16.1 so gcc46 and gcc47 can build it.

Did you follow that? :-)

I'm finishing building gcc46 and gcc47 with ppl-0.12.1 and cloog-0.16.3 with the isl backend... when tested, will commit.

[adfernandes@…]

Committed in r92829

[jmr@…]

You rev bumped gcc46 and gcc47, but forgot gcc45. Done in r92863.

You also enabled cloog for gcc46 and gcc47, but not gcc48. Should you do it for the latter as well?

[adfernandes@…]

I forgot gcc48, and will do that today.

I hadn't forgotten gcc45 (except to mention it). The graphite optimization stuff (ppl and cloog) has been going through a lot of changes that are not super coordinated between groups.

Sometime around gcc45 the backend for cloog changed from ppl to isl. Technically, isl is still in beta, although it is widely used and tested. ppl is still needed outside of cloog for other parts of the loop opimization pipeline.

It's very confusing, and the docs (if any) are worse. It's so bad that the latest ppl doesn't work at all with any gcc branch. I'm not surprised that the clang folks are rolling their own polyhedral optimization system.

Suffice to say that I just didn't have time or inclination to test gcc45. If it builds and runs - great! But sorry if I sound grumpy; such "incompatible versionitis" on such a core bit of software is really annoying.

[adfernandes@…]

Added to gcc48 in r92877

[jmr@…]

The point was just that the ppl library major version changed, so gcc45 was failing to run and needed to be rebuilt. It's kind of important since it's the default variant for a whole lot of ports.

[adfernandes@…]

Ah! Of course - major version change.

Sigh. Thanks.

Like I said, can't wait until clang has their polyhedral optimizations rolled in!