Re: optllvm broken on master?

OK, that sounds great. Here's the patch to revert the offending line: https://ghc.haskell.org/trac/ghc/ticket/7942#comment:54 And I'll prepare another patch to implement switching on 3.6 shortly. I'm definitely in support of the LLVM-with-GHC plan! Cheers Luke On Thu, Nov 20, 2014 at 7:26 PM, Austin Seipp wrote:

(Resending to list)

Yes, switching on the version would be preferable since we detect the tool version in a few places already if it's possible. Or at least we should back out the regressing change right now until we ship LLVM with GHC as I proposed.

On Thu, Nov 20, 2014 at 3:49 PM, Luke Iannini wrote:

...

On Thu, Nov 20, 2014 at 1:43 PM, Luke Iannini

wrote:

...

...

Whoops, that's my fault; it shouldn't have slipped into the patch -- this is required for LLVM 3.6 (which I was using to work on my ARM64 patch),

but

...

clearly isn't backwards compatible.

I wonder what's best to do for this? Would it be correct to switch on LLVM versions or should this be considered a bug in LLVM? (it was the only change needed for 3.6)

(I mean to say, it was the only change needed atop Ben's work on 3.5 support in https://phabricator.haskell.org/D155)

...

I will submit a ticket to revert that line in the meantime.

Thanks! Luke

On Thu, Nov 20, 2014 at 2:26 AM, Dr. ERDI Gergo wrote:

...

Hi,

Tests fail on master when run the optllvm way, e.g.:

/usr/bin/opt: /tmp/ghc16190_0/ghc16190_2.ll:611:25: error: expected 'global' or 'constant' @newCAF$alias = private alias i8* @newCAF

I think it's the change in d87fa34:

index 7307725..cdc407c 100644 --- a/compiler/llvmGen/Llvm/PpLlvm.hs +++ b/compiler/llvmGen/Llvm/PpLlvm.hs @@ -80,7 +80,7 @@ ppLlvmGlobal (LMGlobal var@(LMGlobalVar _ _ link x

a c)

...

...

dat) = const_link = case c of Global -> ppr link <+> text "global" Constant -> ppr link <+> text "constant" - Alias -> text "alias" <+> ppr link + Alias -> ppr link <+> text "alias"

in ppAssignment var $ const_link <+> rhs <> sect <> align $+$ newLine

--

.--= ULLA! =-----------------. \ http://gergo.erdi.hu \ `---= gergo@erdi.hu =-------' If it ain't broke, I can fix it. _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://www.haskell.org/mailman/listinfo/ghc-devs

-- Regards,

Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/