November 2017 - ghc-devs

Re: [commit: ghc] master: llvmGen: Pass vector arguments in vector registers by default (15f788f)
by Gabor Greif 03 Nov '17

03 Nov '17

IIRC there are no more static flags around. So adding ":type: dynamic" each time seems redundant. Could this be automated somehow? Or should we remove ":type:" altogether? Cheers, Gabor On 11/3/17, git(a)git.haskell.org <git(a)git.haskell.org> wrote: > Repository : ssh://git@git.haskell.org/ghc > > On branch : master > Link : > http://ghc.haskell.org/trac/ghc/changeset/15f788f5e5096641245a4f060600a6db9… > >>--------------------------------------------------------------- > > commit 15f788f5e5096641245a4f060600a6db9cbc2c4e > Author: Ben Gamari <bgamari.foss(a)gmail.com> > Date: Thu Nov 2 17:28:40 2017 -0400 > > llvmGen: Pass vector arguments in vector registers by default > > Earlier this year Edward Kmett requested [1] that we enable passing of > vector values in vector registers by default. The GHC calling convention > changes have been in LLVM for a number of years now so let's just flip > the switch. > > [1] https://mail.haskell.org/pipermail/ghc-devs/2017-March/013905.html > > Reviewers: austin > > Subscribers: rwbarton, thomie > > Differential Revision: https://phabricator.haskell.org/D4142 > > >>--------------------------------------------------------------- > > 15f788f5e5096641245a4f060600a6db9cbc2c4e > compiler/main/DynFlags.hs | 5 +++-- > docs/users_guide/using-optimisation.rst | 12 ++++++++++++ > 2 files changed, 15 insertions(+), 2 deletions(-) > > diff --git a/compiler/main/DynFlags.hs b/compiler/main/DynFlags.hs > index a421284..825497e 100644 > --- a/compiler/main/DynFlags.hs > +++ b/compiler/main/DynFlags.hs > @@ -3757,7 +3757,7 @@ fFlagsDeps = [ > flagSpec "kill-one-shot" Opt_KillOneShot, > flagSpec "late-dmd-anal" Opt_LateDmdAnal, > flagSpec "liberate-case" Opt_LiberateCase, > - flagHiddenSpec "llvm-pass-vectors-in-regs" > Opt_LlvmPassVectorsInRegisters, > + flagSpec "llvm-pass-vectors-in-regs" > Opt_LlvmPassVectorsInRegisters, > flagHiddenSpec "llvm-tbaa" Opt_LlvmTBAA, > flagHiddenSpec "llvm-fill-undef-with-garbage" > Opt_LlvmFillUndefWithGarbage, > flagSpec "loopification" Opt_Loopification, > @@ -4051,7 +4051,8 @@ defaultFlags settings > Opt_RPath, > Opt_SharedImplib, > Opt_SimplPreInlining, > - Opt_VersionMacros > + Opt_VersionMacros, > + Opt_LlvmPassVectorsInRegisters > ] > > ++ [f | (ns,f) <- optLevelFlags, 0 `elem` ns] > diff --git a/docs/users_guide/using-optimisation.rst > b/docs/users_guide/using-optimisation.rst > index 4714de7..fc958e0 100644 > --- a/docs/users_guide/using-optimisation.rst > +++ b/docs/users_guide/using-optimisation.rst > @@ -493,6 +493,18 @@ by saying ``-fno-wombat``. > self-recursive saturated tail calls into local jumps rather than > function calls. > > +.. ghc-flag:: -fllvm-pass-vectors-in-regs > + :shortdesc: Pass vector value in vector registers for function calls > + :type: dynamic > + :reverse: -fno-llvm-pass-vectors-in-regs > + :category: > + > + :default: on > + > + Instructs GHC to use the platform's native vector registers to pass > vector > + arguments during function calls. As with all vector support, this > requires > + :ghc-flag:`-fllvm`. > + > .. ghc-flag:: -fmax-inline-alloc-size=⟨n⟩ > :shortdesc: *default: 128.* Set the maximum size of inline array > allocations > to ⟨n⟩ bytes (default: 128). > > _______________________________________________ > ghc-commits mailing list > ghc-commits(a)haskell.org > http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-commits >

2 1

Re: [ANNOUNCE] GHC 8.2.2 release candidate 2
by Ryan Scott 03 Nov '17

03 Nov '17

I built all of Stackage with this RC the other night and didn't observe any regressions. Hooray! (Thanks to Doug Wilson for helping me get set up with his https://github.com/duog/build-stackage repo.) Ryan S.

2 1

Bringing some observable sharing to GHCi
by David Feuer 02 Nov '17

02 Nov '17

The :sprint, :print, and :force commands in GHCi fall into infinite loops when confronted by cyclical data. This bit me hard in https://phabricator.haskell.org/D4085 because that produces cyclical TypeReps, which is trouble for (e.g.) the test break011 which tries to :force a SomeException (which wraps an Exception dictionary, which has a Typeable constraint). I could try coming up with a fix myself, but I'm rather curious whether some of the work you (or others) have already done on observing GHC data graphs could be yanked into GHCi itself for this purpose. We want 1. To be able to display cyclical data in some sensible way. > x = "hi" : x > x `seq` () > :print x should print some useful representation of x. 2. To be able to force cyclical data without looping. > x = "hi" : x > :force x should print a useful representation of x.

4 3

When does GHC produce unlifted `let` bindings?
by Sebastian Graf 02 Nov '17

02 Nov '17

Hi folks, I was debugging a Core-to-Core transform for TEST=spec-inline <https://github.com/ghc/ghc/blob/fe04f3783b662c52c4a0ff36b2d62a7a575998a5/te…> and was wondering (yet again) why GHC produces unlifted `let` bindings in Core like it seems supposed to be doing <https://github.com/ghc/ghc/blob/fe04f3783b662c52c4a0ff36b2d62a7a575998a5/te…> . - Why doesn't this use `case` instead? - Is there a semantic difference? - Can `case` be used with unlifted types? - And if not, why can `let`? Unlifted `let` seems much close to `case` than to `let`. Some GHC passes seem to agree. <https://github.com/ghc/ghc/search?l=Haskell&q=%22strict+let%22&type=&utf8=%…> Cheers, Sebastian

3 4

RE: Building the docs, again
by Andrey Mokhov 01 Nov '17

01 Nov '17

> Maybe Hadrian will help here? Hadrian can build documentation in various formats but the current implementation is very coarse-grain: we either build all possible docs or no docs at all. It would be relatively easy to add more fine-grain documentation targets and modes. If someone could design/describe the requirements in a ticket that would be great. Cheers, Andrey From: Simon Peyton Jones [mailto:simonpj@microsoft.com] Sent: 01 November 2017 08:28 To: Iavor Diatchki <iavor.diatchki(a)gmail.com>; ghc-devs(a)haskell.org; Andrey Mokhov <andrey.mokhov(a)newcastle.ac.uk> Subject: RE: Building the docs, again It is quite confusing that we can be setting the same variable to different values in different places. It would also be quite helpful if we modified the Makefile to say `documentation build target is disabled` or some such, rather then going ahead and building the whole of GHC, only at the end to inform me that it doesn't know what is `html`. Maybe Hadrian will help here? Simon From: ghc-devs [mailto:ghc-devs-bounces@haskell.org] On Behalf Of Iavor Diatchki Sent: 31 October 2017 21:49 To: ghc-devs(a)haskell.org<mailto:ghc-devs@haskell.org> Subject: Re: Building the docs, again Hello, never mind, I figured it out---apparently the build targets were disabled for the `quick` flavor, which I was using. It is quite confusing that we can be setting the same variable to different values in different places. It would also be quite helpful if we modified the Makefile to say `documentation build target is disabled` or some such, rather then going ahead and building the whole of GHC, only at the end to inform me that it doesn't know what is `html`. -Iavor On Tue, Oct 31, 2017 at 2:26 PM Iavor Diatchki <iavor.diatchki(a)gmail.com<mailto:iavor.diatchki@gmail.com>> wrote: Hello, sometime ago, I asked if it is possible to build just the GHC docs, without building the compiler. I was told to just run `make html`. This does not appear to work, the command seems to just start building GHC. What am I doing wrong? Here is the output I see, running the command from the root of the GHC tree. > make html V=0 ===--- building phase 0 make --no-print-directory -f ghc.mk<https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fghc.mk&data…> phase=0 phase_0_builds make[1]: Nothing to be done for 'phase_0_builds'. ===--- building phase 1 make --no-print-directory -f ghc.mk<https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fghc.mk&data…> phase=1 phase_1_builds HC [stage 0] compiler/stage1/build/BufWrite.o HC [stage 0] compiler/stage1/build/Pretty.o compilation IS NOT required HC [stage 0] compiler/stage1/build/PprColour.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Outputable.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Json.o compilation IS NOT required HC [stage 0] compiler/stage1/build/SrcLoc.o compilation IS NOT required HC [stage 0] compiler/stage1/build/BasicTypes.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Unique.o HC [stage 0] compiler/stage1/build/Packages.o-boot compilation IS NOT required HC [stage 0] compiler/stage1/build/FiniteMap.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Name.o-boot (then I interrupted it). Also, in `mk/config.mk<https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fconfig.mk&d…>` I see this: BUILD_SPHINX_HTML = YES -Iavor

3 2

Building the docs, again
by Iavor Diatchki 01 Nov '17

01 Nov '17

Hello, sometime ago, I asked if it is possible to build just the GHC docs, without building the compiler. I was told to just run `make html`. This does not appear to work, the command seems to just start building GHC. What am I doing wrong? Here is the output I see, running the command from the root of the GHC tree. > make html V=0 ===--- building phase 0 make --no-print-directory -f ghc.mk phase=0 phase_0_builds make[1]: Nothing to be done for 'phase_0_builds'. ===--- building phase 1 make --no-print-directory -f ghc.mk phase=1 phase_1_builds HC [stage 0] compiler/stage1/build/BufWrite.o HC [stage 0] compiler/stage1/build/Pretty.o compilation IS NOT required HC [stage 0] compiler/stage1/build/PprColour.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Outputable.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Json.o compilation IS NOT required HC [stage 0] compiler/stage1/build/SrcLoc.o compilation IS NOT required HC [stage 0] compiler/stage1/build/BasicTypes.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Unique.o HC [stage 0] compiler/stage1/build/Packages.o-boot compilation IS NOT required HC [stage 0] compiler/stage1/build/FiniteMap.o compilation IS NOT required HC [stage 0] compiler/stage1/build/Name.o-boot (then I interrupted it). Also, in `mk/config.mk` I see this: BUILD_SPHINX_HTML = YES -Iavor

2 2