I had a similar WTF moment when I noticed that Lucid's HtmlT over IO was 2x slower than HtmlT over Reader/Identity. Discussions over this mailing list pointed to the lack of INLINE pragmas on various functions, like >>= It was finally resolved via a liberal sprinkling of INLINE -- https://github.com/chrisdone/lucid/pull/67/files Why can't GHC automatically consider *every* function as INLINABLE? Let the GHC heuristics decide what is worthy of being inlined. What's the downside? -- Saurabh. On 3 Mar 2017 5:36 pm, "Matthew Pickering" wrote:

I suspect this is because GHC will not specialise functions when used in other modules unless they are marked INLINABLE.

Matt

On Fri, Mar 3, 2017 at 11:40 AM, Joachim Breitner wrote:

...

Hi,

Am Donnerstag, den 02.03.2017, 18:18 +0300 schrieb David Sorokin:

...

I offer to define the functions of the Control.Category module inlinable:

-- | Right-to-left composition (<<<) :: Category cat => cat b c -> cat a b -> cat a c {-# INLINABLE (<<<) #-} (<<<) = (.)

-- | Left-to-right composition (>>>) :: Category cat => cat a b -> cat b c -> cat a c {-# INLINABLE (>>>) #-} f >>> g = g . f

Perhaps all functions from this module should be marked by this pragma as possible.

I am surprised that GHC does not consider these inlineable automatically. Maybe it should? Would be worth investigating, if someone is interested in some entry-level GHC hacking, why GHC does not treat them as `INLINEABLE` in the first place. They are certainly small enough, I would say…

Joachim

-- Joachim “nomeata” Breitner mail@joachim-breitner.de • https://www.joachim-breitner.de/ XMPP: nomeata@joachim-breitner.de • OpenPGP-Key: 0xF0FBF51F Debian Developer: nomeata@debian.org _______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.

---

Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.

.At a guess, performance of both .hi file handling and analysis for inlining. Don't people already gripe about how "slow" ghc is? In that case, an optional flag which people can turn-on only when preparing the final build to be pushed into production. -- Saurabh.

-fexpose-all-unfoldings will allow all definitions to be inlined across modules but NOT specialised. If you want to turn on more aggressive specialisation then you can also use -fspecialise-aggressively. Using the two together would be more like marking everything as INLINABLE but not exactly the same. On Sat, Mar 4, 2017 at 7:13 PM, Joachim Breitner wrote:

Hi,

Am Samstag, den 04.03.2017, 22:28 +0530 schrieb Saurabh Nanda:

...

I had a similar WTF moment when I noticed that Lucid's HtmlT over IO was 2x slower than HtmlT over Reader/Identity. Discussions over this mailing list pointed to the lack of INLINE pragmas on various functions, like >>=

It was finally resolved via a liberal sprinkling of INLINE -- https:/ /github.com/chrisdone/lucid/pull/67/files

Why can't GHC automatically consider *every* function as INLINABLE? Let the GHC heuristics decide what is worthy of being inlined. What's the downside?

the original post was about INLINEABLE not INLINE – let’s keep these two apart.

INLINEABLE says: „Dear compiler, keep the definition of this function in the interface so that you can, maybe, inline (or specialize) it later.“

INLINE implies INLINEABLE, but also says: „…, and really do inline it“.

BTW, there is a flag -fexpose-all-unfoldings that effectively add INLINEABLE to all functions.

Greetings, Joachim

-- Joachim “nomeata” Breitner mail@joachim-breitner.de • https://www.joachim-breitner.de/ XMPP: nomeata@joachim-breitner.de • OpenPGP-Key: 0xF0FBF51F Debian Developer: nomeata@debian.org

_______________________________________________ Haskell-Cafe mailing list To (un)subscribe, modify options or view archives go to: http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe Only members subscribed via the mailman list are allowed to post.