Commit 22a25aa9 authored by's avatar

Document SPECIALISE for imported functions

This is a really useful new facility, but I'd forgotten to document it.
Pls merge to 7.0 branch
parent 92b30c6b
......@@ -7712,27 +7712,48 @@ itself, so an INLINE pragma is always ignored.</para>
<sect3 id="inlinable-pragma">
<title>INLINABLE pragma</title>
<para>An INLINABLE pragma works very like an INLINE pragma, except that:
<para>An <literal>{-# INLINABLE f #-}</literal> pragma on a
function <literal>f</literal> has the following behaviour:
INLINE says "please inline me", but INLINABLE says "feel free to inline me; use your
While <literal>INLINE</literal> says "please inline me", the <literal>INLINABLE</literal>
says "feel free to inline me; use your
discretion". In other words the choice is left to GHC, which uses the same
rules as for pragma-free functions. Unlike INLINE, That decision is made at
rules as for pragma-free functions. Unlike <literal>INLINE</literal>, that decision is made at
the <emphasis>call site</emphasis>, and
will therefore be affected by the inlining threshold, optimisation level etc.
Like INLINE, the INLINABLE pragma retains a copy of the original RHS for
Like <literal>INLINE</literal>, the <literal>INLINABLE</literal> pragma retains a
copy of the original RHS for
inlining purposes, and persists it in the interface file, regardless of
the size of the RHS.
If you use the special function <literal>inline</literal> (<xref linkend="special-ids"/>)
to force inlining at a
call site, you will get a copy of the the original RHS.
Indeed, if you intend to use <literal>inline f</literal> it
is a good idea to mark the definition of <literal>f</literal> INLINABLE,
One way to use <literal>INLINABLE</literal> is in conjunction with
the special function <literal>inline</literal> (<xref linkend="special-ids"/>).
The call <literal>inline f</literal> tries very hard to inline <literal>f</literal>.
To make sure that <literal>f</literal> can be inlined,
it is a good idea to mark the definition
of <literal>f</literal> as <literal>INLINABLE</literal>,
so that GHC guarantees to expose an unfolding regardless of how big it is.
Moreover, by annotating <literal>f</literal> as <literal>INLINABLE</literal>,
you ensure that <literal>f</literal>'s original RHS is inlined, rather than
whatever random optimised version of <literal>f</literal> GHC's optimiser
has produced.
The <literal>INLINABLE</literal> pragma also works with <literal>SPECIALISE</literal>:
if you mark function <literal>f</literal> as <literal>INLINABLE</literal>, then
you can subsequently <literal>SPECIALISE</literal> in another module
(see <xref linkend="specialize-pragma"/>).</para></listitem>
Unlike <literal>INLINE</literal>, it is OK to use
an <literal>INLINABLE</literal> pragma on a recursive function.
The principal reason do to so to allow later use of <literal>SPECIALISE</literal>
......@@ -7995,6 +8016,9 @@ RULE with a somewhat-complex left-hand side (try it yourself), so it might not f
well. If you use this kind of specialisation, let us know how well it works.
<sect3 id="specialize-inline">
<title>SPECIALIZE INLINE</title>
<para>A <literal>SPECIALIZE</literal> pragma can optionally be followed with a
<literal>INLINE</literal> or <literal>NOINLINE</literal> pragma, optionally
followed by a phase, as described in <xref linkend="inline-noinline-pragma"/>.
......@@ -8023,6 +8047,66 @@ specialisation, whose body is also inlined. The result is a type-based
unrolling of the indexing function.</para>
<para>Warning: you can make GHC diverge by using <literal>SPECIALISE INLINE</literal>
on an ordinarily-recursive function.</para>
<sect3><title>SPECIALIZE for imported functions</title>
Generally, you can only give a <literal>SPECIALIZE</literal> pragma
for a function defined in the same module.
However if a function <literal>f</literal> is given an <literal>INLINABLE</literal>
pragma at its definition site, then it can subequently be specialised by
importing modules (see <xref linkend="inlinable-pragma"/>).
For example
module Map( lookup, blah blah ) where
lookup :: Ord key => [(key,a)] -> key -> Maybe a
lookup = ...
{-# INLINABLE lookup #-}
module Client where
import Map( lookup )
data T = T1 | T2 deriving( Eq, Ord )
{-# SPECIALISE lookup :: [(T,a)] -> T -> Maybe a
Here, <literal>lookup</literal> is declared <literal>INLINABLE</literal>, but
it cannot be specialised for type <literal>T</literal> at its definition site,
because that type does not exist yet. Instead a client module can define <literal>T</literal>
and then specialise <literal>lookup</literal> at that type.
Moreover, every module that imports <literal>Client</literal> (or imports a module
that imports <literal>Client</literal>, transitively) will "see", and make use of,
the specialised version of <literal>lookup</literal>. You don't need to put
a <literal>SPECIALIZE</literal> pragma in every module.
Moreover you often don't even need the <literal>SPECIALIZE</literal> pragma in the
first place. When compiling a module M,
GHC's optimiser (with -O) automatically considers each top-level
overloaded function declared in M, and specialises it
for the different types at which it is called in M. The optimiser
<emphasis>also</emphasis> considers each <emphasis>imported</emphasis>
<literal>INLINABLE</literal> overloaded function, and specialises it
for the different types at which it is called in M.
So in our example, it would be enough for <literal>lookup</literal> to
be called at type <literal>T</literal>:
module Client where
import Map( lookup )
data T = T1 | T2 deriving( Eq, Ord )
findT1 :: [(T,a)] -> Maybe a
findT1 m = lookup m T1 -- A call of lookup at type T
However, sometimes there are no such calls, in which case the
pragma can be useful.
<sect3><title>Obselete SPECIALIZE syntax</title>
<para>Note: In earlier versions of GHC, it was possible to provide your own
specialised function for a given type:
......@@ -8033,6 +8117,7 @@ on an ordinarily-recursive function.</para>
This feature has been removed, as it is now subsumed by the
<literal>RULES</literal> pragma (see <xref linkend="rule-spec"/>).</para>
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment