[Git][ghc/ghc][wip/doc-unpack] users-guide: Document field coalescence
Ben Gamari (@bgamari)
gitlab at gitlab.haskell.org
Wed Oct 9 14:59:09 UTC 2024
Ben Gamari pushed to branch wip/doc-unpack at Glasgow Haskell Compiler / GHC
Commits:
b380ddad by Ben Gamari at 2024-10-09T10:59:02-04:00
users-guide: Document field coalescence
- - - - -
1 changed file:
- docs/users_guide/exts/pragmas.rst
Changes:
=====================================
docs/users_guide/exts/pragmas.rst
=====================================
@@ -959,6 +959,35 @@ effect of adding ``{-# UNPACK #-}`` to every strict constructor field which is
of a single-constructor data type. Sum types won't be unpacked automatically
by this though, only with the explicit pragma.
+Also note that GHC will coalesce adjacent sub-word size fields into
+words. For instance, consider ::
+
+ data T = T {-# UNPACK #-} !Word32 {-# UNPACK #-} !Word32
+
+As ``Word16`` is represented by the unlifted 32-bit ``Word32#`` type, the ``T``
+constructor will be represent its two ``Word32`` fields using only a single
+64-bit word.
+
+Note that during coalescence padding will be inserted to ensure that each field
+remains naturally aligned. For instance, on a 32-bit platform ::
+
+ data T = T {-# UNPACK #-} !Word16
+ {-# UNPACK #-} !Word8
+ {-# UNPACK #-} !Word16
+
+will require two words since padding is necessary after the ``Word8`` to
+ensure that the subsequent ``Word32`` is naturally aligned:
+
+.. code-block:: none
+
+ ┌───────────────────────────────────┐
+ │ Header │
+ ├─────────────────┬────────┬────────┤
+ │ Word16 │ Word8 │ padding│
+ ├─────────────────┴────────┴────────┤
+ │ Word16 │
+ └───────────────────────────────────┘
+
.. [1]
In fact, :pragma:`UNPACK` has no effect without :ghc-flag:`-O`, for technical
reasons (see :ghc-ticket:`5252`).
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/b380ddadb5e5376326e4128634cbfd72ab51e0dd
--
View it on GitLab: https://gitlab.haskell.org/ghc/ghc/-/commit/b380ddadb5e5376326e4128634cbfd72ab51e0dd
You're receiving this email because of your account on gitlab.haskell.org.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/ghc-commits/attachments/20241009/0fc6a4eb/attachment-0001.html>
More information about the ghc-commits
mailing list