The many arrays of GHC
Sometimes when profiling Haskell code you will find yourself wondering how some large array will manifest itself in the cost-center profiler (namely in the “closure description” heap profile variety, +RTS -hd). Here is a brief summary.
| Type | What appears in -hd profile? |
Size |
|---|---|---|
| Primitive arrays | ||
GHC.Prim.ArrayArray# |
ARR_PTRS |
\(n\) words |
GHC.Prim.MutableArrayArray# |
MUT_ARR_PTRS_* |
\(n\) words |
GHC.Prim.SmallArray# |
SMALL_ARR_PTRS_* |
\(n\) words |
GHC.Prim.SmallMutableArray# |
SMALL_MUT_ARR_PTRS_* |
\(n\) words |
GHC.Prim.ByteArray# |
ARR_WORDS |
\(n\) bytes |
GHC.Prim.MutableByteArray# |
MUT_ARR_WORDS_* |
\(n\) bytes |
GHC.Prim.Array# |
ARR_PTRS |
\(n\) words |
GHC.Prim.MutableArray# |
MUT_ARR_PTRS_* |
\(n\) words |
array package |
||
Data.Array |
ARR_WORDS |
|
*bytestring package |
||
Data.ByteString.ByteString |
ARR_WORDS |
\(n\) bytes |
Data.ByteString.Short.ShortByteString |
ARR_WORDS |
\(n\) bytes |
vector package |
||
Data.Vector.Vector |
ARR_PTRS |
\(s * n\) bytes + \(n\) words |
Data.Vector.Vector.Mutable.MVector |
ARR_PTRS |
\(s * n\) bytes + \(n\) words |
Data.Vector.Storable.Vector |
ARR_WORDS |
\(s * n\) bytes |
Data.Vector.Storable.Vector.Mutable.MVector |
ARR_WORDS |
\(s * n\) bytes |
Data.Vector.Unboxed.Vector |
ARR_WORDS |
\(s * n\) bytes |
Data.Vector.Unboxed.Vector.Mutable.MVector |
ARR_WORDS |
\(s * n\) bytes |
text package |
||
Data.Text.Text |
ARR_WORDS |
\(\approx 2*n\) bytes |
In the above \(n\) refers to the length of the array, \(s\) refers to the element size in bytes (where appropriate).
Of course, it would be great if the profiler could give us more helpful output. This issue has been tracked as GHC #7275 for several years now. Maybe you could be the one to fix it!