Change `-fprof-late` to insert cost centres after unfolding creation.

The former behaviour of adding cost centres after optimization but
before unfoldings are created is not available via the flag
`prof-late-inline` instead.

I also reduced the overhead of -fprof-late* by pushing the cost centres
into lambdas. This means the cost centres will only account for
execution of functions and not their partial application.

Further I made LATE_CC cost centres it's own CC flavour so they now
won't clash with user defined ones if a user uses the same string for
a custom scc.

LateCC: Don't put cost centres inside constructor workers.

With -fprof-late they are rarely useful as the worker is usually
inlined. Even if the worker is not inlined or we use -fprof-late-linline
they are generally not helpful but bloat compile and run time
significantly. So we just don't add sccs inside constructor workers.

-------------------------
Metric Decrease:
    T13701
-------------------------

2 files changed, 39 insertions, 5 deletions

diff --git a/docs/users_guide/debugging.rst b/docs/users_guide/debugging.rst
index 0c09c4c3ec..87a689c268 100644
--- a/docs/users_guide/debugging.rst
+++ b/docs/users_guide/debugging.rst
@@ -446,6 +446,12 @@ subexpression elimination pass.
 
     Dump output of Core preparation pass
 
+.. ghc-flag:: -ddump-late-cc
+    :shortdesc: Dump core with late cost centres added
+    :type: dynamic
+
+    Dump output of LateCC pass after cost centres have been added.
+
 .. ghc-flag:: -ddump-view-pattern-commoning
     :shortdesc: Dump commoned view patterns
     :type: dynamic
diff --git a/docs/users_guide/profiling.rst b/docs/users_guide/profiling.rst
index 418c9b0bb0..1c2f458f10 100644
--- a/docs/users_guide/profiling.rst
+++ b/docs/users_guide/profiling.rst
@@ -439,19 +439,47 @@ compiled program.
     details.
 
 .. ghc-flag:: -fprof-late
-    :shortdesc: Auto-add ``SCC``\\ s to all top level bindings *after* the optimizer has run.
+    :shortdesc: Auto-add ``SCC``\\ s to all top level bindings *after* the core pipeline has run.
     :type: dynamic
     :reverse: -fno-prof-late
     :category:
 
     :since: 9.4.1
 
-    Adds an automatic ``SCC`` annotation to all top level bindings late in the core pipeline after
-    the optimizer has run. This means these cost centres will not interfere with core-level optimizations
+    Adds an automatic ``SCC`` annotation to all top level bindings late in the compilation pipeline after
+    the optimizer has run and unfoldings have been created. This means these cost centres will not interfere with core-level optimizations
     and the resulting profile will be closer to the performance profile of an optimized non-profiled
     executable.
-    While the results of this are generally very informative some of the compiler internal names
-    will leak into the profile.
+    While the results of this are generally informative, some of the compiler internal names
+    will leak into the profile. Further if a function is inlined into a use site it's costs will be counted against the
+    caller's cost center.
+
+    For example if we have this code:
+
+    .. code-block:: haskell
+
+        {-# INLINE mysum #-}
+        mysum = sum
+        main = print $ mysum [1..9999999]
+
+    Then ``mysum`` will not show up in the profile since it will be inlined into main and therefore
+    it's associated costs will be attributed to mains implicit cost centre.
+
+.. ghc-flag:: -fprof-late-inline
+    :shortdesc: Auto-add ``SCC``\\ s to all top level bindings *after* the optimizer has run and retain them when inlining.
+    :type: dynamic
+    :reverse: -fno-prof-late-inline
+    :category:
+
+    :since: 9.4.1
+
+    Adds an automatic ``SCC`` annotation to all top level bindings late in the core pipeline after
+    the optimizer has run. This is the same as :ghc-flag:`-fprof-late` except that cost centers are included in some unfoldings.
+
+    The result of which is that cost centers *can* inhibit core optimizations to some degree at use sites
+    after inlining. Further there can be significant overhead from cost centres added to small functions if they are inlined often.
+
+    You can try this mode if :ghc-flag:`-fprof-late` results in a profile that's too hard to interpret.
 
 .. ghc-flag:: -fprof-cafs
     :shortdesc: Auto-add ``SCC``\\ s to all CAFs