fixup

Author: harendra-kumar

docs/intro.md

@@ -35,7 +35,14 @@ TBD: document the exact limitations and differences.
 
 ## threadCPUTime# prim op
 
-Available in the 9.2.8 RTS patch.
+Available in the
+[GHC 9.2.8 RTS patch](https://github.com/composewell/ghc/releases/tag/ghc-9.2.8-perf-counters-1-rc1).
+
+Install the patched GHC using:
+
+```
+ghcup install -u https://github.com/composewell/ghc/releases/download/ghc-9.2.8-perf-counters-1-rc1/ghc-9.2.8.20231130-x86_64-unknown-linux.tar.xz ghc
+```
 
 This is a very simple and easy to use mechanism. The RTS is modified
 such that we record the accurate time and allocation information in a
@@ -49,11 +56,14 @@ and B in a program, diff will tell us the time spent and allocations
 between the two points.
 
 We have to ensure that we are diffing the data for the same thread id at
-both the points. See examples directory for a working example.
-
-The API has some measurement overhead but it is not very high.  If we are
-nesting measurements be aware that outer measurement will measure the
-measurement overhead of the inner one.
+both the points. See [this example program](./threadCPUTime.hs).
+
+The API has some measurement overhead but it is not very high.  If we
+are nesting measurements be aware that outer measurement will measure
+the measurement overhead of the inner one. If you are measuring a
+relatively small amount of time then reduce the overhead (approx 2
+microseconds and 300 byte allocations, measure the exact value using an
+empty code block).
 
 This is very useful in micro-measurements and analysis of the CPU cost
 different segments of code in a particular Haskell thread without worrying

docs/threadCPUTime.hs

@@ -30,4 +30,4 @@ main = do
     (nsec2, bytes2, slices2) <- stat
     putStrLn $ "cpu time nsec: " ++ show (nsec2 - nsec1)
     putStrLn $ "allocated bytes: " ++ show (bytes2 - bytes1)
-    putStrLn $ "sched count: " ++ show (slices2 - slices1)
+    putStrLn $ "sched-out count: " ++ show (slices2 - slices1)