Add a perf test for the generics code pattern from #21839.wip/andreask/perf-test-binary

This code showed a strong shift between compile time (got worse) and
run time (got a lot better) recently which is perfectly acceptable.

However it wasn't clear why the compile time regression was happening
initially so I'm adding this test to make it easier to track such changes
in the future.

6 files changed, 109 insertions, 0 deletions

diff --git a/testsuite/tests/perf/compiler/T21839c.hs b/testsuite/tests/perf/compiler/T21839c.hs
new file mode 100644
index 0000000000..594459da03
--- /dev/null
+++ b/testsuite/tests/perf/compiler/T21839c.hs
@@ -0,0 +1,46 @@
+-- For in depth details see the ticket #21839. The short version:
+
+-- We noticed that GHC got slower compiling Cabal the libary.
+-- Eventually I narrowed it down to the pattern below of deriving Generics
+-- for a Enum, and then deriving a Binary instance for that Enum via Generics.
+-- A pattern very frequently used in Cabal.
+-- However this turned out to be a classic compile vs runtime tradeoff.
+-- In benchmarks I found the resulting code for the Binary instance was running
+-- more than twice as fast!
+-- So we decided to merely document this change and add a test representing this behaviour
+-- rather than trying to coax ghc back into its old behaviour.
+
+{-# LANGUAGE DeriveGeneric #-}
+
+{-# OPTIONS_GHC #-}
+module Main
+  ( main
+  ) where
+
+import GHC.Generics
+import Data.Typeable
+import Data.Binary
+import qualified Data.ByteString.Lazy as BL
+import qualified Data.ByteString as BS
+
+data PathTemplateVariable =
+
+       Var0
+     | Var1
+     | Var2
+     | Var3
+     | Var4
+     | Var5
+     | Var6
+     | Var7
+     | Var8
+     | Var9
+  deriving (Generic,Enum)
+
+instance Binary PathTemplateVariable
+
+main :: IO ()
+main = do
+  let lists = replicate 10000 Var0
+      lbs = encode lists
+  print $ BS.length $ BS.toStrict lbs
diff --git a/testsuite/tests/perf/compiler/T21839c.stdout b/testsuite/tests/perf/compiler/T21839c.stdout
new file mode 100644
index 0000000000..b95ab72342
--- /dev/null
+++ b/testsuite/tests/perf/compiler/T21839c.stdout
@@ -0,0 +1 @@
+10008
diff --git a/testsuite/tests/perf/compiler/all.T b/testsuite/tests/perf/compiler/all.T
index cac2ebd48f..a70b93df01 100644
--- a/testsuite/tests/perf/compiler/all.T
+++ b/testsuite/tests/perf/compiler/all.T
@@ -643,3 +643,10 @@ test ('T20261',
       [collect_compiler_stats('all')],
       compile,
       [''])
+
+# Track perf of generics based binary instances
+test('T21839c',
+    [   collect_compiler_stats('all', 1),
+        only_ways(['normal'])],
+    compile,
+    ['-O'])
\ No newline at end of file
diff --git a/testsuite/tests/perf/should_run/T21839r.hs b/testsuite/tests/perf/should_run/T21839r.hs
new file mode 100644
index 0000000000..594459da03
--- /dev/null
+++ b/testsuite/tests/perf/should_run/T21839r.hs
@@ -0,0 +1,46 @@
+-- For in depth details see the ticket #21839. The short version:
+
+-- We noticed that GHC got slower compiling Cabal the libary.
+-- Eventually I narrowed it down to the pattern below of deriving Generics
+-- for a Enum, and then deriving a Binary instance for that Enum via Generics.
+-- A pattern very frequently used in Cabal.
+-- However this turned out to be a classic compile vs runtime tradeoff.
+-- In benchmarks I found the resulting code for the Binary instance was running
+-- more than twice as fast!
+-- So we decided to merely document this change and add a test representing this behaviour
+-- rather than trying to coax ghc back into its old behaviour.
+
+{-# LANGUAGE DeriveGeneric #-}
+
+{-# OPTIONS_GHC #-}
+module Main
+  ( main
+  ) where
+
+import GHC.Generics
+import Data.Typeable
+import Data.Binary
+import qualified Data.ByteString.Lazy as BL
+import qualified Data.ByteString as BS
+
+data PathTemplateVariable =
+
+       Var0
+     | Var1
+     | Var2
+     | Var3
+     | Var4
+     | Var5
+     | Var6
+     | Var7
+     | Var8
+     | Var9
+  deriving (Generic,Enum)
+
+instance Binary PathTemplateVariable
+
+main :: IO ()
+main = do
+  let lists = replicate 10000 Var0
+      lbs = encode lists
+  print $ BS.length $ BS.toStrict lbs
diff --git a/testsuite/tests/perf/should_run/T21839r.stdout b/testsuite/tests/perf/should_run/T21839r.stdout
new file mode 100644
index 0000000000..b95ab72342
--- /dev/null
+++ b/testsuite/tests/perf/should_run/T21839r.stdout
@@ -0,0 +1 @@
+10008
diff --git a/testsuite/tests/perf/should_run/all.T b/testsuite/tests/perf/should_run/all.T
index eb5124a155..9059a6f92f 100644
--- a/testsuite/tests/perf/should_run/all.T
+++ b/testsuite/tests/perf/should_run/all.T
@@ -395,3 +395,11 @@ test('T19347',
     compile_and_run,
     ['-O'])
 
+# Track perf of generics based binary instances
+test('T21839r',
+    [   collect_stats('bytes allocated', 10),
+        collect_runtime_residency(10),
+        collect_compiler_stats('bytes allocated', 1),
+        only_ways(['normal'])],
+    compile_and_run,
+    ['-O'])