1 files changed, 67 insertions, 0 deletions
diff --git a/README.benchmark.md b/README.benchmark.md
new file mode 100644
index 0000000..808828d
--- /dev/null
+++ b/README.benchmark.md
@@ -0,0 +1,67 @@
+# Benchmark improvement  progress
+
+## The initial benchmark timing based on v1.0-home-cooked-binary
+
+~~~~
+===== Execution time benchmark ======
+     7.5 uS     empty function
+ 10983.6 uS     toBits(4000000)
+   332.0 uS     toNum
+   336.2 uS     freq2regestersValues
+ 96886.3 uS     copySettings
+    37.2 uS     pad2NumBitsInPlace
+  8796.9 uS     bitSlice
+229184.0 uS     FreqToSettings
+~~~~
+
+## Skipping settings clone in FreqToSettings
+
+Skipping the setting cloning via copySettings in FreqToSettings
+shaved about 100 uS out of initial 230 uS. It is ok
+to affect global workplace for settings since it is used
+everywhere anyway.
+
+~~~~
+===== Execution time benchmark ======
+     7.5 uS     empty function
+  2785.8 uS     empty table creation
+  4153.9 uS     table expand by 1 element
+  5536.3 uS     table expand by 2 element
+   146.1 uS     bit.rshift(4000000,1)
+ 10686.6 uS     toBits(4000000)
+   331.9 uS     toNum
+   333.4 uS     freq2regestersValues
+ 94721.6 uS     copySettings
+    37.1 uS     pad2NumBitsInPlace
+  8567.5 uS     bitSlice
+123669.0 uS     FreqToSettings  <-- changed
+~~~~
+
+
+## toBits improvement with bitwise operation failed
+
+Attempt to use bitmask and bitshifts  instead of calculating
+reminder of division by 2 was unsuccessful. It increased
+execution time from 10mS to about 15mS. I am guessing
+that calling a function is longer then do division.
+Binary representation table creation and expansion still
+takes most of the time.
+
+## Skipping table binary representation and working with bits directly
+
+I abandoned a cleaner bits in tables approach, and use binary representation
+of masks and appropriate bits from calculated settings. The code is much
+harder to read but speed up of about 40 times totally worse it.
+
+~~~~
+> dofile("bench_lmx2487.lua")
+===== Execution time benchmark ======
+     8.2 uS     empty function
+  2534.5 uS     empty table creation
+  3768.0 uS     table expand by 1 element
+  5032.4 uS     table expand by 2 element
+   130.8 uS     bit.rshift(4000000,1)
+   336.0 uS     freq2regestersValues
+  2840.7 uS     FreqToSettings <-- changed
+  5047.9 uS     setFreq
+~~~~