(Compiler) Optimization of fp expressions

1. run paranoia with different optimization flags: what's  Kahan opinion?
   
2. Write simple kernels and inspect the generated object code: identify the various operation at assembler level
3.  compile code with different optimization options and verify the generated code
• use also pragmas and function attributes to specify the optimization at different level of granularity
• find cases when optimization either changes exceptions' behaviour or produces wrong results
  
4. Learn how to avoid common pitfalls
• undesired conversions
• redundant calls
• non inlining
• loss of precision
  
5. Verify the gain in performance when taking the "correct" approach
6. Apply all this to the minimization exemple

Code

Hints

// setenv OPTFLAGS "-fassociative-math -freciprocal-math -fno-math-errno -fno-signed-zeros -fno-trapping-math -ffinite-math-only"
// use extern to avoid inlining (and compile with -fPIC)
// "volatile" can be used to avoid eager optimization
// compile paranoia with -DVOLATILE and/or -DNOSUB 

pfmon  --long-smpl-period=5000 --resolve-addresses --smpl-per-function --smpl-show-top=20 ./a.out
pfmon -e UNHALTED_CORE_CYCLES,ARITH:CYCLES_DIV_BUSY,SSEX_UOPS_RETIRED:SCALAR_SINGLE,SSEX_UOPS_RETIRED:PACKED_SINGLE  ./a.out k


 objdump -S -r -C --no-show-raw-insn -w test.o | less
(on MacOS: otool -t -v  -V -X test.o  | c++filt | less)

References

• Function Attributes section
• Optimize Options section