Hello!
It seems that I've found a bug in Pelles C compiler.
The attached computer program gives different results at different compiler settings. The difference is very far beyond rounding errors and the code doesn't include precision-sensitive things such as numeric differentiation. It very intensively uses floating point arithmetics.
The next compilation keys (enabled speed optimization for x64 platform with SSE) cause problems
pocc x64_glitch.c /Tx64-coff /Ot /arch:SSE2
polink x64_glitch.obj /OUT:x64_glitch.exe
The next variant (without optimization) causes no problems.
pocc x64_glitch.c /Tx64-coff
polink x64_glitch.obj /OUT:x64_noglitch.exe
The correctness/incorrectness was checked using other compilers (MinGW, MSVC, Open Watcom) and scientific literature (this piece of code reproduces a fragment of water density equation).