user12155310 wrote:Thanks for fast answer :)
In Oracle 11g you have a Java 1.6 JVM and you can load and run your demo on the database JVM.
Kayaman wrote:From memory the slowest part of that algorithm is the SHA1 which is used for each iteration in the iteration count. Maybe using inline assembler for just the SHA1 will be all that is needed to make a significant difference. We will have to see but I doubt it will make more than a factor of 2 difference.
I would imagine the C implementation uses inline assembler code to use the most optimized opcodes available. I'm not sure if JIT is able to recognize complex enough structures to compile things into for example SSE3 instructions.