tcmalloc using a "GangBoard" mechanism which need to change some opcodes at the begining
of the hook function.

But at least on 32bit and 64bit windows, there is a more elegant and effective hook
mechanism: we can change the module's (exe or dll) IAT (Import address table) entry
directly to implement the hook. 

This method doesn't need the disassembler nor changing any opcode, Just rewrite the
function pointer of corresponding IAT entries. 

And because the hooked function will be invocked directly, there is no need to do a
"GangBoard" jump for each call, so it will gains more efficient.