The copyResult function is called when there is a result mismatch between the checker and the cpu for a load instruction (in most of the case, this mismatch is due to a store forwarding). This copyResult function calls the setFloatRegBits function that writes to the floating point registers of the checker. As the used index starts at theISA::FP_Base_DepTag, the assert assert(flatIndex < TheISA::NumFloatRegs) breaks. This patch substract the theISA::FP_Base_DepTag value to the index so that the assert does not break.

It seems to correct the bug I had with the checker. However, I have not done extensive tests.