Good point, an "all-powerful" FloatingPoint unit isnt exactly the most realistic hardware.
The problem is, right now the code will default to the "all-powerful" ALU. >From a generic standpoint, we are still waiting a X number of cycles for an operation to finish (perhaps pipelined), so you can just instantiate a copy of the same module but only direct FP accesses to it. Or as you state in the next thread, assigning different latencies inside that module could work. On Fri, May 27, 2011 at 9:42 AM, Ali Saidi <[email protected]> wrote: > Normally, the hw that does an integer div or mul is very different the hw > executes the fp version. Float mult/div should probably be handled by a > different unit. > > Ali > > Sent from my ARM powered device > > On May 27, 2011, at 7:09 AM, Maximilien Breughe < > [email protected]> wrote: > > > Hello, > > > > I am using M5's inorder model (in ALPHA_SE mode). > > I found out that the InOrderCPU::createBackEndSked only makes schedules > for the MultDivUnit IF the instruction is an Integer > multiplication/division. It doesn't do this for floating point mul/divs. Is > there a special purpose for this? > > > > Changing it to allow for fp mul/div only involves changing the following > lines: > > in cpu/inorder/cpu.cc in the function InOrderCPU::createBackEndSked: > > ---else if (inst->opClass() == IntMultOp || inst->opClass() == IntDivOp) > { > > X.needs(MDU, MultDivUnit::StartMultDiv); > > > > +++else if (inst->opClass() == IntMultOp || inst->opClass() == IntDivOp > || inst->opClass() == FloatDivOp|| inst->opClass() == FloatMultOp ) { > > X.needs(MDU, MultDivUnit::StartMultDiv); > > > > and similar lines in cpu/inorder/resources/mult_div_unit.cc (in functions > MultDivUnit::getSlot, MultDivUnit::execute and MultDivUnit::exeMulDiv). > > > > > > Kind regards, > > > > Max > > _______________________________________________ > > gem5-users mailing list > > [email protected] > > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > -- - Korey
_______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
