sry for bringing up this ancient thread but I got a question about this: shouldn't soft-float be 100% safe because it's a software implementations and thus never throw exceptions?
Thanks Michael On Tue, Dec 8, 2015 at 6:28 PM, Shubha Ramani <[email protected]> wrote: > Thanks for your well-thought out and detailed answer !I guess the safe bet > is "No". Stay away from floating point on EDK2. > Shubha Shubha D. [email protected] > [email protected] > > > On Saturday, December 5, 2015 4:24 PM, Scott Duplichan < > [email protected]> wrote: > > > Daryl McDaniel [mailto:[email protected]] wrote: > > ]Sent: Saturday, December 05, 2015 11:00 AM > ]To: 'Shubha Ramani' <[email protected]>; [email protected] > ]Subject: Re: [edk2] Double or floating point on edk2 ? > ] > ]Hello, > ] > ]Doing floating point operations within the UEFI environment > ]is not recommended. There are many reasons, but the main one > ]is that floating point exceptions are unlikely to be handled. > > I think the need to enable and trap floating point exceptions > is rare. For most applications, the normal masked response > works well. Here is Intel's official position: > > "Exception-handling software is often difficult to write, > and the masked responses have been tailored to deliver the > most reasonable result for each condition. For the majority > of applications, masking all exceptions yields satisfactory > results with the least programming effort." > > ]The default FPU mode (rounding, precision, ...) may also not > ]be what you would want. > > I think the default FPU mode meets the requirements of the C > programming language. But if not, it could be changed. > > ]There is also a chance that one or more of the UEFI functions > ]may destroy a floating point or vector register that you are using. > > Is this really true? I think the compiler will preserve the registers > when needed for C code. ASM code could certainly destroy floating > point registers. A problem would occur if SMI or the timer interrupt > code were to use floating point registers. Could it be possible > for timer callback code to use floating point and destroy registers > of the interrupted application? > > ]That said, there is nothing that would prevent you from using > ]floating point types and operations in your Shell application. > ]If you do, it is your responsibility to ensure that the environment > ]is properly set up for your needs. The UEFI specification details > ]what registers are preserved, or not. > ] > ]Using StdLib does not alleviate this responsibility. The only thing > ]StdLib does is ensure that the correct rounding, floor, and NAN modes > ]are set in the FPU in order to conform to the ISO/IEC 9899 Addendum-1 > ](C95) specification. The sole purpose of StdLib's floating point > ]support is to facilitate porting existing C applications to the UEFI > ]environment. > ] > ]Unfortunately, using floating point is entirely at your own risk. > > The good thing is the problem with unresolved external symbol __dtoui3 > is gone with the current EDK2. This problem used to occur when building > AppPkg for IA32 using VS2013 or newer. It no longer occurs because of > "BaseTools: Add /arch:IA32 option in VS2012 and VS2013". Though that > change was made to prevent the compiler from generating the cmov > instruction, it also has the side effect of turning off SSE2 math. > > Thanks, > Scott > > > ]Sincerely, > ]Daryl McDaniel > > > > -----Original Message----- > > From: edk2-devel [mailto:[email protected]] On Behalf Of > Shubha Ramani > > Sent: Friday, December 04, 2015 4:39 PM > > To: [email protected] > > Subject: [edk2] Double or floating point on edk2 ? > > > > I know this question has been answered before. Do I need to convert my > application to a UEFI StdLib applicationto do double or > > floating point math ? Can't do it in a regular Shell App ? > > Please illuminate. > > Thanks, > > Shubha Shubha D. [email protected] > > [email protected] > > _______________________________________________ > > edk2-devel mailing list > > [email protected] > > https://lists.01.org/mailman/listinfo/edk2-devel > > _______________________________________________ > edk2-devel mailing list > [email protected] > https://lists.01.org/mailman/listinfo/edk2-devel > > > > _______________________________________________ > edk2-devel mailing list > [email protected] > https://lists.01.org/mailman/listinfo/edk2-devel > _______________________________________________ edk2-devel mailing list [email protected] https://lists.01.org/mailman/listinfo/edk2-devel
