On 17/05/2020 17:17, Freddie Chopin wrote: > Hello! > > I have some functions which accept a `std::chrono::duration` or > `std::chrono::time_point` arguments, both based on an 64-bit integers > `long long`. I'm compiling my code for ARM Cortex-M4 microcontroler, > which has "fpv4-sp-d16" FPU. On all GCCs up to (and including) GCC 9 > such arguments were passed in core registers or via stack. In GCC 10 > sometimes they are passed via d7 VFP register, sometimes via stack, > possibly sometimes via core registers (did not check that yet [; ). > Only d7? No, that couldn't be right. d7 would only be used if d0-d6 had also been used. > I suppose this change is related to this entry from > https://gcc.gnu.org/gcc-10/changes.html : > > "The handling of 64-bit integer operations has been significantly > reworked and improved leading to improved performance and reduced stack > usage when using 64-bit integral data types. The option -mneon-for- > 64bits is now deprecated and will be removed in a future release." No, those changes are for handling of 64-bit integral values where we no-longer use Neon to perform those options and have improved the way code is generated to handle them using the GP registers. > > If that's the case, then is there any way I could control this > behaviour and disable it (obviously other than disabling FPU completely > (; )? Maybe this change is related to something else instead (maybe > "The ABI of passing and returning certain C++ classes by value changed > on several targets in GCC 10, including AArch64, ARM, PowerPC ELFv2, > S/390 and Itanium.")? > > I have a very specific use case for that - for my bare-metal RTOS I > have a test application that verifies correct behaviour of my code for > context stacking and restoring in situation where threads use/don't-use > FPU, when threads that use/don't-use FPU are interrupted by chip > interrupts using/not-using FPU and so on. My problem with the new > behaviour in my very specific tests is that I can no longer control > whether the thread uses the FPU or not, which severely limits possible > combinations of test cases. To be more precse - I can no longer write a > thread that doesn't use FPU, even if I never use any floating point > operations explicitly. That's why I'm wondering whether there's a way > to disable this only for my test application. > > Thanks in advance! > > Regards, > FCh > Testcase needed. R.
