On Fri, May 23, 2025 at 10:12 AM Richard Biener <richard.guent...@gmail.com> wrote:
> > > > Am 23.05.2025 um 17:06 schrieb Joel Sherrill via Gcc <gcc@gcc.gnu.org>: > > > > Hi > > > > In the SPARC port of RTEMS, there is a global variable assigned to a > > register for performance reasons. This is the near decade old line of > code: > > > > register struct Per_CPU_Control *_SPARC_Per_CPU_current __asm__( "g6" ); > > > > Per https://gcc.gnu.org/onlinedocs/gcc/Global-Register-Variables.html, > this > > is still the correct way to accomplish this. But -pedantic and specifying > > C17 result in this warning: > > > > ../../../cpukit/score/cpu/sparc/include/rtems/score/cpuimpl.h:230:17: > > warning: file-scope declaration of '_SPARC_Per_CPU_current' specifies > > 'register' [-Wpedantic] > > 230 | register struct Per_CPU_Control *_SPARC_Per_CPU_current __asm__( > > "g6" ); > > | > > > > We do not want to move this pointer out of a register and the GCC docs > > section cited above doesn't mention anything about the register keyword > > being moved to reserved keyword in C17 and C++17 with undefined meaning. > > The example in the GCC manual is similar to our use. > > > > Is there more current guidance on having a global register variable? > Other > > than disabling the pedantic warning around that one line, how should the > > warning be addressed? > > Does it work by writing __extension__ before it? This is a GNU extension, > so complaining with -pedantic when compiling with -std=cXY as opposed to > -std=gnuXY is what I would expect. > Adding __extension__ did work. Thanks! Any idea if that's accepted by clang? Do you think the GCC documentation on this should include this? --Joel Sherrill > > Richard > > > > > Thanks. > > > > --Joel Sherrill >
