On 1/18/21 12:34 PM, Martin Storsjö wrote: > On Mon, 18 Jan 2021, Zebediah Figura (she/her) wrote: > >> On 1/17/21 2:05 PM, Martin Storsjö wrote: >>> Signed-off-by: Martin Storsjö <[email protected]> >>> --- >>> Added memory clobbers for MemoryBarrier(). >>> --- >>> mingw-w64-headers/crt/_mingw.h.in | 8 ++++++++ >>> mingw-w64-headers/include/winnt.h | 6 ++++++ >>> 2 files changed, 14 insertions(+) >>> >>> diff --git a/mingw-w64-headers/crt/_mingw.h.in >>> b/mingw-w64-headers/crt/_mingw.h.in >>> index 01eff66fa..25513d7da 100644 >>> --- a/mingw-w64-headers/crt/_mingw.h.in >>> +++ b/mingw-w64-headers/crt/_mingw.h.in >>> @@ -580,7 +580,15 @@ extern "C" { >>> void __cdecl __debugbreak(void); >>> __MINGW_INTRIN_INLINE void __cdecl __debugbreak(void) >>> { >>> +#if defined(__i386__) || defined(__x86_64__) >>> __asm__ __volatile__("int {$}3":); >>> +#elif defined(__arm__) >>> + __asm__ __volatile__("udf #1"); >>> +#elif defined(__aarch64__) >>> + __asm__ __volatile__("brk #0xf000"); >>> +#else >>> + __asm__ __volatile__("unimplemented"); >>> +#endif >>> } >>> #endif >>> #endif >>> diff --git a/mingw-w64-headers/include/winnt.h >>> b/mingw-w64-headers/include/winnt.h >>> index 0decf495a..558423745 100644 >>> --- a/mingw-w64-headers/include/winnt.h >>> +++ b/mingw-w64-headers/include/winnt.h >>> @@ -1846,6 +1846,9 @@ extern "C" { >>> #define InterlockedCompareExchangePointerAcquire >>> _InterlockedCompareExchangePointer >>> #define InterlockedCompareExchangePointerRelease >>> _InterlockedCompareExchangePointer >>> >>> +#define YieldProcessor() __asm__ __volatile__("dmb ishst\n\tyield") >>> +#define MemoryBarrier() __asm__ __volatile__("dmb":::"memory") >>> + >>> #ifdef __cplusplus >>> } >>> #endif >>> @@ -2055,6 +2058,9 @@ extern "C" { >>> #define InterlockedCompareExchangePointerAcquire >>> _InterlockedCompareExchangePointer >>> #define InterlockedCompareExchangePointerRelease >>> _InterlockedCompareExchangePointer >>> >>> +#define YieldProcessor() __asm__ __volatile__("dmb ishst\n\tyield") >>> +#define MemoryBarrier() __asm__ __volatile__("dmb sy":::"memory") >>> + >>> #ifdef __cplusplus >>> } >>> #endif >>> >> >> Should YieldProcessor() have a compiler barrier? Unfortunately I can't >> seem to find any reference as to whether it should, and I don't have >> Visual Studio to test with. > > I have Visual Studio - how do you suggest to investigate it?
My best guess is to play around with code and optimization until I can find a sequence that the compiler reorders, and see if YieldProcessor() prevents that. But of course that could be difficult, and may not prove anything for certain. I guess I ask because it's not clear to me whether it should be necessary either. It doesn't seem to be present for x86, but there might be code out there that does something like "while (!var) YieldProcessor();" and expects it to work. I'm assuming the "dmb ishst" would prevent CPU reordering enough for that to work, but it's not clear to me that the compiler wouldn't optimize out the loop somehow... > > // Martin > > _______________________________________________ > Mingw-w64-public mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
OpenPGP_signature
Description: OpenPGP digital signature
_______________________________________________ Mingw-w64-public mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
