[libc++][Apple] Use CLOCK_MONOTONIC_RAW instead of CLOCK_UPTIME_RAW for steady_clock

[ldionne]

Summary

In D27429, we switched the Apple implementation of steady_clock::now()
from clock_gettime(CLOCK_MONOTONIC) to clock_gettime(CLOCK_UPTIME_RAW).
The purpose was to get nanosecond precision, and also to improve the
performance of the implementation.

However, it appears that CLOCK_UPTIME_RAW does not satisfy the requirements
of the Standard, since it is not strictly speaking monotonic. Indeed, the
clock does not increment while the system is asleep, which had been
mentioned in D27429 but somehow not addressed.

This patch switches to CLOCK_MONOTONIC_RAW, which is monotonic, increased
during sleep, and also has nanosecond precision.

https://llvm.org/PR44773

[ldionne]

Specifically, I'm looking to understand why Eric's comment at that time (https://reviews.llvm.org/D27429#618296) about using CLOCK_MONOTONIC_RAW wasn't acted upon -- did we simply overlook it or is there a reason why CLOCK_MONOTONIC_RAW is not a suitable implementation?

[dexonsmith]

In D74341#1867559, @ldionne wrote:

Specifically, I'm looking to understand why Eric's comment at that time (https://reviews.llvm.org/D27429#618296) about using CLOCK_MONOTONIC_RAW wasn't acted upon -- did we simply overlook it or is there a reason why CLOCK_MONOTONIC_RAW is not a suitable implementation?

Looking at the discussion I think Howard had concerns with CLOCK_MONOTONIC. They probably didn't apply to CLOCK_MONOTONIC_RAW, but we didn't test that.

One benefit of using CLOCK_UPTIME_RAW is that it matches libc++'s historical behaviour on Apple platforms, since it exactly matches mach_absolute_time(). The main purpose of the above commit was to fix a binary compatibility issue caused by moving away from mach_absolute_time() and this was the conservative fix.

[ldionne]

In D74341#1867605, @dexonsmith wrote:

In D74341#1867559, @ldionne wrote:

Specifically, I'm looking to understand why Eric's comment at that time (https://reviews.llvm.org/D27429#618296) about using CLOCK_MONOTONIC_RAW wasn't acted upon -- did we simply overlook it or is there a reason why CLOCK_MONOTONIC_RAW is not a suitable implementation?

Looking at the discussion I think Howard had concerns with CLOCK_MONOTONIC. They probably didn't apply to CLOCK_MONOTONIC_RAW, but we didn't test that.

One benefit of using CLOCK_UPTIME_RAW is that it matches libc++'s historical behaviour on Apple platforms, since it exactly matches mach_absolute_time(). The main purpose of the above commit was to fix a binary compatibility issue caused by moving away from mach_absolute_time() and this was the conservative fix.

I still do think that CLOCK_MONOTONIC_RAW is the correct behavior w.r.t. the Standard. I'll go ahead and check that it doesn't result in the bincompat issue we were seeing with CLOCK_MONOTONIC.

[EricWF]

LGTM.

I'm not sure why CLOCK_MONOTONIC_RAW wasn't used originally, but that appears to be the correct approach.

[EricWF]

Can we assume that all supported Apple platforms have clock_gettime now? (It was added in 10.12)

If so, can we remove the fallback?

[ldionne]

In D74341#1872286, @EricWF wrote:

Can we assume that all supported Apple platforms have clock_gettime now? (It was added in 10.12)

If so, can we remove the fallback?

I think it would be fine to remove the fallback, as the only thing that'll do is require macOS >= 10.12 in order to build the libc++ dylib. Since pretty much only vendors build the dylib, I think that's a reasonable thing to do. Will do in a followup patch.

libcxx/src/chrono.cpp

 // Warning:  If this is not truly steady, then it is non-conforming.  It is
 //  better for it to not exist and have the rest of libc++ use system_clock
 //  instead.
 
 const bool steady_clock::is_steady;
 
 #if defined(__APPLE__)
 
-// Darwin libc versions >= 1133 provide ns precision via CLOCK_UPTIME_RAW
-#if defined(_LIBCPP_USE_CLOCK_GETTIME) && defined(CLOCK_UPTIME_RAW)
+// Darwin libc versions >= 1133 provide ns precision via CLOCK_MONOTONIC_RAW
+#if defined(_LIBCPP_USE_CLOCK_GETTIME) && defined(CLOCK_MONOTONIC_RAW)
 steady_clock::time_point
 steady_clock::now() _NOEXCEPT
 {
     struct timespec tp;
-    if (0 != clock_gettime(CLOCK_UPTIME_RAW, &tp))
-        __throw_system_error(errno, "clock_gettime(CLOCK_UPTIME_RAW) failed");
+    if (0 != clock_gettime(CLOCK_MONOTONIC_RAW, &tp))
+        __throw_system_error(errno, "clock_gettime(CLOCK_MONOTONIC_RAW) failed");
     return time_point(seconds(tp.tv_sec) + nanoseconds(tp.tv_nsec));
 }
 
 #else
 //   mach_absolute_time() * MachInfo.numer / MachInfo.denom is the number of
 //   nanoseconds since the computer booted up.  MachInfo.numer and MachInfo.denom
 //   are run time constants supplied by the OS.  This clock has no relationship
 //   to the Gregorian calendar.  It's main use is as a high resolution timer.

 }
 
 steady_clock::time_point
 steady_clock::now() _NOEXCEPT
 {
     static FP fp = init_steady_clock();
     return time_point(duration(fp()));
 }
-#endif // defined(_LIBCPP_USE_CLOCK_GETTIME) && defined(CLOCK_UPTIME_RAW)
+#endif // defined(_LIBCPP_USE_CLOCK_GETTIME) && defined(CLOCK_MONOTONIC_RAW)
 
 #elif defined(_LIBCPP_WIN32API)
 
 // https://msdn.microsoft.com/en-us/library/windows/desktop/ms644905(v=vs.85).aspx says:
 //    If the function fails, the return value is zero. <snip>
 //    On systems that run Windows XP or later, the function will always succeed
 //      and will thus never return zero.
 
steady_clock::now() _NOEXCEPT
 
   LARGE_INTEGER counter;
   (void) QueryPerformanceCounter(&counter);
   return time_point(duration(counter.QuadPart * nano::den / freq.QuadPart));
 }
 
 #elif defined(CLOCK_MONOTONIC)
 
-// On Apple platforms only CLOCK_UPTIME_RAW or mach_absolute_time are able to
-// time functions in the nanosecond range. Thus, they are the only acceptable
-// implementations of steady_clock.
+// On Apple platforms only CLOCK_UPTIME_RAW, CLOCK_MONOTONIC_RAW or
+// mach_absolute_time are able to time functions in the nanosecond range.
+// Thus, they are the only acceptable implementations of steady_clock.
 #ifdef __APPLE__
 #error "Never use CLOCK_MONOTONIC for steady_clock::now on Apple platforms"
 #endif
 
 steady_clock::time_point
 steady_clock::now() _NOEXCEPT
 {
     struct timespec tp;