Improve support for Apple Silicon and macOS #275

thorpe-dev · 2022-01-26T18:51:19Z

Will file internal issue - this was done on my personal M1 laptop

Use clock_gettime_nsec_np over mach_absolute_time
Extend support for ARM CPUs to include ARM v8 (which includes Apple Silicon) and ARM v9 (to a lesser extent)
Fix a few places where "CPU != X86" has been considered to be a big-endian processor
use fstat/stat in place of fstat64/stat64 on macOS, as the latter were deprecated in macOS 10.6, and are not available on Apple Silicon
Fix alignment of doubles on arm
Disable JIT support for bdlpcre_regex: there is upstream work to be done (ref sys: disable jit on apple aarch64 BurntSushi/rust-pcre2#16 and others)

Testing performed

Existing unit tests, some of which still fail
bsls_unspecifiedbool.t.cpp fails to build at all

Failing unit tests

The following tests FAILED:
	106 - balm_publicationscheduler.t (Failed)
	117 - balst_stacktraceprintutil.t (Failed)
	160 - balxml_minireader.t (Failed)
	324 - bdldfp_decimalimputil.t (Failed)
	327 - bdldfp_decimalutil.t (Failed)
	359 - bdlma_concurrentmultipool.t (Failed)
	373 - bdlma_multipool.t (Failed)
	395 - bdls_memoryutil.t (Failed)
	399 - bdls_processutil.t (Failed)
	532 - bslim_formatguard.t (Failed)
	635 - bslmf_istransparentpredicate.t (Failed)
	706 - bslmt_threadutil.t (Failed)
	711 - bslmt_timedsemaphore.t (Failed)
	787 - bsls_stopwatch.t (Failed)
	793 - bsls_unspecifiedbool.t (Failed)
	811 - bslstl_chrono.t (Failed)
	822 - bslstl_function.t (Failed)
	824 - bslstl_function_invokerutil.t (Failed)
	881 - bslstl_stringview.t (Failed)

bsls_unspecifiedbool.t.cpp build

/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:418:34: error: type 'BoolType' (aka 'int BloombergLP::bsls::UnspecifiedBool<int>::*') cannot be narrowed to 'bool' in initializer list [-Wc++11-narrowing]
            const bool bar[] = { bt };
                                 ^~
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:418:34: note: insert an explicit cast to silence this issue
            const bool bar[] = { bt };
                                 ^~
                                 static_cast<bool>( )
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:424:23: error: type 'BoolType' (aka 'int BloombergLP::bsls::UnspecifiedBool<int>::*') cannot be narrowed to 'bool' in initializer list [-Wc++11-narrowing]
            } bag = { bt };
                      ^~
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:424:23: note: insert an explicit cast to silence this issue
            } bag = { bt };
                      ^~
                      static_cast<bool>( )
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:621:34: error: type '(anonymous namespace)::Booleable::BoolType' (aka 'int BloombergLP::bsls::UnspecifiedBool<(anonymous namespace)::Booleable>::*') cannot be narrowed to 'bool' in initializer list [-Wc++11-narrowing]
            const bool bar[] = { babel };
                                 ^~~~~
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:621:34: note: insert an explicit cast to silence this issue
            const bool bar[] = { babel };
                                 ^~~~~
                                 static_cast<bool>( )
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:627:23: error: type '(anonymous namespace)::Booleable::BoolType' (aka 'int BloombergLP::bsls::UnspecifiedBool<(anonymous namespace)::Booleable>::*') cannot be narrowed to 'bool' in initializer list [-Wc++11-narrowing]
            } bag = { babel };
                      ^~~~~
/Users/michaelthorpe/programming/bde/groups/bsl/bsls/bsls_unspecifiedbool.t.cpp:627:23: note: insert an explicit cast to silence this issue
            } bag = { babel };
                      ^~~~~
                      static_cast<bool>( )
4 errors generated.

mversche · 2022-01-26T19:06:41Z

This overlaps with #273

notadragon

Overall this seems good, well past our baseline target of just compiling on M1. If you can address my feedback (and verify that things still work on your M1) then we can check this PR on other platforms and land it.

notadragon · 2022-01-26T20:08:02Z

groups/bsl/bsls/bsls_platform.h

@@ -911,6 +919,8 @@ struct Platform {
    struct CpuArmv5 : CpuArm {};
    struct CpuArmv6 : CpuArm {};
    struct CpuArmv7 : CpuArm {};
+    struct CpuArmv8 : CpuArm {};
+    struct CpuArmv9 : CpuArm {};


v9 seems unused, we can remove that.

notadragon · 2022-01-26T20:44:14Z

groups/bsl/bsls/bsls_timeutil.cpp

@@ -448,114 +448,57 @@ struct MachTimerUtil {

  private:


This is fairly extensive. Looking at it I think it's fine, but needs a few changes (which i can't comment on directly because of github):

I suspect the include of <mach/mach_time.h> is no longer the right include... it looks like it should just be <time.h>?

Please change "MachTimerUtil" to "DarwinTimerUtil".

It seems based on documentation that this should also work on non-m1 darwin machines, but we'll have to try to verify that.

cameron314 · 2022-01-26T19:03:10Z

groups/bsl/bslh/bslh_siphashalgorithm.cpp

@@ -121,7 +121,7 @@ static u64 u8to64_le(const u8* p)
 {
    BSLS_ASSERT(p);

-#if defined(BSLS_PLATFORM_CPU_X86) || defined(BSLS_PLATFORM_CPU_X86_64)
+#if defined(BSLS_PLATFORM_IS_LITTLE_ENDIAN)
    return *reinterpret_cast<const u64 *>(p);  // Ignore alignment.


What about alignment?

Huh, seems AArch64 typically allows unaligned accesses in user space.

Actually, that is a good point - i wouldn't widen the existing assumption to all little endian machines.

notadragon · 2022-01-26T22:48:30Z

groups/bsl/bslh/bslh_spookyhashalgorithm.h

@@ -380,7 +380,7 @@ SpookyHashAlgorithm::SpookyHashAlgorithm()
 inline
 SpookyHashAlgorithm::SpookyHashAlgorithm(const char *seed)
 : d_state(
-#if !defined(BSLS_PLATFORM_CPU_X86_64) && !defined(BSLS_PLATFORM_CPU_X86)
+#if !defined(BSLS_PLATFORM_IS_LITTLE_ENDIAN)


This should also not be checking for little_endian - this is about whether the cpu handles unaligned reads properly. For now i would leave this as-is and we can engage the optimization at a later date if we make the m1 a more fully supported platform.

emeryberger · 2022-02-05T00:37:02Z

Built from this branch on my Mac Mini M1. All I needed to do to get bsls_unspecifiedbool.t.cpp was adding static_cast<bool> around bt and babel; good error messages for a change!

I had an almost completely different set of failed unit tests. - denotes the tests that failed given in @thorpe-dev's top comment; + denotes the tests that failed in my run.

Click to see the diff

The following tests FAILED:
             106 - balm_publicationscheduler.t (Failed)
     -       117 - balst_stacktraceprintutil.t (Failed)
     -       160 - balxml_minireader.t (Failed)
     -       324 - bdldfp_decimalimputil.t (Failed)
     -       327 - bdldfp_decimalutil.t (Failed)
     -       359 - bdlma_concurrentmultipool.t (Failed)
     -       373 - bdlma_multipool.t (Failed)
     -       395 - bdls_memoryutil.t (Failed)
     -       399 - bdls_processutil.t (Failed)
     -       532 - bslim_formatguard.t (Failed)
     -       635 - bslmf_istransparentpredicate.t (Failed)
     -       706 - bslmt_threadutil.t (Failed)
     -       711 - bslmt_timedsemaphore.t (Failed)
     -       787 - bsls_stopwatch.t (Failed)
     -       793 - bsls_unspecifiedbool.t (Failed)
     -       811 - bslstl_chrono.t (Failed)
     -       822 - bslstl_function.t (Failed)
     -       824 - bslstl_function_invokerutil.t (Failed)
     -       881 - bslstl_stringview.t (Failed)
     +       118 - balst_stacktraceprintutil.t (Failed)
     +       161 - balxml_minireader.t (Failed)
     +       274 - bdlcc_boundedqueue.t (Failed)
     +       325 - bdldfp_decimalimputil.t (Failed)
     +       328 - bdldfp_decimalutil.t (Failed)
     +       360 - bdlma_concurrentmultipool.t (Failed)
     +       374 - bdlma_multipool.t (Failed)
     +       396 - bdls_memoryutil.t (Failed)
     +       400 - bdls_processutil.t (Failed)
     +       534 - bslim_formatguard.t (Failed)
     +       637 - bslmf_istransparentpredicate.t (Failed)
     +       713 - bslmt_timedsemaphore.t (Timeout)
     +       789 - bsls_stopwatch.t (Failed)
     +       813 - bslstl_chrono.t (Failed)
     +       824 - bslstl_function.t (Failed)
     +       826 - bslstl_function_invokerutil.t (Failed)
     +       883 - bslstl_stringview.t (Failed)

Platform: macOS Monterey Version 12.1, Mac mini (M1, 2020)

Darwin minim 21.2.0 Darwin Kernel Version 21.2.0: Sun Nov 28 20:29:10 PST 2021; root:xnu-8019.61.5~1/RELEASE_ARM64_T8101 arm64

Apple clang version 13.0.0 (clang-1300.0.29.30)
Target: arm64-apple-darwin21.2.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin

Improve support for Apple Silicon and macOS

3f21760

mversche mentioned this pull request Jan 26, 2022

Added support for Apple Silicon (M1). #273

Closed

notadragon suggested changes Jan 26, 2022

View reviewed changes

cameron314 reviewed Jan 26, 2022

View reviewed changes

notadragon suggested changes Jan 26, 2022

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improve support for Apple Silicon and macOS #275

Improve support for Apple Silicon and macOS #275

thorpe-dev commented Jan 26, 2022

mversche commented Jan 26, 2022

notadragon left a comment

notadragon Jan 26, 2022

notadragon Jan 26, 2022

cameron314 Jan 26, 2022

cameron314 Jan 26, 2022

notadragon Jan 26, 2022

notadragon Jan 26, 2022

emeryberger commented Feb 5, 2022 •

edited

Improve support for Apple Silicon and macOS #275

Are you sure you want to change the base?

Improve support for Apple Silicon and macOS #275

Conversation

thorpe-dev commented Jan 26, 2022

mversche commented Jan 26, 2022

notadragon left a comment

Choose a reason for hiding this comment

notadragon Jan 26, 2022

Choose a reason for hiding this comment

notadragon Jan 26, 2022

Choose a reason for hiding this comment

cameron314 Jan 26, 2022

Choose a reason for hiding this comment

cameron314 Jan 26, 2022

Choose a reason for hiding this comment

notadragon Jan 26, 2022

Choose a reason for hiding this comment

notadragon Jan 26, 2022

Choose a reason for hiding this comment

emeryberger commented Feb 5, 2022 • edited

emeryberger commented Feb 5, 2022 •

edited