Fix crash in invoking microtask on ios arm64

[nihui]

Summary

invoking microtask cause segment fault on ios platform
So I tried to modify __kmp_invoke_microtask() a little to use AArch64 assembly implementation, and the crash solved

cross-compiling for ios
host: fedora linux, clang 3.9.1

[AndreyChurbanov]

Couple notes.

1. Your patch breaks non-Linux ARM implementations I think, like FreeBSD ARM or NetBSD ARM. It would be safer if you changed KMP_OS_LINUX to (KMP_OS_LINUX || KMP_OS_DARWIN) instead of removing KMP_OS_LINUX, so that other platforms were not affected.

2. Please upload the full diff as opposed to stripped, so that people could see more context around your changes.

[nihui]

change KMP_OS_LINUX to (KMP_OS_LINUX || KMP_OS_DARWIN)
upload full diff

Thanks

[AndreyChurbanov]

LGTM

[nihui]

hello
I do not have commit access
please commit the change for me
thanks a lot ;)

runtime/src/z_Linux_asm.s

 # endif // KMP_OS_DARWIN
 #endif // KMP_ARCH_X86 || KMP_ARCH_x86_64
 
-#if KMP_OS_LINUX && KMP_ARCH_AARCH64
+#if KMP_ARCH_AARCH64
 
+# if KMP_OS_DARWIN
+#  define KMP_PREFIX_UNDERSCORE(x) _##x  // extra underscore for OS X* symbols
+#  define KMP_LABEL(x) L_##x             // form the name of label
+
+.macro ALIGN
+  .align $0
+.endmacro
+
+.macro DEBUG_INFO
+/* Not sure what .size does in icc, not sure if we need to do something
+   similar for OS X*.
+*/
+.endmacro
+
+.macro PROC
+  ALIGN  4
+  .globl KMP_PREFIX_UNDERSCORE($0)
+KMP_PREFIX_UNDERSCORE($0):
+.endmacro
+# else // KMP_OS_DARWIN
 #  define KMP_PREFIX_UNDERSCORE(x) x  // no extra underscore for Linux* OS symbols
 // Format labels so that they don't override function names in gdb's backtraces
 #  define KMP_LABEL(x) .L_##x         // local label hidden from backtraces

 KMP_PREFIX_UNDERSCORE(\proc):
   .cfi_startproc
 .endm
+# endif // KMP_OS_DARWIN
 
-#endif // KMP_OS_LINUX && KMP_ARCH_AARCH64
+#endif // KMP_ARCH_AARCH64
 
 // -----------------------------------------------------------------------
 // data

 #endif /* KMP_ARCH_X86_64 */
 
 // '
-#if KMP_OS_LINUX && KMP_ARCH_AARCH64
+#if KMP_ARCH_AARCH64
 
 //------------------------------------------------------------------------
 //

   DEBUG_INFO __kmp_invoke_microtask
 // -- End  __kmp_invoke_microtask
 
-#endif /* KMP_OS_LINUX && KMP_ARCH_AARCH64 */
+#endif /* KMP_ARCH_AARCH64 */
 
 #if KMP_ARCH_PPC64
 

runtime/src/z_Linux_util.cpp

 
 #endif // USE_LOAD_BALANCE
 
-#if !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_MIC || (KMP_OS_LINUX && KMP_ARCH_AARCH64) || KMP_ARCH_PPC64)
+#if !(KMP_ARCH_X86 || KMP_ARCH_X86_64 || KMP_MIC || KMP_ARCH_AARCH64 || KMP_ARCH_PPC64)
 
 // we really only need the case with 1 argument, because CLANG always build
 // a struct of pointers to shared variables referenced in the outlined function