Optionally disable thread-local storage. (#735)

Details:
- Implemented a new configure option, --disable-tls, which allows the
  user to optionally disable the use of thread-local storage qualifiers
  on static variables in BLIS. This option will rarely be needed, but
  in some situations may allow BLIS to compile when TLS is unavailable.
  Thanks to Nick Knight for suggesting this option.
- Unlike the --disable-system option, --disable-tls does not forcibly
  disable threading. Instead, warnings of the possible consequences of
  using threading with TLS disabled are added to:
  - the output of './configure --help';
  - the output of 'configure' the --disable-tls option is parsed;
  - the informational header output by the testsuite.
  Thanks to Minh Quan Ho for suggesting these warnings.
- Modified frame/include/bli_lang_defs.h so that BLIS_THREAD_LOCAL is
  defined to nothing when BLIS_ENABLE_TLS is not defined.
- Defined bli_info_get_enable_tls(), which returns whether the cpp macro
  BLIS_ENABLE_TLS was defined.
- Edited --disable-system configure status output for clarity.
- Whitespace updates.

build/bli_config.h.in

@@ -53,6 +53,12 @@
 #define BLIS_DISABLE_SYSTEM
 #endif
 
+#if @enable_tls@
+#define BLIS_ENABLE_TLS
+#else
+#define BLIS_DISABLE_TLS
+#endif
+
 #if @enable_openmp@
 #define BLIS_ENABLE_OPENMP
 #if @enable_openmp_as_def@

configure

@@ -199,6 +199,19 @@ print_usage()
                  library. When disabled, this option also forces the use
                  of --disable-threading.
 
+   --enable-tls, --disable-tls
+
+                 Enable thread-local storage (TLS) for static variables
+                 in BLIS. The default state is enabled. However, like with
+                 --disable-system, there may be rare situations, such as
+                 when --disable-system is appropriate, where thread-local
+                 storage is unsupported by the compiler. In those cases,
+                 disabling TLS may be a suitable workaround.
+		         WARNING: DISABLING TLS IS DANGEROUS AND MAY CAUSE RACE
+                 CONDITIONS! Please try combining --disable-tls with
+                 --disable-threading if you suspect any correctness or
+                 deadlock issues.
+
    --disable-pba-pools, --enable-pba-pools
    --disable-sba-pools, --enable-sba-pools
 
@@ -2512,6 +2525,9 @@ main()
 	# The system flag.
 	enable_system='yes'
 
+	# The thread-local storage flag.
+	enable_tls='yes'
+
 	# The threading flag.
 	threading_model='off'
 
@@ -2692,6 +2708,13 @@ main()
 							enable_system='no'
 							;;
 
+						enable-tls)
+							enable_tls='yes'
+							;;
+						disable-tls)
+							enable_tls='no'
+							;;
+
 						enable-threading=*)
 							threading_model=${OPTARG#*=}
 							;;
@@ -3543,13 +3566,25 @@ main()
 		enable_system_01=1
 	else
 		echo "${script_name}: disabling operating system support."
-		echo "${script_name}: WARNING: all threading will be disabled!"
+		echo "${script_name}: WARNING: disabling OS support forcibly disables all threading!"
 		enable_system_01=0
 
 		# Force threading to be disabled.
 		threading_model='off'
 	fi
 
+	# Check if we are building with or without thread-local storage support.
+	if [[ ${enable_tls} = yes ]]; then
+		echo "${script_name}: enabling thread-local storage (TLS) support."
+		enable_tls_01=1
+	else
+		echo "${script_name}: disabling thread-local storage (TLS) support."
+		echo "${script_name}: WARNING: THIS IS DANGEROUS! Disabling TLS may cause race conditions!"
+		echo "${script_name}: WARNING: Please try --disable-threading if you suspect any correctness"
+		echo "${script_name}: WARNING: or deadlock issues."
+		enable_tls_01=0
+	fi
+
 	# Check the threading model flag and standardize its value, if needed.
 	# Note that single-threaded mode will always be enabled, but not necessarily
 	# by default.
@@ -4202,6 +4237,7 @@ main()
 	| sed >"${bli_config_h_out_path}"                                    \
 	-e "s/@version@/${version_esc}/g"                                    \
 	-e "s/@enable_system@/${enable_system_01}/g"                         \
+	-e "s/@enable_tls@/${enable_tls_01}/g"                               \
 	-e "s/@enable_openmp@/${enable_openmp_01}/g"                         \
 	-e "s/@enable_openmp_as_def@/${enable_openmp_as_def_01}/g"           \
 	-e "s/@enable_pthreads@/${enable_pthreads_01}/g"                     \

frame/base/bli_info.c

@@ -103,10 +103,10 @@ gint_t bli_info_get_enable_sba_pools( void )
 }
 gint_t bli_info_get_enable_threading( void )
 {
-	if ( bli_info_get_enable_openmp() ||
+	if ( bli_info_get_enable_openmp()   ||
 	     bli_info_get_enable_pthreads() ||
 	     bli_info_get_enable_hpx() ) return 1;
-	else                                  return 0;
+	else                             return 0;
 }
 gint_t bli_info_get_enable_openmp( void )
 {
@@ -180,6 +180,14 @@ gint_t bli_info_get_thread_jrir_tlb( void )
 	return 0;
 #endif
 }
+gint_t bli_info_get_enable_tls( void )
+{
+#ifdef BLIS_ENABLE_TLS
+	return 1;
+#else
+	return 0;
+#endif
+}
 gint_t bli_info_get_enable_memkind( void )
 {
 #ifdef BLIS_ENABLE_MEMKIND

frame/base/bli_info.h

@@ -77,6 +77,7 @@ BLIS_EXPORT_BLIS gint_t bli_info_get_enable_hpx_as_default( void );
 BLIS_EXPORT_BLIS gint_t bli_info_get_thread_jrir_slab( void );
 BLIS_EXPORT_BLIS gint_t bli_info_get_thread_jrir_rr( void );
 BLIS_EXPORT_BLIS gint_t bli_info_get_thread_jrir_tlb( void );
+BLIS_EXPORT_BLIS gint_t bli_info_get_enable_tls( void );
 BLIS_EXPORT_BLIS gint_t bli_info_get_enable_memkind( void );
 BLIS_EXPORT_BLIS gint_t bli_info_get_enable_sandbox( void );
 

frame/include/bli_lang_defs.h

@@ -73,7 +73,10 @@
 // doesn't support __thread, as __GNUC__ is not quite unique to GCC.
 // But the possibility of someone using such non-main-stream compiler
 // for building BLIS is low.
-#if defined(__GNUC__) || defined(__clang__) || defined(__ICC) || defined(__IBMC__)
+#if defined(BLIS_ENABLE_TLS) && ( defined(__GNUC__)  || \
+                                  defined(__clang__) || \
+                                  defined(__ICC)     || \
+                                  defined(__IBMC__) )
   #define BLIS_THREAD_LOCAL __thread
 #else
   #define BLIS_THREAD_LOCAL

testsuite/src/test_libblis.c

@@ -867,6 +867,9 @@ void libblis_test_output_params_struct( FILE* os, test_params_t* params )
 	bli_rntm_set_ways_for_op( BLIS_TRSM, BLIS_LEFT,  m, n, k, &trsm_l );
 	bli_rntm_set_ways_for_op( BLIS_TRSM, BLIS_RIGHT, m, n, k, &trsm_r );
 
+	const bool tls_enabled = bli_info_get_enable_tls();
+	const bool thr_enabled = bli_info_get_enable_threading();
+
 	// Output some system parameters.
 	libblis_test_fprintf_c( os, "\n" );
 	libblis_test_fprintf_c( os, "--- BLIS library info -------------------------------------\n" );
@@ -916,6 +919,17 @@ void libblis_test_output_params_struct( FILE* os, test_params_t* params )
 	libblis_test_fprintf_c( os, "\n" );
 	libblis_test_fprintf_c( os, "--- BLIS parallelization info ---\n" );
 	libblis_test_fprintf_c( os, "\n" );
+	libblis_test_fprintf_c( os, "thread-local storage (TLS)     %d\n", ( int )tls_enabled );
+	if ( !tls_enabled && thr_enabled )
+	{
+	libblis_test_fprintf_c( os, "\n" );
+	libblis_test_fprintf_c( os, "[WARNING] BLIS was compiled with TLS disabled. We assume you know what\n" );
+	libblis_test_fprintf_c( os, "[WARNING] you're doing! Multithreaded race conditions, correctness\n" );
+	libblis_test_fprintf_c( os, "[WARNING] issues, and deadlocks may occur. If any of these happen,\n" );
+	libblis_test_fprintf_c( os, "[WARNING] please consider reconfiguring with --disable-threading.\n" );
+
+	}
+	libblis_test_fprintf_c( os, "\n" );
 	libblis_test_fprintf_c( os, "multithreading modes           %s\n", impl_str );
 	libblis_test_fprintf_c( os, "  default mode                 %s\n", def_impl_unset_str );
 	libblis_test_fprintf_c( os, "  current mode                 %s\n", def_impl_set_str );