Make bound checking faster.

With this commit bound checking will be about 20% .. 25% faster.
(Tested with running testsuites with bound checking.
 About 1000 files.)
The speedup is made possible because of TLS implementation.
TLS is not possible with -run so create a new bcheck_run.o

Makefile:lib/Makefile:win32/build-tcc.bat
- Compile new bcheck_run.o
tccelf.c:tccpe.c
- Use bcheck_run.o
lib/bcheck.c
- Use TLS code

Makefile

@@ -372,7 +372,7 @@ IR = $(IM) mkdir -p $2 && cp -r $1/. $2
 IM = @echo "-> $2 : $1" ;
 BINCHECK = $(if $(wildcard $(PROGS) *-tcc$(EXESUF)),,@echo "Makefile: nothing found to install" && exit 1)
 
-EXTRA_O = runmain.o bt-exe.o bt-dll.o bt-log.o bcheck.o
+EXTRA_O = runmain.o bt-exe.o bt-dll.o bt-log.o bcheck.o bcheck_run.o
 
 # install progs & libs
 install-unx:

lib/Makefile

@@ -53,10 +53,10 @@ Cbc = $(Cbt)$(subst yes,,$(CONFIG_bcheck))
 $(Nat)COMMON_O += runmain.o tcov.o
 $(Cbt)COMMON_O += bt-exe.o bt-log.o
 $(Cbt)WIN_O += bt-dll.o
-$(Cbc)COMMON_O += bcheck.o
+$(Cbc)COMMON_O += bcheck.o bcheck_run.o
 
 # not in libtcc1.a
-EXTRA_O = runmain.o bt-exe.o bt-dll.o bt-log.o bcheck.o
+EXTRA_O = runmain.o bt-exe.o bt-dll.o bt-log.o bcheck.o bcheck_run.o
 
 OBJ-i386 = $(I386_O) pic86.o $(LIN_O)
 OBJ-x86_64 = $(X86_64_O) va_list.o $(LIN_O)
@@ -95,6 +95,8 @@ $(TOP)/%.o : %.c $(TCC)
 	$S$(XCC) -c $< -o $@ $(XFLAGS)
 
 $(TOP)/bcheck.o : XFLAGS += $(BFLAGS)
+$(TOP)/bcheck_run.o : bcheck.c $(TCC)
+	$S$(XCC) -c bcheck.c -o $@ $(XFLAGS)  $(BFLAGS) -DBCHECK_RUN
 
 $(X)crt1w.o : crt1.c
 $(X)wincrt1w.o : wincrt1.c

lib/bcheck.c

@@ -53,7 +53,11 @@
   /* an __attribute__ macro is defined in the system headers */
   #undef __attribute__ 
 #endif
+#ifdef __i386__
 #define FASTCALL __attribute__((regparm(3)))
+#else
+#define FASTCALL
+#endif
 
 #ifdef _WIN32
 # define DLL_EXPORT __declspec(dllexport)
@@ -157,12 +161,12 @@ static pthread_spinlock_t bounds_spin;
 #define HAVE_SIGNAL            (1)
 #define HAVE_SIGACTION         (1)
 #define HAVE_FORK              (1)
-#if !defined(__APPLE__) && defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
-#define HAVE_TLS_FUNC          (0)
-#define HAVE_TLS_VAR           (1)
-#else
+#if defined(__APPLE__) || defined(__arm__) || defined(__GNUC__) || defined(BCHECK_RUN)
 #define HAVE_TLS_FUNC          (1)
 #define HAVE_TLS_VAR           (0)
+#else
+#define HAVE_TLS_FUNC          (0)
+#define HAVE_TLS_VAR           (1)
 #endif
 #if defined CONFIG_TCC_MUSL || defined __ANDROID__
 # undef HAVE_CTYPE

tccelf.c

@@ -1839,6 +1839,9 @@ ST_FUNC void tcc_add_runtime(TCCState *s1)
 
 #ifdef CONFIG_TCC_BCHECK
         if (s1->do_bounds_check && s1->output_type != TCC_OUTPUT_DLL) {
+	    if (s1->output_type == TCC_OUTPUT_MEMORY)
+                tcc_add_support(s1, "bcheck_run.o");
+	    else
                 tcc_add_support(s1, "bcheck.o");
 # if !(TARGETOS_OpenBSD || TARGETOS_NetBSD)
             tcc_add_library(s1, "dl");

tccpe.c

@@ -2131,8 +2131,12 @@ static void pe_add_runtime(TCCState *s1, struct pe_info *pe)
 #ifdef CONFIG_TCC_BACKTRACE
     if (s1->do_backtrace) {
 #ifdef CONFIG_TCC_BCHECK
-        if (s1->do_bounds_check && s1->output_type != TCC_OUTPUT_DLL)
+        if (s1->do_bounds_check && s1->output_type != TCC_OUTPUT_DLL) {
+	    if (s1->output_type == TCC_OUTPUT_MEMORY)
+                tcc_add_support(s1, "bcheck_run.o");
+	    else
                 tcc_add_support(s1, "bcheck.o");
+	}
 #endif
         if (s1->output_type == TCC_OUTPUT_EXE)
             tcc_add_support(s1, "bt-exe.o");

win32/build-tcc.bat

@@ -201,6 +201,7 @@ exit /B %ERRORLEVEL%
 .\%1tcc -B. -c ../lib/builtin.c
 .\%1tcc -ar lib/%1libtcc1.a %LIBTCC1%.o crt1.o crt1w.o wincrt1.o wincrt1w.o dllcrt1.o dllmain.o winex.o chkstk.o alloca.o alloca-bt.o stdatomic.o atomic.o builtin.o
 .\%1tcc -B. -c ../lib/bcheck.c -o lib/%1bcheck.o -bt -I..
+.\%1tcc -B. -c ../lib/bcheck.c -o lib/%1bcheck_run.o -bt -I.. -DBCHECK_RUN
 .\%1tcc -B. -c ../lib/bt-exe.c -o lib/%1bt-exe.o
 .\%1tcc -B. -c ../lib/bt-log.c -o lib/%1bt-log.o
 .\%1tcc -B. -c ../lib/bt-dll.c -o lib/%1bt-dll.o