i386: use __fixdfdi instead of __tcc_cvt_ftol

Variants __fixsfdi/__fixxfdi are not needed for now because
the value is converted to double always.

Also:
- remove __tcc_fpinit for unix as it seems redundant by the
  __setfpucw call in the startup code
- avoid reference to s->runtime_main in cross compilers
- configure: fix --with-libgcc help
- tcctok.h: cleanup

configure

@@ -273,7 +273,7 @@ Advanced options (experts only):
   --strip-binaries         strip symbol tables from resulting binaries
   --disable-static         make libtcc.so instead of libtcc.a
   --disable-rpath          disable use of -rpath with the above
-  --with-libgcc            use libgcc_s.so.1 instead of libtcc.a in dynamic link
+  --with-libgcc            use libgcc_s.so.1 instead of libtcc1.a in dynamic link
   --enable-mingw32         build windows version on linux with mingw32
   --enable-cygwin          build windows version on windows with cygwin
   --enable-cross           build cross compilers

i386-gen.c

@@ -581,14 +581,6 @@ ST_FUNC void gfunc_prolog(CType *func_type)
         func_bound_offset = lbounds_section->data_offset;
     }
 #endif
-
-#ifndef CONFIG_USE_LIBGCC
-#ifndef TCC_TARGET_PE
-    if (0 == strcmp(funcname, "main"))
-        gen_static_call(TOK___tcc_fpinit);
-#endif
-#endif
-
 }
 
 /* generate function epilog */
@@ -988,16 +980,20 @@ ST_FUNC void gen_cvt_itof(int t)
 }
 
 /* convert fp to int 't' type */
-/* XXX: handle long long case */
 ST_FUNC void gen_cvt_ftoi(int t)
 {
-    gv(RC_FLOAT);
-    save_reg(TREG_EAX);
-    save_reg(TREG_EDX);
-    gen_static_call(TOK___tcc_cvt_ftol);
-    vtop->r = TREG_EAX; /* mark reg as used */
-    if (t == VT_LLONG)
-        vtop->r2 = TREG_EDX;
+    int bt = vtop->type.t & VT_BTYPE;
+    if (bt == VT_FLOAT)
+        vpush_global_sym(&func_old_type, TOK___fixsfdi);
+    else if (bt == VT_LDOUBLE)
+        vpush_global_sym(&func_old_type, TOK___fixxfdi);
+    else
+        vpush_global_sym(&func_old_type, TOK___fixdfdi);
+    vswap();
+    gfunc_call(1);
+    vpushi(0);
+    vtop->r = REG_IRET;
+    vtop->r2 = REG_LRET;
 }
 
 /* convert from one floating point type to another */

lib/libtcc1.c

@@ -478,24 +478,6 @@ long long __ashldi3(long long a, int b)
 #endif
 }
 
-#ifndef _WIN32
-void __tcc_fpinit(void)
-{
-    unsigned c = 0x137F;
-    __asm__ __volatile__ ("fldcw %0" : : "m" (c));
-}
-#endif
-long long __tcc_cvt_ftol(long double x)
-{
-    unsigned c0, c1;
-    long long ret;
-    __asm__ __volatile__ ("fnstcw %0" : "=m" (c0));
-    c1 = c0 | 0x0C00;
-    __asm__ __volatile__ ("fldcw %0" : : "m" (c1));
-    __asm__ __volatile__ ("fistpll %0"  : "=m" (ret));
-    __asm__ __volatile__ ("fldcw %0" : : "m" (c0));
-    return ret;
-}
 #endif /* !__x86_64__ */
 
 /* XXX: fix tcc's code generator to do this instead */
@@ -616,6 +598,27 @@ unsigned long long __fixunsxfdi (long double a1)
         return 0;
 }
 
+long long __fixsfdi (float a1)
+{
+    long long ret; int s;
+    ret = __fixunssfdi((s = a1 >= 0) ? a1 : -a1);
+    return s ? ret : -ret;
+}
+
+long long __fixdfdi (double a1)
+{
+    long long ret; int s;
+    ret = __fixunsdfdi((s = a1 >= 0) ? a1 : -a1);
+    return s ? ret : -ret;
+}
+
+long long __fixxfdi (long double a1)
+{
+    long long ret; int s;
+    ret = __fixunsxfdi((s = a1 >= 0) ? a1 : -a1);
+    return s ? ret : -ret;
+}
+
 #if defined(__x86_64__) && !defined(_WIN64)
 
 #ifndef __TINYC__

tccelf.c

@@ -178,11 +178,11 @@ LIBTCCAPI void *tcc_get_symbol(TCCState *s, const char *name)
     return (void*)(uintptr_t)get_elf_sym_addr(s, name, 0);
 }
 
-#ifdef TCC_IS_NATIVE
+#if defined TCC_IS_NATIVE || defined TCC_TARGET_PE
 /* return elf symbol value or error */
 ST_FUNC void* tcc_get_symbol_err(TCCState *s, const char *name)
 {
-    return (void*)get_elf_sym_addr(s, name, 1);
+    return (void*)(uintptr_t)get_elf_sym_addr(s, name, 1);
 }
 #endif
 

tccpe.c

@@ -1800,7 +1800,9 @@ static void pe_add_runtime(TCCState *s1, struct pe_info *pe)
 
     if (TCC_OUTPUT_MEMORY == s1->output_type) {
         pe_type = PE_RUN;
+#ifdef TCC_IS_NATIVE
         s1->runtime_main = start_symbol;
+#endif
     } else {
         pe->start_addr = (DWORD)tcc_get_symbol_err(s1, start_symbol);
     }

tcctok.h

@@ -143,6 +143,27 @@
 #endif
 
 /* builtin functions or variables */
+#ifndef TCC_ARM_EABI
+     DEF(TOK_memcpy, "memcpy")
+     DEF(TOK_memset, "memset")
+     DEF(TOK___divdi3, "__divdi3")
+     DEF(TOK___moddi3, "__moddi3")
+     DEF(TOK___udivdi3, "__udivdi3")
+     DEF(TOK___umoddi3, "__umoddi3")
+     DEF(TOK___ashrdi3, "__ashrdi3")
+     DEF(TOK___lshrdi3, "__lshrdi3")
+     DEF(TOK___ashldi3, "__ashldi3")
+     DEF(TOK___floatundisf, "__floatundisf")
+     DEF(TOK___floatundidf, "__floatundidf")
+# ifndef TCC_ARM_VFP
+     DEF(TOK___floatundixf, "__floatundixf")
+     DEF(TOK___fixunsxfdi, "__fixunsxfdi")
+# endif
+     DEF(TOK___fixunssfdi, "__fixunssfdi")
+     DEF(TOK___fixunsdfdi, "__fixunsdfdi")
+#endif
+
+#if defined TCC_TARGET_ARM
 # ifdef TCC_ARM_EABI
      DEF(TOK_memcpy, "__aeabi_memcpy")
      DEF(TOK_memcpy4, "__aeabi_memcpy4")
@@ -150,16 +171,6 @@
      DEF(TOK_memset, "__aeabi_memset")
      DEF(TOK___aeabi_ldivmod, "__aeabi_ldivmod")
      DEF(TOK___aeabi_uldivmod, "__aeabi_uldivmod")
-#else
-     DEF(TOK_memcpy, "memcpy")
-     DEF(TOK_memset, "memset")
-     DEF(TOK___divdi3, "__divdi3")
-     DEF(TOK___moddi3, "__moddi3")
-     DEF(TOK___udivdi3, "__udivdi3")
-     DEF(TOK___umoddi3, "__umoddi3")
-#endif
-#if defined(TCC_TARGET_ARM)
-#ifdef TCC_ARM_EABI
      DEF(TOK___aeabi_idivmod, "__aeabi_idivmod")
      DEF(TOK___aeabi_uidivmod, "__aeabi_uidivmod")
      DEF(TOK___divsi3, "__aeabi_idiv")
@@ -168,6 +179,13 @@
      DEF(TOK___floatdidf, "__aeabi_l2d")
      DEF(TOK___fixsfdi, "__aeabi_f2lz")
      DEF(TOK___fixdfdi, "__aeabi_d2lz")
+     DEF(TOK___ashrdi3, "__aeabi_lasr")
+     DEF(TOK___lshrdi3, "__aeabi_llsr")
+     DEF(TOK___ashldi3, "__aeabi_llsl")
+     DEF(TOK___floatundisf, "__aeabi_ul2f")
+     DEF(TOK___floatundidf, "__aeabi_ul2d")
+     DEF(TOK___fixunssfdi, "__aeabi_f2ulz")
+     DEF(TOK___fixunsdfdi, "__aeabi_d2ulz")
 # else
      DEF(TOK___modsi3, "__modsi3")
      DEF(TOK___umodsi3, "__umodsi3")
@@ -185,7 +203,9 @@
      DEF(TOK___fixsfdi, "__fixsfdi")
      DEF(TOK___fixdfdi, "__fixdfdi")
 # endif
-#elif defined(TCC_TARGET_C67)
+#endif
+
+#if defined TCC_TARGET_C67
      DEF(TOK__divi, "_divi")
      DEF(TOK__divu, "_divu")
      DEF(TOK__divf, "_divf")
@@ -193,32 +213,18 @@
      DEF(TOK__remi, "_remi")
      DEF(TOK__remu, "_remu")
 #endif
-#ifdef TCC_TARGET_I386
-     DEF(TOK___tcc_fpinit, "__tcc_fpinit")
-     DEF(TOK___tcc_cvt_ftol, "__tcc_cvt_ftol")
+
+#if defined TCC_TARGET_I386
+     DEF(TOK___fixsfdi, "__fixsfdi")
+     DEF(TOK___fixdfdi, "__fixdfdi")
+     DEF(TOK___fixxfdi, "__fixxfdi")
 #endif
-#ifdef TCC_ARM_EABI
-     DEF(TOK___ashrdi3, "__aeabi_lasr")
-     DEF(TOK___lshrdi3, "__aeabi_llsr")
-     DEF(TOK___ashldi3, "__aeabi_llsl")
-     DEF(TOK___floatundisf, "__aeabi_ul2f")
-     DEF(TOK___floatundidf, "__aeabi_ul2d")
-     DEF(TOK___fixunssfdi, "__aeabi_f2ulz")
-     DEF(TOK___fixunsdfdi, "__aeabi_d2ulz")
-#else
-     DEF(TOK___ashrdi3, "__ashrdi3")
-     DEF(TOK___lshrdi3, "__lshrdi3")
-     DEF(TOK___ashldi3, "__ashldi3")
-     DEF(TOK___floatundisf, "__floatundisf")
-     DEF(TOK___floatundidf, "__floatundidf")
-#ifndef TCC_ARM_VFP
-     DEF(TOK___floatundixf, "__floatundixf")
-     DEF(TOK___fixunsxfdi, "__fixunsxfdi")
+
+#if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
+     DEF(TOK_alloca, "alloca")
 #endif
-     DEF(TOK___fixunssfdi, "__fixunssfdi")
-     DEF(TOK___fixunsdfdi, "__fixunsdfdi")
-#endif
-#ifdef TCC_TARGET_PE
+
+#if defined TCC_TARGET_PE
      DEF(TOK___chkstk, "__chkstk")
 #endif
 
@@ -244,9 +250,6 @@
      DEF(TOK_strlen, "strlen")
      DEF(TOK_strcpy, "strcpy")
 #endif
-#if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
-     DEF(TOK_alloca, "alloca")
-#endif
 
 /* Tiny Assembler */
  DEF_ASM(byte)