arm64-win32 review: fix problems and pass tests

tccpe.c:
- fix arm64 unwind codes (to make native set/longjmp() work)
  sizeof(RUNTIME_FUNCTION) is 8 on arm64 in the first place
  no need to note stack slots if we don't save any registers anyway

arm64-gen.c:
- fix long double reg-move
- fix arm64_hfa() for structs with float arrays
- gfunc_prolog(): setup stackframe eariler (simplifies unwind codes)
- new function gv_addr(RC);

win32/include/setjmp.h:
- provide correct definition for setjmo() (frameoffset = 224)

tccasm.c:
- support ".quad" with symbol & relocation
- support ".size"
- fix ". - symbol" arithmetic

win32/lib/crt1.c and win32/include/stdlib.h:
- do not write to __argc/__argv which reside in msvcrt.dll
  (msvcrt.dll on arm64 does not like that, crashes on unload)

tcc.c,libtcc.c:
- new functions tcc_fopen/fclose to avoid different stdio unstances
  in tcc.exe & libtcc.dll

tests & github workflow:
- add test-win32.bat to run tests with a tcc compiled by build-tcc.bat
- add msvcrt_start.c for gcc/clang to use the same runtime as tcc

  the problem is that newer gcc as well as clang and cl are
  linking to newer runtimes (such as UCRT) that have partially
  different printf format behavior which makes tcctest fail.

  the solution here is to force these compilers to link with
  msvcrt.dll just like tcc.

  Also, there is no gcc on arm64-win32 currently at all.

  Anyway, this approach to running the github CI tests
  does not require msys2.  But It does rely on gnumake
  as well as on some 'sh' shell though which seems to be
  installed somewhere (maybe it is the one from git).

Author: grischka

.github/workflows/build.yml

@@ -34,56 +34,51 @@ jobs:
     timeout-minutes: 6
     steps:
     - uses: actions/checkout@v4
-    - name: make & test tcc (x86_64-win32)
-      shell: cmd
-      run: |
-        echo ::group:: setup msys mingw64-gcc
-        set MSYS2_PATH_TYPE=inherit
-        set MSYSTEM=MINGW64
-        set CHERE_INVOKING=yes
-        C:\msys64\usr\bin\bash -l -c "pacman -S --noconfirm mingw-w64-x86_64-gcc"
-        echo ::endgroup::
-        C:\msys64\usr\bin\bash -l -c "./configure && make && make test -k"
-    - uses: ilammy/msvc-dev-cmd@v1
-      with:
-        arch: amd64
-    - name: build with MSVC (x86_64-win32)
+    - name: test (x86_64-win32)
       shell: cmd
       run: |
         echo ::group:: run build-tcc.bat
         cd win32
-        call build-tcc.bat -t 64 -c cl
+        call "%ProgramFiles%\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\VsDevCmd.bat" -arch=amd64
+        call build-tcc.bat -c cl -t x86_64
+        @echo off
         echo ::endgroup::
-        .\tcc -I.. libtcc.dll -v ../tests/libtcc_test.c -o libtest.exe && .\libtest.exe
-        .\tcc -I.. libtcc.dll -run ../tests/libtcc_test.c
+        cd ..\tests
+        call test-win32.bat all -k
 
   test-i386-win32:
     runs-on: windows-2025
     timeout-minutes: 6
     steps:
     - uses: actions/checkout@v4
-    - name: make & test tcc (i386-win32)
+    - name: test (i386-win32)
       shell: cmd
       run: |
-        echo ::group:: setup msys mingw32-gcc
-        set MSYS2_PATH_TYPE=inherit
-        set MSYSTEM=MINGW32
-        set CHERE_INVOKING=yes
-        C:\msys64\usr\bin\bash -l -c "pacman -S --noconfirm mingw-w64-i686-gcc"
+        echo ::group:: run build-tcc.bat
+        cd win32
+        call "%ProgramFiles%\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\VsDevCmd.bat" -arch=x86
+        call build-tcc.bat -c cl -t i386
+        @echo off
         echo ::endgroup::
-        C:\msys64\usr\bin\bash -l -c "./configure && make && make test -k"
-    - uses: ilammy/msvc-dev-cmd@v1
-      with:
-        arch: x86
-    - name: build with MSVC (i386-win32)
+        cd ..\tests
+        call test-win32.bat -p c:\mingw32\bin all -k
+
+  test-arm64-win32:
+    runs-on: windows-11-arm
+    timeout-minutes: 6
+    steps:
+    - uses: actions/checkout@v4
+    - name: test (arm64-win32)
       shell: cmd
       run: |
         echo ::group:: run build-tcc.bat
         cd win32
-        call build-tcc.bat -t 32 -c cl
+        call "%ProgramFiles%\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\VsDevCmd.bat" -arch=arm64
+        call build-tcc.bat -c cl -t arm64
+        @echo off
         echo ::endgroup::
-        .\tcc -I.. libtcc.dll -v ../tests/libtcc_test.c -o libtest.exe && .\libtest.exe
-        .\tcc -I.. libtcc.dll -run ../tests/libtcc_test.c
+        cd ..\tests
+        call test-win32.bat -c clang all -k
 
   test-armv7-linux:
     runs-on: ubuntu-22.04

Makefile

@@ -31,7 +31,9 @@ ifdef CONFIG_WIN32
   LIBTCCDEF = libtcc.def
  endif
  ifneq ($(CONFIG_debug),yes)
-  LDFLAGS += -s
+  ifneq ($(CC_NAME),clang)
+   LDFLAGS += -s
+  endif
  endif
  NATIVE_TARGET = $(if $(findstring arm64,$(ARCH)),arm64-win32,$(ARCH)-win$(if $(findstring arm,$(ARCH)),ce,32))
 else

arm64-asm.c

@@ -86,11 +86,7 @@ enum {
 #define TREG_X30 30
 #define TREG_SP  31
 
-#ifdef TCC_TARGET_PE
-#define ARM64_FREG_BASE 19
-#else
 #define ARM64_FREG_BASE 20
-#endif
 #define ARM64_FREG_LAST (ARM64_FREG_BASE + 7)
 
 typedef struct Operand {

arm64-gen.c

@@ -579,7 +579,7 @@ ST_FUNC void load(int r, SValue *sv)
     if (svr < VT_CONST) {
         if (IS_FREG(r) && IS_FREG(svr))
             if (svtt == VT_LDOUBLE)
-                o(ARM64_MOV_V16B | fltr(r) | fltr(svr) << 5);
+                o(ARM64_MOV_V16B | fltr(r) | fltr(svr) * 0x10020);
                     // mov v(r).16b,v(svr).16b
             else
                 o(ARM64_FMOV_SCALAR | fltr(r) | fltr(svr) << 5); // fmov d(r),d(svr)
@@ -794,7 +794,7 @@ static int arm64_hfa_aux(CType *type, int *fsize, int num)
             return num;
         }
     }
-    else if ((type->t & VT_ARRAY) && ((type->t & VT_BTYPE) != VT_PTR)) {
+    else if (type->t & VT_ARRAY) { /* handle float array within struct */
         int num1;
         if (!type->ref->c)
             return num;
@@ -811,8 +811,7 @@ static int arm64_hfa_aux(CType *type, int *fsize, int num)
 
 static int arm64_hfa(CType *type, unsigned *fsize)
 {
-    if ((type->t & VT_BTYPE) == VT_STRUCT ||
-        ((type->t & VT_ARRAY) && ((type->t & VT_BTYPE) != VT_PTR))) {
+    if ((type->t & VT_BTYPE) == VT_STRUCT) {
         int sz = 0;
         int n = arm64_hfa_aux(type, &sz, 0);
         if (0 < n && n <= 4) {
@@ -1037,6 +1036,13 @@ static void arm64_sub_sp(uint64_t diff)
     }
 }
 
+static int gv_addr(int r)
+{
+    gaddrof();
+    vtop->type.t = VT_PTR;
+    return gv(r);
+}
+
 ST_FUNC void gfunc_call(int nb_args)
 {
     CType *return_type;
@@ -1147,9 +1153,7 @@ ST_FUNC void gfunc_call(int nb_args)
             else if ((vtop->type.t & VT_BTYPE) == VT_STRUCT) {
                 int align, size = type_size(&vtop->type, &align);
                 if (size) {
-                    vtop->type.t = VT_PTR;
-                    gaddrof();
-                    gv(RC_R(a[i] / 2));
+                    gv_addr(RC_R(a[i] / 2));
                     arm64_ldrs(a[i] / 2, size);
                 }
             }
@@ -1165,9 +1169,7 @@ ST_FUNC void gfunc_call(int nb_args)
                 uint32_t j, sz, n = arm64_hfa(&vtop->type, &sz);
                 if (n > 0) {
                     /* HFA struct - load from memory into float registers */
-                    vtop->type.t = VT_PTR;
-                    gaddrof();
-                    gv(RC_R30);
+                    gv_addr(RC_R30);
                     for (j = 0; j < n; j++)
                         o(0x3d4003c0 |
                           (sz & 16) << 19 | -(sz & 8) << 27 | (sz & 4) << 29 |
@@ -1186,7 +1188,7 @@ ST_FUNC void gfunc_call(int nb_args)
     if ((return_type->t & VT_BTYPE) == VT_STRUCT) {
         if (a[0] == 1) {
             // indirect return: set x8 and discard the stack value
-            gv(RC_R(8));
+            gv_addr(RC_R(8));
             --vtop;
         }
         else
@@ -1206,7 +1208,7 @@ ST_FUNC void gfunc_call(int nb_args)
         int bt = rt & VT_BTYPE;
         if (bt == VT_STRUCT && !(a[0] & 1)) {
             // A struct was returned in registers, so write it out:
-            gv(RC_R(8));
+            gv_addr(RC_R(8));
             --vtop;
             if (a[0] == 0) {
                 int align, size = type_size(return_type, &align);
@@ -1271,6 +1273,7 @@ ST_FUNC void gfunc_prolog(Sym *func_sym)
 
     for (sym = func_type->ref; sym; sym = sym->next)
         ++n;
+
     pcs_n = n - 1;
     c = n + variadic;
     t = tcc_malloc(c * sizeof(*t));
@@ -1321,6 +1324,8 @@ ST_FUNC void gfunc_prolog(Sym *func_sym)
 
     arm64_func_start_offset = ind;
     o(0xa9b27bfd); // stp x29,x30,[sp,#-224]!
+    o(0x910003fd); // mov x29,sp
+
     for (i = 0; i < last_float; i++)
         // stp q0,q1,[sp,#16], stp q2,q3,[sp,#48]
         // stp q4,q5,[sp,#80], stp q6,q7,[sp,#112]
@@ -1368,7 +1373,6 @@ ST_FUNC void gfunc_prolog(Sym *func_sym)
     tcc_free(a);
     tcc_free(t);
 
-    o(0x910003fd); // mov x29,sp
     arm64_func_sub_sp_offset = ind;
     /* In gfunc_epilog these will be replaced with stack setup code. */
     for (i = 0; i < ARM64_FUNC_STACK_SETUP_SLOTS; ++i)
@@ -1384,8 +1388,7 @@ ST_FUNC void gen_va_start(void)
 {
     int r;
     --vtop; // we don't need the "arg"
-    gaddrof();
-    r = intr(gv(RC_INT));
+    r = intr(gv_addr(RC_INT));
 
 #ifdef TCC_TARGET_PE
     if (arm64_func_va_list_stack) {
@@ -1433,13 +1436,12 @@ ST_FUNC void gen_va_arg(CType *t)
     uint32_t r0, r1;
 
 #ifdef TCC_TARGET_PE
-    int indirect = 0, slot = size + 7 & -8;
+    int indirect = 0, slot = (size + 7) & -8;
 
     if (size > 16)
         indirect = 1, slot = 8;
 
-    gaddrof();
-    r0 = intr(gv(RC_INT));
+    r0 = intr(gv_addr(RC_INT));
     r1 = get_reg(RC_INT);
     vtop[0].r = r1 | VT_LVAL;
     r1 = intr(r1);
@@ -1459,7 +1461,6 @@ ST_FUNC void gen_va_arg(CType *t)
         o(0x9100001e | r1 << 5 | slot << 10); // add x30,x(r1),#(slot)
         o(0xf900001e | r0 << 5); // str x30,[x(r0)] // ap += slot
     }
-
     if (indirect)
         o(ARM64_LDR_X | ARM64_RN(r1) | r1); // ldr x(r1),[x(r1)]
 
@@ -1469,8 +1470,7 @@ ST_FUNC void gen_va_arg(CType *t)
     if (!is_float(t->t))
         hfa = arm64_hfa(t, &fsize);
 
-    gaddrof();
-    r0 = intr(gv(RC_INT));
+    r0 = intr(gv_addr(RC_INT));
     r1 = get_reg(RC_INT);
     vtop[0].r = r1 | VT_LVAL;
     r1 = intr(r1);
@@ -1570,8 +1570,7 @@ ST_FUNC void gfunc_return(CType *func_type)
     case 0:
         if ((func_type->t & VT_BTYPE) == VT_STRUCT) {
             int align, size = type_size(func_type, &align);
-            gaddrof();
-            gv(RC_R(0));
+            gv_addr(RC_R(0));
             arm64_ldrs(0, size);
         }
         else
@@ -1590,8 +1589,7 @@ ST_FUNC void gfunc_return(CType *func_type)
         if ((func_type->t & VT_BTYPE) == VT_STRUCT) {
           /* HFA struct return - load from the address on vtop into float registers */
           uint32_t j, sz, n = arm64_hfa(func_type, &sz);
-          gaddrof();
-          gv(RC_R(0));
+          gv_addr(RC_R(0));
           for (j = 0; j < n; j++)
               o(0x3d400000 |
                 (sz & 16) << 19 | -(sz & 8) << 27 | (sz & 4) << 29 |

i386-asm.c

@@ -494,13 +494,6 @@ ST_FUNC void gen_expr32(ExprValue *pe)
 	gen_addr32(pe->sym ? VT_SYM : 0, pe->sym, pe->v);
 }
 
-#ifdef TCC_TARGET_X86_64
-ST_FUNC void gen_expr64(ExprValue *pe)
-{
-    gen_addr64(pe->sym ? VT_SYM : 0, pe->sym, pe->v);
-}
-#endif
-
 /* XXX: unify with C code output ? */
 static void gen_disp32(ExprValue *pe)
 {

lib/Makefile

@@ -41,7 +41,7 @@ ARM_O = libtcc1.o armeabi.o armflush.o $(COMMON_O)
 ARM64_O = lib-arm64.o $(COMMON_O)
 RISCV64_O = lib-arm64.o $(COMMON_O)
 COMMON_O = stdatomic.o atomic.o builtin.o alloca.o alloca-bt.o
-WIN_O = crt1.o crt1w.o wincrt1.o wincrt1w.o dllcrt1.o dllmain.o
+WIN_O = crt1.o crt1w.o wincrt1.o wincrt1w.o dllcrt1.o dllmain.o winex.o
 LIN_O = dsohandle.o
 OSX_O =
 
@@ -63,7 +63,7 @@ OBJ-x86_64 = $(X86_64_O) va_list.o $(LIN_O)
 OBJ-x86_64-osx = $(X86_64_O) va_list.o $(OSX_O)
 OBJ-i386-win32 = $(I386_O) chkstk.o $(WIN_O)
 OBJ-x86_64-win32 = $(X86_64_O) chkstk.o $(WIN_O)
-OBJ-arm64 = $(ARM64_O) $(LIN_O)
+OBJ-arm64 = $(ARM64_O) armflush.o $(LIN_O)
 OBJ-arm64-osx = $(ARM64_O) $(OSX_O)
 OBJ-arm64-win32 = $(ARM64_O) chkstk.o $(WIN_O)
 OBJ-arm = $(ARM_O) $(LIN_O)

lib/armflush.c

@@ -4,6 +4,9 @@
    intrinsic with gcc.  However tcc in order to compile
    itself needs this function */
 
+/* ------------------------------------------------------------- */
+#if defined __arm__
+
 #ifdef __TINYC__
 
 /* syscall wrapper */
@@ -49,3 +52,13 @@ void __clear_cache(void *beginning, void *end)
  * However, there is no ARM asm parser in tcc so we use it for now */
     syscall(__ARM_NR_cacheflush, beginning, end, 0);
 }
+
+/* ------------------------------------------------------------- */
+#elif defined __aarch64__
+void __clear_cache(void *beg, void *end)
+{
+    __arm64_clear_cache(beg, end);
+}
+
+/* ------------------------------------------------------------- */
+#endif

lib/bt-dll.c

@@ -34,7 +34,11 @@
   REDIR(__bound_strncmp) \
   REDIR(__bound_strcat) \
   REDIR(__bound_strchr) \
-  REDIR(__bound_strdup)
+  REDIR(__bound_strdup) \
+  REDIR(__bound_strncat) \
+  REDIR(__bound_strrchr) \
+  REDIR(__bound_setjmp) \
+  REDIR(__bound_longjmp)
 
 #ifdef __leading_underscore
 #define _(s) "_"#s
@@ -45,11 +49,28 @@
 #define REDIR(s) void *s;
 static struct { REDIR_ALL } all_ptrs;
 #undef REDIR
+
 #define REDIR(s) #s"\0"
 static const char all_names[] = REDIR_ALL;
 #undef REDIR
-#define REDIR(s) __asm__(".global " _(s) ";" _(s) ": jmp *%0" : : "m" (all_ptrs.s) );
-static void all_jmps() { REDIR_ALL }
+
+#if __aarch64__
+# define REDIR(s) \
+    __asm__(".global "_(s)";"_(s)":"); \
+    __asm__(".int 0x58000090"); /* ldr x16, [pc, #16] */ \
+    __asm__(".int 0xf9400210"); /* ldr x16, [x16] */ \
+    __asm__(".int 0xd61f0200"); /* br x16 */ \
+    __asm__(".int 0xd503201f"); /* nop for alignment */ \
+    __asm__(".quad all_ptrs + (. - all_jmps - 16) / 24 * 8"); \
+    __asm__(".type "_(s)",function\n.size "_(s)",.-"_(s));
+
+    __asm__(".text\n.align 8\nall_jmps:");
+    REDIR_ALL
+#else
+# define REDIR(s) \
+    __asm__(".global "_(s)";"_(s)":"); goto *all_ptrs.s;
+    static void all_jmps() { REDIR_ALL }
+#endif
 #undef REDIR
 
 void __bt_init_dll(int bcheck)

lib/lib-arm64.c

@@ -23,13 +23,6 @@ typedef unsigned long long uint64_t;
 #include <string.h>
 #endif
 
-#if !defined __riscv && !defined __APPLE__
-void __clear_cache(void *beg, void *end)
-{
-    __arm64_clear_cache(beg, end);
-}
-#endif
-
 typedef union {
     struct { uint64_t x0, x1; };
     long double f;