Struct va_arg fix

lib/va_list.c:
- Handle struct {double, double} correctly

arm64-gen.c:
riscv64-gen.c:
x86_64-gen.c:
- Allow zero sized structs to work with va_arg

tcctest.c:
- Add new va_arg test code

test/bug.c:
- Remove tst2 va_arg test

arm64-gen.c

@@ -1050,11 +1050,13 @@ ST_FUNC void gfunc_call(int nb_args)
             // value in general-purpose registers
             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));
                     arm64_ldrs(a[i] / 2, size);
                 }
+            }
             else
                 gv(RC_R(a[i] / 2));
         }

lib/va_list.c

@@ -23,6 +23,8 @@ typedef struct {
 } __builtin_va_list[1];
 */
 
+extern void *memcpy(void *dest, const void *src, unsigned long n);
+
 void *__va_arg(__builtin_va_list ap,
                int arg_type,
                int size, int align)
@@ -40,9 +42,15 @@ void *__va_arg(__builtin_va_list ap,
     case __va_float_reg:
         if (ap->fp_offset < 128 + 48) {
             ap->fp_offset += 16;
+            if (size == 8)
                 return ap->reg_save_area + ap->fp_offset - 16;
+            if (ap->fp_offset < 128 + 48) {
+                memcpy(ap->reg_save_area + ap->fp_offset - 8,
+                       ap->reg_save_area + ap->fp_offset, 8);
+                ap->fp_offset += 16;
+                return ap->reg_save_area + ap->fp_offset - 32;
+            }
         }
-        size = 8;
         goto use_overflow_area;
 
     case __va_stack:

riscv64-gen.c

@@ -552,7 +552,9 @@ ST_FUNC void gfunc_call(int nb_args)
         if (!sa && align == 2*XLEN && size <= 2*XLEN)
           areg[0] = (areg[0] + 1) & ~1;
         nregs = prc[0];
-        if ((prc[1] == RC_INT && areg[0] >= 8)
+        if (size == 0)
+            info[i] = 0;
+        else if ((prc[1] == RC_INT && areg[0] >= 8)
             || (prc[1] == RC_FLOAT && areg[1] >= 16)
             || (nregs == 2 && prc[1] == RC_FLOAT && prc[2] == RC_FLOAT
                 && areg[1] >= 15)
@@ -651,6 +653,8 @@ ST_FUNC void gfunc_call(int nb_args)
             vrotb(i+1);
             origtype = vtop->type;
             size = type_size(&vtop->type, &align);
+            if (size == 0)
+                goto done;
             loadt = vtop->type.t & VT_BTYPE;
             if (loadt == VT_STRUCT) {
                 loadt = (ii >> 12) & VT_BTYPE;
@@ -701,6 +705,7 @@ ST_FUNC void gfunc_call(int nb_args)
                 EI(0x13, 0, ireg(r2), ireg(vtop->r2), 0); // mv Ra+1, RR2
                 vtop->r2 = r2;
             }
+done:
             vrott(i+1);
         }
     }

tests/bug.c

@@ -1,21 +1,6 @@
 #include <stdio.h>
 #include <stdarg.h>
 
-typedef struct{double x,y;}p;
-
-void tst2(int n,...)
-{
-  /* va_arg for struct double does not work on some targets */
-  int i;
-  va_list args;
-  va_start(args,n);
-  for (i = 0; i < n; i++) {
-    p v = va_arg(args,p);
-    if (v.x != 1 || v.y != 2) printf("%g %g\n", v.x, v.y);
-  }
-  va_end(args);
-}
-
 void tst3(void)
 {
   /* Should VT_SYM be checked for TOK_builtin_constant_p */
@@ -60,7 +45,5 @@ int compile_errors(void)
 int
 main(void)
 {
-  p v = { 1, 2};
-  tst2(1, v);
   tst3();
 }

tests/tcctest.c

@@ -2627,18 +2627,34 @@ void vprintf1(const char *fmt, ...)
 struct myspace {
     short int profile;
 };
+struct myspace2 {
+    char a[0];
+};
+struct myspace3 {
+    char a[1];
+};
+struct myspace4 {
+    char a[2];
+};
 
 void stdarg_for_struct(struct myspace bob, ...)
 {
     struct myspace george, bill;
+    struct myspace2 alex1;
+    struct myspace3 alex2;
+    struct myspace4 alex3;
     va_list ap;
     short int validate;
 
     va_start(ap, bob);
+    alex1    = va_arg(ap, struct myspace2);
+    alex2    = va_arg(ap, struct myspace3);
+    alex3    = va_arg(ap, struct myspace4);
     bill     = va_arg(ap, struct myspace);
     george   = va_arg(ap, struct myspace);
     validate = va_arg(ap, int);
-    printf("stdarg_for_struct: %d %d %d %d\n",
+    printf("stdarg_for_struct: %d %d %d %d %d %d %d\n",
+           alex2.a[0], alex3.a[0], alex3.a[1],
            bob.profile, bill.profile, george.profile, validate);
     va_end(ap);
 }
@@ -2664,10 +2680,40 @@ void stdarg_syntax(int n, ...)
     (va_end(ap));
 }
 
+typedef struct{
+    double x,y;
+} point;
+point pts[]={{1.0,2.0},{3.0,4.0},{5.0,6.0},{7.0,8.0},{9.0,10.0},{11.0,12.0}};
+
+static void stdarg_double_struct(int nargs, int posd,...)
+{
+    int i;
+    double d;
+    point pi;
+    va_list args;
+
+    printf ("stdarg_double_struct: %d\n", posd);
+    va_start(args,posd);
+    for(i = 0; i < nargs; i++) {
+        if (i == posd) {
+            d = va_arg (args, double);
+            printf ("d %d = %g\n", i, d);
+        }
+        else {
+            pi = va_arg (args, point);
+            printf ("pts[%d] = %g %g\n", i, pi.x, pi.y);
+        }
+    }
+    va_end(args);
+}
+
 void stdarg_test(void)
 {
     LONG_DOUBLE ld = 1234567891234LL;
     struct myspace bob;
+    struct myspace2 bob2;
+    struct myspace3 bob3;
+    struct myspace4 bob4;
 
     vprintf1("%d %d %d\n", 1, 2, 3);
     vprintf1("%f %d %f\n", 1.0, 2, 3.0);
@@ -2709,9 +2755,20 @@ void stdarg_test(void)
              42.0, 43.0, ld);
 
     bob.profile = 42;
-    stdarg_for_struct(bob, bob, bob, bob.profile);
+    bob3.a[0] = 1;
+    bob4.a[0] = 2;
+    bob4.a[1] = 3;
+    stdarg_for_struct(bob, bob2, bob3, bob4, bob, bob, bob.profile);
     stdarg_for_libc("stdarg_for_libc: %s %.2f %d\n", "string", 1.23, 456);
     stdarg_syntax(1, 17);
+#ifndef __riscv
+    stdarg_double_struct(6,-1,pts[0],pts[1],pts[2],pts[3],pts[4],pts[5]);
+    stdarg_double_struct(7,1,pts[0],-1.0,pts[1],pts[2],pts[3],pts[4],pts[5]);
+    stdarg_double_struct(7,2,pts[0],pts[1],-1.0,pts[2],pts[3],pts[4],pts[5]);
+    stdarg_double_struct(7,3,pts[0],pts[1],pts[2],-1.0,pts[3],pts[4],pts[5]);
+    stdarg_double_struct(7,4,pts[0],pts[1],pts[2],pts[3],-1.0,pts[4],pts[5]);
+    stdarg_double_struct(7,5,pts[0],pts[1],pts[2],pts[3],pts[4],-1.0,pts[5]);
+#endif
 }
 
 int reltab[3] = { 1, 2, 3 };

x86_64-gen.c

@@ -1244,6 +1244,7 @@ void gfunc_call(int nb_args)
     stack_adjust = 0;
     for(i = nb_args - 1; i >= 0; i--) {
         mode = classify_x86_64_arg(&vtop[-i].type, NULL, &size, &align, &reg_count);
+        if (size == 0) continue;
         if (mode == x86_64_mode_sse && nb_sse_args + reg_count <= 8) {
             nb_sse_args += reg_count;
 	    onstack[i] = 0;
@@ -1278,6 +1279,7 @@ void gfunc_call(int nb_args)
     stack_adjust &= 15;
     for (i = k = 0; i < nb_args;) {
 	mode = classify_x86_64_arg(&vtop[-i].type, NULL, &size, &align, &reg_count);
+	if (size) {
             if (!onstack[i + k]) {
 	        ++i;
 	        continue;
@@ -1293,6 +1295,7 @@ void gfunc_call(int nb_args)
             }
 	    if (onstack[i + k] == 2)
 	        stack_adjust = 1;
+        }
 
 	vrotb(i+1);
 
@@ -1357,6 +1360,7 @@ void gfunc_call(int nb_args)
     assert(sse_reg <= 8);
     for(i = 0; i < nb_args; i++) {
         mode = classify_x86_64_arg(&vtop->type, &type, &size, &align, &reg_count);
+        if (size == 0) continue;
         /* Alter stack entry type so that gv() knows how to treat it */
         vtop->type = type;
         if (mode == x86_64_mode_sse) {