Add arm64 dll support

Most support was already present.

arm64-link.c:
- create_plt_entry:
 - remove DLLs unimplemented!
- relocate:
 - Add TCC_OUTPUT_DLL for R_AARCH64_ABS64/R_AARCH64_ABS32/R_AARCH64_PREL32

tccelf.c:
- prepare_dynamic_rel:
 - Add R_AARCH64_ABS64/R_AARCH64_ABS32/R_AARCH64_PREL32
- fill_got_entry:
 - Change 'TCC_TARGET_X86_64' into 'PTR_SIZE == 8'

arm64-link.c

@@ -88,9 +88,6 @@ ST_FUNC unsigned create_plt_entry(TCCState *s1, unsigned got_offset, struct sym_
     uint8_t *p;
     unsigned plt_offset;
 
-    if (s1->output_type == TCC_OUTPUT_DLL)
-        tcc_error("DLLs unimplemented!");
-
     if (plt->data_offset == 0) {
         section_ptr_add(plt, 32);
     }
@@ -152,19 +149,54 @@ ST_FUNC void relocate_plt(TCCState *s1)
 
 void relocate(TCCState *s1, ElfW_Rel *rel, int type, unsigned char *ptr, addr_t addr, addr_t val)
 {
-    int sym_index = ELFW(R_SYM)(rel->r_info);
+    int sym_index = ELFW(R_SYM)(rel->r_info), esym_index;
 #ifdef DEBUG_RELOC
     ElfW(Sym) *sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
 #endif
 
     switch(type) {
         case R_AARCH64_ABS64:
+            if (s1->output_type == TCC_OUTPUT_DLL) {
+                esym_index = get_sym_attr(s1, sym_index, 0)->dyn_index;
+                qrel->r_offset = rel->r_offset;
+                if (esym_index) {
+                    qrel->r_info = ELFW(R_INFO)(esym_index, R_AARCH64_ABS64);
+                    qrel->r_addend = rel->r_addend;
+                    qrel++;
+                    break;
+                } else {
+                    qrel->r_info = ELFW(R_INFO)(0, R_AARCH64_RELATIVE);
+                    qrel->r_addend = read64le(ptr) + val;
+                    qrel++;
+                }
+            }
             add64le(ptr, val);
             return;
         case R_AARCH64_ABS32:
+            if (s1->output_type == TCC_OUTPUT_DLL) {
+                /* XXX: this logic may depend on TCC's codegen
+                   now TCC uses R_AARCH64_RELATIVE even for a 64bit pointer */
+                qrel->r_offset = rel->r_offset;
+                qrel->r_info = ELFW(R_INFO)(0, R_AARCH64_RELATIVE);
+                /* Use sign extension! */
+                qrel->r_addend = (int)read32le(ptr) + val;
+                qrel++;
+            }
             add32le(ptr, val);
             return;
 	case R_AARCH64_PREL32:
+            if (s1->output_type == TCC_OUTPUT_DLL) {
+                /* DLL relocation */
+                esym_index = get_sym_attr(s1, sym_index, 0)->dyn_index;
+                if (esym_index) {
+                    qrel->r_offset = rel->r_offset;
+                    qrel->r_info = ELFW(R_INFO)(esym_index, R_AARCH64_PREL32);
+                    /* Use sign extension! */
+                    qrel->r_addend = (int)read32le(ptr) + rel->r_addend;
+                    qrel++;
+                    break;
+                }
+            }
 	    write32le(ptr, val - addr);
 	    return;
         case R_AARCH64_MOVW_UABS_G0_NC:

tccelf.c

@@ -1028,7 +1028,7 @@ static void relocate_rel(TCCState *s1, Section *sr)
 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
 {
     int count = 0;
-#if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
+#if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64) || defined(TCC_TARGET_ARM64)
     ElfW_Rel *rel;
     for_each_elem(sr, 0, rel, ElfW_Rel) {
         int sym_index = ELFW(R_SYM)(rel->r_info);
@@ -1046,6 +1046,9 @@ static int prepare_dynamic_rel(TCCState *s1, Section *sr)
         case R_X86_64_32:
         case R_X86_64_32S:
         case R_X86_64_64:
+#elif defined(TCC_TARGET_ARM64)
+        case R_AARCH64_ABS32:
+        case R_AARCH64_ABS64:
 #endif
             count++;
             break;
@@ -1053,6 +1056,8 @@ static int prepare_dynamic_rel(TCCState *s1, Section *sr)
         case R_386_PC32:
 #elif defined(TCC_TARGET_X86_64)
         case R_X86_64_PC32:
+#elif defined(TCC_TARGET_ARM64)
+        case R_AARCH64_PREL32:
 #endif
             if (get_sym_attr(s1, sym_index, 0)->dyn_index)
                 count++;
@@ -1553,7 +1558,7 @@ ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
     if (0 == offset)
         return;
     section_reserve(s1->got, offset + PTR_SIZE);
-#ifdef TCC_TARGET_X86_64
+#if PTR_SIZE == 8
     write64le(s1->got->data + offset, sym->st_value);
 #else
     write32le(s1->got->data + offset, sym->st_value);