8 files changed, 591 insertions, 49 deletions

diff --git a/src/host/buildvm.c b/src/host/buildvm.c
index eb22bbbd..27f90066 100644
--- a/src/host/buildvm.c
+++ b/src/host/buildvm.c
@@ -18,8 +18,10 @@
 #include "lj_obj.h"
 #include "lj_gc.h"
 #include "lj_bc.h"
+#if LJ_HASJIT
 #include "lj_ir.h"
 #include "lj_ircall.h"
+#endif
 #include "lj_frame.h"
 #include "lj_dispatch.h"
 #if LJ_HASFFI
@@ -59,10 +61,10 @@ static int collect_reloc(BuildCtx *ctx, uint8_t *addr, int idx, int type);
 #include "../dynasm/dasm_x86.h"
 #elif LJ_TARGET_ARM
 #include "../dynasm/dasm_arm.h"
+#elif LJ_TARGET_ARM64
+#include "../dynasm/dasm_arm64.h"
 #elif LJ_TARGET_PPC
 #include "../dynasm/dasm_ppc.h"
-#elif LJ_TARGET_PPCSPE
-#include "../dynasm/dasm_ppc.h"
 #elif LJ_TARGET_MIPS
 #include "../dynasm/dasm_mips.h"
 #else
@@ -110,11 +112,11 @@ static const char *sym_decorate(BuildCtx *ctx,
   if (p) {
 #if LJ_TARGET_X86ORX64
     if (!LJ_64 && (ctx->mode == BUILD_coffasm || ctx->mode == BUILD_peobj))
-      name[0] = '@';
+      name[0] = name[1] == 'R' ? '_' : '@';  /* Just for _RtlUnwind@16. */
     else
       *p = '\0';
-#elif (LJ_TARGET_PPC  || LJ_TARGET_PPCSPE) && !LJ_TARGET_CONSOLE
-    /* Keep @plt. */
+#elif LJ_TARGET_PPC && !LJ_TARGET_CONSOLE
+    /* Keep @plt etc. */
 #else
     *p = '\0';
 #endif
@@ -179,6 +181,7 @@ static int build_code(BuildCtx *ctx)
   ctx->nreloc = 0;
 
   ctx->globnames = globnames;
+  ctx->extnames = extnames;
   ctx->relocsym = (const char **)malloc(NRELOCSYM*sizeof(const char *));
   ctx->nrelocsym = 0;
   for (i = 0; i < (int)NRELOCSYM; i++) relocmap[i] = -1;
@@ -249,6 +252,7 @@ BCDEF(BCNAME)
   NULL
 };
 
+#if LJ_HASJIT
 const char *const ir_names[] = {
 #define IRNAME(name, m, m1, m2) #name,
 IRDEF(IRNAME)
@@ -289,7 +293,9 @@ static const char *const trace_errors[] = {
 #include "lj_traceerr.h"
   NULL
 };
+#endif
 
+#if LJ_HASJIT
 static const char *lower(char *buf, const char *s)
 {
   char *p = buf;
@@ -300,6 +306,7 @@ static const char *lower(char *buf, const char *s)
   *p = '\0';
   return buf;
 }
+#endif
 
 /* Emit C source code for bytecode-related definitions. */
 static void emit_bcdef(BuildCtx *ctx)
@@ -317,24 +324,27 @@ static void emit_bcdef(BuildCtx *ctx)
 /* Emit VM definitions as Lua code for debug modules. */
 static void emit_vmdef(BuildCtx *ctx)
 {
+#if LJ_HASJIT
   char buf[80];
+#endif
   int i;
   fprintf(ctx->fp, "-- This is a generated file. DO NOT EDIT!\n\n");
   fprintf(ctx->fp, "assert(require(\"jit\").version == \"%s\", \"LuaJIT core/library version mismatch\")\n\n", LUAJIT_VERSION);
-  fprintf(ctx->fp, "module(...)\n\n");
+  fprintf(ctx->fp, "return {\n\n");
 
   fprintf(ctx->fp, "bcnames = \"");
   for (i = 0; bc_names[i]; i++) fprintf(ctx->fp, "%-6s", bc_names[i]);
-  fprintf(ctx->fp, "\"\n\n");
+  fprintf(ctx->fp, "\",\n\n");
 
+#if LJ_HASJIT
   fprintf(ctx->fp, "irnames = \"");
   for (i = 0; ir_names[i]; i++) fprintf(ctx->fp, "%-6s", ir_names[i]);
-  fprintf(ctx->fp, "\"\n\n");
+  fprintf(ctx->fp, "\",\n\n");
 
   fprintf(ctx->fp, "irfpm = { [0]=");
   for (i = 0; irfpm_names[i]; i++)
     fprintf(ctx->fp, "\"%s\", ", lower(buf, irfpm_names[i]));
-  fprintf(ctx->fp, "}\n\n");
+  fprintf(ctx->fp, "},\n\n");
 
   fprintf(ctx->fp, "irfield = { [0]=");
   for (i = 0; irfield_names[i]; i++) {
@@ -344,17 +354,18 @@ static void emit_vmdef(BuildCtx *ctx)
     if (p) *p = '.';
     fprintf(ctx->fp, "\"%s\", ", buf);
   }
-  fprintf(ctx->fp, "}\n\n");
+  fprintf(ctx->fp, "},\n\n");
 
   fprintf(ctx->fp, "ircall = {\n[0]=");
   for (i = 0; ircall_names[i]; i++)
     fprintf(ctx->fp, "\"%s\",\n", ircall_names[i]);
-  fprintf(ctx->fp, "}\n\n");
+  fprintf(ctx->fp, "},\n\n");
 
   fprintf(ctx->fp, "traceerr = {\n[0]=");
   for (i = 0; trace_errors[i]; i++)
     fprintf(ctx->fp, "\"%s\",\n", trace_errors[i]);
-  fprintf(ctx->fp, "}\n\n");
+  fprintf(ctx->fp, "},\n\n");
+#endif
 }
 
 /* -- Argument parsing ---------------------------------------------------- */
@@ -491,6 +502,7 @@ int main(int argc, char **argv)
   case BUILD_vmdef:
     emit_vmdef(ctx);
     emit_lib(ctx);
+    fprintf(ctx->fp, "}\n\n");
     break;
   case BUILD_ffdef:
   case BUILD_libdef:
diff --git a/src/host/buildvm.h b/src/host/buildvm.h
index 9fe2f0b6..cfdadc2c 100644
--- a/src/host/buildvm.h
+++ b/src/host/buildvm.h
@@ -82,6 +82,7 @@ typedef struct BuildCtx {
   const char *beginsym;
   /* Strings generated by DynASM. */
   const char *const *globnames;
+  const char *const *extnames;
   const char *dasm_ident;
   const char *dasm_arch;
   /* Relocations. */
diff --git a/src/host/buildvm_asm.c b/src/host/buildvm_asm.c
index 7a059762..8cd8727b 100644
--- a/src/host/buildvm_asm.c
+++ b/src/host/buildvm_asm.c
@@ -51,8 +51,8 @@ static const char *const jccnames[] = {
   "js", "jns", "jpe", "jpo", "jl", "jge", "jle", "jg"
 };
 
-/* Emit relocation for the incredibly stupid OSX assembler. */
-static void emit_asm_reloc_mach(BuildCtx *ctx, uint8_t *cp, int n,
+/* Emit x86/x64 text relocations. */
+static void emit_asm_reloc_text(BuildCtx *ctx, uint8_t *cp, int n,
                                 const char *sym)
 {
   const char *opname = NULL;
@@ -71,6 +71,20 @@ err:
     exit(1);
   }
   emit_asm_bytes(ctx, cp, n);
+  if (strncmp(sym+(*sym == '_'), LABEL_PREFIX, sizeof(LABEL_PREFIX)-1)) {
+    /* Various fixups for external symbols outside of our binary. */
+    if (ctx->mode == BUILD_elfasm) {
+      if (LJ_32)
+        fprintf(ctx->fp, "#if __PIC__\n\t%s lj_wrap_%s\n#else\n", opname, sym);
+      fprintf(ctx->fp, "\t%s %s@PLT\n", opname, sym);
+      if (LJ_32)
+        fprintf(ctx->fp, "#endif\n");
+      return;
+    } else if (LJ_32 && ctx->mode == BUILD_machasm) {
+      fprintf(ctx->fp, "\t%s L%s$stub\n", opname, sym);
+      return;
+    }
+  }
   fprintf(ctx->fp, "\t%s %s\n", opname, sym);
 }
 #else
@@ -79,10 +93,14 @@ static void emit_asm_words(BuildCtx *ctx, uint8_t *p, int n)
 {
   int i;
   for (i = 0; i < n; i += 4) {
+    uint32_t ins = *(uint32_t *)(p+i);
+#if LJ_TARGET_ARM64 && LJ_BE
+    ins = lj_bswap(ins);  /* ARM64 instructions are always little-endian. */
+#endif
     if ((i & 15) == 0)
-      fprintf(ctx->fp, "\t.long 0x%08x", *(uint32_t *)(p+i));
+      fprintf(ctx->fp, "\t.long 0x%08x", ins);
     else
-      fprintf(ctx->fp, ",0x%08x", *(uint32_t *)(p+i));
+      fprintf(ctx->fp, ",0x%08x", ins);
     if ((i & 15) == 12) putc('\n', ctx->fp);
   }
   if ((n & 15) != 0) putc('\n', ctx->fp);
@@ -107,7 +125,16 @@ static void emit_asm_wordreloc(BuildCtx *ctx, uint8_t *p, int n,
             ins, sym);
     exit(1);
   }
-#elif LJ_TARGET_PPC || LJ_TARGET_PPCSPE
+#elif LJ_TARGET_ARM64
+  if ((ins >> 26) == 0x25u) {
+    fprintf(ctx->fp, "\tbl %s\n", sym);
+  } else {
+    fprintf(stderr,
+            "Error: unsupported opcode %08x for %s symbol relocation.\n",
+            ins, sym);
+    exit(1);
+  }
+#elif LJ_TARGET_PPC
 #if LJ_TARGET_PS3
 #define TOCPREFIX "."
 #else
@@ -216,6 +243,12 @@ void emit_asm(BuildCtx *ctx)
 
   fprintf(ctx->fp, "\t.file \"buildvm_%s.dasc\"\n", ctx->dasm_arch);
   fprintf(ctx->fp, "\t.text\n");
+#if LJ_TARGET_MIPS32 && !LJ_ABI_SOFTFP
+  fprintf(ctx->fp, "\t.module fp=32\n");
+#endif
+#if LJ_TARGET_MIPS
+  fprintf(ctx->fp, "\t.set nomips16\n\t.abicalls\n\t.set noreorder\n\t.set nomacro\n");
+#endif
   emit_asm_align(ctx, 4);
 
 #if LJ_TARGET_PS3
@@ -228,13 +261,19 @@ void emit_asm(BuildCtx *ctx)
 
 #if LJ_TARGET_ARM && defined(__GNUC__) && !LJ_NO_UNWIND
   /* This should really be moved into buildvm_arm.dasc. */
+#if LJ_ARCH_HASFPU
+  fprintf(ctx->fp,
+          ".fnstart\n"
+          ".save {r5, r6, r7, r8, r9, r10, r11, lr}\n"
+          ".vsave {d8-d15}\n"
+          ".save {r4}\n"
+          ".pad #28\n");
+#else
   fprintf(ctx->fp,
           ".fnstart\n"
           ".save {r4, r5, r6, r7, r8, r9, r10, r11, lr}\n"
           ".pad #28\n");
 #endif
-#if LJ_TARGET_MIPS
-  fprintf(ctx->fp, ".set nomips16\n.abicalls\n.set noreorder\n.set nomacro\n");
 #endif
 
   for (i = rel = 0; i < ctx->nsym; i++) {
@@ -255,8 +294,9 @@ void emit_asm(BuildCtx *ctx)
       BuildReloc *r = &ctx->reloc[rel];
       int n = r->ofs - ofs;
 #if LJ_TARGET_X86ORX64
-      if (ctx->mode == BUILD_machasm && r->type != 0) {
-        emit_asm_reloc_mach(ctx, ctx->code+ofs, n, ctx->relocsym[r->sym]);
+      if (r->type != 0 &&
+          (ctx->mode == BUILD_elfasm || ctx->mode == BUILD_machasm)) {
+        emit_asm_reloc_text(ctx, ctx->code+ofs, n, ctx->relocsym[r->sym]);
       } else {
         emit_asm_bytes(ctx, ctx->code+ofs, n);
         emit_asm_reloc(ctx, r->type, ctx->relocsym[r->sym]);
@@ -290,10 +330,7 @@ void emit_asm(BuildCtx *ctx)
 #if !(LJ_TARGET_PS3 || LJ_TARGET_PSVITA)
     fprintf(ctx->fp, "\t.section .note.GNU-stack,\"\"," ELFASM_PX "progbits\n");
 #endif
-#if LJ_TARGET_PPCSPE
-    /* Soft-float ABI + SPE. */
-    fprintf(ctx->fp, "\t.gnu_attribute 4, 2\n\t.gnu_attribute 8, 3\n");
-#elif LJ_TARGET_PPC && !LJ_TARGET_PS3
+#if LJ_TARGET_PPC && !LJ_TARGET_PS3 && !LJ_ABI_SOFTFP
     /* Hard-float ABI. */
     fprintf(ctx->fp, "\t.gnu_attribute 4, 1\n");
 #endif
@@ -302,6 +339,10 @@ void emit_asm(BuildCtx *ctx)
     fprintf(ctx->fp, "\t.ident \"%s\"\n", ctx->dasm_ident);
     break;
   case BUILD_machasm:
+#if defined(__apple_build_version__) && __apple_build_version__ >= 15000000 && __apple_build_version__ < 15000300
+    /* Workaround for XCode 15.0 - 15.2. */
+    fprintf(ctx->fp, "\t.subsections_via_symbols\n");
+#endif
     fprintf(ctx->fp,
       "\t.cstring\n"
       "\t.ascii \"%s\\0\"\n", ctx->dasm_ident);
diff --git a/src/host/buildvm_fold.c b/src/host/buildvm_fold.c
index 3aa445ac..e0b19e8c 100644
--- a/src/host/buildvm_fold.c
+++ b/src/host/buildvm_fold.c
@@ -5,6 +5,7 @@
 
 #include "buildvm.h"
 #include "lj_obj.h"
+#if LJ_HASJIT
 #include "lj_ir.h"
 
 /* Context for the folding hash table generator. */
@@ -226,4 +227,10 @@ void emit_fold(BuildCtx *ctx)
 
   makehash(ctx);
 }
+#else
+void emit_fold(BuildCtx *ctx)
+{
+  UNUSED(ctx);
+}
+#endif
 
diff --git a/src/host/buildvm_lib.c b/src/host/buildvm_lib.c
index 7147c2c8..51ac5f12 100644
--- a/src/host/buildvm_lib.c
+++ b/src/host/buildvm_lib.c
@@ -5,7 +5,9 @@
 
 #include "buildvm.h"
 #include "lj_obj.h"
+#include "lj_bc.h"
 #include "lj_lib.h"
+#include "buildvm_libbc.h"
 
 /* Context for library definitions. */
 static uint8_t obuf[8192];
@@ -151,6 +153,62 @@ static void libdef_func(BuildCtx *ctx, char *p, int arg)
   regfunc = REGFUNC_OK;
 }
 
+static uint8_t *libdef_uleb128(uint8_t *p, uint32_t *vv)
+{
+  uint32_t v = *p++;
+  if (v >= 0x80) {
+    int sh = 0; v &= 0x7f;
+    do { v |= ((*p & 0x7f) << (sh += 7)); } while (*p++ >= 0x80);
+  }
+  *vv = v;
+  return p;
+}
+
+static void libdef_fixupbc(uint8_t *p)
+{
+  uint32_t i, sizebc;
+  p += 4;
+  p = libdef_uleb128(p, &sizebc);
+  p = libdef_uleb128(p, &sizebc);
+  p = libdef_uleb128(p, &sizebc);
+  for (i = 0; i < sizebc; i++, p += 4) {
+    uint8_t op = p[libbc_endian ? 3 : 0];
+    uint8_t ra = p[libbc_endian ? 2 : 1];
+    uint8_t rc = p[libbc_endian ? 1 : 2];
+    uint8_t rb = p[libbc_endian ? 0 : 3];
+    if (!LJ_DUALNUM && op == BC_ISTYPE && rc == ~LJ_TNUMX+1) {
+      op = BC_ISNUM; rc++;
+    }
+    p[LJ_ENDIAN_SELECT(0, 3)] = op;
+    p[LJ_ENDIAN_SELECT(1, 2)] = ra;
+    p[LJ_ENDIAN_SELECT(2, 1)] = rc;
+    p[LJ_ENDIAN_SELECT(3, 0)] = rb;
+  }
+}
+
+static void libdef_lua(BuildCtx *ctx, char *p, int arg)
+{
+  UNUSED(arg);
+  if (ctx->mode == BUILD_libdef) {
+    int i;
+    for (i = 0; libbc_map[i].name != NULL; i++) {
+      if (!strcmp(libbc_map[i].name, p)) {
+        int ofs = libbc_map[i].ofs;
+        int len = libbc_map[i+1].ofs - ofs;
+        obuf[2]++;  /* Bump hash table size. */
+        *optr++ = LIBINIT_LUA;
+        libdef_name(p, 0);
+        memcpy(optr, libbc_code + ofs, len);
+        libdef_fixupbc(optr);
+        optr += len;
+        return;
+      }
+    }
+    fprintf(stderr, "Error: missing libbc definition for %s\n", p);
+    exit(1);
+  }
+}
+
 static uint32_t find_rec(char *name)
 {
   char *p = (char *)obuf;
@@ -277,6 +335,7 @@ static const LibDefHandler libdef_handlers[] = {
   { "CF(",      ")",            libdef_func,            LIBINIT_CF },
   { "ASM(",     ")",            libdef_func,            LIBINIT_ASM },
   { "ASM_(",    ")",            libdef_func,            LIBINIT_ASM_ },
+  { "LUA(",     ")",            libdef_lua,             0 },
   { "REC(",     ")",            libdef_rec,             0 },
   { "PUSH(",    ")",            libdef_push,            0 },
   { "SET(",     ")",            libdef_set,             0 },
@@ -333,6 +392,8 @@ void emit_lib(BuildCtx *ctx)
           ok = LJ_HASJIT;
         else if (!strcmp(buf, "#if LJ_HASFFI"))
           ok = LJ_HASFFI;
+        else if (!strcmp(buf, "#if LJ_HASBUFFER"))
+          ok = LJ_HASBUFFER;
         if (!ok) {
           int lvl = 1;
           while (fgets(buf, sizeof(buf), fp) != NULL) {
@@ -380,7 +441,7 @@ void emit_lib(BuildCtx *ctx)
       "#ifndef FF_NUM_ASMFUNC\n#define FF_NUM_ASMFUNC %d\n#endif\n\n",
       ffasmfunc);
   } else if (ctx->mode == BUILD_vmdef) {
-    fprintf(ctx->fp, "}\n\n");
+    fprintf(ctx->fp, "},\n\n");
   } else if (ctx->mode == BUILD_bcdef) {
     int i;
     fprintf(ctx->fp, "\n};\n\n");
diff --git a/src/host/buildvm_libbc.h b/src/host/buildvm_libbc.h
new file mode 100644
index 00000000..276463b2
--- /dev/null
+++ b/src/host/buildvm_libbc.h
@@ -0,0 +1,81 @@
+/* This is a generated file. DO NOT EDIT! */
+
+static const int libbc_endian = 0;
+
+static const uint8_t libbc_code[] = {
+#if LJ_FR2
+/* math.deg */ 0,1,2,0,0,1,2,BC_MULVN,1,0,0,BC_RET1,1,2,0,241,135,158,166,3,
+220,203,178,130,4,
+/* math.rad */ 0,1,2,0,0,1,2,BC_MULVN,1,0,0,BC_RET1,1,2,0,243,244,148,165,20,
+198,190,199,252,3,
+/* string.len */ 0,1,2,0,0,0,3,BC_ISTYPE,0,5,0,BC_LEN,1,0,0,BC_RET1,1,2,0,
+/* table.foreachi */ 0,2,10,0,0,0,15,BC_ISTYPE,0,12,0,BC_ISTYPE,1,9,0,
+BC_KSHORT,2,1,0,BC_LEN,3,0,0,BC_KSHORT,4,1,0,BC_FORI,2,8,128,BC_MOV,6,1,0,
+BC_MOV,8,5,0,BC_TGETR,9,5,0,BC_CALL,6,3,2,BC_ISEQP,6,0,0,BC_JMP,7,1,128,
+BC_RET1,6,2,0,BC_FORL,2,248,127,BC_RET0,0,1,0,
+/* table.foreach */ 0,2,11,0,0,1,16,BC_ISTYPE,0,12,0,BC_ISTYPE,1,9,0,BC_KPRI,
+2,0,0,BC_MOV,3,0,0,BC_KNUM,4,0,0,BC_JMP,5,7,128,BC_MOV,7,1,0,BC_MOV,9,5,0,
+BC_MOV,10,6,0,BC_CALL,7,3,2,BC_ISEQP,7,0,0,BC_JMP,8,1,128,BC_RET1,7,2,0,
+BC_ITERN,5,3,3,BC_ITERL,5,247,127,BC_RET0,0,1,0,1,255,255,249,255,15,
+/* table.getn */ 0,1,2,0,0,0,3,BC_ISTYPE,0,12,0,BC_LEN,1,0,0,BC_RET1,1,2,0,
+/* table.remove */ 0,2,10,0,0,2,30,BC_ISTYPE,0,12,0,BC_LEN,2,0,0,BC_ISNEP,1,0,
+0,BC_JMP,3,7,128,BC_ISEQN,2,0,0,BC_JMP,3,23,128,BC_TGETR,3,2,0,BC_KPRI,4,0,0,
+BC_TSETR,4,2,0,BC_RET1,3,2,0,BC_JMP,3,18,128,BC_ISTYPE,1,14,0,BC_KSHORT,3,1,0,
+BC_ISGT,3,1,0,BC_JMP,3,14,128,BC_ISGT,1,2,0,BC_JMP,3,12,128,BC_TGETR,3,1,0,
+BC_ADDVN,4,1,1,BC_MOV,5,2,0,BC_KSHORT,6,1,0,BC_FORI,4,4,128,BC_SUBVN,8,1,7,
+BC_TGETR,9,7,0,BC_TSETR,9,8,0,BC_FORL,4,252,127,BC_KPRI,4,0,0,BC_TSETR,4,2,0,
+BC_RET1,3,2,0,BC_RET0,0,1,0,0,2,
+/* table.move */ 0,5,12,0,0,0,35,BC_ISTYPE,0,12,0,BC_ISTYPE,1,14,0,BC_ISTYPE,
+2,14,0,BC_ISTYPE,3,14,0,BC_ISNEP,4,0,0,BC_JMP,5,1,128,BC_MOV,4,0,0,BC_ISTYPE,
+4,12,0,BC_ISGT,1,2,0,BC_JMP,5,24,128,BC_SUBVV,5,1,3,BC_ISLT,2,3,0,BC_JMP,6,4,
+128,BC_ISLE,3,1,0,BC_JMP,6,2,128,BC_ISEQV,4,0,0,BC_JMP,6,9,128,BC_MOV,6,1,0,
+BC_MOV,7,2,0,BC_KSHORT,8,1,0,BC_FORI,6,4,128,BC_ADDVV,10,5,9,BC_TGETR,11,9,0,
+BC_TSETR,11,10,4,BC_FORL,6,252,127,BC_JMP,6,8,128,BC_MOV,6,2,0,BC_MOV,7,1,0,
+BC_KSHORT,8,255,255,BC_FORI,6,4,128,BC_ADDVV,10,5,9,BC_TGETR,11,9,0,BC_TSETR,
+11,10,4,BC_FORL,6,252,127,BC_RET1,4,2,0,
+#else
+/* math.deg */ 0,1,2,0,0,1,2,BC_MULVN,1,0,0,BC_RET1,1,2,0,241,135,158,166,3,
+220,203,178,130,4,
+/* math.rad */ 0,1,2,0,0,1,2,BC_MULVN,1,0,0,BC_RET1,1,2,0,243,244,148,165,20,
+198,190,199,252,3,
+/* string.len */ 0,1,2,0,0,0,3,BC_ISTYPE,0,5,0,BC_LEN,1,0,0,BC_RET1,1,2,0,
+/* table.foreachi */ 0,2,9,0,0,0,15,BC_ISTYPE,0,12,0,BC_ISTYPE,1,9,0,
+BC_KSHORT,2,1,0,BC_LEN,3,0,0,BC_KSHORT,4,1,0,BC_FORI,2,8,128,BC_MOV,6,1,0,
+BC_MOV,7,5,0,BC_TGETR,8,5,0,BC_CALL,6,3,2,BC_ISEQP,6,0,0,BC_JMP,7,1,128,
+BC_RET1,6,2,0,BC_FORL,2,248,127,BC_RET0,0,1,0,
+/* table.foreach */ 0,2,10,0,0,1,16,BC_ISTYPE,0,12,0,BC_ISTYPE,1,9,0,BC_KPRI,
+2,0,0,BC_MOV,3,0,0,BC_KNUM,4,0,0,BC_JMP,5,7,128,BC_MOV,7,1,0,BC_MOV,8,5,0,
+BC_MOV,9,6,0,BC_CALL,7,3,2,BC_ISEQP,7,0,0,BC_JMP,8,1,128,BC_RET1,7,2,0,
+BC_ITERN,5,3,3,BC_ITERL,5,247,127,BC_RET0,0,1,0,1,255,255,249,255,15,
+/* table.getn */ 0,1,2,0,0,0,3,BC_ISTYPE,0,12,0,BC_LEN,1,0,0,BC_RET1,1,2,0,
+/* table.remove */ 0,2,10,0,0,2,30,BC_ISTYPE,0,12,0,BC_LEN,2,0,0,BC_ISNEP,1,0,
+0,BC_JMP,3,7,128,BC_ISEQN,2,0,0,BC_JMP,3,23,128,BC_TGETR,3,2,0,BC_KPRI,4,0,0,
+BC_TSETR,4,2,0,BC_RET1,3,2,0,BC_JMP,3,18,128,BC_ISTYPE,1,14,0,BC_KSHORT,3,1,0,
+BC_ISGT,3,1,0,BC_JMP,3,14,128,BC_ISGT,1,2,0,BC_JMP,3,12,128,BC_TGETR,3,1,0,
+BC_ADDVN,4,1,1,BC_MOV,5,2,0,BC_KSHORT,6,1,0,BC_FORI,4,4,128,BC_SUBVN,8,1,7,
+BC_TGETR,9,7,0,BC_TSETR,9,8,0,BC_FORL,4,252,127,BC_KPRI,4,0,0,BC_TSETR,4,2,0,
+BC_RET1,3,2,0,BC_RET0,0,1,0,0,2,
+/* table.move */ 0,5,12,0,0,0,35,BC_ISTYPE,0,12,0,BC_ISTYPE,1,14,0,BC_ISTYPE,
+2,14,0,BC_ISTYPE,3,14,0,BC_ISNEP,4,0,0,BC_JMP,5,1,128,BC_MOV,4,0,0,BC_ISTYPE,
+4,12,0,BC_ISGT,1,2,0,BC_JMP,5,24,128,BC_SUBVV,5,1,3,BC_ISLT,2,3,0,BC_JMP,6,4,
+128,BC_ISLE,3,1,0,BC_JMP,6,2,128,BC_ISEQV,4,0,0,BC_JMP,6,9,128,BC_MOV,6,1,0,
+BC_MOV,7,2,0,BC_KSHORT,8,1,0,BC_FORI,6,4,128,BC_ADDVV,10,5,9,BC_TGETR,11,9,0,
+BC_TSETR,11,10,4,BC_FORL,6,252,127,BC_JMP,6,8,128,BC_MOV,6,2,0,BC_MOV,7,1,0,
+BC_KSHORT,8,255,255,BC_FORI,6,4,128,BC_ADDVV,10,5,9,BC_TGETR,11,9,0,BC_TSETR,
+11,10,4,BC_FORL,6,252,127,BC_RET1,4,2,0,
+#endif
+0
+};
+
+static const struct { const char *name; int ofs; } libbc_map[] = {
+{"math_deg",0},
+{"math_rad",25},
+{"string_len",50},
+{"table_foreachi",69},
+{"table_foreach",136},
+{"table_getn",213},
+{"table_remove",232},
+{"table_move",361},
+{NULL,508}
+};
+
diff --git a/src/host/buildvm_peobj.c b/src/host/buildvm_peobj.c
index 7cbe1438..247b9d2d 100644
--- a/src/host/buildvm_peobj.c
+++ b/src/host/buildvm_peobj.c
@@ -9,7 +9,7 @@
 #include "buildvm.h"
 #include "lj_bc.h"
 
-#if LJ_TARGET_X86ORX64 || LJ_TARGET_PPC
+#if LJ_TARGET_WINDOWS || LJ_TARGET_CYGWIN
 
 /* Context for PE object emitter. */
 static char *strtab;
@@ -93,12 +93,17 @@ typedef struct PEsymaux {
 #define PEOBJ_RELOC_ADDR32NB    0x03
 #define PEOBJ_RELOC_OFS         0
 #define PEOBJ_TEXT_FLAGS        0x60500020  /* 60=r+x, 50=align16, 20=code. */
-#elif LJ_TARGET_PPC
-#define PEOBJ_ARCH_TARGET       0x01f2
-#define PEOBJ_RELOC_REL32       0x06
-#define PEOBJ_RELOC_DIR32       0x02
+#define PEOBJ_PDATA_NRELOC      6
+#define PEOBJ_XDATA_SIZE        (8*2+4+6*2)
+#elif LJ_TARGET_ARM64
+#define PEOBJ_ARCH_TARGET       0xaa64
+#define PEOBJ_RELOC_REL32       0x03  /* MS: BRANCH26. */
+#define PEOBJ_RELOC_DIR32       0x01
+#define PEOBJ_RELOC_ADDR32NB    0x02
 #define PEOBJ_RELOC_OFS         (-4)
-#define PEOBJ_TEXT_FLAGS        0x60400020  /* 60=r+x, 40=align8, 20=code. */
+#define PEOBJ_TEXT_FLAGS        0x60500020  /* 60=r+x, 50=align16, 20=code. */
+#define PEOBJ_PDATA_NRELOC      4
+#define PEOBJ_XDATA_SIZE        (4+24+4 +4+8)
 #endif
 
 /* Section numbers (0-based). */
@@ -106,9 +111,11 @@ enum {
   PEOBJ_SECT_ABS = -2,
   PEOBJ_SECT_UNDEF = -1,
   PEOBJ_SECT_TEXT,
-#if LJ_TARGET_X64
+#ifdef PEOBJ_PDATA_NRELOC
   PEOBJ_SECT_PDATA,
   PEOBJ_SECT_XDATA,
+#elif LJ_TARGET_X86
+  PEOBJ_SECT_SXDATA,
 #endif
   PEOBJ_SECT_RDATA_Z,
   PEOBJ_NSECTIONS
@@ -179,6 +186,9 @@ void emit_peobj(BuildCtx *ctx)
   uint32_t sofs;
   int i, nrsym;
   union { uint8_t b; uint32_t u; } host_endian;
+#ifdef PEOBJ_PDATA_NRELOC
+  uint32_t fcofs = (uint32_t)ctx->sym[ctx->nsym-1].ofs;
+#endif
 
   sofs = sizeof(PEheader) + PEOBJ_NSECTIONS*sizeof(PEsection);
 
@@ -192,22 +202,29 @@ void emit_peobj(BuildCtx *ctx)
   /* Flags: 60 = read+execute, 50 = align16, 20 = code. */
   pesect[PEOBJ_SECT_TEXT].flags = PEOBJ_TEXT_FLAGS;
 
-#if LJ_TARGET_X64
+#ifdef PEOBJ_PDATA_NRELOC
   memcpy(pesect[PEOBJ_SECT_PDATA].name, ".pdata", sizeof(".pdata")-1);
   pesect[PEOBJ_SECT_PDATA].ofs = sofs;
-  sofs += (pesect[PEOBJ_SECT_PDATA].size = 6*4);
+  sofs += (pesect[PEOBJ_SECT_PDATA].size = PEOBJ_PDATA_NRELOC*4);
   pesect[PEOBJ_SECT_PDATA].relocofs = sofs;
-  sofs += (pesect[PEOBJ_SECT_PDATA].nreloc = 6) * PEOBJ_RELOC_SIZE;
+  sofs += (pesect[PEOBJ_SECT_PDATA].nreloc = PEOBJ_PDATA_NRELOC) * PEOBJ_RELOC_SIZE;
   /* Flags: 40 = read, 30 = align4, 40 = initialized data. */
   pesect[PEOBJ_SECT_PDATA].flags = 0x40300040;
 
   memcpy(pesect[PEOBJ_SECT_XDATA].name, ".xdata", sizeof(".xdata")-1);
   pesect[PEOBJ_SECT_XDATA].ofs = sofs;
-  sofs += (pesect[PEOBJ_SECT_XDATA].size = 8*2+4+6*2);  /* See below. */
+  sofs += (pesect[PEOBJ_SECT_XDATA].size = PEOBJ_XDATA_SIZE);  /* See below. */
   pesect[PEOBJ_SECT_XDATA].relocofs = sofs;
   sofs += (pesect[PEOBJ_SECT_XDATA].nreloc = 1) * PEOBJ_RELOC_SIZE;
   /* Flags: 40 = read, 30 = align4, 40 = initialized data. */
   pesect[PEOBJ_SECT_XDATA].flags = 0x40300040;
+#elif LJ_TARGET_X86
+  memcpy(pesect[PEOBJ_SECT_SXDATA].name, ".sxdata", sizeof(".sxdata")-1);
+  pesect[PEOBJ_SECT_SXDATA].ofs = sofs;
+  sofs += (pesect[PEOBJ_SECT_SXDATA].size = 4);
+  pesect[PEOBJ_SECT_SXDATA].relocofs = sofs;
+  /* Flags: 40 = read, 30 = align4, 02 = lnk_info, 40 = initialized data. */
+  pesect[PEOBJ_SECT_SXDATA].flags = 0x40300240;
 #endif
 
   memcpy(pesect[PEOBJ_SECT_RDATA_Z].name, ".rdata$Z", sizeof(".rdata$Z")-1);
@@ -231,8 +248,8 @@ void emit_peobj(BuildCtx *ctx)
   */
   nrsym = ctx->nrelocsym;
   pehdr.nsyms = 1+PEOBJ_NSECTIONS*2 + 1+ctx->nsym + nrsym;
-#if LJ_TARGET_X64
-  pehdr.nsyms += 1;  /* Symbol for lj_err_unwind_win64. */
+#ifdef PEOBJ_PDATA_NRELOC
+  pehdr.nsyms += 1;  /* Symbol for lj_err_unwind_win. */
 #endif
 
   /* Write PE object header and all sections. */
@@ -242,15 +259,8 @@ void emit_peobj(BuildCtx *ctx)
   /* Write .text section. */
   host_endian.u = 1;
   if (host_endian.b != LJ_ENDIAN_SELECT(1, 0)) {
-#if LJ_TARGET_PPC
-    uint32_t *p = (uint32_t *)ctx->code;
-    int n = (int)(ctx->codesz >> 2);
-    for (i = 0; i < n; i++, p++)
-      *p = lj_bswap(*p);  /* Byteswap .text section. */
-#else
     fprintf(stderr, "Error: different byte order for host and target\n");
     exit(1);
-#endif
   }
   owrite(ctx, ctx->code, ctx->codesz);
   for (i = 0; i < ctx->nreloc; i++) {
@@ -263,7 +273,6 @@ void emit_peobj(BuildCtx *ctx)
 
 #if LJ_TARGET_X64
   { /* Write .pdata section. */
-    uint32_t fcofs = (uint32_t)ctx->sym[ctx->nsym-1].ofs;
     uint32_t pdata[3];  /* Start of .text, end of .text and .xdata. */
     PEreloc reloc;
     pdata[0] = 0; pdata[1] = fcofs; pdata[2] = 0;
@@ -312,6 +321,100 @@ void emit_peobj(BuildCtx *ctx)
     reloc.type = PEOBJ_RELOC_ADDR32NB;
     owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
   }
+#elif LJ_TARGET_ARM64
+  /* https://learn.microsoft.com/en-us/cpp/build/arm64-exception-handling */
+  { /* Write .pdata section. */
+    uint32_t pdata[4];
+    PEreloc reloc;
+    pdata[0] = 0;
+    pdata[1] = 0;
+    pdata[2] = fcofs;
+    pdata[3] = 4+24+4;
+    owrite(ctx, &pdata, sizeof(pdata));
+    /* Start of .text and start of .xdata. */
+    reloc.vaddr = 0; reloc.symidx = 1+2+nrsym+2+2+1;
+    reloc.type = PEOBJ_RELOC_ADDR32NB;
+    owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
+    reloc.vaddr = 4; reloc.symidx = 1+2+nrsym+2;
+    reloc.type = PEOBJ_RELOC_ADDR32NB;
+    owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
+    /* Start of vm_ffi_call and start of second part of .xdata. */
+    reloc.vaddr = 8; reloc.symidx = 1+2+nrsym+2+2+1;
+    reloc.type = PEOBJ_RELOC_ADDR32NB;
+    owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
+    reloc.vaddr = 12; reloc.symidx = 1+2+nrsym+2;
+    reloc.type = PEOBJ_RELOC_ADDR32NB;
+    owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
+  }
+  { /* Write .xdata section. */
+    uint32_t u32;
+    uint8_t *p, uwc[24];
+    PEreloc reloc;
+
+#define CBE16(x)        (*p = ((x) >> 8) & 0xff, p[1] = (x) & 0xff, p += 2)
+#define CALLOC_S(s)     (*p++ = ((s) >> 4))  /* s < 512 */
+#define CSAVE_FPLR(o)   (*p++ = 0x40 | ((o) >> 3))  /* o <= 504 */
+#define CSAVE_REGP(r,o) CBE16(0xc800 | (((r) - 19) << 6) | ((o) >> 3))
+#define CSAVE_REGS(r1,r2,o1) do { \
+  int r, o; for (r = r1, o = o1; r <= r2; r += 2, o -= 16) CSAVE_REGP(r, o); \
+} while (0)
+#define CSAVE_REGPX(r,o) CBE16(0xcc00 | (((r) - 19) << 6) | (~(o) >> 3))
+#define CSAVE_FREGP(r,o) CBE16(0xd800 | (((r) - 8) << 6) | ((o) >> 3))
+#define CSAVE_FREGS(r1,r2,o1) do { \
+  int r, o; for (r = r1, o = o1; r <= r2; r += 2, o -= 16) CSAVE_FREGP(r, o); \
+} while (0)
+#define CADD_FP(s)      CBE16(0xe200 | ((s) >> 3))  /* s < 8*256 */
+#define CODE_NOP        0xe3
+#define CODE_END        0xe4
+#define CEND_ALIGN      do { \
+  *p++ = CODE_END; \
+  while ((p - uwc) & 3) *p++ = CODE_NOP; \
+} while (0)
+
+    /* Unwind codes for .text section with handler. */
+    p = uwc;
+    CADD_FP(192);               /* +2 */
+    CSAVE_REGS(19, 28, 176);    /* +5*2 */
+    CSAVE_FREGS(8, 15, 96);     /* +4*2 */
+    CSAVE_FPLR(192);            /* +1 */
+    CALLOC_S(208);              /* +1 */
+    CEND_ALIGN;                 /* +1 +1 -> 24 */
+
+    u32 = ((24u >> 2) << 27) | (1u << 20) | (fcofs >> 2);
+    owrite(ctx, &u32, 4);
+    owrite(ctx, &uwc, 24);
+
+    u32 = 0;  /* Handler RVA to be relocated at 4 + 24. */
+    owrite(ctx, &u32, 4);
+
+    /* Unwind codes for vm_ffi_call without handler. */
+    p = uwc;
+    CADD_FP(16);                /* +2 */
+    CSAVE_FPLR(16);             /* +1 */
+    CSAVE_REGPX(19, -32);       /* +2 */
+    CEND_ALIGN;                 /* +1 +2 -> 8 */
+
+    u32 = ((8u >> 2) << 27) | (((uint32_t)ctx->codesz - fcofs) >> 2);
+    owrite(ctx, &u32, 4);
+    owrite(ctx, &uwc, 8);
+
+    reloc.vaddr = 4 + 24; reloc.symidx = 1+2+nrsym+2+2;
+    reloc.type = PEOBJ_RELOC_ADDR32NB;
+    owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
+  }
+#elif LJ_TARGET_X86
+  /* Write .sxdata section. */
+  for (i = 0; i < nrsym; i++) {
+    if (!strcmp(ctx->relocsym[i], "_lj_err_unwind_win")) {
+      uint32_t symidx = 1+2+i;
+      owrite(ctx, &symidx, 4);
+      break;
+    }
+  }
+  if (i == nrsym) {
+    fprintf(stderr, "Error: extern lj_err_unwind_win not used\n");
+    exit(1);
+  }
 #endif
 
   /* Write .rdata$Z section. */
@@ -330,11 +433,13 @@ void emit_peobj(BuildCtx *ctx)
       emit_peobj_sym(ctx, ctx->relocsym[i], 0,
                      PEOBJ_SECT_UNDEF, PEOBJ_TYPE_FUNC, PEOBJ_SCL_EXTERN);
 
-#if LJ_TARGET_X64
+#ifdef PEOBJ_PDATA_NRELOC
     emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_PDATA);
     emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_XDATA);
-    emit_peobj_sym(ctx, "lj_err_unwind_win64", 0,
+    emit_peobj_sym(ctx, "lj_err_unwind_win", 0,
                    PEOBJ_SECT_UNDEF, PEOBJ_TYPE_FUNC, PEOBJ_SCL_EXTERN);
+#elif LJ_TARGET_X86
+    emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_SXDATA);
 #endif
 
     emit_peobj_sym(ctx, ctx->beginsym, 0,
diff --git a/src/host/genlibbc.lua b/src/host/genlibbc.lua
new file mode 100644
index 00000000..3eb1b827
--- /dev/null
+++ b/src/host/genlibbc.lua
@@ -0,0 +1,234 @@
+----------------------------------------------------------------------------
+-- Lua script to dump the bytecode of the library functions written in Lua.
+-- The resulting 'buildvm_libbc.h' is used for the build process of LuaJIT.
+----------------------------------------------------------------------------
+-- Copyright (C) 2005-2026 Mike Pall. All rights reserved.
+-- Released under the MIT license. See Copyright Notice in luajit.h
+----------------------------------------------------------------------------
+
+local ffi = require("ffi")
+local bit = require("bit")
+local vmdef = require("jit.vmdef")
+local bcnames = vmdef.bcnames
+
+local format = string.format
+
+local isbe = (string.byte(string.dump(function() end), 5) % 2 == 1)
+
+local function usage(arg)
+  io.stderr:write("Usage: ", arg and arg[0] or "genlibbc",
+                  " [-o buildvm_libbc.h] lib_*.c\n")
+  os.exit(1)
+end
+
+local function parse_arg(arg)
+  local outfile = "-"
+  if not (arg and arg[1]) then
+    usage(arg)
+  end
+  if arg[1] == "-o" then
+    outfile = arg[2]
+    if not outfile then usage(arg) end
+    table.remove(arg, 1)
+    table.remove(arg, 1)
+  end
+  return outfile
+end
+
+local function read_files(names)
+  local src = ""
+  for _,name in ipairs(names) do
+    local fp = assert(io.open(name))
+    src = src .. fp:read("*a")
+    fp:close()
+  end
+  return src
+end
+
+local function transform_lua(code)
+  local fixup = {}
+  local n = -30000
+  code = string.gsub(code, "CHECK_(%w*)%((.-)%)", function(tp, var)
+    n = n + 1
+    fixup[n] = { "CHECK", tp }
+    return format("%s=%d", var, n)
+  end)
+  code = string.gsub(code, "PAIRS%((.-)%)", function(var)
+    fixup.PAIRS = true
+    return format("nil, %s, 0x4dp80", var)
+  end)
+  return "return "..code, fixup
+end
+
+local function read_uleb128(p)
+  local v = p[0]; p = p + 1
+  if v >= 128 then
+    local sh = 7; v = v - 128
+    repeat
+      local r = p[0]
+      v = v + bit.lshift(bit.band(r, 127), sh)
+      sh = sh + 7
+      p = p + 1
+    until r < 128
+  end
+  return p, v
+end
+
+-- ORDER LJ_T
+local name2itype = {
+  str = 5, func = 9, tab = 12, int = 14, num = 15
+}
+
+local BC, BCN = {}, {}
+for i=0,#bcnames/6-1 do
+  local name = bcnames:sub(i*6+1, i*6+6):gsub(" ", "")
+  BC[name] = i
+  BCN[i] = name
+end
+local xop, xra = isbe and 3 or 0, isbe and 2 or 1
+local xrc, xrb = isbe and 1 or 2, isbe and 0 or 3
+
+local function fixup_dump(dump, fixup)
+  local buf = ffi.new("uint8_t[?]", #dump+1, dump)
+  local p = buf+5
+  local n, sizebc
+  p, n = read_uleb128(p)
+  local start = p
+  p = p + 4
+  p = read_uleb128(p)
+  p = read_uleb128(p)
+  p, sizebc = read_uleb128(p)
+  local startbc = tonumber(p - start)
+  local rawtab = {}
+  for i=0,sizebc-1 do
+    local op = p[xop]
+    if op == BC.KSHORT then
+      local rd = p[xrc] + 256*p[xrb]
+      rd = bit.arshift(bit.lshift(rd, 16), 16)
+      local f = fixup[rd]
+      if f then
+        if f[1] == "CHECK" then
+          local tp = f[2]
+          if tp == "tab" then rawtab[p[xra]] = true end
+          p[xop] = tp == "num" and BC.ISNUM or BC.ISTYPE
+          p[xrb] = 0
+          p[xrc] = name2itype[tp]
+        else
+          error("unhandled fixup type: "..f[1])
+        end
+      end
+    elseif op == BC.TGETV then
+      if rawtab[p[xrb]] then
+        p[xop] = BC.TGETR
+      end
+    elseif op == BC.TSETV then
+      if rawtab[p[xrb]] then
+        p[xop] = BC.TSETR
+      end
+    elseif op == BC.ITERC then
+      if fixup.PAIRS then
+        p[xop] = BC.ITERN
+      end
+    end
+    p = p + 4
+  end
+  local ndump = ffi.string(start, n)
+  -- Fixup hi-part of 0x4dp80 to LJ_KEYINDEX.
+  ndump = ndump:gsub("\x80\x80\xcd\xaa\x04", "\xff\xff\xf9\xff\x0f")
+  return { dump = ndump, startbc = startbc, sizebc = sizebc }
+end
+
+local function find_defs(src, mode)
+  local defs = {}
+  for name, code in string.gmatch(src, "LJLIB_LUA%(([^)]*)%)%s*/%*(.-)%*/") do
+    local tcode, fixup = transform_lua(code)
+    local func = assert(load(tcode, "", mode))
+    defs[name] = fixup_dump(string.dump(func, mode), fixup)
+    defs[#defs+1] = name
+  end
+  return defs
+end
+
+local function gen_header(defs32, defs64)
+  local t = {}
+  local function w(x) t[#t+1] = x end
+  w("/* This is a generated file. DO NOT EDIT! */\n\n")
+  w("static const int libbc_endian = ") w(isbe and 1 or 0) w(";\n\n")
+  for j,defs in ipairs{defs64, defs32} do
+    local s, sb = "", ""
+    for i,name in ipairs(defs) do
+      local d = defs[name]
+      s = s .. d.dump
+      sb = sb .. string.char(i) .. ("\0"):rep(d.startbc - 1)
+              .. (isbe and "\0\0\0\255" or "\255\0\0\0"):rep(d.sizebc)
+              .. ("\0"):rep(#d.dump - d.startbc - d.sizebc*4)
+    end
+    if j == 1 then
+      w("static const uint8_t libbc_code[] = {\n#if LJ_FR2\n")
+    else
+      w("\n#else\n")
+    end
+    local n = 0
+    for i=1,#s do
+      local x = string.byte(s, i)
+      local xb = string.byte(sb, i)
+      if xb == 255 then
+        local name = BCN[x]
+        local m = #name + 4
+        if n + m > 78 then n = 0; w("\n") end
+        n = n + m
+        w("BC_"); w(name)
+      else
+        local m = x < 10 and 2 or (x < 100 and 3 or 4)
+        if xb == 0 then
+          if n + m > 78 then n = 0; w("\n") end
+        else
+          local name = defs[xb]:gsub("_", ".")
+          if n ~= 0 then w("\n") end
+          w("/* "); w(name); w(" */ ")
+          n = #name + 7
+        end
+        n = n + m
+        w(x)
+      end
+      w(",")
+    end
+  end
+  w("\n#endif\n0\n};\n\n")
+  w("static const struct { const char *name; int ofs; } libbc_map[] = {\n")
+  local m32, m64 = 0, 0
+  for i,name in ipairs(defs32) do
+    assert(name == defs64[i])
+    w('{"'); w(name); w('",'); w(m32) w('},\n')
+    m32 = m32 + #defs32[name].dump
+    m64 = m64 + #defs64[name].dump
+    assert(m32 == m64)
+  end
+  w("{NULL,"); w(m32); w("}\n};\n\n")
+  return table.concat(t)
+end
+
+local function write_file(name, data)
+  if name == "-" then
+    assert(io.write(data))
+    assert(io.flush())
+  else
+    local fp = io.open(name)
+    if fp then
+      local old = fp:read("*a")
+      fp:close()
+      if data == old then return end
+    end
+    fp = assert(io.open(name, "w"))
+    assert(fp:write(data))
+    assert(fp:close())
+  end
+end
+
+local outfile = parse_arg(arg)
+local src = read_files(arg)
+local defs32 = find_defs(src, "Wdts")
+local defs64 = find_defs(src, "Xdts")
+local hdr = gen_header(defs32, defs64)
+write_file(outfile, hdr)
+