allow endian specification of mrb files by mrbc -e/-E

`mruby -b` now accepts both big/little endian mrb (compiled binary) files.

`mrbc` generates mrb files in big endian for .mrb files and in native endian
for C files (with -B option specified) by default.  If you are cross compiling,
you need to specify target endian by -e/-E options if it is different from
host endian.

Author: matz

include/mruby/dump.h

@@ -14,10 +14,16 @@ extern "C" {
 #include "mruby.h"
 #include "mruby/irep.h"
 
-int mrb_dump_irep(mrb_state *mrb, mrb_irep *irep, int debug_info, uint8_t **bin, size_t *bin_size);
+#define DUMP_DEBUG_INFO 1
+#define DUMP_ENDIAN_BIG 2
+#define DUMP_ENDIAN_LIL 4
+#define DUMP_ENDIAN_NAT 6 
+#define DUMP_ENDIAN_MASK 6 
+
+int mrb_dump_irep(mrb_state *mrb, mrb_irep *irep, uint8_t flags, uint8_t **bin, size_t *bin_size);
 #ifdef ENABLE_STDIO
-int mrb_dump_irep_binary(mrb_state*, mrb_irep*, int, FILE*);
-int mrb_dump_irep_cfunc(mrb_state *mrb, mrb_irep*, int, FILE *f, const char *initname);
+int mrb_dump_irep_binary(mrb_state*, mrb_irep*, uint8_t, FILE*);
+int mrb_dump_irep_cfunc(mrb_state *mrb, mrb_irep*, uint8_t flags, FILE *f, const char *initname);
 mrb_irep *mrb_read_irep_file(mrb_state*, FILE*);
 MRB_API mrb_value mrb_load_irep_file(mrb_state*,FILE*);
 MRB_API mrb_value mrb_load_irep_file_cxt(mrb_state*, FILE*, mrbc_context*);
@@ -42,25 +48,26 @@ MRB_API mrb_irep *mrb_read_irep(mrb_state*, const uint8_t*);
 #define MRB_DUMP_NULL_SYM_LEN         0xFFFF
 
 /* Rite Binary File header */
-#define RITE_BINARY_IDENTIFIER         "RITE"
+#define RITE_BINARY_IDENT              "RITE"
+#define RITE_BINARY_IDENT_LIL          "ETIR"
 #define RITE_BINARY_FORMAT_VER         "0003"
 #define RITE_COMPILER_NAME             "MATZ"
 #define RITE_COMPILER_VERSION          "0000"
 
 #define RITE_VM_VER                    "0000"
 
 #define RITE_BINARY_EOF                "END\0"
-#define RITE_SECTION_IREP_IDENTIFIER   "IREP"
-#define RITE_SECTION_LINENO_IDENTIFIER "LINE"
-#define RITE_SECTION_DEBUG_IDENTIFIER  "DBG\0"
-#define RITE_SECTION_LV_IDENTIFIER     "LVAR"
+#define RITE_SECTION_IREP_IDENT        "IREP"
+#define RITE_SECTION_LINENO_IDENT      "LINE"
+#define RITE_SECTION_DEBUG_IDENT       "DBG\0"
+#define RITE_SECTION_LV_IDENT          "LVAR"
 
 #define MRB_DUMP_DEFAULT_STR_LEN      128
 #define MRB_DUMP_ALIGNMENT            sizeof(uint32_t)
 
 /* binary header */
 struct rite_binary_header {
-  uint8_t binary_identify[4]; /* Binary Identifier */
+  uint8_t binary_ident[4];    /* Binary Identifier */
   uint8_t binary_version[4];  /* Binary Format Version */
   uint8_t binary_crc[2];      /* Binary CRC */
   uint8_t binary_size[4];     /* Binary Size */
@@ -70,7 +77,7 @@ struct rite_binary_header {
 
 /* section header */
 #define RITE_SECTION_HEADER \
-  uint8_t section_identify[4]; \
+  uint8_t section_ident[4]; \
   uint8_t section_size[4]
 
 struct rite_section_header {
@@ -101,6 +108,17 @@ struct rite_binary_footer {
   RITE_SECTION_HEADER;
 };
 
+static inline int
+bigendian_p()
+{
+  int i;
+  char *p;
+
+  i = 1;
+  p = (char*)&i;
+  return p[0]?0:1;
+}
+
 static inline size_t
 uint8_to_bin(uint8_t s, uint8_t *bin)
 {
@@ -126,6 +144,16 @@ uint32_to_bin(uint32_t l, uint8_t *bin)
   return sizeof(uint32_t);
 }
 
+static inline size_t
+uint32l_to_bin(uint32_t l, uint8_t *bin)
+{
+  bin[3] = (l >> 24) & 0xff;
+  bin[2] = (l >> 16) & 0xff;
+  bin[1] = (l >> 8) & 0xff;
+  bin[0] = l & 0xff;
+  return sizeof(uint32_t);
+}
+
 static inline uint32_t
 bin_to_uint32(const uint8_t *bin)
 {
@@ -135,6 +163,15 @@ bin_to_uint32(const uint8_t *bin)
          (uint32_t)bin[3];
 }
 
+static inline uint32_t
+bin_to_uint32l(const uint8_t *bin)
+{
+  return (uint32_t)bin[3] << 24 |
+         (uint32_t)bin[2] << 16 |
+         (uint32_t)bin[1] << 8  |
+         (uint32_t)bin[0];
+}
+
 static inline uint16_t
 bin_to_uint16(const uint8_t *bin)
 {

src/dump.c

@@ -80,14 +80,25 @@ write_iseq_block(mrb_state *mrb, mrb_irep *irep, uint8_t *buf, uint8_t flags)
 
   cur += uint32_to_bin(irep->ilen, cur); /* number of opcode */
   cur += write_padding(cur);
-  if (flags & FLAG_BYTEORDER_NATIVE) {
-    memcpy(cur, irep->iseq, irep->ilen * sizeof(mrb_code));
-    cur += irep->ilen * sizeof(mrb_code);
-  }
-  else {
+  switch (flags & DUMP_ENDIAN_NAT) {
+  case DUMP_ENDIAN_BIG:
+    if (bigendian_p()) goto native;
     for (iseq_no = 0; iseq_no < irep->ilen; iseq_no++) {
       cur += uint32_to_bin(irep->iseq[iseq_no], cur); /* opcode */
     }
+    break;
+  case DUMP_ENDIAN_LIL:
+    if (!bigendian_p()) goto native;
+    for (iseq_no = 0; iseq_no < irep->ilen; iseq_no++) {
+      cur += uint32l_to_bin(irep->iseq[iseq_no], cur); /* opcode */
+    }
+    break;
+
+  native:
+  case DUMP_ENDIAN_NAT:
+    memcpy(cur, irep->iseq, irep->ilen * sizeof(mrb_code));
+    cur += irep->ilen * sizeof(mrb_code);
+    break;
   }
 
   return cur - buf;
@@ -323,7 +334,7 @@ write_footer(mrb_state *mrb, uint8_t *bin)
 {
   struct rite_binary_footer footer;
 
-  memcpy(footer.section_identify, RITE_BINARY_EOF, sizeof(footer.section_identify));
+  memcpy(footer.section_ident, RITE_BINARY_EOF, sizeof(footer.section_ident));
   uint32_to_bin(sizeof(struct rite_binary_footer), footer.section_size);
   memcpy(bin, &footer, sizeof(struct rite_binary_footer));
 
@@ -336,7 +347,7 @@ write_section_irep_header(mrb_state *mrb, size_t section_size, uint8_t *bin)
 {
   struct rite_section_irep_header *header = (struct rite_section_irep_header*)bin;
 
-  memcpy(header->section_identify, RITE_SECTION_IREP_IDENTIFIER, sizeof(header->section_identify));
+  memcpy(header->section_ident, RITE_SECTION_IREP_IDENT, sizeof(header->section_ident));
 
   mrb_assert_int_fit(size_t, section_size, uint32_t, UINT32_MAX);
   uint32_to_bin((uint32_t)section_size, header->section_size);
@@ -373,7 +384,7 @@ write_section_lineno_header(mrb_state *mrb, size_t section_size, uint8_t *bin)
 {
   struct rite_section_lineno_header *header = (struct rite_section_lineno_header*)bin;
 
-  memcpy(header->section_identify, RITE_SECTION_LINENO_IDENTIFIER, sizeof(header->section_identify));
+  memcpy(header->section_ident, RITE_SECTION_LINENO_IDENT, sizeof(header->section_ident));
   uint32_to_bin((uint32_t)section_size, header->section_size);
 
   return MRB_DUMP_OK;
@@ -664,7 +675,7 @@ write_section_debug(mrb_state *mrb, mrb_irep *irep, uint8_t *cur, mrb_sym const
   dlen = write_debug_record(mrb, irep, cur, filenames, filenames_len);
   section_size += dlen;
 
-  memcpy(header->section_identify, RITE_SECTION_DEBUG_IDENTIFIER, sizeof(header->section_identify));
+  memcpy(header->section_ident, RITE_SECTION_DEBUG_IDENT, sizeof(header->section_ident));
   mrb_assert(section_size <= INT32_MAX);
   uint32_to_bin(section_size, header->section_size);
 
@@ -803,7 +814,7 @@ write_section_lv(mrb_state *mrb, mrb_irep *irep, uint8_t *start, mrb_sym const *
     goto lv_section_exit;
   }
 
-  memcpy(header->section_identify, RITE_SECTION_LV_IDENTIFIER, sizeof(header->section_identify));
+  memcpy(header->section_ident, RITE_SECTION_LV_IDENT, sizeof(header->section_ident));
 
   diff = cur - start;
   mrb_assert_int_fit(ptrdiff_t, diff, size_t, SIZE_MAX);
@@ -820,19 +831,23 @@ write_rite_binary_header(mrb_state *mrb, size_t binary_size, uint8_t *bin, uint8
   uint16_t crc;
   uint32_t offset;
 
-  if (flags & FLAG_BYTEORDER_NATIVE) {
-    uint32_t ident = 0;
-    size_t i;
-
-    for(i=0; i<sizeof(ident); i++) {
-      ident<<=8;
-      ident|=RITE_BINARY_IDENTIFIER[i];
-    }
-    memcpy(header->binary_identify, (char*)&ident, sizeof(uint32_t));
-  }
-  else {
-    memcpy(header->binary_identify, RITE_BINARY_IDENTIFIER, sizeof(header->binary_identify));
+  switch (flags & DUMP_ENDIAN_NAT) {
+  default:
+  endian_big:
+  case DUMP_ENDIAN_BIG:
+    memcpy(header->binary_ident, RITE_BINARY_IDENT, sizeof(header->binary_ident));
+    break;
+  endian_little:
+  case DUMP_ENDIAN_LIL:
+    memcpy(header->binary_ident, RITE_BINARY_IDENT_LIL, sizeof(header->binary_ident));
+    break;
+
+  case DUMP_ENDIAN_NAT:
+    if (bigendian_p()) goto endian_big;
+    goto endian_little;
+    break;
   }
+
   memcpy(header->binary_version, RITE_BINARY_FORMAT_VER, sizeof(header->binary_version));
   memcpy(header->compiler_name, RITE_COMPILER_NAME, sizeof(header->compiler_name));
   memcpy(header->compiler_version, RITE_COMPILER_VERSION, sizeof(header->compiler_version));
@@ -872,8 +887,23 @@ is_lv_defined(mrb_irep *irep)
   return FALSE;
 }
 
+static uint8_t
+dump_flags(uint8_t flags, uint8_t native)
+{
+  if (native == FLAG_BYTEORDER_NATIVE) {
+    if ((flags & DUMP_ENDIAN_NAT) == 0) {
+      return (flags & DUMP_DEBUG_INFO) | DUMP_ENDIAN_NAT;
+    }
+    return flags;
+  }
+  if ((flags & DUMP_ENDIAN_NAT) == 0) {
+    return (flags & DUMP_DEBUG_INFO) | DUMP_ENDIAN_BIG;
+  }
+  return flags;
+}
+
 static int
-dump_irep(mrb_state *mrb, mrb_irep *irep, int debug_info, uint8_t **bin, size_t *bin_size, uint8_t flags)
+dump_irep(mrb_state *mrb, mrb_irep *irep, uint8_t flags, uint8_t **bin, size_t *bin_size)
 {
   int result = MRB_DUMP_GENERAL_FAILURE;
   size_t malloc_size;
@@ -893,7 +923,7 @@ dump_irep(mrb_state *mrb, mrb_irep *irep, int debug_info, uint8_t **bin, size_t
   section_irep_size += get_irep_record_size(mrb, irep);
 
   /* DEBUG section size */
-  if (debug_info) {
+  if (flags & DUMP_DEBUG_INFO) {
     if (debug_info_defined) {
       section_lineno_size += sizeof(struct rite_section_debug_header);
       /* filename table */
@@ -933,7 +963,7 @@ dump_irep(mrb_state *mrb, mrb_irep *irep, int debug_info, uint8_t **bin, size_t
               sizeof(struct rite_binary_footer);
 
   /* write DEBUG section */
-  if (debug_info) {
+  if (flags & DUMP_DEBUG_INFO) {
     if (debug_info_defined) {
       result = write_section_debug(mrb, irep, cur, filenames, filenames_len);
     }
@@ -972,13 +1002,13 @@ dump_irep(mrb_state *mrb, mrb_irep *irep, int debug_info, uint8_t **bin, size_t
 }
 
 int
-mrb_dump_irep(mrb_state *mrb, mrb_irep *irep, int debug_info, uint8_t **bin, size_t *bin_size)
+mrb_dump_irep(mrb_state *mrb, mrb_irep *irep, uint8_t flags, uint8_t **bin, size_t *bin_size)
 {
-  return dump_irep(mrb, irep, debug_info, bin, bin_size, FLAG_BYTEORDER_NONATIVE);
+  return dump_irep(mrb, irep, dump_flags(flags, FLAG_BYTEORDER_NONATIVE), bin, bin_size);
 }
 
 int
-mrb_dump_irep_binary(mrb_state *mrb, mrb_irep *irep, int debug_info, FILE* fp)
+mrb_dump_irep_binary(mrb_state *mrb, mrb_irep *irep, uint8_t flags, FILE* fp)
 {
   uint8_t *bin = NULL;
   size_t bin_size = 0;
@@ -988,7 +1018,7 @@ mrb_dump_irep_binary(mrb_state *mrb, mrb_irep *irep, int debug_info, FILE* fp)
     return MRB_DUMP_INVALID_ARGUMENT;
   }
 
-  result = dump_irep(mrb, irep, debug_info, &bin, &bin_size, FLAG_BYTEORDER_NONATIVE);
+  result = dump_irep(mrb, irep, dump_flags(flags, FLAG_BYTEORDER_NONATIVE), &bin, &bin_size);
   if (result == MRB_DUMP_OK) {
     if (fwrite(bin, sizeof(bin[0]), bin_size, fp) != bin_size) {
       result = MRB_DUMP_WRITE_FAULT;
@@ -1015,8 +1045,22 @@ is_valid_c_symbol_name(const char *name)
    return TRUE;
 }
 
+static int
+dump_bigendian_p(uint8_t flags)
+{
+  switch (flags & DUMP_ENDIAN_NAT) {
+  case DUMP_ENDIAN_BIG:
+    return TRUE;
+  case DUMP_ENDIAN_LIL:
+    return FALSE;
+  default:
+  case DUMP_ENDIAN_NAT:
+    return bigendian_p();
+  }
+}
+
 int
-mrb_dump_irep_cfunc(mrb_state *mrb, mrb_irep *irep, int debug_info, FILE *fp, const char *initname)
+mrb_dump_irep_cfunc(mrb_state *mrb, mrb_irep *irep, uint8_t flags, FILE *fp, const char *initname)
 {
   uint8_t *bin = NULL;
   size_t bin_size = 0, bin_idx = 0;
@@ -1025,9 +1069,23 @@ mrb_dump_irep_cfunc(mrb_state *mrb, mrb_irep *irep, int debug_info, FILE *fp, co
   if (fp == NULL || initname == NULL || !is_valid_c_symbol_name(initname)) {
     return MRB_DUMP_INVALID_ARGUMENT;
   }
-
-  result = dump_irep(mrb, irep, debug_info, &bin, &bin_size, FLAG_BYTEORDER_NATIVE);
+  flags = dump_flags(flags, FLAG_BYTEORDER_NATIVE);
+  result = dump_irep(mrb, irep, flags, &bin, &bin_size);
   if (result == MRB_DUMP_OK) {
+    if (!dump_bigendian_p(flags)) {
+      if (fprintf(fp, "/* dumped in little endian order.\n"
+                  "   use `mrbc -E` option for big endian CPU. */\n") < 0) {
+        mrb_free(mrb, bin);
+        return MRB_DUMP_WRITE_FAULT;
+      }
+    }
+    else {
+      if (fprintf(fp, "/* dumped in big endian order.\n"
+                  "   use `mrbc -e` option for better performance on little endian CPU. */\n") < 0) {
+        mrb_free(mrb, bin);
+        return MRB_DUMP_WRITE_FAULT;
+      }
+    }
     if (fprintf(fp, "#include <stdint.h>\n") < 0) { /* for uint8_t under at least Darwin */
       mrb_free(mrb, bin);
       return MRB_DUMP_WRITE_FAULT;