Add missing GC steps for io.* functions.
MIPS: Fix cache flush/sync for JIT-compiled code jump area.
ARM: Fix cache flush/sync for exit stubs of JIT-compiled code.
Fix MSVC intrinsics for older versions.
Fix memory access check for fast string interning.

--- a/src/lib_io.c
+++ b/src/lib_io.c
@@ -17,6 +17,7 @@
 #include "lualib.h"
 
 #include "lj_obj.h"
+#include "lj_gc.h"
 #include "lj_err.h"
 #include "lj_str.h"
 #include "lj_state.h"
@@ -152,6 +153,7 @@ static int io_file_readline(lua_State *L, FILE *fp, MSize chop)
     if (n >= m - 64) m += m;
   }
   setstrV(L, L->top++, lj_str_new(L, buf, (size_t)n));
+  lj_gc_check(L);
   return (int)ok;
 }
 
@@ -163,6 +165,7 @@ static void io_file_readall(lua_State *L, FILE *fp)
     n += (MSize)fread(buf+n, 1, m-n, fp);
     if (n != m) {
       setstrV(L, L->top++, lj_str_new(L, buf, (size_t)n));
+      lj_gc_check(L);
       return;
     }
   }
@@ -174,6 +177,7 @@ static int io_file_readlen(lua_State *L, FILE *fp, MSize m)
     char *buf = lj_str_needbuf(L, &G(L)->tmpbuf, m);
     MSize n = (MSize)fread(buf, 1, m, fp);
     setstrV(L, L->top++, lj_str_new(L, buf, (size_t)n));
+    lj_gc_check(L);
     return (n > 0 || m == 0);
   } else {
     int c = getc(fp);
--- a/src/lj_asm_arm.h
+++ b/src/lj_asm_arm.h
@@ -91,6 +91,7 @@ static MCode *asm_exitstub_gen(ASMState *as, ExitNo group)
   *mxp++ = group*EXITSTUBS_PER_GROUP;
   for (i = 0; i < EXITSTUBS_PER_GROUP; i++)
     *mxp++ = ARMI_B|((-6-i)&0x00ffffffu);
+  lj_mcode_sync(as->mcbot, mxp);
   lj_mcode_commitbot(as->J, mxp);
   as->mcbot = mxp;
   as->mclim = as->mcbot + MCLIM_REDZONE;
--- a/src/lj_asm_mips.h
+++ b/src/lj_asm_mips.h
@@ -71,6 +71,7 @@ static void asm_sparejump_setup(ASMState *as)
     memset(mxp+2, 0, MIPS_SPAREJUMP*8);
     mxp += MIPS_SPAREJUMP*2;
     lua_assert(mxp < as->mctop);
+    lj_mcode_sync(as->mcbot, mxp);
     lj_mcode_commitbot(as->J, mxp);
     as->mcbot = mxp;
     as->mclim = as->mcbot + MCLIM_REDZONE;
--- a/src/lj_def.h
+++ b/src/lj_def.h
@@ -243,17 +243,17 @@ static LJ_AINLINE uint32_t lj_getu32(const void *p)
 #endif
 
 #ifdef _M_PPC
-#pragma intrinsic(_CountLeadingZeros)
 unsigned int _CountLeadingZeros(long);
+#pragma intrinsic(_CountLeadingZeros)
 static LJ_AINLINE uint32_t lj_fls(uint32_t x)
 {
   return _CountLeadingZeros(x) ^ 31;
 }
 #else
-#pragma intrinsic(_BitScanForward)
-#pragma intrinsic(_BitScanReverse)
 unsigned char _BitScanForward(uint32_t *, unsigned long);
 unsigned char _BitScanReverse(uint32_t *, unsigned long);
+#pragma intrinsic(_BitScanForward)
+#pragma intrinsic(_BitScanReverse)
 
 static LJ_AINLINE uint32_t lj_ffs(uint32_t x)
 {
--- a/src/lj_str.c
+++ b/src/lj_str.c
@@ -48,7 +48,7 @@ static LJ_AINLINE int str_fastcmp(const char *a, const char *b, MSize len)
 {
   MSize i = 0;
   lua_assert(len > 0);
-  lua_assert((((uintptr_t)a + len) & (LJ_PAGESIZE-1)) <= LJ_PAGESIZE-4);
+  lua_assert((((uintptr_t)a+len-1) & (LJ_PAGESIZE-1)) <= LJ_PAGESIZE-4);
   do {  /* Note: innocuous access up to end of string + 3. */
     uint32_t v = lj_getu32(a+i) ^ *(const uint32_t *)(b+i);
     if (v) {
@@ -121,7 +121,7 @@ GCstr *lj_str_new(lua_State *L, const char *str, size_t lenx)
   h ^= b; h -= lj_rol(b, 16);
   /* Check if the string has already been interned. */
   o = gcref(g->strhash[h & g->strmask]);
-  if (LJ_LIKELY((((uintptr_t)str + len) & (LJ_PAGESIZE-1)) <= LJ_PAGESIZE-4)) {
+  if (LJ_LIKELY((((uintptr_t)str+len-1) & (LJ_PAGESIZE-1)) <= LJ_PAGESIZE-4)) {
     while (o != NULL) {
       GCstr *sx = gco2str(o);
       if (sx->len == len && str_fastcmp(str, strdata(sx), len) == 0) {