ARM: Ensure relative GG_State element alignment differently.

Thanks to jojo59516 and dwing4g.

src/lj_dispatch.h

@@ -70,12 +70,20 @@ typedef uint16_t HotCount;
 typedef struct GG_State {
   lua_State L;				/* Main thread. */
   global_State g;			/* Global state. */
+#if LJ_TARGET_ARM
+  /* Make g reachable via K12 encoded DISPATCH-relative addressing. */
+  uint8_t align1[(16-sizeof(global_State))&15];
+#endif
 #if LJ_TARGET_MIPS
   ASMFunction got[LJ_GOT__MAX];		/* Global offset table. */
 #endif
 #if LJ_HASJIT
   jit_State J;				/* JIT state. */
   HotCount hotcount[HOTCOUNT_SIZE];	/* Hot counters. */
+#if LJ_TARGET_ARM
+  /* Ditto for J. */
+  uint8_t align2[(16-sizeof(jit_State)-sizeof(HotCount)*HOTCOUNT_SIZE)&15];
+#endif
 #endif
   ASMFunction dispatch[GG_LEN_DISP];	/* Instruction dispatch tables. */
   BCIns bcff[GG_NUM_ASMFF];		/* Bytecode for ASM fast functions. */

src/lj_jit.h

@@ -406,11 +406,7 @@ typedef struct jit_State {
   size_t szallmcarea;	/* Total size of all allocated mcode areas. */
 
   TValue errinfo;	/* Additional info element for trace errors. */
-}
-#if LJ_TARGET_ARM
-LJ_ALIGN(16)		/* For DISPATCH-relative addresses in assembler part. */
-#endif
-jit_State;
+} jit_State;
 
 /* Trivial PRNG e.g. used for penalty randomization. */
 static LJ_AINLINE uint32_t LJ_PRNG_BITS(jit_State *J, int bits)