Fix some portability issues with the JIT compiler.

6 files changed, 38 insertions, 17 deletions

diff --git a/src/lib_jit.c b/src/lib_jit.c
index 8f809875..e2fc8067 100644
--- a/src/lib_jit.c
+++ b/src/lib_jit.c
@@ -368,12 +368,16 @@ LJLIB_CF(jit_util_tracemc)
 /* local addr = jit.util.traceexitstub(idx) */
 LJLIB_CF(jit_util_traceexitstub)
 {
+#ifdef EXITSTUBS_PER_GROUP
   ExitNo exitno = (ExitNo)lj_lib_checkint(L, 1);
   jit_State *J = L2J(L);
   if (exitno < EXITSTUBS_PER_GROUP*LJ_MAX_EXITSTUBGR) {
     setintptrV(L->top-1, (intptr_t)(void *)exitstub_addr(J, exitno));
     return 1;
   }
+#else
+  UNUSED(L);
+#endif
   return 0;
 }
 
diff --git a/src/lj_dispatch.c b/src/lj_dispatch.c
index 42027e7d..425eeaa2 100644
--- a/src/lj_dispatch.c
+++ b/src/lj_dispatch.c
@@ -211,10 +211,15 @@ int luaJIT_setmode(lua_State *L, int idx, int mode)
     } else {
       if (!(mode & LUAJIT_MODE_ON))
         G2J(g)->flags &= ~(uint32_t)JIT_F_ON;
+#if LJ_TARGET_X86ORX64
       else if ((G2J(g)->flags & JIT_F_SSE2))
         G2J(g)->flags |= (uint32_t)JIT_F_ON;
       else
         return 0;  /* Don't turn on JIT compiler without SSE2 support. */
+#else
+      else
+        G2J(g)->flags |= (uint32_t)JIT_F_ON;
+#endif
       lj_dispatch_update(g);
     }
     break;
diff --git a/src/lj_jit.h b/src/lj_jit.h
index cced4813..22f1d581 100644
--- a/src/lj_jit.h
+++ b/src/lj_jit.h
@@ -27,7 +27,8 @@
 #define JIT_F_CPU_FIRST         JIT_F_CMOV
 #define JIT_F_CPUSTRING         "\4CMOV\4SSE2\4SSE3\6SSE4.1\2P4\3AMD\2K8\4ATOM"
 #else
-#error "Missing CPU-specific JIT engine flags"
+#define JIT_F_CPU_FIRST         0
+#define JIT_F_CPUSTRING         ""
 #endif
 
 /* Optimization flags. */
@@ -118,7 +119,11 @@ typedef enum {
 } PostProc;
 
 /* Machine code type. */
+#if LJ_TARGET_X86ORX64
 typedef uint8_t MCode;
+#else
+typedef uint32_t MCode;
+#endif
 
 /* Stack snapshot header. */
 typedef struct SnapShot {
@@ -252,6 +257,13 @@ enum {
 #define lj_resetsplit(J)        UNUSED(J)
 #endif
 
+/* Exit stubs. */
+#if LJ_TARGET_X86ORX64
+/* Limited by the range of a short fwd jump (127): (2+2)*(32-1)-2 = 122. */
+#define EXITSTUB_SPACING        (2+2)
+#define EXITSTUBS_PER_GROUP     32
+#endif
+
 /* Fold state is used to fold instructions on-the-fly. */
 typedef struct FoldState {
   IRIns ins;            /* Currently emitted instruction. */
@@ -318,7 +330,9 @@ typedef struct jit_State {
 
   int32_t param[JIT_P__MAX];  /* JIT engine parameters. */
 
+#if LJ_TARGET_X86ORX64
   MCode *exitstubgroup[LJ_MAX_EXITSTUBGR];  /* Exit stub group addresses. */
+#endif
 
   HotPenalty penalty[PENALTY_SLOTS];  /* Penalty slots. */
   uint32_t penaltyslot; /* Round-robin index into penalty slots. */
@@ -344,7 +358,7 @@ typedef struct jit_State {
   size_t szallmcarea;   /* Total size of all allocated mcode areas. */
 
   TValue errinfo;       /* Additional info element for trace errors. */
-} jit_State;
+} LJ_ALIGN(16) jit_State;
 
 /* Trivial PRNG e.g. used for penalty randomization. */
 static LJ_AINLINE uint32_t LJ_PRNG_BITS(jit_State *J, int bits)
@@ -354,21 +368,14 @@ static LJ_AINLINE uint32_t LJ_PRNG_BITS(jit_State *J, int bits)
   return J->prngstate >> (32-bits);
 }
 
-/* Exit stubs. */
-#if LJ_TARGET_X86ORX64
-/* Limited by the range of a short fwd jump (127): (2+2)*(32-1)-2 = 122. */
-#define EXITSTUB_SPACING        (2+2)
-#define EXITSTUBS_PER_GROUP     32
-#else
-#error "Missing CPU-specific exit stub definitions"
-#endif
-
+#ifdef EXITSTUBS_PER_GROUP
 /* Return the address of an exit stub. */
 static LJ_AINLINE MCode *exitstub_addr(jit_State *J, ExitNo exitno)
 {
   lua_assert(J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] != NULL);
-  return J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] +
-         EXITSTUB_SPACING*(exitno % EXITSTUBS_PER_GROUP);
+  return (MCode *)((char *)J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] +
+                   EXITSTUB_SPACING*(exitno % EXITSTUBS_PER_GROUP));
 }
+#endif
 
 #endif
diff --git a/src/lj_mcode.c b/src/lj_mcode.c
index 104e4925..b3ce1f7f 100644
--- a/src/lj_mcode.c
+++ b/src/lj_mcode.c
@@ -274,7 +274,7 @@ MCode *lj_mcode_patch(jit_State *J, MCode *ptr, int finish)
   } else {
     MCode *mc = J->mcarea;
     /* Try current area first to use the protection cache. */
-    if (ptr >= mc && ptr < mc + J->szmcarea) {
+    if (ptr >= mc && ptr < (MCode *)((char *)mc + J->szmcarea)) {
       mcode_protect(J, MCPROT_GEN);
       return mc;
     }
@@ -282,7 +282,7 @@ MCode *lj_mcode_patch(jit_State *J, MCode *ptr, int finish)
     for (;;) {
       mc = ((MCLink *)mc)->next;
       lua_assert(mc != NULL);
-      if (ptr >= mc && ptr < mc + ((MCLink *)mc)->size) {
+      if (ptr >= mc && ptr < (MCode *)((char *)mc + ((MCLink *)mc)->size)) {
         mcode_setprot(mc, ((MCLink *)mc)->size, MCPROT_GEN);
         return mc;
       }
diff --git a/src/lj_snap.c b/src/lj_snap.c
index c044923a..a3c081e1 100644
--- a/src/lj_snap.c
+++ b/src/lj_snap.c
@@ -406,7 +406,10 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
         if (irt_isinteger(t)) {
           setintV(o, (int32_t)ex->gpr[r-RID_MIN_GPR]);
         } else if (irt_isnum(t)) {
-          setnumV(o, ex->fpr[r-RID_MIN_FPR]);
+          if (RID_NUM_FPR)
+            setnumV(o, ex->fpr[r-RID_MIN_FPR]);
+          else
+            setnumV(o, *(double *)&ex->gpr[r-RID_MIN_GPR]);
 #if LJ_64
         } else if (irt_islightud(t)) {
           /* 64 bit lightuserdata which may escape already has the tag bits. */
diff --git a/src/lj_trace.c b/src/lj_trace.c
index ab75c9d2..ecfbfed6 100644
--- a/src/lj_trace.c
+++ b/src/lj_trace.c
@@ -278,7 +278,9 @@ int lj_trace_flushall(lua_State *L)
   memset(J->penalty, 0, sizeof(J->penalty));
   /* Free the whole machine code and invalidate all exit stub groups. */
   lj_mcode_free(J);
+#ifdef EXITSTUBS_PER_GROUP
   memset(J->exitstubgroup, 0, sizeof(J->exitstubgroup));
+#endif
   lj_vmevent_send(L, TRACE,
     setstrV(L, L->top++, lj_str_newlit(L, "flush"));
   );
@@ -700,7 +702,7 @@ int LJ_FASTCALL lj_trace_exit(jit_State *J, void *exptr)
 
   lj_vmevent_send(L, TEXIT,
     ExitState *ex = (ExitState *)exptr;
-    uint32_t i;
+    int32_t i;
     lj_state_checkstack(L, 4+RID_NUM_GPR+RID_NUM_FPR+LUA_MINSTACK);
     setintV(L->top++, J->parent);
     setintV(L->top++, J->exitno);