Logical 'not' must be sign-extended for address operands.

src/buildvm_x86.dasc

@@ -37,19 +37,27 @@
 |.endif
 |
 |.define RA,		ecx
-|.if X64; .define RAa, rcx; .else; .define RAa, RA; .endif
 |.define RAL,		cl
 |.define RB,		ebp		// Must be ebp (C callee-save).
-|.if X64; .define RBa, rbp; .else; .define RBa, RB; .endif
 |.define RC,		eax		// Must be eax (fcomparepp and others).
 |.define RCW,		ax
 |.define RCH,		ah
 |.define RCL,		al
 |.define OP,		RB
 |.define RD,		RC
-|.if X64; .define RDa, rax; .else; .define RDa, RD; .endif
 |.define RDW,		RCW
 |.define RDL,		RCL
+|.if X64
+|.define RAa, rcx
+|.define RBa, rbp
+|.define RCa, rax
+|.define RDa, rax
+|.else
+|.define RAa, RA
+|.define RBa, RB
+|.define RCa, RC
+|.define RDa, RD
+|.endif
 |
 |.if not X64
 |.define FCARG1,	ecx		// x86 fastcall arguments.
@@ -237,7 +245,7 @@
 |.macro ins_ABC; movzx RB, RCH; movzx RC, RCL; .endmacro
 |.macro ins_AB_; movzx RB, RCH; .endmacro
 |.macro ins_A_C; movzx RC, RCL; .endmacro
-|.macro ins_AND; not RD; .endmacro
+|.macro ins_AND; not RDa; .endmacro
 |
 |// Instruction decode+dispatch. Carefully tuned (nope, lodsd is not faster).
 |.macro ins_NEXT
@@ -898,7 +906,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |->vmeta_tgets:
   |  mov TMP1, RC			// RC = GCstr *
   |  mov TMP2, LJ_TSTR
-  |  lea RC, TMP1			// Store temp. TValue in TMP1/TMP2.
+  |  lea RCa, TMP1			// Store temp. TValue in TMP1/TMP2.
   |  cmp PC_OP, BC_GGET
   |  jne >1
   |  lea RA, [DISPATCH+DISPATCH_GL(tmptv)]  // Store fn->l.env in g->tmptv.
@@ -919,7 +927,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
     |  fstp TMPQ
     |.endif
   }
-  |  lea RC, TMP1			// Store temp. TValue in TMP1/TMP2.
+  |  lea RCa, TMP1			// Store temp. TValue in TMP1/TMP2.
   |  jmp >1
   |
   |->vmeta_tgetv:
@@ -933,7 +941,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  mov L:CARG1d, SAVE_L
   |  mov L:CARG1d->base, BASE		// Caveat: CARG2d/CARG3d may be BASE.
   |  mov CARG2d, RB
-  |  mov CARG3d, RC
+  |  mov CARG3, RCa			// May be 64 bit ptr to stack.
   |  mov L:RB, L:CARG1d
   |.else
   |  mov ARG2, RB
@@ -971,7 +979,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |->vmeta_tsets:
   |  mov TMP1, RC			// RC = GCstr *
   |  mov TMP2, LJ_TSTR
-  |  lea RC, TMP1			// Store temp. TValue in TMP1/TMP2.
+  |  lea RCa, TMP1			// Store temp. TValue in TMP1/TMP2.
   |  cmp PC_OP, BC_GSET
   |  jne >1
   |  lea RA, [DISPATCH+DISPATCH_GL(tmptv)]  // Store fn->l.env in g->tmptv.
@@ -992,7 +1000,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
     |  fstp TMPQ
     |.endif
   }
-  |  lea RC, TMP1			// Store temp. TValue in TMP1/TMP2.
+  |  lea RCa, TMP1			// Store temp. TValue in TMP1/TMP2.
   |  jmp >1
   |
   |->vmeta_tsetv:
@@ -1006,7 +1014,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  mov L:CARG1d, SAVE_L
   |  mov L:CARG1d->base, BASE		// Caveat: CARG2d/CARG3d may be BASE.
   |  mov CARG2d, RB
-  |  mov CARG3d, RC
+  |  mov CARG3, RCa			// May be 64 bit ptr to stack.
   |  mov L:RB, L:CARG1d
   |.else
   |  mov ARG2, RB
@@ -1887,7 +1895,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |
   |7:  // Non-standard return case.
   |  mov BASE, RA
-  |  mov RA, -8				// Results start at BASE+RA = BASE-8.
+  |  mov RAa, -8			// Results start at BASE+RA = BASE-8.
   |  jmp ->vm_return
   |
   if (sse) {
@@ -3906,8 +3914,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     |  ins_AD	// RA = dst, RD = hbits|asize
     |.if X64
     |  mov L:CARG1d, SAVE_L
-    |1:
     |  mov L:CARG1d->base, BASE		// Caveat: CARG2d/CARG3d may be BASE.
+    |1:
     |  mov CARG3d, RD
     |  and RD, 0x7ff
     |  shr CARG3d, 11
@@ -3982,7 +3990,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     |  mov L:FCARG1, L:RB
     |  call extern lj_gc_step_fixtop@4	// (lua_State *L)
     |  movzx RD, PC_RD			// Need to reload RD.
-    |  not RD
+    |  not RDa
     |  jmp <2
     break;
 
@@ -4052,7 +4060,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     break;
   case BC_TGETS:
     |  ins_ABC	// RA = dst, RB = table, RC = str const (~)
-    |  not RC
+    |  not RCa
     |  mov STR:RC, [KBASE+RC*4]
     |  checktab RB, ->vmeta_tgets
     |  mov TAB:RB, [BASE+RB*8]
@@ -4182,7 +4190,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     break;
   case BC_TSETS:
     |  ins_ABC	// RA = src, RB = table, RC = str const (~)
-    |  not RC
+    |  not RCa
     |  mov STR:RC, [KBASE+RC*4]
     |  checktab RB, ->vmeta_tsets
     |  mov TAB:RB, [BASE+RB*8]
@@ -4427,7 +4435,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     |  test PC, FRAME_TYPE		// Lua frame below?
     |  jnz <4
     |  movzx RD, PC_RA			// Need to prepare BASE/KBASE.
-    |  not RD
+    |  not RDa
     |  lea BASE, [BASE+RD*8]
     |  mov LFUNC:KBASE, [BASE-8]
     |  mov PROTO:KBASE, LFUNC:KBASE->pt
@@ -4591,7 +4599,7 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
       break;
     }
     |  movzx RA, PC_RA
-    |  not RA				// Note: ~RA = -(RA+1)
+    |  not RAa				// Note: ~RA = -(RA+1)
     |  lea BASE, [BASE+RA*8]		// base = base - (RA+1)*8
     |  mov LFUNC:KBASE, [BASE-8]
     |  mov PROTO:KBASE, LFUNC:KBASE->pt

src/lib_jit.c

@@ -224,14 +224,14 @@ LJLIB_CF(jit_util_funcbc)
 LJLIB_CF(jit_util_funck)
 {
   GCproto *pt = check_Lproto(L, 0);
-  MSize idx = (MSize)lj_lib_checkint(L, 2);
+  ptrdiff_t idx = (ptrdiff_t)lj_lib_checkint(L, 2);
-  if ((int32_t)idx >= 0) {
+  if (idx >= 0) {
-    if (idx < pt->sizekn) {
+    if (idx < (ptrdiff_t)pt->sizekn) {
       setnumV(L->top-1, pt->k.n[idx]);
       return 1;
     }
   } else {
-    if (~idx < pt->sizekgc) {
+    if (~idx < (ptrdiff_t)pt->sizekgc) {
       GCobj *gc = gcref(pt->k.gc[idx]);
       setgcV(L, L->top-1, &gc->gch, ~gc->gch.gct);
       return 1;

src/lj_err.c

@@ -107,10 +107,10 @@ restart:
 	if (ra == slot) { slot = bc_d(ins); goto restart; }
 	break;
       case BC_GGET:
-	*name = strdata(gco2str(gcref(pt->k.gc[~bc_d(ins)])));
+	*name = strdata(gco2str(gcref(pt->k.gc[~(ptrdiff_t)bc_d(ins)])));
 	return "global";
       case BC_TGETS:
-	*name = strdata(gco2str(gcref(pt->k.gc[~bc_c(ins)])));
+	*name = strdata(gco2str(gcref(pt->k.gc[~(ptrdiff_t)bc_c(ins)])));
 	if (ip > pt->bc) {
 	  BCIns insp = ip[-1];
 	  if (bc_op(insp) == BC_MOV && bc_a(insp) == ra+1 &&

src/lj_func.c

@@ -169,7 +169,7 @@ GCfunc *lj_func_newL_gc(lua_State *L, GCproto *pt, GCfuncL *parent)
   nuv = fn->l.nupvalues;
   base = L->base;
   for (i = 0; i < nuv; i++) {
-    int v = pt->uv[i];
+    ptrdiff_t v = pt->uv[i];
     GCupval *uv = v < 0 ? &gcref(puv[~v])->uv : func_finduv(L, base + v);
     setgcref(fn->l.uvptr[i], obj2gco(uv));
   }

src/lj_meta.c

@@ -295,7 +295,7 @@ TValue *lj_meta_equal(lua_State *L, GCobj *o1, GCobj *o2, int ne)
     top = curr_top(L);
     setcont(top, ne ? lj_cont_condf : lj_cont_condt);
     copyTV(L, top+1, mo);
-    it = ~o1->gch.gct;
+    it = ~(int32_t)o1->gch.gct;
     setgcV(L, top+2, &o1->gch, it);
     setgcV(L, top+3, &o2->gch, it);
     return top+2;  /* Trigger metamethod call. */

src/lj_record.c

@@ -1751,7 +1751,7 @@ void lj_record_ins(jit_State *J)
   case BCMpri: setitype(rcv, (int32_t)~rc); rc = TREF_PRI(IRT_NIL+rc); break;
   case BCMnum: { lua_Number n = J->pt->k.n[rc];
     setnumV(rcv, n); ix.key = rc = lj_ir_knumint(J, n); } break;
-  case BCMstr: { GCstr *s = strref(J->pt->k.gc[~rc]);
+  case BCMstr: { GCstr *s = strref(J->pt->k.gc[~(ptrdiff_t)rc]);
     setstrV(J->L, rcv, s); ix.key = rc = lj_ir_kstr(J, s); } break;
   default: break;  /* Handled later. */
   }
@@ -1943,7 +1943,7 @@ void lj_record_ins(jit_State *J)
     break;
   case BC_TDUP:
     rc = emitir(IRT(IR_TDUP, IRT_TAB),
-		lj_ir_ktab(J, tabref(J->pt->k.gc[~rc])), 0);
+		lj_ir_ktab(J, tabref(J->pt->k.gc[~(ptrdiff_t)rc])), 0);
     break;
 
   /* -- Calls and vararg handling ----------------------------------------- */