Merge 1dab08b1ab into eec7a8016c

Thanks to Kacper Michajłow. #1351 #1355

src/Makefile

@@ -302,6 +302,9 @@ endif
 ifneq (,$(INSTALL_LJLIBD))
   TARGET_XCFLAGS+= -DLUA_LJDIR=\"$(INSTALL_LJLIBD)\"
 endif
+ifeq (,$(shell $(TARGET_CC) -o /dev/null -c -x c /dev/null -fno-strict-float-cast-overflow 2>/dev/null || echo 1))
+  TARGET_XCFLAGS+= -fno-strict-float-cast-overflow
+endif
 
 ##############################################################################
 # Target system detection.

src/lj_asm_arm.h

@@ -1927,7 +1927,7 @@ static void asm_hiop(ASMState *as, IRIns *ir)
   } else if ((ir-1)->o == IR_MIN || (ir-1)->o == IR_MAX) {
     as->curins--;  /* Always skip the loword min/max. */
     if (uselo || usehi)
-      asm_sfpmin_max(as, ir-1, (ir-1)->o == IR_MIN ? CC_PL : CC_LE);
+      asm_sfpmin_max(as, ir-1, (ir-1)->o == IR_MIN ? CC_HS : CC_LS);
     return;
 #elif LJ_HASFFI
   } else if ((ir-1)->o == IR_CONV) {

src/lj_asm_x86.h

@@ -886,32 +886,38 @@ static void asm_conv(ASMState *as, IRIns *ir)
       asm_tointg(as, ir, ra_alloc1(as, lref, RSET_FPR));
     } else {
       Reg dest = ra_dest(as, ir, RSET_GPR);
+      Reg tmp = ra_noreg(IR(lref)->r) ? ra_alloc1(as, lref, RSET_FPR) :
+					ra_scratch(as, RSET_FPR);
       x86Op op = st == IRT_NUM ? XO_CVTTSD2SI : XO_CVTTSS2SI;
-      if (LJ_64 ? irt_isu64(ir->t) : irt_isu32(ir->t)) {
-	/* LJ_64: For inputs >= 2^63 add -2^64, convert again. */
-	/* LJ_32: For inputs >= 2^31 add -2^31, convert again and add 2^31. */
-	Reg tmp = ra_noreg(IR(lref)->r) ? ra_alloc1(as, lref, RSET_FPR) :
-					  ra_scratch(as, RSET_FPR);
-	MCLabel l_end = emit_label(as);
-	if (LJ_32)
-	  emit_gri(as, XG_ARITHi(XOg_ADD), dest, (int32_t)0x80000000);
-	emit_rr(as, op, dest|REX_64, tmp);
-	if (st == IRT_NUM)
-	  emit_rma(as, XO_ADDSD, tmp, &as->J->k64[LJ_K64_M2P64_31]);
-	else
-	  emit_rma(as, XO_ADDSS, tmp, &as->J->k32[LJ_K32_M2P64_31]);
-	emit_sjcc(as, CC_NS, l_end);
-	emit_rr(as, XO_TEST, dest|REX_64, dest);  /* Check if dest negative. */
-	emit_rr(as, op, dest|REX_64, tmp);
-	ra_left(as, tmp, lref);
-      } else {
-	if (LJ_64 && irt_isu32(ir->t))
-	  emit_rr(as, XO_MOV, dest, dest);  /* Zero hiword. */
-	emit_mrm(as, op,
-		 dest|((LJ_64 &&
-			(irt_is64(ir->t) || irt_isu32(ir->t))) ? REX_64 : 0),
-		 asm_fuseload(as, lref, RSET_FPR));
+      Reg r64 = (LJ_64 && irt_is64 (ir->t)) ? REX_64 : 0;
+      if (LJ_64 && (irt_isu32(ir->t) || irt_isint(ir->t)))
+	emit_rr(as, XO_MOV, dest, dest);  /* Zero hiword. */
+      if (irt_isu64(ir->t) || irt_isu32(ir->t)) {
+        /* The cvtsd2si family of instructions operates on the signed integers,
+           producing INT_MIN on error.  However we're converting to an unsigned
+           integer, so we want to accept the whole unsigned integer range.
+           Convert both the number and the number minus INT_MIN, choosing the
+           first result if successful and the second otherwise.  */
+        Reg dest2 = ra_scratch(as, rset_exclude(RSET_GPR, dest));
+        Reg tmp2 = ra_scratch(as, rset_exclude(RSET_FPR, tmp));
+        x86Op sub_op;
+        void *krange;
+        if (st == IRT_NUM) {
+          sub_op = XO_SUBSD;
+          krange = &as->J->k64[irt_isu64(ir->t) ? LJ_K64_2P64 : LJ_K64_2P32];
+        } else {
+          sub_op = XO_SUBSS;
+          krange = &as->J->k32[irt_isu64(ir->t) ? LJ_K32_2P64 : LJ_K32_2P32];
+        }
+        emit_rr(as, XO_CMOV + (CC_O<<24), dest|r64, dest2|r64);
+        emit_i8(as, 1);
+        emit_rr(as, XO_ARITHi8, XOg_CMP|r64, dest);
+        emit_rr(as, op, dest2|r64, tmp2);
+        emit_rma(as, sub_op, tmp2, krange);
+        emit_rr(as, XO_MOVAPS, tmp2, tmp);
       }
+      emit_rr(as, op, dest|r64, tmp);
+      ra_left(as, tmp, lref);
     }
   } else if (st >= IRT_I8 && st <= IRT_U16) {  /* Extend to 32 bit integer. */
     Reg left, dest = ra_dest(as, ir, RSET_GPR);

src/lj_ccall.c

@@ -781,17 +781,24 @@ static unsigned int ccall_classify_struct(CTState *cts, CType *ct)
 {
   CTSize sz = ct->size;
   unsigned int r = 0, n = 0, isu = (ct->info & CTF_UNION);
-  while (ct->sib) {
+  while (ct->sib && n <= 4) {
+    unsigned int m = 1;
     CType *sct;
     ct = ctype_get(cts, ct->sib);
     if (ctype_isfield(ct->info)) {
       sct = ctype_rawchild(cts, ct);
+      if (ctype_isarray(sct->info)) {
+	CType *cct = ctype_rawchild(cts, sct);
+	if (!cct->size) continue;
+	m = sct->size / cct->size;
+	sct = cct;
+      }
       if (ctype_isfp(sct->info)) {
 	r |= sct->size;
-	if (!isu) n++; else if (n == 0) n = 1;
+	if (!isu) n += m; else if (n < m) n = m;
       } else if (ctype_iscomplex(sct->info)) {
 	r |= (sct->size >> 1);
-	if (!isu) n += 2; else if (n < 2) n = 2;
+	if (!isu) n += 2*m; else if (n < 2*m) n = 2*m;
       } else if (ctype_isstruct(sct->info)) {
 	goto substruct;
       } else {
@@ -803,10 +810,11 @@ static unsigned int ccall_classify_struct(CTState *cts, CType *ct)
       sct = ctype_rawchild(cts, ct);
     substruct:
       if (sct->size > 0) {
-	unsigned int s = ccall_classify_struct(cts, sct);
+	unsigned int s = ccall_classify_struct(cts, sct), sn;
 	if (s <= 1) goto noth;
 	r |= (s & 255);
-	if (!isu) n += (s >> 8); else if (n < (s >>8)) n = (s >> 8);
+	sn = (s >> 8) * m;
+	if (!isu) n += sn; else if (n < sn) n = sn;
       }
     }
   }

src/lj_jit.h

@@ -359,14 +359,10 @@ enum {
 
 enum {
 #if LJ_TARGET_X86ORX64
+  LJ_K64_2P32,		/* 2^32 */
   LJ_K64_TOBIT,		/* 2^52 + 2^51 */
   LJ_K64_2P64,		/* 2^64 */
   LJ_K64_M2P64,		/* -2^64 */
-#if LJ_32
-  LJ_K64_M2P64_31,	/* -2^64 or -2^31 */
-#else
-  LJ_K64_M2P64_31 = LJ_K64_M2P64,
-#endif
 #endif
 #if LJ_TARGET_MIPS
   LJ_K64_2P31,		/* 2^31 */
@@ -381,7 +377,8 @@ enum {
 
 enum {
 #if LJ_TARGET_X86ORX64
-  LJ_K32_M2P64_31,	/* -2^64 or -2^31 */
+  LJ_K32_2P32,		/* 2^32 */
+  LJ_K32_2P64,		/* 2^64 */
 #endif
 #if LJ_TARGET_PPC
   LJ_K32_2P52_2P31,	/* 2^52 + 2^31 */

src/lj_target_x86.h

@@ -300,6 +300,7 @@ typedef enum {
   XO_CVTSS2SD =	XO_f30f(5a),
   XO_CVTSD2SS =	XO_f20f(5a),
   XO_ADDSS =	XO_f30f(58),
+  XO_SUBSS =	XO_f30f(5c),
   XO_MOVD =	XO_660f(6e),
   XO_MOVDto =	XO_660f(7e),
 

src/lj_trace.c

@@ -318,12 +318,11 @@ void lj_trace_initstate(global_State *g)
 
   /* Initialize 32/64 bit constants. */
 #if LJ_TARGET_X86ORX64
+  J->k64[LJ_K64_2P32].u64 = U64x(41f00000,00000000);
   J->k64[LJ_K64_TOBIT].u64 = U64x(43380000,00000000);
-#if LJ_32
-  J->k64[LJ_K64_M2P64_31].u64 = U64x(c1e00000,00000000);
-#endif
   J->k64[LJ_K64_2P64].u64 = U64x(43f00000,00000000);
-  J->k32[LJ_K32_M2P64_31] = LJ_64 ? 0xdf800000 : 0xcf000000;
+  J->k32[LJ_K32_2P64] = 0x5f800000;
+  J->k32[LJ_K32_2P32] = 0x4f800000;
 #endif
 #if LJ_TARGET_X86ORX64 || LJ_TARGET_MIPS64
   J->k64[LJ_K64_M2P64].u64 = U64x(c3f00000,00000000);

src/vm_arm.dasc

@@ -1717,8 +1717,8 @@ static void build_subroutines(BuildCtx *ctx)
   |.endif
   |.endmacro
   |
-  |  math_minmax math_min, gt, pl
-  |  math_minmax math_max, lt, le
+  |  math_minmax math_min, gt, hs
+  |  math_minmax math_max, lt, ls
   |
   |//-- String library -----------------------------------------------------
   |