hipe: test unit size match in bs_append

This feature was previously missing and expressions such as
<<<<1:1>>/binary>> would succeed construction when compiled with HiPE.

A primop is_divisible is introduced to handle the case when the unit
size is not a power of two.

6 files changed, 48 insertions, 12 deletions

diff --git a/erts/emulator/hipe/hipe_bif0.tab b/erts/emulator/hipe/hipe_bif0.tab
index e3328c7d2c..5ce254314a 100644
--- a/erts/emulator/hipe/hipe_bif0.tab
+++ b/erts/emulator/hipe/hipe_bif0.tab
@@ -142,4 +142,4 @@ atom bs_validate_unicode
 atom bs_validate_unicode_retract
 atom emulate_fpe
 atom emasculate_binary
-
+atom is_divisible
diff --git a/erts/emulator/hipe/hipe_bif_list.m4 b/erts/emulator/hipe/hipe_bif_list.m4
index 6aa0c9a32e..7240280345 100644
--- a/erts/emulator/hipe/hipe_bif_list.m4
+++ b/erts/emulator/hipe/hipe_bif_list.m4
@@ -193,6 +193,7 @@ standard_bif_interface_2(nbif_rethrow, hipe_rethrow)
 standard_bif_interface_3(nbif_find_na_or_make_stub, hipe_find_na_or_make_stub)
 standard_bif_interface_2(nbif_nonclosure_address, hipe_nonclosure_address)
 nocons_nofail_primop_interface_0(nbif_fclearerror_error, hipe_fclearerror_error)
+standard_bif_interface_2(nbif_is_divisible, hipe_is_divisible)
 
 /*
  * Mbox primops with implicit P parameter.
diff --git a/erts/emulator/hipe/hipe_native_bif.c b/erts/emulator/hipe/hipe_native_bif.c
index 688378b2fe..119b0b0895 100644
--- a/erts/emulator/hipe/hipe_native_bif.c
+++ b/erts/emulator/hipe/hipe_native_bif.c
@@ -504,6 +504,18 @@ int hipe_bs_validate_unicode_retract(ErlBinMatchBuffer* mb, Eterm arg)
     return 1;
 }
 
+BIF_RETTYPE hipe_is_divisible(BIF_ALIST_2)
+{
+    /* Arguments are Eterm-sized unsigned integers */
+    Uint dividend = BIF_ARG_1;
+    Uint divisor = BIF_ARG_2;
+    if (dividend % divisor) {
+	BIF_ERROR(BIF_P, BADARG);
+    } else {
+	return NIL;
+    }
+}
+
 /* This is like the loop_rec_fr BEAM instruction
  */
 Eterm hipe_check_get_msg(Process *c_p)
diff --git a/erts/emulator/hipe/hipe_native_bif.h b/erts/emulator/hipe/hipe_native_bif.h
index 0e1a75f7eb..55a0d3bb1b 100644
--- a/erts/emulator/hipe/hipe_native_bif.h
+++ b/erts/emulator/hipe/hipe_native_bif.h
@@ -68,6 +68,7 @@ AEXTERN(Eterm,nbif_bs_put_utf16le,(Process*,Eterm,byte*,unsigned int));
 AEXTERN(Eterm,nbif_bs_get_utf16,(void));
 AEXTERN(Eterm,nbif_bs_validate_unicode,(Process*,Eterm));
 AEXTERN(Eterm,nbif_bs_validate_unicode_retract,(void));
+AEXTERN(void,nbif_is_divisible,(Process*,Uint,Uint));
 
 AEXTERN(void,nbif_select_msg,(Process*));
 AEXTERN(Eterm,nbif_cmp_2,(void));
@@ -93,6 +94,7 @@ BIF_RETTYPE hipe_bs_put_utf16le(BIF_ALIST_3);
 BIF_RETTYPE hipe_bs_validate_unicode(BIF_ALIST_1);
 struct erl_bin_match_buffer;
 int hipe_bs_validate_unicode_retract(struct erl_bin_match_buffer*, Eterm);
+BIF_RETTYPE hipe_is_divisible(BIF_ALIST_2);
 
 #ifdef NO_FPE_SIGNALS
 AEXTERN(void,nbif_emulate_fpe,(Process*));
diff --git a/erts/emulator/hipe/hipe_primops.h b/erts/emulator/hipe/hipe_primops.h
index adf7b1f382..0bec677574 100644
--- a/erts/emulator/hipe/hipe_primops.h
+++ b/erts/emulator/hipe/hipe_primops.h
@@ -68,6 +68,8 @@ PRIMOP_LIST(am_bs_get_utf16, &nbif_bs_get_utf16)
 PRIMOP_LIST(am_bs_validate_unicode, &nbif_bs_validate_unicode)
 PRIMOP_LIST(am_bs_validate_unicode_retract, &nbif_bs_validate_unicode_retract)
 
+PRIMOP_LIST(am_is_divisible, &nbif_is_divisible)
+
 PRIMOP_LIST(am_cmp_2, &nbif_cmp_2)
 PRIMOP_LIST(am_op_exact_eqeq_2, &nbif_eq_2)
 
diff --git a/lib/hipe/rtl/hipe_rtl_binary_construct.erl b/lib/hipe/rtl/hipe_rtl_binary_construct.erl
index d01625b4f9..97351d0142 100644
--- a/lib/hipe/rtl/hipe_rtl_binary_construct.erl
+++ b/lib/hipe/rtl/hipe_rtl_binary_construct.erl
@@ -313,15 +313,16 @@ gen_rtl(BsOP, Dst, Args, TrueLblName, FalseLblName, SystemLimitLblName, ConstTab
 	     hipe_rtl:mk_move(DstVar, Bin),
 	     hipe_rtl:mk_goto(TrueLblName)]; 
 
-	  {bs_append, _U, _F, _B, _Bla} -> 
+	  {bs_append, _U, _F, Unit, _Bla} ->
 	    [Size, Bin] = Args, 
 	    [DstVar, Base, Offset] = Dst, 
 	    [ProcBin] = create_vars(1),
 	    [Flags, SizeReg, IsWritable, EndSubSize, EndSubBitSize] = 
 	      create_regs(5),
-	    [ContLbl,ContLbl2,ContLbl3,WritableLbl,NotWritableLbl] = Lbls =
-	      create_lbls(5),
-	    [ContLblName, ContLbl2Name, ContLbl3Name, Writable, NotWritable] =
+	    [ContLbl,ContLbl2,ContLbl3,ContLbl4,WritableLbl,NotWritableLbl] =
+	      Lbls = create_lbls(6),
+	    [ContLblName, ContLbl2Name, ContLbl3Name, ContLbl4Name,
+	     Writable, NotWritable] =
 	      [hipe_rtl:label_name(Lbl) || Lbl <- Lbls],
 	    Zero = hipe_rtl:mk_imm(0),
 	    SubIsWritable = {sub_binary, is_writable},
@@ -339,17 +340,19 @@ gen_rtl(BsOP, Dst, Args, TrueLblName, FalseLblName, SystemLimitLblName, ConstTab
 	     get_field_from_term({proc_bin, flags}, ProcBin, Flags),
 	     hipe_rtl:mk_alub(Flags, Flags, 'and',
 			      hipe_rtl:mk_imm(?PB_IS_WRITABLE), 
-			      eq, NotWritable, Writable, 0.01),
+			      eq, NotWritable, ContLbl4Name, 0.01),
+	     ContLbl4,
+	     calculate_sizes(Bin, SizeReg, Offset, EndSubSize, EndSubBitSize),
+	     is_divisible(Offset, Unit, Writable, FalseLblName),
 	     WritableLbl,
 	     set_field_from_term(SubIsWritable, Bin, Zero),
 	     realloc_binary(SizeReg, ProcBin, Base),
-	     calculate_sizes(Bin, SizeReg, Offset, EndSubSize, EndSubBitSize),
 	     hipe_tagscheme:mk_sub_binary(DstVar, EndSubSize, Zero,
 					  EndSubBitSize, Zero,
 					  hipe_rtl:mk_imm(1), ProcBin),
 	     hipe_rtl:mk_goto(TrueLblName),
 	     NotWritableLbl,
-	     not_writable_code(Bin, SizeReg, DstVar, Base, Offset, 
+	     not_writable_code(Bin, SizeReg, DstVar, Base, Offset, Unit,
 			       TrueLblName, FalseLblName)]
 	end,
       {Code, ConstTab}
@@ -361,13 +364,15 @@ gen_rtl(BsOP, Dst, Args, TrueLblName, FalseLblName, SystemLimitLblName, ConstTab
 %%
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 
-not_writable_code(Bin, SizeReg, Dst, Base, Offset, 
+not_writable_code(Bin, SizeReg, Dst, Base, Offset, Unit,
 		  TrueLblName, FalseLblName) ->
   [SrcBase] = create_unsafe_regs(1),
   [SrcOffset, SrcSize, TotSize, TotBytes, UsedBytes] = create_regs(5),
-  [IncLbl,AllLbl] = Lbls = create_lbls(2),
-  [IncLblName,AllLblName] = get_label_names(Lbls),
+  [DivLbl,IncLbl,AllLbl] = Lbls = create_lbls(3),
+  [DivLblName,IncLblName,AllLblName] = get_label_names(Lbls),
   [get_base_offset_size(Bin, SrcBase, SrcOffset, SrcSize, FalseLblName),
+   is_divisible(SrcSize, Unit, DivLblName, FalseLblName),
+   DivLbl,
    hipe_rtl:mk_alu(TotSize, SrcSize, add, SizeReg),
    hipe_rtl:mk_alu(TotBytes, TotSize, add, ?LOW_BITS),
    hipe_rtl:mk_alu(TotBytes, TotBytes, srl, ?BYTE_SHIFT),
@@ -1376,4 +1381,18 @@ first_part(Var, Register, FalseLblName) ->
   SuccessLbl2,
   hipe_tagscheme:untag_fixnum(Register, Var)].
 
-
+is_divisible(_Dividend, 1, SuccLbl, _FailLbl) ->
+  [hipe_rtl:mk_goto(SuccLbl)];
+is_divisible(Dividend, Divisor, SuccLbl, FailLbl) ->
+  Log2 = floorlog2(Divisor),
+  case Divisor =:= 1 bsl Log2 of
+    true -> %% Divisor is a power of 2
+      %% Test that the Log2-1 lowest bits are clear
+      Mask = hipe_rtl:mk_imm(Divisor - 1),
+      [Tmp] = create_regs(1),
+      [hipe_rtl:mk_alub(Tmp, Dividend, 'and', Mask, eq, SuccLbl, FailLbl, 0.99)];
+    false ->
+      %% We need division, fall back to a primop
+      [hipe_rtl:mk_call([], is_divisible, [Dividend, hipe_rtl:mk_imm(Divisor)],
+			SuccLbl, FailLbl, not_remote)]
+  end.