hipe_rtl: unify branch and alub

branch and alub overlap in their use cases, but the backends rely on
knowing that the result is unused in their lowering of branch. By
extending alub so that the destination is optional, it can fully replace
branch.

This simplifies rtl by reducing code duplication and the number of
instructions.

Also, in the x86 and arm backends, we can now use 'test' and
{'tst','mvn','teq'} to lower some alubs without destinations. This is
particularly good for x86, as sequences such as 'is_boxed' type tests
now get shorter (both from not needing a mov to copy the variable, but
also from the fact that 'testb' encodes shorter than 'andq').

2 files changed, 46 insertions, 28 deletions

diff --git a/lib/hipe/x86/hipe_rtl_to_x86.erl b/lib/hipe/x86/hipe_rtl_to_x86.erl
index 4c8c98551c..ccb9b7632b 100644
--- a/lib/hipe/x86/hipe_rtl_to_x86.erl
+++ b/lib/hipe/x86/hipe_rtl_to_x86.erl
@@ -91,26 +91,31 @@ conv_insn(I, Map, Data) ->
     #alub{} ->
       %% dst = src1 op src2; if COND goto label
       BinOp = conv_binop(hipe_rtl:alub_op(I)),
-      {Dst, Map0} = conv_dst(hipe_rtl:alub_dst(I), Map),
-      {FixSrc1, Src1, Map1} = conv_src(hipe_rtl:alub_src1(I), Map0),
-      {FixSrc2, Src2, Map2} = conv_src(hipe_rtl:alub_src2(I), Map1),
+      {FixSrc1, Src1, Map0} = conv_src(hipe_rtl:alub_src1(I), Map),
+      {FixSrc2, Src2, Map1} = conv_src(hipe_rtl:alub_src2(I), Map0),
       Cc = conv_cond(hipe_rtl:alub_cond(I)),
-      I1 = [hipe_x86:mk_pseudo_jcc(Cc,
-				   hipe_rtl:alub_true_label(I),
-				   hipe_rtl:alub_false_label(I),
-				   hipe_rtl:alub_pred(I))],
-      I2 = conv_alu(Dst, Src1, BinOp, Src2, I1),
-      {FixSrc1++FixSrc2++I2, Map2, Data};
-    #branch{} ->
-      %% <unused> = src1 - src2; if COND goto label
-      {FixSrc1, Src1, Map0} = conv_src(hipe_rtl:branch_src1(I), Map),
-      {FixSrc2, Src2, Map1} = conv_src(hipe_rtl:branch_src2(I), Map0),
-      Cc = conv_cond(hipe_rtl:branch_cond(I)),
-      I2 = conv_branch(Src1, Cc, Src2,
-		       hipe_rtl:branch_true_label(I),
-		       hipe_rtl:branch_false_label(I),
-		       hipe_rtl:branch_pred(I)),
-      {FixSrc1++FixSrc2++I2, Map1, Data};
+      BranchOp = conv_branchop(BinOp),
+      HasDst = hipe_rtl:alub_has_dst(I),
+      {I2, Map3} =
+	case (not HasDst) andalso BranchOp =/= none of
+	  true ->
+	    {conv_branch(Src1, BranchOp, Src2, Cc,
+			 hipe_rtl:alub_true_label(I),
+			 hipe_rtl:alub_false_label(I),
+			 hipe_rtl:alub_pred(I)), Map1};
+	  false ->
+	    {Dst, Map2} =
+	      case HasDst of
+		false -> {new_untagged_temp(), Map1};
+		true -> conv_dst(hipe_rtl:alub_dst(I), Map1)
+	      end,
+	    I1 = [hipe_x86:mk_pseudo_jcc(Cc,
+					 hipe_rtl:alub_true_label(I),
+					 hipe_rtl:alub_false_label(I),
+					 hipe_rtl:alub_pred(I))],
+	    {conv_alu(Dst, Src1, BinOp, Src2, I1), Map2}
+	end,
+      {FixSrc1++FixSrc2++I2, Map3, Data};
     #call{} ->
       %%	push <arg1>
       %%	...
@@ -360,28 +365,41 @@ conv_shift(Dst, Src1, BinOp, Src2) ->
 %%% Finalise the conversion of a conditional branch operation, taking
 %%% care to not introduce more temps and moves than necessary.
 
-conv_branch(Src1, Cc, Src2, TrueLab, FalseLab, Pred) ->
+conv_branchop('sub') -> 'cmp';
+conv_branchop('and') ->  'test';
+conv_branchop(_) -> none.
+
+branchop_commutes('cmp') -> false;
+branchop_commutes('test') -> true.
+
+conv_branch(Src1, Op, Src2, Cc, TrueLab, FalseLab, Pred) ->
   case hipe_x86:is_imm(Src1) of
     false ->
-      mk_branch(Src1, Cc, Src2, TrueLab, FalseLab, Pred);
+      mk_branch(Src1, Op, Src2, Cc, TrueLab, FalseLab, Pred);
     true ->
       case hipe_x86:is_imm(Src2) of
 	false ->
-	  NewCc = commute_cc(Cc),
-	  mk_branch(Src2, NewCc, Src1, TrueLab, FalseLab, Pred);
+	  NewCc = case branchop_commutes(Op) of
+		    true -> Cc;
+		    false -> commute_cc(Cc)
+		  end,
+	  mk_branch(Src2, Op, Src1, NewCc, TrueLab, FalseLab, Pred);
 	true ->
 	  %% two immediates, let the optimiser clean it up
 	  Tmp = new_untagged_temp(),
 	  [hipe_x86:mk_move(Src1, Tmp) |
-	   mk_branch(Tmp, Cc, Src2, TrueLab, FalseLab, Pred)]
+	   mk_branch(Tmp, Op, Src2, Cc, TrueLab, FalseLab, Pred)]
       end
   end.
 
-mk_branch(Src1, Cc, Src2, TrueLab, FalseLab, Pred) ->
+mk_branch(Src1, Op, Src2, Cc, TrueLab, FalseLab, Pred) ->
   %% PRE: not(is_imm(Src1))
-  [hipe_x86:mk_cmp(Src2, Src1),
+  [mk_branchtest(Src1, Op, Src2),
    hipe_x86:mk_pseudo_jcc(Cc, TrueLab, FalseLab, Pred)].
 
+mk_branchtest(Src1, cmp, Src2) -> hipe_x86:mk_cmp(Src2, Src1);
+mk_branchtest(Src1, test, Src2) -> hipe_x86:mk_test(Src2, Src1).
+
 %%% Convert an RTL ALU or ALUB binary operator.
 
 conv_binop(BinOp) ->
diff --git a/lib/hipe/x86/hipe_x86.erl b/lib/hipe/x86/hipe_x86.erl
index 33d7f77cf1..e7c4497cda 100644
--- a/lib/hipe/x86/hipe_x86.erl
+++ b/lib/hipe/x86/hipe_x86.erl
@@ -201,7 +201,7 @@
 	 shift_src/1,
 	 shift_dst/1,
 
-	 %% mk_test/2,
+	 mk_test/2,
 	 test_src/1,
 	 test_dst/1,
 
@@ -305,7 +305,7 @@ mk_cmp(Src, Dst) -> #cmp{src=Src, dst=Dst}.
 cmp_src(#cmp{src=Src}) -> Src.
 cmp_dst(#cmp{dst=Dst}) -> Dst.
 
-%% mk_test(Src, Dst) -> #test{src=Src, dst=Dst}.
+mk_test(Src, Dst) -> #test{src=Src, dst=Dst}.
 test_src(#test{src=Src}) -> Src.
 test_dst(#test{dst=Dst}) -> Dst.