compiler: Change map instructions for fetching values

* Combine multiple get values with one instruction
* Combine multiple check keys with one instruction

15 files changed, 95 insertions, 43 deletions

diff --git a/lib/compiler/src/beam_a.erl b/lib/compiler/src/beam_a.erl
index 3dfa67a771..fe4f473846 100644
--- a/lib/compiler/src/beam_a.erl
+++ b/lib/compiler/src/beam_a.erl
@@ -92,6 +92,8 @@ rename_instr({put_map_assoc,Fail,S,D,R,L}) ->
     {put_map,Fail,assoc,S,D,R,L};
 rename_instr({put_map_exact,Fail,S,D,R,L}) ->
     {put_map,Fail,exact,S,D,R,L};
+rename_instr({test,has_map_fields,Fail,Src,{list,List}}) ->
+    {test,has_map_fields,Fail,[Src|List]};
 rename_instr({select_val=I,Reg,Fail,{list,List}}) ->
     {select,I,Reg,Fail,List};
 rename_instr({select_tuple_arity=I,Reg,Fail,{list,List}}) ->
diff --git a/lib/compiler/src/beam_asm.erl b/lib/compiler/src/beam_asm.erl
index 112b087f3c..f8cf178d2e 100644
--- a/lib/compiler/src/beam_asm.erl
+++ b/lib/compiler/src/beam_asm.erl
@@ -324,6 +324,8 @@ make_op({gc_bif,Bif,Fail,Live,Args,Dest}, Dict) ->
     encode_op(BifOp, [Fail,Live,{extfunc,erlang,Bif,Arity}|Args++[Dest]],Dict);
 make_op({bs_add=Op,Fail,[Src1,Src2,Unit],Dest}, Dict) ->
     encode_op(Op, [Fail,Src1,Src2,Unit,Dest], Dict);
+make_op({test,Cond,Fail,Src,{list,_}=Ops}, Dict) ->
+    encode_op(Cond, [Fail,Src,Ops], Dict);
 make_op({test,Cond,Fail,Ops}, Dict) when is_list(Ops) ->
     encode_op(Cond, [Fail|Ops], Dict);
 make_op({test,Cond,Fail,Live,[Op|Ops],Dst}, Dict) when is_list(Ops) ->
diff --git a/lib/compiler/src/beam_block.erl b/lib/compiler/src/beam_block.erl
index 3723cc19e1..7a30c68593 100644
--- a/lib/compiler/src/beam_block.erl
+++ b/lib/compiler/src/beam_block.erl
@@ -154,8 +154,8 @@ collect({get_list,S,D1,D2})  -> {set,[D1,D2],[S],get_list};
 collect(remove_message)      -> {set,[],[],remove_message};
 collect({put_map,F,Op,S,D,R,{list,Puts}}) ->
     {set,[D],[S|Puts],{alloc,R,{put_map,Op,F}}};
-collect({get_map_element,F,S,K,D}) ->
-    {set,[D],[S],{get_map_element,K,F}};
+collect({get_map_elements,F,S,{list,Gets}}) ->
+    {set,Gets,[S],{get_map_elements,F}};
 collect({'catch',R,L})       -> {set,[R],[],{'catch',L}};
 collect(fclearerror)         -> {set,[],[],fclearerror};
 collect({fcheckerror,{f,0}}) -> {set,[],[],fcheckerror};
@@ -240,7 +240,7 @@ move_allocates_2(Alloc, [], Acc) ->
 
 alloc_may_pass({set,_,_,{alloc,_,_}}) -> false;
 alloc_may_pass({set,_,_,{set_tuple_element,_}}) -> false;
-alloc_may_pass({set,_,_,{get_map_element,_,_}}) -> false;
+alloc_may_pass({set,_,_,{get_map_elements,_}}) -> false;
 alloc_may_pass({set,_,_,put_list}) -> false;
 alloc_may_pass({set,_,_,put}) -> false;
 alloc_may_pass({set,_,_,_}) -> true.
@@ -291,7 +291,11 @@ opt_moves([X0,Y0], Is0) ->
 	not_possible -> {[X,Y0],Is2};
 	{X,_} -> {[X,Y0],Is2};
 	{Y,Is} -> {[X,Y],Is}
-    end.
+    end;
+opt_moves(Ds, Is) ->
+    %% multiple destinations -> pass through
+    {Ds,Is}.
+
 
 %% opt_move(Dest, [Instruction]) -> {UpdatedDest,[Instruction]} | not_possible
 %%  If there is a {move,Dest,FinalDest} instruction
diff --git a/lib/compiler/src/beam_clean.erl b/lib/compiler/src/beam_clean.erl
index 55f985ad0e..b653998252 100644
--- a/lib/compiler/src/beam_clean.erl
+++ b/lib/compiler/src/beam_clean.erl
@@ -262,8 +262,8 @@ replace([{bs_utf16_size=I,{f,Lbl},Src,Dst}|Is], Acc, D) when Lbl =/= 0 ->
 replace([{put_map=I,{f,Lbl},Op,Src,Dst,Live,List}|Is], Acc, D)
   when Lbl =/= 0 ->
     replace(Is, [{I,{f,label(Lbl, D)},Op,Src,Dst,Live,List}|Acc], D);
-replace([{get_map_element=I,{f,Lbl},Src,Key,Dst}|Is], Acc, D) when Lbl =/= 0 ->
-    replace(Is, [{I,{f,label(Lbl, D)},Src,Key,Dst}|Acc], D);
+replace([{get_map_elements=I,{f,Lbl},Src,List}|Is], Acc, D) when Lbl =/= 0 ->
+    replace(Is, [{I,{f,label(Lbl, D)},Src,List}|Acc], D);
 replace([I|Is], Acc, D) ->
     replace(Is, [I|Acc], D);
 replace([], Acc, _) -> Acc.
diff --git a/lib/compiler/src/beam_disasm.erl b/lib/compiler/src/beam_disasm.erl
index 57fdf95677..3f5e9df5fd 100644
--- a/lib/compiler/src/beam_disasm.erl
+++ b/lib/compiler/src/beam_disasm.erl
@@ -1151,12 +1151,12 @@ resolve_inst({is_map,Args0},_,_,_) ->
     {test, is_map, FLbl, Args};
 
 resolve_inst({has_map_field,Args0},_,_,_) ->
-    [FLbl|Args] = resolve_args(Args0),
-    {test,has_map_field,FLbl,Args};
+    [FLbl,Src,{u,_Len}|Args] = resolve_args(Args0),
+    {test,has_map_field,FLbl,Src,{list,Args}};
 
-resolve_inst({get_map_element,Args},_,_,_) ->
-    [FLbl,Src,Key,Dst] = resolve_args(Args),
-    {get_map_element,FLbl,Src,Key,Dst};
+resolve_inst({get_map_element,Args0},_,_,_) ->
+    [FLbl,Src,{u,_Len}|Args] = resolve_args(Args0),
+    {get_map_element,FLbl,Src,{list,Args}};
 
 %%
 %% Catches instructions that are not yet handled.
diff --git a/lib/compiler/src/beam_flatten.erl b/lib/compiler/src/beam_flatten.erl
index 534bc6d954..46835bece1 100644
--- a/lib/compiler/src/beam_flatten.erl
+++ b/lib/compiler/src/beam_flatten.erl
@@ -63,8 +63,8 @@ norm({set,[],[S,D],{set_tuple_element,I}}) -> {set_tuple_element,S,D,I};
 norm({set,[D1,D2],[S],get_list})          -> {get_list,S,D1,D2};
 norm({set,[D],[S|Puts],{alloc,R,{put_map,Op,F}}}) ->
     {put_map,F,Op,S,D,R,{list,Puts}};
-norm({set,[D],[S],{get_map_element,K,F}}) ->
-    {get_map_element,F,S,K,D};
+norm({set,Gets,[S],{get_map_elements,F}}) ->
+    {get_map_elements,F,S,{list,Gets}};
 norm({set,[],[],remove_message})   -> remove_message;
 norm({set,[],[],fclearerror}) -> fclearerror;
 norm({set,[],[],fcheckerror}) -> {fcheckerror,{f,0}}.
diff --git a/lib/compiler/src/beam_jump.erl b/lib/compiler/src/beam_jump.erl
index 1f720b94c3..0fc8d45c80 100644
--- a/lib/compiler/src/beam_jump.erl
+++ b/lib/compiler/src/beam_jump.erl
@@ -529,7 +529,7 @@ ulbl({bs_put,Lbl,_,_}, Used) ->
     mark_used(Lbl, Used);
 ulbl({put_map,Lbl,_Op,_Src,_Dst,_Live,_List}, Used) ->
     mark_used(Lbl, Used);
-ulbl({get_map_element,Lbl,_Src,_Key,_Dst}, Used) ->
+ulbl({get_map_elements,Lbl,_Src,_List}, Used) ->
     mark_used(Lbl, Used);
 ulbl(_, Used) -> Used.
 
diff --git a/lib/compiler/src/beam_split.erl b/lib/compiler/src/beam_split.erl
index 638a4826ea..688bba9a94 100644
--- a/lib/compiler/src/beam_split.erl
+++ b/lib/compiler/src/beam_split.erl
@@ -53,9 +53,9 @@ split_block([{set,[D],[S|Puts],{alloc,R,{put_map,Op,{f,Lbl}=Fail}}}|Is],
 	    Bl, Acc) when Lbl =/= 0 ->
     split_block(Is, [], [{put_map,Fail,Op,S,D,R,{list,Puts}}|
 			 make_block(Bl, Acc)]);
-split_block([{set,[D],[S],{get_map_element,K,{f,Lbl}=Fail}}|Is], Bl, Acc)
+split_block([{set,Gets,[S],{get_map_elements,{f,Lbl}=Fail}}|Is], Bl, Acc)
   when Lbl =/= 0 ->
-    split_block(Is, [], [{get_map_element,Fail,S,K,D}|make_block(Bl, Acc)]);
+    split_block(Is, [], [{get_map_elements,Fail,S,{list,Gets}}|make_block(Bl, Acc)]);
 split_block([{set,[R],[],{'catch',L}}|Is], Bl, Acc) ->
     split_block(Is, [], [{'catch',R,L}|make_block(Bl, Acc)]);
 split_block([{set,[],[],{line,_}=Line}|Is], Bl, Acc) ->
diff --git a/lib/compiler/src/beam_utils.erl b/lib/compiler/src/beam_utils.erl
index a3f16cfa8f..27034aecce 100644
--- a/lib/compiler/src/beam_utils.erl
+++ b/lib/compiler/src/beam_utils.erl
@@ -185,6 +185,7 @@ is_pure_test({test,is_lt,_,[_,_]}) -> true;
 is_pure_test({test,is_nil,_,[_]}) -> true;
 is_pure_test({test,is_nonempty_list,_,[_]}) -> true;
 is_pure_test({test,test_arity,_,[_,_]}) -> true;
+is_pure_test({test,has_map_fields,_,[_,{list,_}]}) -> true;
 is_pure_test({test,Op,_,Ops}) -> 
     erl_internal:new_type_test(Op, length(Ops)).
 
diff --git a/lib/compiler/src/beam_validator.erl b/lib/compiler/src/beam_validator.erl
index 682f7adbc2..9f8f1cd3f5 100644
--- a/lib/compiler/src/beam_validator.erl
+++ b/lib/compiler/src/beam_validator.erl
@@ -770,6 +770,9 @@ valfun_4({test,is_nonempty_list,{f,Lbl},[Cons]}, Vst) ->
 valfun_4({test,test_arity,{f,Lbl},[Tuple,Sz]}, Vst) when is_integer(Sz) ->
     assert_type(tuple, Tuple, Vst),
     set_type_reg({tuple,Sz}, Tuple, branch_state(Lbl, Vst));
+valfun_4({test,_Op,{f,Lbl},Src,{list,_}}, Vst) ->
+    validate_src([Src], Vst),
+    branch_state(Lbl, Vst);
 valfun_4({test,_Op,{f,Lbl},Src}, Vst) ->
     validate_src(Src, Vst),
     branch_state(Lbl, Vst);
@@ -871,14 +874,21 @@ valfun_4({put_map_assoc,{f,Fail},Src,Dst,Live,{list,List}}, Vst) ->
     verify_put_map(Fail, Src, Dst, Live, List, Vst);
 valfun_4({put_map_exact,{f,Fail},Src,Dst,Live,{list,List}}, Vst) ->
     verify_put_map(Fail, Src, Dst, Live, List, Vst);
-valfun_4({get_map_element,{f,Fail},Src,Key,Dst}, Vst0) ->
-    assert_term(Src, Vst0),
-    assert_term(Key, Vst0),
-    Vst = branch_state(Fail, Vst0),
-    set_type_reg(term, Dst, Vst);
+valfun_4({get_map_elements,{f,Fail},Src,{list,List}}, Vst) ->
+    verify_get_map(Fail, Src, List, Vst);
 valfun_4(_, _) ->
     error(unknown_instruction).
 
+verify_get_map(Fail, Src, List, Vst0) ->
+    assert_term(Src, Vst0),
+    Vst1 = branch_state(Fail, Vst0),
+    verify_get_map_pair(List,Vst0,Vst1).
+
+verify_get_map_pair([],_,Vst) -> Vst;
+verify_get_map_pair([Src,Dst|Vs],Vst0,Vsti) ->
+    assert_term(Src, Vst0),
+    verify_get_map_pair(Vs,Vst0,set_type_reg(term,Dst,Vsti)).
+
 verify_put_map(Fail, Src, Dst, Live, List, Vst0) ->
     verify_live(Live, Vst0),
     verify_y_init(Vst0),
diff --git a/lib/compiler/src/beam_z.erl b/lib/compiler/src/beam_z.erl
index 9953a48710..c204eac25d 100644
--- a/lib/compiler/src/beam_z.erl
+++ b/lib/compiler/src/beam_z.erl
@@ -78,6 +78,8 @@ undo_rename({put_map,Fail,assoc,S,D,R,L}) ->
     {put_map_assoc,Fail,S,D,R,L};
 undo_rename({put_map,Fail,exact,S,D,R,L}) ->
     {put_map_exact,Fail,S,D,R,L};
+undo_rename({test,has_map_fields,Fail,[Src|List]}) ->
+    {test,has_map_fields,Fail,Src,{list,List}};
 undo_rename({select,I,Reg,Fail,List}) ->
     {I,Reg,Fail,{list,List}};
 undo_rename(I) -> I.
diff --git a/lib/compiler/src/genop.tab b/lib/compiler/src/genop.tab
index 79b467f949..7d6bf56ccb 100755
--- a/lib/compiler/src/genop.tab
+++ b/lib/compiler/src/genop.tab
@@ -529,10 +529,10 @@ BEAM_FORMAT_NUMBER=0
 
 153: line/1
 
-# R16
+# R17
 
 154: put_map_assoc/5
 155: put_map_exact/5
 156: is_map/2
-157: has_map_field/3
-158: get_map_element/4
+157: has_map_fields/3
+158: get_map_elements/3
diff --git a/lib/compiler/src/v3_codegen.erl b/lib/compiler/src/v3_codegen.erl
index 4d155c0fd0..e00ee1f3ad 100644
--- a/lib/compiler/src/v3_codegen.erl
+++ b/lib/compiler/src/v3_codegen.erl
@@ -938,22 +938,39 @@ select_map_val(V, Es, B, Fail, I, Vdb, Bef, St0) ->
     {Bis,Aft,St2} = match_cg(B, Fail, Int, St1),
     {Eis++Bis,Aft,St2}.
 
+select_extract_map(_, [], _, _, _, Bef, St) -> {[],Bef,St};
 select_extract_map(Src, Vs, Fail, I, Vdb, Bef, St) ->
-    F = fun ({map_pair,Key,{var,V}}, Int0) ->
-		Rsrc = fetch_var(Src, Int0),
+    %% First split the instruction flow
+    %% We want one set of each
+    %% 1) has_map_fields (no target registers)
+    %% 2) get_map_elements (with target registers)
+    %% Assume keys are term-sorted
+    Rsrc = fetch_var(Src, Bef),
+
+    {{HasKs,GetVs},Aft} = lists:foldr(fun
+	    ({map_pair,Key,{var,V}},{{HasKsi,GetVsi},Int0}) ->
 		case vdb_find(V, Vdb) of
 		    {V,_,L} when L =< I ->
-			{[{test,has_map_field,{f,Fail},[Rsrc,Key]}],Int0};
+			{{[Key|HasKsi],GetVsi},Int0};
 		    _Other ->
 			Reg1 = put_reg(V, Int0#sr.reg),
 			Int1 = Int0#sr{reg=Reg1},
-			{[{get_map_element,{f,Fail},
-			   Rsrc,Key,fetch_reg(V, Reg1)}],
-			 Int1}
+			{{HasKsi,[Key,fetch_reg(V, Reg1)|GetVsi]},Int1}
 		end
-	end,
-    {Es,Aft} = flatmapfoldl(F, Bef, Vs),
-    {Es,Aft,St}.
+	end, {{[],[]},Bef}, Vs),
+
+    Code = case {HasKs,GetVs} of
+	{[],[]} -> {[],Aft,St};
+	{HasKs,[]} ->
+	    [{test,has_map_fields,{f,Fail},Rsrc,{list,HasKs}}];
+	{[],GetVs} ->
+	    [{get_map_elements,   {f,Fail},Rsrc,{list,GetVs}}];
+	{HasKs,GetVs} ->
+	    [{test,has_map_fields,{f,Fail},Rsrc,{list,HasKs}},
+	     {get_map_elements,   {f,Fail},Rsrc,{list,GetVs}}]
+    end,
+    {Code, Aft, St}.
+
 
 select_extract_cons(Src, [{var,Hd}, {var,Tl}], I, Vdb, Bef, St) ->
     {Es,Aft} = case {vdb_find(Hd, Vdb), vdb_find(Tl, Vdb)} of
diff --git a/lib/compiler/src/v3_kernel.erl b/lib/compiler/src/v3_kernel.erl
index bc5ca0314a..5675572092 100644
--- a/lib/compiler/src/v3_kernel.erl
+++ b/lib/compiler/src/v3_kernel.erl
@@ -496,7 +496,6 @@ translate_match_fail_1(Anno, As, Sub, #kern{ff=FF}) ->
 translate_fc(Args) ->
     [#c_literal{val=function_clause},make_list(Args)].
 
-    %{Kes,Ep,St2} = map_pairs(Ces, Sub, St1),
 map_split_pairs(A, Var, Ces, Sub, St0) ->
     %% two steps
     %% 1. force variables
@@ -718,12 +717,8 @@ pattern(#c_tuple{anno=A,es=Ces}, Isub, Osub0, St0) ->
     {Kes,Osub1,St1} = pattern_list(Ces, Isub, Osub0, St0),
     {#k_tuple{anno=A,es=Kes},Osub1,St1};
 pattern(#c_map{anno=A,es=Ces}, Isub, Osub0, St0) ->
-    {Kes,Osub1,St1} = pattern_list(Ces, Isub, Osub0, St0),
+    {Kes,Osub1,St1} = pattern_map_pairs(Ces, Isub, Osub0, St0),
     {#k_map{anno=A,op=exact,es=Kes},Osub1,St1};
-pattern(#c_map_pair{op=#c_literal{val=exact},anno=A,key=Ck,val=Cv},Isub, Osub0, St0) ->
-    {Kk,Osub1,St1} = pattern(Ck, Isub, Osub0, St0),
-    {Kv,Osub2,St2} = pattern(Cv, Isub, Osub1, St1),
-    {#k_map_pair{anno=A,key=Kk,val=Kv},Osub2,St2};
 pattern(#c_binary{anno=A,segments=Cv}, Isub, Osub0, St0) ->
     {Kv,Osub1,St1} = pattern_bin(Cv, Isub, Osub0, St0),
     {#k_binary{anno=A,segs=Kv},Osub1,St1};
@@ -738,6 +733,25 @@ flatten_alias(#c_alias{var=V,pat=P}) ->
     {[V|Vs],Pat};
 flatten_alias(Pat) -> {[],Pat}.
 
+pattern_map_pairs(Ces0, Isub, Osub0, St0) ->
+    %% It is assumed that all core keys are literals
+    %% It is later assumed that these keys are term sorted
+    %% so we need to sort them here
+    Ces1 = lists:sort(fun
+	    (#c_map_pair{key=CkA},#c_map_pair{key=CkB}) ->
+		A = core_lib:literal_value(CkA),
+		B = core_lib:literal_value(CkB),
+		erts_internal:cmp_term(A,B) < 0
+	end, Ces0),
+    %% pattern the pair keys and values as normal
+    {Kes,{Osub1,St1}} = lists:mapfoldl(fun
+	    (#c_map_pair{anno=A,key=Ck,val=Cv},{Osubi0,Sti0}) ->
+		{Kk,Osubi1,Sti1} = pattern(Ck, Isub, Osubi0, Sti0),
+		{Kv,Osubi2,Sti2} = pattern(Cv, Isub, Osubi1, Sti1),
+		{#k_map_pair{anno=A,key=Kk,val=Kv},{Osubi2,Sti2}}
+	end, {Osub0, St0}, Ces1),
+    {Kes,Osub1,St1}.
+
 pattern_bin(Es, Isub, Osub0, St0) ->
     {Kbin,{_,Osub},St} = pattern_bin_1(Es, Isub, Osub0, St0),
     {Kbin,Osub,St}.
@@ -1388,13 +1402,13 @@ get_match(#k_bin_int{}=BinInt, St0) ->
 get_match(#k_tuple{es=Es}, St0) ->
     {Mes,St1} = new_vars(length(Es), St0),
     {#k_tuple{es=Mes},Mes,St1};
-get_match(#k_map{es=Es0}, St0) ->
+get_match(#k_map{op=exact,es=Es0}, St0) ->
     {Mes,St1} = new_vars(length(Es0), St0),
     {Es,_} = mapfoldl(fun
 	    (#k_map_pair{}=Pair, [V|Vs]) ->
 		{Pair#k_map_pair{val=V},Vs}
 	end, Mes, Es0),
-    {#k_map{es=Es},Mes,St1};
+    {#k_map{op=exact,es=Es},Mes,St1};
 get_match(M, St) ->
     {M,[],St}.
 
@@ -1519,7 +1533,7 @@ arg_val(Arg, C) ->
 		    end || Pair <- Es],
 	    %% multiple keys may have the same name
 	    %% do not use ordsets
-	    lists:sort(Keys)
+	    lists:sort(fun(A,B) -> erts_internal:cmp_term(A,B) < 0 end, Keys)
     end.
 
 %% ubody_used_vars(Expr, State) -> [UsedVar]
diff --git a/lib/compiler/src/v3_kernel_pp.erl b/lib/compiler/src/v3_kernel_pp.erl
index 639c6737e2..b4e486f97c 100644
--- a/lib/compiler/src/v3_kernel_pp.erl
+++ b/lib/compiler/src/v3_kernel_pp.erl
@@ -115,7 +115,7 @@ format_1(#k_map{op=assoc,es=Es}, Ctxt) ->
      format_hseq(Es, ",", ctxt_bump_indent(Ctxt, 1), fun format/2),
      "}~"
     ];
-format_1(#k_map{op=exact,es=Es}, Ctxt) ->
+format_1(#k_map{es=Es}, Ctxt) ->
     ["::{",
      format_hseq(Es, ",", ctxt_bump_indent(Ctxt, 1), fun format/2),
      "}::"