v3_codegen: Reduce cost for fixing up bs_match_string instructions

Commit b76588fb5a introduced an optimization of the compile time of
huge functions with many bs_match_string instructions. The
optimization is done in two passes. The first pass coalesces adjacent
bs_match_string instructions. To avoid copying bitstrings multiple
times, the bitstrings in the instructions are combined in to a (deep)
list. The second pass goes through all instructions in the function
and combines the list of bitstrings to a single bitstring in all
bs_match_string instructions.

The second pass (fix_bs_match_string) is run on all instructions in
each function, even if there are no bs_match_instructions in the
function. While fix_bs_match_string is not a bottleneck (it is a
linear pass), its execution time is noticeable when profiling some
modules.

Move the execution of the second pass to the select_binary()
function so that it will only be executed for instructions that
do binary matching. Also take the opportunity to optimize away
uses of bs_restore2 that occour directly after a bs_save2. That
optimimization is currently done in beam_block, but it can be
done essentially for free in the same pass that fixes up
bs_match_string instructions.

2 files changed, 24 insertions, 27 deletions

diff --git a/lib/compiler/src/beam_block.erl b/lib/compiler/src/beam_block.erl
index 5216f39296..e2639e9cac 100644
--- a/lib/compiler/src/beam_block.erl
+++ b/lib/compiler/src/beam_block.erl
@@ -61,15 +61,6 @@ blockify(Is) ->
 blockify([{loop_rec,{f,Fail},{x,0}},{loop_rec_end,_Lbl},{label,Fail}|Is], Acc) ->
     %% Useless instruction sequence.
     blockify(Is, Acc);
-
-%% New bit syntax matching.
-blockify([{bs_save2,R,Point}=I,{bs_restore2,R,Point}|Is], Acc) ->
-    blockify([I|Is], Acc);
-blockify([{bs_save2,R,Point}=I,{test,is_eq_exact,_,_}=Test,
-	  {bs_restore2,R,Point}|Is], Acc) ->
-    blockify([I,Test|Is], Acc);
-
-%% Do other peep-hole optimizations.
 blockify([{test,is_atom,{f,Fail},[Reg]}=I|
 	  [{select,select_val,Reg,{f,Fail},
 	    [{atom,false},{f,_}=BrFalse,
diff --git a/lib/compiler/src/v3_codegen.erl b/lib/compiler/src/v3_codegen.erl
index eb7926d3ab..15a54a5886 100644
--- a/lib/compiler/src/v3_codegen.erl
+++ b/lib/compiler/src/v3_codegen.erl
@@ -121,24 +121,15 @@ cg_fun(Les, Hvs, Vdb, AtomMod, NameArity, Anno, St0) ->
 					   put_reg(V, Reg)
 				   end, [], Hvs),
 			 stk=[]}, 0, Vdb),
-    {B0,_Aft,St} = cg_list(Les, 0, Vdb, Bef,
+    {B,_Aft,St} = cg_list(Les, 0, Vdb, Bef,
 			  St3#cg{bfail=0,
 				 ultimate_failure=UltimateMatchFail,
 				 is_top_block=true}),
-    B = fix_bs_match_strings(B0),
     {Name,Arity} = NameArity,
     Asm = [{label,Fi},line(Anno),{func_info,AtomMod,{atom,Name},Arity},
 	   {label,Fl}|B++[{label,UltimateMatchFail},if_end]],
     {Asm,Fl,St}.
 
-fix_bs_match_strings([{test,bs_match_string,F,[Ctx,BinList]}|Is])
-  when is_list(BinList) ->
-    I = {test,bs_match_string,F,[Ctx,list_to_bitstring(BinList)]},
-    [I|fix_bs_match_strings(Is)];
-fix_bs_match_strings([I|Is]) ->
-    [I|fix_bs_match_strings(Is)];
-fix_bs_match_strings([]) -> [].
-
 %% cg(Lkexpr, Vdb, StackReg, State) -> {[Ainstr],StackReg,State}.
 %%  Generate code for a kexpr.
 %%  Split function into two steps for clarity, not efficiency.
@@ -717,22 +708,37 @@ select_nil(#l{ke={val_clause,nil,B}}, V, Tf, Vf, Bef, St0) ->
 select_binary(#l{ke={val_clause,{binary,{var,V}},B},i=I,vdb=Vdb},
 	      V, Tf, Vf, Bef, St0) ->
     Int0 = clear_dead(Bef#sr{reg=Bef#sr.reg}, I, Vdb),
-    {Bis,Aft,St1} = match_cg(B, Vf, Int0, St0),
+    {Bis0,Aft,St1} = match_cg(B, Vf, Int0, St0),
     CtxReg = fetch_var(V, Int0),
     Live = max_reg(Bef#sr.reg),
-    {[{test,bs_start_match2,{f,Tf},Live,[CtxReg,V],CtxReg},
-      {bs_save2,CtxReg,{V,V}}|Bis],
-     Aft,St1};
+    Bis1 = [{test,bs_start_match2,{f,Tf},Live,[CtxReg,V],CtxReg},
+	    {bs_save2,CtxReg,{V,V}}|Bis0],
+    Bis = finish_select_binary(Bis1),
+    {Bis,Aft,St1};
 select_binary(#l{ke={val_clause,{binary,{var,Ivar}},B},i=I,vdb=Vdb},
 	      V, Tf, Vf, Bef, St0) ->
     Regs = put_reg(Ivar, Bef#sr.reg),
     Int0 = clear_dead(Bef#sr{reg=Regs}, I, Vdb),
-    {Bis,Aft,St1} = match_cg(B, Vf, Int0, St0),
+    {Bis0,Aft,St1} = match_cg(B, Vf, Int0, St0),
     CtxReg = fetch_var(Ivar, Int0),
     Live = max_reg(Bef#sr.reg),
-    {[{test,bs_start_match2,{f,Tf},Live,[fetch_var(V, Bef),Ivar],CtxReg},
-      {bs_save2,CtxReg,{Ivar,Ivar}}|Bis],
-     Aft,St1}.
+    Bis1 = [{test,bs_start_match2,{f,Tf},Live,[fetch_var(V, Bef),Ivar],CtxReg},
+	    {bs_save2,CtxReg,{Ivar,Ivar}}|Bis0],
+    Bis = finish_select_binary(Bis1),
+    {Bis,Aft,St1}.
+
+finish_select_binary([{bs_save2,R,Point}=I,{bs_restore2,R,Point}|Is]) ->
+    [I|finish_select_binary(Is)];
+finish_select_binary([{bs_save2,R,Point}=I,{test,is_eq_exact,_,_}=Test,
+		      {bs_restore2,R,Point}|Is]) ->
+    [I,Test|finish_select_binary(Is)];
+finish_select_binary([{test,bs_match_string,F,[Ctx,BinList]}|Is])
+  when is_list(BinList) ->
+    I = {test,bs_match_string,F,[Ctx,list_to_bitstring(BinList)]},
+    [I|finish_select_binary(Is)];
+finish_select_binary([I|Is]) ->
+    [I|finish_select_binary(Is)];
+finish_select_binary([]) -> [].
 
 %% New instructions for selection of binary segments.