Eliminate use of GetArg1() in the select_val instruction

Instead of having one i_select_val_sfI instruction that uses
the GetArg1() macro to fetch the controlling expression, use
three separate instructions for each of the register types.
That will save one word when selecting on the {x,0} register.
It should also be slightly faster since a conditional branch
is eliminated.

Although it seems that the BEAM compiler will never generate
a constant controlling expression (even with optimizations
turned off), we still make sure that they will work by
evaluating the select_val instruction at load time.

Handle the select_tuple_arity instruction in the same way.

4 files changed, 63 insertions, 24 deletions

diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 6c16c24d0d..46b831fa88 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -555,7 +555,9 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
     unpacked = ap;
     ap = addr + size;
     switch (op) {
-    case op_i_select_val_sfI:
+    case op_i_select_val_rfI:
+    case op_i_select_val_xfI:
+    case op_i_select_val_yfI:
 	{
 	    int n = ap[-1];
 
@@ -567,7 +569,9 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    }
 	}
 	break;
-    case op_i_select_tuple_arity_sfI:
+    case op_i_select_tuple_arity_rfI:
+    case op_i_select_tuple_arity_xfI:
+    case op_i_select_tuple_arity_yfI:
 	{
 	    int n = ap[-1];
 
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 09f6fc8cef..6cd7fa78e0 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -1865,9 +1865,19 @@ void process_main(void)
  {
      Eterm select_val;
 
- OpCase(i_select_tuple_arity_sfI):
-     GetArg1(0, select_val);
+ OpCase(i_select_tuple_arity_xfI):
+     select_val = xb(Arg(0));
+     goto do_select_tuple_arity;
 
+ OpCase(i_select_tuple_arity_yfI):
+     select_val = yb(Arg(0));
+     goto do_select_tuple_arity;
+
+ OpCase(i_select_tuple_arity_rfI):
+     select_val = r(0);
+     I--;
+
+ do_select_tuple_arity:
      if (is_tuple(select_val)) {
 	 select_val = *tuple_val(select_val);
 	 goto do_binary_search;
@@ -1875,9 +1885,17 @@ void process_main(void)
      SET_I((BeamInstr *) Arg(1));
      Goto(*I);
 
+ OpCase(i_select_val_xfI):
+     select_val = xb(Arg(0));
+     goto do_binary_search;
 
- OpCase(i_select_val_sfI):
-     GetArg1(0, select_val);
+ OpCase(i_select_val_yfI):
+     select_val = yb(Arg(0));
+     goto do_binary_search;
+     
+ OpCase(i_select_val_rfI):
+     select_val = r(0);
+     I--;
 
  do_binary_search:
  {
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 73f057929e..54c8ad0eb1 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -3282,7 +3282,6 @@ const_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail,
     int i;
 
     ASSERT(Size.type == TAG_u);
-    ASSERT(S.type == TAG_q);
 
     NEW_GENOP(stp, op);
     op->next = NULL;
@@ -3293,18 +3292,32 @@ const_select_val(LoaderState* stp, GenOpArg S, GenOpArg Fail,
      * Search for a literal matching the controlling expression.
      */
 
-    if (S.type == TAG_q) {
-	Eterm expr = stp->literals[S.val].term;
-	for (i = 0; i < Size.val; i += 2) {
-	    if (Rest[i].type == TAG_q) {
-		Eterm term = stp->literals[Rest[i].val].term;
-		if (eq(term, expr)) {
-		    ASSERT(Rest[i+1].type == TAG_f);
-		    op->a[0] = Rest[i+1];
-		    return op;
+    switch (S.type) {
+    case TAG_q:
+	{
+	    Eterm expr = stp->literals[S.val].term;
+	    for (i = 0; i < Size.val; i += 2) {
+		if (Rest[i].type == TAG_q) {
+		    Eterm term = stp->literals[Rest[i].val].term;
+		    if (eq(term, expr)) {
+			ASSERT(Rest[i+1].type == TAG_f);
+			op->a[0] = Rest[i+1];
+			return op;
+		    }
 		}
 	    }
 	}
+	break;
+    case TAG_i:
+    case TAG_a:
+	for (i = 0; i < Size.val; i += 2) {
+	    if (Rest[i].val == S.val && Rest[i].type == S.type) {
+		ASSERT(Rest[i+1].type == TAG_f);
+		op->a[0] = Rest[i+1];
+		return op;
+	    }
+	}
+	break;
     }
 
     /*
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index f629cc80b6..45b2e197b4 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -124,7 +124,7 @@ init Y1 | init Y2 => init2 Y1 Y2
 
 # Selecting values
 
-select_val S=q Fail=f Size=u Rest=* => const_select_val(S, Fail, Size, Rest)
+select_val S=aiq Fail=f Size=u Rest=* => const_select_val(S, Fail, Size, Rest)
 
 select_val S=s Fail=f Size=u Rest=* | use_jump_tab(Size, Rest) => \
   gen_jump_tab(S, Fail, Size, Rest)
@@ -139,26 +139,30 @@ is_integer TypeFail=f S | select_val S=s Fail=f Size=u Rest=* | \
 select_val S=s Fail=f Size=u Rest=* | mixed_types(Size, Rest) => \
   gen_split_values(S, Fail, Fail, Size, Rest)
 
-is_integer Fail=f S | select_val S=s Fail=f Size=u Rest=* | \
+is_integer Fail=f S | select_val S=d Fail=f Size=u Rest=* | \
   fixed_size_values(Size, Rest) => gen_select_val(S, Fail, Size, Rest)
 
-is_atom Fail=f S | select_val S=s Fail=f Size=u Rest=* | \
+is_atom Fail=f S | select_val S=d Fail=f Size=u Rest=* | \
   fixed_size_values(Size, Rest) => gen_select_val(S, Fail, Size, Rest)
 
 select_val S=s Fail=f Size=u Rest=* | floats_or_bignums(Size, Rest) => \
   gen_select_literals(S, Fail, Size, Rest)
 
-select_val S=s Fail=f Size=u Rest=* | fixed_size_values(Size, Rest) => \
+select_val S=d Fail=f Size=u Rest=* | fixed_size_values(Size, Rest) => \
   gen_select_val(S, Fail, Size, Rest)
 
-is_tuple Fail=f S | select_tuple_arity S=s Fail=f Size=u Rest=* => \
+is_tuple Fail=f S | select_tuple_arity S=d Fail=f Size=u Rest=* => \
   gen_select_tuple_arity(S, Fail, Size, Rest)
 
-select_tuple_arity S=s Fail=f Size=u Rest=* => \
+select_tuple_arity S=d Fail=f Size=u Rest=* => \
   gen_select_tuple_arity(S, Fail, Size, Rest)
 
-i_select_val s f I
-i_select_tuple_arity s f I
+i_select_val r f I
+i_select_val x f I
+i_select_val y f I
+i_select_tuple_arity r f I
+i_select_tuple_arity x f I
+i_select_tuple_arity y f I
 
 i_jump_on_val_zero s f I
 i_jump_on_val s f I I