erts: Add move window instruction

Move an entire region of x registers to the stack.
This reduces the dispatch pressure of move instructions.

Also introduce a move2 specialization for some common move patterns:

move r y | move x y -> move2 : As above, moving regions to the stack
move x r | move x y -> move2 : A seemingly common pattern

2 files changed, 71 insertions, 0 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 8b9e271068..b9b2094e6c 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -1503,6 +1503,40 @@ void process_main(void)
 	Next(2);
     }
 
+ OpCase(move_window3_xxxy): {
+     BeamInstr *next;
+     Eterm *y;
+     PreFetch(4, next);
+     y = (Eterm *)(((unsigned char *)E) + (Arg(3)));
+     y[0] = xb(Arg(0));
+     y[1] = xb(Arg(1));
+     y[2] = xb(Arg(2));
+     NextPF(4, next);
+ }
+ OpCase(move_window4_xxxxy): {
+     BeamInstr *next;
+     Eterm *y;
+     PreFetch(5, next);
+     y = (Eterm *)(((unsigned char *)E) + (Arg(4)));
+     y[0] = xb(Arg(0));
+     y[1] = xb(Arg(1));
+     y[2] = xb(Arg(2));
+     y[3] = xb(Arg(3));
+     NextPF(5, next);
+ }
+ OpCase(move_window5_xxxxxy): {
+     BeamInstr *next;
+     Eterm *y;
+     PreFetch(6, next);
+     y = (Eterm *)(((unsigned char *)E) + (Arg(5)));
+     y[0] = xb(Arg(0));
+     y[1] = xb(Arg(1));
+     y[2] = xb(Arg(2));
+     y[3] = xb(Arg(3));
+     y[4] = xb(Arg(4));
+     NextPF(6, next);
+ }
+
  OpCase(i_move_call_only_fcr): {
      r(0) = Arg(1);
  }
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 66339f8ace..3f7e69e07e 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -298,12 +298,44 @@ move_jump f c
 move_jump f x
 move_jump f y
 
+
+# Movement to and from the stack is common
+# Try to pack as much as we can into one instruction
+
+# Window move
+move_window/5
+move_window/6
+
+# x -> y
+
+move S1=r S2=y | move X1=x Y1=y => move2 S1 S2 X1 Y1
+
+move X1=x Y1=y | move X2=x Y2=y | move X3=x Y3=y | succ(Y1,Y2) | succ(Y2,Y3) => \
+    move_window X1 X2 X3 Y1 Y3
+
+move_window X1=x X2=x X3=x Y1=y Y3=y | move X4=x Y4=y | succ(Y3,Y4) => \
+    move_window X1 X2 X3 X4 Y1 Y4
+
+move_window X1=x X2=x X3=x X4=x Y1=y Y4=y | move X5=x Y5=y | succ(Y4,Y5) => \
+    move_window5 X1 X2 X3 X4 X5 Y1
+
+move_window X1=x X2=x X3=x Y1=y Y3=y => move_window3 X1 X2 X3 Y1
+move_window X1=x X2=x X3=x X4=x Y1=y Y4=y => move_window4 X1 X2 X3 X4 Y1
+
+move_window3 x x x y
+move_window4 x x x x y
+move_window5 x x x x x y
+
 move X1=x Y1=y | move X2=x Y2=y => move2 X1 Y1 X2 Y2
 move Y1=y X1=x | move Y2=y X2=x => move2 Y1 X1 Y2 X2
 move X1=x X2=x | move X3=x X4=x => move2 X1 X2 X3 X4
 
+move S1=x S2=r | move X1=x Y1=y => move2 S1 S2 X1 Y1
+move S1=y S2=r | move X1=x Y1=y => move2 S1 S2 X1 Y1
+
 move2 X1=x Y1=y X2=x Y2=y | move X3=x Y3=y => move3 X1 Y1 X2 Y2 X3 Y3
 move2 Y1=y X1=x Y2=y X2=x | move Y3=y X3=x => move3 Y1 X1 Y2 X2 Y3 X3
+move2 X1=x X2=x X3=x X4=x | move X5=x X6=x => move3 X1 X2 X3 X4 X5 X6
 
 move C=aiq X=x==1 => move_x1 C
 move C=aiq X=x==2 => move_x2 C
@@ -316,6 +348,11 @@ move2 x y x y
 move2 y x y x
 move2 x x x x
 
+move2 x r x y
+move2 r y x y
+move2 y r x y
+
+
 %macro: move3 Move3
 move3 x y x y x y
 move3 y x y x y x