Teach beam_makeops to pack operands for move3 and move_window

It is currently only possible to pack up to 4 operands. However,
the move_window4 instrucion has 5 operands and move_window5 and
move3 instrucations have 6 operands.

Teach beam_makeops to pack instructions with 5 or 6 operands.
Also rewrite the move_window instructions in beam_emu.c to macros
to allow their operands to get packed.

3 files changed, 68 insertions, 56 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 60a399a5ac..a3b38adc4b 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -576,6 +576,48 @@ void** beam_ops;
 
 #define Move3(S1, D1, S2, D2, S3, D3) D1 = (S1); D2 = (S2); D3 = (S3)
 
+#define MoveWindow3(S1, S2, S3, D)		\
+  do {						\
+      Eterm xt0, xt1, xt2;			\
+      Eterm *y = &D;				\
+      xt0  = S1;				\
+      xt1  = S2;				\
+      xt2  = S3;				\
+      y[0] = xt0;				\
+      y[1] = xt1;				\
+      y[2] = xt2;				\
+ } while (0)
+
+#define MoveWindow4(S1, S2, S3, S4, D)		\
+  do {						\
+      Eterm xt0, xt1, xt2, xt3;			\
+      Eterm *y = &D;				\
+      xt0  = S1;				\
+      xt1  = S2;				\
+      xt2  = S3;				\
+      xt3  = S4;				\
+      y[0] = xt0;				\
+      y[1] = xt1;				\
+      y[2] = xt2;				\
+      y[3] = xt3;				\
+ } while (0)
+
+#define MoveWindow5(S1, S2, S3, S4, S5, D)	\
+  do {						\
+      Eterm xt0, xt1, xt2, xt3, xt4;		\
+      Eterm *y = &D;				\
+      xt0  = S1;				\
+      xt1  = S2;				\
+      xt2  = S3;				\
+      xt3  = S4;				\
+      xt4  = S5;				\
+      y[0] = xt0;				\
+      y[1] = xt1;				\
+      y[2] = xt2;				\
+      y[3] = xt3;				\
+      y[4] = xt4;				\
+ } while (0)
+
 #define MoveReturn(Src)				\
     x(0) = (Src);				\
     I = c_p->cp;				\
@@ -1466,52 +1508,6 @@ void process_main(void)
 	Next(3);
     }
 
- OpCase(move_window3_xxxy): {
-     BeamInstr *next;
-     Eterm xt0, xt1, xt2;
-     Eterm *y = (Eterm *)(((unsigned char *)E) + (Arg(3)));
-     PreFetch(4, next);
-     xt0  = xb(Arg(0));
-     xt1  = xb(Arg(1));
-     xt2  = xb(Arg(2));
-     y[0] = xt0;
-     y[1] = xt1;
-     y[2] = xt2;
-     NextPF(4, next);
- }
- OpCase(move_window4_xxxxy): {
-     BeamInstr *next;
-     Eterm xt0, xt1, xt2, xt3;
-     Eterm *y = (Eterm *)(((unsigned char *)E) + (Arg(4)));
-     PreFetch(5, next);
-     xt0  = xb(Arg(0));
-     xt1  = xb(Arg(1));
-     xt2  = xb(Arg(2));
-     xt3  = xb(Arg(3));
-     y[0] = xt0;
-     y[1] = xt1;
-     y[2] = xt2;
-     y[3] = xt3;
-     NextPF(5, next);
- }
- OpCase(move_window5_xxxxxy): {
-     BeamInstr *next;
-     Eterm xt0, xt1, xt2, xt3, xt4;
-     Eterm *y = (Eterm *)(((unsigned char *)E) + (Arg(5)));
-     PreFetch(6, next);
-     xt0  = xb(Arg(0));
-     xt1  = xb(Arg(1));
-     xt2  = xb(Arg(2));
-     xt3  = xb(Arg(3));
-     xt4  = xb(Arg(4));
-     y[0] = xt0;
-     y[1] = xt1;
-     y[2] = xt2;
-     y[3] = xt3;
-     y[4] = xt4;
-     NextPF(6, next);
- }
-
  OpCase(i_move_call_only_fc): {
      r(0) = Arg(1);
  }
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 18202a4200..bce6bacd5c 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -292,6 +292,10 @@ move_window X1=x X2=x X3=x X4=x Y1=y Y4=y | move X5=x Y5=y | succ(Y4,Y5) => \
 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
 
+%macro: move_window3 MoveWindow3 -pack
+%macro: move_window4 MoveWindow4 -pack
+%macro: move_window5 MoveWindow5 -pack
+
 move_window3 x x x y
 move_window4 x x x x y
 move_window5 x x x x x y
@@ -390,7 +394,7 @@ move2_par y x x y
 move2_par x x y x
 move2_par y x x x
 
-%macro: move3 Move3
+%macro: move3 Move3 -pack
 move3 x y x y x y
 move3 y x y x y x
 move3 x x x x x x
diff --git a/erts/emulator/utils/beam_makeops b/erts/emulator/utils/beam_makeops
index 974b29e9d8..44977ecffc 100755
--- a/erts/emulator/utils/beam_makeops
+++ b/erts/emulator/utils/beam_makeops
@@ -54,6 +54,11 @@ $pack_mask[4]  = ['BEAM_LOOSE_MASK', # Only for 64 bit wordsize
 		  'BEAM_LOOSE_MASK',
 		  $WHOLE_WORD];
 
+# Mapping from packagable arguments to number of packed arguments per
+# word.  Initialized after the wordsize is known.
+
+my @args_per_word;
+
 # There are two types of instructions: generic and specific.
 # The generic instructions are those generated by the Beam compiler.
 # Corresponding to each generic instruction, there is generally a
@@ -237,6 +242,20 @@ while (@ARGV && $ARGV[0] =~ /^-(.*)/) {
 }
 
 #
+# Initialize number of arguments per packed word.
+#
+
+$args_per_word[2] = 2;
+$args_per_word[3] = 3;
+$args_per_word[4] = 2;
+$args_per_word[5] = 3;
+$args_per_word[6] = 3;
+
+if ($wordsize == 64) {
+    $args_per_word[4] = 4;
+}
+
+#
 # Parse the input files.
 #
 
@@ -736,7 +755,7 @@ sub init_item {
 	    print "${sep}NULL";
 	} elsif (/^\{/) {
 	    print "$sep$_";
-	} elsif (/^-?\d/) {
+	} elsif (/^-?\d+$/) {
 	    print "$sep$_";
 	} else {
 	    print "$sep\"$_\"";
@@ -1105,7 +1124,6 @@ sub do_pack {
     # Get out of here if too few or too many arguments.
     #
     return ('', '', @args) if $packable_args < 2;
-    &error("too many packable arguments") if $packable_args > 4;
 
     my($size) = 0;
     my($pack_prefix) = '';
@@ -1113,14 +1131,8 @@ sub do_pack {
 				# beginning).
     my($up) = '';		# Pack commands (storing back while
 				# moving forward).
-    my $args_per_word;
-    if ($packable_args < 4 or $wordsize == 64) {
-	$args_per_word = $packable_args;
-    } else {
-	# 4 packable argument, 32 bit wordsize. Need 2 words.
-	$args_per_word = 2;
-    }
 
+    my $args_per_word = $args_per_word[$packable_args];
     my @shift;
     my @mask;
     my @instr;