From 54189af4a519b72883dde3263135baf6aba0f81f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 1 Jul 2015 12:18:54 +0200
Subject: beam_emu.c: Remove unused MoveGenDest macro

---
 erts/emulator/beam/beam_emu.c | 3 ---
 1 file changed, 3 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 806b31654c..0f9bfa1871 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -566,9 +566,6 @@ void** beam_ops;
 #define Move2(S1, D1, S2, D2) D1 = (S1); D2 = (S2)
 #define Move3(S1, D1, S2, D2, S3, D3) D1 = (S1); D2 = (S2); D3 = (S3)
 
-#define MoveGenDest(src, dstp) \
-   if ((dstp) == NULL) { r(0) = (src); } else { *(dstp) = src; }
-
 #define MoveReturn(Src, Dest)       \
     (Dest) = (Src);                 \
     I = c_p->cp;                    \
-- 
cgit v1.2.3


From 45f469ca089096c8a59ca53d948d05d1a998386e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Fri, 30 Mar 2012 11:05:16 +0200
Subject: Change the meaning of 'x' in a transformation

The purpose of this series of commits is to improve code generation
for the Clang compiler.

As a first step we want to change the meaning of 'x' in a
transformation such as:

  operation Literal=q => move Literal x | operation x

Currently, a plain 'x' means reg[0] or x(0), which is the first
element in the X register array. That element is distinct from
r(0) which is a variable in process_main(). Therefore, since r(0)
and x(0) are currently distinct it is fine to use x(0) as a
scratch register.

However, in the next commit we will eliminate the separate variable
for storing the contents of X register zero (thus, x(0) and r(0)
will point to the same location in the X register array). Therefore,
we must use another scratch register in transformation. Redefine
a plain 'x' in a transformation to mean x(1023). Also define
SCRATCH_X_REG so that we can refer to the register by name from
C code.
---
 erts/emulator/beam/beam_load.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 10baab506a..f5f1147261 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -4322,7 +4322,7 @@ gen_has_map_fields(LoaderState* stp, GenOpArg Fail, GenOpArg Src,
     for (i = 0; i < n; i++) {
 	op->a[3+2*i] = Rest[i];
 	op->a[3+2*i+1].type = TAG_x;
-	op->a[3+2*i+1].val = 0;	/* x(0); normally not used */
+	op->a[3+2*i+1].val = SCRATCH_X_REG; /* Ignore result */
     }
     return op;
 }
-- 
cgit v1.2.3


From 1f996cc46a8c935b8f4e33a4a96804f9fb852da3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Fri, 30 Mar 2012 11:38:37 +0200
Subject: Store r(0) and x(0) in the same location

As part of improving code generation for clang, we want to
eliminate the special variable that stores the content of X
register zero most of the time. In a future, that will allow us
to eliminate the special case of handling r(0) for most
instructions, thus reducing the code size and allow other
simplifcations.

Therefore, in this commit, eliminate the variable that is used
to store r(0) and make r(0) as synonym for x(0). I have chosen
to keep the r(0) define to keep the size of the diff managable.
---
 erts/emulator/beam/beam_emu.c | 100 +++---------------------------------------
 1 file changed, 7 insertions(+), 93 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 0f9bfa1871..9a34d5f11a 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -99,10 +99,7 @@ do {									\
 do {                                     \
   int i_;                                \
   int Arity_ = PC[-1];                   \
-  if (Arity_ > 0) {                      \
-	CHECK_TERM(r(0));                \
-  }                                      \
-  for (i_ = 1; i_ < Arity_; i_++) {      \
+  for (i_ = 0; i_ < Arity_; i_++) {      \
 	CHECK_TERM(x(i_));               \
   }                                      \
 } while (0)
@@ -293,7 +290,7 @@ void** beam_ops;
 #define Ib(N) (N)
 #define x(N) reg[N]
 #define y(N) E[N]
-#define r(N) x##N
+#define r(N) x(N)
 
 /*
  * Makes sure that there are StackNeed + HeapNeed + 1 words available
@@ -309,12 +306,10 @@ void** beam_ops;
      needed = (StackNeed) + 1; \
      if (E - HTOP < (needed + (HeapNeed))) { \
            SWAPOUT; \
-           reg[0] = r(0); \
            PROCESS_MAIN_CHK_LOCKS(c_p); \
            FCALLS -= erts_garbage_collect(c_p, needed + (HeapNeed), reg, (M)); \
            ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); \
            PROCESS_MAIN_CHK_LOCKS(c_p); \
-           r(0) = reg[0]; \
            SWAPIN; \
      } \
      E -= needed; \
@@ -363,12 +358,10 @@ void** beam_ops;
     unsigned need = (Nh);                                       		\
     if ((E - HTOP < need) || (MSO(c_p).overhead + (VNh) >= BIN_VHEAP_SZ(c_p))) {\
        SWAPOUT;                                                 		\
-       reg[0] = r(0);                                           		\
        PROCESS_MAIN_CHK_LOCKS(c_p);                             		\
        FCALLS -= erts_garbage_collect(c_p, need, reg, (Live));  		\
        ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);					\
        PROCESS_MAIN_CHK_LOCKS(c_p);                             		\
-       r(0) = reg[0];                                           		\
        SWAPIN;                                                  		\
     }                                                           		\
     HEAP_SPACE_VERIFIED(need);                                                  \
@@ -386,12 +379,10 @@ void** beam_ops;
     unsigned need = (Nh);                                       \
     if (E - HTOP < need) {                                      \
        SWAPOUT;                                                 \
-       reg[0] = r(0);                                           \
        PROCESS_MAIN_CHK_LOCKS(c_p);                             \
        FCALLS -= erts_garbage_collect(c_p, need, reg, (Live));  \
        ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);			\
        PROCESS_MAIN_CHK_LOCKS(c_p);                             \
-       r(0) = reg[0];                                           \
        SWAPIN;                                                  \
     }                                                           \
     HEAP_SPACE_VERIFIED(need);                             \
@@ -408,15 +399,11 @@ void** beam_ops;
     unsigned need = (Nh);						\
     if (E - HTOP < need) {						\
        SWAPOUT;								\
-       reg[0] = r(0);							\
        reg[Live] = Extra;						\
        PROCESS_MAIN_CHK_LOCKS(c_p);					\
        FCALLS -= erts_garbage_collect(c_p, need, reg, (Live)+1);	\
        ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);				\
        PROCESS_MAIN_CHK_LOCKS(c_p);					\
-       if (Live > 0) {							\
-	   r(0) = reg[0];						\
-       }								\
        Extra = reg[Live];						\
        SWAPIN;								\
     }									\
@@ -439,7 +426,6 @@ void** beam_ops;
 #define MakeFun(FunP, NumFree)					\
   do {								\
      SWAPOUT;							\
-     reg[0] = r(0);						\
      r(0) = new_fun(c_p, reg, (ErlFunEntry *) FunP, NumFree);	\
      SWAPIN;							\
   } while (0)
@@ -992,7 +978,6 @@ init_emulator(void)
  */
 
 #if defined(__GNUC__) && defined(sparc) && !defined(DEBUG)
-#  define REG_x0 asm("%l0")
 #  define REG_xregs asm("%l1")
 #  define REG_htop asm("%l2")
 #  define REG_stop asm("%l3")
@@ -1001,7 +986,6 @@ init_emulator(void)
 #  define REG_tmp_arg1 asm("%l6")
 #  define REG_tmp_arg2 asm("%l7")
 #else
-#  define REG_x0
 #  define REG_xregs
 #  define REG_htop
 #  define REG_stop
@@ -1126,11 +1110,6 @@ void process_main(void)
     ERTS_DECLARE_DUMMY(Eterm pid);
 #endif
 
-    /*
-     * X register zero; also called r(0)
-     */
-    register Eterm x0 REG_x0 = NIL;
-
     /* Pointer to X registers: x(1)..x(N); reg[0] is used when doing GC,
      * in all other cases x0 is used.
      */
@@ -1262,7 +1241,7 @@ void process_main(void)
 	int i;
 
 	argp = c_p->arg_reg;
-	for (i = c_p->arity - 1; i > 0; i--) {
+	for (i = c_p->arity - 1; i >= 0; i--) {
 	    reg[i] = argp[i];
 	    CHECK_TERM(reg[i]);
 	}
@@ -1286,12 +1265,6 @@ void process_main(void)
 	}
 
 	next = (BeamInstr *) *I;
-	r(0) = c_p->arg_reg[0];
-#ifdef HARDDEBUG
-	if (c_p->arity > 0) {
-	    CHECK_TERM(r(0));
-	}
-#endif
 	SWAPIN;
 	ASSERT(VALID_INSTR(next));
 
@@ -1365,11 +1338,9 @@ void process_main(void)
 
 	    live = Arg(2);
 	    SWAPOUT;
-	    reg[0] = r(0);
 	    reg[live] = increment_reg_val;
 	    reg[live+1] = make_small(increment_val);
 	    result = erts_gc_mixed_plus(c_p, reg, live);
-	    r(0) = reg[0];
 	    SWAPIN;
 	    ERTS_HOLE_CHECK(c_p);
 	    if (is_value(result)) {
@@ -1383,11 +1354,9 @@ void process_main(void)
     do {                                 \
         Uint live = Arg(1);              \
         SWAPOUT;                         \
-        reg[0] = r(0);                   \
         reg[live] = (Arg1);              \
         reg[live+1] = (Arg2);            \
         result = (Func)(c_p, reg, live); \
-        r(0) = reg[0];                   \
         SWAPIN;                          \
         ERTS_HOLE_CHECK(c_p);            \
         if (is_value(result)) {          \
@@ -1706,7 +1675,6 @@ void process_main(void)
 
      PRE_BIF_SWAPOUT(c_p);
      c_p->fcalls = FCALLS - 1;
-     reg[0] = r(0);
      result = erl_send(c_p, r(0), x(1));
      PreFetch(0, next);
      ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
@@ -1714,7 +1682,6 @@ void process_main(void)
      PROCESS_MAIN_CHK_LOCKS(c_p);
      if (c_p->mbuf || MSO(c_p).overhead >= BIN_VHEAP_SZ(c_p)) {
 	 result = erts_gc_after_bif_call(c_p, result, reg, 2);
-	 r(0) = reg[0];
 	 E = c_p->stop;
      }
      HTOP = HEAP_TOP(c_p);
@@ -1727,7 +1694,6 @@ void process_main(void)
 	 SET_CP(c_p, I+1);
 	 SET_I(c_p->i);
 	 SWAPIN;
-	 r(0) = reg[0];
 	 Dispatch();
      }
      goto find_func_info;
@@ -1935,13 +1901,11 @@ void process_main(void)
      ErtsMoveMsgAttachmentIntoProc(msgp, c_p, E, HTOP, FCALLS,
 				   {
 				       SWAPOUT;
-				       reg[0] = r(0);
 				       PROCESS_MAIN_CHK_LOCKS(c_p);
 				   },
 				   {
 				       ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 				       PROCESS_MAIN_CHK_LOCKS(c_p);
-				       r(0) = reg[0];
 				       SWAPIN;
 				   });
      if (is_non_value(ERL_MESSAGE_TERM(msgp))) {
@@ -2448,11 +2412,9 @@ void process_main(void)
  OpCase(new_map_dII): {
      Eterm res;
 
-     x(0) = r(0);
      SWAPOUT;
      res = new_map(c_p, reg, I-1);
      SWAPIN;
-     r(0) = x(0);
      StoreResult(res, Arg(0));
      Next(3+Arg(2));
  }
@@ -2543,12 +2505,10 @@ do {								\
      Eterm map;
 
      GetArg1(1, map);
-     x(0) = r(0);
      SWAPOUT;
      res = update_map_assoc(c_p, reg, map, I);
      SWAPIN;
      if (is_value(res)) {
-	 r(0) = x(0);
 	 StoreResult(res, Arg(2));
 	 Next(5+Arg(4));
      } else {
@@ -2567,12 +2527,10 @@ do {								\
      Eterm map;
 
      GetArg1(1, map);
-     x(0) = r(0);
      SWAPOUT;
      res = update_map_exact(c_p, reg, map, I);
      SWAPIN;
      if (is_value(res)) {
-	 r(0) = x(0);
 	 StoreResult(res, Arg(2));
 	 Next(5+Arg(4));
      } else {
@@ -2659,7 +2617,6 @@ do {								\
 	Uint live = (Uint) Arg(3);
 
 	GetArg1(2, arg);
-	reg[0] = r(0);
 	reg[live] = arg;
 	bf = (GcBifFunction) Arg(1);
 	c_p->fcalls = FCALLS;
@@ -2671,7 +2628,6 @@ do {								\
 	ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	SWAPIN;
-	r(0) = reg[0];
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
 	if (is_value(result)) {
@@ -2694,7 +2650,6 @@ do {								\
 	Eterm result;
 	Uint live = (Uint) Arg(2);
 
-	reg[0] = r(0);
 	reg[live++] = tmp_arg1;
 	reg[live] = tmp_arg2;
 	bf = (GcBifFunction) Arg(1);
@@ -2707,7 +2662,6 @@ do {								\
 	ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	SWAPIN;
-	r(0) = reg[0];
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
 	if (is_value(result)) {
@@ -2732,7 +2686,6 @@ do {								\
 	Uint live = (Uint) Arg(3);
 
 	GetArg1(2, arg);
-	reg[0] = r(0);
 	reg[live++] = arg;
 	reg[live++] = tmp_arg1;
 	reg[live] = tmp_arg2;
@@ -2746,7 +2699,6 @@ do {								\
 	ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	SWAPIN;
-	r(0) = reg[0];
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
 	if (is_value(result)) {
@@ -2835,7 +2787,6 @@ do {								\
 	PreFetch(1, next);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
-	reg[0] = r(0);
 	result = (*bf)(c_p, reg, I);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result));
 	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
@@ -2857,7 +2808,6 @@ do {								\
 	    SET_CP(c_p, I+2);
 	    SET_I(c_p->i);
 	    SWAPIN;
-	    r(0) = reg[0];
 	    Dispatch();
 	}
 
@@ -2967,11 +2917,9 @@ do {								\
      Uint live = Arg(1);
 
      SWAPOUT;
-     reg[0] = r(0);
      reg[live] = tmp_arg1;
      reg[live+1] = tmp_arg2;
      result = arith_func(c_p, reg, live);
-     r(0) = reg[0];
      SWAPIN;
      ERTS_HOLE_CHECK(c_p);
      if (is_value(result)) {
@@ -3166,10 +3114,8 @@ do {								\
      } else {
 	 Uint live = Arg(2);
 	 SWAPOUT;
-	 reg[0] = r(0);
 	 reg[live] = bnot_val;
 	 bnot_val = erts_gc_bnot(c_p, reg, live);
-	 r(0) = reg[0];
 	 SWAPIN;
 	 ERTS_HOLE_CHECK(c_p);
 	 if (is_nil(bnot_val)) {
@@ -3189,7 +3135,6 @@ do {								\
      next = apply(c_p, r(0), x(1), x(2), reg);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, I+1);
 	 SET_I(next);
 	 Dispatch();
@@ -3204,7 +3149,6 @@ do {								\
      next = apply(c_p, r(0), x(1), x(2), reg);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, (BeamInstr *) E[0]);
 	 E = ADD_BYTE_OFFSET(E, Arg(0));
 	 SET_I(next);
@@ -3220,7 +3164,6 @@ do {								\
      next = apply(c_p, r(0), x(1), x(2), reg);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_I(next);
 	 Dispatch();
      }
@@ -3231,12 +3174,10 @@ do {								\
  OpCase(apply_I): {
      BeamInstr *next;
 
-     reg[0] = r(0);
      SWAPOUT;
      next = fixed_apply(c_p, reg, Arg(0));
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, I+2);
 	 SET_I(next);
 	 Dispatch();
@@ -3248,12 +3189,10 @@ do {								\
  OpCase(apply_last_IP): {
      BeamInstr *next;
 
-     reg[0] = r(0);
      SWAPOUT;
      next = fixed_apply(c_p, reg, Arg(0));
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, (BeamInstr *) E[0]);
 	 E = ADD_BYTE_OFFSET(E, Arg(1));
 	 SET_I(next);
@@ -3270,7 +3209,6 @@ do {								\
      next = apply_fun(c_p, r(0), x(1), reg);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, I+1);
 	 SET_I(next);
 	 Dispatchfun();
@@ -3285,7 +3223,6 @@ do {								\
      next = apply_fun(c_p, r(0), x(1), reg);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, (BeamInstr *) E[0]);
 	 E = ADD_BYTE_OFFSET(E, Arg(0));
 	 SET_I(next);
@@ -3301,7 +3238,6 @@ do {								\
      next = apply_fun(c_p, r(0), x(1), reg);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_I(next);
 	 Dispatchfun();
      }
@@ -3312,12 +3248,9 @@ do {								\
      BeamInstr *next;
 
      SWAPOUT;
-     reg[0] = r(0);
-
      next = call_fun(c_p, Arg(0), reg, THE_NON_VALUE);
      SWAPIN;
      if (next != NULL) {
-	 r(0) = reg[0];
 	 SET_CP(c_p, I+2);
 	 SET_I(next);
 	 Dispatchfun();
@@ -3329,11 +3262,9 @@ do {								\
      BeamInstr *next;
 
      SWAPOUT;
-     reg[0] = r(0);
      next = call_fun(c_p, Arg(0), reg, THE_NON_VALUE);
      SWAPIN;
      if (next != NULL) {
-	r(0) = reg[0];
 	SET_CP(c_p, (BeamInstr *) E[0]);
 	E = ADD_BYTE_OFFSET(E, Arg(1));
 	SET_I(next);
@@ -3418,10 +3349,9 @@ do {								\
       */
 
      argp = c_p->arg_reg;
-     for (i = c_p->arity - 1; i > 0; i--) {
+     for (i = c_p->arity - 1; i >= 0; i--) {
 	 argp[i] = reg[i];
      }
-     c_p->arg_reg[0] = r(0);
      SWAPOUT;
      c_p->i = I;
      goto do_schedule1;
@@ -3531,7 +3461,6 @@ do {								\
  /* Fall through here */
 
  find_func_info: {
-     reg[0] = r(0);
      SWAPOUT;
      I = handle_error(c_p, I, reg, NULL);
      goto post_error_handling;
@@ -3549,9 +3478,7 @@ do {								\
      * code[4]: Not used
      */
     SWAPOUT;
-    reg[0] = r(0);
     I = call_error_handler(c_p, I-3, reg, am_undefined_function);
-    r(0) = reg[0];
     SWAPIN;
     if (I) {
 	Goto(*I);
@@ -3560,14 +3487,12 @@ do {								\
  /* Fall through */
  OpCase(error_action_code): {
     handle_error:
-     reg[0] = r(0);
      SWAPOUT;
      I = handle_error(c_p, NULL, reg, NULL);
  post_error_handling:
      if (I == 0) {
 	 goto do_schedule;
      } else {
-	 r(0) = reg[0];
 	 ASSERT(!is_value(r(0)));
 	 if (c_p->mbuf) {
 	     erts_garbage_collect(c_p, 0, reg+1, 3);
@@ -3610,7 +3535,6 @@ do {								\
 		NifF* fp = vbf = (NifF*) I[1];
 		struct enif_environment_t env;
 		erts_pre_nif(&env, c_p, (struct erl_module_nif*)I[2]);
-		reg[0] = r(0);
 		nif_bif_result = (*fp)(&env, bif_nif_arity, reg);
 		if (env.exception_thrown)
 		    nif_bif_result = THE_NON_VALUE;
@@ -3649,7 +3573,7 @@ do {								\
 	    bif_nif_arity = I[-1];
             ASSERT(bif_nif_arity <= 4);
 	    ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
-	    reg[0] = r(0);
+	    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	    {
 		Eterm (*bf)(Process*, Eterm*, BeamInstr*) = vbf;
 		ASSERT(!ERTS_PROC_IS_EXITING(c_p));
@@ -3679,7 +3603,6 @@ do {								\
 		Goto(*I);
 	    } else if (c_p->freason == TRAP) {
 		SET_I(c_p->i);
-		r(0) = reg[0];
 		if (c_p->flags & F_HIBERNATE_SCHED) {
 		    c_p->flags &= ~F_HIBERNATE_SCHED;
 		    goto do_schedule;
@@ -4126,10 +4049,8 @@ do {								\
      Uint res;
 
      SWAPOUT;
-     reg[0] = r(0);
      reg[live] = tmp_arg2;
      res = erts_bs_append(c_p, reg, live, tmp_arg1, Arg(1), Arg(3));
-     r(0) = reg[0];
      SWAPIN;
      if (is_non_value(res)) {
 	 /* c_p->freason is already set (may be either BADARG or SYSTEM_LIMIT). */
@@ -4706,7 +4627,7 @@ do {								\
      Uint hole_size;
 
      OpCase(bs_context_to_binary_r): {
-	 context_to_binary_context = x0;
+	 context_to_binary_context = r(0);
 	 I -= 2;
 	 goto do_context_to_binary;
      }
@@ -4737,7 +4658,7 @@ do {								\
      Next(2);
 
      OpCase(i_bs_get_binary_all_reuse_rfI): {
-	 context_to_binary_context = x0;
+	 context_to_binary_context = r(0);
 	 goto do_bs_get_binary_all_reuse;
      }
 
@@ -4883,9 +4804,7 @@ do {								\
      BeamInstr real_I;
      ASSERT(I[-5] == (BeamInstr) BeamOp(op_i_func_info_IaaI));
      SWAPOUT;
-     reg[0] = r(0);
      real_I = erts_generic_breakpoint(c_p, I, reg);
-     r(0) = reg[0];
      SWAPIN;
      ASSERT(VALID_INSTR(real_I));
      Goto(real_I);
@@ -5103,7 +5022,6 @@ do {								\
      SWAPOUT;
      c_p->fcalls = FCALLS;
      c_p->def_arg_reg[4] = -neg_o_reds;
-     reg[0] = r(0);
      c_p = hipe_mode_switch(c_p, cmd, reg);
      reg = ERTS_PROC_GET_SCHDATA(c_p)->x_reg_array;
      freg = ERTS_PROC_GET_SCHDATA(c_p)->f_reg_array;
@@ -5120,7 +5038,6 @@ do {								\
 	 /*fall through*/
        case HIPE_MODE_SWITCH_RES_CALL_BEAM:
 	 SET_I(c_p->i);
-	 r(0) = reg[0];
 	 Dispatch();
        case HIPE_MODE_SWITCH_RES_CALL_CLOSURE:
 	 /* This can be used to call any function value, but currently it's
@@ -5131,7 +5048,6 @@ do {								\
 	     next = call_fun(c_p, c_p->arity - 1, reg, THE_NON_VALUE);
 	     SWAPIN;
 	     if (next != NULL) {
-		 r(0) = reg[0];
 		 SET_I(next);
 		 Dispatchfun();
 	     }
@@ -5191,9 +5107,7 @@ do {								\
 
  OpCase(i_debug_breakpoint): {
      SWAPOUT;
-     reg[0] = r(0);
      I = call_error_handler(c_p, I-3, reg, am_breakpoint);
-     r(0) = reg[0];
      SWAPIN;
      if (I) {
 	 Goto(*I);
-- 
cgit v1.2.3


From e1019cbba7a66788a068b6df1a1caf2d643ef65b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Fri, 30 Mar 2012 11:54:04 +0200
Subject: Eliminate R_REG_DEF

---
 erts/emulator/beam/beam_debug.c | 22 ++--------------------
 erts/emulator/beam/beam_emu.c   | 14 --------------
 erts/emulator/beam/beam_load.c  |  8 ++------
 erts/emulator/beam/erl_term.h   |  2 --
 4 files changed, 4 insertions(+), 42 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 8a35ad17c6..00d0b27178 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -452,21 +452,13 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	case 's':		/* Any source (tagged constant or register) */
 	    tag = beam_reg_tag(*ap);
 	    if (tag == X_REG_DEF) {
-		if (reg_index(*ap) == 0) {
-		    erts_print(to, to_arg, "x[0]");
-		} else {
-		    erts_print(to, to_arg, "x(%d)", reg_index(*ap));
-		}
+		erts_print(to, to_arg, "x(%d)", reg_index(*ap));
 		ap++;
 		break;
 	    } else if (tag == Y_REG_DEF) {
 		erts_print(to, to_arg, "y(%d)", reg_index(*ap) - CP_SIZE);
 		ap++;
 		break;
-	    } else if (tag == R_REG_DEF) {
-		erts_print(to, to_arg, "x(0)");
-		ap++;
-		break;
 	    }
 	    /*FALLTHROUGH*/
 	case 'a':		/* Tagged atom */
@@ -483,18 +475,11 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	case 'd':		/* Destination (x(0), x(N), y(N)) */
 	    switch (beam_reg_tag(*ap)) {
 	    case X_REG_DEF:
-		if (reg_index(*ap) == 0) {
-		    erts_print(to, to_arg, "x[0]");
-		} else {
-		    erts_print(to, to_arg, "x(%d)", reg_index(*ap));
-		}
+		erts_print(to, to_arg, "x(%d)", reg_index(*ap));
 		break;
 	    case Y_REG_DEF:
 		erts_print(to, to_arg, "y(%d)", reg_index(*ap) - CP_SIZE);
 		break;
-	    case R_REG_DEF:
-		erts_print(to, to_arg, "x(0)");
-		break;
 	    }
 	    ap++;
 	    break;
@@ -677,9 +662,6 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 		    erts_print(to, to_arg, " %T", (Eterm) ap[0]);
 		} else {
 		    switch ((ap[0] >> 2) & 0x03) {
-		    case R_REG_DEF:
-			erts_print(to, to_arg, " x(0)");
-			break;
 		    case X_REG_DEF:
 			erts_print(to, to_arg, " x(%d)", ap[0] >> 4);
 			break;
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 9a34d5f11a..059c443450 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -160,8 +160,6 @@ do {                                     \
     stb_reg = (DestDesc);                           \
     CHECK_TERM(Result);                             \
     switch (beam_reg_tag(stb_reg)) {                \
-    case R_REG_DEF:                                 \
-      r(0) = (Result); break;                       \
     case X_REG_DEF:                                 \
       xb(x_reg_offset(stb_reg)) = (Result); break;  \
     default:                                        \
@@ -186,8 +184,6 @@ do {                                     \
     stb_next = (BeamInstr *) *I;                                 \
     CHECK_TERM(Result);                                      \
     switch (beam_reg_tag(stb_reg)) {                         \
-    case R_REG_DEF:                                          \
-      r(0) = (Result); Goto(stb_next);                       \
     case X_REG_DEF:                                          \
       xb(x_reg_offset(stb_reg)) = (Result); Goto(stb_next);  \
     default:                                                 \
@@ -521,7 +517,6 @@ void** beam_ops;
    do { \
      tr = Arg(pos); \
      switch (beam_reg_tag(tr)) { \
-     case R_REG_DEF: tr = r(0); break; \
      case X_REG_DEF: tr = xb(x_reg_offset(tr)); break; \
      case Y_REG_DEF: ASSERT(y_reg_offset(tr) >= 1); tr = yb(y_reg_offset(tr)); break; \
      } \
@@ -2391,9 +2386,6 @@ void process_main(void)
      do {
 	 Eterm term = *I++;
 	 switch (term & _TAG_IMMED1_MASK) {
-	 case (R_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:
-	     *hp++ = r(0);
-	     break;
 	 case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:
 	     *hp++ = x(term >> _TAG_IMMED1_SIZE);
 	     break;
@@ -2422,9 +2414,6 @@ void process_main(void)
 #define PUT_TERM_REG(term, desc)				\
 do {								\
     switch ((desc) & _TAG_IMMED1_MASK) {			\
-    case (R_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
-	r(0) = (term);						\
-	break;							\
     case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
 	x((desc) >> _TAG_IMMED1_SIZE) = (term);			\
 	break;							\
@@ -6502,9 +6491,6 @@ static Eterm get_map_element_hash(Eterm map, Eterm key, Uint32 hx)
 do {								\
     Eterm src = (Eterm)(term);					\
     switch (src & _TAG_IMMED1_MASK) {				\
-    case (R_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
-	dest = x(0);						\
-	break;							\
     case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
 	dest = x(src >> _TAG_IMMED1_SIZE);			\
 	break;							\
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index f5f1147261..495e92d600 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2167,7 +2167,7 @@ load_code(LoaderState* stp)
 	    case 's':	/* Any source (tagged constant or register) */
 		switch (tag) {
 		case TAG_r:
-		    code[ci++] = make_rreg();
+		    code[ci++] = make_xreg(0);
 		    break;
 		case TAG_x:
 		    code[ci++] = make_xreg(tmp_op->a[arg].val);
@@ -2193,7 +2193,7 @@ load_code(LoaderState* stp)
 	    case 'd':	/* Destination (x(0), x(N), y(N) */
 		switch (tag) {
 		case TAG_r:
-		    code[ci++] = make_rreg();
+		    code[ci++] = make_xreg(0);
 		    break;
 		case TAG_x:
 		    code[ci++] = make_xreg(tmp_op->a[arg].val);
@@ -2357,10 +2357,6 @@ load_code(LoaderState* stp)
 		ci++;
 		break;
 	    case TAG_r:
-		CodeNeed(1);
-		code[ci++] = (R_REG_DEF << _TAG_PRIMARY_SIZE) |
-		    TAG_PRIMARY_HEADER;
-		break;
 	    case TAG_x:
 		CodeNeed(1);
 		code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) |
diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h
index b928e4ca1e..cff2430547 100644
--- a/erts/emulator/beam/erl_term.h
+++ b/erts/emulator/beam/erl_term.h
@@ -1037,11 +1037,9 @@ _ET_DECLARE_CHECKED(Uint,catch_val,Eterm)
 
 #define X_REG_DEF	0
 #define Y_REG_DEF	1
-#define R_REG_DEF	2
 
 #define beam_reg_tag(x)	((x) & 3)
 
-#define make_rreg()	R_REG_DEF
 #define make_xreg(ix)	(((ix) * sizeof(Eterm)) | X_REG_DEF)
 #define make_yreg(ix)	(((ix) * sizeof(Eterm)) | Y_REG_DEF)
 
-- 
cgit v1.2.3


From f0923b143ecfdacfe4873fc1e96c8ec69726a4d9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 16 Jun 2015 06:39:25 +0200
Subject: Allow X and Y registers to be overloaded with any literal

Consider the try_case_end instruction:

  try_case_end s

The 's' operand type means that the operand can either be a
literal of one of the types atom, integer, or empty list, or
a register. That worked well before R12. In R12 additional
types of literals where introduced. Because of way the
overloading was done, an 's' operand cannot handle the
new types of literals. Therefore, code such as the following
is necessary in ops.tab to avoid giving an 's' operand a
literal:

  try_case_end Literal=q => move Literal x | try_case_end x

While this work, it is error-prone in that it is easy to
forget to add that kind of rule. It would also be complicated
in case we wanted to introduce a new kind of addition operator
such as:

  i_plus jssd

Since there are two 's' operands, two scratch registers and
two 'move' instructions would be needed.

Therefore, we'll need to find a smarter way to find tag
register operands. We will overload the pid and port tags
for X and Y register, respectively. That works because pids
and port are immediate values (fit in one word), and there
are no literals for pids and ports.
---
 erts/emulator/beam/beam_debug.c |  77 +++++++++++-----------
 erts/emulator/beam/beam_emu.c   | 139 +++++++++++++++++++++-------------------
 erts/emulator/beam/beam_load.c  |  22 ++++---
 erts/emulator/beam/erl_term.c   |   6 +-
 erts/emulator/beam/erl_term.h   |  50 +++++++--------
 erts/emulator/beam/ops.tab      |  29 +--------
 6 files changed, 153 insertions(+), 170 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 00d0b27178..243ccb9afd 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -432,31 +432,33 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
     while (*sign) {
 	switch (*sign) {
 	case 'r':		/* x(0) */
-	    erts_print(to, to_arg, "x(0)");
+	    erts_print(to, to_arg, "r(0)");
 	    break;
 	case 'x':		/* x(N) */
-	    if (reg_index(ap[0]) == 0) {
-		erts_print(to, to_arg, "x[0]");
-	    } else {
-		erts_print(to, to_arg, "x(%d)", reg_index(ap[0]));
+	    {
+		Uint n = ap[0] / sizeof(Eterm);
+		erts_print(to, to_arg, "x(%d)", n);
+		ap++;
 	    }
-	    ap++;
 	    break;
 	case 'y':		/* y(N) */
-	    erts_print(to, to_arg, "y(%d)", reg_index(ap[0]) - CP_SIZE);
-	    ap++;
+	    {
+		Uint n = ap[0] / sizeof(Eterm) - CP_SIZE;
+		erts_print(to, to_arg, "y(%d)", n);
+		ap++;
+	    }
 	    break;
 	case 'n':		/* Nil */
 	    erts_print(to, to_arg, "[]");
 	    break;
 	case 's':		/* Any source (tagged constant or register) */
-	    tag = beam_reg_tag(*ap);
-	    if (tag == X_REG_DEF) {
-		erts_print(to, to_arg, "x(%d)", reg_index(*ap));
+	    tag = loader_tag(*ap);
+	    if (tag == LOADER_X_REG) {
+		erts_print(to, to_arg, "x(%d)", loader_x_reg_index(*ap));
 		ap++;
 		break;
-	    } else if (tag == Y_REG_DEF) {
-		erts_print(to, to_arg, "y(%d)", reg_index(*ap) - CP_SIZE);
+	    } else if (tag == LOADER_Y_REG) {
+		erts_print(to, to_arg, "y(%d)", loader_y_reg_index(*ap) - CP_SIZE);
 		ap++;
 		break;
 	    }
@@ -473,12 +475,14 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    ap++;
 	    break;
 	case 'd':		/* Destination (x(0), x(N), y(N)) */
-	    switch (beam_reg_tag(*ap)) {
-	    case X_REG_DEF:
-		erts_print(to, to_arg, "x(%d)", reg_index(*ap));
+	    switch (loader_tag(*ap)) {
+	    case LOADER_X_REG:
+		erts_print(to, to_arg, "x(%d)",
+			   loader_x_reg_index(*ap));
 		break;
-	    case Y_REG_DEF:
-		erts_print(to, to_arg, "y(%d)", reg_index(*ap) - CP_SIZE);
+	    case LOADER_Y_REG:
+		erts_print(to, to_arg, "y(%d)",
+			   loader_y_reg_index(*ap) - CP_SIZE);
 		break;
 	    }
 	    ap++;
@@ -546,7 +550,7 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    ap++;
 	    break;
 	case 'l':		/* fr(N) */
-	    erts_print(to, to_arg, "fr(%d)", reg_index(ap[0]));
+	    erts_print(to, to_arg, "fr(%d)", loader_reg_index(ap[0]));
 	    ap++;
 	    break;
 	default:
@@ -658,17 +662,16 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    int n = unpacked[-1];
 
 	    while (n > 0) {
-		if (!is_header(ap[0])) {
+		switch (loader_tag(ap[0])) {
+		case LOADER_X_REG:
+		    erts_print(to, to_arg, " x(%d)", loader_x_reg_index(ap[0]));
+		    break;
+		case LOADER_Y_REG:
+		    erts_print(to, to_arg, " x(%d)", loader_y_reg_index(ap[0]));
+		    break;
+		default:
 		    erts_print(to, to_arg, " %T", (Eterm) ap[0]);
-		} else {
-		    switch ((ap[0] >> 2) & 0x03) {
-		    case X_REG_DEF:
-			erts_print(to, to_arg, " x(%d)", ap[0] >> 4);
-			break;
-		    case Y_REG_DEF:
-			erts_print(to, to_arg, " y(%d)", ap[0] >> 4);
-			break;
-		    }
+		    break;
 		}
 		ap++, size++, n--;
 	    }
@@ -681,18 +684,16 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    while (n > 0) {
 		if (n % 3 == 1) {
 		    erts_print(to, to_arg, " %X", ap[0]);
-		} else if (!is_header(ap[0])) {
-		    erts_print(to, to_arg, " %T", (Eterm) ap[0]);
 		} else {
-		    switch ((ap[0] >> 2) & 0x03) {
-		    case R_REG_DEF:
-			erts_print(to, to_arg, " x(0)");
+		    switch (loader_tag(ap[0])) {
+		    case LOADER_X_REG:
+			erts_print(to, to_arg, " x(%d)", loader_x_reg_index(ap[0]));
 			break;
-		    case X_REG_DEF:
-			erts_print(to, to_arg, " x(%d)", ap[0] >> 4);
+		    case LOADER_Y_REG:
+			erts_print(to, to_arg, " y(%d)", loader_y_reg_index(ap[0]));
 			break;
-		    case Y_REG_DEF:
-			erts_print(to, to_arg, " y(%d)", ap[0] >> 4);
+		    default:
+			erts_print(to, to_arg, " %T", (Eterm) ap[0]);
 			break;
 		    }
 		}
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 059c443450..a74da2d872 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -154,17 +154,19 @@ do {                                     \
  * Store a result into a register given a destination descriptor.
  */
 
-#define StoreResult(Result, DestDesc)               \
-  do {                                              \
-    Eterm stb_reg;                                  \
-    stb_reg = (DestDesc);                           \
-    CHECK_TERM(Result);                             \
-    switch (beam_reg_tag(stb_reg)) {                \
-    case X_REG_DEF:                                 \
-      xb(x_reg_offset(stb_reg)) = (Result); break;  \
-    default:                                        \
-      yb(y_reg_offset(stb_reg)) = (Result); break;  \
-    }                                               \
+#define StoreResult(Result, DestDesc)			\
+  do {							\
+    Eterm stb_reg;					\
+    stb_reg = (DestDesc);				\
+    CHECK_TERM(Result);					\
+    switch (loader_tag(stb_reg)) {			\
+    case LOADER_X_REG:					\
+      x(loader_x_reg_index(stb_reg)) = (Result);	\
+      break;						\
+    default:						\
+      y(loader_y_reg_index(stb_reg)) = (Result);	\
+      break;						\
+    }							\
   } while (0)
 
 #define StoreSimpleDest(Src, Dest) Dest = (Src)
@@ -175,20 +177,22 @@ do {                                     \
  * be just before the next instruction.
  */
  
-#define StoreBifResult(Dst, Result)                          \
-  do {                                                       \
-    BeamInstr* stb_next;                                         \
-    Eterm stb_reg;                                           \
-    stb_reg = Arg(Dst);                                      \
-    I += (Dst) + 2;                                          \
-    stb_next = (BeamInstr *) *I;                                 \
-    CHECK_TERM(Result);                                      \
-    switch (beam_reg_tag(stb_reg)) {                         \
-    case X_REG_DEF:                                          \
-      xb(x_reg_offset(stb_reg)) = (Result); Goto(stb_next);  \
-    default:                                                 \
-      yb(y_reg_offset(stb_reg)) = (Result); Goto(stb_next);  \
-    }                                                        \
+#define StoreBifResult(Dst, Result)			\
+  do {							\
+    BeamInstr* stb_next;				\
+    Eterm stb_reg;					\
+    stb_reg = Arg(Dst);					\
+    I += (Dst) + 2;					\
+    stb_next = (BeamInstr *) *I;			\
+    CHECK_TERM(Result);					\
+    switch (loader_tag(stb_reg)) {			\
+    case LOADER_X_REG:					\
+      x(loader_x_reg_index(stb_reg)) = (Result);	\
+      Goto(stb_next);					\
+    default:						\
+      y(loader_y_reg_index(stb_reg)) = (Result);	\
+      Goto(stb_next);					\
+    }							\
   } while (0)
 
 #define ClauseFail() goto jump_f
@@ -513,14 +517,19 @@ void** beam_ops;
     ASSERT(VALID_INSTR(Dst));  \
     Goto(Dst)
 
-#define GetR(pos, tr) \
-   do { \
-     tr = Arg(pos); \
-     switch (beam_reg_tag(tr)) { \
-     case X_REG_DEF: tr = xb(x_reg_offset(tr)); break; \
-     case Y_REG_DEF: ASSERT(y_reg_offset(tr) >= 1); tr = yb(y_reg_offset(tr)); break; \
-     } \
-     CHECK_TERM(tr); \
+#define GetR(pos, tr)				\
+   do {						\
+     tr = Arg(pos);				\
+     switch (loader_tag(tr)) {			\
+     case LOADER_X_REG:				\
+        tr = x(loader_x_reg_index(tr));		\
+        break;					\
+     case LOADER_Y_REG:				\
+        ASSERT(loader_y_reg_index(tr) >= 1);	\
+        tr = y(loader_y_reg_index(tr));		\
+        break;					\
+     }						\
+     CHECK_TERM(tr);				\
    } while (0)
 
 #define GetArg1(N, Dst) GetR((N), Dst)
@@ -2385,12 +2394,12 @@ void process_main(void)
 
      do {
 	 Eterm term = *I++;
-	 switch (term & _TAG_IMMED1_MASK) {
-	 case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:
-	     *hp++ = x(term >> _TAG_IMMED1_SIZE);
+	 switch (loader_tag(term)) {
+	 case LOADER_X_REG:
+	     *hp++ = x(loader_x_reg_index(term));
 	     break;
-	 case (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:
-	     *hp++ = y(term >> _TAG_IMMED1_SIZE);
+	 case LOADER_Y_REG:
+	     *hp++ = y(loader_y_reg_index(term));
 	     break;
 	 default:
 	     *hp++ = term;
@@ -2411,19 +2420,19 @@ void process_main(void)
      Next(3+Arg(2));
  }
 
-#define PUT_TERM_REG(term, desc)				\
-do {								\
-    switch ((desc) & _TAG_IMMED1_MASK) {			\
-    case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
-	x((desc) >> _TAG_IMMED1_SIZE) = (term);			\
-	break;							\
-    case (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
-	y((desc) >> _TAG_IMMED1_SIZE) = (term);			\
-	break;							\
-    default:							\
-	ASSERT(0);						\
-	break;							\
-    }								\
+#define PUT_TERM_REG(term, desc)		\
+do {						\
+    switch (loader_tag(desc)) {			\
+    case LOADER_X_REG:				\
+	x(loader_x_reg_index(desc)) = (term);	\
+	break;					\
+    case LOADER_Y_REG:				\
+	y(loader_y_reg_index(desc)) = (term);	\
+	break;					\
+    default:					\
+	ASSERT(0);				\
+	break;					\
+    }						\
 } while(0)
 
  OpCase(i_get_map_elements_fsI): {
@@ -6487,20 +6496,20 @@ static Eterm get_map_element_hash(Eterm map, Eterm key, Uint32 hx)
     return vs ? *vs : THE_NON_VALUE;
 }
 
-#define GET_TERM(term, dest)					\
-do {								\
-    Eterm src = (Eterm)(term);					\
-    switch (src & _TAG_IMMED1_MASK) {				\
-    case (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
-	dest = x(src >> _TAG_IMMED1_SIZE);			\
-	break;							\
-    case (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER:	\
-	dest = y(src >> _TAG_IMMED1_SIZE);			\
-	break;							\
-    default:							\
-	dest = src;						\
-	break;							\
-    }								\
+#define GET_TERM(term, dest)			\
+do {						\
+    Eterm src = (Eterm)(term);			\
+    switch (loader_tag(src)) {			\
+    case LOADER_X_REG:				\
+        dest = x(loader_x_reg_index(src));	\
+	break;					\
+    case LOADER_Y_REG:				\
+        dest = y(loader_y_reg_index(src));	\
+	break;					\
+    default:					\
+	dest = src;				\
+	break;					\
+    }						\
 } while(0)
 
 
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 495e92d600..3cd80c4ec2 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2167,13 +2167,13 @@ load_code(LoaderState* stp)
 	    case 's':	/* Any source (tagged constant or register) */
 		switch (tag) {
 		case TAG_r:
-		    code[ci++] = make_xreg(0);
+		    code[ci++] = make_loader_x_reg(0);
 		    break;
 		case TAG_x:
-		    code[ci++] = make_xreg(tmp_op->a[arg].val);
+		    code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
 		    break;
 		case TAG_y:
-		    code[ci++] = make_yreg(tmp_op->a[arg].val);
+		    code[ci++] = make_loader_y_reg(tmp_op->a[arg].val);
 		    break;
 		case TAG_i:
 		    code[ci++] = (BeamInstr) make_small((Uint)tmp_op->a[arg].val);
@@ -2184,6 +2184,10 @@ load_code(LoaderState* stp)
 		case TAG_n:
 		    code[ci++] = NIL;
 		    break;
+		case TAG_q:
+		    new_literal_patch(stp, ci);
+		    code[ci++] = tmp_op->a[arg].val;
+		    break;
 		default:
 		    LoadError1(stp, "bad tag %d for general source",
 			       tmp_op->a[arg].type);
@@ -2193,13 +2197,13 @@ load_code(LoaderState* stp)
 	    case 'd':	/* Destination (x(0), x(N), y(N) */
 		switch (tag) {
 		case TAG_r:
-		    code[ci++] = make_xreg(0);
+		    code[ci++] = make_loader_x_reg(0);
 		    break;
 		case TAG_x:
-		    code[ci++] = make_xreg(tmp_op->a[arg].val);
+		    code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
 		    break;
 		case TAG_y:
-		    code[ci++] = make_yreg(tmp_op->a[arg].val);
+		    code[ci++] = make_loader_y_reg(tmp_op->a[arg].val);
 		    break;
 		default:
 		    LoadError1(stp, "bad tag %d for destination",
@@ -2359,13 +2363,11 @@ load_code(LoaderState* stp)
 	    case TAG_r:
 	    case TAG_x:
 		CodeNeed(1);
-		code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) |
-		    (X_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER;
+		code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
 		break;
 	    case TAG_y:
 		CodeNeed(1);
-		code[ci++] = (tmp_op->a[arg].val << _TAG_IMMED1_SIZE) |
-		    (Y_REG_DEF << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER;
+		code[ci++] = make_loader_y_reg(tmp_op->a[arg].val);
 		break;
 	    case TAG_n:
 		CodeNeed(1);
diff --git a/erts/emulator/beam/erl_term.c b/erts/emulator/beam/erl_term.c
index 89459fb278..3a5fbcc284 100644
--- a/erts/emulator/beam/erl_term.c
+++ b/erts/emulator/beam/erl_term.c
@@ -173,9 +173,7 @@ ET_DEFINE_CHECKED(Uint,external_thing_data_words,ExternalThing*,is_thing_ptr);
 ET_DEFINE_CHECKED(Eterm,make_cp,UWord *,_is_taggable_pointer);
 ET_DEFINE_CHECKED(UWord *,cp_val,Eterm,is_CP);
 ET_DEFINE_CHECKED(Uint,catch_val,Eterm,is_catch);
-ET_DEFINE_CHECKED(Uint,x_reg_offset,Uint,_is_xreg);
-ET_DEFINE_CHECKED(Uint,y_reg_offset,Uint,_is_yreg);
-ET_DEFINE_CHECKED(Uint,x_reg_index,Uint,_is_xreg);
-ET_DEFINE_CHECKED(Uint,y_reg_index,Uint,_is_yreg);
+ET_DEFINE_CHECKED(Uint,loader_x_reg_index,Uint,_is_loader_x_reg);
+ET_DEFINE_CHECKED(Uint,loader_y_reg_index,Uint,_is_loader_y_reg);
 
 #endif	/* ET_DEBUG */
diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h
index cff2430547..089eecf024 100644
--- a/erts/emulator/beam/erl_term.h
+++ b/erts/emulator/beam/erl_term.h
@@ -1027,42 +1027,40 @@ _ET_DECLARE_CHECKED(Uint,catch_val,Eterm)
 /*
  * Overloaded tags.
  *
- * SMALL = 15
- * ATOM/NIL=7
+ * In the loader, we want to tag a term in a way so that it can
+ * be any literal (atom/integer/float/tuple/list/binary) or a
+ * register.
  *
- * Note that the two least significant bits in SMALL/ATOM/NIL always are 3;
- * thus, we can distinguish register from literals by looking at only these
- * two bits.
+ * We can achive that by overloading the PID and PORT tags to
+ * mean X and Y registers. That works because there are no
+ * pid or port literals.
  */
 
-#define X_REG_DEF	0
-#define Y_REG_DEF	1
+#define _LOADER_TAG_XREG _TAG_IMMED1_PID
+#define _LOADER_TAG_YREG _TAG_IMMED1_PORT
+#define _LOADER_TAG_SIZE _TAG_IMMED1_SIZE
+#define _LOADER_MASK _TAG_IMMED1_MASK
 
-#define beam_reg_tag(x)	((x) & 3)
+#define LOADER_X_REG _LOADER_TAG_XREG
+#define LOADER_Y_REG _LOADER_TAG_YREG
 
-#define make_xreg(ix)	(((ix) * sizeof(Eterm)) | X_REG_DEF)
-#define make_yreg(ix)	(((ix) * sizeof(Eterm)) | Y_REG_DEF)
+#define make_loader_x_reg(R) (((R) << _LOADER_TAG_SIZE) | _LOADER_TAG_XREG)
+#define make_loader_y_reg(R) (((R) << _LOADER_TAG_SIZE) | _LOADER_TAG_YREG)
 
-#define _is_xreg(x)	(beam_reg_tag(x) == X_REG_DEF)
-#define _is_yreg(x)	(beam_reg_tag(x) == Y_REG_DEF)
+#define loader_reg_index(R) ((R) >> _LOADER_TAG_SIZE)
 
-#define _unchecked_x_reg_offset(R)	((R) - X_REG_DEF)
-_ET_DECLARE_CHECKED(Uint,x_reg_offset,Uint)
-#define x_reg_offset(R)	_ET_APPLY(x_reg_offset,(R))
+#define loader_tag(T) ((T) & _LOADER_MASK)
 
-#define _unchecked_y_reg_offset(R)	((R) - Y_REG_DEF)
-_ET_DECLARE_CHECKED(Uint,y_reg_offset,Uint)
-#define y_reg_offset(R)	_ET_APPLY(y_reg_offset,(R))
+#define _is_loader_x_reg(x) (loader_tag(x) == _LOADER_TAG_XREG)
+#define _is_loader_y_reg(x) (loader_tag(x) == _LOADER_TAG_YREG)
 
-#define reg_index(R) ((R) / sizeof(Eterm))
+#define _unchecked_loader_x_reg_index(R) ((R) >> _LOADER_TAG_SIZE)
+_ET_DECLARE_CHECKED(Uint,loader_x_reg_index,Uint)
+#define loader_x_reg_index(R) _ET_APPLY(loader_x_reg_index,(R))
 
-#define _unchecked_x_reg_index(R)	((R) >> 2)
-_ET_DECLARE_CHECKED(Uint,x_reg_index,Uint)
-#define x_reg_index(R)	_ET_APPLY(x_reg_index,(R))
-
-#define _unchecked_y_reg_index(R)	((R) >> 2)
-_ET_DECLARE_CHECKED(Uint,y_reg_index,Uint)
-#define y_reg_index(R)	_ET_APPLY(y_reg_index,(R))
+#define _unchecked_loader_y_reg_index(R)  ((R) >> _LOADER_TAG_SIZE)
+_ET_DECLARE_CHECKED(Uint,loader_y_reg_index,Uint)
+#define loader_y_reg_index(R) _ET_APPLY(loader_y_reg_index,(R))
 
 /*
  * Backwards compatibility definitions:
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 1d32e72247..4fa6a087e2 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -237,12 +237,10 @@ try Y F => catch Y F
 try_case Y => try_end Y
 try_end y
 
-try_case_end Literal=q => move Literal x | try_case_end x
 try_case_end s
 
 # Destructive set tuple element
 
-set_tuple_element Lit=q Tuple Pos => move Lit x | set_tuple_element x Tuple Pos
 set_tuple_element s d P
 
 # Get tuple element
@@ -273,8 +271,8 @@ is_number Fail Literal=q => move Literal x | is_number Fail x
 
 jump f
 
-case_end Literal=cq => move Literal x | case_end x
-badmatch Literal=cq => move Literal x | badmatch x
+case_end Literal=c => move Literal x | case_end x
+badmatch Literal=c => move Literal x | badmatch x
 
 case_end r
 case_end x
@@ -849,7 +847,6 @@ is_boolean f y
 
 is_function2 Fail=f acq Arity => jump Fail
 is_function2 Fail=f Fun a => jump Fail
-is_function2 Fail Fun Literal=q => move Literal x | is_function2 Fail Fun x
 
 is_function2 f s s
 %macro: is_function2 IsFunction2 -fail_action
@@ -1122,11 +1119,8 @@ bif1 Fail Bif=u$bif:erlang:get/1 Src=s Dst=d => i_get Src Dst
 
 bif2 Jump=j u$bif:erlang:element/2 S1=s S2=rxy Dst=d => gen_element(Jump, S1, S2, Dst)
 
-bif1 Fail Bif Literal=q Dst => move Literal x | bif1 Fail Bif x Dst
 bif1 p Bif S1 Dst => bif1_body Bif S1 Dst
 
-bif1_body Bif Literal=q Dst => move Literal x | bif1_body Bif x Dst
-
 bif2 p Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2_body Bif Dst
 bif2 Fail Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2 Fail Bif Dst
 
@@ -1439,8 +1433,6 @@ i_bs_private_append j I d
 # Storing integers into binaries.
 #
 
-bs_put_integer Fail=j Sz=s Unit=u Flags=u Literal=q => \
-	move Literal x | bs_put_integer Fail Sz Unit Flags x
 bs_put_integer Fail=j Sz=sq Unit=u Flags=u Src=s => \
 			gen_put_integer(Fail, Sz, Unit, Flags, Src)
 
@@ -1454,32 +1446,22 @@ i_new_bs_put_integer_imm j I I s
 # Utf8/utf16/utf32 support. (R12B-5)
 #
 
-bs_utf8_size Fail Literal=q Dst=d => \
-	move Literal x | bs_utf8_size Fail x Dst
 bs_utf8_size j Src=s Dst=d => i_bs_utf8_size Src Dst
 
 i_bs_utf8_size s d
 
-bs_utf16_size Fail Literal=q Dst=d => \
-	move Literal x | bs_utf16_size Fail x Dst
 bs_utf16_size j Src=s Dst=d => i_bs_utf16_size Src Dst
 
 i_bs_utf16_size s d
 
-bs_put_utf8 Fail=j Flags=u Literal=q => \
-	move Literal x | bs_put_utf8 Fail Flags x
 bs_put_utf8 Fail u Src=s => i_bs_put_utf8 Fail Src
 
 i_bs_put_utf8 j s
 
-bs_put_utf16 Fail=j Flags=u Literal=q => \
-	move Literal x | bs_put_utf16 Fail Flags x
 bs_put_utf16 Fail Flags=u Src=s => i_bs_put_utf16 Fail Flags Src
 
 i_bs_put_utf16 j I s
 
-bs_put_utf32 Fail=j Flags=u Literal=q => \
-	move Literal x | bs_put_utf32 Fail Flags x
 bs_put_utf32 Fail=j Flags=u Src=s => \
    i_bs_validate_unicode Fail Src | bs_put_integer Fail i=32 u=1 Flags Src
 
@@ -1490,9 +1472,6 @@ i_bs_validate_unicode j s
 #
 bs_put_float Fail Sz=q Unit Flags Val => badarg Fail
 
-bs_put_float Fail=j Sz Unit=u Flags=u Literal=q => \
-	move Literal x | bs_put_float Fail Sz Unit Flags x
-
 bs_put_float Fail=j Sz=s Unit=u Flags=u Src=s => \
 			gen_put_float(Fail, Sz, Unit, Flags, Src)
 
@@ -1506,8 +1485,6 @@ i_new_bs_put_float_imm j I I s
 # Storing binaries into binaries.
 #
 
-bs_put_binary Fail Sz Unit Flags Literal=q => \
- 	move Literal x | bs_put_binary Fail Sz Unit Flags x
 bs_put_binary Fail=j Sz=s Unit=u Flags=u Src=s => \
 			gen_put_binary(Fail, Sz, Unit, Flags, Src)
 
@@ -1731,8 +1708,6 @@ gc_bif2 Fail I Bif S1 S2 Dst => \
 gc_bif3 Fail I Bif S1 S2 S3 Dst => \
 	gen_guard_bif3(Fail, I, Bif, S1, S2, S3, Dst)
 
-i_gc_bif1 Fail Bif V=q Live D => move V x | i_gc_bif1 Fail Bif x Live D
-
 i_gc_bif1 j I s I d
 
 ii_gc_bif2/6
-- 
cgit v1.2.3


From 135c9821e8806defa455a47208fb897fff1a3ff8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 23 Jun 2015 08:53:50 +0200
Subject: Make the 'r' operand type optional

The 'r' type is now mandatory. That means in order to handle
both of the following instructions:

  move x(0) y(7)
  move x(1) y(7)

we would need to define two specific operations in ops.tab:

  move r y
  move x y

We want to make 'r' operands optional. That is, if we have
only this specific instruction:

  move x y

it will match both of the following instructions:

  move x(0) y(7)
  move x(1) y(7)

Make 'r' optional allows us to save code space when we don't
want to make handling of x(0) a special case, but we can still
use 'r' to optimize commonly used instructions.
---
 erts/emulator/beam/beam_debug.c |   6 -
 erts/emulator/beam/beam_emu.c   | 355 ++++++-------------------------
 erts/emulator/beam/beam_load.c  |  66 ++++--
 erts/emulator/beam/erl_vm.h     |   1 +
 erts/emulator/beam/ops.tab      | 456 ++++++++++++++--------------------------
 5 files changed, 271 insertions(+), 613 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 243ccb9afd..30e3765502 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -569,7 +569,6 @@ 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_lins_rfI:
     case op_i_select_val_lins_xfI:
     case op_i_select_val_lins_yfI:
 	{
@@ -589,7 +588,6 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    }
 	}
 	break;
-    case op_i_select_val_bins_rfI:
     case op_i_select_val_bins_xfI:
     case op_i_select_val_bins_yfI:
 	{
@@ -603,7 +601,6 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    }
 	}
 	break;
-    case op_i_select_tuple_arity_rfI:
     case op_i_select_tuple_arity_xfI:
     case op_i_select_tuple_arity_yfI:
         {
@@ -628,7 +625,6 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
             }
         }
         break;
-    case op_i_jump_on_val_rfII:
     case op_i_jump_on_val_xfII:
     case op_i_jump_on_val_yfII:
 	{
@@ -640,7 +636,6 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    }
 	}
 	break;
-    case op_i_jump_on_val_zero_rfI:
     case op_i_jump_on_val_zero_xfI:
     case op_i_jump_on_val_zero_yfI:
 	{
@@ -652,7 +647,6 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    }
 	}
 	break;
-    case op_i_put_tuple_rI:
     case op_i_put_tuple_xI:
     case op_i_put_tuple_yI:
     case op_new_map_dII:
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index a74da2d872..e5179830de 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -556,12 +556,12 @@ void** beam_ops;
 #define Move2(S1, D1, S2, D2) D1 = (S1); D2 = (S2)
 #define Move3(S1, D1, S2, D2, S3, D3) D1 = (S1); D2 = (S2); D3 = (S3)
 
-#define MoveReturn(Src, Dest)       \
-    (Dest) = (Src);                 \
-    I = c_p->cp;                    \
-    ASSERT(VALID_INSTR(*c_p->cp));  \
-    c_p->cp = 0;                    \
-    CHECK_TERM(r(0));               \
+#define MoveReturn(Src)				\
+    x(0) = (Src);				\
+    I = c_p->cp;				\
+    ASSERT(VALID_INSTR(*c_p->cp));		\
+    c_p->cp = 0;				\
+    CHECK_TERM(r(0));				\
     Goto(*I)
 
 #define DeallocateReturn(Deallocate)       \
@@ -573,26 +573,26 @@ void** beam_ops;
     Goto(*I);                              \
   } while (0)
 
-#define MoveDeallocateReturn(Src, Dest, Deallocate)  \
-    (Dest) = (Src);                                  \
+#define MoveDeallocateReturn(Src, Deallocate)	\
+    x(0) = (Src);				\
     DeallocateReturn(Deallocate)
 
-#define MoveCall(Src, Dest, CallDest, Size)	\
-    (Dest) = (Src);				\
+#define MoveCall(Src, CallDest, Size)		\
+    x(0) = (Src);				\
     SET_CP(c_p, I+Size+1);			\
-    SET_I((BeamInstr *) CallDest);			\
+    SET_I((BeamInstr *) CallDest);		\
     Dispatch();
 
-#define MoveCallLast(Src, Dest, CallDest, Deallocate)	\
-    (Dest) = (Src);					\
-    RESTORE_CP(E);					\
-    E = ADD_BYTE_OFFSET(E, (Deallocate));		\
-    SET_I((BeamInstr *) CallDest);				\
+#define MoveCallLast(Src, CallDest, Deallocate)	\
+    x(0) = (Src);				\
+    RESTORE_CP(E);				\
+    E = ADD_BYTE_OFFSET(E, (Deallocate));	\
+    SET_I((BeamInstr *) CallDest);		\
     Dispatch();
 
-#define MoveCallOnly(Src, Dest, CallDest)	\
-    (Dest) = (Src);				\
-    SET_I((BeamInstr *) CallDest);			\
+#define MoveCallOnly(Src, CallDest)		\
+    x(0) = (Src);				\
+    SET_I((BeamInstr *) CallDest);		\
     Dispatch();
 
 #define MoveJump(Src)				\
@@ -676,8 +676,8 @@ void** beam_ops;
     if (is_not_list(Src)) { Fail; }                     \
     A(Need, Alive)
 
-#define IsNonemptyListTestHeap(Src, Need, Alive, Fail)  \
-    if (is_not_list(Src)) { Fail; }                     \
+#define IsNonemptyListTestHeap(Need, Alive, Fail)  \
+    if (is_not_list(x(0))) { Fail; }			\
     TestHeap(Need, Alive)
 
 #define IsTuple(X, Action) if (is_not_tuple(X)) Action
@@ -1316,17 +1316,13 @@ void process_main(void)
 	    Uint live;
 	    Eterm result;
 
-	OpCase(i_increment_yIId):
-	    increment_reg_val = yb(Arg(0));
-	    goto do_increment;
-
 	OpCase(i_increment_xIId):
 	    increment_reg_val = xb(Arg(0));
 	    goto do_increment;
 
-	OpCase(i_increment_rIId):
-	    increment_reg_val = r(0);
-	    I--;
+	OpCase(i_increment_yIId):
+	    increment_reg_val = yb(Arg(0));
+	    goto do_increment;
 
 	do_increment:
 	    increment_val = Arg(1);
@@ -1335,7 +1331,6 @@ void process_main(void)
 		ASSERT(MY_IS_SSMALL(i) == IS_SSMALL(i));
 		if (MY_IS_SSMALL(i)) {
 		    result = make_small(i);
-		store_result:
 		    StoreBifResult(3, result);
 		}
 	    }
@@ -1348,7 +1343,7 @@ void process_main(void)
 	    SWAPIN;
 	    ERTS_HOLE_CHECK(c_p);
 	    if (is_value(result)) {
-		goto store_result;
+		StoreBifResult(3, result);
 	    }
 	    ASSERT(c_p->freason != BADMATCH || is_value(c_p->fvalue));
 	    goto find_func_info;
@@ -1464,47 +1459,37 @@ void process_main(void)
     {
 	Eterm is_eq_exact_lit_val;
 
-    OpCase(i_is_eq_exact_literal_xfc):
-	is_eq_exact_lit_val = xb(Arg(0));
-	I++;
+    OpCase(i_is_eq_exact_literal_fxc):
+	is_eq_exact_lit_val = xb(Arg(1));
 	goto do_is_eq_exact_literal;
 
-    OpCase(i_is_eq_exact_literal_yfc):
-	is_eq_exact_lit_val = yb(Arg(0));
-	I++;
+    OpCase(i_is_eq_exact_literal_fyc):
+	is_eq_exact_lit_val = yb(Arg(1));
 	goto do_is_eq_exact_literal;
 
-    OpCase(i_is_eq_exact_literal_rfc):
-	is_eq_exact_lit_val = r(0);
-
     do_is_eq_exact_literal:
-	if (!eq(Arg(1), is_eq_exact_lit_val)) {
+	if (!eq(Arg(2), is_eq_exact_lit_val)) {
 	    ClauseFail();
 	}
-	Next(2);
+	Next(3);
     }
 
     {
 	Eterm is_ne_exact_lit_val;
 
-    OpCase(i_is_ne_exact_literal_xfc):
-	is_ne_exact_lit_val = xb(Arg(0));
-	I++;
+    OpCase(i_is_ne_exact_literal_fxc):
+	is_ne_exact_lit_val = xb(Arg(1));
 	goto do_is_ne_exact_literal;
 
-    OpCase(i_is_ne_exact_literal_yfc):
-	is_ne_exact_lit_val = yb(Arg(0));
-	I++;
+    OpCase(i_is_ne_exact_literal_fyc):
+	is_ne_exact_lit_val = yb(Arg(1));
 	goto do_is_ne_exact_literal;
 
-    OpCase(i_is_ne_exact_literal_rfc):
-	is_ne_exact_lit_val = r(0);
-
     do_is_ne_exact_literal:
-	if (eq(Arg(1), is_ne_exact_lit_val)) {
+	if (eq(Arg(2), is_ne_exact_lit_val)) {
 	    ClauseFail();
 	}
-	Next(2);
+	Next(3);
     }
 
  OpCase(move_window3_xxxy): {
@@ -1553,7 +1538,7 @@ void process_main(void)
      NextPF(6, next);
  }
 
- OpCase(i_move_call_only_fcr): {
+ OpCase(i_move_call_only_fc): {
      r(0) = Arg(1);
  }
  /* FALL THROUGH */
@@ -1563,7 +1548,7 @@ void process_main(void)
      Dispatch();
  }
 
- OpCase(i_move_call_last_fPcr): {
+ OpCase(i_move_call_last_fPc): {
      r(0) = Arg(2);
  }
  /* FALL THROUGH */
@@ -1575,7 +1560,7 @@ void process_main(void)
      Dispatch();
  }
 
- OpCase(i_move_call_crf): {
+ OpCase(i_move_call_cf): {
      r(0) = Arg(0);
      I++;
  }
@@ -1587,7 +1572,7 @@ void process_main(void)
      Dispatch();
  }
 
- OpCase(i_move_call_ext_last_ePcr): {
+ OpCase(i_move_call_ext_last_ePc): {
      r(0) = Arg(2);
  }
  /* FALL THROUGH */
@@ -1603,7 +1588,7 @@ void process_main(void)
     DTRACE_GLOBAL_CALL_FROM_EXPORT(c_p, Arg(0));
     Dispatchx();
 
- OpCase(i_move_call_ext_cre): {
+ OpCase(i_move_call_ext_ce): {
      r(0) = Arg(0);
      I++;
  }
@@ -1613,7 +1598,7 @@ void process_main(void)
     DTRACE_GLOBAL_CALL_FROM_EXPORT(c_p, Arg(0));
     Dispatchx();
 
- OpCase(i_move_call_ext_only_ecr): {
+ OpCase(i_move_call_ext_only_ec): {
      r(0) = Arg(1);
  }
  /* FALL THROUGH */
@@ -1707,29 +1692,23 @@ void process_main(void)
 	Eterm element_index;
 	Eterm element_tuple;
 
-    OpCase(i_element_xjsd):
-	element_tuple = xb(Arg(0));
-	I++;
+    OpCase(i_element_jxsd):
+	element_tuple = xb(Arg(1));
 	goto do_element;
 
-    OpCase(i_element_yjsd):
-	element_tuple = yb(Arg(0));
-	I++;
+    OpCase(i_element_jysd):
+	element_tuple = yb(Arg(1));
 	goto do_element;
 
-    OpCase(i_element_rjsd):
-	element_tuple = r(0);
-	/* Fall through */
-
     do_element:
-	GetArg1(1, element_index);
+	GetArg1(2, element_index);
 	if (is_small(element_index) && is_tuple(element_tuple)) {
 	    Eterm* tp = tuple_val(element_tuple);
 
 	    if ((signed_val(element_index) >= 1) &&
 		(signed_val(element_index) <= arityval(*tp))) {
 		Eterm result = tp[signed_val(element_index)];
-		StoreBifResult(2, result);
+		StoreBifResult(3, result);
 	    }
 	}
     }
@@ -1743,29 +1722,24 @@ void process_main(void)
     {
 	Eterm fast_element_tuple;
 
-    OpCase(i_fast_element_rjId):
-	fast_element_tuple = r(0);
+    OpCase(i_fast_element_jxId):
+     fast_element_tuple = xb(Arg(1));
+     goto do_fast_element;
+
+    OpCase(i_fast_element_jyId):
+     fast_element_tuple = yb(Arg(1));
+     goto do_fast_element;
 
     do_fast_element:
 	if (is_tuple(fast_element_tuple)) {
 	    Eterm* tp = tuple_val(fast_element_tuple);
-	    Eterm pos = Arg(1);	/* Untagged integer >= 1 */
+	    Eterm pos = Arg(2);	/* Untagged integer >= 1 */
 	    if (pos <= arityval(*tp)) {
 		Eterm result = tp[pos];
-		StoreBifResult(2, result);
+		StoreBifResult(3, result);
 	    }
 	}
      goto badarg;
-
-    OpCase(i_fast_element_xjId):
-     fast_element_tuple = xb(Arg(0));
-     I++;
-     goto do_fast_element;
-
-    OpCase(i_fast_element_yjId):
-     fast_element_tuple = yb(Arg(0));
-     I++;
-     goto do_fast_element;
  }
 
  OpCase(catch_yf):
@@ -1871,7 +1845,7 @@ void process_main(void)
      * Pick up the next message and place it in x(0).
      * If no message, jump to a wait or wait_timeout instruction.
      */
- OpCase(i_loop_rec_fr):
+ OpCase(i_loop_rec_f):
  {
      BeamInstr *next;
      ErlMessage* msgp;
@@ -2183,10 +2157,6 @@ void process_main(void)
      select_val2 = xb(Arg(0));
      goto do_select_tuple_arity2;
 
- OpCase(i_select_tuple_arity2_rfAAff):
-     select_val2 = r(0);
-     I--;
-
  do_select_tuple_arity2:
      if (is_not_tuple(select_val2)) {
 	 goto select_val2_fail;
@@ -2202,10 +2172,6 @@ void process_main(void)
      select_val2 = xb(Arg(0));
      goto do_select_val2;
 
- OpCase(i_select_val2_rfccff):
-     select_val2 = r(0);
-     I--;
-
  do_select_val2:
      if (select_val2 == Arg(2)) {
 	 I += 3;
@@ -2229,10 +2195,6 @@ void process_main(void)
      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);
@@ -2249,10 +2211,6 @@ void process_main(void)
      select_val = yb(Arg(0));
      goto do_linear_search;
 
- OpCase(i_select_val_lins_rfI):
-     select_val = r(0);
-     I--;
-
  do_linear_search: {
      BeamInstr *vs = &Arg(3);
      int ix = 0;
@@ -2279,10 +2237,6 @@ void process_main(void)
      select_val = yb(Arg(0));
      goto do_binary_search;
      
- OpCase(i_select_val_bins_rfI):
-     select_val = r(0);
-     I--;
-
  do_binary_search:
  {
      struct Pairs {
@@ -2343,10 +2297,6 @@ void process_main(void)
      jump_on_val_zero_index = xb(Arg(0));
      goto do_jump_on_val_zero_index;
 
- OpCase(i_jump_on_val_zero_rfI):
-     jump_on_val_zero_index = r(0);
-     I--;
-
  do_jump_on_val_zero_index:
      if (is_small(jump_on_val_zero_index)) {
 	 jump_on_val_zero_index = signed_val(jump_on_val_zero_index);
@@ -2371,10 +2321,6 @@ void process_main(void)
      jump_on_val_index = xb(Arg(0));
      goto do_jump_on_val_index;
 
- OpCase(i_jump_on_val_rfII):
-     jump_on_val_index = r(0);
-     I--;
-
  do_jump_on_val_index:
      if (is_small(jump_on_val_index)) {
 	 jump_on_val_index = (Uint) (signed_val(jump_on_val_index) - Arg(3));
@@ -3441,18 +3387,8 @@ do {						\
     {
 	Eterm badmatch_val;
 
-    OpCase(badmatch_y):
-	badmatch_val = yb(Arg(0));
-	goto do_badmatch;
-
     OpCase(badmatch_x):
 	badmatch_val = xb(Arg(0));
-	goto do_badmatch;
-
-    OpCase(badmatch_r):
-	badmatch_val = r(0);
-
-    do_badmatch:
 	c_p->fvalue = badmatch_val;
 	c_p->freason = BADMATCH;
     }
@@ -3627,16 +3563,6 @@ do {						\
 
     OpCase(case_end_x):
 	case_end_val = xb(Arg(0));
-	goto do_case_end;
-
-    OpCase(case_end_y):
-	case_end_val = yb(Arg(0));
-	goto do_case_end;
-
-    OpCase(case_end_r):
-	case_end_val = r(0);
-
-    do_case_end:
 	c_p->fvalue = case_end_val;
 	c_p->freason = EXC_CASE_CLAUSE;
 	goto find_func_info;
@@ -3692,11 +3618,6 @@ do {						\
 	 goto do_bs_init_bits;
      }
 
-     OpCase(i_bs_init_bits_fail_rjId): {
-	 num_bits_term = r(0);
-	 alloc = 0;
-	 goto do_bs_init_bits;
-     }
      OpCase(i_bs_init_bits_fail_yjId): {
 	 num_bits_term = yb(Arg(0));
 	 I++;
@@ -3823,12 +3744,6 @@ do {						\
 	 goto do_bs_init;
      }
 
-     OpCase(i_bs_init_fail_rjId): {
-	 tmp_arg1 = r(0);
-	 tmp_arg2 = 0;
-	 goto do_bs_init;
-     }
-
      OpCase(i_bs_init_fail_yjId): {
 	 tmp_arg1 = yb(Arg(0));
 	 tmp_arg2 = 0;
@@ -4202,9 +4117,6 @@ do {						\
      Uint slots;
      Eterm context;
 
-     OpCase(i_bs_start_match2_rfIId): {
-	 context = r(0);
-
      do_start_match:
 	 slots = Arg(2);
 	 if (!is_boxed(context)) {
@@ -4251,7 +4163,7 @@ do {						\
 	     ClauseFail();
 	 }
 	 NextPF(4, next);
-     }
+
      OpCase(i_bs_start_match2_xfIId): {
 	 context = xb(Arg(0));
 	 I++;
@@ -4264,18 +4176,6 @@ do {						\
      }
  }
 
- OpCase(bs_test_zero_tail2_fr): {
-     BeamInstr *next;
-     ErlBinMatchBuffer *_mb;
-     
-     PreFetch(1, next);
-     _mb = (ErlBinMatchBuffer*) ms_matchbuffer(r(0));
-     if (_mb->size != _mb->offset) {
-	 ClauseFail();
-     }
-     NextPF(1, next);
- }
-
  OpCase(bs_test_zero_tail2_fx): {
      BeamInstr *next;
      ErlBinMatchBuffer *_mb;
@@ -4288,16 +4188,6 @@ do {						\
      NextPF(2, next);
  }
 
- OpCase(bs_test_tail_imm2_frI): {
-     BeamInstr *next;
-     ErlBinMatchBuffer *_mb;
-     PreFetch(2, next);
-     _mb = ms_matchbuffer(r(0));
-     if (_mb->size - _mb->offset != Arg(1)) {
-	 ClauseFail();
-     }
-     NextPF(2, next);
- }
  OpCase(bs_test_tail_imm2_fxI): {
      BeamInstr *next;
      ErlBinMatchBuffer *_mb;
@@ -4309,16 +4199,6 @@ do {						\
      NextPF(3, next);
  }
 
- OpCase(bs_test_unit_frI): {
-     BeamInstr *next;
-     ErlBinMatchBuffer *_mb;
-     PreFetch(2, next);
-     _mb = ms_matchbuffer(r(0));
-     if ((_mb->size - _mb->offset) % Arg(1)) {
-	 ClauseFail();
-     }
-     NextPF(2, next);
- }
  OpCase(bs_test_unit_fxI): {
      BeamInstr *next;
      ErlBinMatchBuffer *_mb;
@@ -4330,16 +4210,6 @@ do {						\
      NextPF(3, next);
  }
 
- OpCase(bs_test_unit8_fr): {
-     BeamInstr *next;
-     ErlBinMatchBuffer *_mb;
-     PreFetch(1, next);
-     _mb = ms_matchbuffer(r(0));
-     if ((_mb->size - _mb->offset) & 7) {
-	 ClauseFail();
-     }
-     NextPF(1, next);
- }
  OpCase(bs_test_unit8_fx): {
      BeamInstr *next;
      ErlBinMatchBuffer *_mb;
@@ -4354,19 +4224,11 @@ do {						\
  {
      Eterm bs_get_integer8_context;
 
- OpCase(i_bs_get_integer_8_rfd): {
-	 bs_get_integer8_context = r(0);
-	 goto do_bs_get_integer_8;
-     }
-
  OpCase(i_bs_get_integer_8_xfd): {
-	 bs_get_integer8_context = xb(Arg(0));
-	 I++;
-     }
-
- do_bs_get_integer_8: {
 	 ErlBinMatchBuffer *_mb;
 	 Eterm _result;
+	 bs_get_integer8_context = xb(Arg(0));
+	 I++;
 	 _mb = ms_matchbuffer(bs_get_integer8_context);
 	 if (_mb->size - _mb->offset < 8) {
 	     ClauseFail();
@@ -4384,15 +4246,10 @@ do {						\
  {
      Eterm bs_get_integer_16_context;
 
- OpCase(i_bs_get_integer_16_rfd):
-     bs_get_integer_16_context = r(0);
-     goto do_bs_get_integer_16;
-
  OpCase(i_bs_get_integer_16_xfd):
      bs_get_integer_16_context = xb(Arg(0));
      I++;
 
- do_bs_get_integer_16:
      {
 	 ErlBinMatchBuffer *_mb;
 	 Eterm _result;
@@ -4413,17 +4270,10 @@ do {						\
  {
      Eterm bs_get_integer_32_context;
 
- OpCase(i_bs_get_integer_32_rfId):
-     bs_get_integer_32_context = r(0);
-     goto do_bs_get_integer_32;
-
-     
  OpCase(i_bs_get_integer_32_xfId):
      bs_get_integer_32_context = xb(Arg(0));
      I++;
 
-
- do_bs_get_integer_32:
      {
 	 ErlBinMatchBuffer *_mb;
 	 Uint32 _integer;
@@ -4452,13 +4302,6 @@ do {						\
      }
  }
 
- /* Operands: Size Live Fail Flags Dst */
- OpCase(i_bs_get_integer_imm_rIIfId): {
-     tmp_arg1 = r(0);
-     /* Operands: Size Live Fail Flags Dst */
-     goto do_bs_get_integer_imm_test_heap;
- }
-
  /* Operands: x(Reg) Size Live Fail Flags Dst */
  OpCase(i_bs_get_integer_imm_xIIfId): {
      tmp_arg1 = xb(Arg(0));
@@ -4481,15 +4324,6 @@ do {						\
      goto do_bs_get_integer_imm;
  }
 
- /* Operands: Size Fail Flags Dst */
- OpCase(i_bs_get_integer_small_imm_rIfId): {
-     tmp_arg1 = r(0);
-     tmp_arg2 = Arg(0);
-     I++;
-     /* Operands: Fail Flags Dst */
-     goto do_bs_get_integer_imm;
- }
-
  /* Operands: x(Reg) Size Fail Flags Dst */
  OpCase(i_bs_get_integer_small_imm_xIfId): {
      tmp_arg1 = xb(Arg(0));
@@ -4563,11 +4397,6 @@ do {						\
      Eterm get_utf8_context;
 
      /* Operands: MatchContext Fail Dst */
- OpCase(i_bs_get_utf8_rfd): {
-	 get_utf8_context = r(0);
-	 goto do_bs_get_utf8;
-     }
-
  OpCase(i_bs_get_utf8_xfd): {
 	 get_utf8_context = xb(Arg(0));
 	 I++;
@@ -4578,7 +4407,7 @@ do {						\
       * Operands: Fail Dst
       */
 
- do_bs_get_utf8: {
+     {
 	 Eterm result = erts_bs_get_utf8(ms_matchbuffer(get_utf8_context));
 	 if (is_non_value(result)) {
 	     ClauseFail();
@@ -4591,12 +4420,7 @@ do {						\
      Eterm get_utf16_context;
 
      /* Operands: MatchContext Fail Flags Dst */
- OpCase(i_bs_get_utf16_rfId): {
-	 get_utf16_context = r(0);
-	 goto do_bs_get_utf16;
-     }
-
- OpCase(i_bs_get_utf16_xfId): {
+     OpCase(i_bs_get_utf16_xfId): {
 	 get_utf16_context = xb(Arg(0));
 	 I++;
      }
@@ -4605,7 +4429,7 @@ do {						\
       * get_utf16_context = match_context
       * Operands: Fail Flags Dst
       */
- do_bs_get_utf16: {
+     {
 	 Eterm result = erts_bs_get_utf16(ms_matchbuffer(get_utf16_context),
 					  Arg(1));
 	 if (is_non_value(result)) {
@@ -4624,26 +4448,10 @@ do {						\
      Uint orig;
      Uint hole_size;
 
-     OpCase(bs_context_to_binary_r): {
-	 context_to_binary_context = r(0);
-	 I -= 2;
-	 goto do_context_to_binary;
-     }
-
-     /* Unfortunately, inlining can generate this instruction. */
-     OpCase(bs_context_to_binary_y): {
-	 context_to_binary_context = yb(Arg(0));
-	 goto do_context_to_binary0;
-     }
-
-     OpCase(bs_context_to_binary_x): {
+     OpCase(bs_context_to_binary_x):
 	 context_to_binary_context = xb(Arg(0));
-     
-     do_context_to_binary0:
 	 I--;
-     }
 
- do_context_to_binary:
      if (is_boxed(context_to_binary_context) &&
 	 header_is_bin_matchstate(*boxed_val(context_to_binary_context))) {
 	 ErlBinMatchState* ms;
@@ -4655,17 +4463,11 @@ do {						\
      }
      Next(2);
 
-     OpCase(i_bs_get_binary_all_reuse_rfI): {
-	 context_to_binary_context = r(0);
-	 goto do_bs_get_binary_all_reuse;
-     }
-
      OpCase(i_bs_get_binary_all_reuse_xfI): {
 	 context_to_binary_context = xb(Arg(0));
 	 I++;
      }
 
- do_bs_get_binary_all_reuse:
      mb = ms_matchbuffer(context_to_binary_context);
      size = mb->size - mb->offset;
      if (size % Arg(1) != 0) {
@@ -4693,16 +4495,11 @@ do {						\
  {
      Eterm match_string_context;
 
-     OpCase(i_bs_match_string_rfII): {
-	 match_string_context = r(0);
-	 goto do_bs_match_string;
-     }
      OpCase(i_bs_match_string_xfII): {
 	 match_string_context = xb(Arg(0));
 	 I++;
      }
 
- do_bs_match_string:
      {
 	 BeamInstr *next;
 	 byte* bytes;
@@ -4730,14 +4527,6 @@ do {						\
      }
  }
 
- OpCase(i_bs_save2_rI): {
-     BeamInstr *next;
-     ErlBinMatchState *_ms;
-     PreFetch(1, next);
-     _ms = (ErlBinMatchState*) boxed_val((Eterm) r(0));
-     _ms->save_offset[Arg(0)] = _ms->mb.offset;
-     NextPF(1, next);
- }
  OpCase(i_bs_save2_xI): {
      BeamInstr *next;
      ErlBinMatchState *_ms;
@@ -4747,14 +4536,6 @@ do {						\
      NextPF(2, next);
  }
 
- OpCase(i_bs_restore2_rI): {
-     BeamInstr *next;
-     ErlBinMatchState *_ms;
-     PreFetch(1, next);
-     _ms = (ErlBinMatchState*) boxed_val((Eterm) r(0));
-     _ms->mb.offset = _ms->save_offset[Arg(0)];
-     NextPF(1, next);
- }
  OpCase(i_bs_restore2_xI): {
      BeamInstr *next;
      ErlBinMatchState *_ms;
@@ -5030,7 +4811,9 @@ do {						\
      switch( c_p->def_arg_reg[3] ) {
        case HIPE_MODE_SWITCH_RES_RETURN:
 	 ASSERT(is_value(reg[0]));
-	 MoveReturn(reg[0], r(0));
+	 SET_I(c_p->cp);
+	 c_p->cp = 0;
+	 Goto(*I);
        case HIPE_MODE_SWITCH_RES_CALL_EXPORTED:
 	 c_p->i = c_p->hipe.u.callee_exp->addressv[erts_active_code_ix()];
 	 /*fall through*/
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 3cd80c4ec2..3671baa552 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -1847,9 +1847,7 @@ load_code(LoaderState* stp)
 	    case TAG_o:
 		break;
 	    case TAG_x:
-		if (last_op->a[arg].val == 0) {
-		    last_op->a[arg].type = TAG_r;
-		} else if (last_op->a[arg].val >= MAX_REG) {
+		if (last_op->a[arg].val >= MAX_REG) {
 		    LoadError1(stp, "invalid x register number: %u",
 			       last_op->a[arg].val);
 		}
@@ -2055,7 +2053,42 @@ load_code(LoaderState* stp)
 		if (((opc[specific].mask[0] & mask[0]) == mask[0]) &&
 		    ((opc[specific].mask[1] & mask[1]) == mask[1]) &&
 		    ((opc[specific].mask[2] & mask[2]) == mask[2])) {
-		    break;
+
+		    if (!opc[specific].involves_r) {
+			break;	/* No complications - match */
+		    }
+
+		    /*
+		     * The specific operation uses the 'r' operand,
+		     * which is shorthand for x(0). Now things
+		     * get complicated. First we must check whether
+		     * all operands that should be of type 'r' use
+		     * x(0) (as opposed to some other X register).
+		     */
+		    for (arg = 0; arg < arity; arg++) {
+			if (opc[specific].involves_r & (1 << arg) &&
+			    tmp_op->a[arg].type == TAG_x) {
+			    if (tmp_op->a[arg].val != 0) {
+				break; /* Other X register than 0 */
+			    }
+			}
+		    }
+
+		    if (arg == arity) {
+			/*
+			 * All 'r' operands use x(0) in the generic
+			 * operation. That means a match. Now we
+			 * will need to rewrite the generic instruction
+			 * to actually use 'r' instead of 'x(0)'.
+			 */
+			for (arg = 0; arg < arity; arg++) {
+			    if (opc[specific].involves_r & (1 << arg) &&
+				tmp_op->a[arg].type == TAG_x) {
+				tmp_op->a[arg].type = TAG_r;
+			    }
+			}
+			break;	/* Match */
+		    }
 		}
 		specific++;
 	    }
@@ -2166,9 +2199,6 @@ load_code(LoaderState* stp)
 		break;
 	    case 's':	/* Any source (tagged constant or register) */
 		switch (tag) {
-		case TAG_r:
-		    code[ci++] = make_loader_x_reg(0);
-		    break;
 		case TAG_x:
 		    code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
 		    break;
@@ -2196,9 +2226,6 @@ load_code(LoaderState* stp)
 		break;
 	    case 'd':	/* Destination (x(0), x(N), y(N) */
 		switch (tag) {
-		case TAG_r:
-		    code[ci++] = make_loader_x_reg(0);
-		    break;
 		case TAG_x:
 		    code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
 		    break;
@@ -2360,7 +2387,6 @@ load_code(LoaderState* stp)
 		stp->labels[tmp_op->a[arg].val].patches = ci;
 		ci++;
 		break;
-	    case TAG_r:
 	    case TAG_x:
 		CodeNeed(1);
 		code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
@@ -2455,7 +2481,6 @@ load_code(LoaderState* stp)
 	    stp->on_load = ci;
 	    break;
 	case op_bs_put_string_II:
-	case op_i_bs_match_string_rfII:
 	case op_i_bs_match_string_xfII:
 	    new_string_patch(stp, ci-1);
 	    break;
@@ -2693,17 +2718,17 @@ gen_element(LoaderState* stp, GenOpArg Fail, GenOpArg Index,
     op->next = NULL;
 
     if (Index.type == TAG_i && Index.val > 0 &&
-	(Tuple.type == TAG_r || Tuple.type == TAG_x || Tuple.type == TAG_y)) {
+	(Tuple.type == TAG_x || Tuple.type == TAG_y)) {
 	op->op = genop_i_fast_element_4;
-	op->a[0] = Tuple;
-	op->a[1] = Fail;
+	op->a[0] = Fail;
+	op->a[1] = Tuple;
 	op->a[2].type = TAG_u;
 	op->a[2].val = Index.val;
 	op->a[3] = Dst;
     } else {
 	op->op = genop_i_element_4;
-	op->a[0] = Tuple;
-	op->a[1] = Fail;
+	op->a[0] = Fail;
+	op->a[1] = Tuple;
 	op->a[2] = Index;
 	op->a[3] = Dst;
     }
@@ -4798,7 +4823,8 @@ transform_engine(LoaderState* st)
 	    if (var[i].type != instr->a[ap].type)
 		goto restart;
 	    switch (var[i].type) {
-	    case TAG_r: case TAG_n: break;
+	    case TAG_n:
+		break;
 	    default:
 		if (var[i].val != instr->a[ap].val)
 		    goto restart;
@@ -5865,7 +5891,7 @@ make_stub(BeamInstr* fp, Eterm mod, Eterm func, Uint arity, Uint native, BeamIns
     fp[4] = arity;
 #ifdef HIPE
     if (native) {
-	fp[5] = BeamOpCode(op_move_return_nr);
+	fp[5] = BeamOpCode(op_move_return_n);
 	hipe_mfa_save_orig_beam_op(mod, func, arity, fp+5);
     }
 #endif
@@ -6301,7 +6327,7 @@ erts_make_stub_module(Process* p, Eterm Mod, Eterm Beam, Eterm Info)
 #ifdef HIPE
 	op = (Eterm) BeamOpCode(op_hipe_trap_call); /* Might be changed later. */
 #else
-	op = (Eterm) BeamOpCode(op_move_return_nr);
+	op = (Eterm) BeamOpCode(op_move_return_n);
 #endif
 	fp = make_stub(fp, Mod, func, arity, (Uint)native_address, op);
     }
diff --git a/erts/emulator/beam/erl_vm.h b/erts/emulator/beam/erl_vm.h
index cfc978b5a5..98f27a1725 100644
--- a/erts/emulator/beam/erl_vm.h
+++ b/erts/emulator/beam/erl_vm.h
@@ -141,6 +141,7 @@
 typedef struct op_entry {
    char* name;			/* Name of instruction. */
    Uint32 mask[3];		/* Signature mask. */
+   unsigned involves_r;		/* Needs special attention when matching. */
    int sz;			/* Number of loaded words. */
    char* pack;			/* Instructions for packing engine. */
    char* sign;			/* Signature string. */
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 4fa6a087e2..335f8e2db5 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -39,8 +39,8 @@ too_old_compiler | never() =>
 # necessary.) Since the instructions don't work correctly in R12B, simply
 # refuse to load the module.
 
-func_info M=a a==am_module_info A=u==0 | label L | move n r => too_old_compiler
-func_info M=a a==am_module_info A=u==1 | label L | move n r => too_old_compiler
+func_info M=a a==am_module_info A=u==0 | label L | move n x==0 => too_old_compiler
+func_info M=a a==am_module_info A=u==1 | label L | move n x==0 => too_old_compiler
 
 # The undocumented and unsupported guard BIF is_constant/1 was removed
 # in R13. The is_constant/2 operation is marked as obsolete in genop.tab,
@@ -77,14 +77,14 @@ return
 # there is no line/1 instruction between the move and the call.
 #
 
-move S r | line Loc | call_ext Ar Func => \
-     line Loc | move S r | call_ext Ar Func
-move S r | line Loc | call_ext_last Ar Func=u$is_bif D => \
-     line Loc | move S r | call_ext_last Ar Func D
-move S r | line Loc | call_ext_only Ar Func=u$is_bif => \
-     line Loc | move S r | call_ext_only Ar Func
-move S r | line Loc | call Ar Func => \
-     line Loc | move S r | call Ar Func
+move S X0=x==0 | line Loc | call_ext Ar Func => \
+     line Loc | move S X0 | call_ext Ar Func
+move S X0=x==0 | line Loc | call_ext_last Ar Func=u$is_bif D => \
+     line Loc | move S X0 | call_ext_last Ar Func D
+move S X0=x==0 | line Loc | call_ext_only Ar Func=u$is_bif => \
+     line Loc | move S X0 | call_ext_only Ar Func
+move S X0=x==0 | line Loc | call Ar Func => \
+     line Loc | move S X0 | call Ar Func
 
 #
 # A tail-recursive call to an external function (non-BIF) will
@@ -167,31 +167,24 @@ is_tuple Fail=f S | select_tuple_arity S=d 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_bins r f I
 i_select_val_bins x f I
 i_select_val_bins y f I
 
-i_select_val_lins r f I
 i_select_val_lins x f I
 i_select_val_lins y f I
 
-i_select_val2 r f c c f f
 i_select_val2 x f c c f f
 i_select_val2 y f c c f f
 
-i_select_tuple_arity r f I
 i_select_tuple_arity x f I
 i_select_tuple_arity y f I
 
-i_select_tuple_arity2 r f A A f f
 i_select_tuple_arity2 x f A A f f
 i_select_tuple_arity2 y f A A f f
 
-i_jump_on_val_zero r f I
 i_jump_on_val_zero x f I
 i_jump_on_val_zero y f I
 
-i_jump_on_val r f I I
 i_jump_on_val x f I I
 i_jump_on_val y f I I
 
@@ -203,30 +196,13 @@ is_ne_exact L1 S1 S2 | jump Fail | label L2 | same_label(L1, L2) => \
 %macro: get_list GetList -pack
 get_list x x x
 get_list x x y
-get_list x x r
 get_list x y x
 get_list x y y
-get_list x y r
-get_list x r x
-get_list x r y
 
 get_list y x x
 get_list y x y
-get_list y x r
 get_list y y x
 get_list y y y
-get_list y y r
-get_list y r x
-get_list y r y
-
-get_list r x x
-get_list r x y
-get_list r x r
-get_list r y x
-get_list r y y
-get_list r y r
-get_list r r x
-get_list r r y
 
 # Old-style catch.
 catch y f
@@ -247,21 +223,15 @@ set_tuple_element s d P
 
 %macro: i_get_tuple_element GetTupleElement -pack
 i_get_tuple_element x P x
-i_get_tuple_element r P x
 i_get_tuple_element y P x
-i_get_tuple_element x P r
-i_get_tuple_element y P r
 
 %cold
-i_get_tuple_element r P r
 i_get_tuple_element x P y
-i_get_tuple_element r P y
 i_get_tuple_element y P y
 %hot
 
 %macro: is_number IsNumber -fail_action
 %cold
-is_number f r
 is_number f x
 is_number f y
 %hot
@@ -271,16 +241,12 @@ is_number Fail Literal=q => move Literal x | is_number Fail x
 
 jump f
 
-case_end Literal=c => move Literal x | case_end x
-badmatch Literal=c => move Literal x | badmatch x
+case_end NotInX=cy => move NotInX x | case_end x
+badmatch NotInX=cy => move NotInX x | badmatch x
 
-case_end r
 case_end x
-case_end y
 
-badmatch r
 badmatch x
-badmatch y
 
 if_end
 raise s s
@@ -289,7 +255,7 @@ raise s s
 badarg j
 system_limit j
 
-move C=cxy r | jump Lbl => move_jump Lbl C
+move C=cxy x==0 | jump Lbl => move_jump Lbl C
 
 %macro: move_jump MoveJump -nonext
 move_jump f n
@@ -307,8 +273,6 @@ 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
 
@@ -329,12 +293,13 @@ 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 S3=x S4=x => move2 S1 S2 S3 S4
-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
+move X1=x X2=x | move X3=x Y1=y => move2 X1 X2 X3 Y1
+
+move S1=x S2=x | move X1=x Y1=y => move2 S1 S2 X1 Y1
+move S1=y S2=x | move X1=x Y1=y => move2 S1 S2 X1 Y1
 
-move Y1=y X1=x | move S1=r D1=x => move2 Y1 X1 S1 D1
-move S1=r D1=x | move Y1=y X1=x => move2 S1 D1 Y1 X1
+move Y1=y X1=x | move S1=x D1=x => move2 Y1 X1 S1 D1
+move S1=x D1=x | move Y1=y X1=x => move2 S1 D1 Y1 X1
 
 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
@@ -351,14 +316,13 @@ move2 x y x y
 move2 y x y x
 move2 x x x x
 
-move2 x r x x
+move2 x x x y
 
-move2 x r x y
-move2 r y x y
-move2 y r x y
+move2 x y x y
+move2 y x x y
 
-move2 r x y x
-move2 y x r x
+move2 x x y x
+move2 y x x x
 
 %macro: move3 Move3
 move3 x y x y x y
@@ -373,20 +337,14 @@ move S=c D=y => move S x | move x D
 %macro:move Move -pack -gen_dest
 move x x
 move x y
-move x r
 move y x
-move y r
-move r x
-move r y
-move c r
 move c x
 move n x
-move n r
 move y y
 
 # Receive operations.
 
-loop_rec Fail Src | smp_mark_target_label(Fail) => i_loop_rec Fail Src
+loop_rec Fail x==0 | smp_mark_target_label(Fail) => i_loop_rec Fail
 
 label L | wait_timeout Fail Src | smp_already_locked(L) => label L | i_wait_timeout_locked Fail Src
 wait_timeout Fail Src => i_wait_timeout Fail Src
@@ -401,7 +359,7 @@ label L | timeout | smp_already_locked(L) => label L | timeout_locked
 remove_message
 timeout
 timeout_locked
-i_loop_rec f r
+i_loop_rec f
 loop_rec_end f
 wait f
 wait_locked f
@@ -419,29 +377,25 @@ send
 # Optimized comparisons with one immediate/literal operand.
 #
 
-is_eq_exact Lbl R=rxy C=ian => i_is_eq_exact_immed Lbl R C
-is_eq_exact Lbl R=rxy C=q => i_is_eq_exact_literal R Lbl C
+is_eq_exact Lbl R=xy C=ian => i_is_eq_exact_immed Lbl R C
+is_eq_exact Lbl R=xy C=q => i_is_eq_exact_literal Lbl R C
 
-is_ne_exact Lbl R=rxy C=ian => i_is_ne_exact_immed Lbl R C
-is_ne_exact Lbl R=rxy C=q => i_is_ne_exact_literal R Lbl C
+is_ne_exact Lbl R=xy C=ian => i_is_ne_exact_immed Lbl R C
+is_ne_exact Lbl R=xy C=q => i_is_ne_exact_literal Lbl R C
 
 %macro: i_is_eq_exact_immed EqualImmed -fail_action
-i_is_eq_exact_immed f r c
 i_is_eq_exact_immed f x c
 i_is_eq_exact_immed f y c
 
-i_is_eq_exact_literal r f c
-i_is_eq_exact_literal x f c
-i_is_eq_exact_literal y f c
+i_is_eq_exact_literal f x c
+i_is_eq_exact_literal f y c
 
 %macro: i_is_ne_exact_immed NotEqualImmed -fail_action
-i_is_ne_exact_immed f r c
 i_is_ne_exact_immed f x c
 i_is_ne_exact_immed f y c
 
-i_is_ne_exact_literal r f c
-i_is_ne_exact_literal x f c
-i_is_ne_exact_literal y f c
+i_is_ne_exact_literal f x c
+i_is_ne_exact_literal f y c
 
 #
 # Common Compare Specializations
@@ -449,31 +403,21 @@ i_is_ne_exact_literal y f c
 # to keep the instruction set small-ish
 #
 
-is_eq_exact Lbl S1=xy S2=r => is_eq_exact Lbl S2 S1
-is_eq_exact Lbl S1=rx S2=xy => i_is_eq_exact_spec Lbl S1 S2
+is_eq_exact Lbl S1=y S2=x => is_eq_exact Lbl S2 S1
+is_eq_exact Lbl S1=x S2=xy => i_is_eq_exact_spec Lbl S1 S2
 %macro: i_is_eq_exact_spec EqualExact -fail_action
 
 i_is_eq_exact_spec f x x
 i_is_eq_exact_spec f x y
-i_is_eq_exact_spec f r x
-i_is_eq_exact_spec f r y
-%cold
-i_is_eq_exact_spec f r r
-%hot
 
-is_lt Lbl S1=rxc S2=rxc  => i_is_lt_spec Lbl S1 S2
+is_lt Lbl S1=xc S2=xc  => i_is_lt_spec Lbl S1 S2
 
 %macro: i_is_lt_spec IsLessThan -fail_action
 
 i_is_lt_spec f x x
-i_is_lt_spec f x r
 i_is_lt_spec f x c
-i_is_lt_spec f r x
-i_is_lt_spec f r c
 i_is_lt_spec f c x
-i_is_lt_spec f c r
 %cold
-i_is_lt_spec f r r
 i_is_lt_spec f c c
 %hot
 
@@ -523,7 +467,6 @@ i_put_tuple Dst Arity Puts=* | put S => \
 i_put_tuple/2
 
 %macro:i_put_tuple PutTuple -pack -goto:do_put_tuple
-i_put_tuple r I
 i_put_tuple x I
 i_put_tuple y I
 
@@ -540,48 +483,25 @@ put_list x n x
 put_list y n x
 put_list x x x
 put_list y x x
-put_list x x r
-put_list y r r
 
 put_list y y x
 put_list x y x
-put_list r x x
-put_list r y x
-put_list r x r
-put_list y y r
-put_list y r x
-put_list r n x
-
-put_list x r x
-put_list x y r
-put_list y x r
-put_list y x x
 
-put_list x r r
+put_list y x x
 
 # put_list SrcReg Constant Dst
-put_list r c r
-put_list r c x
-put_list r c y
 
-put_list x c r
 put_list x c x
 put_list x c y
 
-put_list y c r
 put_list y c x
 put_list y c y
 
 # put_list Constant SrcReg Dst
-put_list c r r
-put_list c r x
-put_list c r y
 
-put_list c x r
 put_list c x x
 put_list c x y
 
-put_list c y r
 put_list c y x
 put_list c y y
 
@@ -590,18 +510,12 @@ put_list s s d
 %hot
 
 %macro: i_fetch FetchArgs -pack
-i_fetch c r
 i_fetch c x
 i_fetch c y
-i_fetch r c
-i_fetch r x
-i_fetch r y
 i_fetch x c
-i_fetch x r
 i_fetch x x
 i_fetch x y
 i_fetch y c
-i_fetch y r
 i_fetch y x
 i_fetch y y
 
@@ -629,27 +543,27 @@ return_trace
 # Note: There is no 'move_return y r', since there never are any y registers
 # when we do move_return (if we have y registers, we must do move_deallocate_return).
 
-move S r | return => move_return S r
+move S x==0 | return => move_return S
 
 %macro: move_return MoveReturn -nonext
-move_return x r
-move_return c r
-move_return n r
+move_return x
+move_return c
+move_return n
 
-move S r | deallocate D | return => move_deallocate_return S r D
+move S x==0 | deallocate D | return => move_deallocate_return S D
 
 %macro: move_deallocate_return MoveDeallocateReturn -pack -nonext
-move_deallocate_return x r Q
-move_deallocate_return y r Q
-move_deallocate_return c r Q
-move_deallocate_return n r Q
+move_deallocate_return x Q
+move_deallocate_return y Q
+move_deallocate_return c Q
+move_deallocate_return n Q
 
 deallocate D | return => deallocate_return D
 
 %macro: deallocate_return DeallocateReturn -nonext
 deallocate_return Q
 
-test_heap Need u==1 | put_list Y=y r r => test_heap_1_put_list Need Y
+test_heap Need u==1 | put_list Y=y x==0 x==0 => test_heap_1_put_list Need Y
 
 %macro: test_heap_1_put_list TestHeapPutList -pack
 test_heap_1_put_list I y
@@ -658,18 +572,16 @@ test_heap_1_put_list I y
 
 is_tuple Fail Literal=q => move Literal x | is_tuple Fail x
 is_tuple Fail=f c => jump Fail
-is_tuple Fail=f S=rxy | test_arity Fail=f S=rxy Arity => is_tuple_of_arity Fail S Arity
+is_tuple Fail=f S=xy | test_arity Fail=f S=xy Arity => is_tuple_of_arity Fail S Arity
 
 %macro:is_tuple_of_arity IsTupleOfArity -fail_action
 
 is_tuple_of_arity f x A
 is_tuple_of_arity f y A
-is_tuple_of_arity f r A
 
 %macro: is_tuple IsTuple -fail_action
 is_tuple f x
 is_tuple f y
-is_tuple f r
 
 test_arity Fail Literal=q Arity => move Literal x | test_arity Fail x Arity
 test_arity Fail=f c Arity => jump Fail
@@ -677,7 +589,6 @@ test_arity Fail=f c Arity => jump Fail
 %macro: test_arity IsArity -fail_action
 test_arity f x A
 test_arity f y A
-test_arity f r A
 
 is_tuple_of_arity Fail=f Reg Arity | get_tuple_element Reg P=u==0 Dst=xy => \
   is_tuple_of_arity Fail Reg Arity | extract_next_element Dst | original_reg Reg P
@@ -726,46 +637,40 @@ is_integer Fail=f i =>
 is_integer Fail=f an => jump Fail
 is_integer Fail Literal=q => move Literal x | is_integer Fail x
 
-is_integer Fail=f S=rx | allocate Need Regs => is_integer_allocate Fail S Need Regs
+is_integer Fail=f S=x | allocate Need Regs => is_integer_allocate Fail S Need Regs
 
 %macro: is_integer_allocate IsIntegerAllocate -fail_action
 is_integer_allocate f x I I
-is_integer_allocate f r I I
 
 %macro: is_integer IsInteger -fail_action
 is_integer f x
 is_integer f y
-is_integer f r
 
 is_list Fail=f n =>
 is_list Fail Literal=q => move Literal x | is_list Fail x
 is_list Fail=f c => jump Fail
 %macro: is_list IsList -fail_action
-is_list f r
 is_list f x
 %cold
 is_list f y
 %hot
 
-is_nonempty_list Fail=f S=rx | allocate Need Rs => is_nonempty_list_allocate Fail S Need Rs
+is_nonempty_list Fail=f S=x | allocate Need Rs => is_nonempty_list_allocate Fail S Need Rs
 
 %macro:is_nonempty_list_allocate IsNonemptyListAllocate -fail_action -pack
 is_nonempty_list_allocate f x I t
-is_nonempty_list_allocate f r I t
 
-is_nonempty_list F=f r | test_heap I1 I2 => is_non_empty_list_test_heap F r I1 I2
+is_nonempty_list F=f x==0 | test_heap I1 I2 => is_non_empty_list_test_heap F I1 I2
 
 %macro: is_non_empty_list_test_heap IsNonemptyListTestHeap -fail_action -pack
-is_non_empty_list_test_heap f r I t
+is_non_empty_list_test_heap f I t
 
 %macro: is_nonempty_list IsNonemptyList -fail_action
 is_nonempty_list f x
 is_nonempty_list f y
-is_nonempty_list f r
 
 %macro: is_atom IsAtom -fail_action
 is_atom f x
-is_atom f r
 %cold
 is_atom f y
 %hot
@@ -773,7 +678,6 @@ is_atom Fail=f a =>
 is_atom Fail=f niq => jump Fail
 
 %macro: is_float IsFloat -fail_action
-is_float f r
 is_float f x
 %cold
 is_float f y
@@ -787,12 +691,10 @@ is_nil Fail=f qia => jump Fail
 %macro: is_nil IsNil -fail_action
 is_nil f x
 is_nil f y
-is_nil f r
 
 is_binary Fail Literal=q => move Literal x | is_binary Fail x
 is_binary Fail=f c => jump Fail
 %macro: is_binary IsBinary -fail_action
-is_binary f r
 is_binary f x
 %cold
 is_binary f y
@@ -804,7 +706,6 @@ is_bitstr Fail Term => is_bitstring Fail Term
 is_bitstring Fail Literal=q => move Literal x | is_bitstring Fail x
 is_bitstring Fail=f c => jump Fail
 %macro: is_bitstring IsBitstring -fail_action
-is_bitstring f r
 is_bitstring f x
 %cold
 is_bitstring f y
@@ -812,7 +713,6 @@ is_bitstring f y
 
 is_reference Fail=f cq => jump Fail
 %macro: is_reference IsRef -fail_action
-is_reference f r
 is_reference f x
 %cold
 is_reference f y
@@ -820,7 +720,6 @@ is_reference f y
 
 is_pid Fail=f cq => jump Fail
 %macro: is_pid IsPid -fail_action
-is_pid f r
 is_pid f x
 %cold
 is_pid f y
@@ -828,7 +727,6 @@ is_pid f y
 
 is_port Fail=f cq => jump Fail
 %macro: is_port IsPort -fail_action
-is_port f r
 is_port f x
 %cold
 is_port f y
@@ -840,7 +738,6 @@ is_boolean Fail=f ac => jump Fail
 
 %cold
 %macro: is_boolean IsBoolean -fail_action
-is_boolean f r
 is_boolean f x
 is_boolean f y
 %hot
@@ -986,76 +883,76 @@ call_ext_only u==3 u$func:erlang:hibernate/3 => i_hibernate
 %unless USE_VM_PROBES
 
 call_ext Arity u$func:erlang:dt_get_tag/0 => \
-    move a=am_undefined r
+    move a=am_undefined x=0
 call_ext_last Arity u$func:erlang:dt_get_tag/0 D => \
-    move a=am_undefined r | deallocate D | return
+    move a=am_undefined x=0 | deallocate D | return
 call_ext_only Arity u$func:erlang:dt_get_tag/0 => \
-    move a=am_undefined r | return
-
-move Any r | call_ext Arity u$func:erlang:dt_put_tag/1 => \
-    move a=am_undefined r
-move Any r | call_ext_last Arity u$func:erlang:dt_put_tag/1 D => \
-    move a=am_undefined r | deallocate D | return
-move Any r | call_ext_only Arity u$func:erlang:dt_put_tag/1 => \
-    move a=am_undefined r | return
+    move a=am_undefined x=0 | return
+
+move Any x==0 | call_ext Arity u$func:erlang:dt_put_tag/1 => \
+    move a=am_undefined x=0
+move Any x==0 | call_ext_last Arity u$func:erlang:dt_put_tag/1 D => \
+    move a=am_undefined x=0 | deallocate D | return
+move Any x==0 | call_ext_only Arity u$func:erlang:dt_put_tag/1 => \
+    move a=am_undefined x=0 | return
 call_ext Arity u$func:erlang:dt_put_tag/1 => \
-    move a=am_undefined r
+    move a=am_undefined x=0
 call_ext_last Arity u$func:erlang:dt_put_tag/1 D => \
-    move a=am_undefined r | deallocate D | return
+    move a=am_undefined x=0 | deallocate D | return
 call_ext_only Arity u$func:erlang:dt_put_tag/1 => \
-    move a=am_undefined r | return
+    move a=am_undefined x=0 | return
 
 call_ext Arity u$func:erlang:dt_get_tag_data/0 => \
-    move a=am_undefined r
+    move a=am_undefined x=0
 call_ext_last Arity u$func:erlang:dt_get_tag_data/0 D => \
-    move a=am_undefined r | deallocate D | return
+    move a=am_undefined x=0 | deallocate D | return
 call_ext_only Arity u$func:erlang:dt_get_tag_data/0 => \
-    move a=am_undefined r | return
-
-move Any r | call_ext Arity u$func:erlang:dt_spread_tag/1 => \
-    move a=am_true r
-move Any r | call_ext_last Arity u$func:erlang:dt_spread_tag/1 D => \
-    move a=am_true r | deallocate D | return
-move Any r | call_ext_only Arity u$func:erlang:dt_spread_tag/1 => \
-    move a=am_true r | return
+    move a=am_undefined x=0 | return
+
+move Any x==0 | call_ext Arity u$func:erlang:dt_spread_tag/1 => \
+    move a=am_true x=0
+move Any x==0 | call_ext_last Arity u$func:erlang:dt_spread_tag/1 D => \
+    move a=am_true x=0 | deallocate D | return
+move Any x==0 | call_ext_only Arity u$func:erlang:dt_spread_tag/1 => \
+    move a=am_true x=0 | return
 call_ext Arity u$func:erlang:dt_spread_tag/1 => \
-    move a=am_true r
+    move a=am_true x=0
 call_ext_last Arity u$func:erlang:dt_spread_tag/1 D => \
-    move a=am_true r | deallocate D | return
+    move a=am_true x=0 | deallocate D | return
 call_ext_only Arity u$func:erlang:dt_spread_tag/1 => \
-    move a=am_true r | return
-
-move Any r | call_ext Arity u$func:erlang:dt_restore_tag/1 => \
-    move a=am_true r
-move Any r | call_ext_last Arity u$func:erlang:dt_restore_tag/1 D => \
-    move a=am_true r | deallocate D | return
-move Any r | call_ext_only Arity u$func:erlang:dt_restore_tag/1 => \
-    move a=am_true r | return
+    move a=am_true x=0 | return
+
+move Any x==0 | call_ext Arity u$func:erlang:dt_restore_tag/1 => \
+    move a=am_true x=0
+move Any x==0 | call_ext_last Arity u$func:erlang:dt_restore_tag/1 D => \
+    move a=am_true x=0 | deallocate D | return
+move Any x==0 | call_ext_only Arity u$func:erlang:dt_restore_tag/1 => \
+    move a=am_true x=0 | return
 call_ext Arity u$func:erlang:dt_restore_tag/1 => \
-    move a=am_true r
+    move a=am_true x=0
 call_ext_last Arity u$func:erlang:dt_restore_tag/1 D => \
-    move a=am_true r | deallocate D | return
+    move a=am_true x=0 | deallocate D | return
 call_ext_only Arity u$func:erlang:dt_restore_tag/1 => \
-    move a=am_true r | return
-
-move Any r | call_ext Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
-    move Any r
-move Any r | call_ext_last Arity u$func:erlang:dt_prepend_vm_tag_data/1 D => \
-    move Any r | deallocate D | return
-move Any r | call_ext_only Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
-    move Any r | return
+    move a=am_true x=0 | return
+
+move Any x==0 | call_ext Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
+    move Any x=0
+move Any x==0 | call_ext_last Arity u$func:erlang:dt_prepend_vm_tag_data/1 D => \
+    move Any x=0 | deallocate D | return
+move Any x==0 | call_ext_only Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
+    move Any x=0 | return
 call_ext Arity u$func:erlang:dt_prepend_vm_tag_data/1 =>
 call_ext_last Arity u$func:erlang:dt_prepend_vm_tag_data/1 D => \
     deallocate D | return
 call_ext_only Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
     return
 
-move Any r | call_ext Arity u$func:erlang:dt_append_vm_tag_data/1 => \
-    move Any r
-move Any r | call_ext_last Arity u$func:erlang:dt_append_vm_tag_data/1 D => \
-    move Any r | deallocate D | return
-move Any r | call_ext_only Arity u$func:erlang:dt_append_vm_tag_data/1 => \
-    move Any r | return
+move Any x==0 | call_ext Arity u$func:erlang:dt_append_vm_tag_data/1 => \
+    move Any x=0
+move Any x==0 | call_ext_last Arity u$func:erlang:dt_append_vm_tag_data/1 D => \
+    move Any x=0 | deallocate D | return
+move Any x==0 | call_ext_only Arity u$func:erlang:dt_append_vm_tag_data/1 => \
+    move Any x=0 | return
 call_ext Arity u$func:erlang:dt_append_vm_tag_data/1 =>
 call_ext_last Arity u$func:erlang:dt_append_vm_tag_data/1 D => \
     deallocate D | return
@@ -1063,7 +960,7 @@ call_ext_only Arity u$func:erlang:dt_append_vm_tag_data/1 => \
     return
 
 # Can happen after one of the transformations above.
-move Discarded r | move Something r => move Something r
+move Discarded x==0 | move Something x==0 => move Something x=0
 
 %endif
 
@@ -1088,9 +985,9 @@ call_ext_only Ar=u Bif=u$is_bif => \
 # with call instructions.
 #
 
-move S=c r | call_ext Ar=u Func=u$is_not_bif => i_move_call_ext S r Func
-move S=c r | call_ext_last Ar=u Func=u$is_not_bif D => i_move_call_ext_last Func D S r
-move S=c r | call_ext_only Ar=u Func=u$is_not_bif => i_move_call_ext_only Func S r 
+move S=c x==0 | call_ext Ar=u Func=u$is_not_bif => i_move_call_ext S Func
+move S=c x==0 | call_ext_last Ar=u Func=u$is_not_bif D => i_move_call_ext_last Func D S
+move S=c x==0 | call_ext_only Ar=u Func=u$is_not_bif => i_move_call_ext_only Func S
 
 call_ext Ar Func        => i_call_ext Func
 call_ext_last Ar Func D => i_call_ext_last Func D
@@ -1117,7 +1014,7 @@ bif0 u$bif:erlang:node/0 Dst=d => node Dst
 
 bif1 Fail Bif=u$bif:erlang:get/1 Src=s Dst=d => i_get Src Dst
 
-bif2 Jump=j u$bif:erlang:element/2 S1=s S2=rxy Dst=d => gen_element(Jump, S1, S2, Dst)
+bif2 Jump=j u$bif:erlang:element/2 S1=s S2=xy Dst=d => gen_element(Jump, S1, S2, Dst)
 
 bif1 p Bif S1 Dst => bif1_body Bif S1 Dst
 
@@ -1127,24 +1024,20 @@ bif2 Fail Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2 Fail Bif Dst
 i_get s d
 
 %macro: self Self
-self r
 self x
 self y
 
 %macro: node Node
-node r
 node x
 %cold
 node y
 %hot
 
-i_fast_element r j I d
-i_fast_element x j I d
-i_fast_element y j I d
+i_fast_element j x I d
+i_fast_element j y I d
 
-i_element r j s d
-i_element x j s d
-i_element y j s d
+i_element j x s d
+i_element j y s d
 
 bif1 f b s d
 bif1_body b s d
@@ -1155,37 +1048,37 @@ i_bif2_body b d
 # Internal calls.
 #
 
-move S=c r | call Ar P=f => i_move_call S r P
-move S=s r | call Ar P=f => move_call S r P
+move S=c x==0 | call Ar P=f => i_move_call S P
+move S=s x==0 | call Ar P=f => move_call S P
 
-i_move_call c r f
+i_move_call c f
 
 %macro:move_call MoveCall -arg_f -size -nonext
-move_call/3
+move_call/2
 
-move_call x r f
-move_call y r f
+move_call x f
+move_call y f
 
-move S=c r | call_last Ar P=f D => i_move_call_last P D S r
-move S r | call_last Ar P=f D => move_call_last S r P D
+move S=c x==0 | call_last Ar P=f D => i_move_call_last P D S
+move S x==0 | call_last Ar P=f D => move_call_last S P D
 
-i_move_call_last f P c r
+i_move_call_last f P c
 
 %macro:move_call_last MoveCallLast -arg_f -nonext -pack
 
-move_call_last/4
-move_call_last x r f Q
-move_call_last y r f Q
+move_call_last/3
+move_call_last x f Q
+move_call_last y f Q
 
-move S=c r | call_only Ar P=f => i_move_call_only P S r
-move S=x r | call_only Ar P=f => move_call_only S r P
+move S=c x==0 | call_only Ar P=f => i_move_call_only P S
+move S=x x==0 | call_only Ar P=f => move_call_only S P
 
-i_move_call_only f c r
+i_move_call_only f c
 
 %macro:move_call_only MoveCallOnly -arg_f -nonext
-move_call_only/3
+move_call_only/2
 
-move_call_only x r f
+move_call_only x f
 
 call Ar Func        => i_call Func
 call_last Ar Func D => i_call_last Func D
@@ -1199,9 +1092,9 @@ i_call_ext e
 i_call_ext_last e P
 i_call_ext_only e
 
-i_move_call_ext c r e
-i_move_call_ext_last e P c r
-i_move_call_ext_only e c r
+i_move_call_ext c e
+i_move_call_ext_last e P c
+i_move_call_ext_only e c
 
 # Fun calls.
 
@@ -1221,7 +1114,6 @@ i_make_fun I t
 %macro: is_function IsFunction -fail_action
 is_function f x
 is_function f y
-is_function f r
 is_function Fail=f c => jump Fail
 
 func_info M F A => i_func_info u M F A
@@ -1233,126 +1125,102 @@ func_info M F A => i_func_info u M F A
 %cold
 bs_start_match2 Fail=f ica X Y D => jump Fail
 bs_start_match2 Fail Bin X Y D => i_bs_start_match2 Bin Fail X Y D
-i_bs_start_match2 r f I I d
 i_bs_start_match2 x f I I d
 i_bs_start_match2 y f I I d
 
 bs_save2 Reg Index => gen_bs_save(Reg, Index)
-i_bs_save2 r I
 i_bs_save2 x I
 
 bs_restore2 Reg Index => gen_bs_restore(Reg, Index)
-i_bs_restore2 r I
 i_bs_restore2 x I
 
 # Matching integers
 bs_match_string Fail Ms Bits Val => i_bs_match_string Ms Fail Bits Val
 
-i_bs_match_string r f I I
 i_bs_match_string x f I I
 
 # Fetching integers from binaries.
-bs_get_integer2 Fail=f Ms=rx Live=u Sz=sq Unit=u Flags=u Dst=d => \
+bs_get_integer2 Fail=f Ms=x Live=u Sz=sq Unit=u Flags=u Dst=d => \
 			gen_get_integer2(Fail, Ms, Live, Sz, Unit, Flags, Dst)
 
-i_bs_get_integer_small_imm r I f I d
 i_bs_get_integer_small_imm x I f I d
-i_bs_get_integer_imm r I I f I d
 i_bs_get_integer_imm x I I f I d
 i_bs_get_integer f I I d
-i_bs_get_integer_8 r f d
 i_bs_get_integer_8 x f d
-i_bs_get_integer_16 r f d
 i_bs_get_integer_16 x f d
-i_bs_get_integer_32 r f I d
 i_bs_get_integer_32 x f I d
 
 # Fetching binaries from binaries.
-bs_get_binary2 Fail=f Ms=rx Live=u Sz=sq Unit=u Flags=u Dst=d => \
+bs_get_binary2 Fail=f Ms=x Live=u Sz=sq Unit=u Flags=u Dst=d => \
 			gen_get_binary2(Fail, Ms, Live, Sz, Unit, Flags, Dst)
 
 %macro: i_bs_get_binary_imm2 BsGetBinaryImm_2 -fail_action -gen_dest
 %macro: i_bs_get_binary2 BsGetBinary_2 -fail_action -gen_dest
 %macro: i_bs_get_binary_all2 BsGetBinaryAll_2 -fail_action -gen_dest
 
-i_bs_get_binary_imm2 f r I I I d
 i_bs_get_binary_imm2 f x I I I d
-i_bs_get_binary2 f r I s I d
 i_bs_get_binary2 f x I s I d
-i_bs_get_binary_all2 f r I I d
 i_bs_get_binary_all2 f x I I d
-i_bs_get_binary_all_reuse r f I
 i_bs_get_binary_all_reuse x f I
 
 # Fetching float from binaries.
-bs_get_float2 Fail=f Ms=rx Live=u Sz=s Unit=u Flags=u Dst=d => \
+bs_get_float2 Fail=f Ms=x Live=u Sz=s Unit=u Flags=u Dst=d => \
 		gen_get_float2(Fail, Ms, Live, Sz, Unit, Flags, Dst)
 
-bs_get_float2 Fail=f Ms=rx Live=u Sz=q Unit=u Flags=u Dst=d => jump Fail
+bs_get_float2 Fail=f Ms=x Live=u Sz=q Unit=u Flags=u Dst=d => jump Fail
 
 %macro: i_bs_get_float2 BsGetFloat2 -fail_action -gen_dest
-i_bs_get_float2 f r I s I d
 i_bs_get_float2 f x I s I d
 
 # Miscellanous
 
-bs_skip_bits2 Fail=f Ms=rx Sz=s Unit=u Flags=u => \
-			gen_skip_bits2(Fail, Ms, Sz, Unit, Flags)
-bs_skip_bits2 Fail=f Ms=rx Sz=q Unit=u Flags=u => \
+bs_skip_bits2 Fail=f Ms=x Sz=sq Unit=u Flags=u => \
 			gen_skip_bits2(Fail, Ms, Sz, Unit, Flags)
 
 %macro: i_bs_skip_bits_imm2 BsSkipBitsImm2 -fail_action
-i_bs_skip_bits_imm2 f r I
 i_bs_skip_bits_imm2 f x I
 
 %macro: i_bs_skip_bits2 BsSkipBits2 -fail_action
-i_bs_skip_bits2 f r x I
-i_bs_skip_bits2 f r y I
 i_bs_skip_bits2 f x x I
-i_bs_skip_bits2 f x r I
 i_bs_skip_bits2 f x y I
 
 %macro: i_bs_skip_bits_all2 BsSkipBitsAll2 -fail_action
-i_bs_skip_bits_all2 f r I
 i_bs_skip_bits_all2 f x I 
 
-bs_test_tail2 Fail=f Ms=rx Bits=u==0 => bs_test_zero_tail2 Fail Ms
-bs_test_tail2 Fail=f Ms=rx Bits=u => bs_test_tail_imm2 Fail Ms Bits
-bs_test_zero_tail2 f r
+bs_test_tail2 Fail=f Ms=x Bits=u==0 => bs_test_zero_tail2 Fail Ms
+bs_test_tail2 Fail=f Ms=x Bits=u => bs_test_tail_imm2 Fail Ms Bits
 bs_test_zero_tail2 f x
-bs_test_tail_imm2 f r I
 bs_test_tail_imm2 f x I
 
 bs_test_unit F Ms Unit=u==8 => bs_test_unit8 F Ms
-bs_test_unit f r I
 bs_test_unit f x I
-bs_test_unit8 f r
 bs_test_unit8 f x
 
-bs_context_to_binary r
+# An y register operand for bs_context_to_binary is rare,
+# but can happen because of inlining.
+
+bs_context_to_binary Y=y => move Y x | bs_context_to_binary x
+
 bs_context_to_binary x
-bs_context_to_binary y
 
 #
 # Utf8/utf16/utf32 support. (R12B-5)
 #
-bs_get_utf8 Fail=f Ms=rx u u Dst=d => i_bs_get_utf8 Ms Fail Dst
-i_bs_get_utf8 r f d
+bs_get_utf8 Fail=f Ms=x u u Dst=d => i_bs_get_utf8 Ms Fail Dst
 i_bs_get_utf8 x f d
 
-bs_skip_utf8 Fail=f Ms=rx u u => i_bs_get_utf8 Ms Fail x
+bs_skip_utf8 Fail=f Ms=x u u => i_bs_get_utf8 Ms Fail x
 
-bs_get_utf16 Fail=f Ms=rx u Flags=u Dst=d => i_bs_get_utf16 Ms Fail Flags Dst
-bs_skip_utf16 Fail=f Ms=rx u Flags=u => i_bs_get_utf16 Ms Fail Flags x
+bs_get_utf16 Fail=f Ms=x u Flags=u Dst=d => i_bs_get_utf16 Ms Fail Flags Dst
+bs_skip_utf16 Fail=f Ms=x u Flags=u => i_bs_get_utf16 Ms Fail Flags x
 
-i_bs_get_utf16 r f I d
 i_bs_get_utf16 x f I d
 
-bs_get_utf32 Fail=f Ms=rx Live=u Flags=u Dst=d => \
+bs_get_utf32 Fail=f Ms=x Live=u Flags=u Dst=d => \
 	bs_get_integer2 Fail Ms Live i=32 u=1 Flags Dst | \
 	i_fetch Dst Ms | \
 	i_bs_validate_unicode_retract Fail
-bs_skip_utf32 Fail=f Ms=rx Live=u Flags=u => \
+bs_skip_utf32 Fail=f Ms=x Live=u Flags=u => \
 	bs_get_integer2 Fail Ms Live i=32 u=1 Flags x | \
 	i_fetch x Ms | \
 	i_bs_validate_unicode_retract Fail
@@ -1379,9 +1247,8 @@ bs_init2 Fail Sz=u Words Regs Flags Dst => \
 bs_init2 Fail Sz Words=u==0 Regs Flags Dst => \
   i_bs_init_fail Sz Fail Regs Dst
 bs_init2 Fail Sz Words Regs Flags Dst => \
-  i_fetch Sz r | i_bs_init_fail_heap Words Fail Regs Dst
+  i_fetch Sz x=0 | i_bs_init_fail_heap Words Fail Regs Dst
 
-i_bs_init_fail r j I d
 i_bs_init_fail x j I d
 i_bs_init_fail y j I d
 
@@ -1402,9 +1269,8 @@ bs_init_bits Fail Sz=u Words Regs Flags Dst =>  i_bs_init_bits_heap Sz Words Reg
 bs_init_bits Fail Sz Words=u==0 Regs Flags Dst => \
   i_bs_init_bits_fail Sz Fail Regs Dst
 bs_init_bits Fail Sz Words Regs Flags Dst => \
-  i_fetch Sz r | i_bs_init_bits_fail_heap Words Fail Regs Dst
+  i_fetch Sz x=0 | i_bs_init_bits_fail_heap Words Fail Regs Dst
 
-i_bs_init_bits_fail r j I d
 i_bs_init_bits_fail x j I d
 i_bs_init_bits_fail y j I d
 
@@ -1577,7 +1443,6 @@ is_map Fail Lit=q | literal_is_map(Lit) =>
 is_map Fail cq => jump Fail
 
 %macro: is_map IsMap -fail_action
-is_map f r
 is_map f x
 is_map f y
 
@@ -1588,35 +1453,25 @@ has_map_fields Fail Src Size Rest=* => \
 
 ## Transform get_map_elements(s) #{ K1 := V1, K2 := V2 }
 
-get_map_elements Fail Src=rxy Size=u==2 Rest=* => \
+get_map_elements Fail Src=xy Size=u==2 Rest=* => \
     gen_get_map_element(Fail, Src, Size, Rest)
 get_map_elements Fail Src Size Rest=* | map_key_sort(Size, Rest) => \
    gen_get_map_elements(Fail, Src, Size, Rest)
 
 i_get_map_elements f s I
 
-i_get_map_element Fail Src=rxy Key=ry Dst => \
+i_get_map_element Fail Src=xy Key=y Dst => \
     move Key x | i_get_map_element Fail Src x Dst
 
 %macro: i_get_map_element_hash GetMapElementHash -fail_action
-i_get_map_element_hash f r c I r
-i_get_map_element_hash f x c I r
-i_get_map_element_hash f y c I r
-i_get_map_element_hash f r c I x
 i_get_map_element_hash f x c I x
 i_get_map_element_hash f y c I x
-i_get_map_element_hash f r c I y
 i_get_map_element_hash f x c I y
 i_get_map_element_hash f y c I y
 
 %macro: i_get_map_element GetMapElement -fail_action
-i_get_map_element f r x r
-i_get_map_element f x x r
-i_get_map_element f y x r
-i_get_map_element f r x x
 i_get_map_element f x x x
 i_get_map_element f y x x
-i_get_map_element f r x y
 i_get_map_element f x x y
 i_get_map_element f y x y
 
@@ -1625,9 +1480,9 @@ i_get_map_element f y x y
 # the i_increment/4 instruction (in bodies, not in guards).
 #
 
-gc_bif2 p Live u$bif:erlang:splus/2 Int=i Reg=d Dst => \
+gc_bif2 p Live u$bif:erlang:splus/2 Int=i Reg=x Dst => \
 	gen_increment(Reg, Int, Live, Dst)
-gc_bif2 p Live u$bif:erlang:splus/2 Reg=d Int=i Dst => \
+gc_bif2 p Live u$bif:erlang:splus/2 Reg=x Int=i Dst => \
 	gen_increment(Reg, Int, Live, Dst)
 
 gc_bif2 p Live u$bif:erlang:sminus/2 Reg=d Int=i Dst | \
@@ -1662,7 +1517,6 @@ gc_bif1 Fail I u$bif:erlang:bnot/1 Src Dst=d => i_int_bnot Fail Src I Dst
 gc_bif1 Fail I u$bif:erlang:sminus/1 Src Dst=d => i_fetch i Src | i_minus Fail I Dst
 gc_bif1 Fail I u$bif:erlang:splus/1 Src Dst=d => i_fetch i Src | i_plus Fail I Dst
 
-i_increment r I I d
 i_increment x I I d
 i_increment y I I d
 
-- 
cgit v1.2.3


From c76b297463feee133adad510128d312536150392 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 16 Jun 2015 10:06:20 +0200
Subject: Eliminate the use of i_fetch in arithmetic instructions

The i_fetch instruction fetches two operands and places them
in the tmp_arg1 and tmp_arg2 variables. The next instruction
(such as i_plus) does not have to handle different types of
operands, but can get get them simply from the tmp_arg*
variables. Thus, i_fetch was introduced as a way to temper
a potentail combinatorial explosion.

Unfortunately, clang will generate terrible code because of
the tmp_arg1 and tmp_arg2 variables being live across multiple
instructions. Note that Clang has no way to predict the control
flow from one instruction to another. Clang must assume that
any instruction can jump to any other instruction. Somehow GCC
manages to cope with this situation much better.

Therefore, to improve the quality of the code generated by clang, we
must eliminate all uses of the tmp_arg1 and tmp_arg2 variables. This
commit eliminates the use of i_fetch in combination with the
arithmetic and logical instructions.

While we are touching the code for the bsr and bsl instructions,
also move the tmp_big[] array from top scope of process main into
the block that encloses the bsr and bsl instructions.
---
 erts/emulator/beam/beam_emu.c | 319 ++++++++++++++++++------------------------
 erts/emulator/beam/ops.tab    |  93 +++++++-----
 2 files changed, 200 insertions(+), 212 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index e5179830de..82cc030878 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -1145,7 +1145,6 @@ void process_main(void)
      */
     register Eterm tmp_arg1 REG_tmp_arg1 = NIL;
     register Eterm tmp_arg2 REG_tmp_arg2 = NIL;
-    Eterm tmp_big[2];
 
     /*
      * X registers and floating point registers are located in
@@ -1158,8 +1157,6 @@ void process_main(void)
      */
     int neg_o_reds = 0;
 
-    Eterm (*arith_func)(Process* p, Eterm* reg, Uint live);
-
 #ifdef ERTS_OPCODE_COUNTER_SUPPORT
     static void* counting_opcodes[] = { DEFINE_COUNTING_OPCODES };
 #else
@@ -1307,9 +1304,6 @@ void process_main(void)
 #endif
 #include "beam_hot.h"
 
-#define STORE_ARITH_RESULT(res) StoreBifResult(2, (res));
-#define ARITH_FUNC(name) erts_gc_##name
-
 	{
 	    Eterm increment_reg_val;
 	    Eterm increment_val;
@@ -1349,81 +1343,71 @@ void process_main(void)
 	    goto find_func_info;
 	}
 
-#define DO_BIG_ARITH(Func,Arg1,Arg2)     \
-    do {                                 \
-        Uint live = Arg(1);              \
-        SWAPOUT;                         \
-        reg[live] = (Arg1);              \
-        reg[live+1] = (Arg2);            \
-        result = (Func)(c_p, reg, live); \
-        SWAPIN;                          \
-        ERTS_HOLE_CHECK(c_p);            \
-        if (is_value(result)) {          \
-            StoreBifResult(4,result);    \
-        }                                \
-        goto lb_Cl_error;                \
-    } while(0)
+#define DO_OUTLINED_ARITH_2(name, Op1, Op2)	\
+ do {						\
+     Eterm result;				\
+     Uint live = Arg(1);			\
+						\
+     SWAPOUT;					\
+     reg[live] = Op1;				\
+     reg[live+1] = Op2;				\
+     result = erts_gc_##name(c_p, reg, live);	\
+     SWAPIN;					\
+     ERTS_HOLE_CHECK(c_p);			\
+     if (is_value(result)) {			\
+	 StoreBifResult(4, result);		\
+     }						\
+     goto lb_Cl_error;				\
+ } while (0)
 
- OpCase(i_plus_jIxxd):
  {
+     Eterm PlusOp1, PlusOp2;
      Eterm result;
 
-     if (is_both_small(xb(Arg(2)), xb(Arg(3)))) {
-	 Sint i = signed_val(xb(Arg(2))) + signed_val(xb(Arg(3)));
-	 ASSERT(MY_IS_SSMALL(i) == IS_SSMALL(i));
-	 if (MY_IS_SSMALL(i)) {
-	     result = make_small(i);
-             StoreBifResult(4, result);
-	 }
-     }
-     DO_BIG_ARITH(ARITH_FUNC(mixed_plus), xb(Arg(2)), xb(Arg(3)));
- }
+ OpCase(i_plus_jIxxd):
+     PlusOp1 = xb(Arg(2));
+     PlusOp2 = xb(Arg(3));
+     goto do_plus;
 
- OpCase(i_plus_jId):
- {
-     Eterm result;
+ OpCase(i_plus_jIssd):
+     GetArg2(2, PlusOp1, PlusOp2);
+     goto do_plus;
 
-     if (is_both_small(tmp_arg1, tmp_arg2)) {
-	 Sint i = signed_val(tmp_arg1) + signed_val(tmp_arg2);
+ do_plus:
+     if (is_both_small(PlusOp1, PlusOp2)) {
+	 Sint i = signed_val(PlusOp1) + signed_val(PlusOp2);
 	 ASSERT(MY_IS_SSMALL(i) == IS_SSMALL(i));
 	 if (MY_IS_SSMALL(i)) {
 	     result = make_small(i);
-	     STORE_ARITH_RESULT(result);
+             StoreBifResult(4, result);
 	 }
      }
-     arith_func = ARITH_FUNC(mixed_plus);
-     goto do_big_arith2;
+     DO_OUTLINED_ARITH_2(mixed_plus, PlusOp1, PlusOp2);
  }
 
- OpCase(i_minus_jIxxd):
  {
+     Eterm MinusOp1, MinusOp2;
      Eterm result;
 
-     if (is_both_small(xb(Arg(2)), xb(Arg(3)))) {
-	 Sint i = signed_val(xb(Arg(2))) - signed_val(xb(Arg(3)));
-	 ASSERT(MY_IS_SSMALL(i) == IS_SSMALL(i));
-	 if (MY_IS_SSMALL(i)) {
-	     result = make_small(i);
-             StoreBifResult(4, result);
-	 }
-     }
-     DO_BIG_ARITH(ARITH_FUNC(mixed_minus), xb(Arg(2)), xb(Arg(3)));
- }
+ OpCase(i_minus_jIxxd):
+     MinusOp1 = xb(Arg(2));
+     MinusOp2 = xb(Arg(3));
+     goto do_minus;
 
- OpCase(i_minus_jId):
- {
-     Eterm result;
+ OpCase(i_minus_jIssd):
+     GetArg2(2, MinusOp1, MinusOp2);
+     goto do_minus;
 
-     if (is_both_small(tmp_arg1, tmp_arg2)) {
-	 Sint i = signed_val(tmp_arg1) - signed_val(tmp_arg2);
+ do_minus:
+     if (is_both_small(MinusOp1, MinusOp2)) {
+	 Sint i = signed_val(MinusOp1) - signed_val(MinusOp2);
 	 ASSERT(MY_IS_SSMALL(i) == IS_SSMALL(i));
 	 if (MY_IS_SSMALL(i)) {
 	     result = make_small(i);
-	     STORE_ARITH_RESULT(result);
+             StoreBifResult(4, result);
 	 }
      }
-     arith_func = ARITH_FUNC(mixed_minus);
-     goto do_big_arith2;
+     DO_OUTLINED_ARITH_2(mixed_minus, MinusOp1, MinusOp2);
  }
 
  OpCase(i_is_lt_f):
@@ -2767,109 +2751,81 @@ do {						\
   * Arithmetic operations.
   */
 
- OpCase(i_times_jId):
+ OpCase(i_times_jIssd):
  {
-     arith_func = ARITH_FUNC(mixed_times);
-     goto do_big_arith2;
+     Eterm Op1, Op2;
+     GetArg2(2, Op1, Op2);
+     DO_OUTLINED_ARITH_2(mixed_times, Op1, Op2);
  }
 
- OpCase(i_m_div_jId):
+ OpCase(i_m_div_jIssd):
  {
-     arith_func = ARITH_FUNC(mixed_div);
-     goto do_big_arith2;
+     Eterm Op1, Op2;
+     GetArg2(2, Op1, Op2);
+     DO_OUTLINED_ARITH_2(mixed_div, Op1, Op2);
  }
 
- OpCase(i_int_div_jId):
+ OpCase(i_int_div_jIssd):
  {
-     Eterm result;
+     Eterm Op1, Op2;
 
-     if (tmp_arg2 == SMALL_ZERO) {
+     GetArg2(2, Op1, Op2);
+     if (Op2 == SMALL_ZERO) {
 	 goto badarith;
-     } else if (is_both_small(tmp_arg1, tmp_arg2)) {
-	 Sint ires = signed_val(tmp_arg1) / signed_val(tmp_arg2);
+     } else if (is_both_small(Op1, Op2)) {
+	 Sint ires = signed_val(Op1) / signed_val(Op2);
 	 if (MY_IS_SSMALL(ires)) {
-	     result = make_small(ires);
-	     STORE_ARITH_RESULT(result);
+	     Eterm result = make_small(ires);
+	     StoreBifResult(4, result);
 	 }
      }
-     arith_func = ARITH_FUNC(int_div);
-     goto do_big_arith2;
+     DO_OUTLINED_ARITH_2(int_div, Op1, Op2);
  }
 
- OpCase(i_rem_jIxxd):
  {
-     Eterm result;
+     Eterm RemOp1, RemOp2;
 
-     if (xb(Arg(3)) == SMALL_ZERO) {
-	 goto badarith;
-     } else if (is_both_small(xb(Arg(2)), xb(Arg(3)))) {
-	 result = make_small(signed_val(xb(Arg(2))) % signed_val(xb(Arg(3))));
+ OpCase(i_rem_jIxxd):
+     RemOp1 = xb(Arg(2));
+     RemOp2 = xb(Arg(3));
+     goto do_rem;
+
+ OpCase(i_rem_jIssd):
+     GetArg2(2, RemOp1, RemOp2);
+     goto do_rem;
+
+ do_rem:
+     if (RemOp2 == SMALL_ZERO) {
+         goto badarith;
+     } else if (is_both_small(RemOp1, RemOp2)) {
+         Eterm result = make_small(signed_val(RemOp1) % signed_val(RemOp2));
          StoreBifResult(4, result);
+     } else {
+	 DO_OUTLINED_ARITH_2(int_rem, RemOp1, RemOp2);
      }
-     DO_BIG_ARITH(ARITH_FUNC(int_rem),xb(Arg(2)),xb(Arg(3)));
  }
 
- OpCase(i_rem_jId):
  {
-     Eterm result;
-
-     if (tmp_arg2 == SMALL_ZERO) {
-	 goto badarith;
-     } else if (is_both_small(tmp_arg1, tmp_arg2)) {
-	 result = make_small(signed_val(tmp_arg1) % signed_val(tmp_arg2));
-	 STORE_ARITH_RESULT(result);
-     } else {
-	 arith_func = ARITH_FUNC(int_rem);
-	 goto do_big_arith2;
-     }
- }
+     Eterm BandOp1, BandOp2;
 
  OpCase(i_band_jIxcd):
- {
-     Eterm result;
+     BandOp1 = xb(Arg(2));
+     BandOp2 = Arg(3);
+     goto do_band;
+
+ OpCase(i_band_jIssd):
+     GetArg2(2, BandOp1, BandOp2);
+     goto do_band;
 
-     if (is_both_small(xb(Arg(2)), Arg(3))) {
+ do_band:
+     if (is_both_small(BandOp1, BandOp2)) {
          /*
           * No need to untag -- TAG & TAG == TAG.
           */
-         result = xb(Arg(2)) & Arg(3);
+         Eterm result = BandOp1 & BandOp2;
          StoreBifResult(4, result);
      }
-     DO_BIG_ARITH(ARITH_FUNC(band),xb(Arg(2)),Arg(3));
- }
-
- OpCase(i_band_jId):
- {
-     Eterm result;
-
-     if (is_both_small(tmp_arg1, tmp_arg2)) {
-	 /*
-	  * No need to untag -- TAG & TAG == TAG.
-	  */
-	 result = tmp_arg1 & tmp_arg2;
-	 STORE_ARITH_RESULT(result);
-     }
-     arith_func = ARITH_FUNC(band);
-     goto do_big_arith2;
- }
-
-#undef DO_BIG_ARITH
-
- do_big_arith2:
- {
-     Eterm result;
-     Uint live = Arg(1);
-
-     SWAPOUT;
-     reg[live] = tmp_arg1;
-     reg[live+1] = tmp_arg2;
-     result = arith_func(c_p, reg, live);
-     SWAPIN;
-     ERTS_HOLE_CHECK(c_p);
-     if (is_value(result)) {
-	 STORE_ARITH_RESULT(result);
-     }
-     goto lb_Cl_error;
+     DO_OUTLINED_ARITH_2(band, BandOp1, BandOp2);
  }
 
  /*
@@ -2890,97 +2846,102 @@ do {						\
      goto find_func_info;
  }
 
- OpCase(i_bor_jId):
+ OpCase(i_bor_jIssd):
  {
-     Eterm result;
+     Eterm Op1, Op2;
 
-     if (is_both_small(tmp_arg1, tmp_arg2)) {
+     GetArg2(2, Op1, Op2);
+     if (is_both_small(Op1, Op2)) {
 	 /*
 	  * No need to untag -- TAG | TAG == TAG.
 	  */
-	 result = tmp_arg1 | tmp_arg2;
-	 STORE_ARITH_RESULT(result);
+	 Eterm result = Op1 | Op2;
+	 StoreBifResult(4, result);
      }
-     arith_func = ARITH_FUNC(bor);
-     goto do_big_arith2;
+     DO_OUTLINED_ARITH_2(bor, Op1, Op2);
  }
 
- OpCase(i_bxor_jId):
+ OpCase(i_bxor_jIssd):
  {
-     Eterm result;
+     Eterm Op1, Op2;
 
-     if (is_both_small(tmp_arg1, tmp_arg2)) {
+     GetArg2(2, Op1, Op2);
+     if (is_both_small(Op1, Op2)) {
 	 /*
 	  * We could extract the tag from one argument, but a tag extraction
 	  * could mean a shift.  Therefore, play it safe here.
 	  */
-	 result = make_small(signed_val(tmp_arg1) ^ signed_val(tmp_arg2));
-	 STORE_ARITH_RESULT(result);
+	 Eterm result = make_small(signed_val(Op1) ^ signed_val(Op2));
+	 StoreBifResult(4, result);
      }
-     arith_func = ARITH_FUNC(bxor);
-     goto do_big_arith2;
+     DO_OUTLINED_ARITH_2(bxor, Op1, Op2);
  }
 
  {
+     Eterm Op1, Op2;
      Sint i;
      Sint ires;
      Eterm* bigp;
+     Eterm tmp_big[2];
 
-     OpCase(i_bsr_jId):
-	 if (is_small(tmp_arg2)) {
-	     i = -signed_val(tmp_arg2);
-	     if (is_small(tmp_arg1)) {
+     OpCase(i_bsr_jIssd):
+         GetArg2(2, Op1, Op2);
+	 if (is_small(Op2)) {
+	     i = -signed_val(Op2);
+	     if (is_small(Op1)) {
 		 goto small_shift;
-	     } else if (is_big(tmp_arg1)) {
+	     } else if (is_big(Op1)) {
 		 if (i == 0) {
-		     StoreBifResult(2, tmp_arg1);
+		     StoreBifResult(4, Op1);
 		 }
 		 goto big_shift;
 	     }
-	 } else if (is_big(tmp_arg2)) {
+	 } else if (is_big(Op2)) {
 	     /*
 	      * N bsr NegativeBigNum == N bsl MAX_SMALL
 	      * N bsr PositiveBigNum == N bsl MIN_SMALL
 	      */
-	     tmp_arg2 = make_small(bignum_header_is_neg(*big_val(tmp_arg2)) ?
+	     Op2 = make_small(bignum_header_is_neg(*big_val(Op2)) ?
 				   MAX_SMALL : MIN_SMALL);
 	     goto do_bsl;
 	}
      goto badarith;
      
-     OpCase(i_bsl_jId):
+     OpCase(i_bsl_jIssd):
+         GetArg2(2, Op1, Op2);
+
  do_bsl:
-	 if (is_small(tmp_arg2)) {
-	     i = signed_val(tmp_arg2);
+	 if (is_small(Op2)) {
+	     i = signed_val(Op2);
 
-	     if (is_small(tmp_arg1)) {
+	     if (is_small(Op1)) {
 	     small_shift:
-		 ires = signed_val(tmp_arg1);
+		 ires = signed_val(Op1);
 	     
 		 if (i == 0 || ires == 0) {
-		     StoreBifResult(2, tmp_arg1);
+		     StoreBifResult(4, Op1);
 		 } else if (i < 0)  { /* Right shift */
 		     i = -i;
 		     if (i >= SMALL_BITS-1) {
-			 tmp_arg1 = (ires < 0) ? SMALL_MINUS_ONE : SMALL_ZERO;
+			 Op1 = (ires < 0) ? SMALL_MINUS_ONE : SMALL_ZERO;
 		     } else {
-			 tmp_arg1 = make_small(ires >> i);
+			 Op1 = make_small(ires >> i);
 		     }
-		     StoreBifResult(2, tmp_arg1);
+		     StoreBifResult(4, Op1);
 		 } else if (i < SMALL_BITS-1) { /* Left shift */
 		     if ((ires > 0 && ((~(Uint)0 << ((SMALL_BITS-1)-i)) & ires) == 0) ||
 			 ((~(Uint)0 << ((SMALL_BITS-1)-i)) & ~ires) == 0) {
-			 tmp_arg1 = make_small(ires << i);
-			 StoreBifResult(2, tmp_arg1);
+			 Op1 = make_small(ires << i);
+			 StoreBifResult(4, Op1);
 		     }
 		 }
-		 tmp_arg1 = small_to_big(ires, tmp_big);
+		 Op1 = small_to_big(ires, tmp_big);
 
 	     big_shift:
 		 if (i > 0) {	/* Left shift. */
-		     ires = big_size(tmp_arg1) + (i / D_EXP);
+		     ires = big_size(Op1) + (i / D_EXP);
 		 } else {	/* Right shift. */
-		     ires = big_size(tmp_arg1);
+		     ires = big_size(Op1);
 		     if (ires <= (-i / D_EXP))
 			 ires = 3; /* ??? */
 		     else
@@ -2998,14 +2959,14 @@ do {						\
 			 c_p->freason = SYSTEM_LIMIT;
 			 goto lb_Cl_error;
 		     }
-		     TestHeapPreserve(ires+1, Arg(1), tmp_arg1);
+		     TestHeapPreserve(ires+1, Arg(1), Op1);
 		     bigp = HTOP;
-		     tmp_arg1 = big_lshift(tmp_arg1, i, bigp);
-		     if (is_big(tmp_arg1)) {
+		     Op1 = big_lshift(Op1, i, bigp);
+		     if (is_big(Op1)) {
 			 HTOP += bignum_header_arity(*HTOP) + 1;
 		     }
 		     HEAP_SPACE_VERIFIED(0);
-		     if (is_nil(tmp_arg1)) {
+		     if (is_nil(Op1)) {
 			 /*
 			  * This result must have been only slight larger
 			  * than allowed since it wasn't caught by the
@@ -3015,25 +2976,25 @@ do {						\
 			 goto lb_Cl_error;
 		     }
 		     ERTS_HOLE_CHECK(c_p);
-		     StoreBifResult(2, tmp_arg1);
+		     StoreBifResult(4, Op1);
 		 }
-	     } else if (is_big(tmp_arg1)) {
+	     } else if (is_big(Op1)) {
 		 if (i == 0) {
-		     StoreBifResult(2, tmp_arg1);
+		     StoreBifResult(4, Op1);
 		 }
 		 goto big_shift;
 	     }
-	 } else if (is_big(tmp_arg2)) {
-	     if (bignum_header_is_neg(*big_val(tmp_arg2))) {
+	 } else if (is_big(Op2)) {
+	     if (bignum_header_is_neg(*big_val(Op2))) {
 		 /*
 		  * N bsl NegativeBigNum is either 0 or -1, depending on
 		  * the sign of N. Since we don't believe this case
 		  * is common, do the calculation with the minimum
 		  * amount of code.
 		  */
-		 tmp_arg2 = make_small(MIN_SMALL);
+		 Op2 = make_small(MIN_SMALL);
 		 goto do_bsl;
-	     } else if (is_small(tmp_arg1) || is_big(tmp_arg1)) {
+	     } else if (is_small(Op1) || is_big(Op1)) {
 		 /*
 		  * N bsl PositiveBigNum is too large to represent.
 		  */
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 335f8e2db5..e4a757ec8b 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -1475,68 +1475,95 @@ i_get_map_element f y x x
 i_get_map_element f x x y
 i_get_map_element f y x y
 
+#
+# Convert the plus operations to a generic plus instruction.
+#
+gen_plus/5
+gen_minus/5
+
+gc_bif1 Fail Live u$bif:erlang:splus/1 Src Dst => \
+   gen_plus Fail Live Src i Dst
+gc_bif2 Fail Live u$bif:erlang:splus/2 S1 S2 Dst => \
+   gen_plus Fail Live S1 S2 Dst
+
+gc_bif1 Fail Live u$bif:erlang:sminus/1 Src Dst => \
+   gen_minus Fail Live i Src Dst
+gc_bif2 Fail Live u$bif:erlang:sminus/2 S1 S2 Dst => \
+   gen_minus Fail Live S1 S2 Dst
+
 #
 # Optimize addition and subtraction of small literals using
 # the i_increment/4 instruction (in bodies, not in guards).
 #
 
-gc_bif2 p Live u$bif:erlang:splus/2 Int=i Reg=x Dst => \
+gen_plus p Live Int=i Reg=d Dst => \
 	gen_increment(Reg, Int, Live, Dst)
-gc_bif2 p Live u$bif:erlang:splus/2 Reg=x Int=i Dst => \
+gen_plus p Live Reg=d Int=i Dst => \
 	gen_increment(Reg, Int, Live, Dst)
 
-gc_bif2 p Live u$bif:erlang:sminus/2 Reg=d Int=i Dst | \
-	negation_is_small(Int) => \
+gen_minus p Live Reg=d Int=i Dst | negation_is_small(Int) => \
 	gen_increment_from_minus(Reg, Int, Live, Dst)
 
 #
 # GCing arithmetic instructions.
 #
 
-gc_bif2 Fail I u$bif:erlang:splus/2 S1=x S2=x Dst=d => i_plus Fail I S1 S2 Dst
-gc_bif2 Fail I u$bif:erlang:splus/2 S1 S2 Dst=d => i_fetch S1 S2 | i_plus Fail I Dst
-gc_bif2 Fail I u$bif:erlang:sminus/2 S1=x S2=x Dst=d => i_minus Fail I S1 S2 Dst
-gc_bif2 Fail I u$bif:erlang:sminus/2 S1 S2 Dst=d => i_fetch S1 S2 | i_minus Fail I Dst
-gc_bif2 Fail I u$bif:erlang:stimes/2 S1 S2 Dst=d => i_fetch S1 S2 | i_times Fail I Dst
-gc_bif2 Fail I u$bif:erlang:div/2 S1 S2 Dst=d => i_fetch S1 S2 | i_m_div Fail I Dst
+gen_plus Fail Live S1 S2 Dst => i_plus Fail Live S1 S2 Dst
 
-gc_bif2 Fail I u$bif:erlang:intdiv/2 S1 S2 Dst=d => i_fetch S1 S2 | i_int_div Fail I Dst
-gc_bif2 Fail I u$bif:erlang:rem/2 S1=x S2=x Dst=d => i_rem Fail I S1 S2 Dst
-gc_bif2 Fail I u$bif:erlang:rem/2 S1 S2 Dst=d => i_fetch S1 S2 | i_rem Fail I Dst
+gen_minus Fail Live S1 S2 Dst => i_minus Fail Live S1 S2 Dst
 
-gc_bif2 Fail I u$bif:erlang:bsl/2 S1 S2 Dst=d => i_fetch S1 S2 | i_bsl Fail I Dst
-gc_bif2 Fail I u$bif:erlang:bsr/2 S1 S2 Dst=d => i_fetch S1 S2 | i_bsr Fail I Dst
+gc_bif2 Fail Live u$bif:erlang:stimes/2 S1 S2 Dst => \
+  i_times Fail Live S1 S2 Dst
 
-gc_bif2 Fail I u$bif:erlang:band/2 S1=x S2=c Dst=d => i_band Fail I S1 S2 Dst
-gc_bif2 Fail I u$bif:erlang:band/2 S1 S2 Dst=d => i_fetch S1 S2 | i_band Fail I Dst
-gc_bif2 Fail I u$bif:erlang:bor/2 S1 S2 Dst=d => i_fetch S1 S2 | i_bor Fail I Dst
-gc_bif2 Fail I u$bif:erlang:bxor/2 S1 S2 Dst=d => i_fetch S1 S2 | i_bxor Fail I Dst
+gc_bif2 Fail Live u$bif:erlang:div/2 S1 S2 Dst => \
+  i_m_div Fail Live S1 S2 Dst
+gc_bif2 Fail Live u$bif:erlang:intdiv/2 S1 S2 Dst => \
+  i_int_div Fail Live S1 S2 Dst
 
-gc_bif1 Fail I u$bif:erlang:bnot/1 Src Dst=d => i_int_bnot Fail Src I Dst
+gc_bif2 Fail Live u$bif:erlang:rem/2 S1 S2 Dst => \
+  i_rem Fail Live S1 S2 Dst
+
+gc_bif2 Fail Live u$bif:erlang:bsl/2 S1 S2 Dst => \
+  i_bsl Fail Live S1 S2 Dst
+gc_bif2 Fail Live u$bif:erlang:bsr/2 S1 S2 Dst => \
+  i_bsr Fail Live S1 S2 Dst
+
+gc_bif2 Fail Live u$bif:erlang:band/2 S1 S2 Dst => \
+  i_band Fail Live S1 S2 Dst
 
-gc_bif1 Fail I u$bif:erlang:sminus/1 Src Dst=d => i_fetch i Src | i_minus Fail I Dst
-gc_bif1 Fail I u$bif:erlang:splus/1 Src Dst=d => i_fetch i Src | i_plus Fail I Dst
+gc_bif2 Fail Live u$bif:erlang:bor/2 S1 S2 Dst => \
+  i_bor Fail Live S1 S2 Dst
+
+gc_bif2 Fail Live u$bif:erlang:bxor/2 S1 S2 Dst => \
+  i_bxor Fail Live S1 S2 Dst
+
+gc_bif1 Fail I u$bif:erlang:bnot/1 Src Dst=d => i_int_bnot Fail Src I Dst
 
 i_increment x I I d
 i_increment y I I d
 
 i_plus j I x x d
-i_plus j I d
+i_plus j I s s d
+
 i_minus j I x x d
-i_minus j I d
-i_times j I d
-i_m_div j I d
-i_int_div j I d
+i_minus j I s s d
+
+i_times j I s s d
+
+i_m_div j I s s d
+i_int_div j I s s d
+
 i_rem j I x x d
-i_rem j I d
+i_rem j I s s d
 
-i_bsl j I d
-i_bsr j I d
+i_bsl j I s s d
+i_bsr j I s s d
 
 i_band j I x c d
-i_band j I d
-i_bor j I d
-i_bxor j I d
+i_band j I s s d
+
+i_bor j I s s d
+i_bxor j I s s d
 
 i_int_bnot j s I d
 
-- 
cgit v1.2.3


From 3c583d590589b64ab543102f2542bda0337ce2f9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 17 Jun 2015 06:52:50 +0200
Subject: Eliminate the use of i_fetch for relational operators

---
 erts/emulator/beam/beam_emu.c | 39 ++-----------------------
 erts/emulator/beam/ops.tab    | 67 ++++++++++++++-----------------------------
 2 files changed, 25 insertions(+), 81 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 82cc030878..b63fa038d2 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -651,6 +651,9 @@ void** beam_ops;
 #define EqualImmed(X, Y, Action) if (X != Y) { Action; }
 #define NotEqualImmed(X, Y, Action) if (X == Y) { Action; }
 #define EqualExact(X, Y, Action) if (!EQ(X,Y)) { Action; }
+#define NotEqualExact(X, Y, Action) if (EQ(X,Y)) { Action; }
+#define Equal(X, Y, Action) if (!CMP_EQ(X,Y)) { Action; }
+#define NotEqual(X, Y, Action) if (!CMP_NE(X,Y)) { Action; }
 #define IsLessThan(X, Y, Action) if (CMP_GE(X, Y)) { Action; }
 #define IsGreaterEqual(X, Y, Action) if (CMP_LT(X, Y)) { Action; }
 
@@ -1410,36 +1413,6 @@ void process_main(void)
      DO_OUTLINED_ARITH_2(mixed_minus, MinusOp1, MinusOp2);
  }
 
- OpCase(i_is_lt_f):
-    if (CMP_GE(tmp_arg1, tmp_arg2)) {
-	ClauseFail();
-    }
-    Next(1);
-
- OpCase(i_is_ge_f):
-    if (CMP_LT(tmp_arg1, tmp_arg2)) {
-	ClauseFail();
-    }
-    Next(1);
-
- OpCase(i_is_eq_f):
-    if (CMP_NE(tmp_arg1, tmp_arg2)) {
-	ClauseFail();
-    }
-    Next(1);
-
- OpCase(i_is_ne_f):
-    if (CMP_EQ(tmp_arg1, tmp_arg2)) {
-	ClauseFail();
-    }
-    Next(1);
-
- OpCase(i_is_eq_exact_f):
-    if (!EQ(tmp_arg1, tmp_arg2)) {
-	ClauseFail();
-    }
-    Next(1);
-
     {
 	Eterm is_eq_exact_lit_val;
 
@@ -3276,12 +3249,6 @@ do {						\
      NextPF(3, next);
  }
 
- OpCase(i_is_ne_exact_f):
-    if (EQ(tmp_arg1, tmp_arg2)) {
-	ClauseFail();
-    }
-    Next(1);
-
  OpCase(normal_exit): {
      SWAPOUT;
      c_p->freason = EXC_NORMAL;
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index e4a757ec8b..57fb1ea026 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -397,59 +397,36 @@ i_is_ne_exact_immed f y c
 i_is_ne_exact_literal f x c
 i_is_ne_exact_literal f y c
 
-#
-# Common Compare Specializations
-# We don't do all of them since we want
-# to keep the instruction set small-ish
-#
-
-is_eq_exact Lbl S1=y S2=x => is_eq_exact Lbl S2 S1
-is_eq_exact Lbl S1=x S2=xy => i_is_eq_exact_spec Lbl S1 S2
-%macro: i_is_eq_exact_spec EqualExact -fail_action
-
-i_is_eq_exact_spec f x x
-i_is_eq_exact_spec f x y
-
-is_lt Lbl S1=xc S2=xc  => i_is_lt_spec Lbl S1 S2
-
-%macro: i_is_lt_spec IsLessThan -fail_action
-
-i_is_lt_spec f x x
-i_is_lt_spec f x c
-i_is_lt_spec f c x
+is_eq_exact Lbl Y=y X=x => is_eq_exact Lbl X Y
+%macro: is_eq_exact EqualExact -fail_action
+is_eq_exact f x x
+is_eq_exact f x y
+is_eq_exact f s s
+
+%macro: is_lt IsLessThan -fail_action
+is_lt f x x
+is_lt f x c
+is_lt f c x
 %cold
-i_is_lt_spec f c c
+is_lt f s s
 %hot
 
-is_ge Lbl S1=xc S2=xc => i_is_ge_spec Lbl S1 S2
-
-%macro: i_is_ge_spec IsGreaterEqual -fail_action
-
-i_is_ge_spec f x x
-i_is_ge_spec f x c
-i_is_ge_spec f c x
+%macro: is_ge IsGreaterEqual -fail_action
+is_ge f x x
+is_ge f x c
+is_ge f c x
 %cold
-i_is_ge_spec f c c
+is_ge f s s
 %hot
 
-#
-# All other comparisons.
-#
-
-is_eq_exact Lbl S1 S2 => i_fetch S1 S2 | i_is_eq_exact Lbl
-is_ne_exact Lbl S1 S2 => i_fetch S1 S2 | i_is_ne_exact Lbl
+%macro: is_ne_exact NotEqualExact -fail_action
+is_ne_exact f s s
 
-is_lt Lbl S1 S2 => i_fetch S1 S2 | i_is_lt Lbl
-is_ge Lbl S1 S2 => i_fetch S1 S2 | i_is_ge Lbl
-is_eq Lbl S1 S2 => i_fetch S1 S2 | i_is_eq Lbl
-is_ne Lbl S1 S2 => i_fetch S1 S2 | i_is_ne Lbl
+%macro: is_eq Equal -fail_action
+is_eq f s s
 
-i_is_eq_exact f
-i_is_ne_exact f
-i_is_lt f
-i_is_ge f
-i_is_eq f
-i_is_ne f
+%macro: is_ne NotEqual -fail_action
+is_ne f s s
 
 #
 # Putting things.
-- 
cgit v1.2.3


From 30204739a047ab96d2b7d59ae461d4cbb2131509 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 17 Jun 2015 10:34:39 +0200
Subject: Eliminate the use of i_fetch for BIF instructions

---
 erts/emulator/beam/beam_emu.c  | 68 ++++++++++++++++++++++++------------------
 erts/emulator/beam/beam_load.c | 28 ++++++-----------
 erts/emulator/beam/ops.tab     | 21 +++++++------
 3 files changed, 58 insertions(+), 59 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index b63fa038d2..d260c74255 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -2513,12 +2513,10 @@ do {						\
     {
 	typedef Eterm (*GcBifFunction)(Process*, Eterm*, Uint);
 	GcBifFunction bf;
-	Eterm arg;
 	Eterm result;
 	Uint live = (Uint) Arg(3);
 
-	GetArg1(2, arg);
-	reg[live] = arg;
+	GetArg1(2, x(live));
 	bf = (GcBifFunction) Arg(1);
 	c_p->fcalls = FCALLS;
 	SWAPOUT;
@@ -2538,12 +2536,12 @@ do {						\
 	    SET_I((BeamInstr *) Arg(0));
 	    Goto(*I);
 	}
-	reg[0] = arg;
+	x(0) = x(live);
 	I = handle_error(c_p, I, reg, translate_gc_bif((void *) bf));
 	goto post_error_handling;
     }
 
- OpCase(i_gc_bif2_jIId): /* Note, one less parameter than the i_gc_bif1
+ OpCase(i_gc_bif2_jIIssd): /* Note, one less parameter than the i_gc_bif1
 			    and i_gc_bif3 */
     {
 	typedef Eterm (*GcBifFunction)(Process*, Eterm*, Uint);
@@ -2551,8 +2549,13 @@ do {						\
 	Eterm result;
 	Uint live = (Uint) Arg(2);
 
-	reg[live++] = tmp_arg1;
-	reg[live] = tmp_arg2;
+	GetArg2(3, x(live), x(live+1));
+	/*
+	 * XXX This calling convention does not make sense. 'live'
+	 * should point out the first argument, not the second
+	 * (i.e. 'live' should not be incremented below).
+	 */
+	live++;
 	bf = (GcBifFunction) Arg(1);
 	c_p->fcalls = FCALLS;
 	SWAPOUT;
@@ -2566,30 +2569,34 @@ do {						\
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
 	if (is_value(result)) {
-	    StoreBifResult(3, result);
+	    StoreBifResult(5, result);
 	}
 	if (Arg(0) != 0) {
 	    SET_I((BeamInstr *) Arg(0));
 	    Goto(*I);
 	}
-	reg[0] = tmp_arg1;
-	reg[1] = tmp_arg2;
+	live--;
+	x(0) = x(live);
+	x(1) = x(live+1);
 	I = handle_error(c_p, I, reg, translate_gc_bif((void *) bf));
 	goto post_error_handling;
     }
 
- OpCase(i_gc_bif3_jIsId):
+ OpCase(i_gc_bif3_jIIssd):
     {
 	typedef Eterm (*GcBifFunction)(Process*, Eterm*, Uint);
 	GcBifFunction bf;
-	Eterm arg;
 	Eterm result;
-	Uint live = (Uint) Arg(3);
+	Uint live = (Uint) Arg(2);
 
-	GetArg1(2, arg);
-	reg[live++] = arg;
-	reg[live++] = tmp_arg1;
-	reg[live] = tmp_arg2;
+	x(live) = x(SCRATCH_X_REG);
+	GetArg2(3, x(live+1), x(live+2));
+	/*
+	 * XXX This calling convention does not make sense. 'live'
+	 * should point out the first argument, not the third
+	 * (i.e. 'live' should not be incremented below).
+	 */
+	live += 2;
 	bf = (GcBifFunction) Arg(1);
 	c_p->fcalls = FCALLS;
 	SWAPOUT;
@@ -2603,15 +2610,16 @@ do {						\
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
 	if (is_value(result)) {
-	    StoreBifResult(4, result);
+	    StoreBifResult(5, result);
 	}
 	if (Arg(0) != 0) {
 	    SET_I((BeamInstr *) Arg(0));
 	    Goto(*I);
 	}
-	reg[0] = arg;
-	reg[1] = tmp_arg1;
-	reg[2] = tmp_arg2;
+	live -= 2;
+	x(0) = x(live);
+	x(1) = x(live+1);
+	x(2) = x(live+2);
 	I = handle_error(c_p, I, reg, translate_gc_bif((void *) bf));
 	goto post_error_handling;
     }
@@ -2619,12 +2627,13 @@ do {						\
  /*
   * Guards bifs and, or, xor in guards.
   */
- OpCase(i_bif2_fbd):
+ OpCase(i_bif2_fbssd):
     {
-	Eterm tmp_reg[2] = {tmp_arg1, tmp_arg2};
+	Eterm tmp_reg[2];
 	Eterm (*bf)(Process*, Eterm*);
 	Eterm result;
 
+	GetArg2(2, tmp_reg[0], tmp_reg[1]);
 	bf = (BifFunction) Arg(1);
 	c_p->fcalls = FCALLS;
 	PROCESS_MAIN_CHK_LOCKS(c_p);
@@ -2636,7 +2645,7 @@ do {						\
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
 	if (is_value(result)) {
-	    StoreBifResult(2, result);
+	    StoreBifResult(4, result);
 	}
 	SET_I((BeamInstr *) Arg(0));
 	Goto(*I);
@@ -2645,12 +2654,13 @@ do {						\
  /*
   * Guards bifs and, or, xor, relational operators in body.
   */
- OpCase(i_bif2_body_bd):
+ OpCase(i_bif2_body_bssd):
     {
-	Eterm tmp_reg[2] = {tmp_arg1, tmp_arg2};
+	Eterm tmp_reg[2];
 	Eterm (*bf)(Process*, Eterm*);
 	Eterm result;
 
+	GetArg2(1, tmp_reg[0], tmp_reg[1]);
 	bf = (BifFunction) Arg(0);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
@@ -2661,10 +2671,10 @@ do {						\
 	ERTS_HOLE_CHECK(c_p);
 	if (is_value(result)) {
 	    ASSERT(!is_CP(result));
-	    StoreBifResult(1, result);
+	    StoreBifResult(3, result);
 	}
-	reg[0] = tmp_arg1;
-	reg[1] = tmp_arg2;
+	reg[0] = tmp_reg[0];
+	reg[1] = tmp_reg[1];
 	SWAPOUT;
 	I = handle_error(c_p, I, reg, bf);
 	goto post_error_handling;
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 3671baa552..c90576333a 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -3921,9 +3921,7 @@ gen_make_fun2(LoaderState* stp, GenOpArg idx)
 /*
  * Rewrite gc_bifs with one parameter (the common case). Utilized
  * in ops.tab to rewrite instructions calling bif's in guards
- * to use a garbage collecting implementation. The instructions
- * are sometimes once again rewritten to handle literals (putting the
- * parameter in the mostly unused r[0] before the instruction is executed).
+ * to use a garbage collecting implementation.
  */
 static GenOp*
 gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
@@ -3978,10 +3976,6 @@ gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
 
 /*
  * This is used by the ops.tab rule that rewrites gc_bifs with two parameters.
- * The instruction returned is then again rewritten to an i_load instruction
- * followed by i_gc_bif2_jIId, to handle literals properly.
- * As opposed to the i_gc_bif1_jIsId, the instruction  i_gc_bif2_jIId is
- * always rewritten, regardless of if there actually are any literals.
  */
 static GenOp*
 gen_guard_bif2(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
@@ -4008,23 +4002,19 @@ gen_guard_bif2(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
 	op->a[2].val = stp->import[Bif.val].arity;
 	return op;
     }
-    op->op = genop_ii_gc_bif2_6;
+    op->op = genop_i_gc_bif2_6;
     op->arity = 6;
     op->a[0] = Fail;
     op->a[1].type = TAG_u;
-    op->a[2] = S1;
-    op->a[3] = S2;
-    op->a[4] = Live;
+    op->a[2] = Live;
+    op->a[3] = S1;
+    op->a[4] = S2;
     op->a[5] = Dst;
     return op;
 }
 
 /*
  * This is used by the ops.tab rule that rewrites gc_bifs with three parameters.
- * The instruction returned is then again rewritten to a move instruction that
- * uses r[0] for temp storage, followed by an i_load instruction,
- * followed by i_gc_bif3_jIsId, to handle literals properly. Rewriting
- * always occur, as with the gc_bif2 counterpart.
  */
 static GenOp*
 gen_guard_bif3(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
@@ -4055,10 +4045,10 @@ gen_guard_bif3(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
     op->arity = 7;
     op->a[0] = Fail;
     op->a[1].type = TAG_u;
-    op->a[2] = S1;
-    op->a[3] = S2;
-    op->a[4] = S3;
-    op->a[5] = Live;
+    op->a[2] = Live;
+    op->a[3] = S1;
+    op->a[4] = S2;
+    op->a[5] = S3;
     op->a[6] = Dst;
     op->next = NULL;
     return op;
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 57fb1ea026..492d4d240a 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -995,8 +995,8 @@ bif2 Jump=j u$bif:erlang:element/2 S1=s S2=xy Dst=d => gen_element(Jump, S1, S2,
 
 bif1 p Bif S1 Dst => bif1_body Bif S1 Dst
 
-bif2 p Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2_body Bif Dst
-bif2 Fail Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2 Fail Bif Dst
+bif2 p Bif S1 S2 Dst => i_bif2_body Bif S1 S2 Dst
+bif2 Fail Bif S1 S2 Dst => i_bif2 Fail Bif S1 S2 Dst
 
 i_get s d
 
@@ -1018,8 +1018,8 @@ i_element j y s d
 
 bif1 f b s d
 bif1_body b s d
-i_bif2 f b d
-i_bif2_body b d
+i_bif2 f b s s d
+i_bif2_body b s s d
 
 #
 # Internal calls.
@@ -1568,17 +1568,16 @@ gc_bif3 Fail I Bif S1 S2 S3 Dst => \
 
 i_gc_bif1 j I s I d
 
-ii_gc_bif2/6
-
-ii_gc_bif2 Fail Bif S1 S2 Live D => i_fetch S1 S2 | i_gc_bif2 Fail Bif Live D
-
-i_gc_bif2 j I I d
+i_gc_bif2 j I I s s d
 
 ii_gc_bif3/7
 
-ii_gc_bif3 Fail Bif S1 S2 S3 Live D => move S1 x | i_fetch S2 S3 | i_gc_bif3 Fail Bif x Live D
+# A specific instruction can only have 6 operands, so we must
+# pass one of the arguments in an x register.
+ii_gc_bif3 Fail Bif Live S1 S2 S3 Dst => \
+  move S1 x | i_gc_bif3 Fail Bif Live S2 S3 Dst
 
-i_gc_bif3 j I s I d
+i_gc_bif3 j I I s s d
 
 #
 # The following instruction is specially handled in beam_load.c
-- 
cgit v1.2.3


From bede3941be8629efa4d91755c085a91b1416d432 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 17 Jun 2015 15:31:36 +0200
Subject: Eliminate use of i_fetch for bit syntax instructions

---
 erts/emulator/beam/beam_emu.c  | 207 +++++++++++++++++++++--------------------
 erts/emulator/beam/beam_load.c |  27 ++----
 erts/emulator/beam/ops.tab     |  29 +++---
 3 files changed, 129 insertions(+), 134 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index d260c74255..26e28832cf 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -3548,11 +3548,10 @@ do {						\
 	 goto do_bs_init_bits_known;
      }
 
-     OpCase(i_bs_init_bits_fail_heap_IjId): {
-	 /* tmp_arg1 was fetched by an i_fetch instruction */
-	 num_bits_term = tmp_arg1;
-	 alloc = Arg(0);
-	 I++;
+     OpCase(i_bs_init_bits_fail_heap_sIjId): {
+	 GetArg1(0, num_bits_term);
+	 alloc = Arg(1);
+	 I += 2;
 	 goto do_bs_init_bits;
      }
 
@@ -3675,46 +3674,48 @@ do {						\
  }
 
  {
-     OpCase(i_bs_init_fail_heap_IjId): {
-	 /* tmp_arg1 was fetched by an i_fetch instruction */
-	 tmp_arg2 = Arg(0);
-	 I++;
+     Eterm BsOp1, BsOp2;
+
+     OpCase(i_bs_init_fail_heap_sIjId): {
+	 GetArg1(0, BsOp1);
+	 BsOp2 = Arg(1);
+	 I += 2;
 	 goto do_bs_init;
      }
 
      OpCase(i_bs_init_fail_yjId): {
-	 tmp_arg1 = yb(Arg(0));
-	 tmp_arg2 = 0;
+	 BsOp1 = yb(Arg(0));
+	 BsOp2 = 0;
 	 I++;
 	 goto do_bs_init;
      }
 
      OpCase(i_bs_init_fail_xjId): {
-	 tmp_arg1 = xb(Arg(0));
-	 tmp_arg2 = 0;
+	 BsOp1 = xb(Arg(0));
+	 BsOp2 = 0;
 	 I++;
      }
 	 /* FALL THROUGH */
      do_bs_init:
-         if (is_small(tmp_arg1)) {
-	     Sint size = signed_val(tmp_arg1);
+         if (is_small(BsOp1)) {
+	     Sint size = signed_val(BsOp1);
 	     if (size < 0) {
 		 goto badarg;
 	     }
-	     tmp_arg1 = (Eterm) size;
+	     BsOp1 = (Eterm) size;
 	 } else {
 	     Uint bytes;
 
-	     if (!term_to_Uint(tmp_arg1, &bytes)) {
+	     if (!term_to_Uint(BsOp1, &bytes)) {
 		 c_p->freason = bytes;
 		 goto lb_Cl_error;
 	     }
 	     if ((bytes >> (8*sizeof(Uint)-3)) != 0) {
 		 goto system_limit;
 	     }
-	     tmp_arg1 = (Eterm) bytes;
+	     BsOp1 = (Eterm) bytes;
 	 }
-	 if (tmp_arg1 <= ERL_ONHEAP_BIN_LIMIT) {
+	 if (BsOp1 <= ERL_ONHEAP_BIN_LIMIT) {
 	     goto do_heap_bin_alloc;
 	 } else {
 	     goto do_proc_bin_alloc;
@@ -3722,15 +3723,15 @@ do {						\
 
 
      OpCase(i_bs_init_heap_IIId): {
-	 tmp_arg1 = Arg(0);
-	 tmp_arg2 = Arg(1);
+	 BsOp1 = Arg(0);
+	 BsOp2 = Arg(1);
 	 I++;
 	 goto do_proc_bin_alloc;
      }
 
      OpCase(i_bs_init_IId): {
-	 tmp_arg1 = Arg(0);
-	 tmp_arg2 = 0;
+	 BsOp1 = Arg(0);
+	 BsOp2 = 0;
      }
      /* FALL THROUGH */
      do_proc_bin_alloc: {
@@ -3739,13 +3740,13 @@ do {						\
 
 	 erts_bin_offset = 0;
 	 erts_writable_bin = 0;
-	 TestBinVHeap(tmp_arg1 / sizeof(Eterm), 
-	 	      tmp_arg2 + PROC_BIN_SIZE + ERL_SUB_BIN_SIZE, Arg(1));
+	 TestBinVHeap(BsOp1 / sizeof(Eterm),
+		      BsOp2 + PROC_BIN_SIZE + ERL_SUB_BIN_SIZE, Arg(1));
 
 	 /*
 	  * Allocate the binary struct itself.
 	  */
-	 bptr = erts_bin_nrml_alloc(tmp_arg1);
+	 bptr = erts_bin_nrml_alloc(BsOp1);
 	 erts_refc_init(&bptr->refc, 1);
 	 erts_current_bin = (byte *) bptr->orig_bytes;
 
@@ -3755,28 +3756,28 @@ do {						\
 	 pb = (ProcBin *) HTOP;
 	 HTOP += PROC_BIN_SIZE;
 	 pb->thing_word = HEADER_PROC_BIN;
-	 pb->size = tmp_arg1;
+	 pb->size = BsOp1;
 	 pb->next = MSO(c_p).first;
 	 MSO(c_p).first = (struct erl_off_heap_header*) pb;
 	 pb->val = bptr;
 	 pb->bytes = (byte*) bptr->orig_bytes;
 	 pb->flags = 0;
 	 
-	 OH_OVERHEAD(&(MSO(c_p)), tmp_arg1 / sizeof(Eterm));
+	 OH_OVERHEAD(&(MSO(c_p)), BsOp1 / sizeof(Eterm));
 
 	 StoreBifResult(2, make_binary(pb));
      }
 
      OpCase(i_bs_init_heap_bin_heap_IIId): {
-	 tmp_arg1 = Arg(0);
-	 tmp_arg2 = Arg(1);
+	 BsOp1 = Arg(0);
+	 BsOp2 = Arg(1);
 	 I++;
 	 goto do_heap_bin_alloc;
      }
 
      OpCase(i_bs_init_heap_bin_IId): {
-	 tmp_arg1 = Arg(0);
-	 tmp_arg2 = 0;
+	 BsOp1 = Arg(0);
+	 BsOp2 = 0;
      }
      /* Fall through */
      do_heap_bin_alloc:
@@ -3784,33 +3785,36 @@ do {						\
 	     ErlHeapBin* hb;
 	     Uint bin_need;
 
-	     bin_need = heap_bin_size(tmp_arg1);
+	     bin_need = heap_bin_size(BsOp1);
 	     erts_bin_offset = 0;
 	     erts_writable_bin = 0;
-	     TestHeap(bin_need+tmp_arg2+ERL_SUB_BIN_SIZE, Arg(1));
+	     TestHeap(bin_need+BsOp2+ERL_SUB_BIN_SIZE, Arg(1));
 	     hb = (ErlHeapBin *) HTOP;
 	     HTOP += bin_need;
-	     hb->thing_word = header_heap_bin(tmp_arg1);
-	     hb->size = tmp_arg1;
+	     hb->thing_word = header_heap_bin(BsOp1);
+	     hb->size = BsOp1;
 	     erts_current_bin = (byte *) hb->data;
-	     tmp_arg1 = make_binary(hb);
-	     StoreBifResult(2, tmp_arg1);
+	     BsOp1 = make_binary(hb);
+	     StoreBifResult(2, BsOp1);
 	 }
  }
 
- OpCase(i_bs_add_jId): {
-     Uint Unit = Arg(1);
-     if (is_both_small(tmp_arg1, tmp_arg2)) {
-	 Sint Arg1 = signed_val(tmp_arg1);
-	 Sint Arg2 = signed_val(tmp_arg2);
+ OpCase(bs_add_jssId): {
+     Eterm Op1, Op2;
+     Uint Unit = Arg(3);
+
+     GetArg2(1, Op1, Op2);
+     if (is_both_small(Op1, Op2)) {
+	 Sint Arg1 = signed_val(Op1);
+	 Sint Arg2 = signed_val(Op2);
 
 	 if (Arg1 >= 0 && Arg2 >= 0) {
-	     BsSafeMul(Arg2, Unit, goto system_limit, tmp_arg1);
-	     tmp_arg1 += Arg1;
+	     BsSafeMul(Arg2, Unit, goto system_limit, Op1);
+	     Op1 += Arg1;
 
 	 store_bs_add_result:
-	     if (MY_IS_SSMALL((Sint) tmp_arg1)) {
-		 tmp_arg1 = make_small(tmp_arg1);
+	     if (MY_IS_SSMALL((Sint) Op1)) {
+		 Op1 = make_small(Op1);
 	     } else {
 		 /*
 		  * May generate a heap fragment, but in this
@@ -3822,10 +3826,10 @@ do {						\
 		  * references (such as the heap).
 		  */
 		 SWAPOUT;
-		 tmp_arg1 = erts_make_integer(tmp_arg1, c_p);
+		 Op1 = erts_make_integer(Op1, c_p);
 		 HTOP = HEAP_TOP(c_p);
 	     }
-	     StoreBifResult(2, tmp_arg1);
+	     StoreBifResult(4, Op1);
 	 }
 	 goto badarg;
      } else {
@@ -3848,16 +3852,16 @@ do {						\
 	  * an Uint, the reason is SYSTEM_LIMIT.
 	  */
 
-	 if (!term_to_Uint(tmp_arg1, &a)) {
+	 if (!term_to_Uint(Op1, &a)) {
 	     if (a == BADARG) {
 		 goto badarg;
 	     }
-	     if (!term_to_Uint(tmp_arg2, &b)) {
+	     if (!term_to_Uint(Op2, &b)) {
 		 c_p->freason = b;
 		 goto lb_Cl_error;
 	     }
 	     goto system_limit;
-	 } else if (!term_to_Uint(tmp_arg2, &b)) {
+	 } else if (!term_to_Uint(Op2, &b)) {
 	     c_p->freason = b;
 	     goto lb_Cl_error;
 	 }
@@ -3867,8 +3871,8 @@ do {						\
 	  */
 	 
 	 BsSafeMul(b, Unit, goto system_limit, c);
-	 tmp_arg1 = a + c;
-	 if (tmp_arg1 < a) {
+	 Op1 = a + c;
+	 if (Op1 < a) {
 	     /*
 	      * If the result is less than one of the
 	      * arguments, there must have been an overflow.
@@ -3890,46 +3894,43 @@ do {						\
     }
 
  /*
-  * tmp_arg1 = Number of bytes to build
-  * tmp_arg2 = Source binary
-  * Operands: Fail ExtraHeap Live Unit Dst
+  * x(SCRATCH_X_REG);
+  * Operands: Fail ExtraHeap Live Unit Size Dst
   */
 
- OpCase(i_bs_append_jIIId): {
+ OpCase(i_bs_append_jIIIsd): {
      Uint live = Arg(2);
      Uint res;
+     Eterm Size;
 
+     GetArg1(4, Size);
      SWAPOUT;
-     reg[live] = tmp_arg2;
-     res = erts_bs_append(c_p, reg, live, tmp_arg1, Arg(1), Arg(3));
+     reg[live] = x(SCRATCH_X_REG);
+     res = erts_bs_append(c_p, reg, live, Size, Arg(1), Arg(3));
      SWAPIN;
      if (is_non_value(res)) {
 	 /* c_p->freason is already set (may be either BADARG or SYSTEM_LIMIT). */
 	 goto lb_Cl_error;
      }
-     StoreBifResult(4, res);
+     StoreBifResult(5, res);
  }
 
  /*
-  * tmp_arg1 = Number of bytes to build
-  * tmp_arg2 = Source binary
-  * Operands: Fail Unit Dst
+  * Operands: Fail Size Src Unit Dst
   */
- OpCase(i_bs_private_append_jId): {
+ OpCase(i_bs_private_append_jIssd): {
      Eterm res;
+     Eterm Size, Src;
 
-     res = erts_bs_private_append(c_p, tmp_arg2, tmp_arg1, Arg(1));
+     GetArg2(2, Size, Src);
+     res = erts_bs_private_append(c_p, Src, Size, Arg(1));
      if (is_non_value(res)) {
 	 /* c_p->freason is already set (may be either BADARG or SYSTEM_LIMIT). */
 	 goto lb_Cl_error;
      }
-     StoreBifResult(2, res);
+     StoreBifResult(4, res);
  }
 
- /*
-  * tmp_arg1 = Initial size of writable binary
-  * Operands: Live Dst
-  */
  OpCase(bs_init_writable): {
      SWAPOUT;
      r(0) = erts_bs_init_writable(c_p, r(0));
@@ -4024,26 +4025,29 @@ do {						\
 
  /*
   * Only used for validating a value matched out. 
-  *
-  * tmp_arg1 = Integer to validate
-  * tmp_arg2 = Match context
   */
- OpCase(i_bs_validate_unicode_retract_j): {
-     /*
-      * There is no need to untag the integer, but it IS necessary
-      * to make sure it is small (a bignum pointer could fall in
-      * the valid range).
-      */
-     if (is_not_small(tmp_arg1) || tmp_arg1 > make_small(0x10FFFFUL) ||
-	 (make_small(0xD800UL) <= tmp_arg1 &&
-	  tmp_arg1 <= make_small(0xDFFFUL))) {
-	 ErlBinMatchBuffer *mb = ms_matchbuffer(tmp_arg2);
+ OpCase(i_bs_validate_unicode_retract_jss): {
+	Eterm i; 		/* Integer to validate */
 
-	 mb->offset -= 32;
-	 goto badarg;
-     }
-     Next(1);
- }
+	/*
+	 * There is no need to untag the integer, but it IS necessary
+	 * to make sure it is small (a bignum pointer could fall in
+	 * the valid range).
+	 */
+
+	GetArg1(1, i);
+	if (is_not_small(i) || i > make_small(0x10FFFFUL) ||
+	    (make_small(0xD800UL) <= i && i <= make_small(0xDFFFUL))) {
+	    Eterm ms;		/* Match context */
+	    ErlBinMatchBuffer* mb;
+
+	    GetArg1(2, ms);
+	    mb = ms_matchbuffer(ms);
+	    mb->offset -= 32;
+	    goto badarg;
+	}
+	Next(3);
+    }
 
  /*
   * Matching of binaries.
@@ -4292,35 +4296,36 @@ do {						\
  }
 
  /*
-  * tmp_arg1 = Match context
-  * tmp_arg2 = Size field
-  * Operands: Fail Live FlagsAndUnit Dst
+  * Operands: Fail Live FlagsAndUnit Ms Sz Dst
   */
- OpCase(i_bs_get_integer_fIId): {
+ OpCase(i_bs_get_integer_fIIssd): {
      Uint flags;
      Uint size;
+     Eterm Ms;
+     Eterm Sz;
      ErlBinMatchBuffer* mb;
      Eterm result;
 
      flags = Arg(2);
-     BsGetFieldSize(tmp_arg2, (flags >> 3), ClauseFail(), size);
+     GetArg2(3, Ms, Sz);
+     BsGetFieldSize(Sz, (flags >> 3), ClauseFail(), size);
      if (size >= SMALL_BITS) {
 	 Uint wordsneeded;
-	 /* check bits size before potential gc.
+	 /* Check bits size before potential gc.
 	  * We do not want a gc and then realize we don't need
-	  * the allocated space (i.e. if the op fails)
+	  * the allocated space (i.e. if the op fails).
 	  *
-	  * remember to reacquire the matchbuffer after gc.
+	  * Remember to re-acquire the matchbuffer after gc.
 	  */
 
-	 mb = ms_matchbuffer(tmp_arg1);
+	 mb = ms_matchbuffer(Ms);
 	 if (mb->size - mb->offset < size) {
 	     ClauseFail();
 	 }
 	 wordsneeded = 1+WSIZE(NBYTES((Uint) size));
-	 TestHeapPreserve(wordsneeded, Arg(1), tmp_arg1);
+	 TestHeapPreserve(wordsneeded, Arg(1), Ms);
      }
-     mb = ms_matchbuffer(tmp_arg1);
+     mb = ms_matchbuffer(Ms);
      LIGHT_SWAPOUT;
      result = erts_bs_get_integer_2(c_p, size, flags, mb);
      LIGHT_SWAPIN;
@@ -4328,7 +4333,7 @@ do {						\
      if (is_non_value(result)) {
 	 ClauseFail();
      }
-     StoreBifResult(3, result);
+     StoreBifResult(5, result);
  }
 
  {
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index c90576333a..871c2b1f55 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2854,23 +2854,16 @@ gen_get_integer2(LoaderState* stp, GenOpArg Fail, GenOpArg Ms, GenOpArg Live,
 	    goto generic;
 	}
     } else {
-	GenOp* op2;
-	NEW_GENOP(stp, op2);
-	
-	op->op = genop_i_fetch_2;
-	op->arity = 2;
-	op->a[0] = Ms;
-	op->a[1] = Size;
-	op->next = op2;
-
-	op2->op = genop_i_bs_get_integer_4;
-	op2->arity = 4;
-	op2->a[0] = Fail;
-	op2->a[1] = Live;
-	op2->a[2].type = TAG_u;
-	op2->a[2].val = (Unit.val << 3) | Flags.val;
-	op2->a[3] = Dst;
-	op2->next = NULL;
+	op->op = genop_i_bs_get_integer_6;
+	op->arity = 6;
+	op->a[0] = Fail;
+	op->a[1] = Live;
+	op->a[2].type = TAG_u;
+	op->a[2].val = (Unit.val << 3) | Flags.val;
+	op->a[3] = Ms;
+	op->a[4] = Size;
+	op->a[5] = Dst;
+	op->next = NULL;
 	return op;
     }
     op->next = NULL;
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 492d4d240a..637cef0a22 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -1122,7 +1122,7 @@ bs_get_integer2 Fail=f Ms=x Live=u Sz=sq Unit=u Flags=u Dst=d => \
 
 i_bs_get_integer_small_imm x I f I d
 i_bs_get_integer_imm x I I f I d
-i_bs_get_integer f I I d
+i_bs_get_integer f I I s s d
 i_bs_get_integer_8 x f d
 i_bs_get_integer_16 x f d
 i_bs_get_integer_32 x f I d
@@ -1195,14 +1195,12 @@ i_bs_get_utf16 x f I d
 
 bs_get_utf32 Fail=f Ms=x Live=u Flags=u Dst=d => \
 	bs_get_integer2 Fail Ms Live i=32 u=1 Flags Dst | \
-	i_fetch Dst Ms | \
-	i_bs_validate_unicode_retract Fail
+	i_bs_validate_unicode_retract Fail Dst Ms
 bs_skip_utf32 Fail=f Ms=x Live=u Flags=u => \
 	bs_get_integer2 Fail Ms Live i=32 u=1 Flags x | \
-	i_fetch x Ms | \
-	i_bs_validate_unicode_retract Fail
+	i_bs_validate_unicode_retract Fail x Ms
 
-i_bs_validate_unicode_retract j
+i_bs_validate_unicode_retract j s s
 %hot
 
 #
@@ -1224,12 +1222,12 @@ bs_init2 Fail Sz=u Words Regs Flags Dst => \
 bs_init2 Fail Sz Words=u==0 Regs Flags Dst => \
   i_bs_init_fail Sz Fail Regs Dst
 bs_init2 Fail Sz Words Regs Flags Dst => \
-  i_fetch Sz x=0 | i_bs_init_fail_heap Words Fail Regs Dst
+  i_bs_init_fail_heap Sz Words Fail Regs Dst
 
 i_bs_init_fail x j I d
 i_bs_init_fail y j I d
 
-i_bs_init_fail_heap I j I d
+i_bs_init_fail_heap s I j I d
 
 i_bs_init I I d
 i_bs_init_heap_bin I I d
@@ -1246,31 +1244,30 @@ bs_init_bits Fail Sz=u Words Regs Flags Dst =>  i_bs_init_bits_heap Sz Words Reg
 bs_init_bits Fail Sz Words=u==0 Regs Flags Dst => \
   i_bs_init_bits_fail Sz Fail Regs Dst
 bs_init_bits Fail Sz Words Regs Flags Dst => \
-  i_fetch Sz x=0 | i_bs_init_bits_fail_heap Words Fail Regs Dst
+  i_bs_init_bits_fail_heap Sz Words Fail Regs Dst
 
 i_bs_init_bits_fail x j I d
 i_bs_init_bits_fail y j I d
 
-i_bs_init_bits_fail_heap I j I d
+i_bs_init_bits_fail_heap s I j I d
 
 i_bs_init_bits I I d
 i_bs_init_bits_heap I I I d
 
 bs_add Fail S1=i==0 S2 Unit=u==1 D => move S2 D
-bs_add Fail S1 S2 Unit D => i_fetch S1 S2 | i_bs_add Fail Unit D
 
-i_bs_add j I d
+bs_add j s s I d
 
 bs_append Fail Size Extra Live Unit Bin Flags Dst => \
-  i_fetch Size Bin | i_bs_append Fail Extra Live Unit Dst
+  move Bin x | i_bs_append Fail Extra Live Unit Size Dst
 
 bs_private_append Fail Size Unit Bin Flags Dst => \
-  i_fetch Size Bin | i_bs_private_append Fail Unit Dst
+  i_bs_private_append Fail Unit Size Bin Dst
 
 bs_init_writable
 
-i_bs_append j I I I d
-i_bs_private_append j I d
+i_bs_append j I I I s d
+i_bs_private_append j I s s d
 
 #
 # Storing integers into binaries.
-- 
cgit v1.2.3


From bf2b06d3edd2c855071529be17c0c35e778a88fa Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 17 Jun 2015 16:33:19 +0200
Subject: Remove the i_fetch instruction

---
 erts/emulator/beam/beam_emu.c |  2 --
 erts/emulator/beam/ops.tab    | 15 ---------------
 2 files changed, 17 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 26e28832cf..fe3d8aea28 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -148,8 +148,6 @@ do {                                     \
    ASSERT(VALID_INSTR(* (Eterm *)(ip))); \
    I = (ip)
 
-#define FetchArgs(S1, S2) tmp_arg1 = (S1); tmp_arg2 = (S2)
-
 /*
  * Store a result into a register given a destination descriptor.
  */
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 637cef0a22..631e10215c 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -486,21 +486,6 @@ put_list c y y
 put_list s s d
 %hot
 
-%macro: i_fetch FetchArgs -pack
-i_fetch c x
-i_fetch c y
-i_fetch x c
-i_fetch x x
-i_fetch x y
-i_fetch y c
-i_fetch y x
-i_fetch y y
-
-%cold
-i_fetch c c
-i_fetch s s
-%hot
-
 #
 # Some more only used by the emulator
 #
-- 
cgit v1.2.3


From 227269c6b37b88fe106d27a5c0ba148c31472822 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Wed, 17 Jun 2015 19:30:20 +0200
Subject: Eliminate use of tmp_arg1 and tmp_arg2 in bit syntax

---
 erts/emulator/beam/beam_emu.c | 81 +++++++++++++++++++------------------------
 1 file changed, 36 insertions(+), 45 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index fe3d8aea28..89234a9a80 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -989,7 +989,6 @@ init_emulator(void)
 #  define REG_I asm("%l4")
 #  define REG_fcalls asm("%l5")
 #  define REG_tmp_arg1 asm("%l6")
-#  define REG_tmp_arg2 asm("%l7")
 #else
 #  define REG_xregs
 #  define REG_htop
@@ -997,7 +996,6 @@ init_emulator(void)
 #  define REG_I
 #  define REG_fcalls
 #  define REG_tmp_arg1
-#  define REG_tmp_arg2
 #endif
 
 #ifdef USE_VM_PROBES
@@ -1145,7 +1143,6 @@ void process_main(void)
      * Temporaries used for picking up arguments for instructions.
      */
     register Eterm tmp_arg1 REG_tmp_arg1 = NIL;
-    register Eterm tmp_arg2 REG_tmp_arg2 = NIL;
 
     /*
      * X registers and floating point registers are located in
@@ -4242,55 +4239,49 @@ do {						\
      }
  }
 
- /* Operands: x(Reg) Size Live Fail Flags Dst */
- OpCase(i_bs_get_integer_imm_xIIfId): {
-     tmp_arg1 = xb(Arg(0));
-     I++;
-     /* Operands: Size Live Fail Flags Dst */
-     goto do_bs_get_integer_imm_test_heap;
- }
+ {
+     Eterm Ms, Sz;
 
- /*
-  * tmp_arg1 = match context
-  * Operands: Size Live Fail Flags Dst
-  */
- do_bs_get_integer_imm_test_heap: {
-     Uint wordsneeded;
-     tmp_arg2 = Arg(0);
-     wordsneeded = 1+WSIZE(NBYTES(tmp_arg2));
-     TestHeapPreserve(wordsneeded, Arg(1), tmp_arg1);
-     I += 2;
-     /* Operands: Fail Flags Dst */
-     goto do_bs_get_integer_imm;
- }
+     /* Operands: x(Reg) Size Live Fail Flags Dst */
+ OpCase(i_bs_get_integer_imm_xIIfId): {
+	 Uint wordsneeded;
+	 Ms = xb(Arg(0));
+	 Sz = Arg(1);
+	 wordsneeded = 1+WSIZE(NBYTES(Sz));
+	 TestHeapPreserve(wordsneeded, Arg(2), Ms);
+	 I += 3;
+	 /* Operands: Fail Flags Dst */
+	 goto do_bs_get_integer_imm;
+     }
 
- /* Operands: x(Reg) Size Fail Flags Dst */
+     /* Operands: x(Reg) Size Fail Flags Dst */
  OpCase(i_bs_get_integer_small_imm_xIfId): {
-     tmp_arg1 = xb(Arg(0));
-     tmp_arg2 = Arg(1);
-     I += 2;
-     /* Operands: Fail Flags Dst */
-     goto do_bs_get_integer_imm;
- }
+	 Ms = xb(Arg(0));
+	 Sz = Arg(1);
+	 I += 2;
+	 /* Operands: Fail Flags Dst */
+	 goto do_bs_get_integer_imm;
+     }
  
- /*
-  * tmp_arg1 = match context
-  * tmp_arg2 = size of field
-  * Operands: Fail Flags Dst
-  */
+     /*
+      * Ms = match context
+      * Sz = size of field
+      * Operands: Fail Flags Dst
+      */
  do_bs_get_integer_imm: {
-     ErlBinMatchBuffer* mb;
-     Eterm result;
+	 ErlBinMatchBuffer* mb;
+	 Eterm result;
 
-     mb = ms_matchbuffer(tmp_arg1);
-     LIGHT_SWAPOUT;
-     result = erts_bs_get_integer_2(c_p, tmp_arg2, Arg(1), mb);
-     LIGHT_SWAPIN;
-     HEAP_SPACE_VERIFIED(0);
-     if (is_non_value(result)) {
-	 ClauseFail();
+	 mb = ms_matchbuffer(Ms);
+	 LIGHT_SWAPOUT;
+	 result = erts_bs_get_integer_2(c_p, Sz, Arg(1), mb);
+	 LIGHT_SWAPIN;
+	 HEAP_SPACE_VERIFIED(0);
+	 if (is_non_value(result)) {
+	     ClauseFail();
+	 }
+	 StoreBifResult(2, result);
      }
-     StoreBifResult(2, result);
  }
 
  /*
-- 
cgit v1.2.3


From dddd225934cfe79c052e9013a8a2c1d6a29c8ae0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 23 Jun 2015 15:24:07 +0200
Subject: Remove the last use of tmp_arg1

---
 erts/emulator/beam/beam_emu.c | 102 +++++++++++++++++++-----------------------
 erts/emulator/beam/ops.tab    |  53 ++++++----------------
 2 files changed, 59 insertions(+), 96 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 89234a9a80..7dbf14de02 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -603,47 +603,39 @@ void** beam_ops;
    H = CAR(tmp_ptr);				\
    T = CDR(tmp_ptr); } while (0)
 
-#define GetTupleElement(Src, Element, Dest)				\
-  do {									\
-    tmp_arg1 = (Eterm) (((unsigned char *) tuple_val(Src)) + (Element));\
-    (Dest) = (*(Eterm *) tmp_arg1);	                		\
-  } while (0)
-
-#define ExtractNextElement(Dest)					  \
-    tmp_arg1 += sizeof(Eterm);						  \
-    (Dest) = (* (Eterm *) (((unsigned char *) tmp_arg1)))
-
-#define ExtractNextElement2(Dest)				\
-  do {								\
-    Eterm* ene_dstp = &(Dest);					\
-    ene_dstp[0] = ((Eterm *) tmp_arg1)[1];	\
-    ene_dstp[1] = ((Eterm *) tmp_arg1)[2];	\
-    tmp_arg1 += sizeof(Eterm) + sizeof(Eterm);			\
-  } while (0)
-
-#define ExtractNextElement3(Dest)		\
-  do {						\
-    Eterm* ene_dstp = &(Dest);			\
-    ene_dstp[0] = ((Eterm *) tmp_arg1)[1];	\
-    ene_dstp[1] = ((Eterm *) tmp_arg1)[2];	\
-    ene_dstp[2] = ((Eterm *) tmp_arg1)[3];	\
-    tmp_arg1 += 3*sizeof(Eterm);		\
+#define GetTupleElement(Src, Element, Dest)		\
+  do {							\
+    Eterm* src;						\
+    src = ADD_BYTE_OFFSET(tuple_val(Src), (Element));	\
+    (Dest) = *src;					\
   } while (0)
 
-#define ExtractNextElement4(Dest)		\
-  do {						\
-    Eterm* ene_dstp = &(Dest);			\
-    ene_dstp[0] = ((Eterm *) tmp_arg1)[1];	\
-    ene_dstp[1] = ((Eterm *) tmp_arg1)[2];	\
-    ene_dstp[2] = ((Eterm *) tmp_arg1)[3];	\
-    ene_dstp[3] = ((Eterm *) tmp_arg1)[4];	\
-    tmp_arg1 += 4*sizeof(Eterm);		\
+#define GetTupleElement2(Src, Element, Dest)		\
+  do {							\
+    Eterm* src;						\
+    Eterm* dst;						\
+    Eterm E1, E2;					\
+    src = ADD_BYTE_OFFSET(tuple_val(Src), (Element));	\
+    dst = &(Dest);					\
+    E1 = src[0];					\
+    E2 = src[1];					\
+    dst[0] = E1;					\
+    dst[1] = E2;					\
   } while (0)
 
-#define ExtractElement(Element, Dest)		\
-  do {						\
-     tmp_arg1 += (Element);			\
-     (Dest) = (* (Eterm *) tmp_arg1);		\
+#define GetTupleElement3(Src, Element, Dest)		\
+  do {							\
+    Eterm* src;						\
+    Eterm* dst;						\
+    Eterm E1, E2, E3;					\
+    src = ADD_BYTE_OFFSET(tuple_val(Src), (Element));	\
+    dst = &(Dest);					\
+    E1 = src[0];					\
+    E2 = src[1];					\
+    E3 = src[2];					\
+    dst[0] = E1;					\
+    dst[1] = E2;					\
+    dst[2] = E3;					\
   } while (0)
 
 #define EqualImmed(X, Y, Action) if (X != Y) { Action; }
@@ -683,11 +675,9 @@ void** beam_ops;
 
 #define IsTuple(X, Action) if (is_not_tuple(X)) Action
 
-#define IsArity(Pointer, Arity, Fail)					  \
-    if (*(Eterm *)							  \
-	(tmp_arg1 = (Eterm) (tuple_val(Pointer))) != (Arity))             \
-    { 									  \
-        Fail; 								  \
+#define IsArity(Pointer, Arity, Fail)		\
+    if (*tuple_val(Pointer) != (Arity)) {	\
+        Fail;					\
     }
 
 #define IsMap(Src, Fail) if (!is_map(Src)) { Fail; }
@@ -724,14 +714,21 @@ void** beam_ops;
      }							\
   } while (0)
 
-#define IsTupleOfArity(Src, Arity, Fail)				      \
-  do {									      \
-    if (is_not_tuple(Src) || 						      \
-	*(Eterm *)							      \
-	(tmp_arg1 = (Eterm) (tuple_val(Src))) != Arity) {                     \
-        Fail;								      \
-    }									      \
+#ifdef DEBUG
+#define IsTupleOfArity(Src, Arityval, Fail)			\
+  do {								\
+    if (!(is_tuple(Src) && *tuple_val(Src) == Arityval)) {	\
+        Fail;							\
+    }								\
   } while (0)
+#else
+#define IsTupleOfArity(Src, Arityval, Fail)			\
+  do {								\
+    if (!(is_boxed(Src) && *tuple_val(Src) == Arityval)) {	\
+        Fail;							\
+    }								\
+  } while (0)
+#endif
 
 #define IsBoolean(X, Fail) if ((X) != am_true && (X) != am_false) { Fail; }
 
@@ -988,14 +985,12 @@ init_emulator(void)
 #  define REG_stop asm("%l3")
 #  define REG_I asm("%l4")
 #  define REG_fcalls asm("%l5")
-#  define REG_tmp_arg1 asm("%l6")
 #else
 #  define REG_xregs
 #  define REG_htop
 #  define REG_stop
 #  define REG_I
 #  define REG_fcalls
-#  define REG_tmp_arg1
 #endif
 
 #ifdef USE_VM_PROBES
@@ -1139,11 +1134,6 @@ void process_main(void)
      */
     register Sint FCALLS REG_fcalls = 0;
 
-    /*
-     * Temporaries used for picking up arguments for instructions.
-     */
-    register Eterm tmp_arg1 REG_tmp_arg1 = NIL;
-
     /*
      * X registers and floating point registers are located in
      * scheduler specific data.
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 631e10215c..853ba82f83 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -230,6 +230,12 @@ i_get_tuple_element x P y
 i_get_tuple_element y P y
 %hot
 
+%macro: i_get_tuple_element2 GetTupleElement2 -pack
+i_get_tuple_element2 x P x
+
+%macro: i_get_tuple_element3 GetTupleElement3 -pack
+i_get_tuple_element3 x P x
+
 %macro: is_number IsNumber -fail_action
 %cold
 is_number f x
@@ -552,48 +558,15 @@ test_arity Fail=f c Arity => jump Fail
 test_arity f x A
 test_arity f y A
 
-is_tuple_of_arity Fail=f Reg Arity | get_tuple_element Reg P=u==0 Dst=xy => \
-  is_tuple_of_arity Fail Reg Arity | extract_next_element Dst | original_reg Reg P
-
-test_arity Fail Reg Arity | get_tuple_element Reg P=u==0 Dst=xy => \
-  test_arity Fail Reg Arity | extract_next_element Dst | original_reg Reg P
-
-original_reg Reg P1 | get_tuple_element Reg P2 Dst=xy | succ(P1, P2) => \
-  extract_next_element Dst | original_reg Reg P2
-
-get_tuple_element Reg P Dst => i_get_tuple_element Reg P Dst | original_reg Reg P
-
-original_reg Reg Pos =>
-
-original_reg/2
-
-extract_next_element D1=xy | original_reg Reg P1 | get_tuple_element Reg P2 D2=xy | \
-succ(P1, P2) | succ(D1, D2) => \
-  extract_next_element2 D1 | original_reg Reg P2
-
-extract_next_element2 D1=xy | original_reg Reg P1 | get_tuple_element Reg P2 D2=xy | \
-succ(P1, P2) | succ2(D1, D2) => \
-  extract_next_element3 D1 | original_reg Reg P2
-
-#extract_next_element3 D1=xy | original_reg Reg P1 | get_tuple_element Reg P2 D2=xy | \
-#succ(P1, P2) | succ3(D1, D2) => \
-#  extract_next_element4 D1 | original_reg Reg P2
-
-%macro: extract_next_element ExtractNextElement -pack
-extract_next_element x
-extract_next_element y
-
-%macro: extract_next_element2 ExtractNextElement2 -pack
-extract_next_element2 x
-extract_next_element2 y
+get_tuple_element Reg=x P1 D1=x | get_tuple_element Reg=x P2 D2=x | \
+   get_tuple_element Reg=x P3 D3=x | \
+   succ(P1, P2) | succ(P2, P3) | \
+   succ(D1, D2) | succ(D2, D3) => i_get_tuple_element3 Reg P1 D1
 
-%macro: extract_next_element3 ExtractNextElement3 -pack
-extract_next_element3 x
-extract_next_element3 y
+get_tuple_element Reg=x P1 D1=x | get_tuple_element Reg=x P2 D2=x | \
+   succ(P1, P2) | succ(D1, D2) => i_get_tuple_element2 Reg P1 D1
 
-#%macro: extract_next_element4 ExtractNextElement4 -pack
-#extract_next_element4 x
-#extract_next_element4 y
+get_tuple_element Reg P Dst => i_get_tuple_element Reg P Dst
 
 is_integer Fail=f i =>
 is_integer Fail=f an => jump Fail
-- 
cgit v1.2.3


From d252130b8e880f2f7f826217fe806da76fbcbb7c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 10 Apr 2012 12:28:11 +0200
Subject: Rewrite the hipe_mode_switch instructions

The 'cmd' variable that were shared by several hipe_mode_switch
instructions would cause clang to produce sub-optimal code,
probably because it considered the instructions as part of of
loop that needed to be optimized.

What would was that 'cmd' would be assigned to the ESI register
(lower 32 bits of the RSI register). It would use ESI for other
purposes in instructions, but at the end of every instruction
it would set ESI to 1 just in case the next instruction happened
to be hipe_trap_return. This can be seen clearly if this commit
is omitted and the define HIPE_MODE_SWITCH_CMD_RETURN in
hipe/hipe_mode_switch.h is changed from 1 to some other number
such as 42. You will see that 42 is assigned to ESI at the end
of every instruction.

Eliminate this problem by elimininating the shared 'cmd' variable.
---
 erts/emulator/beam/beam_emu.c | 32 +++++++++++++++-----------------
 1 file changed, 15 insertions(+), 17 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 7dbf14de02..362c6e4826 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -4681,7 +4681,12 @@ do {						\
 
 #ifdef HIPE
  {
-     unsigned cmd;
+#define HIPE_MODE_SWITCH(Cmd)			\
+     SWAPOUT;					\
+     c_p->fcalls = FCALLS;			\
+     c_p->def_arg_reg[4] = -neg_o_reds;		\
+     c_p = hipe_mode_switch(c_p, Cmd, reg);     \
+     goto L_post_hipe_mode_switch
 
      OpCase(hipe_trap_call): {
 	 /*
@@ -4695,38 +4700,31 @@ do {						\
 	  */
 	 ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI));
 	 c_p->hipe.u.ncallee = (void(*)(void)) I[-4];
-	 cmd = HIPE_MODE_SWITCH_CMD_CALL | (I[-1] << 8);
 	 ++hipe_trap_count;
-	 goto L_hipe_mode_switch;
+	 HIPE_MODE_SWITCH(HIPE_MODE_SWITCH_CMD_CALL | (I[-1] << 8));
      }
      OpCase(hipe_trap_call_closure): {
        ASSERT(I[-5] == (Uint) OpCode(i_func_info_IaaI));
        c_p->hipe.u.ncallee = (void(*)(void)) I[-4];
-       cmd = HIPE_MODE_SWITCH_CMD_CALL_CLOSURE | (I[-1] << 8);
        ++hipe_trap_count;
-       goto L_hipe_mode_switch;
+       HIPE_MODE_SWITCH(HIPE_MODE_SWITCH_CMD_CALL_CLOSURE | (I[-1] << 8));
      }
      OpCase(hipe_trap_return): {
-	 cmd = HIPE_MODE_SWITCH_CMD_RETURN;
-	 goto L_hipe_mode_switch;
+	 HIPE_MODE_SWITCH(HIPE_MODE_SWITCH_CMD_RETURN);
      }
      OpCase(hipe_trap_throw): {
-	 cmd = HIPE_MODE_SWITCH_CMD_THROW;
-	 goto L_hipe_mode_switch;
+	 HIPE_MODE_SWITCH(HIPE_MODE_SWITCH_CMD_THROW);
      }
      OpCase(hipe_trap_resume): {
-	 cmd = HIPE_MODE_SWITCH_CMD_RESUME;
-	 goto L_hipe_mode_switch;
+	 HIPE_MODE_SWITCH(HIPE_MODE_SWITCH_CMD_RESUME);
      }
- L_hipe_mode_switch:
-     /* XXX: this abuse of def_arg_reg[] is horrid! */
-     SWAPOUT;
-     c_p->fcalls = FCALLS;
-     c_p->def_arg_reg[4] = -neg_o_reds;
-     c_p = hipe_mode_switch(c_p, cmd, reg);
+#undef HIPE_MODE_SWITCH
+
+ L_post_hipe_mode_switch:
      reg = ERTS_PROC_GET_SCHDATA(c_p)->x_reg_array;
      freg = ERTS_PROC_GET_SCHDATA(c_p)->f_reg_array;
      ERL_BITS_RELOAD_STATEP(c_p);
+     /* XXX: this abuse of def_arg_reg[] is horrid! */
      neg_o_reds = -c_p->def_arg_reg[4];
      FCALLS = c_p->fcalls;
      SWAPIN;
-- 
cgit v1.2.3


From 1f73d45327bb13a615f2f0a8d9d4888ddacb95a5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 23 Jun 2015 09:38:23 +0200
Subject: Add back frequently used x(0) instructions

---
 erts/emulator/beam/beam_emu.c | 10 ++++++++++
 erts/emulator/beam/ops.tab    | 37 +++++++++++++++++++++++++++++++++----
 2 files changed, 43 insertions(+), 4 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 362c6e4826..bac7057abe 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -1298,6 +1298,11 @@ void process_main(void)
 	    Uint live;
 	    Eterm result;
 
+	OpCase(i_increment_rIId):
+	    increment_reg_val = x(0);
+	    I--;
+	    goto do_increment;
+
 	OpCase(i_increment_xIId):
 	    increment_reg_val = xb(Arg(0));
 	    goto do_increment;
@@ -1357,6 +1362,11 @@ void process_main(void)
      PlusOp2 = xb(Arg(3));
      goto do_plus;
 
+ OpCase(i_plus_jIxyd):
+     PlusOp1 = xb(Arg(2));
+     PlusOp2 = yb(Arg(3));
+     goto do_plus;
+
  OpCase(i_plus_jIssd):
      GetArg2(2, PlusOp1, PlusOp2);
      goto do_plus;
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 853ba82f83..4ab08faae3 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -204,6 +204,14 @@ get_list y x y
 get_list y y x
 get_list y y y
 
+# The following get_list instructions using x(0) are frequently used.
+get_list r x x
+get_list r r y
+get_list x r x
+get_list r x y
+get_list r y r
+get_list r x r
+
 # Old-style catch.
 catch y f
 catch_end y
@@ -348,6 +356,14 @@ move c x
 move n x
 move y y
 
+# The following move instructions using x(0) are frequently used.
+
+move x r
+move r x
+move y r
+move c r
+move r y
+
 # Receive operations.
 
 loop_rec Fail x==0 | smp_mark_target_label(Fail) => i_loop_rec Fail
@@ -390,6 +406,7 @@ is_ne_exact Lbl R=xy C=ian => i_is_ne_exact_immed Lbl R C
 is_ne_exact Lbl R=xy C=q => i_is_ne_exact_literal Lbl R C
 
 %macro: i_is_eq_exact_immed EqualImmed -fail_action
+i_is_eq_exact_immed f r c
 i_is_eq_exact_immed f x c
 i_is_eq_exact_immed f y c
 
@@ -478,15 +495,21 @@ put_list x c x
 put_list x c y
 
 put_list y c x
-put_list y c y
 
 # put_list Constant SrcReg Dst
 
 put_list c x x
-put_list c x y
-
 put_list c y x
-put_list c y y
+
+# The following put_list instructions using x(0) are frequently used.
+
+put_list y r r
+put_list x r r
+put_list r n r
+put_list r n x
+put_list r x x
+put_list r x r
+put_list x x r
 
 %cold
 put_list s s d
@@ -544,10 +567,12 @@ is_tuple Fail=f S=xy | test_arity Fail=f S=xy Arity => is_tuple_of_arity Fail S
 
 %macro:is_tuple_of_arity IsTupleOfArity -fail_action
 
+is_tuple_of_arity f r A
 is_tuple_of_arity f x A
 is_tuple_of_arity f y A
 
 %macro: is_tuple IsTuple -fail_action
+is_tuple f r
 is_tuple f x
 is_tuple f y
 
@@ -593,6 +618,7 @@ is_list f y
 is_nonempty_list Fail=f S=x | allocate Need Rs => is_nonempty_list_allocate Fail S Need Rs
 
 %macro:is_nonempty_list_allocate IsNonemptyListAllocate -fail_action -pack
+is_nonempty_list_allocate f r I t
 is_nonempty_list_allocate f x I t
 
 is_nonempty_list F=f x==0 | test_heap I1 I2 => is_non_empty_list_test_heap F I1 I2
@@ -1440,6 +1466,7 @@ gen_minus p Live Reg=d Int=i Dst | negation_is_small(Int) => \
 # GCing arithmetic instructions.
 #
 
+gen_plus Fail Live Y=y X=x Dst => i_plus Fail Live X Y Dst
 gen_plus Fail Live S1 S2 Dst => i_plus Fail Live S1 S2 Dst
 
 gen_minus Fail Live S1 S2 Dst => i_minus Fail Live S1 S2 Dst
@@ -1471,10 +1498,12 @@ gc_bif2 Fail Live u$bif:erlang:bxor/2 S1 S2 Dst => \
 
 gc_bif1 Fail I u$bif:erlang:bnot/1 Src Dst=d => i_int_bnot Fail Src I Dst
 
+i_increment r I I d
 i_increment x I I d
 i_increment y I I d
 
 i_plus j I x x d
+i_plus j I x y d
 i_plus j I s s d
 
 i_minus j I x x d
-- 
cgit v1.2.3


From 3dc2bee53b4d36f41821a6ab512cf01c958c11f9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Tue, 23 Jun 2015 17:31:49 +0200
Subject: Introduce specialized versions of move2

Currently, move2/2 does the two moves sequentially to ensure
that the instruction will always work correctly.

We can do better than that. If the two move instructions have
any registers in common, we can introduce simpler and slightly
more efficient instructions to handle those cases:

  move_shift/3
  move_dup/3

For the remaining cases when the the move instructions
have no common registers, the move2/4 instruction can perform
the moves in parallel which is probably slightly more efficient.
For clarity's sake, we will remain the instruction to move2_par/4.
---
 erts/emulator/beam/beam_emu.c | 18 ++++++++++-
 erts/emulator/beam/ops.tab    | 71 +++++++++++++++++++++++++++++++------------
 2 files changed, 68 insertions(+), 21 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index bac7057abe..2c10e7ae7c 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -551,7 +551,23 @@ void** beam_ops;
        Store(term, Dst);           \
    } while (0)
 
-#define Move2(S1, D1, S2, D2) D1 = (S1); D2 = (S2)
+#define Move2Par(S1, D1, S2, D2)		\
+  do {						\
+      Eterm V1, V2;				\
+      V1 = (S1); V2 = (S2); D1 = V1; D2 = V2;	\
+  } while (0)
+
+#define MoveShift(Src, SD, D)			\
+  do {						\
+    Eterm V;					\
+    V = Src; D = SD; SD = V;			\
+  } while (0)
+
+#define MoveDup(Src, D1, D2)			\
+  do {						\
+    D1 = D2 = (Src);				\
+  } while (0)
+
 #define Move3(S1, D1, S2, D2, S3, D3) D1 = (S1); D2 = (S2); D3 = (S3)
 
 #define MoveReturn(Src)				\
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 4ab08faae3..97525c3b72 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -303,21 +303,40 @@ 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 Src=x D1=x | move Src=x D2=x => move_dup Src D1 D2
+move Src=x SD=x | move SD=x D=x   => move_dup Src SD D
+move Src=x D1=x | move Src=x D2=y => move_dup Src D1 D2
+move Src=y SD=x | move SD=x D=y   => move_dup Src SD D
+move Src=x SD=x | move SD=x D=y   => move_dup Src SD D
+move Src=y SD=x | move SD=x D=x   => move_dup Src SD D
 
-move X1=x X2=x | move X3=x Y1=y => move2 X1 X2 X3 Y1
+move SD=x D=x | move Src=xy SD=x => move_shift Src SD D
+move SD=y D=x | move Src=x  SD=y => move_shift Src SD D
+move SD=x D=y | move Src=x  SD=x => move_shift Src SD D
 
-move S1=x S2=x | move X1=x Y1=y => move2 S1 S2 X1 Y1
-move S1=y S2=x | move X1=x Y1=y => move2 S1 S2 X1 Y1
+# The transformations above guarantee that the source for
+# the second move is not the same as the destination for
+# the first move. That means that we can do the moves in
+# parallel (fetch both values, then store them) which could
+# be faster.
 
-move Y1=y X1=x | move S1=x D1=x => move2 Y1 X1 S1 D1
-move S1=x D1=x | move Y1=y X1=x => move2 S1 D1 Y1 X1
+move X1=x Y1=y | move X2=x Y2=y => move2_par X1 Y1 X2 Y2
+move Y1=y X1=x | move Y2=y X2=x => move2_par Y1 X1 Y2 X2
 
-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 X1=x X2=x | move X3=x X4=x => move2_par X1 X2 X3 X4
+
+move X1=x X2=x | move X3=x Y1=y => move2_par X1 X2 X3 Y1
+
+move S1=x S2=x | move X1=x Y1=y => move2_par S1 S2 X1 Y1
+
+move S1=y S2=x | move X1=x Y1=y => move2_par S1 S2 X1 Y1
+
+move Y1=y X1=x | move S1=x D1=x => move2_par Y1 X1 S1 D1
+move S1=x D1=x | move Y1=y X1=x => move2_par S1 D1 Y1 X1
+
+move2_par X1=x Y1=y X2=x Y2=y | move X3=x Y3=y => move3 X1 Y1 X2 Y2 X3 Y3
+move2_par Y1=y X1=x Y2=y X2=x | move Y3=y X3=x => move3 Y1 X1 Y2 X2 Y3 X3
+move2_par 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
@@ -325,18 +344,30 @@ move C=aiq X=x==2 => move_x2 C
 move_x1 c
 move_x2 c
 
-%macro: move2 Move2 -pack
-move2 x y x y
-move2 y x y x
-move2 x x x x
+%macro: move_shift MoveShift -pack
+move_shift x x x
+move_shift y x x
+move_shift x y x
+move_shift x x y
+
+%macro: move_dup MoveDup -pack
+move_dup x x x
+move_dup x x y
+move_dup y x x
+move_dup y x y
+
+%macro: move2_par Move2Par -pack
+
+move2_par x y x y
+move2_par y x y x
+move2_par x x x x
 
-move2 x x x y
+move2_par x x x y
 
-move2 x y x y
-move2 y x x y
+move2_par y x x y
 
-move2 x x y x
-move2 y x x x
+move2_par x x y x
+move2_par y x x x
 
 %macro: move3 Move3
 move3 x y x y x y
-- 
cgit v1.2.3


From 62473daf8169a04a07409f344d938bc51a4536c3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Thu, 25 Jun 2015 13:21:14 +0200
Subject: Introduce swap_temp/3 and swap/2

Sequences of three move instructionst that effectively swap the
contents of two registers are fairly common. We can replace them
with a swap_temp/3 instruction. The third operand is the temporary
register to be used for swapping, since the temporary register
may actually be used.

If swap_temp/3 instruction is followed by a call, the temporary
register will often (but not always) be killed by the call. If
it is killed, we can replace the swap_temp/3 instruction with a
slightly cheaper swap/2 instruction.
---
 erts/emulator/beam/beam_emu.c  | 14 ++++++++++++++
 erts/emulator/beam/beam_load.c |  6 ++++++
 erts/emulator/beam/ops.tab     | 28 ++++++++++++++++++++++++++++
 3 files changed, 48 insertions(+)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 2c10e7ae7c..4e6e7aa48a 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -545,6 +545,20 @@ void** beam_ops;
    HTOP += 2;                      \
   } while (0)
 
+#define Swap(R1, R2)				\
+  do {						\
+    Eterm V = R1;				\
+    R1 = R2;					\
+    R2 = V;					\
+  } while (0)
+
+#define SwapTemp(R1, R2, Tmp)			\
+  do {						\
+    Eterm V = R1;				\
+    R1 = R2;					\
+    R2 = Tmp = V;				\
+  } while (0)
+
 #define Move(Src, Dst, Store)      \
    do {                            \
        Eterm term = (Src);         \
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 871c2b1f55..1e3690fcd6 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2702,6 +2702,12 @@ same_label(LoaderState* stp, GenOpArg Target, GenOpArg Label)
 	Target.val == Label.val;
 }
 
+static int
+is_killed(LoaderState* stp, GenOpArg Reg, GenOpArg Live)
+{
+    return Reg.type == TAG_x && Live.type == TAG_u &&
+	Live.val <= Reg.val;
+}
 
 /*
  * Generate an instruction for element/2.
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 97525c3b72..386de0a658 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -303,6 +303,34 @@ move_window3 x x x y
 move_window4 x x x x y
 move_window5 x x x x x y
 
+# Swap registers.
+move R1=x Tmp=x | move R2=xy R1 | move Tmp R2 => swap_temp R1 R2 Tmp
+
+swap_temp R1 R2 Tmp | line Loc | apply Live | is_killed(Tmp, Live) => \
+  swap R1 R2 | line Loc | apply Live
+
+swap_temp R1 R2 Tmp | line Loc | call Live Addr | is_killed(Tmp, Live) => \
+  swap R1 R2 | line Loc | call Live Addr
+swap_temp R1 R2 Tmp | call_only Live Addr | \
+  is_killed(Tmp, Live) => swap R1 R2 | call_only Live Addr
+swap_temp R1 R2 Tmp | call_last Live Addr D | \
+  is_killed(Tmp, Live) => swap R1 R2 | call_last Live Addr D
+
+swap_temp R1 R2 Tmp | line Loc | call_ext Live Addr | is_killed(Tmp, Live) => \
+  swap R1 R2 | line Loc | call_ext Live Addr
+swap_temp R1 R2 Tmp | line Loc | call_ext_only Live Addr | \
+  is_killed(Tmp, Live) => swap R1 R2 | line Loc | call_ext_only Live Addr
+swap_temp R1 R2 Tmp | line Loc | call_ext_last Live Addr D | \
+  is_killed(Tmp, Live) => swap R1 R2 | line Loc | call_ext_last Live Addr D
+
+%macro: swap_temp SwapTemp -pack
+swap_temp x x x
+swap_temp x y x
+
+%macro: swap Swap -pack
+swap x x
+swap x y
+
 move Src=x D1=x | move Src=x D2=x => move_dup Src D1 D2
 move Src=x SD=x | move SD=x D=x   => move_dup Src SD D
 move Src=x D1=x | move Src=x D2=y => move_dup Src D1 D2
-- 
cgit v1.2.3


From 5e83d2ae89c731bd7e4c32b550150336008d2974 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Thu, 25 Jun 2015 16:18:18 +0200
Subject: Use a cheaper tag scheme for 'd' operands

Since 'd' operands can only either an X register or an Y register,
we only need a single bit to distinguish them. Furthermore, we can
pre-multiply the register number with the word size to speed up
address calculation.
---
 erts/emulator/beam/beam_debug.c | 13 ++++------
 erts/emulator/beam/beam_emu.c   | 57 ++++++++++++++++++-----------------------
 erts/emulator/beam/beam_load.c  |  4 +--
 erts/emulator/beam/beam_load.h  |  1 +
 4 files changed, 33 insertions(+), 42 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 30e3765502..ea1f2cd012 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -475,15 +475,12 @@ print_op(int to, void *to_arg, int op, int size, BeamInstr* addr)
 	    ap++;
 	    break;
 	case 'd':		/* Destination (x(0), x(N), y(N)) */
-	    switch (loader_tag(*ap)) {
-	    case LOADER_X_REG:
-		erts_print(to, to_arg, "x(%d)",
-			   loader_x_reg_index(*ap));
-		break;
-	    case LOADER_Y_REG:
+	    if (*ap & 1) {
 		erts_print(to, to_arg, "y(%d)",
-			   loader_y_reg_index(*ap) - CP_SIZE);
-		break;
+			   *ap / sizeof(Eterm) - CP_SIZE);
+	    } else {
+		erts_print(to, to_arg, "x(%d)",
+			   *ap / sizeof(Eterm));
 	    }
 	    ap++;
 	    break;
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 4e6e7aa48a..60a399a5ac 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -148,23 +148,21 @@ do {                                     \
    ASSERT(VALID_INSTR(* (Eterm *)(ip))); \
    I = (ip)
 
+/*
+ * Register target (X or Y register).
+ */
+#define REG_TARGET(Target) (*(((Target) & 1) ? &yb(Target-1) : &xb(Target)))
+
 /*
  * Store a result into a register given a destination descriptor.
  */
 
-#define StoreResult(Result, DestDesc)			\
-  do {							\
-    Eterm stb_reg;					\
-    stb_reg = (DestDesc);				\
-    CHECK_TERM(Result);					\
-    switch (loader_tag(stb_reg)) {			\
-    case LOADER_X_REG:					\
-      x(loader_x_reg_index(stb_reg)) = (Result);	\
-      break;						\
-    default:						\
-      y(loader_y_reg_index(stb_reg)) = (Result);	\
-      break;						\
-    }							\
+#define StoreResult(Result, DestDesc)		\
+  do {						\
+    Eterm stb_reg;				\
+    stb_reg = (DestDesc);			\
+    CHECK_TERM(Result);				\
+    REG_TARGET(stb_reg) = (Result);		\
   } while (0)
 
 #define StoreSimpleDest(Src, Dest) Dest = (Src)
@@ -175,22 +173,16 @@ do {                                     \
  * be just before the next instruction.
  */
  
-#define StoreBifResult(Dst, Result)			\
-  do {							\
-    BeamInstr* stb_next;				\
-    Eterm stb_reg;					\
-    stb_reg = Arg(Dst);					\
-    I += (Dst) + 2;					\
-    stb_next = (BeamInstr *) *I;			\
-    CHECK_TERM(Result);					\
-    switch (loader_tag(stb_reg)) {			\
-    case LOADER_X_REG:					\
-      x(loader_x_reg_index(stb_reg)) = (Result);	\
-      Goto(stb_next);					\
-    default:						\
-      y(loader_y_reg_index(stb_reg)) = (Result);	\
-      Goto(stb_next);					\
-    }							\
+#define StoreBifResult(Dst, Result)		\
+  do {						\
+    BeamInstr* stb_next;			\
+    Eterm stb_reg;				\
+    stb_reg = Arg(Dst);				\
+    I += (Dst) + 2;				\
+    stb_next = (BeamInstr *) *I;		\
+    CHECK_TERM(Result);				\
+    REG_TARGET(stb_reg) = (Result);		\
+    Goto(stb_next);				\
   } while (0)
 
 #define ClauseFail() goto jump_f
@@ -3277,7 +3269,8 @@ do {						\
      Eterm* p;
      
      PreFetch(3, next);
-     GetArg2(0, element, tuple);
+     GetArg1(0, element);
+     tuple = REG_TARGET(Arg(1));
      ASSERT(is_tuple(tuple));
      p = (Eterm *) ((unsigned char *) tuple_val(tuple) + Arg(2));
      *p = element;
@@ -4605,7 +4598,7 @@ do {						\
      BeamInstr *next;
 
      PreFetch(2, next);
-     GetR(0, targ1);
+     targ1 = REG_TARGET(Arg(0));
      /* Arg(0) == HEADER_FLONUM */
      GET_DOUBLE(targ1, *(FloatDef*)ADD_BYTE_OFFSET(freg, fr));
      NextPF(2, next);
@@ -4625,7 +4618,7 @@ do {						\
      Eterm fr = Arg(1);
      BeamInstr *next;
 
-     GetR(0, targ1);
+     targ1 = REG_TARGET(Arg(0));
      PreFetch(2, next);
      if (is_small(targ1)) {
 	 fb(fr) = (double) signed_val(targ1);
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 1e3690fcd6..a9d47eb7b0 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2227,10 +2227,10 @@ load_code(LoaderState* stp)
 	    case 'd':	/* Destination (x(0), x(N), y(N) */
 		switch (tag) {
 		case TAG_x:
-		    code[ci++] = make_loader_x_reg(tmp_op->a[arg].val);
+		    code[ci++] = tmp_op->a[arg].val * sizeof(Eterm);
 		    break;
 		case TAG_y:
-		    code[ci++] = make_loader_y_reg(tmp_op->a[arg].val);
+		    code[ci++] = tmp_op->a[arg].val * sizeof(Eterm) + 1;
 		    break;
 		default:
 		    LoadError1(stp, "bad tag %d for destination",
diff --git a/erts/emulator/beam/beam_load.h b/erts/emulator/beam/beam_load.h
index d5af634fad..eedb5ee4cd 100644
--- a/erts/emulator/beam/beam_load.h
+++ b/erts/emulator/beam/beam_load.h
@@ -52,6 +52,7 @@ extern BeamInstr* em_call_error_handler;
 extern BeamInstr* em_apply_bif;
 extern BeamInstr* em_call_nif;
 
+
 /*
  * The following variables keep a sorted list of address ranges for
  * each module.  It allows us to quickly find a function given an
-- 
cgit v1.2.3


From a52bd8a9ebece53c67024e3ad17a899c22cea2ec Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Mon, 29 Jun 2015 15:56:44 +0200
Subject: Ensure that the move_call_ext_{last,only} instructions are used

Update transformations to ensure that the move_call_ext_last
and move_call_ext_last are used.
---
 erts/emulator/beam/ops.tab | 23 ++++++++---------------
 1 file changed, 8 insertions(+), 15 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 386de0a658..18202a4200 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -76,17 +76,6 @@ return
 # with the following call instruction, we need to make sure that
 # there is no line/1 instruction between the move and the call.
 #
-
-move S X0=x==0 | line Loc | call_ext Ar Func => \
-     line Loc | move S X0 | call_ext Ar Func
-move S X0=x==0 | line Loc | call_ext_last Ar Func=u$is_bif D => \
-     line Loc | move S X0 | call_ext_last Ar Func D
-move S X0=x==0 | line Loc | call_ext_only Ar Func=u$is_bif => \
-     line Loc | move S X0 | call_ext_only Ar Func
-move S X0=x==0 | line Loc | call Ar Func => \
-     line Loc | move S X0 | call Ar Func
-
-#
 # A tail-recursive call to an external function (non-BIF) will
 # never be saved on the stack, so there is no reason to keep
 # the line instruction. (The compiler did not remove the line
@@ -94,10 +83,14 @@ move S X0=x==0 | line Loc | call Ar Func => \
 # BIFs and ordinary Erlang functions.)
 #
 
-line Loc | call_ext_last Ar Func=u$is_not_bif D => \
-     call_ext_last Ar Func D
-line Loc | call_ext_only Ar Func=u$is_not_bif => \
-     call_ext_only Ar Func
+move S X0=x==0 | line Loc | call_ext Ar Func => \
+     line Loc | move S X0 | call_ext Ar Func
+move S X0=x==0 | line Loc | call_ext_last Ar Func=u$is_not_bif D => \
+     move S X0 | call_ext_last Ar Func D
+move S X0=x==0 | line Loc | call_ext_only Ar Func=u$is_not_bif => \
+     move S X0 | call_ext_only Ar Func
+move S X0=x==0 | line Loc | call Ar Func => \
+     line Loc | move S X0 | call Ar Func
 
 line Loc | func_info M F A => func_info M F A | line Loc
 
-- 
cgit v1.2.3


From fad052472def54fff1268b21313b15bd666437c3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Sun, 21 Jun 2015 18:09:52 +0200
Subject: 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.
---
 erts/emulator/beam/beam_emu.c | 88 +++++++++++++++++++++----------------------
 erts/emulator/beam/ops.tab    |  6 ++-
 2 files changed, 47 insertions(+), 47 deletions(-)

(limited to 'erts/emulator/beam')

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
-- 
cgit v1.2.3


From 6d6fce257cf666113e982c0887121739dff5bf1d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Sun, 1 Apr 2012 09:28:34 +0200
Subject: Eliminate prefetch for conditional instructions

Not pre-fetching in conditional instructions (instructions that use
-fail_action) seems to improve performance slightly.

The reason for that is that conditional instructions may jump to the
failure label, wasting the pre-fetched instruction. Another reason
is that that many conditional instructions do a function call, and
the register pressure is therefore high. Avoiding the pre-fetch
may reduce the register pressure and pontentially result in more
efficient code.
---
 erts/emulator/beam/ops.tab | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index bce6bacd5c..3d52da77d3 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -477,7 +477,7 @@ i_is_ne_exact_literal f x c
 i_is_ne_exact_literal f y c
 
 is_eq_exact Lbl Y=y X=x => is_eq_exact Lbl X Y
-%macro: is_eq_exact EqualExact -fail_action
+%macro: is_eq_exact EqualExact -fail_action -pack
 is_eq_exact f x x
 is_eq_exact f x y
 is_eq_exact f s s
-- 
cgit v1.2.3


From 7bfff20c48cdcabf1eae4b193a374a72977b9d3e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Thu, 2 Jul 2015 14:50:47 +0200
Subject: Eliminate the variable temp_bits at the top scope of process_main()

---
 erts/emulator/beam/beam_emu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index a3b38adc4b..1ab8c71685 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -819,6 +819,7 @@ void** beam_ops;
 #define BsGetFieldSize(Bits, Unit, Fail, Target)	\
    do {							\
       Sint _signed_size; Uint _uint_size;		\
+      Uint temp_bits;					\
       if (is_small(Bits)) {				\
         _signed_size = signed_val(Bits);		\
          if (_signed_size < 0) { Fail; }		\
@@ -833,6 +834,7 @@ void** beam_ops;
 #define BsGetUncheckedFieldSize(Bits, Unit, Fail, Target)	\
    do {								\
       Sint _signed_size; Uint _uint_size;			\
+      Uint temp_bits;						\
       if (is_small(Bits)) {					\
         _signed_size = signed_val(Bits);			\
          if (_signed_size < 0) { Fail; }			\
@@ -1219,8 +1221,6 @@ void process_main(void)
 #endif
 #endif
 
-    Uint temp_bits; /* Temporary used by BsSkipBits2 & BsGetInteger2 */
-
     Eterm pt_arity;		/* Used by do_put_tuple */
 
     Uint64 start_time = 0;          /* Monitor long schedule */
-- 
cgit v1.2.3


From e820b8534fd440aaf4991f3c57b8990ce836c8b0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Sat, 4 Jul 2015 08:58:33 +0200
Subject: Speed up list matching

The combination is_non_empty_list followed by get_list is extremly
common (but not in estone_SUITE, which is why it has not been noticed
before). Therefore it is worthwile to introduce a combined
instruction.
---
 erts/emulator/beam/beam_emu.c | 11 +++++++++++
 erts/emulator/beam/ops.tab    |  7 +++++++
 2 files changed, 18 insertions(+)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 1ab8c71685..989066a1ca 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -737,6 +737,17 @@ void** beam_ops;
     if (is_not_list(x(0))) { Fail; }			\
     TestHeap(Need, Alive)
 
+#define IsNonemptyListGetList(Src, H, T, Fail)	\
+    if (is_not_list(Src)) {			\
+         Fail;					\
+    } else {					\
+       Eterm* tmp_ptr = list_val(Src);		\
+       Eterm hd, tl;				\
+       hd = CAR(tmp_ptr);			\
+       tl = CDR(tmp_ptr);			\
+       H = hd; T = tl;				\
+    }
+
 #define IsTuple(X, Action) if (is_not_tuple(X)) Action
 
 #define IsArity(Pointer, Arity, Fail)		\
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 3d52da77d3..46fefb88af 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -682,6 +682,13 @@ is_nonempty_list F=f x==0 | test_heap I1 I2 => is_non_empty_list_test_heap F I1
 %macro: is_non_empty_list_test_heap IsNonemptyListTestHeap -fail_action -pack
 is_non_empty_list_test_heap f I t
 
+is_nonempty_list Fail=f S=x | get_list S D1=x D2=x => \
+  is_nonempty_list_get_list Fail S D1 D2
+
+%macro: is_nonempty_list_get_list IsNonemptyListGetList -fail_action -pack
+is_nonempty_list_get_list f r x x
+is_nonempty_list_get_list f x x x
+
 %macro: is_nonempty_list IsNonemptyList -fail_action
 is_nonempty_list f x
 is_nonempty_list f y
-- 
cgit v1.2.3


From c5755ae5ded3ba27dc5d884303ead232abc77a40 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bj=C3=B6rn=20Gustavsson?= <bjorn@erlang.org>
Date: Sat, 4 Jul 2015 17:10:14 +0200
Subject: Slightly tweak the peformance for get_list

Fetch the head and tail parts to temporary variables before
writing them to their destinations. That should allow the CPU to
perform the moves in parallel, which might improve performance.
---
 erts/emulator/beam/beam_emu.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

(limited to 'erts/emulator/beam')

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 989066a1ca..a296713a0a 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -662,10 +662,14 @@ void** beam_ops;
      SET_I((BeamInstr *) Arg(0));		\
      Goto(*I);
 
-#define GetList(Src, H, T) do {			\
-   Eterm* tmp_ptr = list_val(Src);		\
-   H = CAR(tmp_ptr);				\
-   T = CDR(tmp_ptr); } while (0)
+#define GetList(Src, H, T)			\
+  do {						\
+    Eterm* tmp_ptr = list_val(Src);		\
+    Eterm hd, tl;				\
+    hd = CAR(tmp_ptr);				\
+    tl = CDR(tmp_ptr);				\
+    H = hd; T = tl;				\
+  } while (0)
 
 #define GetTupleElement(Src, Element, Dest)		\
   do {							\
-- 
cgit v1.2.3