Merge branch 'bjorn/erts/beam-loader-fixes'

* bjorn/erts/beam-loader-fixes:
  Print transformtions sorted on the name of the first instruction
  Share code for call of predicate and transformation functions
  Simplify transformations of gc_bif[123] instructions
  Refactor 'too_old_compiler' handling
  Introduce 'try_me_else_fail'
  Combine 'store_var' with 'next_arg'
  Merge 'next_instr' and 'is_op' into 'next_instr'
  Merge 'new_instr' and 'store_op' into 'new_instr'
  Combine a 'call' instruction with the following 'end' instruction
  Remove some unnecessary type constraints in transformations
  Get rid of redundant 'try_me_else' and 'fail' instructions
  Add some more information in instruction comments
  In transformations, don't store variables that are never used
  Remove redundant 'next_arg' before 'next_instr'
  Generalize and rename is_set_var_instr() to is_instr()
  Remove too_old_compiler handling for very old instruction variants
  Remove the special instructions for the hybrid heap emulator
  Remove optimization of 'move R R'

4 files changed, 257 insertions, 249 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 4b5b5cbdaa..5691f7aec1 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -4863,92 +4863,6 @@ void process_main(void)
  }
 
  /*
-  * Instructions for allocating on the message area.
-  */
-
- OpCase(i_global_cons):
- {
-     BeamInstr *next;
-#ifdef HYBRID
-     Eterm *hp;
-
-     PreFetch(0,next);
-     TestGlobalHeap(2,2,hp);
-     hp[0] = r(0);
-     hp[1] = x(1);
-     r(0) = make_list(hp);
-#ifndef INCREMENTAL
-     global_htop += 2;
-#endif
-     NextPF(0,next);
-#else
-     PreFetch(0,next);
-     c_p->freason = EXC_INTERNAL_ERROR;
-     goto find_func_info;
-#endif
- }
-
- OpCase(i_global_tuple):
- {
-     BeamInstr *next;
-     int len;
-#ifdef HYBRID
-     Eterm list;
-     Eterm *hp;
-#endif
-
-     if ((len = list_length(r(0))) < 0) {
-         goto badarg;
-     }
-
-     PreFetch(0,next);
-#ifdef HYBRID
-     TestGlobalHeap(len + 1,1,hp);
-     list = r(0);
-     r(0) = make_tuple(hp);
-     *hp++ = make_arityval(len);
-     while(is_list(list))
-     {
-         Eterm* cons = list_val(list);
-         *hp++ = CAR(cons);
-         list = CDR(cons);
-     }
-#ifndef INCREMENTAL
-     global_htop += len + 1;
-#endif
-     NextPF(0,next);
-#else
-     c_p->freason = EXC_INTERNAL_ERROR;
-     goto find_func_info;
-#endif
- }
-
- OpCase(i_global_copy):
- {
-     BeamInstr *next;
-     PreFetch(0,next);
-#ifdef HYBRID
-     if (!IS_CONST(r(0)))
-     {
-         BM_SWAP_TIMER(system,copy);
-         SWAPOUT;
-         reg[0] = r(0);
-         reg[1] = NIL;
-         r(0) = copy_struct_lazy(c_p,r(0),0);
-         ASSERT(ma_src_top == 0);
-         ASSERT(ma_dst_top == 0);
-         ASSERT(ma_offset_top == 0);
-         SWAPIN;
-         BM_SWAP_TIMER(copy,system);
-     }
-     NextPF(0,next);
-#else
-     c_p->freason = EXC_INTERNAL_ERROR;
-     goto find_func_info;
-#endif
- }
-
- /*
   * New floating point instructions.
   */
 
@@ -5241,7 +5155,6 @@ void process_main(void)
 
  OpCase(int_code_end):
  OpCase(label_L):
- OpCase(too_old_compiler):
  OpCase(on_load):
  OpCase(line_I):
     erl_exit(1, "meta op\n");
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 16dd5795c7..de4b32b238 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -1887,14 +1887,6 @@ load_code(LoaderState* stp)
 	}
 
 	/*
-	 * Special error message instruction.
-	 */
-	if (stp->genop->op == genop_too_old_compiler_0) {
-	    LoadError0(stp, "please re-compile this module with an " 
-		       ERLANG_OTP_RELEASE " compiler");
-	}
-
-	/*
 	 * From the collected generic instruction, find the specific
 	 * instruction.
 	 */
@@ -1945,7 +1937,27 @@ load_code(LoaderState* stp)
 			       ERLANG_OTP_RELEASE " compiler ");
 		}
 
-		LoadError0(stp, "no specific operation found");
+		/*
+		 * Some generic instructions should have a special
+		 * error message.
+		 */
+		switch (stp->genop->op) {
+		case genop_too_old_compiler_0:
+		    LoadError0(stp, "please re-compile this module with an "
+			       ERLANG_OTP_RELEASE " compiler");
+		case genop_unsupported_guard_bif_3:
+		    {
+			Eterm Mod = (Eterm) stp->genop->a[0].val;
+			Eterm Name = (Eterm) stp->genop->a[1].val;
+			Uint arity = (Uint) stp->genop->a[2].val;
+			FREE_GENOP(stp, stp->genop);
+			stp->genop = 0;
+			LoadError3(stp, "unsupported guard BIF: %T:%T/%d\n",
+				   Mod, Name, arity);
+		    }
+		default:
+		    LoadError0(stp, "no specific operation found");
+		}
 	    }
 
 	    stp->specific_op = specific;
@@ -2409,6 +2421,8 @@ load_code(LoaderState* stp)
 #define no_fpe_signals(St) 0
 #endif
 
+#define never(St) 0
+
 /*
  * Predicate that tests whether a jump table can be used.
  */
@@ -3664,10 +3678,7 @@ gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
     BifFunction bf;
 
     NEW_GENOP(stp, op);
-    op->op = genop_i_gc_bif1_5;
-    op->arity = 5;
-    op->a[0] = Fail;
-    op->a[1].type = TAG_u;
+    op->next = NULL;
     bf = stp->import[Bif.val].bf;
     /* The translations here need to have a reverse counterpart in
        beam_emu.c:translate_gc_bif for error handling to work properly. */
@@ -3688,19 +3699,30 @@ gen_guard_bif1(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
     } else if (bf == trunc_1) {
 	op->a[1].val = (BeamInstr) (void *) erts_gc_trunc_1;
     } else {
-	abort();
+	op->op = genop_unsupported_guard_bif_3;
+	op->arity = 3;
+	op->a[0].type = TAG_a;
+	op->a[0].val = stp->import[Bif.val].module;
+	op->a[1].type = TAG_a;
+	op->a[1].val = stp->import[Bif.val].function;
+	op->a[2].type = TAG_u;
+	op->a[2].val = stp->import[Bif.val].arity;
+	return op;
     }
+    op->op = genop_i_gc_bif1_5;
+    op->arity = 5;
+    op->a[0] = Fail;
+    op->a[1].type = TAG_u;
     op->a[2] = Src;
     op->a[3] = Live;
     op->a[4] = Dst;
-    op->next = NULL;
     return op;
 }
 
 /*
- * This is used by the ops.tab rule that rewrites gc_bifs with two parameters
+ * 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
- * folowed by i_gc_bif2_jIId, to handle literals properly.
+ * 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.
  */
@@ -3712,31 +3734,39 @@ gen_guard_bif2(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
     BifFunction bf;
 
     NEW_GENOP(stp, op);
-    op->op = genop_ii_gc_bif2_6;
-    op->arity = 6;
-    op->a[0] = Fail;
-    op->a[1].type = TAG_u;
+    op->next = NULL;
     bf = stp->import[Bif.val].bf;
     /* The translations here need to have a reverse counterpart in
        beam_emu.c:translate_gc_bif for error handling to work properly. */
     if (bf == binary_part_2) {
 	op->a[1].val = (BeamInstr) (void *) erts_gc_binary_part_2;
     } else {
-	abort();
+	op->op = genop_unsupported_guard_bif_3;
+	op->arity = 3;
+	op->a[0].type = TAG_a;
+	op->a[0].val = stp->import[Bif.val].module;
+	op->a[1].type = TAG_a;
+	op->a[1].val = stp->import[Bif.val].function;
+	op->a[2].type = TAG_u;
+	op->a[2].val = stp->import[Bif.val].arity;
+	return op;
     }
+    op->op = genop_ii_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[5] = Dst;
-    op->next = NULL;
     return op;
 }
 
 /*
- * This is used by the ops.tab rule that rewrites gc_bifs with three parameters
+ * 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,
- * folowed by i_gc_bif3_jIsId, to handle literals properly. Rewriting
+ * followed by i_gc_bif3_jIsId, to handle literals properly. Rewriting
  * always occur, as with the gc_bif2 counterpart.
  */
 static GenOp*
@@ -3747,18 +3777,27 @@ gen_guard_bif3(LoaderState* stp, GenOpArg Fail, GenOpArg Live, GenOpArg Bif,
     BifFunction bf;
 
     NEW_GENOP(stp, op);
-    op->op = genop_ii_gc_bif3_7;
-    op->arity = 7;
-    op->a[0] = Fail;
-    op->a[1].type = TAG_u;
+    op->next = NULL;
     bf = stp->import[Bif.val].bf;
     /* The translations here need to have a reverse counterpart in
        beam_emu.c:translate_gc_bif for error handling to work properly. */
     if (bf == binary_part_3) {
 	op->a[1].val = (BeamInstr) (void *) erts_gc_binary_part_3;
     } else {
-	abort();
+	op->op = genop_unsupported_guard_bif_3;
+	op->arity = 3;
+	op->a[0].type = TAG_a;
+	op->a[0].val = stp->import[Bif.val].module;
+	op->a[1].type = TAG_a;
+	op->a[1].val = stp->import[Bif.val].function;
+	op->a[2].type = TAG_u;
+	op->a[2].val = stp->import[Bif.val].arity;
+	return op;
     }
+    op->op = genop_ii_gc_bif3_7;
+    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;
@@ -4225,6 +4264,7 @@ transform_engine(LoaderState* st)
     GenOp* instr;
     Uint* pc;
     int rval;
+    static Uint restart_fail[1] = {TOP_fail};
 
     ASSERT(gen_opc[st->genop->op].transform != -1);
     pc = op_transform + gen_opc[st->genop->op].transform;
@@ -4238,7 +4278,6 @@ transform_engine(LoaderState* st)
     ASSERT(restart != NULL);
     pc = restart;
     ASSERT(*pc < NUM_TOPS);	/* Valid instruction? */
-    ASSERT(*pc == TOP_try_me_else || *pc == TOP_fail);
     instr = st->genop;
 
 #define RETURN(r) rval = (r); goto do_return;
@@ -4251,7 +4290,9 @@ transform_engine(LoaderState* st)
 	op = *pc++;
 
 	switch (op) {
-	case TOP_is_op:
+	case TOP_next_instr:
+	    instr = instr->next;
+	    ap = 0;
 	    if (instr == NULL) {
 		/*
 		 * We'll need at least one more instruction to decide whether
@@ -4438,10 +4479,6 @@ transform_engine(LoaderState* st)
 	case TOP_next_arg:
 	    ap++;
 	    break;
-	case TOP_next_instr:
-	    instr = instr->next;
-	    ap = 0;
-	    break;
 	case TOP_commit:
 	    instr = instr->next; /* The next_instr was optimized away. */
 
@@ -4459,8 +4496,8 @@ transform_engine(LoaderState* st)
 #endif
 	    break;
 
-#if defined(TOP_call)
-	case TOP_call:
+#if defined(TOP_call_end)
+	case TOP_call_end:
 	    {
 		GenOp** lastp;
 		GenOp* new_instr;
@@ -4497,7 +4534,7 @@ transform_engine(LoaderState* st)
 		*lastp = st->genop;
 		st->genop = new_instr;
 	    }
-	    break;
+	    RETURN(TE_OK);
 #endif
 	case TOP_new_instr:
 	    /*
@@ -4506,12 +4543,10 @@ transform_engine(LoaderState* st)
 	    NEW_GENOP(st, instr);
 	    instr->next = st->genop;
 	    st->genop = instr;
+	    instr->op = op = *pc++;
+	    instr->arity = gen_opc[op].arity;
 	    ap = 0;
 	    break;
-	case TOP_store_op:
-	    instr->op = *pc++;
-	    instr->arity = *pc++;
-	    break;
 	case TOP_store_type:
 	    i = *pc++;
 	    instr->a[ap].type = i;
@@ -4521,21 +4556,25 @@ transform_engine(LoaderState* st)
 	    i = *pc++;
 	    instr->a[ap].val = i;
 	    break;
-	case TOP_store_var:
+	case TOP_store_var_next_arg:
 	    i = *pc++;
 	    ASSERT(i < TE_MAX_VARS);
 	    instr->a[ap].type = var[i].type;
 	    instr->a[ap].val = var[i].val;
+	    ap++;
 	    break;
 	case TOP_try_me_else:
 	    restart = pc + 1;
 	    restart += *pc++;
 	    ASSERT(*pc < NUM_TOPS); /* Valid instruction? */
 	    break;
+	case TOP_try_me_else_fail:
+	    restart = restart_fail;
+	    break;
 	case TOP_end:
 	    RETURN(TE_OK);
 	case TOP_fail:
-	    RETURN(TE_FAIL)
+	    RETURN(TE_FAIL);
 	default:
 	    ASSERT(0);
 	}
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 538f0b94af..34bd5d0653 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -25,30 +25,12 @@
 # instruction transformations; thus, they never occur in BEAM files.
 #
 
-# Special instruction used to generate an error message when
-# trying to load a module compiled by the V1 compiler (R5 & R6).
-# (Specially treated in beam_load.c.)
+# The too_old_compiler/0 instruction is specially handled in beam_load.c
+# to produce a user-friendly message informing the user that the module
+# needs to be re-compiled with a modern compiler.
 
 too_old_compiler/0
-too_old_compiler
-
-#
-# Obsolete instruction usage follow. (Nowdays we use f with
-# a zero label instead of p.)
-#
-
-is_list p S => too_old_compiler
-is_nonempty_list p R => too_old_compiler
-is_nil p R => too_old_compiler
-
-is_tuple p S => too_old_compiler
-test_arity p S Arity => too_old_compiler
-
-is_integer p R => too_old_compiler
-is_float p R => too_old_compiler
-is_atom p R => too_old_compiler
-
-is_eq_exact p S1 S2 => too_old_compiler
+too_old_compiler | never() =>
 
 # In R9C and earlier, the loader used to insert special instructions inside
 # the module_info/0,1 functions. (In R10B and later, the compiler inserts
@@ -88,9 +70,6 @@ i_time_breakpoint
 i_return_time_trace
 i_return_to_trace
 i_yield
-i_global_cons
-i_global_tuple
-i_global_copy
 
 return
 
@@ -310,8 +289,6 @@ raise s s
 badarg j
 system_limit j
 
-move R R =>
-
 move C=cxy r | jump Lbl => move_jump Lbl C
 
 %macro: move_jump MoveJump -nonext
@@ -618,8 +595,6 @@ get_tuple_element Reg P Dst => i_get_tuple_element Reg P Dst | original_reg Reg
 
 original_reg Reg Pos =>
 
-get_tuple_element Reg P Dst => i_get_tuple_element Reg P Dst
-
 original_reg/2
 
 extract_next_element D1=xy | original_reg Reg P1 | get_tuple_element Reg P2 D2=xy | \
@@ -908,23 +883,6 @@ call_ext_last u==3 u$func:erlang:hibernate/3 D => i_hibernate
 call_ext_only u==3 u$func:erlang:hibernate/3 => i_hibernate
 
 #
-# Hybrid memory architecture need special cons and tuple instructions
-# that allocate on the message area. These looks like BIFs in the BEAM code.
-#
-
-call_ext u==2 u$func:hybrid:cons/2 => i_global_cons
-call_ext_last u==2 u$func:hybrid:cons/2 D => i_global_cons | deallocate_return D
-call_ext_only Ar=u==2 u$func:hybrid:cons/2 => i_global_cons | return
-
-call_ext u==1 u$func:hybrid:tuple/1 => i_global_tuple
-call_ext_last u==1 u$func:hybrid:tuple/1 D => i_global_tuple | deallocate_return D
-call_ext_only Ar=u==1 u$func:hybrid:tuple/1 => i_global_tuple | return
-
-call_ext u==1 u$func:hybrid:copy/1 => i_global_copy
-call_ext_last u==1 u$func:hybrid:copy/1 D => i_global_copy | deallocate_return D
-call_ext_only u==1 Ar=u$func:hybrid:copy/1 => i_global_copy | return
-
-#
 # The general case for BIFs that have no special instructions.
 # A BIF used in the tail must be followed by a return instruction.
 #
@@ -961,9 +919,9 @@ move S=c r | call_ext Ar=u Func=u$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 
 
-call_ext Ar=u Func        => i_call_ext Func
-call_ext_last Ar=u Func D => i_call_ext_last Func D
-call_ext_only Ar=u Func   => i_call_ext_only Func
+call_ext Ar Func        => i_call_ext Func
+call_ext_last Ar Func D => i_call_ext_last Func D
+call_ext_only Ar Func   => i_call_ext_only Func
 
 i_apply
 i_apply_last P
@@ -997,7 +955,7 @@ 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=f Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2 Fail Bif Dst
+bif2 Fail Bif S1 S2 Dst => i_fetch S1 S2 | i_bif2 Fail Bif Dst
 
 i_get s d
 
@@ -1080,8 +1038,8 @@ i_move_call_ext_only e c r
 
 # Fun calls.
 
-call_fun Arity=u | deallocate D | return => i_call_fun_last Arity D
-call_fun Arity=u => i_call_fun Arity
+call_fun Arity | deallocate D | return => i_call_fun_last Arity D
+call_fun Arity => i_call_fun Arity
 
 i_call_fun I
 i_call_fun_last I P
@@ -1337,13 +1295,13 @@ 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=j u Src=s => i_bs_put_utf8 Fail Src
+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=j Flags=u Src=s => i_bs_put_utf16 Fail Flags Src
+bs_put_utf16 Fail Flags=u Src=s => i_bs_put_utf16 Fail Flags Src
 
 i_bs_put_utf16 j I s
 
@@ -1508,34 +1466,13 @@ bif1 Fail u$bif:erlang:trunc/1 s d => too_old_compiler
 #
 # Guard BIFs.
 #
-gc_bif1 Fail I Bif=u$bif:erlang:length/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif1 Fail I Bif=u$bif:erlang:size/1 Src Dst=d => \
+gc_bif1 Fail I Bif Src Dst => \
 	gen_guard_bif1(Fail, I, Bif, Src, Dst)
 
-gc_bif1 Fail I Bif=u$bif:erlang:bit_size/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif1 Fail I Bif=u$bif:erlang:byte_size/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif1 Fail I Bif=u$bif:erlang:abs/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif1 Fail I Bif=u$bif:erlang:float/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif1 Fail I Bif=u$bif:erlang:round/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif1 Fail I Bif=u$bif:erlang:trunc/1 Src Dst=d => \
-	gen_guard_bif1(Fail, I, Bif, Src, Dst)
-
-gc_bif2 Fail I Bif=u$bif:erlang:binary_part/2 S1 S2 Dst=d => \
+gc_bif2 Fail I Bif S1 S2 Dst => \
 	gen_guard_bif2(Fail, I, Bif, S1, S2, Dst)
 
-gc_bif3 Fail I Bif=u$bif:erlang:binary_part/3 S1 S2 S3 Dst=d => \
+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
@@ -1553,6 +1490,15 @@ 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
 
 i_gc_bif3 j I s I d
+
+#
+# The following instruction is specially handled in beam_load.c
+# to produce a user-friendly message if an unsupported guard BIF is
+# encountered.
+#
+unsupported_guard_bif/3
+unsupported_guard_bif A B C | never() =>
+
 #
 # R13B03
 #
diff --git a/erts/emulator/utils/beam_makeops b/erts/emulator/utils/beam_makeops
index ebf7db3277..58c36c3bdc 100755
--- a/erts/emulator/utils/beam_makeops
+++ b/erts/emulator/utils/beam_makeops
@@ -105,7 +105,9 @@ my %match_engine_ops;		# All opcodes for the match engine.
 my %gen_transform_offset;
 my @transformations;
 my @call_table;
+my %call_table;
 my @pred_table;
+my %pred_table;
 
 # Operand types for generic instructions.
 
@@ -187,6 +189,12 @@ sub define_type_bit {
 }
 
 #
+# Pre-define the 'fail' instruction. It is used internally
+# by the 'try_me_else_fail' instruction.
+#
+$match_engine_ops{'TOP_fail'} = 1;
+
+#
 # Sanity checks.
 #
 
@@ -1304,7 +1312,8 @@ sub tr_gen {
 
     foreach $ref (@g) {
 	my($line, $orig_transform, $from_ref, $to_ref) = @$ref;
-	my $so_far = tr_gen_from($line, @$from_ref);
+	my $used_ref = used_vars($from_ref, $to_ref);
+	my $so_far = tr_gen_from($line, $used_ref, @$from_ref);
 	tr_gen_to($line, $orig_transform, $so_far, @$to_ref);
     }
 
@@ -1313,9 +1322,22 @@ sub tr_gen {
     #
     my($offset) = 0;
     print "Uint op_transform[] = {\n";
-    foreach $key (keys %gen_transform) {
+    foreach $key (sort keys %gen_transform) {
 	$gen_transform_offset{$key} = $offset;
-	foreach $instr (@{$gen_transform{$key}}) {
+	my @instr = @{$gen_transform{$key}};
+
+	#
+	# If the last instruction is 'fail', remove it and
+	# convert the previous 'try_me_else' to 'try_me_else_fail'.
+	#
+	if (is_instr($instr[$#instr], 'fail')) {
+	    pop(@instr);
+	    my $i = $#instr;
+	    $i-- while !is_instr($instr[$i], 'try_me_else');
+	    $instr[$i] = make_op('', 'try_me_else_fail');
+	}
+
+	foreach $instr (@instr) {
 	    my($size, $instr_ref, $comment) = @$instr;
 	    my($op, @args) = @$instr_ref;
 	    print "    ";
@@ -1342,8 +1364,48 @@ sub tr_gen {
     print "};\n\n";
 }
 
+sub used_vars {
+    my($from_ref,$to_ref) = @_;
+    my %used;
+    my %seen;
+
+    foreach my $ref (@$from_ref) {
+	my($name,$arity,@ops) = @$ref;
+	if ($name =~ /^[.]/) {
+	    foreach my $var (@ops) {
+		$used{$var} = 1;
+	    }
+	} else {
+	    # Any variable that is used at least twice on the
+	    # left-hand side is used. (E.g. "move R R".)
+	    foreach my $op (@ops) {
+		my($var, $type, $type_val) = @$op;
+		next if $var eq '';
+		$used{$var} = 1 if $seen{$var};
+		$seen{$var} = 1;
+	    }
+	}
+    }
+
+    foreach my $ref (@$to_ref) {
+	my($name, $arity, @ops) = @$ref;
+	if ($name =~ /^[.]/) {
+	    foreach my $var (@ops) {
+		$used{$var} = 1;
+	    }
+	} else {
+	    foreach my $op (@ops) {
+		my($var, $type, $type_val) = @$op;
+		next if $var eq '';
+		$used{$var} = 1;
+	    }
+	}
+    }
+    \%used;
+}
+
 sub tr_gen_from {
-    my($line, @tr) = @_;
+    my($line,$used_ref,@tr) = @_;
     my(%var) = ();
     my(%var_type);
     my($var_num) = 0;
@@ -1353,25 +1415,30 @@ sub tr_gen_from {
     my(@fix_pred_funcs);
     my($op, $ref);		# Loop variables.
     my $where = "left side of transformation in line $line: ";
+    my %var_used = %$used_ref;
+    my $may_fail = 0;
+    my $is_first = 1;
 
     foreach $ref (@tr) {
 	my($name, $arity, @ops) = @$ref;
 	my($key) = "$name/$arity";
 	my($opnum);
 
+	$may_fail = 1 unless $is_first;
+	$is_first = 0;
+
 	#
 	# A name starting with a period is a C pred function to be called.
 	#
 
 	if ($name =~ /^\.(\w+)/) {
 	    $name = $1;
+	    $may_fail = 1;
 	    my $var;
 	    my(@args);
 
-	    my $next_instr = pop(@code); # Get rid of 'next_instr'
 	    push(@fix_pred_funcs, scalar(@code));
 	    push(@code, [$name, @ops]);
-	    push(@code, $next_instr);
 	    next;
 	}
 
@@ -1383,17 +1450,21 @@ sub tr_gen_from {
 	    unless defined $gen_opnum{$name,$arity};
 	$opnum = $gen_opnum{$name,$arity};
 
-	push(@code, &make_op("$name/$arity", 'is_op', $opnum));
+	push(@code, make_op("$name/$arity", 'next_instr', $opnum));
 	$min_window++;
 	foreach $op (@ops) {
 	    my($var, $type, $type_val, $cond, $val) = @$op;
+	    my $ignored_var = "$var (ignored)";
 
 	    if ($type ne '' && $type ne '*') {
+		$may_fail = 1;
+
 		#
 		# The is_bif, is_not_bif, and is_func instructions have
 		# their own built-in type test and don't need to
 		# be guarded with a type test instruction.
 		#
+		$ignored_var = '';
 		unless ($cond eq 'is_bif' or
 			$cond eq 'is_not_bif' or
 			$cond eq 'is_func') {
@@ -1407,7 +1478,7 @@ sub tr_gen_from {
 			push(@code, &make_op($types, 'is_type', $type_mask));
 		    } else {
 			$cond = '';
-			push(@code, &make_op($types, 'is_type_eq',
+			push(@code, &make_op("$types== $val", 'is_type_eq',
 					     $type_mask, $val));
 		    }
 		}
@@ -1415,46 +1486,55 @@ sub tr_gen_from {
 
 	    if ($cond eq 'is_func') {
 		my($m, $f, $a) = split(/:/, $val);
+		$ignored_var = '';
+		$may_fail = 1;
 		push(@code, &make_op('', "$cond", "am_$m",
 				     "am_$f", $a));
 	    } elsif ($cond ne '') {
+		$ignored_var = '';
+		$may_fail = 1;
 		push(@code, &make_op('', "$cond", $val));
 	    }
 
 	    if ($var ne '') {
 		if (defined $var{$var}) {
+		    $ignored_var = '';
+		    $may_fail = 1;
 		    push(@code, &make_op($var, 'is_same_var', $var{$var}));
 		} elsif ($type eq '*') {
 		    #
 		    # Reserve a hole for a 'rest_args' instruction.
 		    #
+		    $ignored_var = '';
 		    push(@fix_rest_args, scalar(@code));
 		    push(@code, $var);
-		} else {
+		} elsif ($var_used{$var}) {
+		    $ignored_var = '';
 		    $var_type{$var} = 'scalar';
 		    $var{$var} = $var_num;
 		    $var_num++;
 		    push(@code, &make_op($var, 'set_var', $var{$var}));
 		}
 	    }
-	    if (is_set_var_instr($code[$#code])) {
+	    if (is_instr($code[$#code], 'set_var')) {
 		my $ref = pop @code;
 		my $comment = $ref->[2];
 		my $var = $ref->[1][1];
 		push(@code, make_op($comment, 'set_var_next_arg', $var));
 	    } else {
-		push(@code, &make_op('', 'next_arg'));
+		push(@code, &make_op($ignored_var, 'next_arg'));
 	    }
 	}
-	push(@code, &make_op('', 'next_instr'));
-	pop(@code) if $code[$#code]->[1][0] eq 'next_arg';
+
+	# Remove redundant 'next_arg' instructions before the end
+	# of the instruction.
+	pop(@code) while is_instr($code[$#code], 'next_arg');
     }
 
     #
     # Insert the commit operation.
     #
-    pop(@code);		# Get rid of 'next_instr'
-    push(@code, &make_op('', 'commit'));
+    push(@code, make_op($may_fail ? '' : 'always reached', 'commit'));
 
     #
     # If there is an rest_args instruction, we must insert its correct
@@ -1484,9 +1564,8 @@ sub tr_gen_from {
 		push(@args, "var+$var{$var}");
 	    }
 	}
-	splice(@code, $index, 1, &make_op("$name()",
-					  'pred', scalar(@pred_table)));
-	push(@pred_table, [$name, @args]);
+	my $pi = tr_next_index(\@pred_table, \%pred_table, $name, @args);
+	splice(@code, $index, 1, make_op("$name()", 'pred', $pi));
     }
 
     $te_max_vars = $var_num
@@ -1503,6 +1582,10 @@ sub tr_gen_to {
     my($op, $ref);		# Loop variables.
     my($where) = "right side of transformation in line $line: ";
 
+    my $last_instr = $code[$#code];
+    my $cannot_fail = is_instr($last_instr, 'commit') &&
+	(get_comment($last_instr) =~ /^always/);
+
     foreach $ref (@tr) {
 	my($name, $arity, @ops) = @$ref;
 
@@ -1524,9 +1607,10 @@ sub tr_gen_to {
 		    push(@args, "var+$var{$var}");
 		}
 	    }
-	    pop(@code);	# Get rid of 'next_instr'
-	    push(@code, &make_op("$name()", 'call', scalar(@call_table)));
-	    push(@call_table, [$name, @args]);
+	    pop(@code);	# Get rid of 'commit' instruction
+	    my $index = tr_next_index(\@call_table, \%call_table,
+				      $name, @args);
+	    push(@code, make_op("$name()", 'call_end', $index));
 	    last;
 	}
 
@@ -1543,27 +1627,27 @@ sub tr_gen_to {
 	# Create code to build the generic instruction.
 	#
 
-	push(@code, &make_op('', 'new_instr'));
-	push(@code, &make_op("$name/$arity", 'store_op', $opnum, $arity));
+	push(@code, make_op("$name/$arity", 'new_instr', $opnum));
 	foreach $op (@ops) {
 	    my($var, $type, $type_val) = @$op;
 
 	    if ($var ne '') {
 		&error($where, "variable '$var' unbound")
 		    unless defined $var{$var};
-		push(@code, &make_op($var, 'store_var', $var{$var}));
+		push(@code, &make_op($var, 'store_var_next_arg', $var{$var}));
 	    } elsif ($type ne '') {
 		push(@code, &make_op('', 'store_type', "TAG_$type"));
 		if ($type_val) {
 		    push(@code, &make_op('', 'store_val', $type_val));
 		}
+		push(@code, make_op('', 'next_arg'));
 	    }
-	    push(@code, &make_op('', 'next_arg'));
 	}
-	pop(@code) if $code[$#code]->[1][0] eq 'next_arg';
+	pop(@code) if is_instr($code[$#code], 'next_arg');
     }
 
-    push(@code, &make_op('', 'end'));
+    push(@code, make_op('', 'end'))
+	unless is_instr($code[$#code], 'call_end');
 
     #
     # Chain together all codes segments having the same first operation.
@@ -1575,11 +1659,20 @@ sub tr_gen_to {
     $min_window{$key} = $min_window
 	if $min_window{$key} > $min_window;
 
-    pop(@{$gen_transform{$key}})
+    my $prev_last;
+    $prev_last = pop(@{$gen_transform{$key}})
 	if defined @{$gen_transform{$key}}; # Fail
-    my(@prefix) = (&make_op($comment), &make_op('', 'try_me_else', &tr_code_len(@code)));
-    unshift(@code, @prefix);
-    push(@{$gen_transform{$key}}, @code, &make_op('', 'fail'));
+
+    if ($prev_last && !is_instr($prev_last, 'fail')) {
+	error("Line $line: A previous transformation shadows '$orig_transform'");
+    }
+    unless ($cannot_fail) {
+	unshift(@code, make_op('', 'try_me_else',
+			       tr_code_len(@code)));
+	push(@code, make_op(""), make_op("$key", 'fail'));
+    }
+    unshift(@code, make_op($comment));
+    push(@{$gen_transform{$key}}, @code),
 }
 
 sub tr_code_len {
@@ -1597,21 +1690,38 @@ sub make_op {
     [scalar(@op), [@op], $comment];
 }
 
-sub is_set_var_instr {
-    my($ref) = @_;
+sub is_instr {
+    my($ref,$op) = @_;
     return 0 unless ref($ref) eq 'ARRAY';
-    $ref->[1][0] eq 'set_var';
+    $ref->[1][0] eq $op;
+}
+
+sub get_comment {
+    my($ref,$op) = @_;
+    return '' unless ref($ref) eq 'ARRAY';
+    $ref->[2];
+}
+
+sub tr_next_index {
+    my($lref,$href,$name,@args) = @_;
+    my $code = "RVAL = $name(" . join(', ', 'st', @args) . "); break;\n";
+    my $index;
+
+    if (defined $$href{$code}) {
+	$index = $$href{$code};
+    } else {
+	$index = scalar(@$lref);
+	push(@$lref, $code);
+	$$href{$code} = $index;
+    }
+    $index;
 }
 
 sub tr_gen_call {
     my(@call_table) = @_;
     my($i);
 
-    print "\n";
     for ($i = 0; $i < @call_table; $i++) {
-	my $ref = $call_table[$i];
-	my($name, @args) = @$ref;
-	print "case $i: RVAL = $name(", join(', ', 'st', @args), "); break;\n";
+	print "case $i: $call_table[$i]";
     }
-    print "\n";
 }