Merge branch 'sverk/refactor-encode-size/OTP-12585'

* sverk/refactor-encode-size/OTP-12585:
  erts: Optimize insert and delete for big maps
  erts: Optimize == and /= for unequal big maps
  erts: Refactor encode_size_struct_int

Conflicts:
	erts/emulator/beam/erl_map.c

8 files changed, 139 insertions, 174 deletions

diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c
index bfecac1612..32f3cda4f5 100644
--- a/erts/emulator/beam/dist.c
+++ b/erts/emulator/beam/dist.c
@@ -712,7 +712,7 @@ void erts_dsend_context_dtor(Binary* ctx_bin)
     ErtsSendContext* ctx = ERTS_MAGIC_BIN_DATA(ctx_bin);
     switch (ctx->dss.phase) {
     case ERTS_DSIG_SEND_PHASE_MSG_SIZE:
-	DESTROY_SAVED_ESTACK(&ctx->dss.u.sc.estack);
+	DESTROY_SAVED_WSTACK(&ctx->dss.u.sc.wstack);
 	break;
     case ERTS_DSIG_SEND_PHASE_MSG_ENCODE:
 	DESTROY_SAVED_WSTACK(&ctx->dss.u.ec.wstack);
@@ -1800,7 +1800,7 @@ erts_dsig_send(ErtsDSigData *dsdp, struct erts_dsig_send_context* ctx)
 	    erts_encode_dist_ext_size(ctx->ctl, ctx->flags, ctx->acmp, &ctx->data_size);
 
 	    if (is_value(ctx->msg)) {
-		ctx->u.sc.estack.start = NULL;
+		ctx->u.sc.wstack.wstart = NULL;
 		ctx->u.sc.flags = ctx->flags;
 		ctx->u.sc.level = 0;
 		ctx->phase = ERTS_DSIG_SEND_PHASE_MSG_SIZE;
diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h
index 2a2ba0c83f..cd2cc0ef4a 100644
--- a/erts/emulator/beam/dist.h
+++ b/erts/emulator/beam/dist.h
@@ -282,7 +282,7 @@ typedef struct TTBSizeContext_ {
     int level;
     Uint result;
     Eterm obj;
-    ErtsEStack estack;
+    ErtsWStack wstack;
 } TTBSizeContext;
 
 typedef struct TTBEncodeContext_ {
diff --git a/erts/emulator/beam/erl_bif_lists.c b/erts/emulator/beam/erl_bif_lists.c
index 820ed2385d..e006d57124 100644
--- a/erts/emulator/beam/erl_bif_lists.c
+++ b/erts/emulator/beam/erl_bif_lists.c
@@ -390,7 +390,7 @@ keyfind(int Bif, Process* p, Eterm Key, Eterm Pos, Eterm List)
 		Eterm *tuple_ptr = tuple_val(term);
 		if (pos <= arityval(*tuple_ptr)) {
 		    Eterm element = tuple_ptr[pos];
-		    if (CMP(Key, element) == 0) {
+		    if (CMP_EQ(Key, element)) {
 			return term;
 		    }
 		}
diff --git a/erts/emulator/beam/erl_db_tree.c b/erts/emulator/beam/erl_db_tree.c
index 577da35b75..d90af46659 100644
--- a/erts/emulator/beam/erl_db_tree.c
+++ b/erts/emulator/beam/erl_db_tree.c
@@ -1116,7 +1116,7 @@ static int db_select_tree(Process *p, DbTable *tbl,
     sc.all_objects = mpi.all_objects;
 
     if (!mpi.got_partial && mpi.some_limitation && 
-	CMP(mpi.least,mpi.most) == 0) {
+	CMP_EQ(mpi.least,mpi.most)) {
 	doit_select(tb,mpi.save_term,&sc,0 /* direction doesn't matter */);
 	RET_TO_BIF(sc.accum,DB_ERROR_NONE);
     }
@@ -1324,7 +1324,7 @@ static int db_select_count_tree(Process *p, DbTable *tbl,
     sc.all_objects = mpi.all_objects;
 
     if (!mpi.got_partial && mpi.some_limitation && 
-	CMP(mpi.least,mpi.most) == 0) {
+	CMP_EQ(mpi.least,mpi.most)) {
 	doit_select_count(tb,mpi.save_term,&sc,0 /* dummy */);
 	RET_TO_BIF(erts_make_integer(sc.got,p),DB_ERROR_NONE);
     }
@@ -1429,7 +1429,7 @@ static int db_select_chunk_tree(Process *p, DbTable *tbl,
     sc.all_objects = mpi.all_objects;
 
     if (!mpi.got_partial && mpi.some_limitation && 
-	CMP(mpi.least,mpi.most) == 0) {
+	CMP_EQ(mpi.least,mpi.most)) {
 	doit_select(tb,mpi.save_term,&sc, 0 /* direction doesn't matter */);
 	if (sc.accum != NIL) {
 	    hp=HAlloc(p, 3);
@@ -1673,7 +1673,7 @@ static int db_select_delete_tree(Process *p, DbTable *tbl,
     sc.mp = mpi.mp;
 
     if (!mpi.got_partial && mpi.some_limitation && 
-	CMP(mpi.least,mpi.most) == 0) {
+	CMP_EQ(mpi.least,mpi.most)) {
 	doit_select_delete(tb,mpi.save_term,&sc, 0 /* direction doesn't 
 						      matter */);
 	RET_TO_BIF(erts_make_integer(sc.accum,p),DB_ERROR_NONE);
diff --git a/erts/emulator/beam/erl_map.c b/erts/emulator/beam/erl_map.c
index 20a17bcd24..cca287c753 100644
--- a/erts/emulator/beam/erl_map.c
+++ b/erts/emulator/beam/erl_map.c
@@ -1947,27 +1947,32 @@ int erts_hashmap_insert_down(Uint32 hx, Eterm key, Eterm node, Uint *sz,
 		    case HAMT_SUBTAG_NODE_BITMAP:
 			hval = MAP_HEADER_VAL(hdr);
 			ix   = hashmap_index(hx);
-			bp   = 1 << ix;
-			slot = hashmap_bitcount(hval & (bp - 1));
-			n    = hashmap_bitcount(hval);
+                        bp   = 1 << ix;
+                        if (hval == 0xffff) {
+                            slot = ix;
+                            n = 16;
+                        } else {
+                            slot = hashmap_bitcount(hval & (bp - 1));
+                            n    = hashmap_bitcount(hval);
+                        }
+
+                        ESTACK_PUSH4(*sp, n, bp, slot, node);
+
+                        if (!(bp & hval)) { /* not occupied */
+                            if (is_update) {
+				UnUseTmpHeapNoproc(2);
+                                return 0;
+                            }
+                            size += HAMT_NODE_BITMAP_SZ(n+1);
+                            goto unroll;
+                        }
+
+                        hx    = hashmap_shift_hash(th,hx,lvl,key);
+                        node  = ptr[slot+1];
+                        ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17);
+                        size += HAMT_NODE_BITMAP_SZ(n);
+                        break;
 
-			ESTACK_PUSH4(*sp, n, bp, slot, node);
-
-			/* occupied */
-			if (bp & hval) {
-			    hx    = hashmap_shift_hash(th,hx,lvl,key);
-			    node  = ptr[slot+1];
-			    ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17);
-			    size += HAMT_NODE_BITMAP_SZ(n);
-			    break;
-			}
-			/* not occupied */
-			if (is_update) {
-                            UnUseTmpHeapNoproc(2);
-			    return 0;
-			}
-			size += HAMT_NODE_BITMAP_SZ(n+1);
-			goto unroll;
 		    case HAMT_SUBTAG_HEAD_BITMAP:
 			hval = MAP_HEADER_VAL(hdr);
 			ix   = hashmap_index(hx);
@@ -2197,21 +2202,25 @@ static Eterm hashmap_delete(Process *p, Uint32 hx, Eterm key, Eterm map) {
 			hval = MAP_HEADER_VAL(hdr);
 			ix   = hashmap_index(hx);
 			bp   = 1 << ix;
-			slot = hashmap_bitcount(hval & (bp - 1));
-			n    = hashmap_bitcount(hval);
+                        if (hval == 0xffff) {
+                            slot = ix;
+                            n = 16;
+                        } else if (bp & hval) {
+                            slot = hashmap_bitcount(hval & (bp - 1));
+                            n    = hashmap_bitcount(hval);
+                        } else {
+                            /* not occupied */
+                            goto not_found;
+                        }
 
 			ESTACK_PUSH4(stack, n, bp, slot, node);
 
-			/* occupied */
-			if (bp & hval) {
-			    hx    = hashmap_shift_hash(th,hx,lvl,key);
-			    node  = ptr[slot+1];
-			    ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17);
-			    size += HAMT_NODE_BITMAP_SZ(n);
-			    break;
-			}
-			/* not occupied */
-			goto not_found;
+                        hx    = hashmap_shift_hash(th,hx,lvl,key);
+                        node  = ptr[slot+1];
+                        ASSERT(HAMT_NODE_BITMAP_SZ(n) <= 17);
+                        size += HAMT_NODE_BITMAP_SZ(n);
+                        break;
+
 		    case HAMT_SUBTAG_HEAD_BITMAP:
 			hval = MAP_HEADER_VAL(hdr);
 			ix   = hashmap_index(hx);
diff --git a/erts/emulator/beam/erl_utils.h b/erts/emulator/beam/erl_utils.h
index 7cb8972e29..6a28105cb9 100644
--- a/erts/emulator/beam/erl_utils.h
+++ b/erts/emulator/beam/erl_utils.h
@@ -167,24 +167,26 @@ int eq(Eterm, Eterm);
 #define EQ(x,y) (((x) == (y)) || (is_not_both_immed((x),(y)) && eq((x),(y))))
 
 #if HALFWORD_HEAP
-Sint erts_cmp_rel_opt(Eterm, Eterm*, Eterm, Eterm*, int);
-#define cmp_rel(A,A_BASE,B,B_BASE)       erts_cmp_rel_opt(A,A_BASE,B,B_BASE,0)
-#define cmp_rel_term(A,A_BASE,B,B_BASE)  erts_cmp_rel_opt(A,A_BASE,B,B_BASE,1)
-#define CMP(A,B)                         erts_cmp_rel_opt(A,NULL,B,NULL,0)
-#define CMP_TERM(A,B)                    erts_cmp_rel_opt(A,NULL,B,NULL,1)
+Sint erts_cmp_rel_opt(Eterm, Eterm*, Eterm, Eterm*, int, int);
+#define cmp_rel(A,A_BASE,B,B_BASE)       erts_cmp_rel_opt(A,A_BASE,B,B_BASE,0,0)
+#define cmp_rel_term(A,A_BASE,B,B_BASE)  erts_cmp_rel_opt(A,A_BASE,B,B_BASE,1,0)
+#define CMP(A,B)                         erts_cmp_rel_opt(A,NULL,B,NULL,0,0)
+#define CMP_TERM(A,B)                    erts_cmp_rel_opt(A,NULL,B,NULL,1,0)
+#define CMP_EQ_ONLY(A,B)                 erts_cmp_rel_opt(A,NULL,B,NULL,0,1)
 #else
-Sint cmp(Eterm, Eterm);
-Sint erts_cmp(Eterm, Eterm, int);
-#define cmp_rel(A,A_BASE,B,B_BASE)       erts_cmp(A,B,0)
-#define cmp_rel_term(A,A_BASE,B,B_BASE)  erts_cmp(A,B,1)
-#define CMP(A,B)                         erts_cmp(A,B,0)
-#define CMP_TERM(A,B)                    erts_cmp(A,B,1)
+Sint erts_cmp(Eterm, Eterm, int, int);
+Sint cmp(Eterm a, Eterm b);
+#define cmp_rel(A,A_BASE,B,B_BASE)       erts_cmp(A,B,0,0)
+#define cmp_rel_term(A,A_BASE,B,B_BASE)  erts_cmp(A,B,1,0)
+#define CMP(A,B)                         erts_cmp(A,B,0,0)
+#define CMP_TERM(A,B)                    erts_cmp(A,B,1,0)
+#define CMP_EQ_ONLY(A,B)                 erts_cmp(A,B,0,1)
 #endif
 
 #define cmp_lt(a,b)          (CMP((a),(b)) <  0)
 #define cmp_le(a,b)          (CMP((a),(b)) <= 0)
-#define cmp_eq(a,b)          (CMP((a),(b)) == 0)
-#define cmp_ne(a,b)          (CMP((a),(b)) != 0)
+#define cmp_eq(a,b)          (CMP_EQ_ONLY((a),(b)) == 0)
+#define cmp_ne(a,b)          (CMP_EQ_ONLY((a),(b)) != 0)
 #define cmp_ge(a,b)          (CMP((a),(b)) >= 0)
 #define cmp_gt(a,b)          (CMP((a),(b)) >  0)
 
diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c
index 2117dbec62..fe48298155 100644
--- a/erts/emulator/beam/external.c
+++ b/erts/emulator/beam/external.c
@@ -1779,7 +1779,7 @@ static void ttb_context_destructor(Binary *context_bin)
 	context->alive = 0;
 	switch (context->state) {
 	case TTBSize:
-	    DESTROY_SAVED_ESTACK(&context->s.sc.estack);
+	    DESTROY_SAVED_WSTACK(&context->s.sc.wstack);
 	    break;
 	case TTBEncode:
 	    DESTROY_SAVED_WSTACK(&context->s.ec.wstack);
@@ -1847,7 +1847,7 @@ static Eterm erts_term_to_binary_int(Process* p, Eterm Term, int level, Uint fla
 	/* Setup enough to get started */
 	context->state = TTBSize;
 	context->alive = 1;
-	context->s.sc.estack.start = NULL;
+	context->s.sc.wstack.wstart = NULL;
 	context->s.sc.flags = flags;
 	context->s.sc.level = level;
     } else {
@@ -3962,51 +3962,35 @@ static int
 encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
 		       unsigned dflags, Sint *reds, Uint *res)
 {
-    DECLARE_ESTACK(s);
+    DECLARE_WSTACK(s);
     Uint m, i, arity;
     Uint result = 0;
     Sint r = 0;
 
     if (ctx) {
-	ESTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK);
+	WSTACK_CHANGE_ALLOCATOR(s, ERTS_ALC_T_SAVED_ESTACK);
 	r = *reds;
 
-	if (ctx->estack.start) { /* restore saved stack */
-	    ESTACK_RESTORE(s, &ctx->estack);
+	if (ctx->wstack.wstart) { /* restore saved stack */
+	    WSTACK_RESTORE(s, &ctx->wstack);
 	    result = ctx->result;
 	    obj = ctx->obj;
 	}
     }
 
-    goto L_jump_start;
+#define LIST_TAIL_OP ((0 << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER)
+#define TERM_ARRAY_OP(N) (((N) << _TAG_PRIMARY_SIZE) | TAG_PRIMARY_HEADER)
+#define TERM_ARRAY_OP_DEC(OP) ((OP) - (1 << _TAG_PRIMARY_SIZE))
+
+
+    for (;;) {
+	ASSERT(!is_header(obj));
 
- outer_loop:
-    while (!ESTACK_ISEMPTY(s)) {
-	obj = ESTACK_POP(s);
-    handle_popped_obj:
-	if (is_list(obj)) {
-	    Eterm* cons = list_val(obj);
-	    Eterm tl;
-
-	    tl = CDR(cons);
-	    obj = CAR(cons);
-	    ESTACK_PUSH(s, tl);
-	} else if (is_nil(obj)) {
-	    result++;
-	    goto outer_loop;
-	} else {
-	    /*
-	     * Other term (in the tail of a non-proper list or
-	     * in a fun's environment).
-	     */
-	}
-    
-    L_jump_start:
 	if (ctx && --r == 0) {
 	    *reds = r;
 	    ctx->obj = obj;
 	    ctx->result = result;
-	    ESTACK_SAVE(s, &ctx->estack);
+	    WSTACK_SAVE(s, &ctx->wstack);
 	    return -1;
 	}
 	switch (tag_val_def(obj)) {
@@ -4089,69 +4073,43 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
 		result += m + 2 + 1;
 	    } else {
 		result += 5;
-		goto handle_popped_obj;
+		WSTACK_PUSH2(s, (UWord)CDR(list_val(obj)), (UWord)LIST_TAIL_OP);
+		obj = CAR(list_val(obj));
+		continue; /* big loop */
 	    }
 	    break;
 	case TUPLE_DEF:
 	    {
 		Eterm* ptr = tuple_val(obj);
-		Uint i;
 		arity = arityval(*ptr);
 		if (arity <= 0xff) {
 		    result += 1 + 1;
 		} else {
 		    result += 1 + 4;
 		}
-		for (i = 1; i <= arity; ++i) {
-		    if (is_list(ptr[i])) {
-			if ((m = is_string(ptr[i])) && (m < MAX_STRING_LEN)) {
-			    result += m + 2 + 1;
-			    continue;
-			} else {
-			    result += 5;
-			}
-		    }
-		    ESTACK_PUSH(s,ptr[i]);
+		if (arity > 1) {
+		    WSTACK_PUSH2(s, (UWord) (ptr + 2),
+				    (UWord) TERM_ARRAY_OP(arity-1));
 		}
-		goto outer_loop;
+                else if (arity == 0) {
+		    break;
+                }
+		obj = ptr[1];
+		continue; /* big loop */
 	    }
-	    break;
 	case MAP_DEF:
 	    if (is_flatmap(obj)) {
 		flatmap_t *mp = (flatmap_t*)flatmap_val(obj);
 		Uint size = flatmap_get_size(mp);
-		Uint i;
-		Eterm *ptr;
 
 		result += 1 + 4; /* tag + 4 bytes size */
 
-		/* push values first */
-		ptr = flatmap_get_values(mp);
-		for (i = size; i; i--, ptr++) {
-		    if (is_list(*ptr)) {
-			if ((m = is_string(*ptr)) && (m < MAX_STRING_LEN)) {
-			    result += m + 2 + 1;
-			    continue;
-			} else {
-			    result += 5;
-			}
-		    }
-		    ESTACK_PUSH(s,*ptr);
+                if (size) {
+		    WSTACK_PUSH4(s, (UWord) flatmap_get_values(mp),
+				    (UWord) TERM_ARRAY_OP(size),
+		                    (UWord) flatmap_get_keys(mp),
+				    (UWord) TERM_ARRAY_OP(size));
 		}
-
-		ptr = flatmap_get_keys(mp);
-		for (i = size; i; i--, ptr++) {
-		    if (is_list(*ptr)) {
-			if ((m = is_string(*ptr)) && (m < MAX_STRING_LEN)) {
-			    result += m + 2 + 1;
-			    continue;
-			} else {
-			    result += 5;
-			}
-		    }
-		    ESTACK_PUSH(s,*ptr);
-		}
-		goto outer_loop;
 	    } else {
 		Eterm *ptr;
 		Eterm hdr;
@@ -4178,35 +4136,13 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
 		}
 
 		ptr++;
-		ESTACK_RESERVE(s, node_sz*2);
+		WSTACK_RESERVE(s, node_sz*2);
 		while(node_sz--) {
                     if (is_list(*ptr)) {
-			Eterm* leaf = list_val(*ptr);
-			if (is_not_list(CAR(leaf))) {
-			    ESTACK_FAST_PUSH(s, CAR(leaf));
-			}
-			else {
-			    if ((m = is_string(CAR(leaf))) && (m < MAX_STRING_LEN)) {
-				result += m + 2 + 1;
-			    } else {
-				result += 5;
-				ESTACK_FAST_PUSH(s, CAR(leaf));
-			    }
-			}
-			if (is_not_list(CDR(leaf))) {
-			    ESTACK_FAST_PUSH(s, CDR(leaf));
-			}
-			else {
-			    if ((m = is_string(CDR(leaf))) && (m < MAX_STRING_LEN)) {
-				result += m + 2 + 1;
-			    } else {
-				result += 5;
-				ESTACK_FAST_PUSH(s, CDR(leaf));
-			    }
-			}
-		    }
-		    else {
-			ESTACK_FAST_PUSH(s, *ptr);
+			WSTACK_FAST_PUSH(s, CAR(list_val(*ptr)));
+			WSTACK_FAST_PUSH(s, CDR(list_val(*ptr)));
+                    } else {
+			WSTACK_FAST_PUSH(s, *ptr);
 		    }
 		    ptr++;
 		}
@@ -4262,25 +4198,13 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
 		    result += 2 * (1 + 4); /* Index + Uniq */
 		    result += 1 + (funp->num_free < 0x100 ? 1 : 4);
 		}
-		for (i = 1; i < funp->num_free; i++) {
-		    obj = funp->env[i];
-
-		    if (is_not_list(obj)) {
-			/* Push any non-list terms on the stack */
-			ESTACK_PUSH(s, obj);
-		    } else {
-			/* Lists must be handled specially. */
-			if ((m = is_string(obj)) && (m < MAX_STRING_LEN)) {
-			    result += m + 2 + 1;
-			} else {
-			    result += 5;
-			    ESTACK_PUSH(s, obj);
-			}
-		    }
+		if (funp->num_free > 1) {
+		    WSTACK_PUSH2(s, (UWord) (funp->env + 1),
+				    (UWord) TERM_ARRAY_OP(funp->num_free-1));
 		}
 		if (funp->num_free != 0) {
 		    obj = funp->env[0];
-		    goto L_jump_start;
+		    continue; /* big loop */
 		}
 		break;
 	    }
@@ -4303,11 +4227,40 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
 	    erl_exit(1,"Internal data structure error (in encode_size_struct2)%x\n",
 		     obj);
 	}
+
+	if (WSTACK_ISEMPTY(s)) {
+	    break;
+	}
+	obj = (Eterm) WSTACK_POP(s);
+
+        if (is_header(obj)) {
+            switch (obj) {
+	    case LIST_TAIL_OP:
+		obj = (Eterm) WSTACK_POP(s);
+		if (is_list(obj)) {
+		    Eterm* cons = list_val(obj);
+
+		    WSTACK_PUSH2(s, (UWord)CDR(cons), (UWord)LIST_TAIL_OP);
+		    obj = CAR(cons);
+		}
+		break;
+
+	    case TERM_ARRAY_OP(1):
+		obj = *(Eterm*)WSTACK_POP(s);
+		break;
+	    default: { /* TERM_ARRAY_OP(N) when N > 1 */
+		Eterm* ptr = (Eterm*) WSTACK_POP(s);
+		WSTACK_PUSH2(s, (UWord) (ptr+1),
+			        (UWord) TERM_ARRAY_OP_DEC(obj));
+		obj = *ptr;
+	    }
+	    }
+	}
     }
 
-    DESTROY_ESTACK(s);
+    WSTACK_DESTROY(s);
     if (ctx) {
-	ASSERT(ctx->estack.start == NULL);
+	ASSERT(ctx->wstack.wstart == NULL);
 	*reds = r;
     }
     *res = result;
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index 2dd81e3ca3..bf6d9fff50 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -2911,7 +2911,7 @@ static int cmp_atoms(Eterm a, Eterm b)
  */
 Sint cmp(Eterm a, Eterm b)
 {
-    return erts_cmp(a, b, 0);
+    return erts_cmp(a, b, 0, 0);
 }
 #endif
 
@@ -2920,9 +2920,10 @@ Sint cmp(Eterm a, Eterm b)
  * exact = 0 -> arith-based compare
  */
 #if HALFWORD_HEAP
-Sint erts_cmp_rel_opt(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base, int exact)
+Sint erts_cmp_rel_opt(Eterm a, Eterm* a_base, Eterm b, Eterm* b_base,
+                      int exact, int eq_only)
 #else
-Sint erts_cmp(Eterm a, Eterm b, int exact)
+Sint erts_cmp(Eterm a, Eterm b, int exact, int eq_only)
 #endif
 {
 #define PSTACK_TYPE struct erts_cmp_hashmap_state
@@ -3742,7 +3743,7 @@ pop_next:
     return 0;
 
 not_equal:
-    if (!PSTACK_IS_EMPTY(hmap_stack)) {
+    if (!PSTACK_IS_EMPTY(hmap_stack) && !eq_only) {
         WSTACK_ROLLBACK(stack, PSTACK_TOP(hmap_stack)->wstack_rollback);
         goto pop_next;
     }