Eliminate use of GetArg1() in the fast_element instruction

Use separate instructions for each register type.

3 files changed, 40 insertions, 29 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 1b9fc86871..bdc730e6c1 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -1492,24 +1492,32 @@ void process_main(void)
     c_p->freason = BADARG;
     goto lb_Cl_error;
 
- OpCase(i_fast_element_jIsd): {
-     Eterm tuple;
-
-     /*
-      * Inlined version of element/2 for even more speed.
-      * The first argument is an untagged integer >= 1.
-      * The second argument is guaranteed to be a register operand.
-      */
-     GetArg1(2, tuple);
-     if (is_tuple(tuple)) {
-	 Eterm* tp = tuple_val(tuple);
-	 Eterm pos = Arg(1);
-	 if (pos <= arityval(*tp)) {
-	     Eterm result = tp[pos];
-	     StoreBifResult(3, result);
-	 }
-     }
+    {
+	Eterm fast_element_tuple;
+
+    OpCase(i_fast_element_rjId):
+	fast_element_tuple = r(0);
+
+    do_fast_element:
+	if (is_tuple(fast_element_tuple)) {
+	    Eterm* tp = tuple_val(fast_element_tuple);
+	    Eterm pos = Arg(1);	/* Untagged integer >= 1 */
+	    if (pos <= arityval(*tp)) {
+		Eterm result = tp[pos];
+		StoreBifResult(2, 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):
diff --git a/erts/emulator/beam/beam_load.c b/erts/emulator/beam/beam_load.c
index 54c8ad0eb1..4233e26f54 100644
--- a/erts/emulator/beam/beam_load.c
+++ b/erts/emulator/beam/beam_load.c
@@ -2308,23 +2308,23 @@ gen_element(LoaderState* stp, GenOpArg Fail, GenOpArg Index,
     GenOp* op;
 
     NEW_GENOP(stp, op);
-    op->op = genop_i_element_4;
     op->arity = 4;
-    op->a[0] = Fail;
-    op->a[1] = Index;
-    op->a[2] = Tuple;
-    op->a[3] = Dst;
     op->next = NULL;
 
-    /*
-     * If safe, generate a faster instruction.
-     */
-
     if (Index.type == TAG_i && Index.val > 0 &&
 	(Tuple.type == TAG_r || Tuple.type == TAG_x || Tuple.type == TAG_y)) {
 	op->op = genop_i_fast_element_4;
-	op->a[1].type = TAG_u;
-	op->a[1].val = Index.val;
+	op->a[0] = Tuple;
+	op->a[1] = Fail;
+	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] = Fail;
+	op->a[1] = Index;
+	op->a[2] = Tuple;
+	op->a[3] = Dst;
     }
 
     return op;
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index f158fa9543..491f8f1d90 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -920,7 +920,10 @@ node x
 node y
 %hot
 
-i_fast_element j I s d
+i_fast_element r j I d
+i_fast_element x j I d
+i_fast_element y j I d
+
 i_element j s s d
 
 bif1 f b s d