Optimize continuation pointer management

The BEAM instructions for calling a function don't save the
continuation pointer (return address) on the stack, but to a special
BEAM register called CP. It is the responsibility of the called
function to save CP to the stack frame before calling other functions.

In the earlier implementations of BEAM on Sparc, CP was located in a
CPU register. That meant that the continuation pointer was never
written to memory when calling simple functions that didn't call
other functions at all or ended in a tail-call to another function.

The modern BEAM no longer keeps CP in CPU register. Instead, it is
kept in the `process` struct (in `p->cp`). That means the continuation
pointer must be written to the memory on every call, and if the called
function will call other functions, it will must read the continuation
pointer from `p->cp` and store it on the stack.

This commit eliminates the concept of the CP register and modifies
the call instructions to directly store the continuation pointer on
the stack. That makes allocation and trimming of stack frames slightly
faster. A more important benefit is simplification of code that handles
continuation pointers. Because all continuation pointers are now stored
on the stack, the special case of handling `p->cp` disappears.

Co-authored-by: John Högberg <[email protected]>

1 files changed, 26 insertions, 25 deletions

diff --git a/erts/emulator/beam/instrs.tab b/erts/emulator/beam/instrs.tab
index bc8c1189a8..efdba73057 100644
--- a/erts/emulator/beam/instrs.tab
+++ b/erts/emulator/beam/instrs.tab
@@ -19,7 +19,12 @@
 // %CopyrightEnd%
 //
 
-// Stack manipulation instructions
+//
+// Stack manipulation instructions follow.
+//
+// See the comment for AH() in macros.tab for information about
+// the layout of stack frames.
+//
 
 allocate(NeedStack, Live) {
     $AH($NeedStack, 0, $Live);
@@ -58,15 +63,14 @@ allocate_heap_zero(NeedStack, NeedHeap, Live) {
 
 deallocate(Deallocate) {
     //| -no_prefetch
-    SET_CP(c_p, (BeamInstr *) cp_val(*E));
     E = ADD_BYTE_OFFSET(E, $Deallocate);
 }
 
 deallocate_return(Deallocate) {
     //| -no_next
     int words_to_pop = $Deallocate;
-    SET_I((BeamInstr *) cp_val(*E));
     E = ADD_BYTE_OFFSET(E, words_to_pop);
+    $RETURN();
     CHECK_TERM(x(0));
     DispatchReturn;
 }
@@ -93,13 +97,13 @@ DISPATCH_ABS(CallDest) {
 }
 
 i_call(CallDest) {
-    SET_CP(c_p, $NEXT_INSTRUCTION);
+    $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
     $DISPATCH_REL($CallDest);
 }
 
 move_call(Src, CallDest) {
     x(0) = $Src;
-    SET_CP(c_p, $NEXT_INSTRUCTION);
+    $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
     $DISPATCH_REL($CallDest);
 }
 
@@ -131,7 +135,7 @@ DISPATCHX(Dest) {
 }
 
 i_call_ext(Dest) {
-    SET_CP(c_p, $NEXT_INSTRUCTION);
+    $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
     $DISPATCHX($Dest);
 }
 
@@ -175,7 +179,7 @@ i_apply() {
     BeamInstr *next;
     $APPLY(NULL, 0, next);
     if (ERTS_LIKELY(next != NULL)) {
-        SET_CP(c_p, $NEXT_INSTRUCTION);
+        $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
         $DISPATCH_ABS(next);
     }
     $HANDLE_APPLY_ERROR();
@@ -211,7 +215,7 @@ apply(Arity) {
     BeamInstr *next;
     $FIXED_APPLY($Arity, NULL, 0, next);
     if (ERTS_LIKELY(next != NULL)) {
-        SET_CP(c_p, $NEXT_INSTRUCTION);
+        $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
         $DISPATCH_ABS(next);
     }
     $HANDLE_APPLY_ERROR();
@@ -247,7 +251,7 @@ i_apply_fun() {
     BeamInstr *next;
     $APPLY_FUN(next);
     if (ERTS_LIKELY(next != NULL)) {
-        SET_CP(c_p, $NEXT_INSTRUCTION);
+        $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
         $DISPATCH_FUN(next);
     }
     $HANDLE_APPLY_FUN_ERROR();
@@ -283,7 +287,7 @@ i_call_fun(Fun) {
     BeamInstr *next;
     $CALL_FUN($Fun, next);
     if (ERTS_LIKELY(next != NULL)) {
-        SET_CP(c_p, $NEXT_INSTRUCTION);
+        $SAVE_CONTINUATION_POINTER($NEXT_INSTRUCTION);
         $DISPATCH_FUN(next);
     }
     $HANDLE_APPLY_FUN_ERROR();
@@ -301,15 +305,8 @@ i_call_fun_last(Fun, Deallocate) {
 
 return() {
     //| -no_next
-    SET_I(c_p->cp);
+    $RETURN();
     DTRACE_RETURN_FROM_PC(c_p);
-
-    /*
-     * We must clear the CP to make sure that a stale value do not
-     * create a false module dependcy preventing code upgrading.
-     * It also means that we can use the CP in stack backtraces.
-     */
-    c_p->cp = 0;
     CHECK_TERM(r(0));
     HEAP_SPACE_VERIFIED(0);
     DispatchReturn;
@@ -478,16 +475,21 @@ i_make_fun(FunP, NumFree) {
 }
 
 move_trim(Src, Dst, Words) {
-    Uint cp = E[0];
     $Dst = $Src;
-    E += $Words;
-    E[0] = cp;
+    $i_trim($Words);
 }
 
 i_trim(Words) {
-    Uint cp = E[0];
     E += $Words;
-    E[0] = cp;
+
+    /*
+     * Clear the reserved location for the continuation pointer at
+     * E[0]. This is not strictly necessary for correctness, but if a
+     * GC is triggered before E[0] is overwritten by another
+     * continuation pointer the now dead term at E[0] would be
+     * retained by the GC.
+     */
+    E[0] = NIL;
 }
 
 move(Src, Dst) {
@@ -599,8 +601,7 @@ move_window5(S1, S2, S3, S4, S5, D) {
 move_return(Src) {
     //| -no_next
     x(0) = $Src;
-    SET_I(c_p->cp);
-    c_p->cp = 0;
+    $RETURN();
     DispatchReturn;
 }