From 17e0374d393ce446fd1ec5aac38da1ff31d2ac7e Mon Sep 17 00:00:00 2001
From: Taylor R Campbell <campbell@mumble.net>
Date: Sat, 29 Dec 2018 01:11:00 +0000
Subject: [PATCH] Use CALL/RET for pushing and returning to continuations on
 amd64.

Calls now look like:

  ;; (assign (register #x123) (cons-pointer tag (entry:continuation cont)))
  (CALL (@PCR pushed))
  (JMP (@PCR cont))
pushed:
  (OR Q (@R ,rsp) (&U ,tag))
  ...
  (JMP (@PCR uuo-link))

Returns now look like:

  ;; (pop-return)
  (AND Q (@R ,rsp) (R ,regnum:datum-mask))
  (RET)

These should happen in pairs, so that we can take advantage of the
CPU's return address branch target prediction stack rather than
abusing the indirect jump branch target predictor.

WARNING: This changes the amd64 compiled code interface, so new
compiled code requires a new microcode.  (A new microcode might be
able to handle existing compiled code just fine.)
---
 src/compiler/machines/x86-64/rules1.scm | 31 ++++++++++++++++++++++---
 src/compiler/machines/x86-64/rules3.scm | 10 ++++----
 src/microcode/cmpintmd/x86-64.c         | 11 +++++++++
 src/microcode/cmpintmd/x86-64.h         |  4 +++-
 4 files changed, 46 insertions(+), 10 deletions(-)

diff --git a/src/compiler/machines/x86-64/rules1.scm b/src/compiler/machines/x86-64/rules1.scm
index a327bcb15..5f6acf4d0 100644
--- a/src/compiler/machines/x86-64/rules1.scm
+++ b/src/compiler/machines/x86-64/rules1.scm
@@ -175,7 +175,12 @@ USA.
 
 (define-rule statement
   (ASSIGN (REGISTER (? target)) (ENTRY:CONTINUATION (? label)))
-  (load-pc-relative-address (target-register-reference target) label 8))
+  (let* ((target (target-register-reference target))
+	 (get-pc (generate-label 'GET-PC)))
+    (LAP (CALL (@PCR ,get-pc))
+	 (JMP (@PCRO ,label 8))
+	(LABEL ,get-pc)
+	 (POP Q ,target))))
 
 (define-rule statement
   ;; This is an intermediate rule -- not intended to produce code.
@@ -195,8 +200,28 @@ USA.
 	  (CONS-POINTER (MACHINE-CONSTANT (? type))
 			(ENTRY:CONTINUATION (? label))))
   (assert (= type type-code:compiled-return))
-  (load-pc-relative-address/typed (target-register-reference target)
-				  type label 8))
+  (let* ((target (target-register-reference target))
+	 (temp (temporary-register-reference))
+	 (pushed (generate-label 'PUSHED)))
+    (LAP (CALL (@PCR ,pushed))
+	 (JMP (@PCRO ,label 8))
+	(LABEL ,pushed)
+	 (POP Q ,target)
+	 (MOV Q ,temp (&U ,(make-non-pointer-literal type 0)))
+	 (OR Q ,target ,temp))))
+
+(define-rule statement
+  (ASSIGN (PRE-INCREMENT (REGISTER 4) -1)
+	  (CONS-POINTER (MACHINE-CONSTANT (? type))
+			(ENTRY:CONTINUATION (? label))))
+  (assert (= type type-code:compiled-return))
+  (let* ((temp (temporary-register-reference))
+	 (pushed (generate-label 'PUSHED)))
+    (LAP (CALL (@PCR ,pushed))
+	 (JMP (@PCRO ,label 8))
+	(LABEL ,pushed)
+	 (MOV Q ,temp (&U ,(make-non-pointer-literal type 0)))
+	 (OR Q (@R 4) ,temp))))
 
 (define-rule statement
   (ASSIGN (REGISTER (? target)) (VARIABLE-CACHE (? name)))
diff --git a/src/compiler/machines/x86-64/rules3.scm b/src/compiler/machines/x86-64/rules3.scm
index 8f49c4707..74c747be8 100644
--- a/src/compiler/machines/x86-64/rules3.scm
+++ b/src/compiler/machines/x86-64/rules3.scm
@@ -39,9 +39,8 @@ USA.
     (cond ((null? checks)
 	   (current-bblock-continue!
 	    (make-new-sblock
-	     (LAP (POP Q (R ,rax))	; continuation
-		  (AND Q (R ,rax) (R ,regnum:datum-mask)) ; clear type
-		  (JMP (R ,rax))))))
+	     (LAP (AND Q (@R ,rsp) (R ,regnum:datum-mask))
+		  (RET)))))
 	  ((block-association 'POP-RETURN)
 	   => current-bblock-continue!)
 	  (else
@@ -50,9 +49,8 @@ USA.
 		   (let ((interrupt-label (generate-label 'INTERRUPT)))
 		     (LAP (CMP Q (R ,regnum:free-pointer) ,reg:compiled-memtop)
 			  (JGE (@PCR ,interrupt-label))
-			  (POP Q (R ,rax)) ; continuation
-			  (AND Q (R ,rax) (R ,regnum:datum-mask)) ; clear type
-			  (JMP (R ,rax))
+			  (AND Q (@R ,rsp) (R ,regnum:datum-mask))
+			  (RET)
 			  (LABEL ,interrupt-label)
 			  ,@(invoke-hook
 			     entry:compiler-interrupt-continuation-2))))))
diff --git a/src/microcode/cmpintmd/x86-64.c b/src/microcode/cmpintmd/x86-64.c
index be21835ae..19e168e3b 100644
--- a/src/microcode/cmpintmd/x86-64.c
+++ b/src/microcode/cmpintmd/x86-64.c
@@ -63,6 +63,17 @@ write_cc_entry_offset (cc_entry_offset_t * ceo, insn_t * address)
     = (((ceo->offset) << 1) | ((ceo->continued_p) ? 1 : 0));
   return (false);
 }
+
+insn_t *
+cc_return_address_to_entry_address (insn_t * pc)
+{
+  if ((pc[0]) == 0xeb)		/* JMP rel8 */
+    return ((pc + 2) + (* ((int8_t *) &pc[1])) - 8);
+  else if ((pc[0]) == 0xe9)	/* JMP rel32 */
+    return ((pc + 5) + (* ((int32_t *) &pc[1])) - 8);
+  else
+    return (pc - 8);
+}
 
 /* Compiled closures */
 
diff --git a/src/microcode/cmpintmd/x86-64.h b/src/microcode/cmpintmd/x86-64.h
index d48aee366..4acce5472 100644
--- a/src/microcode/cmpintmd/x86-64.h
+++ b/src/microcode/cmpintmd/x86-64.h
@@ -161,7 +161,9 @@ typedef uint8_t insn_t;
 #define CC_RETURN_ADDRESS_PTR(r)	0
 #define CC_RETURN_ADDRESS_PC(r)		(r)
 
-#define CC_RETURN_ADDRESS_TO_ENTRY_ADDRESS(r)	(((insn_t *) (r)) - 8)
+insn_t * cc_return_address_to_entry_address (insn_t *);
+
+#define CC_RETURN_ADDRESS_TO_ENTRY_ADDRESS cc_return_address_to_entry_address
 
 #define EMBEDDED_CLOSURE_ADDRS_P 1
 
-- 
2.25.1