From: Stas Boukarev <stassats@gmail.com>
Date: Wed, 4 Sep 2013 09:30:17 +0000 (+0400)
Subject: Micro-optimize calling alloc_tramp on x86_64.
X-Git-Url: http://repo.macrolet.net/gitweb/?a=commitdiff_plain;h=aa7b669779e8e88349938ca962229f31ead08af2;p=sbcl.git

Micro-optimize calling alloc_tramp on x86_64.

In allocation-tramp, instead of using R11 temp-reg for storing the
address of alloc_tramp, use alloc-tn, because the result of
alloc_tramp will be stored there and its value doesn't have to be
preserved. And most of the time it will be allocated to a
non-64-bit-mode register, allowing to save 2 bytes on encoding.
---

diff --git a/src/compiler/x86-64/macros.lisp b/src/compiler/x86-64/macros.lisp
index fe896aa..781a270 100644
--- a/src/compiler/x86-64/macros.lisp
+++ b/src/compiler/x86-64/macros.lisp
@@ -168,8 +168,8 @@
 ;;; object.
 (defun allocation-tramp (alloc-tn size lowtag)
   (inst push size)
-  (inst mov temp-reg-tn (make-fixup "alloc_tramp" :foreign))
-  (inst call temp-reg-tn)
+  (inst mov alloc-tn (make-fixup "alloc_tramp" :foreign))
+  (inst call alloc-tn)
   (inst pop alloc-tn)
   (when lowtag
     (inst lea alloc-tn (make-ea :byte :base alloc-tn :disp lowtag)))