From: Julian Seward <jseward@acm.org>
Date: Sat, 26 Mar 2005 12:57:39 +0000 (+0000)
Subject: Update comment.
X-Git-Tag: svn/VALGRIND_3_0_1^2~241
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=db8a81d3dc3b21861c4e4724875e31a037ff1792;p=thirdparty%2Fvalgrind.git

Update comment.

Stare at sanity checks but still fail to figure out how to make them
cheaper.  The register allocator's sanity checks consume 15%-20% of
the total running time of Vex.



git-svn-id: svn://svn.valgrind.org/vex/trunk@1093
---

diff --git a/VEX/priv/host-generic/reg_alloc2.c b/VEX/priv/host-generic/reg_alloc2.c
index ecc82551fe..fbbde4ec1a 100644
--- a/VEX/priv/host-generic/reg_alloc2.c
+++ b/VEX/priv/host-generic/reg_alloc2.c
@@ -618,8 +618,7 @@ HInstrArray* doRegisterAllocation (
    /* --------- Stage 3: allocate spill slots. --------- */
 
    /* Each spill slot is 8 bytes long.  For 128-bit vregs
-      we'll have to allocate two spill slots.  For now, tho,
-      ignore the 128-bit problem.
+      we have to allocate two spill slots.
 
       Do a rank-based allocation of vregs to spill slot numbers.  We
       put as few values as possible in spill slows, but nevertheless
@@ -641,7 +640,9 @@ HInstrArray* doRegisterAllocation (
 
       /* The spill slots are 64 bits in size.  That means, to spill a
          Vec128-class vreg, we'll need to find two adjacent spill
-         slots to use.  */
+         slots to use.  Note, this special-casing needs to happen for
+         all 128-bit sized register classes.  Currently though
+         HRcVector is the only such class. */
 
       if (vreg_lrs[j].reg_class != HRcVec128) {