This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
PATCH: PR libc/10162: Properly recover from shorter read
- From: "H.J. Lu" <hongjiu dot lu at intel dot com>
- To: "H.J. Lu" <hjl dot tools at gmail dot com>
- Cc: GNU C Library <libc-alpha at sourceware dot org>
- Date: Wed, 18 Nov 2009 11:29:13 -0800
- Subject: PATCH: PR libc/10162: Properly recover from shorter read
- References: <20091109195706.GA29166@lucon.org> <20091109203741.GA29411@lucon.org>
- Reply-to: "H.J. Lu" <hjl dot tools at gmail dot com>
On Mon, Nov 09, 2009 at 12:37:41PM -0800, H.J. Lu wrote:
> On Mon, Nov 09, 2009 at 11:57:06AM -0800, H.J. Lu wrote:
> > When data is shorter than software pipeline, recovery may fail. This
> > patch avoids it by using a simple loop on data shorter than software
> > pipeline.
> >
>
> Here is an updated patch. This one properly handles the shorter read
> on speculative load.
>
Here is the updated patch to remove the previous patch and properly
recover from shorter read.
H.J.
---
2009-11-18 H.J. Lu <hongjiu.lu@intel.com>
PR libc/10162
* sysdeps/ia64/memchr.S: Don't use a simple loop on data shorter
than software pipeline. Properly recover from shorter read.
diff --git a/sysdeps/ia64/memchr.S b/sysdeps/ia64/memchr.S
index e721e17..cdd71ca 100644
--- a/sysdeps/ia64/memchr.S
+++ b/sysdeps/ia64/memchr.S
@@ -71,8 +71,7 @@ ENTRY(__memchr)
cmp.ne p7, p0 = r0, r0 // clear p7
extr.u chr = in1, 0, 8 // chr = (unsigned char) in1
mov len = in2
- // use a simple loop on data shorter than software pipeline
- cmp.gtu p6, p0 = ((MEMLAT + 1) * 8), in2
+ cmp.gtu p6, p0 = 16, in2 // use a simple loop for short
(p6) br.cond.spnt .srchfew ;; // searches
sub loopcnt = 8, tmp // loopcnt = 8 - tmp
cmp.eq p6, p0 = tmp, r0
@@ -127,7 +126,16 @@ ENTRY(__memchr)
br.ret.sptk.many b0
.recovery:
- adds ret0 = -((MEMLAT + 1) * 8), ret0;;
+#if MEMLAT != 6
+# error "MEMLAT must be 6!"
+#endif
+(p[MEMLAT-6]) add ret0 = -8, ret0;;
+(p[MEMLAT-5]) add ret0 = -8, ret0;;
+(p[MEMLAT-4]) add ret0 = -8, ret0;;
+(p[MEMLAT-3]) add ret0 = -8, ret0;;
+(p[MEMLAT-2]) add ret0 = -8, ret0;;
+(p[MEMLAT-1]) add ret0 = -8, ret0;;
+(p[MEMLAT]) add ret0 = -8, ret0;;
(p[MEMLAT+1]) add ret0 = -8, ret0;;
(p[MEMLAT+2]) add ret0 = -8, ret0;;
.l4: