Message ID | 20140325020430.GE18201@bubble.grove.modra.org |
---|---|
State | New |
Headers | show |
On 24-03-2014 23:04, Alan Modra wrote: > Typo fix. Discovered when looking at some disassembly, and wondering > why this disassembled as dcbtt. My simple benchmark testing didn't > show any effect of using the wrong hint, but at a guess you'd only see > a difference on a very long memrchr with cold caches. > > * sysdeps/powerpc/powerpc64/power7/memrchr.S: Correct stream hint. > > diff --git a/sysdeps/powerpc/powerpc64/power7/memrchr.S b/sysdeps/powerpc/powerpc64/power7/memrchr.S > index 40e436f..0c01ca2 100644 > --- a/sysdeps/powerpc/powerpc64/power7/memrchr.S > +++ b/sysdeps/powerpc/powerpc64/power7/memrchr.S > @@ -29,7 +29,7 @@ ENTRY (__memrchr) > mr r10,r3 > clrrdi r6,r7,7 > li r9,3<<5 > - dcbt r9,r6,16 /* Stream hint, decreasing addresses. */ > + dcbt r9,r6,8 /* Stream hint, decreasing addresses. */ > > /* Replicate BYTE to doubleword. */ > insrdi r4,r4,8,48 > This patch is ok, thanks Alan.
diff --git a/sysdeps/powerpc/powerpc64/power7/memrchr.S b/sysdeps/powerpc/powerpc64/power7/memrchr.S index 40e436f..0c01ca2 100644 --- a/sysdeps/powerpc/powerpc64/power7/memrchr.S +++ b/sysdeps/powerpc/powerpc64/power7/memrchr.S @@ -29,7 +29,7 @@ ENTRY (__memrchr) mr r10,r3 clrrdi r6,r7,7 li r9,3<<5 - dcbt r9,r6,16 /* Stream hint, decreasing addresses. */ + dcbt r9,r6,8 /* Stream hint, decreasing addresses. */ /* Replicate BYTE to doubleword. */ insrdi r4,r4,8,48