s390: optimize memset implementation

Like for the memset16/32/64 variants avoid that subsequent mvc
instructions depend on each other since that might have negative
performance impacts.

This patch is currently hardly relevant since at least gcc 7.1
generates only inline memset code and not a single memset call.
However there is no reason to not provide an optimized version
just in case gcc generates memset calls again, like it did in
the past.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

diff --git a/arch/s390/lib/mem.S b/arch/s390/lib/mem.S
index f88cf6983849bb30f0b2b9d61b434ad36b8e7a79..9255a087fa96d3114265b509cc0fe94368261e0c 100644 (file)
--- a/arch/s390/lib/mem.S
+++ b/arch/s390/lib/mem.S
@@ -78,21 +78,25 @@ ENTRY(memset)
        ex      %r4,0(%r3)
        br      %r14
 .Lmemset_fill:
-       stc     %r3,0(%r2)
        cghi    %r4,1
        lgr     %r1,%r2
-       ber     %r14
+       je      .Lmemset_fill_exit
        aghi    %r4,-2
-       srlg    %r3,%r4,8
-       ltgr    %r3,%r3
+       srlg    %r5,%r4,8
+       ltgr    %r5,%r5
        jz      .Lmemset_fill_remainder
 .Lmemset_fill_loop:
-       mvc     1(256,%r1),0(%r1)
+       stc     %r3,0(%r1)
+       mvc     1(255,%r1),0(%r1)
        la      %r1,256(%r1)
-       brctg   %r3,.Lmemset_fill_loop
+       brctg   %r5,.Lmemset_fill_loop
 .Lmemset_fill_remainder:
-       larl    %r3,.Lmemset_mvc
-       ex      %r4,0(%r3)
+       stc     %r3,0(%r1)
+       larl    %r5,.Lmemset_mvc
+       ex      %r4,0(%r5)
+       br      %r14
+.Lmemset_fill_exit:
+       stc     %r3,0(%r1)
        br      %r14
 .Lmemset_xc:
        xc      0(1,%r1),0(%r1)