about summary refs log tree commit diff
path: root/sysdeps/s390/s390-32/bcopy.S
diff options
context:
space:
mode:
authorUlrich Drepper <drepper@redhat.com>2001-07-16 06:29:42 +0000
committerUlrich Drepper <drepper@redhat.com>2001-07-16 06:29:42 +0000
commit3c204435e91ba20fbfef438e52fdc147e679ca58 (patch)
treef3cdbfd9405e6844b14d51b05974284ed902099d /sysdeps/s390/s390-32/bcopy.S
parenta6d214835f18ce3232d44ed6f0aba679d090693d (diff)
downloadglibc-3c204435e91ba20fbfef438e52fdc147e679ca58.tar.gz
glibc-3c204435e91ba20fbfef438e52fdc147e679ca58.tar.xz
glibc-3c204435e91ba20fbfef438e52fdc147e679ca58.zip
Use mvcle for big blocks (> 64K) and a mvc loop for small blocks.
Diffstat (limited to 'sysdeps/s390/s390-32/bcopy.S')
-rw-r--r--sysdeps/s390/s390-32/bcopy.S27
1 files changed, 19 insertions, 8 deletions
diff --git a/sysdeps/s390/s390-32/bcopy.S b/sysdeps/s390/s390-32/bcopy.S
index d3ecf1d5e7..ae90dc158e 100644
--- a/sysdeps/s390/s390-32/bcopy.S
+++ b/sysdeps/s390/s390-32/bcopy.S
@@ -34,12 +34,14 @@ ENTRY(__bcopy)
         jnl     .L0
         alr     %r1,%r2
         clr     %r1,%r3
-        jh      .L5
+        jh      .L7
 .L0:	ahi     %r4,-1              # length - 1
 	lr      %r1,%r4
 	srl     %r1,8
-	ltr     %r1,%r1
+	ltr     %r1,%r1             # < 256 bytes to move ?
 	jz      .L2
+	chi     %r1,255             # > 1MB to move ?
+	jh      .L5
 .L1:	mvc     0(256,%r3),0(%r2)   # move in 256 byte chunks
 	la      %r2,256(%r2)
 	la      %r3,256(%r3)
@@ -49,22 +51,31 @@ ENTRY(__bcopy)
 .L3:	ex      %r4,0(%r1)          # execute mvc with length ((%r4)&255)+1
 .L4:	br      %r14
 
-.L5:                                # destructive overlay, can not use mvcle
+	# data copies > 1MB are faster with mvcle.
+.L5:	ahi     %r4,1               # length + 1
+	lr      %r5,%r4	            # source length
+	lr	%r4,%r2             # source address
+	lr	%r2,%r3             # set destination
+	lr	%r3,%r5             # destination length = source length
+.L6:	mvcle	%r2,%r4,0           # thats it, MVCLE is your friend
+	jo	.L6
+	br	%r14
+.L7:                                # destructive overlay, can not use mvcle
         lr     %r1,%r2              # bcopy is called with source,dest
         lr     %r2,%r3              # memmove with dest,source! Oh, well...
         lr     %r3,%r1
         basr   %r1,0
-.L6:
+.L8:
 #ifdef PIC
-        al     %r1,.L7-.L6(%r1)     # get address of global offset table
+        al     %r1,.L9-.L8(%r1)     # get address of global offset table
                                     # load address of memmove
         l      %r1,memmove@GOT12(%r1)
         br     %r1
-.L7:    .long  _GLOBAL_OFFSET_TABLE_-.L6
+.L9:    .long  _GLOBAL_OFFSET_TABLE_-.L8
 #else
-        al     %r1,.L7-.L6(%r1)     # load address of memmove
+        al     %r1,.L9-.L8(%r1)     # load address of memmove
         br     %r1                  # jump to memmove
-.L7:    .long  memmove-.L6
+.L9:    .long  memmove-.L8
 #endif
 
 END(__bcopy)