S/390: Add support for STT_GNU_IFUNC symbols.

Add support for STT_GNU_IFUNC symbols and the new R_390_IRELATIVE
relocation.  Provide optimized version of memcpy, memset, and memcmp
for z10 and z196.

1 files changed, 55 insertions, 31 deletions

diff --git a/sysdeps/s390/s390-64/memcpy.S b/sysdeps/s390/s390-64/memcpy.S
index be916fe..1e5f050 100644
--- a/sysdeps/s390/s390-64/memcpy.S
+++ b/sysdeps/s390/s390-64/memcpy.S
@@ -1,7 +1,6 @@
 /* memcpy - copy a block from source to destination.  64 bit S/390 version.
-   Copyright (C) 2000, 2001, 2003 Free Software Foundation, Inc.
+   Copyright (C) 2012 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
-   Contributed by Martin Schwidefsky (schwidefsky@de.ibm.com).
 
    The GNU C Library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
@@ -17,41 +16,66 @@
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
+
+#include "sysdep.h"
+#include "asm-syntax.h"
+
 /* INPUT PARAMETERS
      %r2 = address of destination memory area
      %r3 = address of source memory area
      %r4 = number of bytes to copy.  */
 
-#include "sysdep.h"
-#include "asm-syntax.h"
 
-	.text
+       .text
+
+#ifdef USE_MULTIARCH
+ENTRY(memcpy_z900)
+#else
 ENTRY(memcpy)
-	ltgr	%r4,%r4
-	jz	.L3
-	aghi	%r4,-1		    # length - 1
-	lgr	%r1,%r2		    # copy destination address
-	srlg	%r5,%r4,8
-	ltgr    %r5,%r5             # < 256 bytes to mvoe ?
-	jz	.L1
-	chi     %r5,255             # > 1 MB to move ?
-	jh	.L4
-.L0:	mvc	0(256,%r1),0(%r3)   # move in 256 byte chunks
-	la	%r1,256(%r1)
-	la	%r3,256(%r3)
-	brctg	%r5,.L0
-.L1:	bras	%r5,.L2		    # setup base pointer for execute
-	mvc	0(1,%r1),0(%r3)	    # instruction for execute
-.L2:	ex	%r4,0(%r5)	    # execute mvc with length ((%r4)&255)+1
-.L3:	br	%r14
-	# data copies > 1MB are faster with mvcle.
-.L4:	aghi    %r4,1               # length + 1
-	lgr	%r5,%r4	            # source length
-	lgr	%r4,%r3             # source address
-	lgr	%r3,%r5             # destination length = source length
-.L5:	mvcle	%r2,%r4,0           # thats it, MVCLE is your friend
-	jo	.L5
-	lgr     %r2,%r1             # return destination address
-	br	%r14
+#endif
+	.machine "z900"
+        ltgr    %r4,%r4
+        je      .L_Z900_4
+        aghi    %r4,-1
+        srlg    %r5,%r4,8
+        ltgr    %r5,%r5
+        lgr     %r1,%r2
+        jne     .L_Z900_13
+.L_Z900_3:
+        larl    %r5,.L_Z900_15
+        ex      %r4,0(%r5)
+.L_Z900_4:
+        br      %r14
+.L_Z900_13:
+        chi	%r5,4096             # Switch to mvcle for copies >1MB
+        jh      memcpy_mvcle
+.L_Z900_12:
+        mvc     0(256,%r1),0(%r3)
+        la      %r1,256(%r1)
+        la      %r3,256(%r3)
+        brctg   %r5,.L_Z900_12
+        j       .L_Z900_3
+.L_Z900_15:
+        mvc     0(1,%r1),0(%r3)
+
+#ifdef USE_MULTIARCH
+END(memcpy_z900)
+#else
 END(memcpy)
 libc_hidden_builtin_def (memcpy)
+#endif
+
+ENTRY(memcpy_mvcle)
+       # Using as standalone function will result in unexpected
+       # results since the length field is incremented by 1 in order to
+       # compensate the changes already done in the functions above.
+       aghi    %r4,1               # length + 1
+       lgr     %r5,%r4             # source length
+       lgr     %r4,%r3             # source address
+       lgr     %r3,%r5             # destination length = source length
+.L_MVCLE_1:
+       mvcle   %r2,%r4,0           # thats it, MVCLE is your friend
+       jo      .L_MVCLE_1
+       lgr     %r2,%r1             # return destination address
+       br      %r14
+END(memcpy_mvcle)