bca718
    Backport of the following patch as a prerequistite for
bca718
    96d6fd6c4060d739abb1822e7ad633af749532b2:
bca718
    commit 69f13dbf06c6195de0ada8632271d58ca3cf55da
bca718
    Author: Adhemerval Zanella <azanella@linux.vnet.ibm.com>
bca718
    Date:   Thu Sep 26 09:29:19 2013 -0500
bca718
    
bca718
        PowerPC: strcpy/stpcpy optimization for PPC64/POWER7
bca718
    
bca718
        This patch intends to unify both strcpy and stpcpy implementationsi
bca718
        for PPC64 and PPC64/POWER7. The idead default powerpc64 implementation
bca718
        is to provide both doubleword and word aligned memory access.
bca718
    
bca718
        For PPC64/POWER7 is also provide doubleword and word memory access,
bca718
        remove the branch hints, use the cmpb instruction for compare
bca718
        doubleword/words, and add an optimization for inputs of same alignment.
bca718
    
bca718
        ChangeLog:
bca718
    
bca718
            2013-10-04  Adhemerval Zanella  <azanella@linux.vnet.ibm.com>
bca718
    
bca718
            * sysdeps/powerpc/powerpc64/strcpy.S (strcpy): Add word load/store
bca718
            to provide a boost for large inputs with word alignment.
bca718
            * sysdeps/powerpc/powerpc64/stpcpy.S (__stpcpy): Rewrite
bca718
            implementation based on optimized PPC64 strcpy.
bca718
            * sysdeps/powerpc/powerpc64/power7/strcpy.S: New file: optimized
bca718
            strcpy for PPC64/POWER7 based on both doubleword and word load/store.
bca718
            * sysdeps/powerpc/powerpc64/power7/stpcpy.S: New file: optimized
bca718
            stpcpy for PPC64/POWER7 based on PPC64/POWER7 strcpy.
bca718
bca718
diff --git a/sysdeps/powerpc/powerpc64/power7/stpcpy.S b/sysdeps/powerpc/powerpc64/power7/stpcpy.S
bca718
new file mode 100644
bca718
index 0000000..727dd06
bca718
--- /dev/null
bca718
+++ b/sysdeps/powerpc/powerpc64/power7/stpcpy.S
bca718
@@ -0,0 +1,24 @@
bca718
+/* Optimized stpcpy implementation for PowerPC64/POWER7.
bca718
+   Copyright (C) 2013 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#define USE_AS_STPCPY
bca718
+#include <sysdeps/powerpc/powerpc64/power7/strcpy.S>
bca718
+
bca718
+weak_alias (__stpcpy, stpcpy)
bca718
+libc_hidden_def (__stpcpy)
bca718
+libc_hidden_builtin_def (stpcpy)
bca718
diff --git a/sysdeps/powerpc/powerpc64/power7/strcpy.S b/sysdeps/powerpc/powerpc64/power7/strcpy.S
bca718
new file mode 100644
bca718
index 0000000..5c341a1
bca718
--- /dev/null
bca718
+++ b/sysdeps/powerpc/powerpc64/power7/strcpy.S
bca718
@@ -0,0 +1,274 @@
bca718
+/* Optimized strcpy/stpcpy implementation for PowerPC64/POWER7.
bca718
+   Copyright (C) 2013 Free Software Foundation, Inc.
bca718
+   This file is part of the GNU C Library.
bca718
+
bca718
+   The GNU C Library is free software; you can redistribute it and/or
bca718
+   modify it under the terms of the GNU Lesser General Public
bca718
+   License as published by the Free Software Foundation; either
bca718
+   version 2.1 of the License, or (at your option) any later version.
bca718
+
bca718
+   The GNU C Library is distributed in the hope that it will be useful,
bca718
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
bca718
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
bca718
+   Lesser General Public License for more details.
bca718
+
bca718
+   You should have received a copy of the GNU Lesser General Public
bca718
+   License along with the GNU C Library; if not, see
bca718
+   <http://www.gnu.org/licenses/>.  */
bca718
+
bca718
+#include <sysdep.h>
bca718
+
bca718
+/* Implements the function
bca718
+
bca718
+   char * [r3] strcpy (char *dest [r3], const char *src [r4])
bca718
+
bca718
+   or
bca718
+
bca718
+   char * [r3] strcpy (char *dest [r3], const char *src [r4])
bca718
+
bca718
+   if USE_AS_STPCPY is defined. It tries to use aligned memory accesses
bca718
+   when possible using the following algorithm:
bca718
+
bca718
+   if (((((uintptr_t)dst & 0x7UL) == 0) && ((uintptr_t)src & 0x7UL) == 0))
bca718
+     goto aligned_doubleword_copy;
bca718
+   if (((((uintptr_t)dst & 0x3UL) == 0) && ((uintptr_t)src & 0x3UL) == 0))
bca718
+     goto aligned_word_copy;
bca718
+   if (((uintptr_t)dst & 0x7UL) == ((uintptr_t)src & 0x7UL))
bca718
+     goto same_alignment;
bca718
+   goto unaligned;
bca718
+
bca718
+   The aligned comparison are made using cmpb instructions.  */
bca718
+
bca718
+#ifdef USE_AS_STPCPY
bca718
+# define FUNC_NAME __stpcpy
bca718
+#else
bca718
+# define FUNC_NAME strcpy
bca718
+#endif
bca718
+
bca718
+	.machine  power7
bca718
+EALIGN (FUNC_NAME, 4, 0)
bca718
+	CALL_MCOUNT 2
bca718
+
bca718
+#define rTMP	r0
bca718
+#ifdef USE_AS_STPCPY
bca718
+#define rRTN	r3	/* pointer to previous word/doubleword in dest */
bca718
+#else
bca718
+#define rRTN	r12	/* pointer to previous word/doubleword in dest */
bca718
+#endif
bca718
+#define rSRC	r4	/* pointer to previous word/doubleword in src */
bca718
+#define rMASK	r5	/* mask 0xffffffff | 0xffffffffffffffff */
bca718
+#define rWORD	r6	/* current word from src */
bca718
+#define rALT	r7	/* alternate word from src */
bca718
+#define rRTNAL	r8	/* alignment of return pointer */
bca718
+#define rSRCAL	r9	/* alignment of source pointer */
bca718
+#define rALCNT	r10	/* bytes to read to reach 8 bytes alignment */
bca718
+#define rSUBAL	r11	/* doubleword minus unaligned displacement */
bca718
+
bca718
+#ifndef USE_AS_STPCPY
bca718
+/* Save the dst pointer to use as return value.  */
bca718
+	mr	rRTN, r3
bca718
+#endif
bca718
+	or	rTMP, rSRC, rRTN
bca718
+	clrldi.	rTMP, rTMP, 61
bca718
+	bne	L(check_word_alignment)
bca718
+	b	L(aligned_doubleword_copy)
bca718
+
bca718
+L(same_alignment):
bca718
+/* Src and dst with same alignment: align both to doubleword.  */
bca718
+	mr	rALCNT, rRTN
bca718
+	lbz	rWORD, 0(rSRC)
bca718
+	subfic	rSUBAL, rRTNAL, 8
bca718
+	addi	rRTN, rRTN, 1
bca718
+	addi	rSRC, rSRC, 1
bca718
+	cmpdi	cr7, rWORD, 0
bca718
+	stb	rWORD, 0(rALCNT)
bca718
+	beq	cr7, L(s2)
bca718
+
bca718
+	add	rALCNT, rALCNT, rSUBAL
bca718
+	subf	rALCNT, rRTN, rALCNT
bca718
+	addi	rALCNT, rALCNT, 1
bca718
+	mtctr	rALCNT
bca718
+	b	L(s1)
bca718
+
bca718
+	.align 4
bca718
+L(s0):
bca718
+	addi	rSRC, rSRC, 1
bca718
+	lbz	rWORD, -1(rSRC)
bca718
+	cmpdi	cr7, rWORD, 0
bca718
+	stb	rWORD, -1(rALCNT)
bca718
+	beqlr	cr7
bca718
+	mr	rRTN, rALCNT
bca718
+L(s1):
bca718
+	addi	rALCNT, rRTN,1
bca718
+	bdnz	L(s0)
bca718
+	b L(aligned_doubleword_copy)
bca718
+	.align 4
bca718
+L(s2):
bca718
+	mr	rRTN, rALCNT
bca718
+	blr
bca718
+
bca718
+/* For doubleword aligned memory, operate using doubleword load and stores.  */
bca718
+	.align 4
bca718
+L(aligned_doubleword_copy):
bca718
+	li	rMASK, 0
bca718
+	addi	rRTN, rRTN, -8
bca718
+	ld	rWORD, 0(rSRC)
bca718
+	b	L(g2)
bca718
+
bca718
+	.align 4
bca718
+L(g0):	ldu	rALT, 8(rSRC)
bca718
+	stdu	rWORD, 8(rRTN)
bca718
+	cmpb	rTMP, rALT, rMASK
bca718
+	cmpdi	rTMP, 0
bca718
+	bne	L(g1)
bca718
+	ldu	rWORD, 8(rSRC)
bca718
+	stdu	rALT, 8(rRTN)
bca718
+L(g2):	cmpb	rTMP, rWORD, rMASK
bca718
+	cmpdi	rTMP, 0		/* If rTMP is 0, no null's have been found.  */
bca718
+	beq	L(g0)
bca718
+
bca718
+	mr	rALT, rWORD
bca718
+/* We've hit the end of the string.  Do the rest byte-by-byte.  */
bca718
+L(g1):
bca718
+#ifdef __LITTLE_ENDIAN__
bca718
+	extrdi.	rTMP, rALT, 8, 56
bca718
+	stbu	rALT, 8(rRTN)
bca718
+	beqlr-
bca718
+	extrdi.	rTMP, rALT, 8, 48
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	extrdi.	rTMP, rALT, 8, 40
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	extrdi.	rTMP, rALT, 8, 32
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	extrdi.	rTMP, rALT, 8, 24
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	extrdi.	rTMP, rALT, 8, 16
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	extrdi.	rTMP, rALT, 8, 8
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	extrdi	rTMP, rALT, 8, 0
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+#else
bca718
+	extrdi.	rTMP, rALT, 8, 0
bca718
+	stbu	rTMP, 8(rRTN)
bca718
+	beqlr
bca718
+	extrdi.	rTMP, rALT, 8, 8
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	extrdi.	rTMP, rALT, 8, 16
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	extrdi.	rTMP, rALT, 8, 24
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	extrdi.	rTMP, rALT, 8, 32
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	extrdi.	rTMP, rALT, 8, 40
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	extrdi.	rTMP, rALT, 8, 48
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	stbu	rALT, 1(rRTN)
bca718
+#endif
bca718
+	blr
bca718
+
bca718
+L(check_word_alignment):
bca718
+	clrldi. rTMP, rTMP, 62
bca718
+	beq	L(aligned_word_copy)
bca718
+	rldicl	rRTNAL, rRTN, 0, 61
bca718
+	rldicl	rSRCAL, rSRC, 0, 61
bca718
+	cmpld	cr7, rSRCAL, rRTNAL
bca718
+	beq	cr7, L(same_alignment)
bca718
+	b	L(unaligned)
bca718
+
bca718
+/* For word aligned memory, operate using word load and stores.  */
bca718
+	.align	4
bca718
+L(aligned_word_copy):
bca718
+	li	rMASK, 0
bca718
+	addi	rRTN, rRTN, -4
bca718
+	lwz	rWORD, 0(rSRC)
bca718
+	b	L(g5)
bca718
+
bca718
+	.align	4
bca718
+L(g3):	lwzu	rALT, 4(rSRC)
bca718
+	stwu	rWORD, 4(rRTN)
bca718
+	cmpb	rTMP, rALT, rMASK
bca718
+	cmpwi	rTMP, 0
bca718
+	bne	L(g4)
bca718
+	lwzu	rWORD, 4(rSRC)
bca718
+	stwu	rALT, 4(rRTN)
bca718
+L(g5):	cmpb	rTMP, rWORD, rMASK
bca718
+	cmpwi	rTMP, 0		/* If rTMP is 0, no null in word.  */
bca718
+	beq	L(g3)
bca718
+
bca718
+	mr      rALT, rWORD
bca718
+/* We've hit the end of the string.  Do the rest byte-by-byte.  */
bca718
+L(g4):
bca718
+#ifdef __LITTLE_ENDIAN__
bca718
+	rlwinm.	rTMP, rALT, 0, 24, 31
bca718
+	stbu	rALT, 4(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm.	rTMP, rALT, 24, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm.	rTMP, rALT, 16, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm	rTMP, rALT, 8, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+#else
bca718
+	rlwinm. rTMP, rALT, 8, 24, 31
bca718
+	stbu    rTMP, 4(rRTN)
bca718
+	beqlr
bca718
+	rlwinm. rTMP, rALT, 16, 24, 31
bca718
+	stbu    rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	rlwinm. rTMP, rALT, 24, 24, 31
bca718
+	stbu    rTMP, 1(rRTN)
bca718
+	beqlr
bca718
+	stbu    rALT, 1(rRTN)
bca718
+#endif
bca718
+	blr
bca718
+
bca718
+/* Oh well.  In this case, we just do a byte-by-byte copy.  */
bca718
+	.align	4
bca718
+L(unaligned):
bca718
+	lbz	rWORD, 0(rSRC)
bca718
+	addi	rRTN, rRTN, -1
bca718
+	cmpdi	rWORD, 0
bca718
+	beq	L(u2)
bca718
+
bca718
+	.align 	5
bca718
+L(u0):	lbzu	rALT, 1(rSRC)
bca718
+	stbu	rWORD, 1(rRTN)
bca718
+	cmpdi	rALT, 0
bca718
+	beq	L(u1)
bca718
+	lbzu	rWORD, 1(rSRC)
bca718
+	stbu	rALT, 1(rRTN)
bca718
+	cmpdi	rWORD, 0
bca718
+	beq	L(u2)
bca718
+	lbzu	rALT, 1(rSRC)
bca718
+	stbu	rWORD, 1(rRTN)
bca718
+	cmpdi	rALT, 0
bca718
+	beq	L(u1)
bca718
+	lbzu	rWORD, 1(rSRC)
bca718
+	stbu	rALT, 1(rRTN)
bca718
+	cmpdi	rWORD, 0
bca718
+	bne	L(u0)
bca718
+L(u2):	stbu	rWORD, 1(rRTN)
bca718
+	blr
bca718
+L(u1):	stbu	rALT, 1(rRTN)
bca718
+	blr
bca718
+END (FUNC_NAME)
bca718
+
bca718
+#ifndef USE_AS_STPCPY
bca718
+libc_hidden_builtin_def (strcpy)
bca718
+#endif
bca718
diff --git a/sysdeps/powerpc/powerpc64/stpcpy.S b/sysdeps/powerpc/powerpc64/stpcpy.S
bca718
index d795b61..09aa3be 100644
bca718
--- a/sysdeps/powerpc/powerpc64/stpcpy.S
bca718
+++ b/sysdeps/powerpc/powerpc64/stpcpy.S
bca718
@@ -1,5 +1,5 @@
bca718
 /* Optimized stpcpy implementation for PowerPC64.
bca718
-   Copyright (C) 1997, 1999, 2000, 2002, 2004 Free Software Foundation, Inc.
bca718
+   Copyright (C) 1997-2013 Free Software Foundation, Inc.
bca718
    This file is part of the GNU C Library.
bca718
 
bca718
    The GNU C Library is free software; you can redistribute it and/or
bca718
@@ -16,123 +16,9 @@
bca718
    License along with the GNU C Library; if not, see
bca718
    <http://www.gnu.org/licenses/>.  */
bca718
 
bca718
-#include <sysdep.h>
bca718
-#include <bp-sym.h>
bca718
-#include <bp-asm.h>
bca718
+#define USE_AS_STPCPY
bca718
+#include <sysdeps/powerpc/powerpc64/strcpy.S>
bca718
 
bca718
-/* See strlen.s for comments on how the end-of-string testing works.  */
bca718
-
bca718
-/* char * [r3] stpcpy (char *dest [r3], const char *src [r4])  */
bca718
-
bca718
-EALIGN (BP_SYM (__stpcpy), 4, 0)
bca718
-	CALL_MCOUNT 2
bca718
-
bca718
-#define rTMP	r0
bca718
-#define rRTN	r3
bca718
-#if __BOUNDED_POINTERS__
bca718
-# define rDEST	r4		/* pointer to previous word in dest */
bca718
-# define rSRC	r5		/* pointer to previous word in src */
bca718
-# define rLOW	r11
bca718
-# define rHIGH	r12
bca718
-#else
bca718
-# define rDEST	r3		/* pointer to previous word in dest */
bca718
-# define rSRC	r4		/* pointer to previous word in src */
bca718
-#endif
bca718
-#define rWORD	r6		/* current word from src */
bca718
-#define rFEFE	r7		/* 0xfefefeff */
bca718
-#define r7F7F	r8		/* 0x7f7f7f7f */
bca718
-#define rNEG	r9		/* ~(word in src | 0x7f7f7f7f) */
bca718
-#define rALT	r10		/* alternate word from src */
bca718
-
bca718
-	CHECK_BOUNDS_LOW (rSRC, rLOW, rHIGH)
bca718
-	CHECK_BOUNDS_LOW (rDEST, rLOW, rHIGH)
bca718
-	STORE_RETURN_BOUNDS (rLOW, rHIGH)
bca718
-
bca718
-	or	rTMP, rSRC, rDEST
bca718
-	clrldi.	rTMP, rTMP, 62
bca718
-	addi	rDEST, rDEST, -4
bca718
-	bne	L(unaligned)
bca718
-
bca718
-	lis	rFEFE, -0x101
bca718
-	lis	r7F7F, 0x7f7f
bca718
-	lwz	rWORD, 0(rSRC)
bca718
-	addi	rFEFE, rFEFE, -0x101
bca718
-	addi	r7F7F, r7F7F, 0x7f7f
bca718
-	b	L(g2)
bca718
-
bca718
-L(g0):	lwzu	rALT, 4(rSRC)
bca718
-	stwu	rWORD, 4(rDEST)
bca718
-	add	rTMP, rFEFE, rALT
bca718
-	nor	rNEG, r7F7F, rALT
bca718
-	and.	rTMP, rTMP, rNEG
bca718
-	bne-	L(g1)
bca718
-	lwzu	rWORD, 4(rSRC)
bca718
-	stwu	rALT, 4(rDEST)
bca718
-L(g2):	add	rTMP, rFEFE, rWORD
bca718
-	nor	rNEG, r7F7F, rWORD
bca718
-	and.	rTMP, rTMP, rNEG
bca718
-	beq+	L(g0)
bca718
-
bca718
-	mr	rALT, rWORD
bca718
-/* We've hit the end of the string.  Do the rest byte-by-byte.  */
bca718
-L(g1):
bca718
-#ifdef __LITTLE_ENDIAN__
bca718
-	rlwinm.	rTMP, rALT, 0, 24, 31
bca718
-	stbu	rALT, 4(rDEST)
bca718
-	beqlr-
bca718
-	rlwinm.	rTMP, rALT, 24, 24, 31
bca718
-	stbu	rTMP, 1(rDEST)
bca718
-	beqlr-
bca718
-	rlwinm.	rTMP, rALT, 16, 24, 31
bca718
-	stbu	rTMP, 1(rDEST)
bca718
-	beqlr-
bca718
-	rlwinm	rTMP, rALT, 8, 24, 31
bca718
-	stbu	rTMP, 1(rDEST)
bca718
-	blr
bca718
-#else
bca718
-	rlwinm.	rTMP, rALT, 8, 24, 31
bca718
-	stbu	rTMP, 4(rDEST)
bca718
-	beqlr-
bca718
-	rlwinm.	rTMP, rALT, 16, 24, 31
bca718
-	stbu	rTMP, 1(rDEST)
bca718
-	beqlr-
bca718
-	rlwinm.	rTMP, rALT, 24, 24, 31
bca718
-	stbu	rTMP, 1(rDEST)
bca718
-	beqlr-
bca718
-	stbu	rALT, 1(rDEST)
bca718
-	CHECK_BOUNDS_HIGH (rDEST, rHIGH, twlgt)
bca718
-	STORE_RETURN_VALUE (rDEST)
bca718
-	blr
bca718
-#endif
bca718
-
bca718
-/* Oh well.  In this case, we just do a byte-by-byte copy.  */
bca718
-	.align 4
bca718
-	nop
bca718
-L(unaligned):
bca718
-	lbz	rWORD, 0(rSRC)
bca718
-	addi	rDEST, rDEST, 3
bca718
-	cmpwi	rWORD, 0
bca718
-	beq-	L(u2)
bca718
-
bca718
-L(u0):	lbzu	rALT, 1(rSRC)
bca718
-	stbu	rWORD, 1(rDEST)
bca718
-	cmpwi	rALT, 0
bca718
-	beq-	L(u1)
bca718
-	nop		/* Let 601 load start of loop.  */
bca718
-	lbzu	rWORD, 1(rSRC)
bca718
-	stbu	rALT, 1(rDEST)
bca718
-	cmpwi	rWORD, 0
bca718
-	bne+	L(u0)
bca718
-L(u2):	stbu	rWORD, 1(rDEST)
bca718
-	CHECK_BOUNDS_HIGH (rDEST, rHIGH, twlgt)
bca718
-	STORE_RETURN_VALUE (rDEST)
bca718
-	blr
bca718
-L(u1):	stbu	rALT, 1(rDEST)
bca718
-	CHECK_BOUNDS_HIGH (rDEST, rHIGH, twlgt)
bca718
-	STORE_RETURN_VALUE (rDEST)
bca718
-	blr
bca718
-END (BP_SYM (__stpcpy))
bca718
-
bca718
-weak_alias (BP_SYM (__stpcpy), BP_SYM (stpcpy))
bca718
+weak_alias (__stpcpy, stpcpy)
bca718
 libc_hidden_def (__stpcpy)
bca718
 libc_hidden_builtin_def (stpcpy)
bca718
diff --git a/sysdeps/powerpc/powerpc64/strcpy.S b/sysdeps/powerpc/powerpc64/strcpy.S
bca718
index 9434c27..793325d 100644
bca718
--- a/sysdeps/powerpc/powerpc64/strcpy.S
bca718
+++ b/sysdeps/powerpc/powerpc64/strcpy.S
bca718
@@ -1,5 +1,5 @@
bca718
 /* Optimized strcpy implementation for PowerPC64.
bca718
-   Copyright (C) 1997, 1999, 2000, 2002, 2003, 2011 Free Software Foundation, Inc.
bca718
+   Copyright (C) 1997-2013 Free Software Foundation, Inc.
bca718
    This file is part of the GNU C Library.
bca718
 
bca718
    The GNU C Library is free software; you can redistribute it and/or
bca718
@@ -17,52 +17,43 @@
bca718
    <http://www.gnu.org/licenses/>.  */
bca718
 
bca718
 #include <sysdep.h>
bca718
-#include <bp-sym.h>
bca718
-#include <bp-asm.h>
bca718
 
bca718
 /* See strlen.s for comments on how the end-of-string testing works.  */
bca718
 
bca718
 /* char * [r3] strcpy (char *dest [r3], const char *src [r4])  */
bca718
 
bca718
-EALIGN (BP_SYM (strcpy), 4, 0)
bca718
+#ifdef USE_AS_STPCPY
bca718
+# define FUNC_NAME __stpcpy
bca718
+#else
bca718
+# define FUNC_NAME strcpy
bca718
+#endif
bca718
+
bca718
+EALIGN (FUNC_NAME, 4, 0)
bca718
 	CALL_MCOUNT 2
bca718
 
bca718
 #define rTMP	r0
bca718
-#define rRTN	r3	/* incoming DEST arg preserved as result */
bca718
-/* Note.  The Bounded pointer support in this code is broken.  This code
bca718
-   was inherited from PPC32 and that support was never completed.
bca718
-   Current PPC gcc does not support -fbounds-check or -fbounded-pointers.
bca718
-   These artifacts are left in the code as a reminder in case we need
bca718
-   bounded pointer support in the future.  */
bca718
-#if __BOUNDED_POINTERS__
bca718
-# define rDEST	r4	/* pointer to previous word in dest */
bca718
-# define rSRC	r5	/* pointer to previous word in src */
bca718
-# define rLOW	r11
bca718
-# define rHIGH	r12
bca718
+#ifdef USE_AS_STPCPY
bca718
+#define rRTN    r3      /* pointer to previous word/doubleword in dest */
bca718
 #else
bca718
-# define rSRC	r4	/* pointer to previous word in src */
bca718
-# define rDEST	r5	/* pointer to previous word in dest */
bca718
+#define rRTN    r12     /* pointer to previous word/doubleword in dest */
bca718
 #endif
bca718
+#define rSRC	r4	/* pointer to previous word/doubleword in src */
bca718
 #define rWORD	r6	/* current word from src */
bca718
-#define rFEFE	r7	/* constant 0xfefefefefefefeff (-0x0101010101010101) */
bca718
-#define r7F7F	r8	/* constant 0x7f7f7f7f7f7f7f7f */
bca718
-#define rNEG	r9	/* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */
bca718
+#define rFEFE	r7	/* constant 0xfefefeff | 0xfefefefefefefeff */
bca718
+#define r7F7F	r8	/* constant 0x7f7f7f7f | 0x7f7f7f7f7f7f7f7f */
bca718
+#define rNEG	r9	/* ~(word in s1 | r7F7F) */
bca718
 #define rALT	r10	/* alternate word from src */
bca718
 
bca718
-	CHECK_BOUNDS_LOW (rSRC, rLOW, rHIGH)
bca718
-	CHECK_BOUNDS_LOW (rDEST, rLOW, rHIGH)
bca718
-	STORE_RETURN_BOUNDS (rLOW, rHIGH)
bca718
-
bca718
-	dcbt	0,rSRC
bca718
+#ifndef USE_AS_STPCPY
bca718
+/* Save the dst pointer to use as return value.  */
bca718
+	mr      rRTN, r3
bca718
+#endif
bca718
 	or	rTMP, rSRC, rRTN
bca718
 	clrldi.	rTMP, rTMP, 61
bca718
-#if __BOUNDED_POINTERS__
bca718
-	addi	rDEST, rDEST, -8
bca718
-#else
bca718
-	addi	rDEST, rRTN, -8
bca718
-#endif
bca718
-	dcbtst	0,rRTN
bca718
-	bne	L(unaligned)
bca718
+	bne	L(check_word_alignment)
bca718
+
bca718
+/* For doubleword aligned memory, operate using doubleword load and stores.  */
bca718
+	addi	rRTN, rRTN, -8
bca718
 
bca718
 	lis	rFEFE, -0x101
bca718
 	lis	r7F7F, 0x7f7f
bca718
@@ -75,13 +66,13 @@ EALIGN (BP_SYM (strcpy), 4, 0)
bca718
 	b	L(g2)
bca718
 
bca718
 L(g0):	ldu	rALT, 8(rSRC)
bca718
-	stdu	rWORD, 8(rDEST)
bca718
+	stdu	rWORD, 8(rRTN)
bca718
 	add	rTMP, rFEFE, rALT
bca718
 	nor	rNEG, r7F7F, rALT
bca718
 	and.	rTMP, rTMP, rNEG
bca718
 	bne-	L(g1)
bca718
 	ldu	rWORD, 8(rSRC)
bca718
-	stdu	rALT, 8(rDEST)
bca718
+	stdu	rALT, 8(rRTN)
bca718
 L(g2):	add	rTMP, rFEFE, rWORD
bca718
 	nor	rNEG, r7F7F, rWORD
bca718
 	and.	rTMP, rTMP, rNEG
bca718
@@ -92,80 +83,134 @@ L(g2):	add	rTMP, rFEFE, rWORD
bca718
 L(g1):
bca718
 #ifdef __LITTLE_ENDIAN__
bca718
 	extrdi.	rTMP, rALT, 8, 56
bca718
-	stb	rALT, 8(rDEST)
bca718
+	stbu	rALT, 8(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 48
bca718
-	stb	rTMP, 9(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 40
bca718
-	stb	rTMP, 10(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 32
bca718
-	stb	rTMP, 11(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 24
bca718
-	stb	rTMP, 12(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 16
bca718
-	stb	rTMP, 13(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 8
bca718
-	stb	rTMP, 14(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi	rTMP, rALT, 8, 0
bca718
-	stb	rTMP, 15(rDEST)
bca718
-	blr
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 #else
bca718
 	extrdi.	rTMP, rALT, 8, 0
bca718
-	stb	rTMP, 8(rDEST)
bca718
+	stbu	rTMP, 8(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 8
bca718
-	stb	rTMP, 9(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 16
bca718
-	stb	rTMP, 10(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 24
bca718
-	stb	rTMP, 11(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 32
bca718
-	stb	rTMP, 12(rDEST)
bca718
-	beqlr-
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr
bca718
 	extrdi.	rTMP, rALT, 8, 40
bca718
-	stb	rTMP, 13(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
 	extrdi.	rTMP, rALT, 8, 48
bca718
-	stb	rTMP, 14(rDEST)
bca718
+	stbu	rTMP, 1(rRTN)
bca718
 	beqlr-
bca718
-	stb	rALT, 15(rDEST)
bca718
-	/* GKM FIXME: check high bound.  */
bca718
+	stbu	rALT, 1(rRTN)
bca718
+#endif
bca718
 	blr
bca718
+
bca718
+L(check_word_alignment):
bca718
+	clrldi. rTMP, rTMP, 62
bca718
+	bne     L(unaligned)
bca718
+
bca718
+/* For word aligned memory, operate using word load and stores.  */
bca718
+	addi	rRTN, rRTN, -4
bca718
+
bca718
+	lis	rFEFE, -0x101
bca718
+	lis	r7F7F, 0x7f7f
bca718
+	lwz	rWORD, 0(rSRC)
bca718
+	addi	rFEFE, rFEFE, -0x101
bca718
+	addi	r7F7F, r7F7F, 0x7f7f
bca718
+	b	L(g5)
bca718
+
bca718
+L(g3):	lwzu	rALT, 4(rSRC)
bca718
+	stwu	rWORD, 4(rRTN)
bca718
+	add	rTMP, rFEFE, rALT
bca718
+	nor	rNEG, r7F7F, rALT
bca718
+	and.	rTMP, rTMP, rNEG
bca718
+	bne-	L(g4)
bca718
+	lwzu	rWORD, 4(rSRC)
bca718
+	stwu	rALT, 4(rRTN)
bca718
+L(g5):	add	rTMP, rFEFE, rWORD
bca718
+	nor	rNEG, r7F7F, rWORD
bca718
+	and.	rTMP, rTMP, rNEG
bca718
+	beq+	L(g3)
bca718
+
bca718
+	mr	rALT, rWORD
bca718
+/* We've hit the end of the string.  Do the rest byte-by-byte.  */
bca718
+L(g4):
bca718
+#ifdef __LITTLE_ENDIAN__
bca718
+	rlwinm.	rTMP, rALT, 0, 24, 31
bca718
+	stbu	rALT, 4(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm.	rTMP, rALT, 24, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm.	rTMP, rALT, 16, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm	rTMP, rALT, 8, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+#else
bca718
+	rlwinm.	rTMP, rALT, 8, 24, 31
bca718
+	stbu	rTMP, 4(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm.	rTMP, rALT, 16, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	rlwinm.	rTMP, rALT, 24, 24, 31
bca718
+	stbu	rTMP, 1(rRTN)
bca718
+	beqlr-
bca718
+	stbu	rALT, 1(rRTN)
bca718
 #endif
bca718
+	blr
bca718
 
bca718
 /* Oh well.  In this case, we just do a byte-by-byte copy.  */
bca718
 	.align 4
bca718
 	nop
bca718
 L(unaligned):
bca718
 	lbz	rWORD, 0(rSRC)
bca718
-	addi	rDEST, rRTN, -1
bca718
+	addi	rRTN, rRTN, -1
bca718
 	cmpwi	rWORD, 0
bca718
 	beq-	L(u2)
bca718
 
bca718
 L(u0):	lbzu	rALT, 1(rSRC)
bca718
-	stbu	rWORD, 1(rDEST)
bca718
+	stbu	rWORD, 1(rRTN)
bca718
 	cmpwi	rALT, 0
bca718
 	beq-	L(u1)
bca718
 	nop		/* Let 601 load start of loop.  */
bca718
 	lbzu	rWORD, 1(rSRC)
bca718
-	stbu	rALT, 1(rDEST)
bca718
+	stbu	rALT, 1(rRTN)
bca718
 	cmpwi	rWORD, 0
bca718
 	bne+	L(u0)
bca718
-L(u2):	stb	rWORD, 1(rDEST)
bca718
-	/* GKM FIXME: check high bound.  */
bca718
+L(u2):	stbu	rWORD, 1(rRTN)
bca718
 	blr
bca718
-L(u1):	stb	rALT, 1(rDEST)
bca718
-	/* GKM FIXME: check high bound.  */
bca718
+L(u1):	stbu	rALT, 1(rRTN)
bca718
 	blr
bca718
+END (FUNC_NAME)
bca718
 
bca718
-END (BP_SYM (strcpy))
bca718
+#ifndef USE_AS_STPCPY
bca718
 libc_hidden_builtin_def (strcpy)
bca718
+#endif