Correct cacheline size to 32-bytes for ppc405 memset.S (bug 14595).

This patch also creates a version of memset.S for the ppc476 processor
which uses a 128-byte cacheline size for dcbz insns.

3 files changed, 169 insertions, 6 deletions

diff --git a/ports/ChangeLog.powerpc b/ports/ChangeLog.powerpc
index 642e7165c6..e22a7333a3 100644
--- a/ports/ChangeLog.powerpc
+++ b/ports/ChangeLog.powerpc
@@ -1,3 +1,12 @@
+2012-09-25  Jason Gunthorpe  <jgunthorpe@obsidianresearch.com>
+	    Ryan S. Arnold  <rsa@linux.vnet.ibm.com>
+
+	[BZ #14595]
+	* sysdeps/powerpc/powerpc32/476/memset.S: New file copied from
+	405/memset.S to preserve 128-byte cacheline size.
+	* sysdeps/powerpc/powerpc32/405/memset.S (memset): Fix cacheline size
+	to 32-bytes for 405, 440, and 464 processors.
+
 2012-10-19  Roland McGrath  <roland@hack.frob.com>
 
 	* sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/nptl/libc.abilist
diff --git a/ports/sysdeps/powerpc/powerpc32/405/memset.S b/ports/sysdeps/powerpc/powerpc32/405/memset.S
index e132ce3652..c2ee6c593c 100644
--- a/ports/sysdeps/powerpc/powerpc32/405/memset.S
+++ b/ports/sysdeps/powerpc/powerpc32/405/memset.S
@@ -1,5 +1,5 @@
-/* Optimized memset implementation for PowerPC476.
-   Copyright (C) 2010 Free Software Foundation, Inc.
+/* Optimized memset for PowerPC405,440,464 (32-byte cacheline).
+   Copyright (C) 2012 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -104,7 +104,7 @@ L(use_dcbz):
        add     r3,r3,r7
 
 L(skip_string_loop):
-       clrlwi  r8,r6,25
+       clrlwi  r8,r6,27
        srwi.   r8,r8,4
        beq     L(dcbz_pre_loop)
        mtctr   r8
@@ -119,14 +119,14 @@ L(word_loop):
        bdnz    L(word_loop)
 
 L(dcbz_pre_loop):
-       srwi    r6,r5,7
+       srwi    r6,r5,5
        mtctr   r6
        addi    r7,0,0
 
 L(dcbz_loop):
        dcbz    r3,r7
-       addi    r3,r3,0x80
-       subi    r5,r5,0x80
+       addi    r3,r3,0x20
+       subi    r5,r5,0x20
        bdnz    L(dcbz_loop)
        srwi.   r6,r5,4
        beq     L(postword2_count_loop)
diff --git a/ports/sysdeps/powerpc/powerpc32/476/memset.S b/ports/sysdeps/powerpc/powerpc32/476/memset.S
new file mode 100644
index 0000000000..8b5750442b
--- /dev/null
+++ b/ports/sysdeps/powerpc/powerpc32/476/memset.S
@@ -0,0 +1,154 @@
+/* Optimized memset for PowerPC476 (128-byte cacheline).
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library.  If not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#include <sysdep.h>
+#include <bp-sym.h>
+#include <bp-asm.h>
+
+/* memset
+
+       r3:destination address and return address
+       r4:source integer to copy
+       r5:byte count
+       r11:sources integer to copy in all 32 bits of reg
+       r12:temp return address
+
+       Save return address in r12
+       If destinationn is unaligned and count is greater tha 255 bytes
+       set 0-3 bytes to make destination aligned
+       If count is greater tha 255 bytes and setting zero to memory
+       use dbcz to set memeory when we can
+       otherwsie do the follwoing
+       If 16 or more words to set we use 16 word copy loop.
+       Finaly we set 0-15 extra bytes with string store. */
+
+EALIGN (BP_SYM (memset), 5, 0)
+       rlwinm  r11,r4,0,24,31
+       rlwimi  r11,r4,8,16,23
+       rlwimi  r11,r11,16,0,15
+       addi    r12,r3,0
+       cmpwi   r5,0x00FF
+       ble     L(preword8_count_loop)
+       cmpwi   r4,0x00
+       beq     L(use_dcbz)
+       neg     r6,r3
+       clrlwi. r6,r6,30
+       beq     L(preword8_count_loop)
+       addi    r8,0,1
+       mtctr   r6
+       subi    r3,r3,1
+
+L(unaligned_bytecopy_loop):
+       stbu    r11,0x1(r3)
+       subf.   r5,r8,r5
+       beq     L(end_memset)
+       bdnz    L(unaligned_bytecopy_loop)
+       addi    r3,r3,1
+
+L(preword8_count_loop):
+       srwi.   r6,r5,4
+       beq     L(preword2_count_loop)
+       mtctr   r6
+       addi    r3,r3,-4
+       mr      r8,r11
+       mr      r9,r11
+       mr      r10,r11
+
+L(word8_count_loop_no_dcbt):
+       stwu    r8,4(r3)
+       stwu    r9,4(r3)
+       subi    r5,r5,0x10
+       stwu    r10,4(r3)
+       stwu    r11,4(r3)
+       bdnz    L(word8_count_loop_no_dcbt)
+       addi    r3,r3,4
+
+L(preword2_count_loop):
+       clrlwi. r7,r5,28
+       beq     L(end_memset)
+       mr      r8,r11
+       mr      r9,r11
+       mr      r10,r11
+       mtxer   r7
+       stswx   r8,0,r3
+
+L(end_memset):
+       addi    r3,r12,0
+       blr
+
+L(use_dcbz):
+       neg     r6,r3
+       clrlwi. r7,r6,28
+       beq     L(skip_string_loop)
+       mr      r8,r11
+       mr      r9,r11
+       mr      r10,r11
+       subf    r5,r7,r5
+       mtxer   r7
+       stswx   r8,0,r3
+       add     r3,r3,r7
+
+L(skip_string_loop):
+       clrlwi  r8,r6,25
+       srwi.   r8,r8,4
+       beq     L(dcbz_pre_loop)
+       mtctr   r8
+
+L(word_loop):
+       stw     r11,0(r3)
+       subi    r5,r5,0x10
+       stw     r11,4(r3)
+       stw     r11,8(r3)
+       stw     r11,12(r3)
+       addi    r3,r3,0x10
+       bdnz    L(word_loop)
+
+L(dcbz_pre_loop):
+       srwi    r6,r5,7
+       mtctr   r6
+       addi    r7,0,0
+
+L(dcbz_loop):
+       dcbz    r3,r7
+       addi    r3,r3,0x80
+       subi    r5,r5,0x80
+       bdnz    L(dcbz_loop)
+       srwi.   r6,r5,4
+       beq     L(postword2_count_loop)
+       mtctr   r6
+
+L(postword8_count_loop):
+       stw     r11,0(r3)
+       subi    r5,r5,0x10
+       stw     r11,4(r3)
+       stw     r11,8(r3)
+       stw     r11,12(r3)
+       addi    r3,r3,0x10
+       bdnz    L(postword8_count_loop)
+
+L(postword2_count_loop):
+       clrlwi. r7,r5,28
+       beq     L(end_memset)
+       mr      r8,r11
+       mr      r9,r11
+       mr      r10,r11
+       mtxer   r7
+       stswx   r8,0,r3
+       b       L(end_memset)
+END (BP_SYM (memset))
+libc_hidden_builtin_def (memset)