|
 |
00db10 |
# commit 759cfef3ac4c07dba1ece0bbc1207e099348816d
|
|
|
00db10 |
# Author: Alan Modra <amodra@gmail.com>
|
|
|
00db10 |
# Date: Sat Aug 17 18:47:22 2013 +0930
|
|
|
00db10 |
#
|
|
|
00db10 |
# PowerPC LE memcpy
|
|
|
00db10 |
# http://sourceware.org/ml/libc-alpha/2013-08/msg00103.html
|
|
|
00db10 |
#
|
|
|
00db10 |
# LIttle-endian support for memcpy. I spent some time cleaning up the
|
|
|
00db10 |
# 64-bit power7 memcpy, in order to avoid the extra alignment traps
|
|
|
00db10 |
# power7 takes for little-endian. It probably would have been better
|
|
|
00db10 |
# to copy the linux kernel version of memcpy.
|
|
|
00db10 |
#
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc32/power4/memcpy.S: Add little endian support.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc32/power6/memcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc32/power7/memcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc32/power7/mempcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc64/memcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc64/power4/memcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc64/power6/memcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc64/power7/memcpy.S: Likewise.
|
|
|
00db10 |
# * sysdeps/powerpc/powerpc64/power7/mempcpy.S: Likewise. Make better
|
|
|
00db10 |
# use of regs. Use power7 mtocrf. Tidy function tails.
|
|
|
00db10 |
#
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power4/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power4/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power4/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power4/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
@@ -205,15 +205,28 @@
|
|
|
00db10 |
blt cr6,5f
|
|
|
00db10 |
srwi 7,6,16
|
|
|
00db10 |
bgt cr6,3f
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ sth 7,0(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
3:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,24
|
|
|
00db10 |
+ stb 6,0(3)
|
|
|
00db10 |
+ sth 7,1(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
stb 7,0(3)
|
|
|
00db10 |
sth 6,1(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
5:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
7:
|
|
|
00db10 |
cmplwi cr1,10,16
|
|
|
00db10 |
@@ -341,13 +354,23 @@
|
|
|
00db10 |
bf 30,1f
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two words to copy, so copy them */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,6,10
|
|
|
00db10 |
+ slw 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,6,10 /* shift 1st src word to left align it in R0 */
|
|
|
00db10 |
srw 8,7,9 /* shift 2nd src word to right align it in R8 */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8 /* or them to get word to store */
|
|
|
00db10 |
lwz 6,8(5) /* load the 3rd src word */
|
|
|
00db10 |
stw 0,0(4) /* store the 1st dst word */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,7,10
|
|
|
00db10 |
+ slw 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,7,10 /* now left align 2nd src word into R0 */
|
|
|
00db10 |
srw 8,6,9 /* shift 3rd src word to right align it in R8 */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8 /* or them to get word to store */
|
|
|
00db10 |
lwz 7,12(5)
|
|
|
00db10 |
stw 0,4(4) /* store the 2nd dst word */
|
|
|
00db10 |
@@ -355,8 +378,13 @@
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
bf 31,4f
|
|
|
00db10 |
/* there is a third word to copy, so copy it */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,6,10
|
|
|
00db10 |
+ slw 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,6,10 /* shift 3rd src word to left align it in R0 */
|
|
|
00db10 |
srw 8,7,9 /* shift 4th src word to right align it in R8 */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8 /* or them to get word to store */
|
|
|
00db10 |
stw 0,0(4) /* store 3rd dst word */
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -366,8 +394,13 @@
|
|
|
00db10 |
b 4f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
1:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,6,10
|
|
|
00db10 |
+ slw 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,6,10 /* shift 1st src word to left align it in R0 */
|
|
|
00db10 |
srw 8,7,9 /* shift 2nd src word to right align it in R8 */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,8
|
|
|
00db10 |
or 0,0,8 /* or them to get word to store */
|
|
|
00db10 |
bf 31,4f
|
|
|
00db10 |
@@ -380,23 +413,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
4:
|
|
|
00db10 |
/* copy 16 bytes at a time */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,6,10
|
|
|
00db10 |
+ slw 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,6,10
|
|
|
00db10 |
srw 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
lwz 6,0(5)
|
|
|
00db10 |
stw 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,7,10
|
|
|
00db10 |
+ slw 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,7,10
|
|
|
00db10 |
srw 8,6,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
lwz 7,4(5)
|
|
|
00db10 |
stw 0,4(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,6,10
|
|
|
00db10 |
+ slw 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,6,10
|
|
|
00db10 |
srw 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
lwz 6,8(5)
|
|
|
00db10 |
stw 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,7,10
|
|
|
00db10 |
+ slw 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,7,10
|
|
|
00db10 |
srw 8,6,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
lwz 7,12(5)
|
|
|
00db10 |
stw 0,12(4)
|
|
|
00db10 |
@@ -405,8 +458,13 @@
|
|
|
00db10 |
bdnz+ 4b
|
|
|
00db10 |
8:
|
|
|
00db10 |
/* calculate and store the final word */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srw 0,6,10
|
|
|
00db10 |
+ slw 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slw 0,6,10
|
|
|
00db10 |
srw 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
stw 0,0(4)
|
|
|
00db10 |
3:
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power6/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power6/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power6/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power6/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
@@ -221,15 +221,28 @@
|
|
|
00db10 |
blt cr6,5f
|
|
|
00db10 |
srwi 7,6,16
|
|
|
00db10 |
bgt cr6,3f
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ sth 7,0(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
3:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,24
|
|
|
00db10 |
+ stb 6,0(3)
|
|
|
00db10 |
+ sth 7,1(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
stb 7,0(3)
|
|
|
00db10 |
sth 6,1(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
5:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
7:
|
|
|
00db10 |
cmplwi cr1,10,16
|
|
|
00db10 |
@@ -579,7 +592,11 @@
|
|
|
00db10 |
lwz 6,-1(4)
|
|
|
00db10 |
cmplwi cr6,31,4
|
|
|
00db10 |
srwi 8,31,5 /* calculate the 32 byte loop count */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srwi 6,6,8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slwi 6,6,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
clrlwi 31,31,27 /* The remaining bytes, < 32. */
|
|
|
00db10 |
blt cr5,L(wdu1_32tail)
|
|
|
00db10 |
mtctr 8
|
|
|
00db10 |
@@ -587,8 +604,12 @@
|
|
|
00db10 |
|
|
|
00db10 |
lwz 8,3(4)
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,24,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
/* Equivalent to: srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
rlwimi 6,8,8,(32-8),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b L(wdu1_loop32x)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(wdu1_loop32):
|
|
|
00db10 |
@@ -597,8 +618,12 @@
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
stw 10,-8(3)
|
|
|
00db10 |
stw 11,-4(3)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,24,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
/* Equivalent to srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
rlwimi 6,8,8,(32-8),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
L(wdu1_loop32x):
|
|
|
00db10 |
lwz 10,8(4)
|
|
|
00db10 |
lwz 11,12(4)
|
|
|
00db10 |
@@ -615,7 +640,11 @@
|
|
|
00db10 |
stw 6,16(3)
|
|
|
00db10 |
stw 7,20(3)
|
|
|
00db10 |
addi 3,3,32
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srwi 6,8,8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slwi 6,8,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
bdnz+ L(wdu1_loop32)
|
|
|
00db10 |
stw 10,-8(3)
|
|
|
00db10 |
stw 11,-4(3)
|
|
|
00db10 |
@@ -626,8 +655,12 @@
|
|
|
00db10 |
blt cr6,L(wdu_4tail)
|
|
|
00db10 |
/* calculate and store the final word */
|
|
|
00db10 |
lwz 8,3(4)
|
|
|
00db10 |
-/* Equivalent to: srwi 8,8,32-9; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,24,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+/* Equivalent to: srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
rlwimi 6,8,8,(32-8),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b L(wdu_32tailx)
|
|
|
00db10 |
|
|
|
00db10 |
L(wdu2_32):
|
|
|
00db10 |
@@ -635,7 +668,11 @@
|
|
|
00db10 |
lwz 6,-2(4)
|
|
|
00db10 |
cmplwi cr6,31,4
|
|
|
00db10 |
srwi 8,31,5 /* calculate the 32 byte loop count */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srwi 6,6,16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slwi 6,6,16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
clrlwi 31,31,27 /* The remaining bytes, < 32. */
|
|
|
00db10 |
blt cr5,L(wdu2_32tail)
|
|
|
00db10 |
mtctr 8
|
|
|
00db10 |
@@ -643,8 +680,11 @@
|
|
|
00db10 |
|
|
|
00db10 |
lwz 8,2(4)
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
-/* Equivalent to: srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,16,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
rlwimi 6,8,16,(32-16),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b L(wdu2_loop32x)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(wdu2_loop32):
|
|
|
00db10 |
@@ -653,8 +693,11 @@
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
stw 10,-8(3)
|
|
|
00db10 |
stw 11,-4(3)
|
|
|
00db10 |
-/* Equivalent to srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,16,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
rlwimi 6,8,16,(32-16),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
L(wdu2_loop32x):
|
|
|
00db10 |
lwz 10,8(4)
|
|
|
00db10 |
lwz 11,12(4)
|
|
|
00db10 |
@@ -672,7 +715,11 @@
|
|
|
00db10 |
stw 6,16(3)
|
|
|
00db10 |
stw 7,20(3)
|
|
|
00db10 |
addi 3,3,32
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srwi 6,8,16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slwi 6,8,16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
bdnz+ L(wdu2_loop32)
|
|
|
00db10 |
stw 10,-8(3)
|
|
|
00db10 |
stw 11,-4(3)
|
|
|
00db10 |
@@ -683,8 +730,11 @@
|
|
|
00db10 |
blt cr6,L(wdu_4tail)
|
|
|
00db10 |
/* calculate and store the final word */
|
|
|
00db10 |
lwz 8,2(4)
|
|
|
00db10 |
-/* Equivalent to: srwi 8,8,32-9; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,16,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
rlwimi 6,8,16,(32-16),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b L(wdu_32tailx)
|
|
|
00db10 |
|
|
|
00db10 |
L(wdu3_32):
|
|
|
00db10 |
@@ -692,7 +742,11 @@
|
|
|
00db10 |
lwz 6,-3(4)
|
|
|
00db10 |
cmplwi cr6,31,4
|
|
|
00db10 |
srwi 8,31,5 /* calculate the 32 byte loop count */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srwi 6,6,24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slwi 6,6,24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
clrlwi 31,31,27 /* The remaining bytes, < 32. */
|
|
|
00db10 |
blt cr5,L(wdu3_32tail)
|
|
|
00db10 |
mtctr 8
|
|
|
00db10 |
@@ -700,8 +754,11 @@
|
|
|
00db10 |
|
|
|
00db10 |
lwz 8,1(4)
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
-/* Equivalent to: srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,8,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
rlwimi 6,8,24,(32-24),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b L(wdu3_loop32x)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(wdu3_loop32):
|
|
|
00db10 |
@@ -710,8 +767,11 @@
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
stw 10,-8(3)
|
|
|
00db10 |
stw 11,-4(3)
|
|
|
00db10 |
-/* Equivalent to srwi 8,8,32-8; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,8,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
rlwimi 6,8,24,(32-24),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
L(wdu3_loop32x):
|
|
|
00db10 |
lwz 10,8(4)
|
|
|
00db10 |
lwz 11,12(4)
|
|
|
00db10 |
@@ -728,7 +788,11 @@
|
|
|
00db10 |
stw 6,16(3)
|
|
|
00db10 |
stw 7,20(3)
|
|
|
00db10 |
addi 3,3,32
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srwi 6,8,24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
slwi 6,8,24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
bdnz+ L(wdu3_loop32)
|
|
|
00db10 |
stw 10,-8(3)
|
|
|
00db10 |
stw 11,-4(3)
|
|
|
00db10 |
@@ -739,8 +803,11 @@
|
|
|
00db10 |
blt cr6,L(wdu_4tail)
|
|
|
00db10 |
/* calculate and store the final word */
|
|
|
00db10 |
lwz 8,1(4)
|
|
|
00db10 |
-/* Equivalent to: srwi 8,8,32-9; or 6,6,8 */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rldimi 6,8,8,32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
rlwimi 6,8,24,(32-24),31
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b L(wdu_32tailx)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(wdu_32tailx):
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
@@ -385,7 +385,7 @@
|
|
|
00db10 |
|
|
|
00db10 |
beq L(copy_GE_32_unaligned_cont)
|
|
|
00db10 |
|
|
|
00db10 |
- /* SRC is not quadword aligned, get it aligned. */
|
|
|
00db10 |
+ /* DST is not quadword aligned, get it aligned. */
|
|
|
00db10 |
|
|
|
00db10 |
mtcrf 0x01,0
|
|
|
00db10 |
subf 31,0,5
|
|
|
00db10 |
@@ -437,13 +437,21 @@
|
|
|
00db10 |
mr 11,12
|
|
|
00db10 |
mtcrf 0x01,9
|
|
|
00db10 |
cmplwi cr6,9,1
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ lvsr 5,0,12
|
|
|
00db10 |
+#else
|
|
|
00db10 |
lvsl 5,0,12
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
lvx 3,0,12
|
|
|
00db10 |
bf 31,L(setup_unaligned_loop)
|
|
|
00db10 |
|
|
|
00db10 |
/* Copy another 16 bytes to align to 32-bytes due to the loop . */
|
|
|
00db10 |
lvx 4,12,6
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 11,12,16
|
|
|
00db10 |
addi 10,3,16
|
|
|
00db10 |
stvx 6,0,3
|
|
|
00db10 |
@@ -463,11 +471,17 @@
|
|
|
00db10 |
vector instructions though. */
|
|
|
00db10 |
|
|
|
00db10 |
lvx 4,11,6 /* vr4 = r11+16. */
|
|
|
00db10 |
- vperm 6,3,4,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr6. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
lvx 3,11,7 /* vr3 = r11+32. */
|
|
|
00db10 |
- vperm 10,4,3,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr10. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 10,3,4,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 10,4,3,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 11,11,32
|
|
|
00db10 |
stvx 6,0,10
|
|
|
00db10 |
stvx 10,10,6
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/mempcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/mempcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/mempcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc32/power7/mempcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
@@ -327,7 +327,7 @@
|
|
|
00db10 |
|
|
|
00db10 |
beq L(copy_GE_32_unaligned_cont)
|
|
|
00db10 |
|
|
|
00db10 |
- /* SRC is not quadword aligned, get it aligned. */
|
|
|
00db10 |
+ /* DST is not quadword aligned, get it aligned. */
|
|
|
00db10 |
|
|
|
00db10 |
mtcrf 0x01,0
|
|
|
00db10 |
subf 31,0,5
|
|
|
00db10 |
@@ -379,13 +379,21 @@
|
|
|
00db10 |
mr 11,12
|
|
|
00db10 |
mtcrf 0x01,9
|
|
|
00db10 |
cmplwi cr6,9,1
|
|
|
00db10 |
- lvsl 5,0,12
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ lvsr 5,0,12
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ lvsl 5,0,12
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
lvx 3,0,12
|
|
|
00db10 |
bf 31,L(setup_unaligned_loop)
|
|
|
00db10 |
|
|
|
00db10 |
/* Copy another 16 bytes to align to 32-bytes due to the loop . */
|
|
|
00db10 |
lvx 4,12,6
|
|
|
00db10 |
- vperm 6,3,4,5
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 11,12,16
|
|
|
00db10 |
addi 10,3,16
|
|
|
00db10 |
stvx 6,0,3
|
|
|
00db10 |
@@ -405,11 +413,17 @@
|
|
|
00db10 |
vector instructions though. */
|
|
|
00db10 |
|
|
|
00db10 |
lvx 4,11,6 /* vr4 = r11+16. */
|
|
|
00db10 |
- vperm 6,3,4,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr6. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
lvx 3,11,7 /* vr3 = r11+32. */
|
|
|
00db10 |
- vperm 10,4,3,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr10. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 10,3,4,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 10,4,3,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 11,11,32
|
|
|
00db10 |
stvx 6,0,10
|
|
|
00db10 |
stvx 10,10,6
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
@@ -214,15 +214,28 @@
|
|
|
00db10 |
blt cr6,5f
|
|
|
00db10 |
srdi 7,6,16
|
|
|
00db10 |
bgt cr6,3f
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ sth 7,0(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
3:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,24
|
|
|
00db10 |
+ stb 6,0(3)
|
|
|
00db10 |
+ sth 7,1(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
stb 7,0(3)
|
|
|
00db10 |
sth 6,1(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
5:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
7:
|
|
|
00db10 |
cmpldi cr1,10,16
|
|
|
00db10 |
@@ -330,7 +343,11 @@
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
subfic 9,10,64
|
|
|
00db10 |
beq 2f
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,6,10
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
cmpldi 11,1
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
addi 4,4,-8
|
|
|
00db10 |
@@ -338,15 +355,25 @@
|
|
|
00db10 |
b 1f
|
|
|
00db10 |
2: addi 5,5,8
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+0: srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
0: sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
cmpldi 11,2
|
|
|
00db10 |
ld 6,8(5)
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
addi 11,11,-2
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,7,10
|
|
|
00db10 |
+1: sld 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,7,10
|
|
|
00db10 |
1: srd 8,6,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
beq 8f
|
|
|
00db10 |
ld 7,16(5)
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power4/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power4/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power4/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power4/memcpy.S 2014-05-29 13:05:51.000000000 -0500
|
|
|
00db10 |
@@ -1,5 +1,5 @@
|
|
|
00db10 |
/* Optimized memcpy implementation for PowerPC64.
|
|
|
00db10 |
- Copyright (C) 2003, 2006, 2011 Free Software Foundation, Inc.
|
|
|
00db10 |
+ Copyright (C) 2003-2014 Free Software Foundation, Inc.
|
|
|
00db10 |
This file is part of the GNU C Library.
|
|
|
00db10 |
|
|
|
00db10 |
The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
@@ -17,26 +17,24 @@
|
|
|
00db10 |
<http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
|
|
|
00db10 |
#include <sysdep.h>
|
|
|
00db10 |
-#include <bp-sym.h>
|
|
|
00db10 |
-#include <bp-asm.h>
|
|
|
00db10 |
|
|
|
00db10 |
/* __ptr_t [r3] memcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
|
|
|
00db10 |
Returns 'dst'.
|
|
|
00db10 |
|
|
|
00db10 |
- Memcpy handles short copies (< 32-bytes) using a binary move blocks
|
|
|
00db10 |
- (no loops) of lwz/stw. The tail (remaining 1-3) bytes is handled
|
|
|
00db10 |
- with the appropriate combination of byte and halfword load/stores.
|
|
|
00db10 |
- There is minimal effort to optimize the alignment of short moves.
|
|
|
00db10 |
+ Memcpy handles short copies (< 32-bytes) using a binary move blocks
|
|
|
00db10 |
+ (no loops) of lwz/stw. The tail (remaining 1-3) bytes is handled
|
|
|
00db10 |
+ with the appropriate combination of byte and halfword load/stores.
|
|
|
00db10 |
+ There is minimal effort to optimize the alignment of short moves.
|
|
|
00db10 |
The 64-bit implementations of POWER3 and POWER4 do a reasonable job
|
|
|
00db10 |
- of handling unligned load/stores that do not cross 32-byte boundries.
|
|
|
00db10 |
+ of handling unaligned load/stores that do not cross 32-byte boundaries.
|
|
|
00db10 |
|
|
|
00db10 |
Longer moves (>= 32-bytes) justify the effort to get at least the
|
|
|
00db10 |
destination doubleword (8-byte) aligned. Further optimization is
|
|
|
00db10 |
- posible when both source and destination are doubleword aligned.
|
|
|
00db10 |
+ possible when both source and destination are doubleword aligned.
|
|
|
00db10 |
Each case has a optimized unrolled loop. */
|
|
|
00db10 |
|
|
|
00db10 |
.machine power4
|
|
|
00db10 |
-EALIGN (BP_SYM (memcpy), 5, 0)
|
|
|
00db10 |
+EALIGN (memcpy, 5, 0)
|
|
|
00db10 |
CALL_MCOUNT 3
|
|
|
00db10 |
|
|
|
00db10 |
cmpldi cr1,5,31
|
|
|
00db10 |
@@ -44,20 +42,20 @@
|
|
|
00db10 |
std 3,-16(1)
|
|
|
00db10 |
std 31,-8(1)
|
|
|
00db10 |
cfi_offset(31,-8)
|
|
|
00db10 |
- andi. 11,3,7 /* check alignement of dst. */
|
|
|
00db10 |
+ andi. 11,3,7 /* check alignment of dst. */
|
|
|
00db10 |
clrldi 0,0,61 /* Number of bytes until the 1st doubleword of dst. */
|
|
|
00db10 |
- clrldi 10,4,61 /* check alignement of src. */
|
|
|
00db10 |
+ clrldi 10,4,61 /* check alignment of src. */
|
|
|
00db10 |
cmpldi cr6,5,8
|
|
|
00db10 |
ble- cr1,.L2 /* If move < 32 bytes use short move code. */
|
|
|
00db10 |
- cmpld cr6,10,11
|
|
|
00db10 |
+ cmpld cr6,10,11
|
|
|
00db10 |
mr 12,4
|
|
|
00db10 |
srdi 9,5,3 /* Number of full double words remaining. */
|
|
|
00db10 |
mtcrf 0x01,0
|
|
|
00db10 |
mr 31,5
|
|
|
00db10 |
beq .L0
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
subf 31,0,5
|
|
|
00db10 |
- /* Move 0-7 bytes as needed to get the destination doubleword alligned. */
|
|
|
00db10 |
+ /* Move 0-7 bytes as needed to get the destination doubleword aligned. */
|
|
|
00db10 |
1: bf 31,2f
|
|
|
00db10 |
lbz 6,0(12)
|
|
|
00db10 |
addi 12,12,1
|
|
|
00db10 |
@@ -74,17 +72,17 @@
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
addi 3,3,4
|
|
|
00db10 |
0:
|
|
|
00db10 |
- clrldi 10,12,61 /* check alignement of src again. */
|
|
|
00db10 |
+ clrldi 10,12,61 /* check alignment of src again. */
|
|
|
00db10 |
srdi 9,31,3 /* Number of full double words remaining. */
|
|
|
00db10 |
-
|
|
|
00db10 |
- /* Copy doublewords from source to destination, assumpting the
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Copy doublewords from source to destination, assuming the
|
|
|
00db10 |
destination is aligned on a doubleword boundary.
|
|
|
00db10 |
|
|
|
00db10 |
At this point we know there are at least 25 bytes left (32-7) to copy.
|
|
|
00db10 |
- The next step is to determine if the source is also doubleword aligned.
|
|
|
00db10 |
+ The next step is to determine if the source is also doubleword aligned.
|
|
|
00db10 |
If not branch to the unaligned move code at .L6. which uses
|
|
|
00db10 |
a load, shift, store strategy.
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
Otherwise source and destination are doubleword aligned, and we can
|
|
|
00db10 |
the optimized doubleword copy loop. */
|
|
|
00db10 |
.L0:
|
|
|
00db10 |
@@ -97,14 +95,14 @@
|
|
|
00db10 |
Use a unrolled loop to copy 4 doubleword (32-bytes) per iteration.
|
|
|
00db10 |
If the copy is not an exact multiple of 32 bytes, 1-3
|
|
|
00db10 |
doublewords are copied as needed to set up the main loop. After
|
|
|
00db10 |
- the main loop exits there may be a tail of 1-7 bytes. These byte are
|
|
|
00db10 |
+ the main loop exits there may be a tail of 1-7 bytes. These byte are
|
|
|
00db10 |
copied a word/halfword/byte at a time as needed to preserve alignment. */
|
|
|
00db10 |
|
|
|
00db10 |
srdi 8,31,5
|
|
|
00db10 |
cmpldi cr1,9,4
|
|
|
00db10 |
cmpldi cr6,11,0
|
|
|
00db10 |
mr 11,12
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
bf 30,1f
|
|
|
00db10 |
ld 6,0(12)
|
|
|
00db10 |
ld 7,8(12)
|
|
|
00db10 |
@@ -115,7 +113,7 @@
|
|
|
00db10 |
addi 10,3,16
|
|
|
00db10 |
bf 31,4f
|
|
|
00db10 |
ld 0,16(12)
|
|
|
00db10 |
- std 0,16(3)
|
|
|
00db10 |
+ std 0,16(3)
|
|
|
00db10 |
blt cr1,3f
|
|
|
00db10 |
addi 11,12,24
|
|
|
00db10 |
addi 10,3,24
|
|
|
00db10 |
@@ -129,7 +127,7 @@
|
|
|
00db10 |
addi 11,12,8
|
|
|
00db10 |
std 6,0(3)
|
|
|
00db10 |
addi 10,3,8
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
4:
|
|
|
00db10 |
ld 6,0(11)
|
|
|
00db10 |
@@ -144,7 +142,7 @@
|
|
|
00db10 |
std 0,24(10)
|
|
|
00db10 |
addi 10,10,32
|
|
|
00db10 |
bdnz 4b
|
|
|
00db10 |
-3:
|
|
|
00db10 |
+3:
|
|
|
00db10 |
|
|
|
00db10 |
rldicr 0,31,0,60
|
|
|
00db10 |
mtcrf 0x01,31
|
|
|
00db10 |
@@ -152,9 +150,9 @@
|
|
|
00db10 |
.L9:
|
|
|
00db10 |
add 3,3,0
|
|
|
00db10 |
add 12,12,0
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
/* At this point we have a tail of 0-7 bytes and we know that the
|
|
|
00db10 |
- destiniation is double word aligned. */
|
|
|
00db10 |
+ destination is double word aligned. */
|
|
|
00db10 |
4: bf 29,2f
|
|
|
00db10 |
lwz 6,0(12)
|
|
|
00db10 |
addi 12,12,4
|
|
|
00db10 |
@@ -173,29 +171,29 @@
|
|
|
00db10 |
ld 31,-8(1)
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
-
|
|
|
00db10 |
-/* Copy up to 31 bytes. This divided into two cases 0-8 bytes and 9-31
|
|
|
00db10 |
- bytes. Each case is handled without loops, using binary (1,2,4,8)
|
|
|
00db10 |
- tests.
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Copy up to 31 bytes. This divided into two cases 0-8 bytes and 9-31
|
|
|
00db10 |
+ bytes. Each case is handled without loops, using binary (1,2,4,8)
|
|
|
00db10 |
+ tests.
|
|
|
00db10 |
+
|
|
|
00db10 |
In the short (0-8 byte) case no attempt is made to force alignment
|
|
|
00db10 |
- of either source or destination. The hardware will handle the
|
|
|
00db10 |
- unaligned load/stores with small delays for crossing 32- 64-byte, and
|
|
|
00db10 |
+ of either source or destination. The hardware will handle the
|
|
|
00db10 |
+ unaligned load/stores with small delays for crossing 32- 64-byte, and
|
|
|
00db10 |
4096-byte boundaries. Since these short moves are unlikely to be
|
|
|
00db10 |
- unaligned or cross these boundaries, the overhead to force
|
|
|
00db10 |
+ unaligned or cross these boundaries, the overhead to force
|
|
|
00db10 |
alignment is not justified.
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
The longer (9-31 byte) move is more likely to cross 32- or 64-byte
|
|
|
00db10 |
boundaries. Since only loads are sensitive to the 32-/64-byte
|
|
|
00db10 |
- boundaries it is more important to align the source then the
|
|
|
00db10 |
+ boundaries it is more important to align the source then the
|
|
|
00db10 |
destination. If the source is not already word aligned, we first
|
|
|
00db10 |
- move 1-3 bytes as needed. Since we are only word aligned we don't
|
|
|
00db10 |
- use double word load/stores to insure that all loads are aligned.
|
|
|
00db10 |
+ move 1-3 bytes as needed. Since we are only word aligned we don't
|
|
|
00db10 |
+ use double word load/stores to insure that all loads are aligned.
|
|
|
00db10 |
While the destination and stores may still be unaligned, this
|
|
|
00db10 |
is only an issue for page (4096 byte boundary) crossing, which
|
|
|
00db10 |
should be rare for these short moves. The hardware handles this
|
|
|
00db10 |
- case automatically with a small delay. */
|
|
|
00db10 |
-
|
|
|
00db10 |
+ case automatically with a small delay. */
|
|
|
00db10 |
+
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
.L2:
|
|
|
00db10 |
mtcrf 0x01,5
|
|
|
00db10 |
@@ -216,15 +214,28 @@
|
|
|
00db10 |
blt cr6,5f
|
|
|
00db10 |
srdi 7,6,16
|
|
|
00db10 |
bgt cr6,3f
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ sth 7,0(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
3:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,24
|
|
|
00db10 |
+ stb 6,0(3)
|
|
|
00db10 |
+ sth 7,1(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
stb 7,0(3)
|
|
|
00db10 |
sth 6,1(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
5:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
7:
|
|
|
00db10 |
cmpldi cr1,10,16
|
|
|
00db10 |
@@ -258,11 +269,11 @@
|
|
|
00db10 |
lwz 6,0(12)
|
|
|
00db10 |
addi 12,12,4
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
- addi 3,3,4
|
|
|
00db10 |
+ addi 3,3,4
|
|
|
00db10 |
2: /* Move 2-3 bytes. */
|
|
|
00db10 |
bf 30,1f
|
|
|
00db10 |
lhz 6,0(12)
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
+ sth 6,0(3)
|
|
|
00db10 |
bf 31,0f
|
|
|
00db10 |
lbz 7,2(12)
|
|
|
00db10 |
stb 7,2(3)
|
|
|
00db10 |
@@ -283,8 +294,8 @@
|
|
|
00db10 |
mr 12,4
|
|
|
00db10 |
bne cr6,4f
|
|
|
00db10 |
/* Would have liked to use use ld/std here but the 630 processors are
|
|
|
00db10 |
- slow for load/store doubles that are not at least word aligned.
|
|
|
00db10 |
- Unaligned Load/Store word execute with only a 1 cycle penaltity. */
|
|
|
00db10 |
+ slow for load/store doubles that are not at least word aligned.
|
|
|
00db10 |
+ Unaligned Load/Store word execute with only a 1 cycle penalty. */
|
|
|
00db10 |
lwz 6,0(4)
|
|
|
00db10 |
lwz 7,4(4)
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
@@ -299,14 +310,14 @@
|
|
|
00db10 |
6:
|
|
|
00db10 |
bf 30,5f
|
|
|
00db10 |
lhz 7,4(4)
|
|
|
00db10 |
- sth 7,4(3)
|
|
|
00db10 |
+ sth 7,4(3)
|
|
|
00db10 |
bf 31,0f
|
|
|
00db10 |
lbz 8,6(4)
|
|
|
00db10 |
stb 8,6(3)
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
-5:
|
|
|
00db10 |
+5:
|
|
|
00db10 |
bf 31,0f
|
|
|
00db10 |
lbz 6,4(4)
|
|
|
00db10 |
stb 6,4(3)
|
|
|
00db10 |
@@ -336,13 +347,23 @@
|
|
|
00db10 |
bf 30,1f
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,7,10
|
|
|
00db10 |
+ sld 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,7,10
|
|
|
00db10 |
srd 8,6,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -351,8 +372,13 @@
|
|
|
00db10 |
blt cr6,8f /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,4f
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -363,8 +389,13 @@
|
|
|
00db10 |
b 4f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
1:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,4f
|
|
|
00db10 |
@@ -375,23 +406,44 @@
|
|
|
00db10 |
addi 4,4,8
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
-4: sld 0,6,10
|
|
|
00db10 |
+4:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,7,10
|
|
|
00db10 |
+ sld 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,7,10
|
|
|
00db10 |
srd 8,6,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,7,10
|
|
|
00db10 |
+ sld 8,6,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,7,10
|
|
|
00db10 |
srd 8,6,9
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -401,9 +453,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
8:
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srd 0,6,10
|
|
|
00db10 |
+ sld 8,7,9
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sld 0,6,10
|
|
|
00db10 |
srd 8,7,9
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
3:
|
|
|
00db10 |
rldicr 0,31,0,60
|
|
|
00db10 |
@@ -413,5 +470,5 @@
|
|
|
00db10 |
ld 31,-8(1)
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
-END_GEN_TB (BP_SYM (memcpy),TB_TOCLESS)
|
|
|
00db10 |
+END_GEN_TB (memcpy,TB_TOCLESS)
|
|
|
00db10 |
libc_hidden_builtin_def (memcpy)
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power6/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power6/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power6/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power6/memcpy.S 2014-05-29 13:05:27.000000000 -0500
|
|
|
00db10 |
@@ -1,5 +1,5 @@
|
|
|
00db10 |
/* Optimized memcpy implementation for PowerPC64.
|
|
|
00db10 |
- Copyright (C) 2003, 2006, 2007, 2011 Free Software Foundation, Inc.
|
|
|
00db10 |
+ Copyright (C) 2003-2014 Free Software Foundation, Inc.
|
|
|
00db10 |
This file is part of the GNU C Library.
|
|
|
00db10 |
|
|
|
00db10 |
The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
@@ -17,52 +17,50 @@
|
|
|
00db10 |
<http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
|
|
|
00db10 |
#include <sysdep.h>
|
|
|
00db10 |
-#include <bp-sym.h>
|
|
|
00db10 |
-#include <bp-asm.h>
|
|
|
00db10 |
|
|
|
00db10 |
/* __ptr_t [r3] memcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
|
|
|
00db10 |
Returns 'dst'.
|
|
|
00db10 |
|
|
|
00db10 |
- Memcpy handles short copies (< 32-bytes) using a binary move blocks
|
|
|
00db10 |
- (no loops) of lwz/stw. The tail (remaining 1-3) bytes is handled
|
|
|
00db10 |
- with the appropriate combination of byte and halfword load/stores.
|
|
|
00db10 |
- There is minimal effort to optimize the alignment of short moves.
|
|
|
00db10 |
+ Memcpy handles short copies (< 32-bytes) using a binary move blocks
|
|
|
00db10 |
+ (no loops) of lwz/stw. The tail (remaining 1-3) bytes is handled
|
|
|
00db10 |
+ with the appropriate combination of byte and halfword load/stores.
|
|
|
00db10 |
+ There is minimal effort to optimize the alignment of short moves.
|
|
|
00db10 |
The 64-bit implementations of POWER3 and POWER4 do a reasonable job
|
|
|
00db10 |
- of handling unligned load/stores that do not cross 32-byte boundries.
|
|
|
00db10 |
+ of handling unaligned load/stores that do not cross 32-byte boundaries.
|
|
|
00db10 |
|
|
|
00db10 |
Longer moves (>= 32-bytes) justify the effort to get at least the
|
|
|
00db10 |
destination doubleword (8-byte) aligned. Further optimization is
|
|
|
00db10 |
- posible when both source and destination are doubleword aligned.
|
|
|
00db10 |
- Each case has a optimized unrolled loop.
|
|
|
00db10 |
-
|
|
|
00db10 |
- For POWER6 unaligned loads will take a 20+ cycle hicup for any
|
|
|
00db10 |
+ possible when both source and destination are doubleword aligned.
|
|
|
00db10 |
+ Each case has a optimized unrolled loop.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ For POWER6 unaligned loads will take a 20+ cycle hiccup for any
|
|
|
00db10 |
L1 cache miss that crosses a 32- or 128-byte boundary. Store
|
|
|
00db10 |
- is more forgiving and does not take a hicup until page or
|
|
|
00db10 |
- segment boundaries. So we require doubleword alignment for
|
|
|
00db10 |
+ is more forgiving and does not take a hiccup until page or
|
|
|
00db10 |
+ segment boundaries. So we require doubleword alignment for
|
|
|
00db10 |
the source but may take a risk and only require word alignment
|
|
|
00db10 |
for the destination. */
|
|
|
00db10 |
|
|
|
00db10 |
.machine "power6"
|
|
|
00db10 |
-EALIGN (BP_SYM (memcpy), 7, 0)
|
|
|
00db10 |
+EALIGN (memcpy, 7, 0)
|
|
|
00db10 |
CALL_MCOUNT 3
|
|
|
00db10 |
|
|
|
00db10 |
cmpldi cr1,5,31
|
|
|
00db10 |
neg 0,3
|
|
|
00db10 |
std 3,-16(1)
|
|
|
00db10 |
std 31,-8(1)
|
|
|
00db10 |
- andi. 11,3,7 /* check alignement of dst. */
|
|
|
00db10 |
+ andi. 11,3,7 /* check alignment of dst. */
|
|
|
00db10 |
clrldi 0,0,61 /* Number of bytes until the 1st doubleword of dst. */
|
|
|
00db10 |
- clrldi 10,4,61 /* check alignement of src. */
|
|
|
00db10 |
+ clrldi 10,4,61 /* check alignment of src. */
|
|
|
00db10 |
cmpldi cr6,5,8
|
|
|
00db10 |
ble- cr1,.L2 /* If move < 32 bytes use short move code. */
|
|
|
00db10 |
mtcrf 0x01,0
|
|
|
00db10 |
- cmpld cr6,10,11
|
|
|
00db10 |
+ cmpld cr6,10,11
|
|
|
00db10 |
srdi 9,5,3 /* Number of full double words remaining. */
|
|
|
00db10 |
beq .L0
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
subf 5,0,5
|
|
|
00db10 |
- /* Move 0-7 bytes as needed to get the destination doubleword alligned.
|
|
|
00db10 |
- Duplicate some code to maximize fall-throught and minimize agen delays. */
|
|
|
00db10 |
+ /* Move 0-7 bytes as needed to get the destination doubleword aligned.
|
|
|
00db10 |
+ Duplicate some code to maximize fall-through and minimize agen delays. */
|
|
|
00db10 |
1: bf 31,2f
|
|
|
00db10 |
lbz 6,0(4)
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
@@ -78,7 +76,7 @@
|
|
|
00db10 |
lwz 6,1(4)
|
|
|
00db10 |
stw 6,1(3)
|
|
|
00db10 |
b 0f
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
2: bf 30,4f
|
|
|
00db10 |
lhz 6,0(4)
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
@@ -86,26 +84,26 @@
|
|
|
00db10 |
lwz 6,2(4)
|
|
|
00db10 |
stw 6,2(3)
|
|
|
00db10 |
b 0f
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
4: bf 29,0f
|
|
|
00db10 |
lwz 6,0(4)
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
-0:
|
|
|
00db10 |
+0:
|
|
|
00db10 |
/* Add the number of bytes until the 1st doubleword of dst to src and dst. */
|
|
|
00db10 |
add 4,4,0
|
|
|
00db10 |
add 3,3,0
|
|
|
00db10 |
-
|
|
|
00db10 |
- clrldi 10,4,61 /* check alignement of src again. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+ clrldi 10,4,61 /* check alignment of src again. */
|
|
|
00db10 |
srdi 9,5,3 /* Number of full double words remaining. */
|
|
|
00db10 |
-
|
|
|
00db10 |
- /* Copy doublewords from source to destination, assumpting the
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Copy doublewords from source to destination, assuming the
|
|
|
00db10 |
destination is aligned on a doubleword boundary.
|
|
|
00db10 |
|
|
|
00db10 |
At this point we know there are at least 25 bytes left (32-7) to copy.
|
|
|
00db10 |
- The next step is to determine if the source is also doubleword aligned.
|
|
|
00db10 |
+ The next step is to determine if the source is also doubleword aligned.
|
|
|
00db10 |
If not branch to the unaligned move code at .L6. which uses
|
|
|
00db10 |
a load, shift, store strategy.
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
Otherwise source and destination are doubleword aligned, and we can
|
|
|
00db10 |
the optimized doubleword copy loop. */
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
@@ -123,14 +121,14 @@
|
|
|
00db10 |
the main loop exits there may be a tail of 1-7 bytes. These byte
|
|
|
00db10 |
are copied a word/halfword/byte at a time as needed to preserve
|
|
|
00db10 |
alignment.
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
For POWER6 the L1 is store-through and the L2 is store-in. The
|
|
|
00db10 |
L2 is clocked at half CPU clock so we can store 16 bytes every
|
|
|
00db10 |
other cycle. POWER6 also has a load/store bypass so we can do
|
|
|
00db10 |
- load, load, store, store every 2 cycles.
|
|
|
00db10 |
-
|
|
|
00db10 |
+ load, load, store, store every 2 cycles.
|
|
|
00db10 |
+
|
|
|
00db10 |
The following code is sensitive to cache line alignment. Do not
|
|
|
00db10 |
- make any change with out first making sure thay don't result in
|
|
|
00db10 |
+ make any change with out first making sure they don't result in
|
|
|
00db10 |
splitting ld/std pairs across a cache line. */
|
|
|
00db10 |
|
|
|
00db10 |
mtcrf 0x02,5
|
|
|
00db10 |
@@ -273,7 +271,7 @@
|
|
|
00db10 |
std 8,16+96(10)
|
|
|
00db10 |
std 0,24+96(10)
|
|
|
00db10 |
ble cr5,L(das_loop_e)
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
mtctr 12
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(das_loop2):
|
|
|
00db10 |
@@ -326,10 +324,10 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(das_tail):
|
|
|
00db10 |
beq cr1,0f
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
L(das_tail2):
|
|
|
00db10 |
/* At this point we have a tail of 0-7 bytes and we know that the
|
|
|
00db10 |
- destiniation is double word aligned. */
|
|
|
00db10 |
+ destination is double word aligned. */
|
|
|
00db10 |
4: bf 29,2f
|
|
|
00db10 |
lwz 6,0(4)
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
@@ -344,7 +342,7 @@
|
|
|
00db10 |
lbz 6,4(4)
|
|
|
00db10 |
stb 6,4(3)
|
|
|
00db10 |
b 0f
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
2: bf 30,1f
|
|
|
00db10 |
lhz 6,0(4)
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
@@ -352,7 +350,7 @@
|
|
|
00db10 |
lbz 6,2(4)
|
|
|
00db10 |
stb 6,2(3)
|
|
|
00db10 |
b 0f
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
1: bf 31,0f
|
|
|
00db10 |
lbz 6,0(4)
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
@@ -361,7 +359,7 @@
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
|
|
|
00db10 |
-/* Copy up to 31 bytes. This divided into two cases 0-8 bytes and 9-31
|
|
|
00db10 |
+/* Copy up to 31 bytes. This divided into two cases 0-8 bytes and 9-31
|
|
|
00db10 |
bytes. Each case is handled without loops, using binary (1,2,4,8)
|
|
|
00db10 |
tests.
|
|
|
00db10 |
|
|
|
00db10 |
@@ -402,15 +400,28 @@
|
|
|
00db10 |
blt cr6,5f
|
|
|
00db10 |
srdi 7,6,16
|
|
|
00db10 |
bgt cr6,3f
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ sth 7,0(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sth 6,0(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
3:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,24
|
|
|
00db10 |
+ stb 6,0(3)
|
|
|
00db10 |
+ sth 7,1(3)
|
|
|
00db10 |
+#else
|
|
|
00db10 |
stb 7,0(3)
|
|
|
00db10 |
sth 6,1(3)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
b 7f
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
5:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ rotlwi 6,6,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
stb 6,0(3)
|
|
|
00db10 |
7:
|
|
|
00db10 |
cmpldi cr1,10,16
|
|
|
00db10 |
@@ -421,7 +432,7 @@
|
|
|
00db10 |
/* At least 6 bytes left and the source is word aligned. This allows
|
|
|
00db10 |
some speculative loads up front. */
|
|
|
00db10 |
/* We need to special case the fall-through because the biggest delays
|
|
|
00db10 |
- are due to address computation not being ready in time for the
|
|
|
00db10 |
+ are due to address computation not being ready in time for the
|
|
|
00db10 |
AGEN. */
|
|
|
00db10 |
lwz 6,0(12)
|
|
|
00db10 |
lwz 7,4(12)
|
|
|
00db10 |
@@ -452,7 +463,7 @@
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
-L(dus_tail16p8): /* less then 8 bytes left. */
|
|
|
00db10 |
+L(dus_tail16p8): /* less than 8 bytes left. */
|
|
|
00db10 |
beq cr1,L(dus_tailX) /* exactly 16 bytes, early exit. */
|
|
|
00db10 |
cmpldi cr1,10,20
|
|
|
00db10 |
bf 29,L(dus_tail16p2)
|
|
|
00db10 |
@@ -466,7 +477,7 @@
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
-L(dus_tail16p4): /* less then 4 bytes left. */
|
|
|
00db10 |
+L(dus_tail16p4): /* less than 4 bytes left. */
|
|
|
00db10 |
addi 12,12,24
|
|
|
00db10 |
addi 3,3,24
|
|
|
00db10 |
bgt cr0,L(dus_tail2)
|
|
|
00db10 |
@@ -474,7 +485,7 @@
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
-L(dus_tail16p2): /* 16 bytes moved, less then 4 bytes left. */
|
|
|
00db10 |
+L(dus_tail16p2): /* 16 bytes moved, less than 4 bytes left. */
|
|
|
00db10 |
addi 12,12,16
|
|
|
00db10 |
addi 3,3,16
|
|
|
00db10 |
b L(dus_tail2)
|
|
|
00db10 |
@@ -499,7 +510,7 @@
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
-L(dus_tail8p4): /* less then 4 bytes left. */
|
|
|
00db10 |
+L(dus_tail8p4): /* less than 4 bytes left. */
|
|
|
00db10 |
addi 12,12,8
|
|
|
00db10 |
addi 3,3,8
|
|
|
00db10 |
bgt cr1,L(dus_tail2)
|
|
|
00db10 |
@@ -510,14 +521,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(dus_tail4): /* Move 4 bytes. */
|
|
|
00db10 |
/* r6 already loaded speculatively. If we are here we know there is
|
|
|
00db10 |
- more then 4 bytes left. So there is no need to test. */
|
|
|
00db10 |
+ more than 4 bytes left. So there is no need to test. */
|
|
|
00db10 |
addi 12,12,4
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
addi 3,3,4
|
|
|
00db10 |
L(dus_tail2): /* Move 2-3 bytes. */
|
|
|
00db10 |
bf 30,L(dus_tail1)
|
|
|
00db10 |
lhz 6,0(12)
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
+ sth 6,0(3)
|
|
|
00db10 |
bf 31,L(dus_tailX)
|
|
|
00db10 |
lbz 7,2(12)
|
|
|
00db10 |
stb 7,2(3)
|
|
|
00db10 |
@@ -537,7 +548,7 @@
|
|
|
00db10 |
.LE8:
|
|
|
00db10 |
mr 12,4
|
|
|
00db10 |
bne cr6,L(dus_4)
|
|
|
00db10 |
-/* Exactly 8 bytes. We may cross a 32-/128-byte boundry and take a ~20
|
|
|
00db10 |
+/* Exactly 8 bytes. We may cross a 32-/128-byte boundary and take a ~20
|
|
|
00db10 |
cycle delay. This case should be rare and any attempt to avoid this
|
|
|
00db10 |
would take most of 20 cycles any way. */
|
|
|
00db10 |
ld 6,0(4)
|
|
|
00db10 |
@@ -552,7 +563,7 @@
|
|
|
00db10 |
stw 6,0(3)
|
|
|
00db10 |
bf 30,L(dus_5)
|
|
|
00db10 |
lhz 7,4(4)
|
|
|
00db10 |
- sth 7,4(3)
|
|
|
00db10 |
+ sth 7,4(3)
|
|
|
00db10 |
bf 31,L(dus_0)
|
|
|
00db10 |
lbz 8,6(4)
|
|
|
00db10 |
stb 8,6(3)
|
|
|
00db10 |
@@ -590,20 +601,31 @@
|
|
|
00db10 |
bge cr0, L(du4_do)
|
|
|
00db10 |
blt cr5, L(du1_do)
|
|
|
00db10 |
beq cr5, L(du2_do)
|
|
|
00db10 |
- b L(du3_do)
|
|
|
00db10 |
-
|
|
|
00db10 |
+ b L(du3_do)
|
|
|
00db10 |
+
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du1_do):
|
|
|
00db10 |
bf 30,L(du1_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+ /* FIXME: can combine last shift and "or" into "rldimi" */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 8
|
|
|
00db10 |
+ sldi 8,7, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 8
|
|
|
00db10 |
srdi 8,7, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 8
|
|
|
00db10 |
+ sldi 8,6, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 8
|
|
|
00db10 |
srdi 8,6, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -612,8 +634,13 @@
|
|
|
00db10 |
blt cr6,L(du1_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du1_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 8
|
|
|
00db10 |
+ sldi 8,7, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 8
|
|
|
00db10 |
srdi 8,7, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -624,8 +651,13 @@
|
|
|
00db10 |
b L(du1_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du1_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 8
|
|
|
00db10 |
+ sldi 8,7, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 8
|
|
|
00db10 |
srdi 8,7, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du1_loop)
|
|
|
00db10 |
@@ -637,23 +669,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du1_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 8
|
|
|
00db10 |
+ sldi 8,7, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 8
|
|
|
00db10 |
srdi 8,7, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 8
|
|
|
00db10 |
+ sldi 8,6, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 8
|
|
|
00db10 |
srdi 8,6, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 8
|
|
|
00db10 |
+ sldi 8,7, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 8
|
|
|
00db10 |
srdi 8,7, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 8
|
|
|
00db10 |
+ sldi 8,6, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 8
|
|
|
00db10 |
srdi 8,6, 64-8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -663,9 +715,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du1_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 8
|
|
|
00db10 |
+ sldi 8,7, 64-8
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 8
|
|
|
00db10 |
srdi 8,7, 64-8
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
|
|
|
00db10 |
@@ -674,13 +731,23 @@
|
|
|
00db10 |
bf 30,L(du2_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 16
|
|
|
00db10 |
+ sldi 8,7, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 16
|
|
|
00db10 |
srdi 8,7, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 16
|
|
|
00db10 |
+ sldi 8,6, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 16
|
|
|
00db10 |
srdi 8,6, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -689,8 +756,13 @@
|
|
|
00db10 |
blt cr6,L(du2_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du2_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 16
|
|
|
00db10 |
+ sldi 8,7, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 16
|
|
|
00db10 |
srdi 8,7, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -701,8 +773,13 @@
|
|
|
00db10 |
b L(du2_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du2_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 16
|
|
|
00db10 |
+ sldi 8,7, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 16
|
|
|
00db10 |
srdi 8,7, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du2_loop)
|
|
|
00db10 |
@@ -714,23 +791,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du2_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 16
|
|
|
00db10 |
+ sldi 8,7, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 16
|
|
|
00db10 |
srdi 8,7, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 16
|
|
|
00db10 |
+ sldi 8,6, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 16
|
|
|
00db10 |
srdi 8,6, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 16
|
|
|
00db10 |
+ sldi 8,7, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 16
|
|
|
00db10 |
srdi 8,7, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 16
|
|
|
00db10 |
+ sldi 8,6, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 16
|
|
|
00db10 |
srdi 8,6, 64-16
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -740,9 +837,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du2_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 16
|
|
|
00db10 |
+ sldi 8,7, 64-16
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 16
|
|
|
00db10 |
srdi 8,7, 64-16
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
|
|
|
00db10 |
@@ -751,13 +853,23 @@
|
|
|
00db10 |
bf 30,L(du3_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 24
|
|
|
00db10 |
+ sldi 8,7, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 24
|
|
|
00db10 |
srdi 8,7, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 24
|
|
|
00db10 |
+ sldi 8,6, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 24
|
|
|
00db10 |
srdi 8,6, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -766,8 +878,13 @@
|
|
|
00db10 |
blt cr6,L(du3_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du3_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 24
|
|
|
00db10 |
+ sldi 8,7, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 24
|
|
|
00db10 |
srdi 8,7, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -778,8 +895,13 @@
|
|
|
00db10 |
b L(du3_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du3_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 24
|
|
|
00db10 |
+ sldi 8,7, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 24
|
|
|
00db10 |
srdi 8,7, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du3_loop)
|
|
|
00db10 |
@@ -791,23 +913,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du3_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 24
|
|
|
00db10 |
+ sldi 8,7, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 24
|
|
|
00db10 |
srdi 8,7, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 24
|
|
|
00db10 |
+ sldi 8,6, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 24
|
|
|
00db10 |
srdi 8,6, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 24
|
|
|
00db10 |
+ sldi 8,7, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 24
|
|
|
00db10 |
srdi 8,7, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 24
|
|
|
00db10 |
+ sldi 8,6, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 24
|
|
|
00db10 |
srdi 8,6, 64-24
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -817,9 +959,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du3_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 24
|
|
|
00db10 |
+ sldi 8,7, 64-24
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 24
|
|
|
00db10 |
srdi 8,7, 64-24
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
|
|
|
00db10 |
@@ -834,13 +981,23 @@
|
|
|
00db10 |
bf 30,L(du4_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 32
|
|
|
00db10 |
+ sldi 8,7, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 32
|
|
|
00db10 |
srdi 8,7, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 32
|
|
|
00db10 |
+ sldi 8,6, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 32
|
|
|
00db10 |
srdi 8,6, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -849,8 +1006,13 @@
|
|
|
00db10 |
blt cr6,L(du4_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du4_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 32
|
|
|
00db10 |
+ sldi 8,7, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 32
|
|
|
00db10 |
srdi 8,7, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -861,8 +1023,13 @@
|
|
|
00db10 |
b L(du4_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du4_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 32
|
|
|
00db10 |
+ sldi 8,7, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 32
|
|
|
00db10 |
srdi 8,7, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du4_loop)
|
|
|
00db10 |
@@ -874,23 +1041,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du4_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 32
|
|
|
00db10 |
+ sldi 8,7, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 32
|
|
|
00db10 |
srdi 8,7, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 32
|
|
|
00db10 |
+ sldi 8,6, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 32
|
|
|
00db10 |
srdi 8,6, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 32
|
|
|
00db10 |
+ sldi 8,7, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 32
|
|
|
00db10 |
srdi 8,7, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 32
|
|
|
00db10 |
+ sldi 8,6, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 32
|
|
|
00db10 |
srdi 8,6, 64-32
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -900,9 +1087,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du4_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 32
|
|
|
00db10 |
+ sldi 8,7, 64-32
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 32
|
|
|
00db10 |
srdi 8,7, 64-32
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
|
|
|
00db10 |
@@ -911,13 +1103,23 @@
|
|
|
00db10 |
bf 30,L(du5_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 40
|
|
|
00db10 |
+ sldi 8,7, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 40
|
|
|
00db10 |
srdi 8,7, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 40
|
|
|
00db10 |
+ sldi 8,6, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 40
|
|
|
00db10 |
srdi 8,6, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -926,8 +1128,13 @@
|
|
|
00db10 |
blt cr6,L(du5_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du5_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 40
|
|
|
00db10 |
+ sldi 8,7, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 40
|
|
|
00db10 |
srdi 8,7, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -938,8 +1145,13 @@
|
|
|
00db10 |
b L(du5_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du5_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 40
|
|
|
00db10 |
+ sldi 8,7, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 40
|
|
|
00db10 |
srdi 8,7, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du5_loop)
|
|
|
00db10 |
@@ -951,23 +1163,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du5_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 40
|
|
|
00db10 |
+ sldi 8,7, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 40
|
|
|
00db10 |
srdi 8,7, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 40
|
|
|
00db10 |
+ sldi 8,6, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 40
|
|
|
00db10 |
srdi 8,6, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 40
|
|
|
00db10 |
+ sldi 8,7, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 40
|
|
|
00db10 |
srdi 8,7, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 40
|
|
|
00db10 |
+ sldi 8,6, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 40
|
|
|
00db10 |
srdi 8,6, 64-40
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -977,9 +1209,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du5_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 40
|
|
|
00db10 |
+ sldi 8,7, 64-40
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 40
|
|
|
00db10 |
srdi 8,7, 64-40
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
|
|
|
00db10 |
@@ -988,13 +1225,23 @@
|
|
|
00db10 |
bf 30,L(du6_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 48
|
|
|
00db10 |
+ sldi 8,7, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 48
|
|
|
00db10 |
srdi 8,7, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 48
|
|
|
00db10 |
+ sldi 8,6, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 48
|
|
|
00db10 |
srdi 8,6, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -1003,8 +1250,13 @@
|
|
|
00db10 |
blt cr6,L(du6_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du6_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 48
|
|
|
00db10 |
+ sldi 8,7, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 48
|
|
|
00db10 |
srdi 8,7, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -1015,8 +1267,13 @@
|
|
|
00db10 |
b L(du6_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du6_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 48
|
|
|
00db10 |
+ sldi 8,7, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 48
|
|
|
00db10 |
srdi 8,7, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du6_loop)
|
|
|
00db10 |
@@ -1028,23 +1285,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du6_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 48
|
|
|
00db10 |
+ sldi 8,7, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 48
|
|
|
00db10 |
srdi 8,7, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 48
|
|
|
00db10 |
+ sldi 8,6, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 48
|
|
|
00db10 |
srdi 8,6, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 48
|
|
|
00db10 |
+ sldi 8,7, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 48
|
|
|
00db10 |
srdi 8,7, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 48
|
|
|
00db10 |
+ sldi 8,6, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 48
|
|
|
00db10 |
srdi 8,6, 64-48
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -1054,9 +1331,14 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du6_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 48
|
|
|
00db10 |
+ sldi 8,7, 64-48
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 48
|
|
|
00db10 |
srdi 8,7, 64-48
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
|
|
|
00db10 |
@@ -1065,13 +1347,23 @@
|
|
|
00db10 |
bf 30,L(du7_1dw)
|
|
|
00db10 |
|
|
|
00db10 |
/* there are at least two DWs to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 56
|
|
|
00db10 |
+ sldi 8,7, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 56
|
|
|
00db10 |
srdi 8,7, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 56
|
|
|
00db10 |
+ sldi 8,6, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 56
|
|
|
00db10 |
srdi 8,6, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
@@ -1080,8 +1372,13 @@
|
|
|
00db10 |
blt cr6,L(du7_fini) /* if total DWs = 3, then bypass loop */
|
|
|
00db10 |
bf 31,L(du7_loop)
|
|
|
00db10 |
/* there is a third DW to copy */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 56
|
|
|
00db10 |
+ sldi 8,7, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 56
|
|
|
00db10 |
srdi 8,7, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
mr 6,7
|
|
|
00db10 |
@@ -1092,8 +1389,13 @@
|
|
|
00db10 |
b L(du7_loop)
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du7_1dw):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 56
|
|
|
00db10 |
+ sldi 8,7, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 56
|
|
|
00db10 |
srdi 8,7, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 5,5,16
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
bf 31,L(du7_loop)
|
|
|
00db10 |
@@ -1105,23 +1407,43 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
/* copy 32 bytes at a time */
|
|
|
00db10 |
L(du7_loop):
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 56
|
|
|
00db10 |
+ sldi 8,7, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 56
|
|
|
00db10 |
srdi 8,7, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,0(5)
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 56
|
|
|
00db10 |
+ sldi 8,6, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 56
|
|
|
00db10 |
srdi 8,6, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,8(5)
|
|
|
00db10 |
std 0,8(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 56
|
|
|
00db10 |
+ sldi 8,7, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 56
|
|
|
00db10 |
srdi 8,7, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 6,16(5)
|
|
|
00db10 |
std 0,16(4)
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,7, 56
|
|
|
00db10 |
+ sldi 8,6, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,7, 56
|
|
|
00db10 |
srdi 8,6, 64-56
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
or 0,0,8
|
|
|
00db10 |
ld 7,24(5)
|
|
|
00db10 |
std 0,24(4)
|
|
|
00db10 |
@@ -1131,12 +1453,17 @@
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du7_fini):
|
|
|
00db10 |
/* calculate and store the final DW */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ srdi 0,6, 56
|
|
|
00db10 |
+ sldi 8,7, 64-56
|
|
|
00db10 |
+#else
|
|
|
00db10 |
sldi 0,6, 56
|
|
|
00db10 |
srdi 8,7, 64-56
|
|
|
00db10 |
- or 0,0,8
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ or 0,0,8
|
|
|
00db10 |
std 0,0(4)
|
|
|
00db10 |
b L(du_done)
|
|
|
00db10 |
-
|
|
|
00db10 |
+
|
|
|
00db10 |
.align 4
|
|
|
00db10 |
L(du_done):
|
|
|
00db10 |
rldicr 0,31,0,60
|
|
|
00db10 |
@@ -1144,9 +1471,9 @@
|
|
|
00db10 |
beq cr1,0f /* If the tail is 0 bytes we are done! */
|
|
|
00db10 |
|
|
|
00db10 |
add 3,3,0
|
|
|
00db10 |
- add 12,12,0
|
|
|
00db10 |
+ add 12,12,0
|
|
|
00db10 |
/* At this point we have a tail of 0-7 bytes and we know that the
|
|
|
00db10 |
- destiniation is double word aligned. */
|
|
|
00db10 |
+ destination is double word aligned. */
|
|
|
00db10 |
4: bf 29,2f
|
|
|
00db10 |
lwz 6,0(12)
|
|
|
00db10 |
addi 12,12,4
|
|
|
00db10 |
@@ -1165,5 +1492,5 @@
|
|
|
00db10 |
ld 31,-8(1)
|
|
|
00db10 |
ld 3,-16(1)
|
|
|
00db10 |
blr
|
|
|
00db10 |
-END_GEN_TB (BP_SYM (memcpy),TB_TOCLESS)
|
|
|
00db10 |
+END_GEN_TB (memcpy,TB_TOCLESS)
|
|
|
00db10 |
libc_hidden_builtin_def (memcpy)
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/memcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/memcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/memcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/memcpy.S 2014-05-29 13:05:40.000000000 -0500
|
|
|
00db10 |
@@ -1,5 +1,5 @@
|
|
|
00db10 |
/* Optimized memcpy implementation for PowerPC64/POWER7.
|
|
|
00db10 |
- Copyright (C) 2010, 2011 Free Software Foundation, Inc.
|
|
|
00db10 |
+ Copyright (C) 2010-2014 Free Software Foundation, Inc.
|
|
|
00db10 |
Contributed by Luis Machado <luisgpm@br.ibm.com>.
|
|
|
00db10 |
This file is part of the GNU C Library.
|
|
|
00db10 |
|
|
|
00db10 |
@@ -18,425 +18,366 @@
|
|
|
00db10 |
<http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
|
|
|
00db10 |
#include <sysdep.h>
|
|
|
00db10 |
-#include <bp-sym.h>
|
|
|
00db10 |
-#include <bp-asm.h>
|
|
|
00db10 |
|
|
|
00db10 |
|
|
|
00db10 |
/* __ptr_t [r3] memcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
|
|
|
00db10 |
Returns 'dst'. */
|
|
|
00db10 |
|
|
|
00db10 |
+#define dst 11 /* Use r11 so r3 kept unchanged. */
|
|
|
00db10 |
+#define src 4
|
|
|
00db10 |
+#define cnt 5
|
|
|
00db10 |
+
|
|
|
00db10 |
.machine power7
|
|
|
00db10 |
-EALIGN (BP_SYM (memcpy), 5, 0)
|
|
|
00db10 |
+EALIGN (memcpy, 5, 0)
|
|
|
00db10 |
CALL_MCOUNT 3
|
|
|
00db10 |
|
|
|
00db10 |
- cmpldi cr1,5,31
|
|
|
00db10 |
+ cmpldi cr1,cnt,31
|
|
|
00db10 |
neg 0,3
|
|
|
00db10 |
- std 3,-16(1)
|
|
|
00db10 |
- std 31,-8(1)
|
|
|
00db10 |
- cfi_offset(31,-8)
|
|
|
00db10 |
ble cr1, L(copy_LT_32) /* If move < 32 bytes use short move
|
|
|
00db10 |
code. */
|
|
|
00db10 |
|
|
|
00db10 |
- andi. 11,3,7 /* Check alignment of DST. */
|
|
|
00db10 |
-
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+/* In little-endian mode, power7 takes an alignment trap on any lxvd2x
|
|
|
00db10 |
+ or stxvd2x crossing a 32-byte boundary, so ensure the aligned_copy
|
|
|
00db10 |
+ loop is only used for quadword aligned copies. */
|
|
|
00db10 |
+ andi. 10,3,15
|
|
|
00db10 |
+ clrldi 11,4,60
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ andi. 10,3,7 /* Check alignment of DST. */
|
|
|
00db10 |
+ clrldi 11,4,61 /* Check alignment of SRC. */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ cmpld cr6,10,11 /* SRC and DST alignments match? */
|
|
|
00db10 |
|
|
|
00db10 |
- clrldi 10,4,61 /* Check alignment of SRC. */
|
|
|
00db10 |
- cmpld cr6,10,11 /* SRC and DST alignments match? */
|
|
|
00db10 |
- mr 12,4
|
|
|
00db10 |
- mr 31,5
|
|
|
00db10 |
+ mr dst,3
|
|
|
00db10 |
bne cr6,L(copy_GE_32_unaligned)
|
|
|
00db10 |
+ beq L(aligned_copy)
|
|
|
00db10 |
|
|
|
00db10 |
- srdi 9,5,3 /* Number of full quadwords remaining. */
|
|
|
00db10 |
-
|
|
|
00db10 |
- beq L(copy_GE_32_aligned_cont)
|
|
|
00db10 |
-
|
|
|
00db10 |
- clrldi 0,0,61
|
|
|
00db10 |
- mtcrf 0x01,0
|
|
|
00db10 |
- subf 31,0,5
|
|
|
00db10 |
-
|
|
|
00db10 |
- /* Get the SRC aligned to 8 bytes. */
|
|
|
00db10 |
-
|
|
|
00db10 |
-1: bf 31,2f
|
|
|
00db10 |
- lbz 6,0(12)
|
|
|
00db10 |
- addi 12,12,1
|
|
|
00db10 |
- stb 6,0(3)
|
|
|
00db10 |
- addi 3,3,1
|
|
|
00db10 |
-2: bf 30,4f
|
|
|
00db10 |
- lhz 6,0(12)
|
|
|
00db10 |
- addi 12,12,2
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
- addi 3,3,2
|
|
|
00db10 |
-4: bf 29,0f
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- addi 12,12,4
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- addi 3,3,4
|
|
|
00db10 |
-0:
|
|
|
00db10 |
- clrldi 10,12,61 /* Check alignment of SRC again. */
|
|
|
00db10 |
- srdi 9,31,3 /* Number of full doublewords remaining. */
|
|
|
00db10 |
-
|
|
|
00db10 |
-L(copy_GE_32_aligned_cont):
|
|
|
00db10 |
-
|
|
|
00db10 |
- clrldi 11,31,61
|
|
|
00db10 |
- mtcrf 0x01,9
|
|
|
00db10 |
-
|
|
|
00db10 |
- srdi 8,31,5
|
|
|
00db10 |
- cmpldi cr1,9,4
|
|
|
00db10 |
- cmpldi cr6,11,0
|
|
|
00db10 |
- mr 11,12
|
|
|
00db10 |
+ mtocrf 0x01,0
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ clrldi 0,0,60
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ clrldi 0,0,61
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
|
|
|
00db10 |
- /* Copy 1~3 doublewords so the main loop starts
|
|
|
00db10 |
- at a multiple of 32 bytes. */
|
|
|
00db10 |
-
|
|
|
00db10 |
- bf 30,1f
|
|
|
00db10 |
- ld 6,0(12)
|
|
|
00db10 |
- ld 7,8(12)
|
|
|
00db10 |
- addi 11,12,16
|
|
|
00db10 |
- mtctr 8
|
|
|
00db10 |
- std 6,0(3)
|
|
|
00db10 |
- std 7,8(3)
|
|
|
00db10 |
- addi 10,3,16
|
|
|
00db10 |
- bf 31,4f
|
|
|
00db10 |
- ld 0,16(12)
|
|
|
00db10 |
- std 0,16(3)
|
|
|
00db10 |
- blt cr1,3f
|
|
|
00db10 |
- addi 11,12,24
|
|
|
00db10 |
- addi 10,3,24
|
|
|
00db10 |
- b 4f
|
|
|
00db10 |
-
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
-1: /* Copy 1 doubleword and set the counter. */
|
|
|
00db10 |
- mr 10,3
|
|
|
00db10 |
- mtctr 8
|
|
|
00db10 |
- bf 31,4f
|
|
|
00db10 |
- ld 6,0(12)
|
|
|
00db10 |
- addi 11,12,8
|
|
|
00db10 |
- std 6,0(3)
|
|
|
00db10 |
- addi 10,3,8
|
|
|
00db10 |
+/* Get the DST and SRC aligned to 8 bytes (16 for little-endian). */
|
|
|
00db10 |
+1:
|
|
|
00db10 |
+ bf 31,2f
|
|
|
00db10 |
+ lbz 6,0(src)
|
|
|
00db10 |
+ addi src,src,1
|
|
|
00db10 |
+ stb 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,1
|
|
|
00db10 |
+2:
|
|
|
00db10 |
+ bf 30,4f
|
|
|
00db10 |
+ lhz 6,0(src)
|
|
|
00db10 |
+ addi src,src,2
|
|
|
00db10 |
+ sth 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,2
|
|
|
00db10 |
+4:
|
|
|
00db10 |
+ bf 29,8f
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ addi src,src,4
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,4
|
|
|
00db10 |
+8:
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ bf 28,16f
|
|
|
00db10 |
+ ld 6,0(src)
|
|
|
00db10 |
+ addi src,src,8
|
|
|
00db10 |
+ std 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,8
|
|
|
00db10 |
+16:
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ subf cnt,0,cnt
|
|
|
00db10 |
|
|
|
00db10 |
+/* Main aligned copy loop. Copies 128 bytes at a time. */
|
|
|
00db10 |
L(aligned_copy):
|
|
|
00db10 |
- /* Main aligned copy loop. Copies up to 128-bytes at a time. */
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
-4:
|
|
|
00db10 |
- /* check for any 32-byte or 64-byte lumps that are outside of a
|
|
|
00db10 |
- nice 128-byte range. R8 contains the number of 32-byte
|
|
|
00db10 |
- lumps, so drop this into the CR, and use the SO/EQ bits to help
|
|
|
00db10 |
- handle the 32- or 64- byte lumps. Then handle the rest with an
|
|
|
00db10 |
- unrolled 128-bytes-at-a-time copy loop. */
|
|
|
00db10 |
- mtocrf 1,8
|
|
|
00db10 |
- li 6,16 # 16() index
|
|
|
00db10 |
- li 7,32 # 32() index
|
|
|
00db10 |
- li 8,48 # 48() index
|
|
|
00db10 |
-
|
|
|
00db10 |
-L(aligned_32byte):
|
|
|
00db10 |
- /* if the SO bit (indicating a 32-byte lump) is not set, move along. */
|
|
|
00db10 |
- bns cr7,L(aligned_64byte)
|
|
|
00db10 |
- lxvd2x 6,0,11
|
|
|
00db10 |
- lxvd2x 7,11,6
|
|
|
00db10 |
- addi 11,11,32
|
|
|
00db10 |
- stxvd2x 6,0,10
|
|
|
00db10 |
- stxvd2x 7,10,6
|
|
|
00db10 |
- addi 10,10,32
|
|
|
00db10 |
-
|
|
|
00db10 |
-L(aligned_64byte):
|
|
|
00db10 |
- /* if the EQ bit (indicating a 64-byte lump) is not set, move along. */
|
|
|
00db10 |
- bne cr7,L(aligned_128setup)
|
|
|
00db10 |
- lxvd2x 6,0,11
|
|
|
00db10 |
- lxvd2x 7,11,6
|
|
|
00db10 |
- lxvd2x 8,11,7
|
|
|
00db10 |
- lxvd2x 9,11,8
|
|
|
00db10 |
- addi 11,11,64
|
|
|
00db10 |
- stxvd2x 6,0,10
|
|
|
00db10 |
- stxvd2x 7,10,6
|
|
|
00db10 |
- stxvd2x 8,10,7
|
|
|
00db10 |
- stxvd2x 9,10,8
|
|
|
00db10 |
- addi 10,10,64
|
|
|
00db10 |
-
|
|
|
00db10 |
-L(aligned_128setup):
|
|
|
00db10 |
- /* Set up for the 128-byte at a time copy loop. */
|
|
|
00db10 |
- srdi 8,31,7
|
|
|
00db10 |
- cmpdi 8,0 # Any 4x lumps left?
|
|
|
00db10 |
- beq 3f # if not, move along.
|
|
|
00db10 |
- lxvd2x 6,0,11
|
|
|
00db10 |
- lxvd2x 7,11,6
|
|
|
00db10 |
- mtctr 8 # otherwise, load the ctr and begin.
|
|
|
00db10 |
- li 8,48 # 48() index
|
|
|
00db10 |
+ li 6,16
|
|
|
00db10 |
+ li 7,32
|
|
|
00db10 |
+ li 8,48
|
|
|
00db10 |
+ mtocrf 0x02,cnt
|
|
|
00db10 |
+ srdi 12,cnt,7
|
|
|
00db10 |
+ cmpdi 12,0
|
|
|
00db10 |
+ beq L(aligned_tail)
|
|
|
00db10 |
+ lxvd2x 6,0,src
|
|
|
00db10 |
+ lxvd2x 7,src,6
|
|
|
00db10 |
+ mtctr 12
|
|
|
00db10 |
b L(aligned_128loop)
|
|
|
00db10 |
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(aligned_128head):
|
|
|
00db10 |
/* for the 2nd + iteration of this loop. */
|
|
|
00db10 |
- lxvd2x 6,0,11
|
|
|
00db10 |
- lxvd2x 7,11,6
|
|
|
00db10 |
+ lxvd2x 6,0,src
|
|
|
00db10 |
+ lxvd2x 7,src,6
|
|
|
00db10 |
L(aligned_128loop):
|
|
|
00db10 |
- lxvd2x 8,11,7
|
|
|
00db10 |
- lxvd2x 9,11,8
|
|
|
00db10 |
- stxvd2x 6,0,10
|
|
|
00db10 |
- addi 11,11,64
|
|
|
00db10 |
- stxvd2x 7,10,6
|
|
|
00db10 |
- stxvd2x 8,10,7
|
|
|
00db10 |
- stxvd2x 9,10,8
|
|
|
00db10 |
- lxvd2x 6,0,11
|
|
|
00db10 |
- lxvd2x 7,11,6
|
|
|
00db10 |
- addi 10,10,64
|
|
|
00db10 |
- lxvd2x 8,11,7
|
|
|
00db10 |
- lxvd2x 9,11,8
|
|
|
00db10 |
- addi 11,11,64
|
|
|
00db10 |
- stxvd2x 6,0,10
|
|
|
00db10 |
- stxvd2x 7,10,6
|
|
|
00db10 |
- stxvd2x 8,10,7
|
|
|
00db10 |
- stxvd2x 9,10,8
|
|
|
00db10 |
- addi 10,10,64
|
|
|
00db10 |
+ lxvd2x 8,src,7
|
|
|
00db10 |
+ lxvd2x 9,src,8
|
|
|
00db10 |
+ stxvd2x 6,0,dst
|
|
|
00db10 |
+ addi src,src,64
|
|
|
00db10 |
+ stxvd2x 7,dst,6
|
|
|
00db10 |
+ stxvd2x 8,dst,7
|
|
|
00db10 |
+ stxvd2x 9,dst,8
|
|
|
00db10 |
+ lxvd2x 6,0,src
|
|
|
00db10 |
+ lxvd2x 7,src,6
|
|
|
00db10 |
+ addi dst,dst,64
|
|
|
00db10 |
+ lxvd2x 8,src,7
|
|
|
00db10 |
+ lxvd2x 9,src,8
|
|
|
00db10 |
+ addi src,src,64
|
|
|
00db10 |
+ stxvd2x 6,0,dst
|
|
|
00db10 |
+ stxvd2x 7,dst,6
|
|
|
00db10 |
+ stxvd2x 8,dst,7
|
|
|
00db10 |
+ stxvd2x 9,dst,8
|
|
|
00db10 |
+ addi dst,dst,64
|
|
|
00db10 |
bdnz L(aligned_128head)
|
|
|
00db10 |
|
|
|
00db10 |
-3:
|
|
|
00db10 |
- /* Check for tail bytes. */
|
|
|
00db10 |
- rldicr 0,31,0,60
|
|
|
00db10 |
- mtcrf 0x01,31
|
|
|
00db10 |
- beq cr6,0f
|
|
|
00db10 |
-
|
|
|
00db10 |
-.L9:
|
|
|
00db10 |
- add 3,3,0
|
|
|
00db10 |
- add 12,12,0
|
|
|
00db10 |
-
|
|
|
00db10 |
- /* At this point we have a tail of 0-7 bytes and we know that the
|
|
|
00db10 |
- destination is doubleword-aligned. */
|
|
|
00db10 |
-4: /* Copy 4 bytes. */
|
|
|
00db10 |
- bf 29,2f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- addi 12,12,4
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- addi 3,3,4
|
|
|
00db10 |
-2: /* Copy 2 bytes. */
|
|
|
00db10 |
- bf 30,1f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lhz 6,0(12)
|
|
|
00db10 |
- addi 12,12,2
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
- addi 3,3,2
|
|
|
00db10 |
-1: /* Copy 1 byte. */
|
|
|
00db10 |
- bf 31,0f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lbz 6,0(12)
|
|
|
00db10 |
- stb 6,0(3)
|
|
|
00db10 |
-0: /* Return original DST pointer. */
|
|
|
00db10 |
- ld 31,-8(1)
|
|
|
00db10 |
- ld 3,-16(1)
|
|
|
00db10 |
+L(aligned_tail):
|
|
|
00db10 |
+ mtocrf 0x01,cnt
|
|
|
00db10 |
+ bf 25,32f
|
|
|
00db10 |
+ lxvd2x 6,0,src
|
|
|
00db10 |
+ lxvd2x 7,src,6
|
|
|
00db10 |
+ lxvd2x 8,src,7
|
|
|
00db10 |
+ lxvd2x 9,src,8
|
|
|
00db10 |
+ addi src,src,64
|
|
|
00db10 |
+ stxvd2x 6,0,dst
|
|
|
00db10 |
+ stxvd2x 7,dst,6
|
|
|
00db10 |
+ stxvd2x 8,dst,7
|
|
|
00db10 |
+ stxvd2x 9,dst,8
|
|
|
00db10 |
+ addi dst,dst,64
|
|
|
00db10 |
+32:
|
|
|
00db10 |
+ bf 26,16f
|
|
|
00db10 |
+ lxvd2x 6,0,src
|
|
|
00db10 |
+ lxvd2x 7,src,6
|
|
|
00db10 |
+ addi src,src,32
|
|
|
00db10 |
+ stxvd2x 6,0,dst
|
|
|
00db10 |
+ stxvd2x 7,dst,6
|
|
|
00db10 |
+ addi dst,dst,32
|
|
|
00db10 |
+16:
|
|
|
00db10 |
+ bf 27,8f
|
|
|
00db10 |
+ lxvd2x 6,0,src
|
|
|
00db10 |
+ addi src,src,16
|
|
|
00db10 |
+ stxvd2x 6,0,dst
|
|
|
00db10 |
+ addi dst,dst,16
|
|
|
00db10 |
+8:
|
|
|
00db10 |
+ bf 28,4f
|
|
|
00db10 |
+ ld 6,0(src)
|
|
|
00db10 |
+ addi src,src,8
|
|
|
00db10 |
+ std 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,8
|
|
|
00db10 |
+4: /* Copies 4~7 bytes. */
|
|
|
00db10 |
+ bf 29,L(tail2)
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ bf 30,L(tail5)
|
|
|
00db10 |
+ lhz 7,4(src)
|
|
|
00db10 |
+ sth 7,4(dst)
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz 8,6(src)
|
|
|
00db10 |
+ stb 8,6(dst)
|
|
|
00db10 |
+ /* Return original DST pointer. */
|
|
|
00db10 |
blr
|
|
|
00db10 |
|
|
|
00db10 |
- /* Handle copies of 0~31 bytes. */
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Handle copies of 0~31 bytes. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(copy_LT_32):
|
|
|
00db10 |
- cmpldi cr6,5,8
|
|
|
00db10 |
- mr 12,4
|
|
|
00db10 |
- mtcrf 0x01,5
|
|
|
00db10 |
+ mr dst,3
|
|
|
00db10 |
+ cmpldi cr6,cnt,8
|
|
|
00db10 |
+ mtocrf 0x01,cnt
|
|
|
00db10 |
ble cr6,L(copy_LE_8)
|
|
|
00db10 |
|
|
|
00db10 |
/* At least 9 bytes to go. */
|
|
|
00db10 |
neg 8,4
|
|
|
00db10 |
- clrrdi 11,4,2
|
|
|
00db10 |
- andi. 0,8,3
|
|
|
00db10 |
- cmpldi cr1,5,16
|
|
|
00db10 |
- mr 10,5
|
|
|
00db10 |
+ andi. 0,8,3
|
|
|
00db10 |
+ cmpldi cr1,cnt,16
|
|
|
00db10 |
beq L(copy_LT_32_aligned)
|
|
|
00db10 |
|
|
|
00db10 |
- /* Force 4-bytes alignment for SRC. */
|
|
|
00db10 |
- mtocrf 0x01,0
|
|
|
00db10 |
- subf 10,0,5
|
|
|
00db10 |
-2: bf 30,1f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lhz 6,0(12)
|
|
|
00db10 |
- addi 12,12,2
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
- addi 3,3,2
|
|
|
00db10 |
-1: bf 31,L(end_4bytes_alignment)
|
|
|
00db10 |
-
|
|
|
00db10 |
- lbz 6,0(12)
|
|
|
00db10 |
- addi 12,12,1
|
|
|
00db10 |
- stb 6,0(3)
|
|
|
00db10 |
- addi 3,3,1
|
|
|
00db10 |
+ /* Force 4-byte alignment for SRC. */
|
|
|
00db10 |
+ mtocrf 0x01,0
|
|
|
00db10 |
+ subf cnt,0,cnt
|
|
|
00db10 |
+2:
|
|
|
00db10 |
+ bf 30,1f
|
|
|
00db10 |
+ lhz 6,0(src)
|
|
|
00db10 |
+ addi src,src,2
|
|
|
00db10 |
+ sth 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,2
|
|
|
00db10 |
+1:
|
|
|
00db10 |
+ bf 31,L(end_4bytes_alignment)
|
|
|
00db10 |
+ lbz 6,0(src)
|
|
|
00db10 |
+ addi src,src,1
|
|
|
00db10 |
+ stb 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,1
|
|
|
00db10 |
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(end_4bytes_alignment):
|
|
|
00db10 |
- cmpldi cr1,10,16
|
|
|
00db10 |
- mtcrf 0x01,10
|
|
|
00db10 |
+ cmpldi cr1,cnt,16
|
|
|
00db10 |
+ mtocrf 0x01,cnt
|
|
|
00db10 |
|
|
|
00db10 |
L(copy_LT_32_aligned):
|
|
|
00db10 |
/* At least 6 bytes to go, and SRC is word-aligned. */
|
|
|
00db10 |
blt cr1,8f
|
|
|
00db10 |
|
|
|
00db10 |
/* Copy 16 bytes. */
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- lwz 7,4(12)
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- lwz 8,8(12)
|
|
|
00db10 |
- stw 7,4(3)
|
|
|
00db10 |
- lwz 6,12(12)
|
|
|
00db10 |
- addi 12,12,16
|
|
|
00db10 |
- stw 8,8(3)
|
|
|
00db10 |
- stw 6,12(3)
|
|
|
00db10 |
- addi 3,3,16
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ lwz 7,4(src)
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ lwz 8,8(src)
|
|
|
00db10 |
+ stw 7,4(dst)
|
|
|
00db10 |
+ lwz 6,12(src)
|
|
|
00db10 |
+ addi src,src,16
|
|
|
00db10 |
+ stw 8,8(dst)
|
|
|
00db10 |
+ stw 6,12(dst)
|
|
|
00db10 |
+ addi dst,dst,16
|
|
|
00db10 |
8: /* Copy 8 bytes. */
|
|
|
00db10 |
- bf 28,4f
|
|
|
00db10 |
+ bf 28,L(tail4)
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ lwz 7,4(src)
|
|
|
00db10 |
+ addi src,src,8
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ stw 7,4(dst)
|
|
|
00db10 |
+ addi dst,dst,8
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+/* Copies 4~7 bytes. */
|
|
|
00db10 |
+L(tail4):
|
|
|
00db10 |
+ bf 29,L(tail2)
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ bf 30,L(tail5)
|
|
|
00db10 |
+ lhz 7,4(src)
|
|
|
00db10 |
+ sth 7,4(dst)
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz 8,6(src)
|
|
|
00db10 |
+ stb 8,6(dst)
|
|
|
00db10 |
+ /* Return original DST pointer. */
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- lwz 7,4(12)
|
|
|
00db10 |
- addi 12,12,8
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- stw 7,4(3)
|
|
|
00db10 |
- addi 3,3,8
|
|
|
00db10 |
-4: /* Copy 4 bytes. */
|
|
|
00db10 |
- bf 29,2f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- addi 12,12,4
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- addi 3,3,4
|
|
|
00db10 |
-2: /* Copy 2-3 bytes. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+/* Copies 2~3 bytes. */
|
|
|
00db10 |
+L(tail2):
|
|
|
00db10 |
bf 30,1f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lhz 6,0(12)
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
- bf 31,0f
|
|
|
00db10 |
- lbz 7,2(12)
|
|
|
00db10 |
- stb 7,2(3)
|
|
|
00db10 |
- ld 3,-16(1)
|
|
|
00db10 |
+ lhz 6,0(src)
|
|
|
00db10 |
+ sth 6,0(dst)
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz 7,2(src)
|
|
|
00db10 |
+ stb 7,2(dst)
|
|
|
00db10 |
blr
|
|
|
00db10 |
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
-1: /* Copy 1 byte. */
|
|
|
00db10 |
- bf 31,0f
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+L(tail5):
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz 6,4(src)
|
|
|
00db10 |
+ stb 6,4(dst)
|
|
|
00db10 |
+ blr
|
|
|
00db10 |
|
|
|
00db10 |
- lbz 6,0(12)
|
|
|
00db10 |
- stb 6,0(3)
|
|
|
00db10 |
-0: /* Return original DST pointer. */
|
|
|
00db10 |
- ld 3,-16(1)
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
+1:
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz 6,0(src)
|
|
|
00db10 |
+ stb 6,0(dst)
|
|
|
00db10 |
+ /* Return original DST pointer. */
|
|
|
00db10 |
blr
|
|
|
00db10 |
|
|
|
00db10 |
- /* Handles copies of 0~8 bytes. */
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Handles copies of 0~8 bytes. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(copy_LE_8):
|
|
|
00db10 |
- bne cr6,4f
|
|
|
00db10 |
+ bne cr6,L(tail4)
|
|
|
00db10 |
|
|
|
00db10 |
/* Though we could've used ld/std here, they are still
|
|
|
00db10 |
slow for unaligned cases. */
|
|
|
00db10 |
|
|
|
00db10 |
- lwz 6,0(4)
|
|
|
00db10 |
- lwz 7,4(4)
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- stw 7,4(3)
|
|
|
00db10 |
- ld 3,-16(1) /* Return original DST pointers. */
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ lwz 7,4(src)
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ stw 7,4(dst)
|
|
|
00db10 |
blr
|
|
|
00db10 |
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
-4: /* Copies 4~7 bytes. */
|
|
|
00db10 |
- bf 29,2b
|
|
|
00db10 |
|
|
|
00db10 |
- lwz 6,0(4)
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- bf 30,5f
|
|
|
00db10 |
- lhz 7,4(4)
|
|
|
00db10 |
- sth 7,4(3)
|
|
|
00db10 |
- bf 31,0f
|
|
|
00db10 |
- lbz 8,6(4)
|
|
|
00db10 |
- stb 8,6(3)
|
|
|
00db10 |
- ld 3,-16(1)
|
|
|
00db10 |
- blr
|
|
|
00db10 |
-
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
-5: /* Copy 1 byte. */
|
|
|
00db10 |
- bf 31,0f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lbz 6,4(4)
|
|
|
00db10 |
- stb 6,4(3)
|
|
|
00db10 |
-
|
|
|
00db10 |
-0: /* Return original DST pointer. */
|
|
|
00db10 |
- ld 3,-16(1)
|
|
|
00db10 |
- blr
|
|
|
00db10 |
-
|
|
|
00db10 |
- /* Handle copies of 32+ bytes where DST is aligned (to quadword) but
|
|
|
00db10 |
- SRC is not. Use aligned quadword loads from SRC, shifted to realign
|
|
|
00db10 |
- the data, allowing for aligned DST stores. */
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
+/* Handle copies of 32+ bytes where DST is aligned (to quadword) but
|
|
|
00db10 |
+ SRC is not. Use aligned quadword loads from SRC, shifted to realign
|
|
|
00db10 |
+ the data, allowing for aligned DST stores. */
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(copy_GE_32_unaligned):
|
|
|
00db10 |
- clrldi 0,0,60 /* Number of bytes until the 1st
|
|
|
00db10 |
- quadword. */
|
|
|
00db10 |
- andi. 11,3,15 /* Check alignment of DST (against
|
|
|
00db10 |
- quadwords). */
|
|
|
00db10 |
- srdi 9,5,4 /* Number of full quadwords remaining. */
|
|
|
00db10 |
+ clrldi 0,0,60 /* Number of bytes until the 1st dst quadword. */
|
|
|
00db10 |
+#ifndef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ andi. 10,3,15 /* Check alignment of DST (against quadwords). */
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ srdi 9,cnt,4 /* Number of full quadwords remaining. */
|
|
|
00db10 |
|
|
|
00db10 |
beq L(copy_GE_32_unaligned_cont)
|
|
|
00db10 |
|
|
|
00db10 |
- /* SRC is not quadword aligned, get it aligned. */
|
|
|
00db10 |
+ /* DST is not quadword aligned, get it aligned. */
|
|
|
00db10 |
|
|
|
00db10 |
- mtcrf 0x01,0
|
|
|
00db10 |
- subf 31,0,5
|
|
|
00db10 |
+ mtocrf 0x01,0
|
|
|
00db10 |
+ subf cnt,0,cnt
|
|
|
00db10 |
|
|
|
00db10 |
/* Vector instructions work best when proper alignment (16-bytes)
|
|
|
00db10 |
is present. Move 0~15 bytes as needed to get DST quadword-aligned. */
|
|
|
00db10 |
-1: /* Copy 1 byte. */
|
|
|
00db10 |
+1:
|
|
|
00db10 |
bf 31,2f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lbz 6,0(12)
|
|
|
00db10 |
- addi 12,12,1
|
|
|
00db10 |
- stb 6,0(3)
|
|
|
00db10 |
- addi 3,3,1
|
|
|
00db10 |
-2: /* Copy 2 bytes. */
|
|
|
00db10 |
+ lbz 6,0(src)
|
|
|
00db10 |
+ addi src,src,1
|
|
|
00db10 |
+ stb 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,1
|
|
|
00db10 |
+2:
|
|
|
00db10 |
bf 30,4f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lhz 6,0(12)
|
|
|
00db10 |
- addi 12,12,2
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
- addi 3,3,2
|
|
|
00db10 |
-4: /* Copy 4 bytes. */
|
|
|
00db10 |
+ lhz 6,0(src)
|
|
|
00db10 |
+ addi src,src,2
|
|
|
00db10 |
+ sth 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,2
|
|
|
00db10 |
+4:
|
|
|
00db10 |
bf 29,8f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- addi 12,12,4
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- addi 3,3,4
|
|
|
00db10 |
-8: /* Copy 8 bytes. */
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ addi src,src,4
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,4
|
|
|
00db10 |
+8:
|
|
|
00db10 |
bf 28,0f
|
|
|
00db10 |
-
|
|
|
00db10 |
- ld 6,0(12)
|
|
|
00db10 |
- addi 12,12,8
|
|
|
00db10 |
- std 6,0(3)
|
|
|
00db10 |
- addi 3,3,8
|
|
|
00db10 |
+ ld 6,0(src)
|
|
|
00db10 |
+ addi src,src,8
|
|
|
00db10 |
+ std 6,0(dst)
|
|
|
00db10 |
+ addi dst,dst,8
|
|
|
00db10 |
0:
|
|
|
00db10 |
- clrldi 10,12,60 /* Check alignment of SRC. */
|
|
|
00db10 |
- srdi 9,31,4 /* Number of full quadwords remaining. */
|
|
|
00db10 |
+ srdi 9,cnt,4 /* Number of full quadwords remaining. */
|
|
|
00db10 |
|
|
|
00db10 |
/* The proper alignment is present, it is OK to copy the bytes now. */
|
|
|
00db10 |
L(copy_GE_32_unaligned_cont):
|
|
|
00db10 |
|
|
|
00db10 |
/* Setup two indexes to speed up the indexed vector operations. */
|
|
|
00db10 |
- clrldi 11,31,60
|
|
|
00db10 |
- li 6,16 /* Index for 16-bytes offsets. */
|
|
|
00db10 |
+ clrldi 10,cnt,60
|
|
|
00db10 |
+ li 6,16 /* Index for 16-bytes offsets. */
|
|
|
00db10 |
li 7,32 /* Index for 32-bytes offsets. */
|
|
|
00db10 |
- cmpldi cr1,11,0
|
|
|
00db10 |
- srdi 8,31,5 /* Setup the loop counter. */
|
|
|
00db10 |
- mr 10,3
|
|
|
00db10 |
- mr 11,12
|
|
|
00db10 |
- mtcrf 0x01,9
|
|
|
00db10 |
- cmpldi cr6,9,1
|
|
|
00db10 |
- lvsl 5,0,12
|
|
|
00db10 |
- lvx 3,0,12
|
|
|
00db10 |
- bf 31,L(setup_unaligned_loop)
|
|
|
00db10 |
-
|
|
|
00db10 |
- /* Copy another 16 bytes to align to 32-bytes due to the loop . */
|
|
|
00db10 |
- lvx 4,12,6
|
|
|
00db10 |
- vperm 6,3,4,5
|
|
|
00db10 |
- addi 11,12,16
|
|
|
00db10 |
- addi 10,3,16
|
|
|
00db10 |
- stvx 6,0,3
|
|
|
00db10 |
+ cmpldi cr1,10,0
|
|
|
00db10 |
+ srdi 8,cnt,5 /* Setup the loop counter. */
|
|
|
00db10 |
+ mtocrf 0x01,9
|
|
|
00db10 |
+ cmpldi cr6,9,1
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ lvsr 5,0,src
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ lvsl 5,0,src
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ lvx 3,0,src
|
|
|
00db10 |
+ li 0,0
|
|
|
00db10 |
+ bf 31,L(setup_unaligned_loop)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ /* Copy another 16 bytes to align to 32-bytes due to the loop. */
|
|
|
00db10 |
+ lvx 4,src,6
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ addi src,src,16
|
|
|
00db10 |
+ stvx 6,0,dst
|
|
|
00db10 |
+ addi dst,dst,16
|
|
|
00db10 |
vor 3,4,4
|
|
|
00db10 |
+ clrrdi 0,src,60
|
|
|
00db10 |
|
|
|
00db10 |
L(setup_unaligned_loop):
|
|
|
00db10 |
- mtctr 8
|
|
|
00db10 |
- ble cr6,L(end_unaligned_loop)
|
|
|
00db10 |
+ mtctr 8
|
|
|
00db10 |
+ ble cr6,L(end_unaligned_loop)
|
|
|
00db10 |
|
|
|
00db10 |
/* Copy 32 bytes at a time using vector instructions. */
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(unaligned_loop):
|
|
|
00db10 |
|
|
|
00db10 |
/* Note: vr6/vr10 may contain data that was already copied,
|
|
|
00db10 |
@@ -444,63 +385,56 @@
|
|
|
00db10 |
some portions again. This is faster than having unaligned
|
|
|
00db10 |
vector instructions though. */
|
|
|
00db10 |
|
|
|
00db10 |
- lvx 4,11,6 /* vr4 = r11+16. */
|
|
|
00db10 |
- vperm 6,3,4,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr6. */
|
|
|
00db10 |
- lvx 3,11,7 /* vr3 = r11+32. */
|
|
|
00db10 |
- vperm 10,4,3,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr10. */
|
|
|
00db10 |
- addi 11,11,32
|
|
|
00db10 |
- stvx 6,0,10
|
|
|
00db10 |
- stvx 10,10,6
|
|
|
00db10 |
- addi 10,10,32
|
|
|
00db10 |
-
|
|
|
00db10 |
+ lvx 4,src,6
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ lvx 3,src,7
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 10,3,4,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 10,4,3,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ addi src,src,32
|
|
|
00db10 |
+ stvx 6,0,dst
|
|
|
00db10 |
+ stvx 10,dst,6
|
|
|
00db10 |
+ addi dst,dst,32
|
|
|
00db10 |
bdnz L(unaligned_loop)
|
|
|
00db10 |
|
|
|
00db10 |
- .align 4
|
|
|
00db10 |
+ clrrdi 0,src,60
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .align 4
|
|
|
00db10 |
L(end_unaligned_loop):
|
|
|
00db10 |
|
|
|
00db10 |
/* Check for tail bytes. */
|
|
|
00db10 |
- rldicr 0,31,0,59
|
|
|
00db10 |
- mtcrf 0x01,31
|
|
|
00db10 |
- beq cr1,0f
|
|
|
00db10 |
+ mtocrf 0x01,cnt
|
|
|
00db10 |
+ beqlr cr1
|
|
|
00db10 |
|
|
|
00db10 |
- add 3,3,0
|
|
|
00db10 |
- add 12,12,0
|
|
|
00db10 |
+ add src,src,0
|
|
|
00db10 |
|
|
|
00db10 |
/* We have 1~15 tail bytes to copy, and DST is quadword aligned. */
|
|
|
00db10 |
-8: /* Copy 8 bytes. */
|
|
|
00db10 |
+ /* Copy 8 bytes. */
|
|
|
00db10 |
bf 28,4f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- lwz 7,4(12)
|
|
|
00db10 |
- addi 12,12,8
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- stw 7,4(3)
|
|
|
00db10 |
- addi 3,3,8
|
|
|
00db10 |
-4: /* Copy 4 bytes. */
|
|
|
00db10 |
- bf 29,2f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lwz 6,0(12)
|
|
|
00db10 |
- addi 12,12,4
|
|
|
00db10 |
- stw 6,0(3)
|
|
|
00db10 |
- addi 3,3,4
|
|
|
00db10 |
-2: /* Copy 2~3 bytes. */
|
|
|
00db10 |
- bf 30,1f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lhz 6,0(12)
|
|
|
00db10 |
- addi 12,12,2
|
|
|
00db10 |
- sth 6,0(3)
|
|
|
00db10 |
- addi 3,3,2
|
|
|
00db10 |
-1: /* Copy 1 byte. */
|
|
|
00db10 |
- bf 31,0f
|
|
|
00db10 |
-
|
|
|
00db10 |
- lbz 6,0(12)
|
|
|
00db10 |
- stb 6,0(3)
|
|
|
00db10 |
-0: /* Return original DST pointer. */
|
|
|
00db10 |
- ld 31,-8(1)
|
|
|
00db10 |
- ld 3,-16(1)
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ lwz 7,4(src)
|
|
|
00db10 |
+ addi src,src,8
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ stw 7,4(dst)
|
|
|
00db10 |
+ addi dst,dst,8
|
|
|
00db10 |
+4: /* Copy 4~7 bytes. */
|
|
|
00db10 |
+ bf 29,L(tail2)
|
|
|
00db10 |
+ lwz 6,0(src)
|
|
|
00db10 |
+ stw 6,0(dst)
|
|
|
00db10 |
+ bf 30,L(tail5)
|
|
|
00db10 |
+ lhz 7,4(src)
|
|
|
00db10 |
+ sth 7,4(dst)
|
|
|
00db10 |
+ bflr 31
|
|
|
00db10 |
+ lbz 8,6(src)
|
|
|
00db10 |
+ stb 8,6(dst)
|
|
|
00db10 |
+ /* Return original DST pointer. */
|
|
|
00db10 |
blr
|
|
|
00db10 |
|
|
|
00db10 |
-END_GEN_TB (BP_SYM (memcpy),TB_TOCLESS)
|
|
|
00db10 |
+END_GEN_TB (memcpy,TB_TOCLESS)
|
|
|
00db10 |
libc_hidden_builtin_def (memcpy)
|
|
|
00db10 |
diff -urN glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/mempcpy.S glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/mempcpy.S
|
|
|
00db10 |
--- glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/mempcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/powerpc/powerpc64/power7/mempcpy.S 2014-05-29 13:04:56.000000000 -0500
|
|
|
00db10 |
@@ -367,13 +367,21 @@
|
|
|
00db10 |
mr 11,12
|
|
|
00db10 |
mtcrf 0x01,9
|
|
|
00db10 |
cmpldi cr6,9,1
|
|
|
00db10 |
- lvsl 5,0,12
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ lvsr 5,0,12
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ lvsl 5,0,12
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
lvx 3,0,12
|
|
|
00db10 |
bf 31,L(setup_unaligned_loop)
|
|
|
00db10 |
|
|
|
00db10 |
/* Copy another 16 bytes to align to 32-bytes due to the loop . */
|
|
|
00db10 |
lvx 4,12,6
|
|
|
00db10 |
- vperm 6,3,4,5
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 11,12,16
|
|
|
00db10 |
addi 10,3,16
|
|
|
00db10 |
stvx 6,0,3
|
|
|
00db10 |
@@ -393,11 +401,17 @@
|
|
|
00db10 |
vector instructions though. */
|
|
|
00db10 |
|
|
|
00db10 |
lvx 4,11,6 /* vr4 = r11+16. */
|
|
|
00db10 |
- vperm 6,3,4,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr6. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 6,4,3,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 6,3,4,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
lvx 3,11,7 /* vr3 = r11+32. */
|
|
|
00db10 |
- vperm 10,4,3,5 /* Merge the correctly-aligned portions
|
|
|
00db10 |
- of vr3/vr4 into vr10. */
|
|
|
00db10 |
+#ifdef __LITTLE_ENDIAN__
|
|
|
00db10 |
+ vperm 10,3,4,5
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+ vperm 10,4,3,5
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
addi 11,11,32
|
|
|
00db10 |
stvx 6,0,10
|
|
|
00db10 |
stvx 10,10,6
|