PowerPC LE memcpy

http://sourceware.org/ml/libc-alpha/2013-08/msg00103.html

LIttle-endian support for memcpy.  I spent some time cleaning up the
64-bit power7 memcpy, in order to avoid the extra alignment traps
power7 takes for little-endian.  It probably would have been better
to copy the linux kernel version of memcpy.

	* sysdeps/powerpc/powerpc32/power4/memcpy.S: Add little endian support.
	* sysdeps/powerpc/powerpc32/power6/memcpy.S: Likewise.
	* sysdeps/powerpc/powerpc32/power7/memcpy.S: Likewise.
	* sysdeps/powerpc/powerpc32/power7/mempcpy.S: Likewise.
	* sysdeps/powerpc/powerpc64/memcpy.S: Likewise.
	* sysdeps/powerpc/powerpc64/power4/memcpy.S: Likewise.
	* sysdeps/powerpc/powerpc64/power6/memcpy.S: Likewise.
	* sysdeps/powerpc/powerpc64/power7/memcpy.S: Likewise.
	* sysdeps/powerpc/powerpc64/power7/mempcpy.S: Likewise.  Make better
	use of regs.  Use power7 mtocrf.  Tidy function tails.

1 files changed, 329 insertions, 0 deletions

diff --git a/sysdeps/powerpc/powerpc64/power6/memcpy.S b/sysdeps/powerpc/powerpc64/power6/memcpy.S
index d6d242d..e3f3d8a 100644
--- a/sysdeps/powerpc/powerpc64/power6/memcpy.S
+++ b/sysdeps/powerpc/powerpc64/power6/memcpy.S
@@ -400,15 +400,28 @@ L(das_tail2):
     blt   cr6,5f
     srdi  7,6,16
     bgt	  cr6,3f
+#ifdef __LITTLE_ENDIAN__
+    sth   7,0(3)
+#else
     sth   6,0(3)
+#endif
     b     7f
     .align  4
 3:
+#ifdef __LITTLE_ENDIAN__
+    rotlwi 6,6,24
+    stb   6,0(3)
+    sth   7,1(3)
+#else
     stb   7,0(3)
     sth   6,1(3)
+#endif
     b     7f
     .align  4
 5:
+#ifdef __LITTLE_ENDIAN__
+    rotlwi 6,6,8
+#endif
     stb   6,0(3)
 7:
     cmpldi	cr1,10,16
@@ -595,13 +608,24 @@ L(du1_do):
     bf      30,L(du1_1dw)
 
     /* there are at least two DWs to copy */
+    /* FIXME: can combine last shift and "or" into "rldimi" */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 8
+    sldi     8,7, 64-8
+#else
     sldi     0,6, 8
     srdi     8,7, 64-8
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 8
+    sldi     8,6, 64-8
+#else
     sldi     0,7, 8
     srdi     8,6, 64-8
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -610,8 +634,13 @@ L(du1_do):
     blt     cr6,L(du1_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du1_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 8
+    sldi     8,7, 64-8
+#else
     sldi     0,6, 8
     srdi     8,7, 64-8
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -622,8 +651,13 @@ L(du1_do):
     b       L(du1_loop)
     .align 4
 L(du1_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 8
+    sldi     8,7, 64-8
+#else
     sldi     0,6, 8
     srdi     8,7, 64-8
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du1_loop)
@@ -635,23 +669,43 @@ L(du1_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du1_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 8
+    sldi   8,7, 64-8
+#else
     sldi   0,6, 8
     srdi   8,7, 64-8
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 8
+    sldi   8,6, 64-8
+#else
     sldi   0,7, 8
     srdi   8,6, 64-8
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 8
+    sldi   8,7, 64-8
+#else
     sldi   0,6, 8
     srdi   8,7, 64-8
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 8
+    sldi   8,6, 64-8
+#else
     sldi   0,7, 8
     srdi   8,6, 64-8
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -661,8 +715,13 @@ L(du1_loop):
     .align 4
 L(du1_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 8
+    sldi   8,7, 64-8
+#else
     sldi   0,6, 8
     srdi   8,7, 64-8
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)
@@ -672,13 +731,23 @@ L(du2_do):
     bf      30,L(du2_1dw)
 
     /* there are at least two DWs to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 16
+    sldi     8,7, 64-16
+#else
     sldi     0,6, 16
     srdi     8,7, 64-16
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 16
+    sldi     8,6, 64-16
+#else
     sldi     0,7, 16
     srdi     8,6, 64-16
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -687,8 +756,13 @@ L(du2_do):
     blt     cr6,L(du2_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du2_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 16
+    sldi     8,7, 64-16
+#else
     sldi     0,6, 16
     srdi     8,7, 64-16
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -699,8 +773,13 @@ L(du2_do):
     b       L(du2_loop)
     .align 4
 L(du2_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 16
+    sldi     8,7, 64-16
+#else
     sldi     0,6, 16
     srdi     8,7, 64-16
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du2_loop)
@@ -712,23 +791,43 @@ L(du2_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du2_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 16
+    sldi   8,7, 64-16
+#else
     sldi   0,6, 16
     srdi   8,7, 64-16
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 16
+    sldi   8,6, 64-16
+#else
     sldi   0,7, 16
     srdi   8,6, 64-16
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 16
+    sldi   8,7, 64-16
+#else
     sldi   0,6, 16
     srdi   8,7, 64-16
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 16
+    sldi   8,6, 64-16
+#else
     sldi   0,7, 16
     srdi   8,6, 64-16
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -738,8 +837,13 @@ L(du2_loop):
     .align 4
 L(du2_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 16
+    sldi   8,7, 64-16
+#else
     sldi   0,6, 16
     srdi   8,7, 64-16
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)
@@ -749,13 +853,23 @@ L(du3_do):
     bf      30,L(du3_1dw)
 
     /* there are at least two DWs to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 24
+    sldi     8,7, 64-24
+#else
     sldi     0,6, 24
     srdi     8,7, 64-24
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 24
+    sldi     8,6, 64-24
+#else
     sldi     0,7, 24
     srdi     8,6, 64-24
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -764,8 +878,13 @@ L(du3_do):
     blt     cr6,L(du3_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du3_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 24
+    sldi     8,7, 64-24
+#else
     sldi     0,6, 24
     srdi     8,7, 64-24
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -776,8 +895,13 @@ L(du3_do):
     b       L(du3_loop)
     .align 4
 L(du3_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 24
+    sldi     8,7, 64-24
+#else
     sldi     0,6, 24
     srdi     8,7, 64-24
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du3_loop)
@@ -789,23 +913,43 @@ L(du3_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du3_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 24
+    sldi   8,7, 64-24
+#else
     sldi   0,6, 24
     srdi   8,7, 64-24
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 24
+    sldi   8,6, 64-24
+#else
     sldi   0,7, 24
     srdi   8,6, 64-24
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 24
+    sldi   8,7, 64-24
+#else
     sldi   0,6, 24
     srdi   8,7, 64-24
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 24
+    sldi   8,6, 64-24
+#else
     sldi   0,7, 24
     srdi   8,6, 64-24
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -815,8 +959,13 @@ L(du3_loop):
     .align 4
 L(du3_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 24
+    sldi   8,7, 64-24
+#else
     sldi   0,6, 24
     srdi   8,7, 64-24
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)
@@ -832,13 +981,23 @@ L(du4_dox):
     bf      30,L(du4_1dw)
 
     /* there are at least two DWs to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 32
+    sldi     8,7, 64-32
+#else
     sldi     0,6, 32
     srdi     8,7, 64-32
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 32
+    sldi     8,6, 64-32
+#else
     sldi     0,7, 32
     srdi     8,6, 64-32
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -847,8 +1006,13 @@ L(du4_dox):
     blt     cr6,L(du4_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du4_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 32
+    sldi     8,7, 64-32
+#else
     sldi     0,6, 32
     srdi     8,7, 64-32
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -859,8 +1023,13 @@ L(du4_dox):
     b       L(du4_loop)
     .align 4
 L(du4_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 32
+    sldi     8,7, 64-32
+#else
     sldi     0,6, 32
     srdi     8,7, 64-32
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du4_loop)
@@ -872,23 +1041,43 @@ L(du4_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du4_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 32
+    sldi   8,7, 64-32
+#else
     sldi   0,6, 32
     srdi   8,7, 64-32
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 32
+    sldi   8,6, 64-32
+#else
     sldi   0,7, 32
     srdi   8,6, 64-32
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 32
+    sldi   8,7, 64-32
+#else
     sldi   0,6, 32
     srdi   8,7, 64-32
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 32
+    sldi   8,6, 64-32
+#else
     sldi   0,7, 32
     srdi   8,6, 64-32
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -898,8 +1087,13 @@ L(du4_loop):
     .align 4
 L(du4_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 32
+    sldi   8,7, 64-32
+#else
     sldi   0,6, 32
     srdi   8,7, 64-32
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)
@@ -909,13 +1103,23 @@ L(du5_do):
     bf      30,L(du5_1dw)
 
     /* there are at least two DWs to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 40
+    sldi     8,7, 64-40
+#else
     sldi     0,6, 40
     srdi     8,7, 64-40
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 40
+    sldi     8,6, 64-40
+#else
     sldi     0,7, 40
     srdi     8,6, 64-40
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -924,8 +1128,13 @@ L(du5_do):
     blt     cr6,L(du5_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du5_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 40
+    sldi     8,7, 64-40
+#else
     sldi     0,6, 40
     srdi     8,7, 64-40
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -936,8 +1145,13 @@ L(du5_do):
     b       L(du5_loop)
     .align 4
 L(du5_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 40
+    sldi     8,7, 64-40
+#else
     sldi     0,6, 40
     srdi     8,7, 64-40
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du5_loop)
@@ -949,23 +1163,43 @@ L(du5_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du5_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 40
+    sldi   8,7, 64-40
+#else
     sldi   0,6, 40
     srdi   8,7, 64-40
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 40
+    sldi   8,6, 64-40
+#else
     sldi   0,7, 40
     srdi   8,6, 64-40
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 40
+    sldi   8,7, 64-40
+#else
     sldi   0,6, 40
     srdi   8,7, 64-40
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 40
+    sldi   8,6, 64-40
+#else
     sldi   0,7, 40
     srdi   8,6, 64-40
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -975,8 +1209,13 @@ L(du5_loop):
     .align 4
 L(du5_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 40
+    sldi   8,7, 64-40
+#else
     sldi   0,6, 40
     srdi   8,7, 64-40
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)
@@ -986,13 +1225,23 @@ L(du6_do):
     bf      30,L(du6_1dw)
 
     /* there are at least two DWs to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 48
+    sldi     8,7, 64-48
+#else
     sldi     0,6, 48
     srdi     8,7, 64-48
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 48
+    sldi     8,6, 64-48
+#else
     sldi     0,7, 48
     srdi     8,6, 64-48
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -1001,8 +1250,13 @@ L(du6_do):
     blt     cr6,L(du6_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du6_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 48
+    sldi     8,7, 64-48
+#else
     sldi     0,6, 48
     srdi     8,7, 64-48
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -1013,8 +1267,13 @@ L(du6_do):
     b       L(du6_loop)
     .align 4
 L(du6_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 48
+    sldi     8,7, 64-48
+#else
     sldi     0,6, 48
     srdi     8,7, 64-48
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du6_loop)
@@ -1026,23 +1285,43 @@ L(du6_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du6_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 48
+    sldi   8,7, 64-48
+#else
     sldi   0,6, 48
     srdi   8,7, 64-48
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 48
+    sldi   8,6, 64-48
+#else
     sldi   0,7, 48
     srdi   8,6, 64-48
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 48
+    sldi   8,7, 64-48
+#else
     sldi   0,6, 48
     srdi   8,7, 64-48
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 48
+    sldi   8,6, 64-48
+#else
     sldi   0,7, 48
     srdi   8,6, 64-48
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -1052,8 +1331,13 @@ L(du6_loop):
     .align 4
 L(du6_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 48
+    sldi   8,7, 64-48
+#else
     sldi   0,6, 48
     srdi   8,7, 64-48
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)
@@ -1063,13 +1347,23 @@ L(du7_do):
     bf      30,L(du7_1dw)
 
     /* there are at least two DWs to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 56
+    sldi     8,7, 64-56
+#else
     sldi     0,6, 56
     srdi     8,7, 64-56
+#endif
     or      0,0,8
     ld      6,16(5)
     std     0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,7, 56
+    sldi     8,6, 64-56
+#else
     sldi     0,7, 56
     srdi     8,6, 64-56
+#endif
     or      0,0,8
     ld      7,24(5)
     std     0,8(4)
@@ -1078,8 +1372,13 @@ L(du7_do):
     blt     cr6,L(du7_fini)  /* if total DWs = 3, then bypass loop */
     bf      31,L(du7_loop)
     /* there is a third DW to copy */
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 56
+    sldi     8,7, 64-56
+#else
     sldi     0,6, 56
     srdi     8,7, 64-56
+#endif
     or      0,0,8
     std     0,0(4)
     mr      6,7
@@ -1090,8 +1389,13 @@ L(du7_do):
     b       L(du7_loop)
     .align 4
 L(du7_1dw):
+#ifdef __LITTLE_ENDIAN__
+    srdi     0,6, 56
+    sldi     8,7, 64-56
+#else
     sldi     0,6, 56
     srdi     8,7, 64-56
+#endif
     addi    5,5,16
     or      0,0,8
     bf      31,L(du7_loop)
@@ -1103,23 +1407,43 @@ L(du7_1dw):
     .align 4
 /* copy 32 bytes at a time */
 L(du7_loop):
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 56
+    sldi   8,7, 64-56
+#else
     sldi   0,6, 56
     srdi   8,7, 64-56
+#endif
     or    0,0,8
     ld    6,0(5)
     std   0,0(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 56
+    sldi   8,6, 64-56
+#else
     sldi   0,7, 56
     srdi   8,6, 64-56
+#endif
     or    0,0,8
     ld    7,8(5)
     std   0,8(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 56
+    sldi   8,7, 64-56
+#else
     sldi   0,6, 56
     srdi   8,7, 64-56
+#endif
     or    0,0,8
     ld    6,16(5)
     std   0,16(4)
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,7, 56
+    sldi   8,6, 64-56
+#else
     sldi   0,7, 56
     srdi   8,6, 64-56
+#endif
     or    0,0,8
     ld    7,24(5)
     std   0,24(4)
@@ -1129,8 +1453,13 @@ L(du7_loop):
     .align 4
 L(du7_fini):
     /* calculate and store the final DW */
+#ifdef __LITTLE_ENDIAN__
+    srdi   0,6, 56
+    sldi   8,7, 64-56
+#else
     sldi   0,6, 56
     srdi   8,7, 64-56
+#endif
     or    0,0,8
     std   0,0(4)
     b     L(du_done)