PowerPC LE strcpy

http://sourceware.org/ml/libc-alpha/2013-08/msg00100.html

The strcpy changes for little-endian are quite straight-forward, just
a matter of rotating the last word differently.

I'll note that the powerpc64 version of stpcpy is just begging to be
converted to use 64-bit loads and stores..

	* sysdeps/powerpc/powerpc64/strcpy.S: Add little-endian support:
	* sysdeps/powerpc/powerpc32/strcpy.S: Likewise.
	* sysdeps/powerpc/powerpc64/stpcpy.S: Likewise.
	* sysdeps/powerpc/powerpc32/stpcpy.S: Likewise.

5 files changed, 85 insertions, 3 deletions

diff --git a/ChangeLog b/ChangeLog
index 540cb94316..bbe5836409 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,12 @@
 2013-10-04  Alan Modra  <amodra@gmail.com>
 
+	* sysdeps/powerpc/powerpc64/strcpy.S: Add little-endian support:
+	* sysdeps/powerpc/powerpc32/strcpy.S: Likewise.
+	* sysdeps/powerpc/powerpc64/stpcpy.S: Likewise.
+	* sysdeps/powerpc/powerpc32/stpcpy.S: Likewise.
+
+2013-10-04  Alan Modra  <amodra@gmail.com>
+
 	* sysdeps/powerpc/powerpc64/strcmp.S (rTMP2): Define as r0.
 	(rTMP): Define as r11.
 	(strcmp): Add little-endian support.  Optimise tail.
diff --git a/sysdeps/powerpc/powerpc32/stpcpy.S b/sysdeps/powerpc/powerpc32/stpcpy.S
index 03c6dddc3a..7e106e0e6c 100644
--- a/sysdeps/powerpc/powerpc32/stpcpy.S
+++ b/sysdeps/powerpc/powerpc32/stpcpy.S
@@ -62,7 +62,22 @@ L(g2):	add	rTMP, rFEFE, rWORD
 
 	mr	rALT, rWORD
 /* We've hit the end of the string.  Do the rest byte-by-byte.  */
-L(g1):	rlwinm.	rTMP, rALT, 8, 24, 31
+L(g1):
+#ifdef __LITTLE_ENDIAN__
+	rlwinm.	rTMP, rALT, 0, 24, 31
+	stbu	rALT, 4(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 24, 24, 31
+	stbu	rTMP, 1(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 16, 24, 31
+	stbu	rTMP, 1(rDEST)
+	beqlr-
+	rlwinm	rTMP, rALT, 8, 24, 31
+	stbu	rTMP, 1(rDEST)
+	blr
+#else
+	rlwinm.	rTMP, rALT, 8, 24, 31
 	stbu	rTMP, 4(rDEST)
 	beqlr-
 	rlwinm.	rTMP, rALT, 16, 24, 31
@@ -73,6 +88,7 @@ L(g1):	rlwinm.	rTMP, rALT, 8, 24, 31
 	beqlr-
 	stbu	rALT, 1(rDEST)
 	blr
+#endif
 
 /* Oh well.  In this case, we just do a byte-by-byte copy.  */
 	.align 4
diff --git a/sysdeps/powerpc/powerpc32/strcpy.S b/sysdeps/powerpc/powerpc32/strcpy.S
index 4ae577dbb6..e938cc42a7 100644
--- a/sysdeps/powerpc/powerpc32/strcpy.S
+++ b/sysdeps/powerpc/powerpc32/strcpy.S
@@ -62,7 +62,22 @@ L(g2):	add	rTMP, rFEFE, rWORD
 
 	mr	rALT, rWORD
 /* We've hit the end of the string.  Do the rest byte-by-byte.  */
-L(g1):	rlwinm.	rTMP, rALT, 8, 24, 31
+L(g1):
+#ifdef __LITTLE_ENDIAN__
+	rlwinm.	rTMP, rALT, 0, 24, 31
+	stb	rALT, 4(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 24, 24, 31
+	stb	rTMP, 5(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 16, 24, 31
+	stb	rTMP, 6(rDEST)
+	beqlr-
+	rlwinm	rTMP, rALT, 8, 24, 31
+	stb	rTMP, 7(rDEST)
+	blr
+#else
+	rlwinm.	rTMP, rALT, 8, 24, 31
 	stb	rTMP, 4(rDEST)
 	beqlr-
 	rlwinm.	rTMP, rALT, 16, 24, 31
@@ -73,6 +88,7 @@ L(g1):	rlwinm.	rTMP, rALT, 8, 24, 31
 	beqlr-
 	stb	rALT, 7(rDEST)
 	blr
+#endif
 
 /* Oh well.  In this case, we just do a byte-by-byte copy.  */
 	.align 4
diff --git a/sysdeps/powerpc/powerpc64/stpcpy.S b/sysdeps/powerpc/powerpc64/stpcpy.S
index 070cd4662f..c0b39729e2 100644
--- a/sysdeps/powerpc/powerpc64/stpcpy.S
+++ b/sysdeps/powerpc/powerpc64/stpcpy.S
@@ -62,7 +62,22 @@ L(g2):	add	rTMP, rFEFE, rWORD
 
 	mr	rALT, rWORD
 /* We've hit the end of the string.  Do the rest byte-by-byte.  */
-L(g1):	rlwinm.	rTMP, rALT, 8, 24, 31
+L(g1):
+#ifdef __LITTLE_ENDIAN__
+	rlwinm.	rTMP, rALT, 0, 24, 31
+	stbu	rALT, 4(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 24, 24, 31
+	stbu	rTMP, 1(rDEST)
+	beqlr-
+	rlwinm.	rTMP, rALT, 16, 24, 31
+	stbu	rTMP, 1(rDEST)
+	beqlr-
+	rlwinm	rTMP, rALT, 8, 24, 31
+	stbu	rTMP, 1(rDEST)
+	blr
+#else
+	rlwinm.	rTMP, rALT, 8, 24, 31
 	stbu	rTMP, 4(rDEST)
 	beqlr-
 	rlwinm.	rTMP, rALT, 16, 24, 31
@@ -73,6 +88,7 @@ L(g1):	rlwinm.	rTMP, rALT, 8, 24, 31
 	beqlr-
 	stbu	rALT, 1(rDEST)
 	blr
+#endif
 
 /* Oh well.  In this case, we just do a byte-by-byte copy.  */
 	.align 4
diff --git a/sysdeps/powerpc/powerpc64/strcpy.S b/sysdeps/powerpc/powerpc64/strcpy.S
index 4c6fd3f9d7..a7fd85bad4 100644
--- a/sysdeps/powerpc/powerpc64/strcpy.S
+++ b/sysdeps/powerpc/powerpc64/strcpy.S
@@ -68,6 +68,32 @@ L(g2):	add	rTMP, rFEFE, rWORD
 	mr	rALT, rWORD
 /* We've hit the end of the string.  Do the rest byte-by-byte.  */
 L(g1):
+#ifdef __LITTLE_ENDIAN__
+	extrdi.	rTMP, rALT, 8, 56
+	stb	rALT, 8(rDEST)
+	beqlr-
+	extrdi.	rTMP, rALT, 8, 48
+	stb	rTMP, 9(rDEST)
+	beqlr-
+	extrdi.	rTMP, rALT, 8, 40
+	stb	rTMP, 10(rDEST)
+	beqlr-
+	extrdi.	rTMP, rALT, 8, 32
+	stb	rTMP, 11(rDEST)
+	beqlr-
+	extrdi.	rTMP, rALT, 8, 24
+	stb	rTMP, 12(rDEST)
+	beqlr-
+	extrdi.	rTMP, rALT, 8, 16
+	stb	rTMP, 13(rDEST)
+	beqlr-
+	extrdi.	rTMP, rALT, 8, 8
+	stb	rTMP, 14(rDEST)
+	beqlr-
+	extrdi	rTMP, rALT, 8, 0
+	stb	rTMP, 15(rDEST)
+	blr
+#else
 	extrdi.	rTMP, rALT, 8, 0
 	stb	rTMP, 8(rDEST)
 	beqlr-
@@ -91,6 +117,7 @@ L(g1):
 	beqlr-
 	stb	rALT, 15(rDEST)
 	blr
+#endif
 
 /* Oh well.  In this case, we just do a byte-by-byte copy.  */
 	.align 4