Merge changes between r23946 and r24305 from /fsf/trunk.

git-svn-id: svn://svn.eglibc.org/trunk@24306 7b3dc134-2b1b-0410-93df-9e9f96275f8d

14 files changed, 1725 insertions, 1283 deletions

diff --git a/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S b/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S
index d0071c765..ebec0e0ba 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_finite.S
@@ -39,10 +39,8 @@ EALIGN (__finite, 4, 0)
 
 	stfd    fp1,-16(r1)   /* Transfer FP to GPR's.  */
 	ori	2,2,0	      /* Force a new dispatch group.  */
-
-	lhz     r4,-16(r1)    /* Fetch the upper portion of the high word of
-			      the FP value (where the exponent and sign bits
-			      are).  */
+	lhz     r4,-16+HISHORT(r1)  /* Fetch the upper 16 bits of the FP value
+				    (biased exponent and sign bit).  */
 	clrlwi  r4,r4,17      /* r4 = abs(r4).  */
 	cmpwi   cr7,r4,0x7ff0 /* r4 == 0x7ff0?  */
 	bltlr   cr7	      /* LT means finite, other non-finite.  */
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S b/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S
index 1aea12383..8d088db5a 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/fpu/s_isinf.S
@@ -38,9 +38,8 @@ EALIGN (__isinf, 4, 0)
 
 	stfd    fp1,-16(r1)   /* Transfer FP to GPR's.  */
 	ori	2,2,0	      /* Force a new dispatch group.  */
-	lhz	r4,-16(r1)    /* Fetch the upper portion of the high word of
-			      the FP value (where the exponent and sign bits
-			      are).  */
+	lhz	r4,-16+HISHORT(r1)  /* Fetch the upper 16 bits of the FP value
+				    (biased exponent and sign bit).  */
 	cmpwi	cr7,r4,0x7ff0 /* r4 == 0x7ff0?  */
 	li	r3,1
 	beqlr   cr7	      /* EQ means INF, otherwise -INF.  */
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/memchr.S b/libc/sysdeps/powerpc/powerpc64/power7/memchr.S
index 3416897f5..5076dd0c1 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/memchr.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/memchr.S
@@ -25,109 +25,112 @@ ENTRY (__memchr)
 	CALL_MCOUNT 2
 	dcbt	0,r3
 	clrrdi  r8,r3,3
-	rlwimi	r4,r4,8,16,23
-	rlwimi	r4,r4,16,0,15
+	insrdi	r4,r4,8,48
 	add	r7,r3,r5      /* Calculate the last acceptable address.  */
+	insrdi	r4,r4,16,32
 	cmpldi	r5,32
+	li	r9, -1
+	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
 	insrdi  r4,r4,32,0
+	addi	r7,r7,-1
+#ifdef __LITTLE_ENDIAN__
+	sld	r9,r9,r6
+#else
+	srd	r9,r9,r6
+#endif
 	ble	L(small_range)
 
-	cmpld	cr7,r3,r7     /* Compare the starting address (r3) with the
-				 ending address (r7).  If (r3 >= r7),
-				 the size passed in was zero or negative.  */
-	ble	cr7,L(proceed)
-
-	li	r7,-1         /* Artificially set our ending address (r7)
-				 such that we will exit early.  */
-
-L(proceed):
-	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
-	cmpldi	cr6,r6,0      /* cr6 == Do we have padding?  */
 	ld	r12,0(r8)     /* Load doubleword from memory.  */
-	cmpb	r10,r12,r4    /* Check for BYTEs in DWORD1.  */
-	beq	cr6,L(proceed_no_padding)
-	sld	r10,r10,r6
-	srd	r10,r10,r6
-L(proceed_no_padding):
-	cmpldi	cr7,r10,0     /* Does r10 indicate we got a hit?  */
+	cmpb	r3,r12,r4     /* Check for BYTEs in DWORD1.  */
+	and	r3,r3,r9
+	clrldi	r5,r7,61      /* Byte count - 1 in last dword.  */
+	clrrdi	r7,r7,3       /* Address of last doubleword.  */
+	cmpldi	cr7,r3,0      /* Does r3 indicate we got a hit?  */
 	bne	cr7,L(done)
 
-	/* See if we are at the last acceptable address yet.  */
-	addi	r9,r8,8
-	cmpld	cr6,r9,r7
-	bge	cr6,L(null)
-
 	mtcrf   0x01,r8
 	/* Are we now aligned to a quadword boundary?  If so, skip to
 	   the main loop.  Otherwise, go through the alignment code.  */
-
 	bt	28,L(loop_setup)
 
 	/* Handle DWORD2 of pair.  */
 	ldu	r12,8(r8)
-	cmpb	r10,r12,r4
-	cmpldi	cr7,r10,0
+	cmpb	r3,r12,r4
+	cmpldi	cr7,r3,0
 	bne	cr7,L(done)
 
-	/* Are we done already?  */
-	addi	r9,r8,8
-	cmpld	cr6,r9,r7
-	bge	cr6,L(null)
-
 L(loop_setup):
-	sub	r5,r7,r9
-	srdi	r6,r5,4	      /* Number of loop iterations.  */
+	/* The last dword we want to read in the loop below is the one
+	   containing the last byte of the string, ie. the dword at
+	   (s + size - 1) & ~7, or r7.  The first dword read is at
+	   r8 + 8, we read 2 * cnt dwords, so the last dword read will
+	   be at r8 + 8 + 16 * cnt - 8.  Solving for cnt gives
+	   cnt = (r7 - r8) / 16  */
+	sub	r6,r7,r8
+	srdi	r6,r6,4	      /* Number of loop iterations.  */
 	mtctr	r6            /* Setup the counter.  */
-	b	L(loop)
-	/* Main loop to look for BYTE backwards in the string.  Since
-	   it's a small loop (< 8 instructions), align it to 32-bytes.  */
-	.p2align  5
+
+	/* Main loop to look for BYTE in the string.  Since
+	   it's a small loop (8 instructions), align it to 32-bytes.  */
+	.align	5
 L(loop):
 	/* Load two doublewords, compare and merge in a
 	   single register for speed.  This is an attempt
 	   to speed up the byte-checking process for bigger strings.  */
 	ld	r12,8(r8)
 	ldu	r11,16(r8)
-	cmpb	r10,r12,r4
+	cmpb	r3,r12,r4
 	cmpb	r9,r11,r4
-	or	r5,r9,r10     /* Merge everything in one doubleword.  */
-	cmpldi	cr7,r5,0
+	or	r6,r9,r3      /* Merge everything in one doubleword.  */
+	cmpldi	cr7,r6,0
 	bne	cr7,L(found)
 	bdnz	L(loop)
 
-	/* We're here because the counter reached 0, and that means we
-	   didn't have any matches for BYTE in the whole range.  */
-	subi	r11,r7,8
-	cmpld	cr6,r8,r11
-	blt	cr6,L(loop_small)
-	b	L(null)
+	/* We may have one more dword to read.  */
+	cmpld	r8,r7
+	beqlr
 
+	ldu	r12,8(r8)
+	cmpb	r3,r12,r4
+	cmpldi	cr6,r3,0
+	bne	cr6,L(done)
+	blr
+
+	.align	4
+L(found):
 	/* OK, one (or both) of the doublewords contains BYTE.  Check
 	   the first doubleword and decrement the address in case the first
 	   doubleword really contains BYTE.  */
-	.align	4
-L(found):
-	cmpldi	cr6,r10,0
+	cmpldi	cr6,r3,0
 	addi	r8,r8,-8
 	bne	cr6,L(done)
 
 	/* BYTE must be in the second doubleword.  Adjust the address
-	   again and move the result of cmpb to r10 so we can calculate the
+	   again and move the result of cmpb to r3 so we can calculate the
 	   pointer.  */
 
-	mr	r10,r9
+	mr	r3,r9
 	addi	r8,r8,8
 
-	/* r10 has the output of the cmpb instruction, that is, it contains
+	/* r3 has the output of the cmpb instruction, that is, it contains
 	   0xff in the same position as BYTE in the original
 	   doubleword from the string.  Use that to calculate the pointer.
 	   We need to make sure BYTE is *before* the end of the range.  */
 L(done):
-	cntlzd	r0,r10	      /* Count leading zeroes before the match.  */
-	srdi	r0,r0,3	      /* Convert leading zeroes to bytes.  */
+#ifdef __LITTLE_ENDIAN__
+	addi    r0,r3,-1
+	andc    r0,r0,r3
+	popcntd	r0,r0	      /* Count trailing zeros.  */
+#else
+	cntlzd	r0,r3	      /* Count leading zeros before the match.  */
+#endif
+	cmpld	r8,r7         /* Are we on the last dword?  */
+	srdi	r0,r0,3	      /* Convert leading/trailing zeros to bytes.  */
 	add	r3,r8,r0
-	cmpld	r3,r7
-	bge	L(null)
+	cmpld	cr7,r0,r5     /* If on the last dword, check byte offset.  */
+	bnelr
+	blelr	cr7
+	li	r3,0
 	blr
 
 	.align	4
@@ -139,67 +142,44 @@ L(null):
 	.align	4
 L(small_range):
 	cmpldi	r5,0
-	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
-	beq	L(null)       /* This branch is for the cmpldi r5,0 above.  */
+	beq	L(null)
 	ld	r12,0(r8)     /* Load word from memory.  */
-	cmpldi	cr6,r6,0      /* cr6 == Do we have padding?  */
-	cmpb	r10,r12,r4    /* Check for BYTE in DWORD1.  */
-			      /* If no padding, skip the shifts.  */
-	beq	cr6,L(small_no_padding)
-	sld	r10,r10,r6
-	srd	r10,r10,r6
-L(small_no_padding):
-	cmpldi	cr7,r10,0
+	cmpb	r3,r12,r4     /* Check for BYTE in DWORD1.  */
+	and	r3,r3,r9
+	cmpldi	cr7,r3,0
+	clrldi	r5,r7,61      /* Byte count - 1 in last dword.  */
+	clrrdi	r7,r7,3       /* Address of last doubleword.  */
+	cmpld	r8,r7         /* Are we done already?  */
 	bne	cr7,L(done)
+	beqlr
 
-	/* Are we done already?  */
-	addi    r9,r8,8
-	cmpld	r9,r7
-	bge	L(null)
-	/* If we're not done, drop through into loop_small.  */
-
-L(loop_small):                /* loop_small has been unrolled.  */
 	ldu	r12,8(r8)
-	cmpb	r10,r12,r4
-	addi	r9,r8,8
-	cmpldi	cr6,r10,0
-	cmpld	r9,r7
+	cmpb	r3,r12,r4
+	cmpldi	cr6,r3,0
+	cmpld	r8,r7
 	bne	cr6,L(done)   /* Found something.  */
-	bge	L(null)       /* Hit end of string (length).  */
+	beqlr		      /* Hit end of string (length).  */
 
 	ldu	r12,8(r8)
-	cmpb	r10,r12,r4
-	addi	r9,r8,8
-	cmpldi	cr6,r10,0
-	cmpld	r9,r7
-	bne	cr6,L(done)   /* Found something.  */
-	bge	L(null)
+	cmpb	r3,r12,r4
+	cmpldi	cr6,r3,0
+	cmpld	r8,r7
+	bne	cr6,L(done)
+	beqlr
 
 	ldu	r12,8(r8)
-	subi	r11,r7,8
-	cmpb	r10,r12,r4
-	cmpldi	cr6,r10,0
-	ori	r2,r2,0       /* Force a dispatch group.  */
+	cmpb	r3,r12,r4
+	cmpldi	cr6,r3,0
+	cmpld	r8,r7
 	bne	cr6,L(done)
+	beqlr
 
-	cmpld	r8,r11        /* At end of range?  */
-	bge	L(null)
-
-	/* For most cases we will never get here.  Under some combinations of
-	   padding + length there is a leftover double that still needs to be
-	   checked.  */
 	ldu	r12,8(r8)
-	cmpb	r10,r12,r4
-	addi	r9,r8,8
-	cmpldi	cr6,r10,0
-	cmpld	r9,r7
-	bne	cr6,L(done)   /* Found something.  */
-
-	/* Save a branch and exit directly.  */
-	li	r3,0
+	cmpb	r3,r12,r4
+	cmpldi	cr6,r3,0
+	bne	cr6,L(done)
 	blr
 
-
 END (__memchr)
 weak_alias (__memchr, memchr)
 libc_hidden_builtin_def (memchr)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/memcmp.S b/libc/sysdeps/powerpc/powerpc64/power7/memcmp.S
index f190c6461..6851cdc75 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/memcmp.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/memcmp.S
@@ -23,10 +23,9 @@
 		    size_t size [r5])  */
 
 	.machine power7
-EALIGN (memcmp,4,0)
+EALIGN (memcmp, 4, 0)
 	CALL_MCOUNT 3
 
-#define rTMP	r0
 #define rRTN	r3
 #define rSTR1	r3	/* first string arg */
 #define rSTR2	r4	/* second string arg */
@@ -37,354 +36,557 @@ EALIGN (memcmp,4,0)
 #define rWORD4	r9	/* next word in s2 */
 #define rWORD5	r10	/* next word in s1 */
 #define rWORD6	r11	/* next word in s2 */
-#define rBITDIF	r12	/* bits that differ in s1 & s2 words */
 #define rWORD7	r30	/* next word in s1 */
 #define rWORD8	r31	/* next word in s2 */
 
-	xor	rTMP,rSTR2,rSTR1
-	cmpldi	cr6,rN,0
-	cmpldi	cr1,rN,12
-	clrldi.	rTMP,rTMP,61
-	clrldi	rBITDIF,rSTR1,61
-	cmpldi	cr5,rBITDIF,0
-	beq-	cr6,L(zeroLength)
-	dcbt	0,rSTR1
-	dcbt	0,rSTR2
+	xor	r0, rSTR2, rSTR1
+	cmpldi	cr6, rN, 0
+	cmpldi	cr1, rN, 12
+	clrldi.	r0, r0, 61
+	clrldi	r12, rSTR1, 61
+	cmpldi	cr5, r12, 0
+	beq-	cr6, L(zeroLength)
+	dcbt	0, rSTR1
+	dcbt	0, rSTR2
 /* If less than 8 bytes or not aligned, use the unaligned
    byte loop.  */
-	blt	cr1,L(bytealigned)
-	std	rWORD8,-8(r1)
-	cfi_offset(rWORD8,-8)
-	std	rWORD7,-16(r1)
-	cfi_offset(rWORD7,-16)
+	blt	cr1, L(bytealigned)
+	std	rWORD8, -8(r1)
+	cfi_offset(rWORD8, -8)
+	std	rWORD7, -16(r1)
+	cfi_offset(rWORD7, -16)
 	bne	L(unaligned)
 /* At this point we know both strings have the same alignment and the
-   compare length is at least 8 bytes.  rBITDIF contains the low order
+   compare length is at least 8 bytes.  r12 contains the low order
    3 bits of rSTR1 and cr5 contains the result of the logical compare
-   of rBITDIF to 0.  If rBITDIF == 0 then we are already double word
-   aligned and can perform the DWaligned loop.
+   of r12 to 0.  If r12 == 0 then we are already double word
+   aligned and can perform the DW aligned loop.
 
    Otherwise we know the two strings have the same alignment (but not
-   yet DW).  So we can force the string addresses to the next lower DW
-   boundary and special case this first DW word using shift left to
+   yet DW).  So we force the string addresses to the next lower DW
+   boundary and special case this first DW using shift left to
    eliminate bits preceding the first byte.  Since we want to join the
-   normal (DWaligned) compare loop, starting at the second double word,
+   normal (DW aligned) compare loop, starting at the second double word,
    we need to adjust the length (rN) and special case the loop
-   versioning for the first DW. This insures that the loop count is
-   correct and the first DW (shifted) is in the expected resister pair.  */
+   versioning for the first DW. This ensures that the loop count is
+   correct and the first DW (shifted) is in the expected register pair.  */
 	.align	4
 L(samealignment):
-	clrrdi	rSTR1,rSTR1,3
-	clrrdi	rSTR2,rSTR2,3
-	beq	cr5,L(DWaligned)
-	add	rN,rN,rBITDIF
-	sldi	r11,rBITDIF,3
-	srdi	rTMP,rN,5	/* Divide by 32 */
-	andi.	rBITDIF,rN,24	/* Get the DW remainder */
-	ld	rWORD1,0(rSTR1)
-	ld	rWORD2,0(rSTR2)
-	cmpldi	cr1,rBITDIF,16
-	cmpldi	cr7,rN,32
-	clrldi	rN,rN,61
+	clrrdi	rSTR1, rSTR1, 3
+	clrrdi	rSTR2, rSTR2, 3
+	beq	cr5, L(DWaligned)
+	add	rN, rN, r12
+	sldi	rWORD6, r12, 3
+	srdi	r0, rN, 5	/* Divide by 32 */
+	andi.	r12, rN, 24	/* Get the DW remainder */
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 0(rSTR1)
+	ld	rWORD2, 0(rSTR2)
+#endif
+	cmpldi	cr1, r12, 16
+	cmpldi	cr7, rN, 32
+	clrldi	rN, rN, 61
 	beq	L(dPs4)
-	mtctr	rTMP
-	bgt	cr1,L(dPs3)
-	beq	cr1,L(dPs2)
+	mtctr	r0
+	bgt	cr1, L(dPs3)
+	beq	cr1, L(dPs2)
 
 /* Remainder is 8 */
 	.align	3
 L(dsP1):
-	sld	rWORD5,rWORD1,r11
-	sld	rWORD6,rWORD2,r11
-	cmpld	cr5,rWORD5,rWORD6
-	blt	cr7,L(dP1x)
+	sld	rWORD5, rWORD1, rWORD6
+	sld	rWORD6, rWORD2, rWORD6
+	cmpld	cr5, rWORD5, rWORD6
+	blt	cr7, L(dP1x)
 /* Do something useful in this cycle since we have to branch anyway.  */
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD2,8(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 8(rSTR1)
+	ld	rWORD2, 8(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
 	b	L(dP1e)
 /* Remainder is 16 */
 	.align	4
 L(dPs2):
-	sld	rWORD5,rWORD1,r11
-	sld	rWORD6,rWORD2,r11
-	cmpld	cr6,rWORD5,rWORD6
-	blt	cr7,L(dP2x)
+	sld	rWORD5, rWORD1, rWORD6
+	sld	rWORD6, rWORD2, rWORD6
+	cmpld	cr6, rWORD5, rWORD6
+	blt	cr7, L(dP2x)
 /* Do something useful in this cycle since we have to branch anyway.  */
-	ld	rWORD7,8(rSTR1)
-	ld	rWORD8,8(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD7, 8(rSTR1)
+	ld	rWORD8, 8(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
 	b	L(dP2e)
 /* Remainder is 24 */
 	.align	4
 L(dPs3):
-	sld	rWORD3,rWORD1,r11
-	sld	rWORD4,rWORD2,r11
-	cmpld	cr1,rWORD3,rWORD4
+	sld	rWORD3, rWORD1, rWORD6
+	sld	rWORD4, rWORD2, rWORD6
+	cmpld	cr1, rWORD3, rWORD4
 	b	L(dP3e)
 /* Count is a multiple of 32, remainder is 0 */
 	.align	4
 L(dPs4):
-	mtctr	rTMP
-	sld	rWORD1,rWORD1,r11
-	sld	rWORD2,rWORD2,r11
-	cmpld	cr0,rWORD1,rWORD2
+	mtctr	r0
+	sld	rWORD1, rWORD1, rWORD6
+	sld	rWORD2, rWORD2, rWORD6
+	cmpld	cr7, rWORD1, rWORD2
 	b	L(dP4e)
 
 /* At this point we know both strings are double word aligned and the
    compare length is at least 8 bytes.  */
 	.align	4
 L(DWaligned):
-	andi.	rBITDIF,rN,24	/* Get the DW remainder */
-	srdi	rTMP,rN,5	/* Divide by 32 */
-	cmpldi	cr1,rBITDIF,16
-	cmpldi	cr7,rN,32
-	clrldi	rN,rN,61
+	andi.	r12, rN, 24	/* Get the DW remainder */
+	srdi	r0, rN, 5	/* Divide by 32 */
+	cmpldi	cr1, r12, 16
+	cmpldi	cr7, rN, 32
+	clrldi	rN, rN, 61
 	beq	L(dP4)
-	bgt	cr1,L(dP3)
-	beq	cr1,L(dP2)
+	bgt	cr1, L(dP3)
+	beq	cr1, L(dP2)
 
 /* Remainder is 8 */
 	.align	4
 L(dP1):
-	mtctr	rTMP
+	mtctr	r0
 /* Normally we'd use rWORD7/rWORD8 here, but since we might exit early
    (8-15 byte compare), we want to use only volatile registers.  This
    means we can avoid restoring non-volatile registers since we did not
    change any on the early exit path.  The key here is the non-early
    exit path only cares about the condition code (cr5), not about which
    register pair was used.  */
-	ld	rWORD5,0(rSTR1)
-	ld	rWORD6,0(rSTR2)
-	cmpld	cr5,rWORD5,rWORD6
-	blt	cr7,L(dP1x)
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD2,8(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 0(rSTR1)
+	ld	rWORD6, 0(rSTR2)
+#endif
+	cmpld	cr5, rWORD5, rWORD6
+	blt	cr7, L(dP1x)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 8(rSTR1)
+	ld	rWORD2, 8(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
 L(dP1e):
-	ld	rWORD3,16(rSTR1)
-	ld	rWORD4,16(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	ld	rWORD5,24(rSTR1)
-	ld	rWORD6,24(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr5,L(dLcr5)
-	bne	cr0,L(dLcr0)
-
-	ldu	rWORD7,32(rSTR1)
-	ldu	rWORD8,32(rSTR2)
-	bne	cr1,L(dLcr1)
-	cmpld	cr5,rWORD7,rWORD8
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 16(rSTR1)
+	ld	rWORD4, 16(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 24(rSTR1)
+	ld	rWORD6, 24(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr5, L(dLcr5x)
+	bne	cr7, L(dLcr7x)
+
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ldu	rWORD7, 32(rSTR1)
+	ldu	rWORD8, 32(rSTR2)
+#endif
+	bne	cr1, L(dLcr1)
+	cmpld	cr5, rWORD7, rWORD8
 	bdnz	L(dLoop)
-	bne	cr6,L(dLcr6)
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
+	bne	cr6, L(dLcr6)
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
 	.align	3
 L(dP1x):
-	sldi.	r12,rN,3
-	bne	cr5,L(dLcr5)
-	subfic	rN,r12,64	/* Shift count is 64 - (rN * 8).  */
+	sldi.	r12, rN, 3
+	bne	cr5, L(dLcr5x)
+	subfic	rN, r12, 64	/* Shift count is 64 - (rN * 8).  */
 	bne	L(d00)
-	li	rRTN,0
+	li	rRTN, 0
 	blr
 
 /* Remainder is 16 */
 	.align	4
 L(dP2):
-	mtctr	rTMP
-	ld	rWORD5,0(rSTR1)
-	ld	rWORD6,0(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	blt	cr7,L(dP2x)
-	ld	rWORD7,8(rSTR1)
-	ld	rWORD8,8(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
+	mtctr	r0
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 0(rSTR1)
+	ld	rWORD6, 0(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	blt	cr7, L(dP2x)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD7, 8(rSTR1)
+	ld	rWORD8, 8(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
 L(dP2e):
-	ld	rWORD1,16(rSTR1)
-	ld	rWORD2,16(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
-	ld	rWORD3,24(rSTR1)
-	ld	rWORD4,24(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	addi	rSTR1,rSTR1,8
-	addi	rSTR2,rSTR2,8
-	bne	cr6,L(dLcr6)
-	bne	cr5,L(dLcr5)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 16(rSTR1)
+	ld	rWORD2, 16(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 24(rSTR1)
+	ld	rWORD4, 24(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#endif
+	bne	cr6, L(dLcr6)
+	bne	cr5, L(dLcr5)
 	b	L(dLoop2)
 /* Again we are on a early exit path (16-23 byte compare), we want to
    only use volatile registers and avoid restoring non-volatile
    registers.  */
 	.align	4
 L(dP2x):
-	ld	rWORD3,8(rSTR1)
-	ld	rWORD4,8(rSTR2)
-	cmpld	cr5,rWORD3,rWORD4
-	sldi.	r12,rN,3
-	bne	cr6,L(dLcr6)
-	addi	rSTR1,rSTR1,8
-	addi	rSTR2,rSTR2,8
-	bne	cr5,L(dLcr5)
-	subfic	rN,r12,64	/* Shift count is 64 - (rN * 8).  */
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 8(rSTR1)
+	ld	rWORD4, 8(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	sldi.	r12, rN, 3
+	bne	cr6, L(dLcr6x)
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#endif
+	bne	cr1, L(dLcr1x)
+	subfic	rN, r12, 64	/* Shift count is 64 - (rN * 8).  */
 	bne	L(d00)
-	li	rRTN,0
+	li	rRTN, 0
 	blr
 
 /* Remainder is 24 */
 	.align	4
 L(dP3):
-	mtctr	rTMP
-	ld	rWORD3,0(rSTR1)
-	ld	rWORD4,0(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
+	mtctr	r0
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 0(rSTR1)
+	ld	rWORD4, 0(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
 L(dP3e):
-	ld	rWORD5,8(rSTR1)
-	ld	rWORD6,8(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	blt	cr7,L(dP3x)
-	ld	rWORD7,16(rSTR1)
-	ld	rWORD8,16(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	ld	rWORD1,24(rSTR1)
-	ld	rWORD2,24(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
-	addi	rSTR1,rSTR1,16
-	addi	rSTR2,rSTR2,16
-	bne	cr1,L(dLcr1)
-	bne	cr6,L(dLcr6)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 8(rSTR1)
+	ld	rWORD6, 8(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	blt	cr7, L(dP3x)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD7, 16(rSTR1)
+	ld	rWORD8, 16(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 24(rSTR1)
+	ld	rWORD2, 24(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 16
+	addi	rSTR2, rSTR2, 16
+#endif
+	bne	cr1, L(dLcr1)
+	bne	cr6, L(dLcr6)
 	b	L(dLoop1)
 /* Again we are on a early exit path (24-31 byte compare), we want to
    only use volatile registers and avoid restoring non-volatile
    registers.  */
 	.align	4
 L(dP3x):
-	ld	rWORD1,16(rSTR1)
-	ld	rWORD2,16(rSTR2)
-	cmpld	cr5,rWORD1,rWORD2
-	sldi.	r12,rN,3
-	bne	cr1,L(dLcr1)
-	addi	rSTR1,rSTR1,16
-	addi	rSTR2,rSTR2,16
-	bne	cr6,L(dLcr6)
-	subfic	rN,r12,64	/* Shift count is 64 - (rN * 8).  */
-	bne	cr5,L(dLcr5)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 16(rSTR1)
+	ld	rWORD2, 16(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+	sldi.	r12, rN, 3
+	bne	cr1, L(dLcr1x)
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 16
+	addi	rSTR2, rSTR2, 16
+#endif
+	bne	cr6, L(dLcr6x)
+	subfic	rN, r12, 64	/* Shift count is 64 - (rN * 8).  */
+	bne	cr7, L(dLcr7x)
 	bne	L(d00)
-	li	rRTN,0
+	li	rRTN, 0
 	blr
 
 /* Count is a multiple of 32, remainder is 0 */
 	.align	4
 L(dP4):
-	mtctr	rTMP
-	ld	rWORD1,0(rSTR1)
-	ld	rWORD2,0(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
+	mtctr	r0
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 0(rSTR1)
+	ld	rWORD2, 0(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
 L(dP4e):
-	ld	rWORD3,8(rSTR1)
-	ld	rWORD4,8(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	ld	rWORD5,16(rSTR1)
-	ld	rWORD6,16(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	ldu	rWORD7,24(rSTR1)
-	ldu	rWORD8,24(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	bne	cr0,L(dLcr0)
-	bne	cr1,L(dLcr1)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 8(rSTR1)
+	ld	rWORD4, 8(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 16(rSTR1)
+	ld	rWORD6, 16(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ldu	rWORD7, 24(rSTR1)
+	ldu	rWORD8, 24(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	bne	cr7, L(dLcr7)
+	bne	cr1, L(dLcr1)
 	bdz-	L(d24)		/* Adjust CTR as we start with +4 */
 /* This is the primary loop */
 	.align	4
 L(dLoop):
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD2,8(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	bne	cr6,L(dLcr6)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 8(rSTR1)
+	ld	rWORD2, 8(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	bne	cr6, L(dLcr6)
 L(dLoop1):
-	ld	rWORD3,16(rSTR1)
-	ld	rWORD4,16(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr5,L(dLcr5)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 16(rSTR1)
+	ld	rWORD4, 16(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr5, L(dLcr5)
 L(dLoop2):
-	ld	rWORD5,24(rSTR1)
-	ld	rWORD6,24(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	bne	cr0,L(dLcr0)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 24(rSTR1)
+	ld	rWORD6, 24(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	bne	cr7, L(dLcr7)
 L(dLoop3):
-	ldu	rWORD7,32(rSTR1)
-	ldu	rWORD8,32(rSTR2)
-	bne	cr1,L(dLcr1)
-	cmpld	cr0,rWORD1,rWORD2
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ldu	rWORD7, 32(rSTR1)
+	ldu	rWORD8, 32(rSTR2)
+#endif
+	bne	cr1, L(dLcr1)
+	cmpld	cr7, rWORD1, rWORD2
 	bdnz	L(dLoop)
 
 L(dL4):
-	cmpld	cr1,rWORD3,rWORD4
-	bne	cr6,L(dLcr6)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr5,L(dLcr5)
-	cmpld	cr5,rWORD7,rWORD8
+	cmpld	cr1, rWORD3, rWORD4
+	bne	cr6, L(dLcr6)
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr5, L(dLcr5)
+	cmpld	cr5, rWORD7, rWORD8
 L(d44):
-	bne	cr0,L(dLcr0)
+	bne	cr7, L(dLcr7)
 L(d34):
-	bne	cr1,L(dLcr1)
+	bne	cr1, L(dLcr1)
 L(d24):
-	bne	cr6,L(dLcr6)
+	bne	cr6, L(dLcr6)
 L(d14):
-	sldi.	r12,rN,3
-	bne	cr5,L(dLcr5)
+	sldi.	r12, rN, 3
+	bne	cr5, L(dLcr5)
 L(d04):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	subfic	rN,r12,64	/* Shift count is 64 - (rN * 8).  */
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+	subfic	rN, r12, 64	/* Shift count is 64 - (rN * 8).  */
 	beq	L(zeroLength)
 /* At this point we have a remainder of 1 to 7 bytes to compare.  Since
    we are aligned it is safe to load the whole double word, and use
    shift right double to eliminate bits beyond the compare length.  */
 L(d00):
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD2,8(rSTR2)
-	srd	rWORD1,rWORD1,rN
-	srd	rWORD2,rWORD2,rN
-	cmpld	cr5,rWORD1,rWORD2
-	bne	cr5,L(dLcr5x)
-	li	rRTN,0
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 8(rSTR1)
+	ld	rWORD2, 8(rSTR2)
+#endif
+	srd	rWORD1, rWORD1, rN
+	srd	rWORD2, rWORD2, rN
+	cmpld	cr7, rWORD1, rWORD2
+	bne	cr7, L(dLcr7x)
+	li	rRTN, 0
 	blr
+
 	.align	4
-L(dLcr0):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
-	bgtlr	cr0
-	li	rRTN,-1
+L(dLcr7):
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+L(dLcr7x):
+	li	rRTN, 1
+	bgtlr	cr7
+	li	rRTN, -1
 	blr
 	.align	4
 L(dLcr1):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+L(dLcr1x):
+	li	rRTN, 1
 	bgtlr	cr1
-	li	rRTN,-1
+	li	rRTN, -1
 	blr
 	.align	4
 L(dLcr6):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+L(dLcr6x):
+	li	rRTN, 1
 	bgtlr	cr6
-	li	rRTN,-1
+	li	rRTN, -1
 	blr
 	.align	4
 L(dLcr5):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
 L(dLcr5x):
-	li	rRTN,1
+	li	rRTN, 1
 	bgtlr	cr5
-	li	rRTN,-1
+	li	rRTN, -1
 	blr
 
 	.align	4
 L(bytealigned):
 	mtctr	rN
-	beq	cr6,L(zeroLength)
+#if 0
+/* Huh?  We've already branched on cr6!  */
+	beq	cr6, L(zeroLength)
+#endif
 
 /* We need to prime this loop.  This loop is swing modulo scheduled
    to avoid pipe delays.  The dependent instruction latencies (load to
@@ -396,38 +598,38 @@ L(bytealigned):
    So we must precondition some registers and condition codes so that
    we don't exit the loop early on the first iteration.  */
 
-	lbz	rWORD1,0(rSTR1)
-	lbz	rWORD2,0(rSTR2)
+	lbz	rWORD1, 0(rSTR1)
+	lbz	rWORD2, 0(rSTR2)
 	bdz	L(b11)
-	cmpld	cr0,rWORD1,rWORD2
-	lbz	rWORD3,1(rSTR1)
-	lbz	rWORD4,1(rSTR2)
+	cmpld	cr7, rWORD1, rWORD2
+	lbz	rWORD3, 1(rSTR1)
+	lbz	rWORD4, 1(rSTR2)
 	bdz	L(b12)
-	cmpld	cr1,rWORD3,rWORD4
-	lbzu	rWORD5,2(rSTR1)
-	lbzu	rWORD6,2(rSTR2)
+	cmpld	cr1, rWORD3, rWORD4
+	lbzu	rWORD5, 2(rSTR1)
+	lbzu	rWORD6, 2(rSTR2)
 	bdz	L(b13)
 	.align	4
 L(bLoop):
-	lbzu	rWORD1,1(rSTR1)
-	lbzu	rWORD2,1(rSTR2)
-	bne	cr0,L(bLcr0)
+	lbzu	rWORD1, 1(rSTR1)
+	lbzu	rWORD2, 1(rSTR2)
+	bne	cr7, L(bLcr7)
 
-	cmpld	cr6,rWORD5,rWORD6
+	cmpld	cr6, rWORD5, rWORD6
 	bdz	L(b3i)
 
-	lbzu	rWORD3,1(rSTR1)
-	lbzu	rWORD4,1(rSTR2)
-	bne	cr1,L(bLcr1)
+	lbzu	rWORD3, 1(rSTR1)
+	lbzu	rWORD4, 1(rSTR2)
+	bne	cr1, L(bLcr1)
 
-	cmpld	cr0,rWORD1,rWORD2
+	cmpld	cr7, rWORD1, rWORD2
 	bdz	L(b2i)
 
-	lbzu	rWORD5,1(rSTR1)
-	lbzu	rWORD6,1(rSTR2)
-	bne	cr6,L(bLcr6)
+	lbzu	rWORD5, 1(rSTR1)
+	lbzu	rWORD6, 1(rSTR2)
+	bne	cr6, L(bLcr6)
 
-	cmpld	cr1,rWORD3,rWORD4
+	cmpld	cr1, rWORD3, rWORD4
 	bdnz	L(bLoop)
 
 /* We speculatively loading bytes before we have tested the previous
@@ -437,542 +639,727 @@ L(bLoop):
    tested.  In this case we must complete the pending operations
    before returning.  */
 L(b1i):
-	bne	cr0,L(bLcr0)
-	bne	cr1,L(bLcr1)
+	bne	cr7, L(bLcr7)
+	bne	cr1, L(bLcr1)
 	b	L(bx56)
 	.align	4
 L(b2i):
-	bne	cr6,L(bLcr6)
-	bne	cr0,L(bLcr0)
+	bne	cr6, L(bLcr6)
+	bne	cr7, L(bLcr7)
 	b	L(bx34)
 	.align	4
 L(b3i):
-	bne	cr1,L(bLcr1)
-	bne	cr6,L(bLcr6)
+	bne	cr1, L(bLcr1)
+	bne	cr6, L(bLcr6)
 	b	L(bx12)
 	.align	4
-L(bLcr0):
-	li	rRTN,1
-	bgtlr	cr0
-	li	rRTN,-1
+L(bLcr7):
+	li	rRTN, 1
+	bgtlr	cr7
+	li	rRTN, -1
 	blr
 L(bLcr1):
-	li	rRTN,1
+	li	rRTN, 1
 	bgtlr	cr1
-	li	rRTN,-1
+	li	rRTN, -1
 	blr
 L(bLcr6):
-	li	rRTN,1
+	li	rRTN, 1
 	bgtlr	cr6
-	li	rRTN,-1
+	li	rRTN, -1
 	blr
 
 L(b13):
-	bne	cr0,L(bx12)
-	bne	cr1,L(bx34)
+	bne	cr7, L(bx12)
+	bne	cr1, L(bx34)
 L(bx56):
-	sub	rRTN,rWORD5,rWORD6
+	sub	rRTN, rWORD5, rWORD6
 	blr
 	nop
 L(b12):
-	bne	cr0,L(bx12)
+	bne	cr7, L(bx12)
 L(bx34):
-	sub	rRTN,rWORD3,rWORD4
+	sub	rRTN, rWORD3, rWORD4
 	blr
 L(b11):
 L(bx12):
-	sub	rRTN,rWORD1,rWORD2
+	sub	rRTN, rWORD1, rWORD2
 	blr
 	.align	4
-L(zeroLengthReturn):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
 L(zeroLength):
-	li	rRTN,0
+	li	rRTN, 0
 	blr
 
 	.align	4
 /* At this point we know the strings have different alignment and the
-   compare length is at least 8 bytes.  rBITDIF contains the low order
+   compare length is at least 8 bytes.  r12 contains the low order
    3 bits of rSTR1 and cr5 contains the result of the logical compare
-   of rBITDIF to 0.  If rBITDIF == 0 then rStr1 is double word
+   of r12 to 0.  If r12 == 0 then rStr1 is double word
    aligned and can perform the DWunaligned loop.
 
    Otherwise we know that rSTR1 is not already DW aligned yet.
    So we can force the string addresses to the next lower DW
-   boundary and special case this first DW word using shift left to
+   boundary and special case this first DW using shift left to
    eliminate bits preceding the first byte.  Since we want to join the
    normal (DWaligned) compare loop, starting at the second double word,
    we need to adjust the length (rN) and special case the loop
-   versioning for the first DW. This insures that the loop count is
+   versioning for the first DW. This ensures that the loop count is
    correct and the first DW (shifted) is in the expected resister pair.  */
-#define rSHL	r29	/* Unaligned shift left count.  */
-#define rSHR	r28	/* Unaligned shift right count.  */
-#define rB		r27	/* Left rotation temp for rWORD2.  */
-#define rD		r26	/* Left rotation temp for rWORD4.  */
-#define rF		r25	/* Left rotation temp for rWORD6.  */
-#define rH		r24	/* Left rotation temp for rWORD8.  */
-#define rA		r0	/* Right rotation temp for rWORD2.  */
-#define rC		r12	/* Right rotation temp for rWORD4.  */
-#define rE		r0	/* Right rotation temp for rWORD6.  */
-#define rG		r12	/* Right rotation temp for rWORD8.  */
+#define rSHL		r29	/* Unaligned shift left count.  */
+#define rSHR		r28	/* Unaligned shift right count.  */
+#define rWORD8_SHIFT	r27	/* Left rotation temp for rWORD2.  */
+#define rWORD2_SHIFT	r26	/* Left rotation temp for rWORD4.  */
+#define rWORD4_SHIFT	r25	/* Left rotation temp for rWORD6.  */
+#define rWORD6_SHIFT	r24	/* Left rotation temp for rWORD8.  */
 L(unaligned):
-	std	r29,-24(r1)
-	cfi_offset(r29,-24)
-	clrldi	rSHL,rSTR2,61
-	beq	cr6,L(duzeroLength)
-	std	r28,-32(r1)
-	cfi_offset(r28,-32)
-	beq	cr5,L(DWunaligned)
-	std	r27,-40(r1)
-	cfi_offset(r27,-40)
-/* Adjust the logical start of rSTR2 ro compensate for the extra bits
+	std	rSHL, -24(r1)
+	cfi_offset(rSHL, -24)
+	clrldi	rSHL, rSTR2, 61
+	beq	cr6, L(duzeroLength)
+	std	rSHR, -32(r1)
+	cfi_offset(rSHR, -32)
+	beq	cr5, L(DWunaligned)
+	std	rWORD8_SHIFT, -40(r1)
+	cfi_offset(rWORD8_SHIFT, -40)
+/* Adjust the logical start of rSTR2 to compensate for the extra bits
    in the 1st rSTR1 DW.  */
-	sub	r27,rSTR2,rBITDIF
+	sub	rWORD8_SHIFT, rSTR2, r12
 /* But do not attempt to address the DW before that DW that contains
    the actual start of rSTR2.  */
-	clrrdi	rSTR2,rSTR2,3
-	std	r26,-48(r1)
-	cfi_offset(r26,-48)
+	clrrdi	rSTR2, rSTR2, 3
+	std	rWORD2_SHIFT, -48(r1)
+	cfi_offset(rWORD2_SHIFT, -48)
 /* Compute the left/right shift counts for the unaligned rSTR2,
    compensating for the logical (DW aligned) start of rSTR1.  */
-	clrldi	rSHL,r27,61
-	clrrdi	rSTR1,rSTR1,3
-	std	r25,-56(r1)
-	cfi_offset(r25,-56)
-	sldi	rSHL,rSHL,3
-	cmpld	cr5,r27,rSTR2
-	add	rN,rN,rBITDIF
-	sldi	r11,rBITDIF,3
-	std	r24,-64(r1)
-	cfi_offset(r24,-64)
-	subfic	rSHR,rSHL,64
-	srdi	rTMP,rN,5	/* Divide by 32 */
-	andi.	rBITDIF,rN,24	/* Get the DW remainder */
+	clrldi	rSHL, rWORD8_SHIFT, 61
+	clrrdi	rSTR1, rSTR1, 3
+	std	rWORD4_SHIFT, -56(r1)
+	cfi_offset(rWORD4_SHIFT, -56)
+	sldi	rSHL, rSHL, 3
+	cmpld	cr5, rWORD8_SHIFT, rSTR2
+	add	rN, rN, r12
+	sldi	rWORD6, r12, 3
+	std	rWORD6_SHIFT, -64(r1)
+	cfi_offset(rWORD6_SHIFT, -64)
+	subfic	rSHR, rSHL, 64
+	srdi	r0, rN, 5	/* Divide by 32 */
+	andi.	r12, rN, 24	/* Get the DW remainder */
 /* We normally need to load 2 DWs to start the unaligned rSTR2, but in
    this special case those bits may be discarded anyway.  Also we
    must avoid loading a DW where none of the bits are part of rSTR2 as
    this may cross a page boundary and cause a page fault.  */
-	li	rWORD8,0
-	blt	cr5,L(dus0)
-	ld	rWORD8,0(rSTR2)
-	la	rSTR2,8(rSTR2)
-	sld	rWORD8,rWORD8,rSHL
+	li	rWORD8, 0
+	blt	cr5, L(dus0)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD8, 0(rSTR2)
+	addi	rSTR2, rSTR2, 8
+#endif
+	sld	rWORD8, rWORD8, rSHL
 
 L(dus0):
-	ld	rWORD1,0(rSTR1)
-	ld	rWORD2,0(rSTR2)
-	cmpldi	cr1,rBITDIF,16
-	cmpldi	cr7,rN,32
-	srd	rG,rWORD2,rSHR
-	clrldi	rN,rN,61
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 0(rSTR1)
+	ld	rWORD2, 0(rSTR2)
+#endif
+	cmpldi	cr1, r12, 16
+	cmpldi	cr7, rN, 32
+	srd	r12, rWORD2, rSHR
+	clrldi	rN, rN, 61
 	beq	L(duPs4)
-	mtctr	rTMP
-	or	rWORD8,rG,rWORD8
-	bgt	cr1,L(duPs3)
-	beq	cr1,L(duPs2)
+	mtctr	r0
+	or	rWORD8, r12, rWORD8
+	bgt	cr1, L(duPs3)
+	beq	cr1, L(duPs2)
 
 /* Remainder is 8 */
 	.align	4
 L(dusP1):
-	sld	rB,rWORD2,rSHL
-	sld	rWORD7,rWORD1,r11
-	sld	rWORD8,rWORD8,r11
-	bge	cr7,L(duP1e)
+	sld	rWORD8_SHIFT, rWORD2, rSHL
+	sld	rWORD7, rWORD1, rWORD6
+	sld	rWORD8, rWORD8, rWORD6
+	bge	cr7, L(duP1e)
 /* At this point we exit early with the first double word compare
    complete and remainder of 0 to 7 bytes.  See L(du14) for details on
    how we handle the remaining bytes.  */
-	cmpld	cr5,rWORD7,rWORD8
-	sldi.	rN,rN,3
-	bne	cr5,L(duLcr5)
-	cmpld	cr7,rN,rSHR
+	cmpld	cr5, rWORD7, rWORD8
+	sldi.	rN, rN, 3
+	bne	cr5, L(duLcr5)
+	cmpld	cr7, rN, rSHR
 	beq	L(duZeroReturn)
-	li	rA,0
-	ble	cr7,L(dutrim)
-	ld	rWORD2,8(rSTR2)
-	srd	rA,rWORD2,rSHR
+	li	r0, 0
+	ble	cr7, L(dutrim)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD2, 8(rSTR2)
+#endif
+	srd	r0, rWORD2, rSHR
 	b	L(dutrim)
 /* Remainder is 16 */
 	.align	4
 L(duPs2):
-	sld	rH,rWORD2,rSHL
-	sld	rWORD5,rWORD1,r11
-	sld	rWORD6,rWORD8,r11
+	sld	rWORD6_SHIFT, rWORD2, rSHL
+	sld	rWORD5, rWORD1, rWORD6
+	sld	rWORD6, rWORD8, rWORD6
 	b	L(duP2e)
 /* Remainder is 24 */
 	.align	4
 L(duPs3):
-	sld	rF,rWORD2,rSHL
-	sld	rWORD3,rWORD1,r11
-	sld	rWORD4,rWORD8,r11
+	sld	rWORD4_SHIFT, rWORD2, rSHL
+	sld	rWORD3, rWORD1, rWORD6
+	sld	rWORD4, rWORD8, rWORD6
 	b	L(duP3e)
 /* Count is a multiple of 32, remainder is 0 */
 	.align	4
 L(duPs4):
-	mtctr	rTMP
-	or	rWORD8,rG,rWORD8
-	sld	rD,rWORD2,rSHL
-	sld	rWORD1,rWORD1,r11
-	sld	rWORD2,rWORD8,r11
+	mtctr	r0
+	or	rWORD8, r12, rWORD8
+	sld	rWORD2_SHIFT, rWORD2, rSHL
+	sld	rWORD1, rWORD1, rWORD6
+	sld	rWORD2, rWORD8, rWORD6
 	b	L(duP4e)
 
 /* At this point we know rSTR1 is double word aligned and the
    compare length is at least 8 bytes.  */
 	.align	4
 L(DWunaligned):
-	std	r27,-40(r1)
-	cfi_offset(r27,-40)
-	clrrdi	rSTR2,rSTR2,3
-	std	r26,-48(r1)
-	cfi_offset(r26,-48)
-	srdi	rTMP,rN,5	/* Divide by 32 */
-	std	r25,-56(r1)
-	cfi_offset(r25,-56)
-	andi.	rBITDIF,rN,24	/* Get the DW remainder */
-	std	r24,-64(r1)
-	cfi_offset(r24,-64)
-	sldi	rSHL,rSHL,3
-	ld	rWORD6,0(rSTR2)
-	ldu	rWORD8,8(rSTR2)
-	cmpldi	cr1,rBITDIF,16
-	cmpldi	cr7,rN,32
-	clrldi	rN,rN,61
-	subfic	rSHR,rSHL,64
-	sld	rH,rWORD6,rSHL
+	std	rWORD8_SHIFT, -40(r1)
+	cfi_offset(rWORD8_SHIFT, -40)
+	clrrdi	rSTR2, rSTR2, 3
+	std	rWORD2_SHIFT, -48(r1)
+	cfi_offset(rWORD2_SHIFT, -48)
+	srdi	r0, rN, 5	/* Divide by 32 */
+	std	rWORD4_SHIFT, -56(r1)
+	cfi_offset(rWORD4_SHIFT, -56)
+	andi.	r12, rN, 24	/* Get the DW remainder */
+	std	rWORD6_SHIFT, -64(r1)
+	cfi_offset(rWORD6_SHIFT, -64)
+	sldi	rSHL, rSHL, 3
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD6, 0(rSTR2)
+	ldu	rWORD8, 8(rSTR2)
+#endif
+	cmpldi	cr1, r12, 16
+	cmpldi	cr7, rN, 32
+	clrldi	rN, rN, 61
+	subfic	rSHR, rSHL, 64
+	sld	rWORD6_SHIFT, rWORD6, rSHL
 	beq	L(duP4)
-	mtctr	rTMP
-	bgt	cr1,L(duP3)
-	beq	cr1,L(duP2)
+	mtctr	r0
+	bgt	cr1, L(duP3)
+	beq	cr1, L(duP2)
 
 /* Remainder is 8 */
 	.align	4
 L(duP1):
-	srd	rG,rWORD8,rSHR
-	ld	rWORD7,0(rSTR1)
-	sld	rB,rWORD8,rSHL
-	or	rWORD8,rG,rH
-	blt	cr7,L(duP1x)
+	srd	r12, rWORD8, rSHR
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	addi	rSTR1, rSTR1, 8
+#else
+	ld	rWORD7, 0(rSTR1)
+#endif
+	sld	rWORD8_SHIFT, rWORD8, rSHL
+	or	rWORD8, r12, rWORD6_SHIFT
+	blt	cr7, L(duP1x)
 L(duP1e):
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD2,8(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	srd	rA,rWORD2,rSHR
-	sld	rD,rWORD2,rSHL
-	or	rWORD2,rA,rB
-	ld	rWORD3,16(rSTR1)
-	ld	rWORD4,16(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
-	srd	rC,rWORD4,rSHR
-	sld	rF,rWORD4,rSHL
-	bne	cr5,L(duLcr5)
-	or	rWORD4,rC,rD
-	ld	rWORD5,24(rSTR1)
-	ld	rWORD6,24(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	srd	rE,rWORD6,rSHR
-	sld	rH,rWORD6,rSHL
-	bne	cr0,L(duLcr0)
-	or	rWORD6,rE,rF
-	cmpld	cr6,rWORD5,rWORD6
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 8(rSTR1)
+	ld	rWORD2, 8(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	srd	r0, rWORD2, rSHR
+	sld	rWORD2_SHIFT, rWORD2, rSHL
+	or	rWORD2, r0, rWORD8_SHIFT
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 16(rSTR1)
+	ld	rWORD4, 16(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+	srd	r12, rWORD4, rSHR
+	sld	rWORD4_SHIFT, rWORD4, rSHL
+	bne	cr5, L(duLcr5)
+	or	rWORD4, r12, rWORD2_SHIFT
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 24(rSTR1)
+	ld	rWORD6, 24(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	srd	r0, rWORD6, rSHR
+	sld	rWORD6_SHIFT, rWORD6, rSHL
+	bne	cr7, L(duLcr7)
+	or	rWORD6, r0, rWORD4_SHIFT
+	cmpld	cr6, rWORD5, rWORD6
 	b	L(duLoop3)
 	.align	4
 /* At this point we exit early with the first double word compare
    complete and remainder of 0 to 7 bytes.  See L(du14) for details on
    how we handle the remaining bytes.  */
 L(duP1x):
-	cmpld	cr5,rWORD7,rWORD8
-	sldi.	rN,rN,3
-	bne	cr5,L(duLcr5)
-	cmpld	cr7,rN,rSHR
+	cmpld	cr5, rWORD7, rWORD8
+	sldi.	rN, rN, 3
+	bne	cr5, L(duLcr5)
+	cmpld	cr7, rN, rSHR
 	beq	L(duZeroReturn)
-	li	rA,0
-	ble	cr7,L(dutrim)
-	ld	rWORD2,8(rSTR2)
-	srd	rA,rWORD2,rSHR
+	li	r0, 0
+	ble	cr7, L(dutrim)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD2, 8(rSTR2)
+#endif
+	srd	r0, rWORD2, rSHR
 	b	L(dutrim)
 /* Remainder is 16 */
 	.align	4
 L(duP2):
-	srd	rE,rWORD8,rSHR
-	ld	rWORD5,0(rSTR1)
-	or	rWORD6,rE,rH
-	sld	rH,rWORD8,rSHL
+	srd	r0, rWORD8, rSHR
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	addi	rSTR1, rSTR1, 8
+#else
+	ld	rWORD5, 0(rSTR1)
+#endif
+	or	rWORD6, r0, rWORD6_SHIFT
+	sld	rWORD6_SHIFT, rWORD8, rSHL
 L(duP2e):
-	ld	rWORD7,8(rSTR1)
-	ld	rWORD8,8(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	srd	rG,rWORD8,rSHR
-	sld	rB,rWORD8,rSHL
-	or	rWORD8,rG,rH
-	blt	cr7,L(duP2x)
-	ld	rWORD1,16(rSTR1)
-	ld	rWORD2,16(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	bne	cr6,L(duLcr6)
-	srd	rA,rWORD2,rSHR
-	sld	rD,rWORD2,rSHL
-	or	rWORD2,rA,rB
-	ld	rWORD3,24(rSTR1)
-	ld	rWORD4,24(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
-	bne	cr5,L(duLcr5)
-	srd	rC,rWORD4,rSHR
-	sld	rF,rWORD4,rSHL
-	or	rWORD4,rC,rD
-	addi	rSTR1,rSTR1,8
-	addi	rSTR2,rSTR2,8
-	cmpld	cr1,rWORD3,rWORD4
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD7, 8(rSTR1)
+	ld	rWORD8, 8(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	srd	r12, rWORD8, rSHR
+	sld	rWORD8_SHIFT, rWORD8, rSHL
+	or	rWORD8, r12, rWORD6_SHIFT
+	blt	cr7, L(duP2x)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 16(rSTR1)
+	ld	rWORD2, 16(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	bne	cr6, L(duLcr6)
+	srd	r0, rWORD2, rSHR
+	sld	rWORD2_SHIFT, rWORD2, rSHL
+	or	rWORD2, r0, rWORD8_SHIFT
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 24(rSTR1)
+	ld	rWORD4, 24(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+	bne	cr5, L(duLcr5)
+	srd	r12, rWORD4, rSHR
+	sld	rWORD4_SHIFT, rWORD4, rSHL
+	or	rWORD4, r12, rWORD2_SHIFT
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#endif
+	cmpld	cr1, rWORD3, rWORD4
 	b	L(duLoop2)
 	.align	4
 L(duP2x):
-	cmpld	cr5,rWORD7,rWORD8
-	addi	rSTR1,rSTR1,8
-	addi	rSTR2,rSTR2,8
-	bne	cr6,L(duLcr6)
-	sldi.	rN,rN,3
-	bne	cr5,L(duLcr5)
-	cmpld	cr7,rN,rSHR
+	cmpld	cr5, rWORD7, rWORD8
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#endif
+	bne	cr6, L(duLcr6)
+	sldi.	rN, rN, 3
+	bne	cr5, L(duLcr5)
+	cmpld	cr7, rN, rSHR
 	beq	L(duZeroReturn)
-	li	rA,0
-	ble	cr7,L(dutrim)
-	ld	rWORD2,8(rSTR2)
-	srd	rA,rWORD2,rSHR
+	li	r0, 0
+	ble	cr7, L(dutrim)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD2, 8(rSTR2)
+#endif
+	srd	r0, rWORD2, rSHR
 	b	L(dutrim)
 
 /* Remainder is 24 */
 	.align	4
 L(duP3):
-	srd	rC,rWORD8,rSHR
-	ld	rWORD3,0(rSTR1)
-	sld	rF,rWORD8,rSHL
-	or	rWORD4,rC,rH
+	srd	r12, rWORD8, rSHR
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	addi	rSTR1, rSTR1, 8
+#else
+	ld	rWORD3, 0(rSTR1)
+#endif
+	sld	rWORD4_SHIFT, rWORD8, rSHL
+	or	rWORD4, r12, rWORD6_SHIFT
 L(duP3e):
-	ld	rWORD5,8(rSTR1)
-	ld	rWORD6,8(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	srd	rE,rWORD6,rSHR
-	sld	rH,rWORD6,rSHL
-	or	rWORD6,rE,rF
-	ld	rWORD7,16(rSTR1)
-	ld	rWORD8,16(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr1,L(duLcr1)
-	srd	rG,rWORD8,rSHR
-	sld	rB,rWORD8,rSHL
-	or	rWORD8,rG,rH
-	blt	cr7,L(duP3x)
-	ld	rWORD1,24(rSTR1)
-	ld	rWORD2,24(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	bne	cr6,L(duLcr6)
-	srd	rA,rWORD2,rSHR
-	sld	rD,rWORD2,rSHL
-	or	rWORD2,rA,rB
-	addi	rSTR1,rSTR1,16
-	addi	rSTR2,rSTR2,16
-	cmpld	cr0,rWORD1,rWORD2
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 8(rSTR1)
+	ld	rWORD6, 8(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	srd	r0, rWORD6, rSHR
+	sld	rWORD6_SHIFT, rWORD6, rSHL
+	or	rWORD6, r0, rWORD4_SHIFT
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD7, 16(rSTR1)
+	ld	rWORD8, 16(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr1, L(duLcr1)
+	srd	r12, rWORD8, rSHR
+	sld	rWORD8_SHIFT, rWORD8, rSHL
+	or	rWORD8, r12, rWORD6_SHIFT
+	blt	cr7, L(duP3x)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 24(rSTR1)
+	ld	rWORD2, 24(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	bne	cr6, L(duLcr6)
+	srd	r0, rWORD2, rSHR
+	sld	rWORD2_SHIFT, rWORD2, rSHL
+	or	rWORD2, r0, rWORD8_SHIFT
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 16
+	addi	rSTR2, rSTR2, 16
+#endif
+	cmpld	cr7, rWORD1, rWORD2
 	b	L(duLoop1)
 	.align	4
 L(duP3x):
-	addi	rSTR1,rSTR1,16
-	addi	rSTR2,rSTR2,16
-	bne	cr1,L(duLcr1)
-	cmpld	cr5,rWORD7,rWORD8
-	bne	cr6,L(duLcr6)
-	sldi.	rN,rN,3
-	bne	cr5,L(duLcr5)
-	cmpld	cr7,rN,rSHR
+#ifndef __LITTLE_ENDIAN__
+	addi	rSTR1, rSTR1, 16
+	addi	rSTR2, rSTR2, 16
+#endif
+#if 0
+/* Huh?  We've already branched on cr1!  */
+	bne	cr1, L(duLcr1)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	bne	cr6, L(duLcr6)
+	sldi.	rN, rN, 3
+	bne	cr5, L(duLcr5)
+	cmpld	cr7, rN, rSHR
 	beq	L(duZeroReturn)
-	li	rA,0
-	ble	cr7,L(dutrim)
-	ld	rWORD2,8(rSTR2)
-	srd	rA,rWORD2,rSHR
+	li	r0, 0
+	ble	cr7, L(dutrim)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD2, 8(rSTR2)
+#endif
+	srd	r0, rWORD2, rSHR
 	b	L(dutrim)
 
 /* Count is a multiple of 32, remainder is 0 */
 	.align	4
 L(duP4):
-	mtctr	rTMP
-	srd	rA,rWORD8,rSHR
-	ld	rWORD1,0(rSTR1)
-	sld	rD,rWORD8,rSHL
-	or	rWORD2,rA,rH
+	mtctr	r0
+	srd	r0, rWORD8, rSHR
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	addi	rSTR1, rSTR1, 8
+#else
+	ld	rWORD1, 0(rSTR1)
+#endif
+	sld	rWORD2_SHIFT, rWORD8, rSHL
+	or	rWORD2, r0, rWORD6_SHIFT
 L(duP4e):
-	ld	rWORD3,8(rSTR1)
-	ld	rWORD4,8(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
-	srd	rC,rWORD4,rSHR
-	sld	rF,rWORD4,rSHL
-	or	rWORD4,rC,rD
-	ld	rWORD5,16(rSTR1)
-	ld	rWORD6,16(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	bne	cr0,L(duLcr0)
-	srd	rE,rWORD6,rSHR
-	sld	rH,rWORD6,rSHL
-	or	rWORD6,rE,rF
-	ldu	rWORD7,24(rSTR1)
-	ldu	rWORD8,24(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr1,L(duLcr1)
-	srd	rG,rWORD8,rSHR
-	sld	rB,rWORD8,rSHL
-	or	rWORD8,rG,rH
-	cmpld	cr5,rWORD7,rWORD8
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 8(rSTR1)
+	ld	rWORD4, 8(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+	srd	r12, rWORD4, rSHR
+	sld	rWORD4_SHIFT, rWORD4, rSHL
+	or	rWORD4, r12, rWORD2_SHIFT
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 16(rSTR1)
+	ld	rWORD6, 16(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	bne	cr7, L(duLcr7)
+	srd	r0, rWORD6, rSHR
+	sld	rWORD6_SHIFT, rWORD6, rSHL
+	or	rWORD6, r0, rWORD4_SHIFT
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ldu	rWORD7, 24(rSTR1)
+	ldu	rWORD8, 24(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr1, L(duLcr1)
+	srd	r12, rWORD8, rSHR
+	sld	rWORD8_SHIFT, rWORD8, rSHL
+	or	rWORD8, r12, rWORD6_SHIFT
+	cmpld	cr5, rWORD7, rWORD8
 	bdz	L(du24)		/* Adjust CTR as we start with +4 */
 /* This is the primary loop */
 	.align	4
 L(duLoop):
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD2,8(rSTR2)
-	cmpld	cr1,rWORD3,rWORD4
-	bne	cr6,L(duLcr6)
-	srd	rA,rWORD2,rSHR
-	sld	rD,rWORD2,rSHL
-	or	rWORD2,rA,rB
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD1, 8(rSTR1)
+	ld	rWORD2, 8(rSTR2)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	bne	cr6, L(duLcr6)
+	srd	r0, rWORD2, rSHR
+	sld	rWORD2_SHIFT, rWORD2, rSHL
+	or	rWORD2, r0, rWORD8_SHIFT
 L(duLoop1):
-	ld	rWORD3,16(rSTR1)
-	ld	rWORD4,16(rSTR2)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr5,L(duLcr5)
-	srd	rC,rWORD4,rSHR
-	sld	rF,rWORD4,rSHL
-	or	rWORD4,rC,rD
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD3, 0, rSTR1
+	ldbrx	rWORD4, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD3, 16(rSTR1)
+	ld	rWORD4, 16(rSTR2)
+#endif
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr5, L(duLcr5)
+	srd	r12, rWORD4, rSHR
+	sld	rWORD4_SHIFT, rWORD4, rSHL
+	or	rWORD4, r12, rWORD2_SHIFT
 L(duLoop2):
-	ld	rWORD5,24(rSTR1)
-	ld	rWORD6,24(rSTR2)
-	cmpld	cr5,rWORD7,rWORD8
-	bne	cr0,L(duLcr0)
-	srd	rE,rWORD6,rSHR
-	sld	rH,rWORD6,rSHL
-	or	rWORD6,rE,rF
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD5, 0, rSTR1
+	ldbrx	rWORD6, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD5, 24(rSTR1)
+	ld	rWORD6, 24(rSTR2)
+#endif
+	cmpld	cr5, rWORD7, rWORD8
+	bne	cr7, L(duLcr7)
+	srd	r0, rWORD6, rSHR
+	sld	rWORD6_SHIFT, rWORD6, rSHL
+	or	rWORD6, r0, rWORD4_SHIFT
 L(duLoop3):
-	ldu	rWORD7,32(rSTR1)
-	ldu	rWORD8,32(rSTR2)
-	cmpld	cr0,rWORD1,rWORD2
-	bne-	cr1,L(duLcr1)
-	srd	rG,rWORD8,rSHR
-	sld	rB,rWORD8,rSHL
-	or	rWORD8,rG,rH
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD7, 0, rSTR1
+	ldbrx	rWORD8, 0, rSTR2
+	addi	rSTR1, rSTR1, 8
+	addi	rSTR2, rSTR2, 8
+#else
+	ldu	rWORD7, 32(rSTR1)
+	ldu	rWORD8, 32(rSTR2)
+#endif
+	cmpld	cr7, rWORD1, rWORD2
+	bne	cr1, L(duLcr1)
+	srd	r12, rWORD8, rSHR
+	sld	rWORD8_SHIFT, rWORD8, rSHL
+	or	rWORD8, r12, rWORD6_SHIFT
 	bdnz	L(duLoop)
 
 L(duL4):
-	bne	cr1,L(duLcr1)
-	cmpld	cr1,rWORD3,rWORD4
-	bne	cr6,L(duLcr6)
-	cmpld	cr6,rWORD5,rWORD6
-	bne	cr5,L(duLcr5)
-	cmpld	cr5,rWORD7,rWORD8
+#if 0
+/* Huh?  We've already branched on cr1!  */
+	bne	cr1, L(duLcr1)
+#endif
+	cmpld	cr1, rWORD3, rWORD4
+	bne	cr6, L(duLcr6)
+	cmpld	cr6, rWORD5, rWORD6
+	bne	cr5, L(duLcr5)
+	cmpld	cr5, rWORD7, rWORD8
 L(du44):
-	bne	cr0,L(duLcr0)
+	bne	cr7, L(duLcr7)
 L(du34):
-	bne	cr1,L(duLcr1)
+	bne	cr1, L(duLcr1)
 L(du24):
-	bne	cr6,L(duLcr6)
+	bne	cr6, L(duLcr6)
 L(du14):
-	sldi.	rN,rN,3
-	bne	cr5,L(duLcr5)
+	sldi.	rN, rN, 3
+	bne	cr5, L(duLcr5)
 /* At this point we have a remainder of 1 to 7 bytes to compare.  We use
    shift right double to eliminate bits beyond the compare length.
-   This allows the use of double word subtract to compute the final
-   result.
 
    However it may not be safe to load rWORD2 which may be beyond the
    string length. So we compare the bit length of the remainder to
    the right shift count (rSHR). If the bit count is less than or equal
    we do not need to load rWORD2 (all significant bits are already in
-   rB).  */
-	cmpld	cr7,rN,rSHR
+   rWORD8_SHIFT).  */
+	cmpld	cr7, rN, rSHR
 	beq	L(duZeroReturn)
-	li	rA,0
-	ble	cr7,L(dutrim)
-	ld	rWORD2,8(rSTR2)
-	srd	rA,rWORD2,rSHR
+	li	r0, 0
+	ble	cr7, L(dutrim)
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD2, 0, rSTR2
+	addi	rSTR2, rSTR2, 8
+#else
+	ld	rWORD2, 8(rSTR2)
+#endif
+	srd	r0, rWORD2, rSHR
 	.align	4
 L(dutrim):
-	ld	rWORD1,8(rSTR1)
-	ld	rWORD8,-8(r1)
-	subfic	rN,rN,64	/* Shift count is 64 - (rN * 8).  */
-	or	rWORD2,rA,rB
-	ld	rWORD7,-16(r1)
-	ld	r29,-24(r1)
-	srd	rWORD1,rWORD1,rN
-	srd	rWORD2,rWORD2,rN
-	ld	r28,-32(r1)
-	ld	r27,-40(r1)
-	li	rRTN,0
-	cmpld	cr0,rWORD1,rWORD2
-	ld	r26,-48(r1)
-	ld	r25,-56(r1)
-	beq	cr0,L(dureturn24)
-	li	rRTN,1
-	ld	r24,-64(r1)
-	bgtlr	cr0
-	li	rRTN,-1
+#ifdef __LITTLE_ENDIAN__
+	ldbrx	rWORD1, 0, rSTR1
+#else
+	ld	rWORD1, 8(rSTR1)
+#endif
+	ld	rWORD8, -8(r1)
+	subfic	rN, rN, 64	/* Shift count is 64 - (rN * 8).  */
+	or	rWORD2, r0, rWORD8_SHIFT
+	ld	rWORD7, -16(r1)
+	ld	rSHL, -24(r1)
+	srd	rWORD1, rWORD1, rN
+	srd	rWORD2, rWORD2, rN
+	ld	rSHR, -32(r1)
+	ld	rWORD8_SHIFT, -40(r1)
+	li	rRTN, 0
+	cmpld	cr7, rWORD1, rWORD2
+	ld	rWORD2_SHIFT, -48(r1)
+	ld	rWORD4_SHIFT, -56(r1)
+	beq	cr7, L(dureturn24)
+	li	rRTN, 1
+	ld	rWORD6_SHIFT, -64(r1)
+	bgtlr	cr7
+	li	rRTN, -1
 	blr
 	.align	4
-L(duLcr0):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
-	bgt	cr0,L(dureturn29)
-	ld	r29,-24(r1)
-	ld	r28,-32(r1)
-	li	rRTN,-1
+L(duLcr7):
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+	li	rRTN, 1
+	bgt	cr7, L(dureturn29)
+	ld	rSHL, -24(r1)
+	ld	rSHR, -32(r1)
+	li	rRTN, -1
 	b	L(dureturn27)
 	.align	4
 L(duLcr1):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
-	bgt	cr1,L(dureturn29)
-	ld	r29,-24(r1)
-	ld	r28,-32(r1)
-	li	rRTN,-1
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+	li	rRTN, 1
+	bgt	cr1, L(dureturn29)
+	ld	rSHL, -24(r1)
+	ld	rSHR, -32(r1)
+	li	rRTN, -1
 	b	L(dureturn27)
 	.align	4
 L(duLcr6):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
-	bgt	cr6,L(dureturn29)
-	ld	r29,-24(r1)
-	ld	r28,-32(r1)
-	li	rRTN,-1
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+	li	rRTN, 1
+	bgt	cr6, L(dureturn29)
+	ld	rSHL, -24(r1)
+	ld	rSHR, -32(r1)
+	li	rRTN, -1
 	b	L(dureturn27)
 	.align	4
 L(duLcr5):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
-	li	rRTN,1
-	bgt	cr5,L(dureturn29)
-	ld	r29,-24(r1)
-	ld	r28,-32(r1)
-	li	rRTN,-1
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
+	li	rRTN, 1
+	bgt	cr5, L(dureturn29)
+	ld	rSHL, -24(r1)
+	ld	rSHR, -32(r1)
+	li	rRTN, -1
 	b	L(dureturn27)
 	.align	3
 L(duZeroReturn):
-	li	rRTN,0
+	li	rRTN, 0
 	.align	4
 L(dureturn):
-	ld	rWORD8,-8(r1)
-	ld	rWORD7,-16(r1)
+	ld	rWORD8, -8(r1)
+	ld	rWORD7, -16(r1)
 L(dureturn29):
-	ld	r29,-24(r1)
-	ld	r28,-32(r1)
+	ld	rSHL, -24(r1)
+	ld	rSHR, -32(r1)
 L(dureturn27):
-	ld	r27,-40(r1)
+	ld	rWORD8_SHIFT, -40(r1)
 L(dureturn26):
-	ld	r26,-48(r1)
+	ld	rWORD2_SHIFT, -48(r1)
 L(dureturn25):
-	ld	r25,-56(r1)
+	ld	rWORD4_SHIFT, -56(r1)
 L(dureturn24):
-	ld	r24,-64(r1)
+	ld	rWORD6_SHIFT, -64(r1)
 	blr
 L(duzeroLength):
-	li	rRTN,0
+	li	rRTN, 0
 	blr
 
 END (memcmp)
 libc_hidden_builtin_def (memcmp)
-weak_alias (memcmp,bcmp)
+weak_alias (memcmp, bcmp)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/memcpy.S b/libc/sysdeps/powerpc/powerpc64/power7/memcpy.S
index 800a9f1bb..e8df75f59 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/memcpy.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/memcpy.S
@@ -23,418 +23,361 @@
 /* __ptr_t [r3] memcpy (__ptr_t dst [r3], __ptr_t src [r4], size_t len [r5]);
    Returns 'dst'.  */
 
+#define dst 11		/* Use r11 so r3 kept unchanged.  */
+#define src 4
+#define cnt 5
+
 	.machine power7
 EALIGN (memcpy, 5, 0)
 	CALL_MCOUNT 3
 
-	cmpldi  cr1,5,31
+	cmpldi	cr1,cnt,31
 	neg	0,3
-	std	3,-16(1)
-	std	31,-8(1)
-	cfi_offset(31,-8)
 	ble	cr1, L(copy_LT_32)  /* If move < 32 bytes use short move
 				    code.  */
 
-	andi.   11,3,7	      /* Check alignment of DST.  */
-
-
-	clrldi  10,4,61       /* Check alignment of SRC.  */
-	cmpld   cr6,10,11     /* SRC and DST alignments match?  */
-	mr	12,4
-	mr	31,5
+#ifdef __LITTLE_ENDIAN__
+/* In little-endian mode, power7 takes an alignment trap on any lxvd2x
+   or stxvd2x crossing a 32-byte boundary, so ensure the aligned_copy
+   loop is only used for quadword aligned copies.  */
+	andi.	10,3,15
+	clrldi	11,4,60
+#else
+	andi.	10,3,7		/* Check alignment of DST.  */
+	clrldi	11,4,61		/* Check alignment of SRC.  */
+#endif
+	cmpld	cr6,10,11	/* SRC and DST alignments match?  */
+
+	mr	dst,3
 	bne	cr6,L(copy_GE_32_unaligned)
+	beq	L(aligned_copy)
 
-	srdi    9,5,3	      /* Number of full quadwords remaining.  */
-
-	beq    L(copy_GE_32_aligned_cont)
-
-	clrldi  0,0,61
-	mtcrf   0x01,0
-	subf    31,0,5
-
-	/* Get the SRC aligned to 8 bytes.  */
-
-1:	bf	31,2f
-	lbz	6,0(12)
-	addi    12,12,1
-	stb	6,0(3)
-	addi    3,3,1
-2:	bf      30,4f
-	lhz     6,0(12)
-	addi    12,12,2
-	sth     6,0(3)
-	addi    3,3,2
-4:	bf      29,0f
-	lwz     6,0(12)
-	addi    12,12,4
-	stw     6,0(3)
-	addi    3,3,4
-0:
-	clrldi  10,12,61      /* Check alignment of SRC again.  */
-	srdi    9,31,3	      /* Number of full doublewords remaining.  */
-
-L(copy_GE_32_aligned_cont):
-
-	clrldi  11,31,61
-	mtcrf   0x01,9
-
-	srdi    8,31,5
-	cmpldi  cr1,9,4
-	cmpldi  cr6,11,0
-	mr	11,12
-
-	/* Copy 1~3 doublewords so the main loop starts
-	at a multiple of 32 bytes.  */
+	mtocrf	0x01,0
+#ifdef __LITTLE_ENDIAN__
+	clrldi	0,0,60
+#else
+	clrldi	0,0,61
+#endif
 
-	bf	30,1f
-	ld      6,0(12)
-	ld      7,8(12)
-	addi    11,12,16
-	mtctr   8
-	std     6,0(3)
-	std     7,8(3)
-	addi    10,3,16
-	bf      31,4f
-	ld      0,16(12)
-	std     0,16(3)
-	blt     cr1,3f
-	addi    11,12,24
-	addi    10,3,24
-	b       4f
-
-	.align  4
-1:	/* Copy 1 doubleword and set the counter.  */
-	mr	10,3
-	mtctr   8
-	bf      31,4f
-	ld      6,0(12)
-	addi    11,12,8
-	std     6,0(3)
-	addi    10,3,8
-
-L(aligned_copy):
-	/* Main aligned copy loop. Copies up to 128-bytes at a time. */
-	.align  4
+/* Get the DST and SRC aligned to 8 bytes (16 for little-endian).  */
+1:
+	bf	31,2f
+	lbz	6,0(src)
+	addi	src,src,1
+	stb	6,0(dst)
+	addi	dst,dst,1
+2:
+	bf	30,4f
+	lhz	6,0(src)
+	addi	src,src,2
+	sth	6,0(dst)
+	addi	dst,dst,2
 4:
-	/* check for any 32-byte or 64-byte lumps that are outside of a
-	   nice 128-byte range.  R8 contains the number of 32-byte
-	   lumps, so drop this into the CR, and use the SO/EQ bits to help
-	   handle the 32- or 64- byte lumps.  Then handle the rest with an
-	   unrolled 128-bytes-at-a-time copy loop. */
-	mtocrf	1,8
-	li	6,16	# 16() index
-	li	7,32	# 32() index
-	li	8,48	# 48() index
-
-L(aligned_32byte):
-	/* if the SO bit (indicating a 32-byte lump) is not set, move along. */
-	bns	cr7,L(aligned_64byte)
-	lxvd2x	6,0,11
-	lxvd2x	7,11,6
-	addi	11,11,32
-	stxvd2x	6,0,10
-	stxvd2x	7,10,6
-	addi	10,10,32
-
-L(aligned_64byte):
-	/* if the EQ bit (indicating a 64-byte lump) is not set, move along. */
-	bne	cr7,L(aligned_128setup)
-	lxvd2x	6,0,11
-	lxvd2x	7,11,6
-	lxvd2x	8,11,7
-	lxvd2x	9,11,8
-	addi	11,11,64
-	stxvd2x	6,0,10
-	stxvd2x	7,10,6
-	stxvd2x	8,10,7
-	stxvd2x	9,10,8
-	addi	10,10,64
-
-L(aligned_128setup):
-	/* Set up for the 128-byte at a time copy loop.  */
-	srdi	8,31,7
-	cmpdi	8,0	# Any 4x lumps left?
-	beq	3f	# if not, move along.
-	lxvd2x	6,0,11
-	lxvd2x	7,11,6
-	mtctr	8	# otherwise, load the ctr and begin.
-	li	8,48	# 48() index
+	bf	29,8f
+	lwz	6,0(src)
+	addi	src,src,4
+	stw	6,0(dst)
+	addi	dst,dst,4
+8:
+#ifdef __LITTLE_ENDIAN__
+	bf	28,16f
+	ld	6,0(src)
+	addi	src,src,8
+	std	6,0(dst)
+	addi	dst,dst,8
+16:
+#endif
+	subf	cnt,0,cnt
+
+/* Main aligned copy loop. Copies 128 bytes at a time. */
+L(aligned_copy):
+	li	6,16
+	li	7,32
+	li	8,48
+	mtocrf	0x02,cnt
+	srdi	12,cnt,7
+	cmpdi	12,0
+	beq	L(aligned_tail)
+	lxvd2x	6,0,src
+	lxvd2x	7,src,6
+	mtctr	12
 	b	L(aligned_128loop)
 
+	.align  4
 L(aligned_128head):
 	/* for the 2nd + iteration of this loop. */
-	lxvd2x	6,0,11
-	lxvd2x	7,11,6
+	lxvd2x	6,0,src
+	lxvd2x	7,src,6
 L(aligned_128loop):
-	lxvd2x	8,11,7
-	lxvd2x	9,11,8
-	stxvd2x	6,0,10
-	addi	11,11,64
-	stxvd2x	7,10,6
-	stxvd2x	8,10,7
-	stxvd2x	9,10,8
-	lxvd2x	6,0,11
-	lxvd2x	7,11,6
-	addi	10,10,64
-	lxvd2x	8,11,7
-	lxvd2x	9,11,8
-	addi	11,11,64
-	stxvd2x	6,0,10
-	stxvd2x	7,10,6
-	stxvd2x	8,10,7
-	stxvd2x	9,10,8
-	addi	10,10,64
+	lxvd2x	8,src,7
+	lxvd2x	9,src,8
+	stxvd2x	6,0,dst
+	addi	src,src,64
+	stxvd2x	7,dst,6
+	stxvd2x	8,dst,7
+	stxvd2x	9,dst,8
+	lxvd2x	6,0,src
+	lxvd2x	7,src,6
+	addi	dst,dst,64
+	lxvd2x	8,src,7
+	lxvd2x	9,src,8
+	addi	src,src,64
+	stxvd2x	6,0,dst
+	stxvd2x	7,dst,6
+	stxvd2x	8,dst,7
+	stxvd2x	9,dst,8
+	addi	dst,dst,64
 	bdnz	L(aligned_128head)
 
-3:
-	/* Check for tail bytes.  */
-	rldicr  0,31,0,60
-	mtcrf   0x01,31
-	beq	cr6,0f
-
-.L9:
-	add	3,3,0
-	add	12,12,0
-
-	/*  At this point we have a tail of 0-7 bytes and we know that the
-	destination is doubleword-aligned.  */
-4:	/* Copy 4 bytes.  */
-	bf	29,2f
-
-	lwz     6,0(12)
-	addi    12,12,4
-	stw     6,0(3)
-	addi    3,3,4
-2:	/* Copy 2 bytes.  */
-	bf	30,1f
-
-	lhz     6,0(12)
-	addi    12,12,2
-	sth     6,0(3)
-	addi    3,3,2
-1:	/* Copy 1 byte.  */
-	bf	31,0f
-
-	lbz	6,0(12)
-	stb	6,0(3)
-0:	/* Return original DST pointer.  */
-	ld	31,-8(1)
-	ld	3,-16(1)
+L(aligned_tail):
+	mtocrf	0x01,cnt
+	bf	25,32f
+	lxvd2x	6,0,src
+	lxvd2x	7,src,6
+	lxvd2x	8,src,7
+	lxvd2x	9,src,8
+	addi	src,src,64
+	stxvd2x	6,0,dst
+	stxvd2x	7,dst,6
+	stxvd2x	8,dst,7
+	stxvd2x	9,dst,8
+	addi	dst,dst,64
+32:
+	bf	26,16f
+	lxvd2x	6,0,src
+	lxvd2x	7,src,6
+	addi	src,src,32
+	stxvd2x	6,0,dst
+	stxvd2x	7,dst,6
+	addi	dst,dst,32
+16:
+	bf	27,8f
+	lxvd2x	6,0,src
+	addi	src,src,16
+	stxvd2x	6,0,dst
+	addi	dst,dst,16
+8:
+	bf	28,4f
+	ld	6,0(src)
+	addi	src,src,8
+	std     6,0(dst)
+	addi	dst,dst,8
+4:	/* Copies 4~7 bytes.  */
+	bf	29,L(tail2)
+	lwz	6,0(src)
+	stw     6,0(dst)
+	bf      30,L(tail5)
+	lhz     7,4(src)
+	sth     7,4(dst)
+	bflr	31
+	lbz     8,6(src)
+	stb     8,6(dst)
+	/* Return original DST pointer.  */
 	blr
 
-	/* Handle copies of 0~31 bytes.  */
-	.align  4
+
+/* Handle copies of 0~31 bytes.  */
+	.align	4
 L(copy_LT_32):
-	cmpldi  cr6,5,8
-	mr	12,4
-	mtcrf   0x01,5
+	mr	dst,3
+	cmpldi	cr6,cnt,8
+	mtocrf	0x01,cnt
 	ble	cr6,L(copy_LE_8)
 
 	/* At least 9 bytes to go.  */
 	neg	8,4
-	clrrdi  11,4,2
-	andi.   0,8,3
-	cmpldi  cr1,5,16
-	mr	10,5
+	andi.	0,8,3
+	cmpldi	cr1,cnt,16
 	beq	L(copy_LT_32_aligned)
 
-	/* Force 4-bytes alignment for SRC.  */
-	mtocrf  0x01,0
-	subf    10,0,5
-2:	bf	30,1f
-
-	lhz	6,0(12)
-	addi    12,12,2
-	sth	6,0(3)
-	addi    3,3,2
-1:	bf	31,L(end_4bytes_alignment)
-
-	lbz	6,0(12)
-	addi    12,12,1
-	stb	6,0(3)
-	addi    3,3,1
-
-	.align  4
+	/* Force 4-byte alignment for SRC.  */
+	mtocrf	0x01,0
+	subf	cnt,0,cnt
+2:
+	bf	30,1f
+	lhz	6,0(src)
+	addi	src,src,2
+	sth	6,0(dst)
+	addi	dst,dst,2
+1:
+	bf	31,L(end_4bytes_alignment)
+	lbz	6,0(src)
+	addi	src,src,1
+	stb	6,0(dst)
+	addi	dst,dst,1
+
+	.align	4
 L(end_4bytes_alignment):
-	cmpldi  cr1,10,16
-	mtcrf   0x01,10
+	cmpldi	cr1,cnt,16
+	mtocrf	0x01,cnt
 
 L(copy_LT_32_aligned):
 	/* At least 6 bytes to go, and SRC is word-aligned.  */
 	blt	cr1,8f
 
 	/* Copy 16 bytes.  */
-	lwz	6,0(12)
-	lwz     7,4(12)
-	stw     6,0(3)
-	lwz     8,8(12)
-	stw     7,4(3)
-	lwz     6,12(12)
-	addi    12,12,16
-	stw     8,8(3)
-	stw     6,12(3)
-	addi    3,3,16
+	lwz	6,0(src)
+	lwz	7,4(src)
+	stw	6,0(dst)
+	lwz	8,8(src)
+	stw	7,4(dst)
+	lwz	6,12(src)
+	addi	src,src,16
+	stw	8,8(dst)
+	stw	6,12(dst)
+	addi	dst,dst,16
 8:	/* Copy 8 bytes.  */
-	bf	28,4f
+	bf	28,L(tail4)
+	lwz	6,0(src)
+	lwz	7,4(src)
+	addi	src,src,8
+	stw	6,0(dst)
+	stw	7,4(dst)
+	addi	dst,dst,8
+
+	.align	4
+/* Copies 4~7 bytes.  */
+L(tail4):
+	bf	29,L(tail2)
+	lwz	6,0(src)
+	stw	6,0(dst)
+	bf	30,L(tail5)
+	lhz	7,4(src)
+	sth	7,4(dst)
+	bflr	31
+	lbz	8,6(src)
+	stb	8,6(dst)
+	/* Return original DST pointer.  */
+	blr
 
-	lwz     6,0(12)
-	lwz     7,4(12)
-	addi    12,12,8
-	stw     6,0(3)
-	stw     7,4(3)
-	addi    3,3,8
-4:	/* Copy 4 bytes.  */
-	bf	29,2f
-
-	lwz     6,0(12)
-	addi    12,12,4
-	stw     6,0(3)
-	addi    3,3,4
-2:	/* Copy 2-3 bytes.  */
+	.align	4
+/* Copies 2~3 bytes.  */
+L(tail2):
 	bf	30,1f
-
-	lhz     6,0(12)
-	sth     6,0(3)
-	bf      31,0f
-	lbz     7,2(12)
-	stb     7,2(3)
-	ld	3,-16(1)
+	lhz	6,0(src)
+	sth	6,0(dst)
+	bflr	31
+	lbz	7,2(src)
+	stb	7,2(dst)
 	blr
 
-	.align  4
-1:	/* Copy 1 byte.  */
-	bf	31,0f
+	.align	4
+L(tail5):
+	bflr	31
+	lbz	6,4(src)
+	stb	6,4(dst)
+	blr
 
-	lbz	6,0(12)
-	stb	6,0(3)
-0:	/* Return original DST pointer.  */
-	ld	3,-16(1)
+	.align	4
+1:
+	bflr	31
+	lbz	6,0(src)
+	stb	6,0(dst)
+	/* Return original DST pointer.  */
 	blr
 
-	/* Handles copies of 0~8 bytes.  */
-	.align  4
+
+/* Handles copies of 0~8 bytes.  */
+	.align	4
 L(copy_LE_8):
-	bne	cr6,4f
+	bne	cr6,L(tail4)
 
 	/* Though we could've used ld/std here, they are still
 	slow for unaligned cases.  */
 
-	lwz	6,0(4)
-	lwz     7,4(4)
-	stw     6,0(3)
-	stw     7,4(3)
-	ld      3,-16(1)      /* Return original DST pointers.  */
+	lwz	6,0(src)
+	lwz	7,4(src)
+	stw	6,0(dst)
+	stw	7,4(dst)
 	blr
 
-	.align  4
-4:	/* Copies 4~7 bytes.  */
-	bf	29,2b
-
-	lwz	6,0(4)
-	stw     6,0(3)
-	bf      30,5f
-	lhz     7,4(4)
-	sth     7,4(3)
-	bf      31,0f
-	lbz     8,6(4)
-	stb     8,6(3)
-	ld	3,-16(1)
-	blr
-
-	.align  4
-5:	/* Copy 1 byte.  */
-	bf	31,0f
-
-	lbz	6,4(4)
-	stb	6,4(3)
-
-0:	/* Return original DST pointer.  */
-	ld	3,-16(1)
-	blr
 
-	/* Handle copies of 32+ bytes where DST is aligned (to quadword) but
-	SRC is not.  Use aligned quadword loads from SRC, shifted to realign
-	the data, allowing for aligned DST stores.  */
-	.align  4
+/* Handle copies of 32+ bytes where DST is aligned (to quadword) but
+   SRC is not.	Use aligned quadword loads from SRC, shifted to realign
+   the data, allowing for aligned DST stores.  */
+	.align	4
 L(copy_GE_32_unaligned):
-	clrldi  0,0,60	      /* Number of bytes until the 1st
-			      quadword.  */
-	andi.   11,3,15       /* Check alignment of DST (against
-			      quadwords).  */
-	srdi    9,5,4	      /* Number of full quadwords remaining.  */
+	clrldi	0,0,60	      /* Number of bytes until the 1st dst quadword.  */
+#ifndef __LITTLE_ENDIAN__
+	andi.	10,3,15	      /* Check alignment of DST (against quadwords).  */
+#endif
+	srdi	9,cnt,4	      /* Number of full quadwords remaining.  */
 
 	beq	L(copy_GE_32_unaligned_cont)
 
-	/* SRC is not quadword aligned, get it aligned.  */
+	/* DST is not quadword aligned, get it aligned.  */
 
-	mtcrf   0x01,0
-	subf    31,0,5
+	mtocrf	0x01,0
+	subf	cnt,0,cnt
 
 	/* Vector instructions work best when proper alignment (16-bytes)
 	is present.  Move 0~15 bytes as needed to get DST quadword-aligned.  */
-1:	/* Copy 1 byte.  */
+1:
 	bf	31,2f
-
-	lbz	6,0(12)
-	addi    12,12,1
-	stb	6,0(3)
-	addi    3,3,1
-2:	/* Copy 2 bytes.  */
+	lbz	6,0(src)
+	addi	src,src,1
+	stb	6,0(dst)
+	addi	dst,dst,1
+2:
 	bf	30,4f
-
-	lhz     6,0(12)
-	addi    12,12,2
-	sth     6,0(3)
-	addi    3,3,2
-4:	/* Copy 4 bytes.  */
+	lhz	6,0(src)
+	addi	src,src,2
+	sth	6,0(dst)
+	addi	dst,dst,2
+4:
 	bf	29,8f
-
-	lwz     6,0(12)
-	addi    12,12,4
-	stw     6,0(3)
-	addi    3,3,4
-8:	/* Copy 8 bytes.  */
+	lwz	6,0(src)
+	addi	src,src,4
+	stw	6,0(dst)
+	addi	dst,dst,4
+8:
 	bf	28,0f
-
-	ld	6,0(12)
-	addi    12,12,8
-	std	6,0(3)
-	addi    3,3,8
+	ld	6,0(src)
+	addi	src,src,8
+	std	6,0(dst)
+	addi	dst,dst,8
 0:
-	clrldi  10,12,60      /* Check alignment of SRC.  */
-	srdi    9,31,4	      /* Number of full quadwords remaining.  */
+	srdi	9,cnt,4	      /* Number of full quadwords remaining.  */
 
 	/* The proper alignment is present, it is OK to copy the bytes now.  */
 L(copy_GE_32_unaligned_cont):
 
 	/* Setup two indexes to speed up the indexed vector operations.  */
-	clrldi  11,31,60
-	li      6,16	      /* Index for 16-bytes offsets.  */
+	clrldi	10,cnt,60
+	li	6,16	      /* Index for 16-bytes offsets.  */
 	li	7,32	      /* Index for 32-bytes offsets.  */
-	cmpldi  cr1,11,0
-	srdi    8,31,5	      /* Setup the loop counter.  */
-	mr      10,3
-	mr      11,12
-	mtcrf   0x01,9
-	cmpldi  cr6,9,1
-	lvsl    5,0,12
-	lvx     3,0,12
-	bf      31,L(setup_unaligned_loop)
-
-	/* Copy another 16 bytes to align to 32-bytes due to the loop .  */
-	lvx     4,12,6
-	vperm   6,3,4,5
-	addi    11,12,16
-	addi    10,3,16
-	stvx    6,0,3
+	cmpldi	cr1,10,0
+	srdi	8,cnt,5	      /* Setup the loop counter.  */
+	mtocrf	0x01,9
+	cmpldi	cr6,9,1
+#ifdef __LITTLE_ENDIAN__
+	lvsr	5,0,src
+#else
+	lvsl	5,0,src
+#endif
+	lvx	3,0,src
+	li	0,0
+	bf	31,L(setup_unaligned_loop)
+
+	/* Copy another 16 bytes to align to 32-bytes due to the loop.  */
+	lvx	4,src,6
+#ifdef __LITTLE_ENDIAN__
+	vperm	6,4,3,5
+#else
+	vperm	6,3,4,5
+#endif
+	addi	src,src,16
+	stvx	6,0,dst
+	addi	dst,dst,16
 	vor	3,4,4
+	clrrdi	0,src,60
 
 L(setup_unaligned_loop):
-	mtctr   8
-	ble     cr6,L(end_unaligned_loop)
+	mtctr	8
+	ble	cr6,L(end_unaligned_loop)
 
 	/* Copy 32 bytes at a time using vector instructions.  */
-	.align  4
+	.align	4
 L(unaligned_loop):
 
 	/* Note: vr6/vr10 may contain data that was already copied,
@@ -442,62 +385,55 @@ L(unaligned_loop):
 	some portions again. This is faster than having unaligned
 	vector instructions though.  */
 
-	lvx	4,11,6	      /* vr4 = r11+16.  */
-	vperm   6,3,4,5	      /* Merge the correctly-aligned portions
-			      of vr3/vr4 into vr6.  */
-	lvx	3,11,7	      /* vr3 = r11+32.  */
-	vperm   10,4,3,5      /* Merge the correctly-aligned portions
-			      of vr3/vr4 into vr10.  */
-	addi    11,11,32
-	stvx    6,0,10
-	stvx    10,10,6
-	addi    10,10,32
-
+	lvx	4,src,6
+#ifdef __LITTLE_ENDIAN__
+	vperm	6,4,3,5
+#else
+	vperm	6,3,4,5
+#endif
+	lvx	3,src,7
+#ifdef __LITTLE_ENDIAN__
+	vperm	10,3,4,5
+#else
+	vperm	10,4,3,5
+#endif
+	addi	src,src,32
+	stvx	6,0,dst
+	stvx	10,dst,6
+	addi	dst,dst,32
 	bdnz	L(unaligned_loop)
 
-	.align  4
+	clrrdi	0,src,60
+
+	.align	4
 L(end_unaligned_loop):
 
 	/* Check for tail bytes.  */
-	rldicr  0,31,0,59
-	mtcrf   0x01,31
-	beq	cr1,0f
+	mtocrf	0x01,cnt
+	beqlr	cr1
 
-	add	3,3,0
-	add	12,12,0
+	add	src,src,0
 
 	/*  We have 1~15 tail bytes to copy, and DST is quadword aligned.  */
-8:	/* Copy 8 bytes.  */
+	/* Copy 8 bytes.  */
 	bf	28,4f
-
-	lwz	6,0(12)
-	lwz	7,4(12)
-	addi    12,12,8
-	stw	6,0(3)
-	stw	7,4(3)
-	addi    3,3,8
-4:	/* Copy 4 bytes.  */
-	bf	29,2f
-
-	lwz	6,0(12)
-	addi    12,12,4
-	stw	6,0(3)
-	addi    3,3,4
-2:	/* Copy 2~3 bytes.  */
-	bf	30,1f
-
-	lhz	6,0(12)
-	addi    12,12,2
-	sth	6,0(3)
-	addi    3,3,2
-1:	/* Copy 1 byte.  */
-	bf	31,0f
-
-	lbz	6,0(12)
-	stb	6,0(3)
-0:	/* Return original DST pointer.  */
-	ld	31,-8(1)
-	ld	3,-16(1)
+	lwz	6,0(src)
+	lwz	7,4(src)
+	addi	src,src,8
+	stw	6,0(dst)
+	stw	7,4(dst)
+	addi	dst,dst,8
+4:	/* Copy 4~7 bytes.  */
+	bf	29,L(tail2)
+	lwz	6,0(src)
+	stw	6,0(dst)
+	bf	30,L(tail5)
+	lhz	7,4(src)
+	sth	7,4(dst)
+	bflr	31
+	lbz	8,6(src)
+	stb	8,6(dst)
+	/* Return original DST pointer.  */
 	blr
 
 END_GEN_TB (memcpy,TB_TOCLESS)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S b/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S
index f20be938d..b93ab7da5 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/mempcpy.S
@@ -365,13 +365,21 @@ L(copy_GE_32_unaligned_cont):
 	mr	11,12
 	mtcrf	0x01,9
 	cmpldi	cr6,9,1
-	lvsl	5,0,12
+#ifdef __LITTLE_ENDIAN__
+	lvsr    5,0,12
+#else
+	lvsl    5,0,12
+#endif
 	lvx	3,0,12
 	bf	31,L(setup_unaligned_loop)
 
 	/* Copy another 16 bytes to align to 32-bytes due to the loop .  */
 	lvx	4,12,6
-	vperm	6,3,4,5
+#ifdef __LITTLE_ENDIAN__
+	vperm   6,4,3,5
+#else
+	vperm   6,3,4,5
+#endif
 	addi	11,12,16
 	addi	10,3,16
 	stvx	6,0,3
@@ -391,11 +399,17 @@ L(unaligned_loop):
 	vector instructions though.  */
 
 	lvx	4,11,6	      /* vr4 = r11+16.  */
-	vperm	6,3,4,5	      /* Merge the correctly-aligned portions
-				 of vr3/vr4 into vr6.  */
+#ifdef __LITTLE_ENDIAN__
+	vperm   6,4,3,5
+#else
+	vperm   6,3,4,5
+#endif
 	lvx	3,11,7	      /* vr3 = r11+32.  */
-	vperm	10,4,3,5      /* Merge the correctly-aligned portions
-				 of vr3/vr4 into vr10.  */
+#ifdef __LITTLE_ENDIAN__
+	vperm   10,3,4,5
+#else
+	vperm   10,4,3,5
+#endif
 	addi	11,11,32
 	stvx	6,0,10
 	stvx	10,10,6
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/memrchr.S b/libc/sysdeps/powerpc/powerpc64/power7/memrchr.S
index c49995210..a9e86cb19 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/memrchr.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/memrchr.S
@@ -23,118 +23,132 @@
 	.machine  power7
 ENTRY (__memrchr)
 	CALL_MCOUNT
-	dcbt	0,r3
-	mr	r7,r3
-	add	r3,r7,r5      /* Calculate the last acceptable address.  */
-	cmpld	cr7,r3,r7     /* Is the address equal or less than r3?  */
+	add	r7,r3,r5      /* Calculate the last acceptable address.  */
+	neg	r0,r7
+	addi	r7,r7,-1
+	mr	r10,r3
+	clrrdi	r6,r7,7
+	li	r9,3<<5
+	dcbt	r9,r6,16      /* Stream hint, decreasing addresses.  */
 
 	/* Replicate BYTE to doubleword.  */
-	rlwimi	r4,r4,8,16,23
-	rlwimi	r4,r4,16,0,15
+	insrdi	r4,r4,8,48
+	insrdi	r4,r4,16,32
 	insrdi  r4,r4,32,0
-	bge	cr7,L(proceed)
-
-	li	r3,-1	      /* Make r11 the biggest if r4 <= 0.  */
-L(proceed):
 	li	r6,-8
-	addi	r9,r3,-1
-	clrrdi  r8,r9,3
-	addi	r8,r8,8
-	neg	r0,r3
+	li	r9,-1
 	rlwinm	r0,r0,3,26,28 /* Calculate padding.  */
-
+	clrrdi	r8,r7,3
+	srd	r9,r9,r0
 	cmpldi	r5,32
+	clrrdi	r0,r10,3
 	ble	L(small_range)
 
-	ldbrx	r12,r8,r6     /* Load reversed doubleword from memory.  */
-	cmpb	r10,r12,r4    /* Check for BYTE in DWORD1.  */
-	sld	r10,r10,r0
-	srd	r10,r10,r0
-	cmpldi	cr7,r10,0     /* If r10 == 0, no BYTEs have been found.  */
+#ifdef __LITTLE_ENDIAN__
+	ldx	r12,0,r8
+#else
+	ldbrx	r12,0,r8      /* Load reversed doubleword from memory.  */
+#endif
+	cmpb	r3,r12,r4     /* Check for BYTE in DWORD1.  */
+	and	r3,r3,r9
+	cmpldi	cr7,r3,0      /* If r3 == 0, no BYTEs have been found.  */
 	bne	cr7,L(done)
 
-	/* Are we done already?  */
-	addi	r9,r8,-8
-	cmpld	cr6,r9,r7
-	ble	cr6,L(null)
-
 	mtcrf   0x01,r8
-	/* Are we now aligned to a doubleword boundary?  If so, skip to
+	/* Are we now aligned to a quadword boundary?  If so, skip to
 	   the main loop.  Otherwise, go through the alignment code.  */
-	mr	r8,r9
-	bt	28,L(loop_setup)
+	bf	28,L(loop_setup)
 
 	/* Handle DWORD2 of pair.  */
+#ifdef __LITTLE_ENDIAN__
+	ldx	r12,r8,r6
+#else
 	ldbrx	r12,r8,r6
-	cmpb	r10,r12,r4
-	cmpldi	cr7,r10,0
-	bne	cr7,L(done)
-
-	/* Are we done already.  */
+#endif
 	addi	r8,r8,-8
-	cmpld	cr6,r8,r7
-	ble	cr6,L(null)
+	cmpb	r3,r12,r4
+	cmpldi	cr7,r3,0
+	bne	cr7,L(done)
 
 L(loop_setup):
-	li	r0,-16
-	sub	r5,r8,r7
-	srdi	r9,r5,4	      /* Number of loop iterations.  */
+	/* The last dword we want to read in the loop below is the one
+	   containing the first byte of the string, ie. the dword at
+	   s & ~7, or r0.  The first dword read is at r8 - 8, we
+	   read 2 * cnt dwords, so the last dword read will be at
+	   r8 - 8 - 16 * cnt + 8.  Solving for cnt gives
+	   cnt = (r8 - r0) / 16  */
+	sub	r5,r8,r0
+	addi	r8,r8,-8
+	srdi	r9,r5,4       /* Number of loop iterations.  */
 	mtctr	r9	      /* Setup the counter.  */
-	b	L(loop)
-	/* Main loop to look for BYTE backwards in the string.  Since it's a
-	   small loop (< 8 instructions), align it to 32-bytes.  */
-	.p2align  5
+
+	/* Main loop to look for BYTE backwards in the string.
+	   FIXME: Investigate whether 32 byte align helps with this
+	   9 instruction loop.  */
+	.align	5
 L(loop):
 	/* Load two doublewords, compare and merge in a
 	   single register for speed.  This is an attempt
 	   to speed up the byte-checking process for bigger strings.  */
 
-	ldbrx	r12,r8,r6
-	ldbrx	r11,r8,r0
-	addi	r8,r8,-8
-	cmpb	r10,r12,r4
+#ifdef __LITTLE_ENDIAN__
+	ldx	r12,0,r8
+	ldx	r11,r8,r6
+#else
+	ldbrx	r12,0,r8
+	ldbrx	r11,r8,r6
+#endif
+	cmpb	r3,r12,r4
 	cmpb	r9,r11,r4
-	or	r5,r9,r10     /* Merge everything in one doubleword.  */
+	or	r5,r9,r3      /* Merge everything in one doubleword.  */
 	cmpldi	cr7,r5,0
 	bne	cr7,L(found)
-	addi	r8,r8,-8
+	addi	r8,r8,-16
 	bdnz	L(loop)
-	/* We're here because the counter reached 0, and that means we
-	   didn't have any matches for BYTE in the whole range.  Just return
-	   the original range.  */
-	addi	r8,r8,8
-	cmpld	cr6,r8,r7
-	bgt	cr6,L(loop_small)
-	b	L(null)
-
-	/* OK, one (or both) of the words contains BYTE.  Check
-	   the first word and decrement the address in case the first
-	   word really contains BYTE.  */
+
+	/* We may have one more word to read.  */
+	cmpld	r8,r0
+	bnelr
+
+#ifdef __LITTLE_ENDIAN__
+	ldx	r12,0,r8
+#else
+	ldbrx	r12,0,r8
+#endif
+	cmpb	r3,r12,r4
+	cmpldi	cr7,r3,0
+	bne	cr7,L(done)
+	blr
+
 	.align	4
 L(found):
-	cmpldi	cr6,r10,0
-	addi	r8,r8,8
+	/* OK, one (or both) of the dwords contains BYTE.  Check
+	   the first dword.  */
+	cmpldi	cr6,r3,0
 	bne	cr6,L(done)
 
 	/* BYTE must be in the second word.  Adjust the address
-	   again and move the result of cmpb to r10 so we can calculate the
+	   again and move the result of cmpb to r3 so we can calculate the
 	   pointer.  */
 
-	mr	r10,r9
+	mr	r3,r9
 	addi	r8,r8,-8
 
-	/* r10 has the output of the cmpb instruction, that is, it contains
-	   0xff in the same position as the BYTE in the original
+	/* r3 has the output of the cmpb instruction, that is, it contains
+	   0xff in the same position as BYTE in the original
 	   word from the string.  Use that to calculate the pointer.
 	   We need to make sure BYTE is *before* the end of the
 	   range.  */
 L(done):
-	cntlzd	r0,r10	      /* Count leading zeroes before the match.  */
-	srdi	r6,r0,3	      /* Convert leading zeroes to bytes.  */
-	addi	r0,r6,1
+	cntlzd	r9,r3	      /* Count leading zeros before the match.  */
+	cmpld	r8,r0         /* Are we on the last word?  */
+	srdi	r6,r9,3	      /* Convert leading zeros to bytes.  */
+	addi	r0,r6,-7
 	sub	r3,r8,r0
-	cmpld	r3,r7
-	blt	L(null)
+	cmpld	cr7,r3,r10
+	bnelr
+	bgelr	cr7
+	li	r3,0
 	blr
 
 	.align	4
@@ -148,29 +162,35 @@ L(small_range):
 	cmpldi	r5,0
 	beq	L(null)
 
-	ldbrx	r12,r8,r6     /* Load reversed doubleword from memory.  */
-	cmpb	r10,r12,r4    /* Check for BYTE in DWORD1.  */
-	sld	r10,r10,r0
-	srd	r10,r10,r0
-	cmpldi	cr7,r10,0
+#ifdef __LITTLE_ENDIAN__
+	ldx	r12,0,r8
+#else
+	ldbrx	r12,0,r8      /* Load reversed doubleword from memory.  */
+#endif
+	cmpb	r3,r12,r4     /* Check for BYTE in DWORD1.  */
+	and	r3,r3,r9
+	cmpldi	cr7,r3,0
 	bne	cr7,L(done)
 
 	/* Are we done already?  */
+	cmpld	r8,r0
 	addi	r8,r8,-8
-	cmpld	r8,r7
-	ble	L(null)
-	b	L(loop_small)
+	beqlr
 
-	.p2align  5
+	.align	5
 L(loop_small):
-	ldbrx	r12,r8,r6
-	cmpb	r10,r12,r4
-	cmpldi	cr6,r10,0
-	bne	cr6,L(done)
+#ifdef __LITTLE_ENDIAN__
+	ldx	r12,0,r8
+#else
+	ldbrx	r12,0,r8
+#endif
+	cmpb	r3,r12,r4
+	cmpld	r8,r0
+	cmpldi	cr7,r3,0
+	bne	cr7,L(done)
 	addi	r8,r8,-8
-	cmpld	r8,r7
-	ble	L(null)
-	b	L(loop_small)
+	bne	L(loop_small)
+	blr
 
 END (__memrchr)
 weak_alias (__memrchr, memrchr)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/memset.S b/libc/sysdeps/powerpc/powerpc64/power7/memset.S
index b24cfa163..8b081e87c 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/memset.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/memset.S
@@ -32,8 +32,8 @@ L(_memset):
 	mr	10,3
 
 	/* Replicate byte to word.  */
-	rlwimi	4,4,8,16,23
-	rlwimi	4,4,16,0,15
+	insrdi	4,4,8,48
+	insrdi	4,4,16,32
 	ble	cr6,L(small)	/* If length <= 8, use short copy code.  */
 
 	neg	0,3
@@ -321,7 +321,7 @@ L(medium):
 	clrldi	0,0,62
 	beq	L(medium_aligned)
 
-	/* Force 4-bytes alignment for SRC.  */
+	/* Force 4-bytes alignment for DST.  */
 	mtocrf	0x01,0
 	subf	5,0,5
 1:	/* Copy 1 byte.  */
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/rawmemchr.S b/libc/sysdeps/powerpc/powerpc64/power7/rawmemchr.S
index 50a33d8fa..547aed771 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/rawmemchr.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/rawmemchr.S
@@ -27,8 +27,8 @@ ENTRY (__rawmemchr)
 	clrrdi	r8,r3,3	      /* Align the address to doubleword boundary.  */
 
 	/* Replicate byte to doubleword.  */
-	rlwimi	r4,r4,8,16,23
-	rlwimi	r4,r4,16,0,15
+	insrdi	r4,r4,8,48
+	insrdi	r4,r4,16,32
 	insrdi	r4,r4,32,0
 
 	/* Now r4 has a doubleword of c bytes.  */
@@ -36,8 +36,13 @@ ENTRY (__rawmemchr)
 	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
 	ld	r12,0(r8)     /* Load doubleword from memory.  */
 	cmpb	r5,r12,r4     /* Compare each byte against c byte.  */
+#ifdef __LITTLE_ENDIAN__
+	srd	r5,r5,r6
+	sld	r5,r5,r6
+#else
 	sld	r5,r5,r6      /* Move left to discard ignored bits.  */
 	srd	r5,r5,r6      /* Bring the bits back as zeros.  */
+#endif
 	cmpdi	cr7,r5,0      /* If r5 == 0, no c bytes have been found.  */
 	bne	cr7,L(done)
 
@@ -91,8 +96,14 @@ L(loop):
 	   doubleword from the string.  Use that fact to find out what is
 	   the position of the byte inside the string.  */
 L(done):
+#ifdef __LITTLE_ENDIAN__
+	addi    r0,r5,-1
+	andc    r0,r0,r5
+	popcntd	r0,r0	      /* Count trailing zeros.  */
+#else
 	cntlzd	r0,r5	      /* Count leading zeros before the match.  */
-	srdi	r0,r0,3	      /* Convert leading zeroes to bytes.  */
+#endif
+	srdi	r0,r0,3	      /* Convert leading zeros to bytes.  */
 	add	r3,r8,r0      /* Return address of the matching char.  */
 	blr
 END (__rawmemchr)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/strchr.S b/libc/sysdeps/powerpc/powerpc64/power7/strchr.S
index 3ffe7a188..4679a158f 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/strchr.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/strchr.S
@@ -35,8 +35,8 @@ ENTRY (strchr)
 	beq	cr7,L(null_match)
 
 	/* Replicate byte to doubleword.  */
-	rlwimi	r4,r4,8,16,23
-	rlwimi	r4,r4,16,0,15
+	insrdi	r4,r4,8,48
+	insrdi	r4,r4,16,32
 	insrdi  r4,r4,32,0
 
 	/* Now r4 has a doubleword of c bytes and r0 has
@@ -47,11 +47,17 @@ ENTRY (strchr)
 
 	/* Move the doublewords left and right to discard the bits that are
 	   not part of the string and bring them back as zeros.  */
-
+#ifdef __LITTLE_ENDIAN__
+	srd	r10,r10,r6
+	srd	r11,r11,r6
+	sld	r10,r10,r6
+	sld	r11,r11,r6
+#else
 	sld	r10,r10,r6
 	sld	r11,r11,r6
 	srd	r10,r10,r6
 	srd	r11,r11,r6
+#endif
 	or	r5,r10,r11    /* OR the results to speed things up.  */
 	cmpdi	cr7,r5,0      /* If r5 == 0, no c or null bytes
 				 have been found.  */
@@ -108,15 +114,24 @@ L(loop):
 	mr	r11,r7
 	addi	r8,r8,8
 
-	/* r5 has the output of the cmpb instruction, that is, it contains
+	/* r10/r11 have the output of the cmpb instructions, that is,
 	   0xff in the same position as the c/null byte in the original
 	   doubleword from the string.  Use that to calculate the pointer.  */
 L(done):
-	cntlzd	r4,r10	      /* Count leading zeroes before c matches.  */
-	cntlzd	r0,r11	      /* Count leading zeroes before null matches.  */
-	cmpld	cr7,r4,r0
+#ifdef __LITTLE_ENDIAN__
+	addi    r3,r10,-1
+	andc    r3,r3,r10
+	popcntd	r0,r3
+	addi    r4,r11,-1
+	andc    r4,r4,r11
+	cmpld	cr7,r3,r4
 	bgt	cr7,L(no_match)
-	srdi	r0,r4,3	      /* Convert leading zeroes to bytes.  */
+#else
+	cntlzd	r0,r10	      /* Count leading zeros before c matches.  */
+	cmpld	cr7,r11,r10
+	bgt	cr7,L(no_match)
+#endif
+	srdi	r0,r0,3	      /* Convert leading zeros to bytes.  */
 	add	r3,r8,r0      /* Return address of the matching c byte
 				 or null in case c was not found.  */
 	blr
@@ -135,9 +150,13 @@ L(null_match):
 
 	/* Move the doublewords left and right to discard the bits that are
 	   not part of the string and bring them back as zeros.  */
-
+#ifdef __LITTLE_ENDIAN__
+	srd	r5,r5,r6
+	sld	r5,r5,r6
+#else
 	sld	r5,r5,r6
 	srd	r5,r5,r6
+#endif
 	cmpdi	cr7,r5,0      /* If r10 == 0, no c or null bytes
 				 have been found.  */
 	bne	cr7,L(done_null)
@@ -192,7 +211,13 @@ L(loop_null):
 	   0xff in the same position as the null byte in the original
 	   doubleword from the string.  Use that to calculate the pointer.  */
 L(done_null):
+#ifdef __LITTLE_ENDIAN__
+	addi    r0,r5,-1
+	andc    r0,r0,r5
+	popcntd	r0,r0
+#else
 	cntlzd	r0,r5	      /* Count leading zeros before the match.  */
+#endif
 	srdi	r0,r0,3	      /* Convert leading zeros to bytes.  */
 	add	r3,r8,r0      /* Return address of the matching null byte.  */
 	blr
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/strchrnul.S b/libc/sysdeps/powerpc/powerpc64/power7/strchrnul.S
index 9dbc51b0d..df457525e 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/strchrnul.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/strchrnul.S
@@ -27,8 +27,8 @@ ENTRY (__strchrnul)
 	clrrdi	r8,r3,3	      /* Align the address to doubleword boundary.  */
 
 	/* Replicate byte to doubleword.  */
-	rlwimi	r4,r4,8,16,23
-	rlwimi	r4,r4,16,0,15
+	insrdi	r4,r4,8,48
+	insrdi	r4,r4,16,32
 	insrdi	r4,r4,32,0
 
 	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
@@ -44,10 +44,17 @@ ENTRY (__strchrnul)
 
 	/* Move the doublewords left and right to discard the bits that are
 	   not part of the string and to bring them back as zeros.  */
+#ifdef __LITTLE_ENDIAN__
+	srd	r10,r10,r6
+	srd	r9,r9,r6
+	sld	r10,r10,r6
+	sld	r9,r9,r6
+#else
 	sld	r10,r10,r6
 	sld	r9,r9,r6
 	srd	r10,r10,r6
 	srd	r9,r9,r6
+#endif
 	or	r5,r9,r10     /* OR the results to speed things up.  */
 	cmpdi	cr7,r5,0      /* If r5 == 0, no c or null bytes
 				 have been found.  */
@@ -97,7 +104,7 @@ L(loop):
 	bne	cr6,L(done)
 
 	/* The c/null byte must be in the second doubleword.  Adjust the
-	   address again and move the result of cmpb to r10 so we can calculate
+	   address again and move the result of cmpb to r5 so we can calculate
 	   the pointer.  */
 	mr	r5,r10
 	addi	r8,r8,8
@@ -106,7 +113,13 @@ L(loop):
 	   0xff in the same position as the c/null byte in the original
 	   doubleword from the string.  Use that to calculate the pointer.  */
 L(done):
+#ifdef __LITTLE_ENDIAN__
+	addi    r0,r5,-1
+	andc    r0,r0,r5
+	popcntd	r0,r0
+#else
 	cntlzd	r0,r5	      /* Count leading zeros before the match.  */
+#endif
 	srdi	r0,r0,3	      /* Convert leading zeros to bytes.  */
 	add	r3,r8,r0      /* Return address of matching c/null byte.  */
 	blr
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/strlen.S b/libc/sysdeps/powerpc/powerpc64/power7/strlen.S
index 343216952..807ef1082 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/strlen.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/strlen.S
@@ -30,7 +30,11 @@ ENTRY (strlen)
 				 with cmpb.  */
 	li	r5,-1	      /* MASK = 0xffffffffffffffff.  */
 	ld	r12,0(r4)     /* Load doubleword from memory.  */
+#ifdef __LITTLE_ENDIAN__
+	sld	r5,r5,r6
+#else
 	srd	r5,r5,r6      /* MASK = MASK >> padding.  */
+#endif
 	orc	r9,r12,r5     /* Mask bits that are not part of the string.  */
 	cmpb	r10,r9,r0     /* Check for null bytes in DWORD1.  */
 	cmpdi	cr7,r10,0     /* If r10 == 0, no null's have been found.  */
@@ -48,9 +52,6 @@ ENTRY (strlen)
 	cmpb	r10,r12,r0
 	cmpdi	cr7,r10,0
 	bne	cr7,L(done)
-	b	L(loop)	      /* We branch here (rather than falling through)
-				 to skip the nops due to heavy alignment
-				 of the loop below.  */
 
 	/* Main loop to look for the end of the string.  Since it's a
 	   small loop (< 8 instructions), align it to 32-bytes.  */
@@ -87,9 +88,15 @@ L(loop):
 	   0xff in the same position as the null byte in the original
 	   doubleword from the string.  Use that to calculate the length.  */
 L(done):
-	cntlzd	r0,r10	      /* Count leading zeroes before the match.  */
+#ifdef __LITTLE_ENDIAN__
+	addi	r9, r10, -1   /* Form a mask from trailing zeros.  */
+	andc	r9, r9, r10
+	popcntd r0, r9	      /* Count the bits in the mask.  */
+#else
+	cntlzd	r0,r10	      /* Count leading zeros before the match.  */
+#endif
 	subf	r5,r3,r4
-	srdi	r0,r0,3	      /* Convert leading zeroes to bytes.  */
+	srdi	r0,r0,3	      /* Convert leading/trailing zeros to bytes.  */
 	add	r3,r5,r0      /* Compute final length.  */
 	blr
 END (strlen)
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/strncmp.S b/libc/sysdeps/powerpc/powerpc64/power7/strncmp.S
index 77ecad5ab..e618b010b 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/strncmp.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/strncmp.S
@@ -27,7 +27,7 @@
 EALIGN (strncmp,5,0)
 	CALL_MCOUNT 3
 
-#define rTMP	r0
+#define rTMP2	r0
 #define rRTN	r3
 #define rSTR1	r3	/* first string arg */
 #define rSTR2	r4	/* second string arg */
@@ -40,6 +40,7 @@ EALIGN (strncmp,5,0)
 #define r7F7F	r9	/* constant 0x7f7f7f7f7f7f7f7f */
 #define rNEG	r10	/* ~(word in s1 | 0x7f7f7f7f7f7f7f7f) */
 #define rBITDIF	r11	/* bits that differ in s1 & s2 words */
+#define rTMP	r12
 
 	dcbt	0,rSTR1
 	nop
@@ -83,12 +84,57 @@ L(g1):	add	rTMP,rFEFE,rWORD1
    we don't compare two strings as different because of gunk beyond
    the end of the strings...  */
 
+#ifdef __LITTLE_ENDIAN__
+L(endstring):
+	addi    rTMP2, rTMP, -1
+	beq	cr1, L(equal)
+	andc    rTMP2, rTMP2, rTMP
+	rldimi	rTMP2, rTMP2, 1, 0
+	and	rWORD2, rWORD2, rTMP2	/* Mask off gunk.  */
+	and	rWORD1, rWORD1, rTMP2
+	cmpd	cr1, rWORD1, rWORD2
+	beq	cr1, L(equal)
+	cmpb	rBITDIF, rWORD1, rWORD2	/* 0xff on equal bytes.  */
+	addi	rNEG, rBITDIF, 1
+	orc	rNEG, rNEG, rBITDIF	/* 0's below LS differing byte.  */
+	sldi	rNEG, rNEG, 8		/* 1's above LS differing byte.  */
+	andc	rWORD1, rWORD1, rNEG	/* mask off MS bytes.  */
+	andc	rWORD2, rWORD2, rNEG
+	xor.	rBITDIF, rWORD1, rWORD2
+	sub	rRTN, rWORD1, rWORD2
+	blt	L(highbit)
+	sradi	rRTN, rRTN, 63		/* must return an int.  */
+	ori	rRTN, rRTN, 1
+	blr
+L(equal):
+	li	rRTN, 0
+	blr
+
+L(different):
+	ld	rWORD1, -8(rSTR1)
+	cmpb	rBITDIF, rWORD1, rWORD2	/* 0xff on equal bytes.  */
+	addi	rNEG, rBITDIF, 1
+	orc	rNEG, rNEG, rBITDIF	/* 0's below LS differing byte.  */
+	sldi	rNEG, rNEG, 8		/* 1's above LS differing byte.  */
+	andc	rWORD1, rWORD1, rNEG	/* mask off MS bytes.  */
+	andc	rWORD2, rWORD2, rNEG
+	xor.	rBITDIF, rWORD1, rWORD2
+	sub	rRTN, rWORD1, rWORD2
+	blt	L(highbit)
+	sradi	rRTN, rRTN, 63
+	ori	rRTN, rRTN, 1
+	blr
+L(highbit):
+	sradi	rRTN, rWORD2, 63
+	ori	rRTN, rRTN, 1
+	blr
+
+#else
 L(endstring):
 	and	rTMP,r7F7F,rWORD1
 	beq	cr1,L(equal)
 	add	rTMP,rTMP,r7F7F
 	xor.	rBITDIF,rWORD1,rWORD2
-
 	andc	rNEG,rNEG,rTMP
 	blt	L(highbit)
 	cntlzd	rBITDIF,rBITDIF
@@ -97,7 +143,7 @@ L(endstring):
 	cmpd	cr1,rNEG,rBITDIF
 	sub	rRTN,rWORD1,rWORD2
 	blt	cr1,L(equal)
-	sradi	rRTN,rRTN,63
+	sradi	rRTN,rRTN,63		/* must return an int.  */
 	ori	rRTN,rRTN,1
 	blr
 L(equal):
@@ -105,7 +151,7 @@ L(equal):
 	blr
 
 L(different):
-	ldu	rWORD1,-8(rSTR1)
+	ld	rWORD1,-8(rSTR1)
 	xor.	rBITDIF,rWORD1,rWORD2
 	sub	rRTN,rWORD1,rWORD2
 	blt	L(highbit)
@@ -113,11 +159,10 @@ L(different):
 	ori	rRTN,rRTN,1
 	blr
 L(highbit):
-	srdi	rWORD2,rWORD2,56
-	srdi	rWORD1,rWORD1,56
-	sub	rRTN,rWORD1,rWORD2
+	sradi	rRTN,rWORD2,63
+	ori	rRTN,rRTN,1
 	blr
-
+#endif
 
 /* Oh well.  In this case, we just do a byte-by-byte comparison.  */
 	.align	4
diff --git a/libc/sysdeps/powerpc/powerpc64/power7/strnlen.S b/libc/sysdeps/powerpc/powerpc64/power7/strnlen.S
index 37c7dbfe8..51591069d 100644
--- a/libc/sysdeps/powerpc/powerpc64/power7/strnlen.S
+++ b/libc/sysdeps/powerpc/powerpc64/power7/strnlen.S
@@ -24,33 +24,29 @@
 ENTRY (__strnlen)
 	CALL_MCOUNT 2
 	dcbt	0,r3
-	clrrdi  r8,r3,3
+	clrrdi	r8,r3,3
 	add	r7,r3,r4      /* Calculate the last acceptable address.  */
 	cmpldi	r4,32
 	li	r0,0	      /* Doubleword with null chars.  */
+	addi	r7,r7,-1
+
 	/* If we have less than 33 bytes to search, skip to a faster code.  */
 	ble	L(small_range)
 
-	cmpld	cr7,r3,r7    /* Is the address equal or less than r3?  If
-				it's equal or less, it means size is either 0
-				or a negative number.  */
-	ble	cr7,L(proceed)
-
-	li	r7,-1	      /* Make r11 the biggest if r4 <= 0.  */
-L(proceed):
 	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
 	ld	r12,0(r8)     /* Load doubleword from memory.  */
 	cmpb	r10,r12,r0    /* Check for null bytes in DWORD1.  */
+#ifdef __LITTLE_ENDIAN__
+	srd	r10,r10,r6
+	sld	r10,r10,r6
+#else
 	sld	r10,r10,r6
 	srd	r10,r10,r6
+#endif
 	cmpldi	cr7,r10,0     /* If r10 == 0, no null's have been found.  */
 	bne	cr7,L(done)
 
-	/* Are we done already?  */
-	addi	r9,r8,8
-	cmpld	cr6,r9,r7
-	bge	cr6,L(end_max)
-
+	clrrdi	r7,r7,3       /* Address of last doubleword.  */
 	mtcrf   0x01,r8
 	/* Are we now aligned to a quadword boundary?  If so, skip to
 	   the main loop.  Otherwise, go through the alignment code.  */
@@ -63,17 +59,18 @@ L(proceed):
 	cmpldi	cr7,r10,0
 	bne	cr7,L(done)
 
-	/* Are we done already?  */
-	addi	r9,r8,8
-	cmpld	cr6,r9,r7
-	bge	cr6,L(end_max)
-
 L(loop_setup):
-	sub	r5,r7,r9
+	/* The last dword we want to read in the loop below is the one
+	   containing the last byte of the string, ie. the dword at
+	   (s + size - 1) & ~7, or r7.  The first dword read is at
+	   r8 + 8, we read 2 * cnt dwords, so the last dword read will
+	   be at r8 + 8 + 16 * cnt - 8.  Solving for cnt gives
+	   cnt = (r7 - r8) / 16  */
+	sub	r5,r7,r8
 	srdi	r6,r5,4	      /* Number of loop iterations.  */
 	mtctr	r6	      /* Setup the counter.  */
-	b	L(loop)
-	/* Main loop to look for the null byte backwards in the string.  Since
+
+	/* Main loop to look for the null byte in the string.  Since
 	   it's a small loop (< 8 instructions), align it to 32-bytes.  */
 	.p2align  5
 L(loop):
@@ -89,15 +86,18 @@ L(loop):
 	cmpldi	cr7,r5,0
 	bne	cr7,L(found)
 	bdnz	L(loop)
-	/* We're here because the counter reached 0, and that means we
-	   didn't have any matches for null in the whole range.  Just return
-	   the original size.  */
-	addi	r9,r8,8
-	cmpld	cr6,r9,r7
-	blt	cr6,L(loop_small)
+
+	/* We may have one more dword to read.  */
+	cmpld	cr6,r8,r7
+	beq	cr6,L(end_max)
+
+	ldu	r12,8(r8)
+	cmpb	r10,r12,r0
+	cmpldi	cr6,r10,0
+	bne	cr6,L(done)
 
 L(end_max):
-	sub	r3,r7,r3
+	mr	r3,r4
 	blr
 
 	/* OK, one (or both) of the doublewords contains a null byte.  Check
@@ -119,52 +119,59 @@ L(found):
 	/* r10 has the output of the cmpb instruction, that is, it contains
 	   0xff in the same position as the null byte in the original
 	   doubleword from the string.  Use that to calculate the length.
-	   We need to make sure the null char is *before* the start of the
-	   range (since we're going backwards).  */
+	   We need to make sure the null char is *before* the end of the
+	   range.  */
 L(done):
-	cntlzd	r0,r10	      /* Count leading zeroes before the match.  */
-	srdi	r0,r0,3	      /* Convert leading zeroes to bytes.  */
-	add	r9,r8,r0
-	sub	r6,r9,r3      /* Length until the match.  */
-	cmpld	r9,r7
-	bgt	L(end_max)
-	mr	r3,r6
-	blr
-
-	.align	4
-L(zero):
-	li	r3,0
+#ifdef __LITTLE_ENDIAN__
+	addi	r0,r10,-1
+	andc	r0,r0,r10
+	popcntd	r0,r0
+#else
+	cntlzd	r0,r10	      /* Count leading zeros before the match.  */
+#endif
+	sub	r3,r8,r3
+	srdi	r0,r0,3	      /* Convert leading/trailing zeros to bytes.  */
+	add	r3,r3,r0      /* Length until the match.  */
+	cmpld	r3,r4
+	blelr
+	mr	r3,r4
 	blr
 
 /* Deals with size <= 32.  */
 	.align	4
 L(small_range):
 	cmpldi	r4,0
-	beq	L(zero)
+	beq	L(end_max)
+
+	clrrdi	r7,r7,3       /* Address of last doubleword.  */
 
 	rlwinm	r6,r3,3,26,28 /* Calculate padding.  */
-	ld	r12,0(r8)     /* Load word from memory.  */
+	ld	r12,0(r8)     /* Load doubleword from memory.  */
 	cmpb	r10,r12,r0    /* Check for null bytes in DWORD1.  */
+#ifdef __LITTLE_ENDIAN__
+	srd	r10,r10,r6
+	sld	r10,r10,r6
+#else
 	sld	r10,r10,r6
 	srd	r10,r10,r6
+#endif
 	cmpldi	cr7,r10,0
 	bne	cr7,L(done)
 
-	addi    r9,r8,8
-	cmpld	r9,r7
-	bge	L(end_max)
-	b	L(loop_small)
+	cmpld	r8,r7
+	beq	L(end_max)
 
 	.p2align  5
 L(loop_small):
 	ldu	r12,8(r8)
 	cmpb	r10,r12,r0
-	addi	r9,r8,8
 	cmpldi	cr6,r10,0
 	bne	cr6,L(done)
-	cmpld	r9,r7
-	bge	L(end_max)
-	b	L(loop_small)
+	cmpld	r8,r7
+	bne	L(loop_small)
+	mr	r3,r4
+	blr
+
 END (__strnlen)
 weak_alias (__strnlen, strnlen)
 libc_hidden_builtin_def (strnlen)