From 1c9d43c2c1af3471c1ef064c3598edf131932b34 Mon Sep 17 00:00:00 2001
From: Xin Qi <xqi@codeaurora.org>
Date: Wed, 12 Feb 2014 16:39:04 -0800
Subject: libc: krait: Fix memmove half-word atomicity

In tailing case of less than 7, if even number of bytes are left, then
use half word load/store to make sure the whole routine is half-word
atomic

Change-Id: I54948ac050ff866986a7e144d4240652981ea87a
---
 libc/arch-arm/krait/bionic/memmove.S | 55 +++++++++++++++---------------------
 1 file changed, 22 insertions(+), 33 deletions(-)

(limited to 'libc/arch-arm/krait/bionic/memmove.S')

diff --git a/libc/arch-arm/krait/bionic/memmove.S b/libc/arch-arm/krait/bionic/memmove.S
index cfa06cef4..b7b77ce7e 100644
--- a/libc/arch-arm/krait/bionic/memmove.S
+++ b/libc/arch-arm/krait/bionic/memmove.S
@@ -94,28 +94,15 @@ _memmove_words:
 	cmpge	r12, r0
 	it	le
 	ble	memcpy
-	cmp	r2, #4
-	it	le
-	ble	.Lneon_b2f_smallcopy_loop
+	cmp	r2, #63
+	ble	.Lneon_b2f_smallcopy
 	push	{r0, lr}
 	add	r0, r0, r2
 	add	r1, r1, r2
-	cmp	r2, #64
-	it	ge
-	bge	.Lneon_b2f_copy_64
-	cmp	r2, #32
-	it	ge
-	bge	.Lneon_b2f_copy_32
-	cmp	r2, #8
-	it	ge
-	bge	.Lneon_b2f_copy_8
-	b	.Lneon_b2f_copy_1
-.Lneon_b2f_copy_64:
 	mov	r12, r2, lsr #6
-	add	r1, r1, #32
 	add	r0, r0, #32
+	add	r1, r1, #32
 	cmp	r12, #PLDTHRESH
-	it	le
 	ble	.Lneon_b2f_copy_64_loop_nopld
 	sub	r12, #PLDOFFS
 	sub	lr, r1, #(PLDOFFS)*PLDSIZE
@@ -129,7 +116,6 @@ _memmove_words:
 	subs	r12, r12, #1
 	vst1.32	{q0, q1}, [r0]!
 	vst1.32	{q2, q3}, [r0]
-	it	ne
 	bne	.Lneon_b2f_copy_64_loop_outer
 	mov	r12, #PLDOFFS
 .Lneon_b2f_copy_64_loop_nopld:
@@ -140,15 +126,12 @@ _memmove_words:
 	subs	r12, r12, #1
 	vst1.32	{q8, q9}, [r0]!
 	vst1.32	{q10, q11}, [r0]
-	it	ne
 	bne	.Lneon_b2f_copy_64_loop_nopld
 	ands	r2, r2, #0x3f
-	it	eq
 	beq	.Lneon_memmove_done
 	sub	r1, r1, #32
 	sub	r0, r0, #32
 	cmp	r2, #32
-	it	lt
 	blt	.Lneon_b2f_copy_8
 .Lneon_b2f_copy_32:
 	sub	r1, r1, #32
@@ -156,12 +139,9 @@ _memmove_words:
 	vld1.32	{q0, q1}, [r1]
 	vst1.32	{q0, q1}, [r0]
 	ands	r2, r2, #0x1f
-	it	eq
 	beq	.Lneon_memmove_done
-.Lneon_b2f_copy_finish:
 .Lneon_b2f_copy_8:
 	movs	r12, r2, lsr #0x3
-	it	eq
 	beq	.Lneon_b2f_copy_1
 .Lneon_b2f_copy_8_loop:
 	sub	r1, r1, #8
@@ -169,30 +149,39 @@ _memmove_words:
 	vld1.32	{d0}, [r1]
 	subs	r12, r12, #1
 	vst1.32	{d0}, [r0]
-	it	ne
 	bne	.Lneon_b2f_copy_8_loop
-.Lneon_b2f_copy_1:
 	ands	r2, r2, #0x7
-	it	eq
 	beq	.Lneon_memmove_done
+.Lneon_b2f_copy_1:
 	sub	r1, r1, r2
 	sub	r0, r0, r2
-.Lneon_b2f_copy_1_loop:
+	ands	r12, r2, #1
+	beq	.Lneon_b2f_copy_halfword_loop
 	subs	r2, r2, #1
 	ldrb	r3, [r1, r2]
 	strb	r3, [r0, r2]
-	it	ne
-	bne	.Lneon_b2f_copy_1_loop
+	beq	.Lneon_memmove_done
+.Lneon_b2f_copy_halfword_loop:
+	subs	r2, r2, #2
+	ldrh	r3, [r1, r2]
+	strh	r3, [r0, r2]
+	bne	.Lneon_b2f_copy_halfword_loop
 .Lneon_memmove_done:
 	pop	{r0, pc}
-.Lneon_b2f_smallcopy_loop:
+.Lneon_b2f_smallcopy:
+	ands	r12, r2, #1
+	beq	.Lneon_b2f_halfword_small_loop
 	subs	r2, r2, #1
 	ldrb	r3, [r1, r2]
 	strb	r3, [r0, r2]
-	it	ne
-	bne	.Lneon_b2f_smallcopy_loop
+	it	eq
+	bxeq	lr
+.Lneon_b2f_halfword_small_loop:
+	subs	r2, r2, #2
+	ldrh	r3, [r1, r2]
+	strh	r3, [r0, r2]
+	bne	.Lneon_b2f_halfword_small_loop
 	bx	lr
         .cfi_endproc
 END(memmove)
 
-
-- 
cgit v1.2.3