20060303a

1 files changed, 170 insertions, 0 deletions

diff --git a/libkern/memmove-power.s b/libkern/memmove-power.s
new file mode 100644
index 00000000..1b27e5d6
--- /dev/null
+++ b/libkern/memmove-power.s
@@ -0,0 +1,170 @@
+#define	BDNZ	BC	16,0,
+	TEXT	memcpy(SB), $0
+	BR	move
+
+	TEXT	memmove(SB), $0
+move:
+
+/*
+ * performance:
+ * (tba)
+ */
+
+	MOVW	R3, s1+0(FP)
+	MOVW	n+8(FP), R9		/* R9 is count */
+	MOVW	R3, R10			/* R10 is to-pointer */
+	CMP	R9, $0
+	BEQ	ret
+	BLT	trap
+	MOVW	s2+4(FP), R11		/* R11 is from-pointer */
+
+/*
+ * if no more than 16 bytes, just use one lsw/stsw
+ */
+	CMP	R9, $16
+	BLE	fout
+
+	ADD	R9,R11, R13		/* R13 is end from-pointer */
+	ADD	R9,R10, R12		/* R12 is end to-pointer */
+
+/*
+ * easiest test is copy backwards if
+ * destination string has higher mem address
+ */
+	CMPU	R10, R11
+	BGT	back
+
+/*
+ * test if both pointers
+ * are similarly word aligned
+ */
+	XOR	R10,R11, R7
+	ANDCC	$3,R7
+	BNE	fbad
+
+/*
+ * move a few bytes to align pointers
+ */
+	ANDCC	$3,R10,R7
+	BEQ	f2
+	SUBC	R7, $4, R7
+	SUB	R7, R9
+	MOVW	R7, XER
+	LSW	(R11), R16
+	ADD	R7, R11
+	STSW	R16, (R10)
+	ADD	R7, R10
+
+/*
+ * turn R14 into doubleword count
+ * copy 16 bytes at a time while there's room.
+ */
+f2:
+	SRAWCC	$4, R9, R14
+	BLE	fout
+	MOVW	R14, CTR
+	SUB	$4, R11
+	SUB	$4, R10
+f3:
+	MOVWU	4(R11), R16
+	MOVWU	R16, 4(R10)
+	MOVWU	4(R11), R17
+	MOVWU	R17, 4(R10)
+	MOVWU	4(R11), R16
+	MOVWU	R16, 4(R10)
+	MOVWU	4(R11), R17
+	MOVWU	R17, 4(R10)
+	BDNZ	f3
+	RLWNMCC	$0, R9, $15, R9	/* residue */
+	BEQ	ret
+	ADD	$4, R11
+	ADD	$4, R10
+
+/*
+ * move up to 16 bytes through R16 .. R19; aligned and unaligned
+ */
+fout:
+	MOVW	R9, XER
+	LSW	(R11), R16
+	STSW	R16, (R10)
+	BR	ret
+
+/*
+ * loop for unaligned copy, then copy up to 15 remaining bytes
+ */
+fbad:
+	SRAWCC	$4, R9, R14
+	BLE	f6
+	MOVW	R14, CTR
+f5:
+	LSW	(R11), $16, R16
+	ADD	$16, R11
+	STSW	R16, $16, (R10)
+	ADD	$16, R10
+	BDNZ	f5
+	RLWNMCC	$0, R9, $15, R9	/* residue */
+	BEQ	ret
+f6:
+	MOVW	R9, XER
+	LSW	(R11), R16
+	STSW	R16, (R10)
+	BR	ret
+
+/*
+ * whole thing repeated for backwards
+ */
+back:
+	CMP	R9, $4
+	BLT	bout
+
+	XOR	R12,R13, R7
+	ANDCC	$3,R7
+	BNE	bout
+b1:
+	ANDCC	$3,R13, R7
+	BEQ	b2
+	MOVBZU	-1(R13), R16
+	MOVBZU	R16, -1(R12)
+	SUB	$1, R9
+	BR	b1
+b2:
+	SRAWCC	$4, R9, R14
+	BLE	b4
+	MOVW	R14, CTR
+b3:
+	MOVWU	-4(R13), R16
+	MOVWU	R16, -4(R12)
+	MOVWU	-4(R13), R17
+	MOVWU	R17, -4(R12)
+	MOVWU	-4(R13), R16
+	MOVWU	R16, -4(R12)
+	MOVWU	-4(R13), R17
+	MOVWU	R17, -4(R12)
+	BDNZ	b3
+	RLWNMCC	$0, R9, $15, R9	/* residue */
+	BEQ	ret
+b4:
+	SRAWCC	$2, R9, R14
+	BLE	bout
+	MOVW	R14, CTR
+b5:
+	MOVWU	-4(R13), R16
+	MOVWU	R16, -4(R12)
+	BDNZ	b5
+	RLWNMCC	$0, R9, $3, R9	/* residue */
+	BEQ	ret
+
+bout:
+	CMPU	R13, R11
+	BLE	ret
+	MOVBZU	-1(R13), R16
+	MOVBZU	R16, -1(R12)
+	BR	bout
+
+trap:
+/*	MOVW	$0, R0	*/
+	MOVW	R0, 0(R0)
+
+ret:
+	MOVW	s1+0(FP), R3
+	RETURN