sh: Add SH-5 optimized memcpy()/memset()/strcpy()/strlen().

Adopted from the uClibc optimized string versions.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

arch/sh/lib64/memset.S

@@ -0,0 +1,91 @@
+/* Cloned and hacked for uClibc by Paul Mundt, December 2003 */
+/* Modified by SuperH, Inc. September 2003 */
+!
+! Fast SH memset
+!
+! by Toshiyasu Morita (tm@netcom.com)
+!
+! SH5 code by J"orn Rennecke (joern.rennecke@superh.com)
+! Copyright 2002 SuperH Ltd.
+!
+
+#if __BYTE_ORDER == __LITTLE_ENDIAN
+#define SHHI shlld
+#define SHLO shlrd
+#else
+#define SHHI shlrd
+#define SHLO shlld
+#endif
+
+	.section .text..SHmedia32,"ax"
+	.globl	memset
+	.type	memset, @function
+
+	.align 5
+
+memset:
+	pta/l multiquad, tr0
+	andi r2, 7, r22
+	ptabs r18, tr2
+	mshflo.b r3,r3,r3
+	add r4, r22, r23
+	mperm.w r3, r63, r3	// Fill pattern now in every byte of r3
+
+	movi 8, r9
+	bgtu/u r23, r9, tr0 // multiquad
+
+	beqi/u r4, 0, tr2       // Return with size 0 - ensures no mem accesses
+	ldlo.q r2, 0, r7
+	shlli r4, 2, r4
+	movi -1, r8
+	SHHI r8, r4, r8
+	SHHI r8, r4, r8
+	mcmv r7, r8, r3
+	stlo.q r2, 0, r3
+	blink tr2, r63
+
+multiquad:
+	pta/l lastquad, tr0
+	stlo.q r2, 0, r3
+	shlri r23, 3, r24
+	add r2, r4, r5
+	beqi/u r24, 1, tr0 // lastquad
+	pta/l loop, tr1
+	sub r2, r22, r25
+	andi r5, -8, r20   // calculate end address and
+	addi r20, -7*8, r8 // loop end address; This might overflow, so we need
+	                   // to use a different test before we start the loop
+	bge/u r24, r9, tr1 // loop
+	st.q r25, 8, r3
+	st.q r20, -8, r3
+	shlri r24, 1, r24
+	beqi/u r24, 1, tr0 // lastquad
+	st.q r25, 16, r3
+	st.q r20, -16, r3
+	beqi/u r24, 2, tr0 // lastquad
+	st.q r25, 24, r3
+	st.q r20, -24, r3
+lastquad:
+	sthi.q r5, -1, r3
+	blink tr2,r63
+
+loop:
+!!!	alloco r25, 32	// QQQ comment out for short-term fix to SHUK #3895.
+			// QQQ commenting out is locically correct, but sub-optimal
+			// QQQ Sean McGoogan - 4th April 2003.
+	st.q r25, 8, r3
+	st.q r25, 16, r3
+	st.q r25, 24, r3
+	st.q r25, 32, r3
+	addi r25, 32, r25
+	bgeu/l r8, r25, tr1 // loop
+
+	st.q r20, -40, r3
+	st.q r20, -32, r3
+	st.q r20, -24, r3
+	st.q r20, -16, r3
+	st.q r20, -8, r3
+	sthi.q r5, -1, r3
+	blink tr2,r63
+
+	.size	memset,.-memset