[PATCH 2/3] sparc: assembly version of memmove for ultra1+

[Patrick McGehearty]

I'll clean up the nits and double check for any other missing delay slot 
spaces.
I expect I'll have it ready for resubmission later today.
- patrick

On 9/27/2017 3:40 PM, Sam Ravnborg wrote:
> Hi Patrick.
>
> Nitpick below.
>
> 	Sam
>
>> +
>> +.Ls2alg:
>> +	lduh	[%o1], %o3	/* know src is 2 byte aligned  */
>> +	inc	2, %o1
>> +	srl	%o3, 8, %o4
>> +	stb	%o4, [%o0]	/* have to do bytes,  */
>> +	stb	%o3, [%o0 + 1]	/* don't know dst alingment  */
>> +	inc	2, %o0
>> +	dec	2, %o2
>> +
>> +.Laldst:
>> +	andcc	%o0, 3, %o5	/* align the destination address  */
> .Lald:	bz,pn	%icc, .Lw4cp
> Label on own line would make patches more readable.
> But src looks OK.
>
>> +	 cmp	%o5, 2
>> +	bz,pn	%icc, .Lw2cp
>> +	 cmp	%o5, 3
>> +.Lw3cp:
>> +	lduw	[%o1], %o4
>> +	inc	4, %o1
>> +	srl	%o4, 24, %o5
>> +	stb	%o5, [%o0]
>> +	bne,pt	%icc, .Lw1cp
>> +	 inc	%o0
>> +	dec	1, %o2
>> +	andn	%o2, 3, %o3	/* i3 is aligned word count  */
>> +	dec	4, %o3		/* avoid reading beyond tail of src  */
>> +	sub	%o1, %o0, %o1	/* i1 gets the difference  */
>> +
>> +1:	sll	%o4, 8, %g1	/* save residual bytes  */
>> +	lduw	[%o1+%o0], %o4
>> +	deccc	4, %o3
>> +	srl	%o4, 24, %o5	/* merge with residual  */
>> +	or	%o5, %g1, %g1
>> +	st	%g1, [%o0]
>> +	bnz,pt	%XCC, 1b
>> +	 inc	4, %o0
>> +	sub	%o1, 3, %o1	/* used one byte of last word read  */
>> +	and	%o2, 3, %o2
>> +	b	7f
>> +	 inc	4, %o2
>> +
>> +.Lw1cp:
>> +	srl	%o4, 8, %o5
>> +	sth	%o5, [%o0]
>> +	inc	2, %o0
>> +	dec	3, %o2
>> +	andn	%o2, 3, %o3
>> +	dec	4, %o3		/* avoid reading beyond tail of src  */
>> +	sub	%o1, %o0, %o1	/* i1 gets the difference  */
>> +
>> +2:	sll	%o4, 24, %g1	/* save residual bytes  */
>> +	lduw	[%o1+%o0], %o4
>> +	deccc	4, %o3
>> +	srl	%o4, 8, %o5	/* merge with residual  */
>> +	or	%o5, %g1, %g1
>> +	st	%g1, [%o0]
>> +	bnz,pt	%XCC, 2b
>> +	 inc	4, %o0
>> +	sub	%o1, 1, %o1	/* used three bytes of last word read  */
>> +	and	%o2, 3, %o2
>> +	b	7f
>> +	inc	4, %o2
> Delay slot - indent instruction with one space.
>
>> +
>> +.Lw2cp:
>> +	lduw	[%o1], %o4
>> +	inc	4, %o1
>> +	srl	%o4, 16, %o5
>> +	sth	%o5, [%o0]
>> +	inc	2, %o0
>> +	dec	2, %o2
>> +	andn	%o2, 3, %o3	/* i3 is aligned word count  */
>> +	dec	4, %o3		/* avoid reading beyond tail of src  */
>> +	sub	%o1, %o0, %o1	/* i1 gets the difference  */
>> +
>> +3:	sll	%o4, 16, %g1	/* save residual bytes  */
>> +	lduw	[%o1+%o0], %o4
>> +	deccc	4, %o3
>> +	srl	%o4, 16, %o5	/* merge with residual  */
>> +	or	%o5, %g1, %g1
>> +	st	%g1, [%o0]
>> +	bnz,pt	%XCC, 3b
>> +	 inc	4, %o0
>> +	sub	%o1, 2, %o1	/* used two bytes of last word read  */
>> +	and	%o2, 3, %o2
>> +	b	7f
>> +	 inc	4, %o2
>> +
>> +.Lw4cp:
>> +	andn	%o2, 3, %o3	/* i3 is aligned word count  */
>> +	sub	%o1, %o0, %o1	/* i1 gets the difference  */
>> +
>> +1:	lduw	[%o1+%o0], %o4	/* read from address  */
>> +	deccc	4, %o3		/* decrement count  */
>> +	st	%o4, [%o0]	/* write at destination address  */
>> +	bg,pt	%XCC, 1b
>> +	 inc	4, %o0		/* increment to address  */
>> +	b	7f
>> +	 and	%o2, 3, %o2	/* number of leftover bytes, if any  */
>> +
>> +/*
>> + * differenced byte copy, works with any alignment
>> + */
>> +.Ldbytecp:
>> +	b	7f
>> +	 sub	%o1, %o0, %o1	/* i1 gets the difference  */
>> +
>> +4:	stb	%o4, [%o0]	/* write to address  */
>> +	inc	%o0		/* inc to address  */
>> +7:	deccc	%o2		/* decrement count  */
>> +	bge,a	%XCC, 4b	/* loop till done  */
>> +	 ldub	[%o1+%o0], %o4	/* read from address  */
>> +	retl
>> +	 mov	%g2, %o0	/* return pointer to destination  */
>> +
>> +/*
>> + * an overlapped copy that must be done "backwards"
>> + */
>> +.Lovbc:
>> +	add	%o1, %o2, %o1	/* get to end of source space  */
>> +	add	%o0, %o2, %o0	/* get to end of destination space  */
>> +	sub	%o1, %o0, %o1	/* i1 gets the difference  */
>> +
>> +5:	dec	%o0		/* decrement to address  */
>> +	ldub	[%o1+%o0], %o3	/* read a byte  */
>> +	deccc	%o2		/* decrement count  */
>> +	bg,pt	%XCC, 5b 	/* loop until done  */
>> +	 stb	%o3, [%o0]	/* write byte  */
>> +	retl
>> +	 mov	%g2, %o0	/* return pointer to destination  */
>> +END(memmove)
>> +
>> +libc_hidden_builtin_def (memmove)