[apple/xnu.git] / osfmk / ppc / commpage / bcopy_g3.s

/*
 * Copyright (c) 2003 Apple Computer, Inc. All rights reserved.
 *
 * @APPLE_LICENSE_HEADER_START@
 * 
 * The contents of this file constitute Original Code as defined in and
 * are subject to the Apple Public Source License Version 1.1 (the
 * "License").  You may not use this file except in compliance with the
 * License.  Please obtain a copy of the License at
 * http://www.apple.com/publicsource and read it before using this file.
 * 
 * This Original Code and all software distributed under the License are
 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
 * License for the specific language governing rights and limitations
 * under the License.
 * 
 * @APPLE_LICENSE_HEADER_END@
 */
/* =======================================
 * BCOPY, MEMCPY, and MEMMOVE for Mac OS X
 * =======================================
 *
 * Version of 2/20/2003, tuned for G3.
 *
 * Register usage.  Note we use R2, so this code will not run in a PEF/CFM
 * environment.
 *
 *   r0  = "w7" or temp
 *   r2  = "w8"
 *   r3  = not used, as memcpy and memmove return 1st parameter as a value
 *   r4  = source ptr ("rs")
 *   r5  = count of bytes to move ("rc")
 *   r6  = "w1"
 *   r7  = "w2"
 *   r8  = "w3"
 *   r9  = "w4"
 *   r10 = "w5"
 *   r11 = "w6"
 *   r12 = destination ptr ("rd")
 * f0-f3 = used for moving 8-byte aligned data
 */
#define rs	r4		// NB: we depend on rs==r4 in "lswx" instructions
#define rd	r12
#define rc	r5

#define w1	r6
#define w2	r7
#define w3	r8
#define	w4	r9
#define	w5	r10
#define	w6	r11
#define	w7	r0
#define	w8	r2

#define	ASSEMBLER
#include <sys/appleapiopts.h>
#include <ppc/asm.h>
#include <machine/cpu_capabilities.h>
#include <machine/commpage.h>

        .text
        .globl 	EXT(bcopy_g3)


#define	kLong	33					// too long for string ops


// Main entry points.

        .align 	5
bcopy_g3:							// void bcopy(const void *src, void *dst, size_t len)
        cmplwi	rc,kLong			// length > 32 bytes?
        sub		w1,r4,r3			// must move in reverse if (rd-rs)<rc
        mr		rd,r4				// start to move source & dest to canonic spot
        bge		LLong0				// skip if long operand
        mtxer	rc					// set length for string ops
        lswx	r5,0,r3				// load bytes into r5-r12
        stswx	r5,0,r4				// store them
        blr

// NB: memcpy() and memmove() must follow bcopy() by 32 bytes, for comm page.

        .align	5
Lmemcpy_g3:							// void* memcpy(void *dst, void *src, size_t len)
Lmemmove_g3:						// void* memmove(void *dst, const void *src, size_t len)
        cmplwi	rc,kLong			// length > 32 bytes?
        sub		w1,r3,rs			// must move in reverse if (rd-rs)<rc
        mr		rd,r3				// must leave r3 alone, it is return value for memcpy etc
        bge		LLong1				// longer than 32 bytes
        mtxer	rc					// set length for string ops
        lswx	r5,0,r4				// load bytes into r5-r12
        stswx	r5,0,r3				// store them
        blr

// Long operands (more than 32 bytes.)
//		w1  = (rd-rs), used to check for alignment

LLong0:								// enter from bcopy()
        mr		rs,r3				// must leave r3 alone (it is return value for memcpy)
LLong1:								// enter from memcpy() and memmove()
        cmplw	cr1,w1,rc			// set cr1 blt iff we must move reverse
        rlwinm	r0,w1,0,0x3			// are operands relatively word-aligned?
        neg		w2,rd				// prepare to align destination
        cmpwi	cr5,r0,0			// set cr5 beq if relatively word aligned
        blt		cr1,LLongReverse	// handle reverse move
        andi.	w4,w2,3				// w4 <- #bytes to word align destination
        beq		cr5,LLongFloat		// relatively aligned so use FPRs
        sub		rc,rc,w4			// adjust count for alignment
        srwi	r0,rc,5				// get #chunks to xfer (>=1)
        rlwinm	rc,rc,0,0x1F		// mask down to leftover bytes
        mtctr	r0					// set up loop count
        beq		1f					// dest already word aligned
    
// Word align the destination.
        
        mtxer	w4					// byte count to xer
        cmpwi	r0,0				// any chunks to xfer?
        lswx	w1,0,rs				// move w4 bytes to align dest
        add		rs,rs,w4
        stswx	w1,0,rd
        add		rd,rd,w4
        beq-	2f					// pathologic case, no chunks to xfer

// Forward, unaligned loop.

1:
        lwz		w1,0(rs)
        lwz		w2,4(rs)
        lwz		w3,8(rs)
        lwz		w4,12(rs)
        lwz		w5,16(rs)
        lwz		w6,20(rs)
        lwz		w7,24(rs)
        lwz		w8,28(rs)
        addi	rs,rs,32
        stw		w1,0(rd)
        stw		w2,4(rd)
        stw		w3,8(rd)
        stw		w4,12(rd)
        stw		w5,16(rd)
        stw		w6,20(rd)
        stw		w7,24(rd)
        stw		w8,28(rd)
        addi	rd,rd,32
        bdnz	1b
2:									// rc = remaining bytes (0-31)
        mtxer	rc					// set up count for string ops
        mr		r0,rd				// move dest ptr out of the way
        lswx	r5,0,rs				// load xer bytes into r5-r12 (rs==r4)
        stswx	r5,0,r0				// store them
        blr
        

// Forward, aligned loop.  We use FPRs.

LLongFloat:
        andi.	w4,w2,7				// W4 <- #bytes to doubleword-align destination
        sub		rc,rc,w4			// adjust count for alignment
        srwi	r0,rc,5				// number of 32-byte chunks to xfer
        rlwinm	rc,rc,0,0x1F		// mask down to leftover bytes
        mtctr	r0					// set up loop count
        beq		1f					// dest already doubleword aligned
    
// Doubleword align the destination.
        
        mtxer	w4					// byte count to xer
        cmpwi	r0,0				// any chunks to xfer?
        lswx	w1,0,rs				// move w4 bytes to align dest
        add		rs,rs,w4
        stswx	w1,0,rd
        add		rd,rd,w4
        beq-	2f					// pathologic case, no chunks to xfer
1:									// loop over 32-byte chunks
        lfd		f0,0(rs)
        lfd		f1,8(rs)
        lfd		f2,16(rs)
        lfd		f3,24(rs)
        addi	rs,rs,32
        stfd	f0,0(rd)
        stfd	f1,8(rd)
        stfd	f2,16(rd)
        stfd	f3,24(rd)
        addi	rd,rd,32
        bdnz	1b
2:									// rc = remaining bytes (0-31)
        mtxer	rc					// set up count for string ops
        mr		r0,rd				// move dest ptr out of the way
        lswx	r5,0,rs				// load xer bytes into r5-r12 (rs==r4)
        stswx	r5,0,r0				// store them
        blr

        
// Long, reverse moves.
//		cr5 = beq if relatively word aligned

LLongReverse:
        add		rd,rd,rc			// point to end of operands + 1
        add		rs,rs,rc
        beq		cr5,LReverseFloat	// aligned operands so can use FPRs
        srwi	r0,rc,5				// get chunk count
        rlwinm	rc,rc,0,0x1F		// mask down to leftover bytes
        mtctr	r0					// set up loop count
        mtxer	rc					// set up for trailing bytes
1:
        lwz		w1,-4(rs)
        lwz		w2,-8(rs)
        lwz		w3,-12(rs)
        lwz		w4,-16(rs)
        stw		w1,-4(rd)
        lwz		w5,-20(rs)
        stw		w2,-8(rd)
        lwz		w6,-24(rs)
        stw		w3,-12(rd)
        lwz		w7,-28(rs)
        stw		w4,-16(rd)
        lwzu	w8,-32(rs)
        stw		w5,-20(rd)
        stw		w6,-24(rd)
        stw		w7,-28(rd)
        stwu	w8,-32(rd)
        bdnz	1b

        sub		r4,rs,rc			// point to 1st (leftmost) leftover byte (0..31)
        sub		r0,rd,rc			// move dest ptr out of way
        lswx	r5,0,r4				// load xer bytes into r5-r12
        stswx	r5,0,r0				// store them
        blr


// Long, reverse aligned moves.  We use FPRs.

LReverseFloat:
        andi.	w4,rd,7				// W3 <- #bytes to doubleword-align destination
        sub		rc,rc,w4			// adjust count for alignment
        srwi	r0,rc,5				// number of 32-byte chunks to xfer
        rlwinm	rc,rc,0,0x1F		// mask down to leftover bytes
        mtctr	r0					// set up loop count
        beq		1f					// dest already doubleword aligned
    
// Doubleword align the destination.
        
        mtxer	w4					// byte count to xer
        cmpwi	r0,0				// any chunks to xfer?
        sub		rs,rs,w4			// point to 1st bytes to xfer
        sub		rd,rd,w4
        lswx	w1,0,rs				// move w3 bytes to align dest
        stswx	w1,0,rd
        beq-	2f					// pathologic case, no chunks to xfer
1:
        lfd		f0,-8(rs)
        lfd		f1,-16(rs)
        lfd		f2,-24(rs)
        lfdu	f3,-32(rs)
        stfd	f0,-8(rd)
        stfd	f1,-16(rd)
        stfd	f2,-24(rd)
        stfdu	f3,-32(rd)
        bdnz	1b
2:									// rc = remaining bytes (0-31)
        mtxer	rc					// set up count for string ops
        sub		r4,rs,rc			// point to 1st (leftmost) leftover byte (0..31)
        sub		r0,rd,rc			// move dest ptr out of way
        lswx	r5,0,r4				// load xer bytes into r5-r12
        stswx	r5,0,r0				// store them
        blr

        COMMPAGE_DESCRIPTOR(bcopy_g3,_COMM_PAGE_BCOPY,0,k64Bit+kHasAltivec,0)
Commit	Line	Data
55e303ae A	1	/*
	2	* Copyright (c) 2003 Apple Computer, Inc. All rights reserved.
	3	*
	4	* @APPLE_LICENSE_HEADER_START@
	5	*
e5568f75 A	6	* The contents of this file constitute Original Code as defined in and
	7	* are subject to the Apple Public Source License Version 1.1 (the
	8	* "License"). You may not use this file except in compliance with the
	9	* License. Please obtain a copy of the License at
	10	* http://www.apple.com/publicsource and read it before using this file.
55e303ae	11	*
e5568f75 A	12	* This Original Code and all software distributed under the License are
e5568f75 A	13	* distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
55e303ae A	14	* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
55e303ae A	15	* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
e5568f75 A	16	* FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
	17	* License for the specific language governing rights and limitations
	18	* under the License.
55e303ae A	19	*
	20	* @APPLE_LICENSE_HEADER_END@
	21	*/
	22	/* =======================================
	23	* BCOPY, MEMCPY, and MEMMOVE for Mac OS X
	24	* =======================================
	25	*
	26	* Version of 2/20/2003, tuned for G3.
	27	*
	28	* Register usage. Note we use R2, so this code will not run in a PEF/CFM
	29	* environment.
	30	*
	31	* r0 = "w7" or temp
	32	* r2 = "w8"
	33	* r3 = not used, as memcpy and memmove return 1st parameter as a value
	34	* r4 = source ptr ("rs")
	35	* r5 = count of bytes to move ("rc")
	36	* r6 = "w1"
	37	* r7 = "w2"
	38	* r8 = "w3"
	39	* r9 = "w4"
	40	* r10 = "w5"
	41	* r11 = "w6"
	42	* r12 = destination ptr ("rd")
	43	* f0-f3 = used for moving 8-byte aligned data
	44	*/
	45	#define rs r4 // NB: we depend on rs==r4 in "lswx" instructions
	46	#define rd r12
	47	#define rc r5
	48
	49	#define w1 r6
	50	#define w2 r7
	51	#define w3 r8
	52	#define w4 r9
	53	#define w5 r10
	54	#define w6 r11
	55	#define w7 r0
	56	#define w8 r2
	57
	58	#define ASSEMBLER
	59	#include <sys/appleapiopts.h>
	60	#include <ppc/asm.h>
	61	#include <machine/cpu_capabilities.h>
	62	#include <machine/commpage.h>
	63
	64	.text
	65	.globl EXT(bcopy_g3)
	66
	67
	68	#define kLong 33 // too long for string ops
	69
	70
	71	// Main entry points.
	72
	73	.align 5
	74	bcopy_g3: // void bcopy(const void src, void dst, size_t len)
	75	cmplwi rc,kLong // length > 32 bytes?
	76	sub w1,r4,r3 // must move in reverse if (rd-rs)<rc
	77	mr rd,r4 // start to move source & dest to canonic spot
	78	bge LLong0 // skip if long operand
	79	mtxer rc // set length for string ops
	80	lswx r5,0,r3 // load bytes into r5-r12
	81	stswx r5,0,r4 // store them
	82	blr
83
84	// NB: memcpy() and memmove() must follow bcopy() by 32 bytes, for comm page.
85
86	.align 5
87	Lmemcpy_g3: // void* memcpy(void dst, void src, size_t len)
88	Lmemmove_g3: // void* memmove(void dst, const void src, size_t len)
89	cmplwi rc,kLong // length > 32 bytes?
90	sub w1,r3,rs // must move in reverse if (rd-rs)<rc
91	mr rd,r3 // must leave r3 alone, it is return value for memcpy etc
92	bge LLong1 // longer than 32 bytes
93	mtxer rc // set length for string ops
94	lswx r5,0,r4 // load bytes into r5-r12
95	stswx r5,0,r3 // store them
96	blr
97
98	// Long operands (more than 32 bytes.)
99	// w1 = (rd-rs), used to check for alignment
100
101	LLong0: // enter from bcopy()
102	mr rs,r3 // must leave r3 alone (it is return value for memcpy)
103	LLong1: // enter from memcpy() and memmove()
104	cmplw cr1,w1,rc // set cr1 blt iff we must move reverse
105	rlwinm r0,w1,0,0x3 // are operands relatively word-aligned?
106	neg w2,rd // prepare to align destination
107	cmpwi cr5,r0,0 // set cr5 beq if relatively word aligned
108	blt cr1,LLongReverse // handle reverse move
109	andi. w4,w2,3 // w4 <- #bytes to word align destination
110	beq cr5,LLongFloat // relatively aligned so use FPRs
111	sub rc,rc,w4 // adjust count for alignment
112	srwi r0,rc,5 // get #chunks to xfer (>=1)
113	rlwinm rc,rc,0,0x1F // mask down to leftover bytes
114	mtctr r0 // set up loop count
115	beq 1f // dest already word aligned
116
117	// Word align the destination.
118
119	mtxer w4 // byte count to xer
120	cmpwi r0,0 // any chunks to xfer?
121	lswx w1,0,rs // move w4 bytes to align dest
122	add rs,rs,w4
123	stswx w1,0,rd
124	add rd,rd,w4
125	beq- 2f // pathologic case, no chunks to xfer
126
127	// Forward, unaligned loop.
128
129	1:
130	lwz w1,0(rs)
131	lwz w2,4(rs)
132	lwz w3,8(rs)
133	lwz w4,12(rs)
134	lwz w5,16(rs)
135	lwz w6,20(rs)
136	lwz w7,24(rs)
137	lwz w8,28(rs)
138	addi rs,rs,32
139	stw w1,0(rd)
140	stw w2,4(rd)
141	stw w3,8(rd)
142	stw w4,12(rd)
143	stw w5,16(rd)
144	stw w6,20(rd)
145	stw w7,24(rd)
146	stw w8,28(rd)
147	addi rd,rd,32
148	bdnz 1b
149	2: // rc = remaining bytes (0-31)
150	mtxer rc // set up count for string ops
151	mr r0,rd // move dest ptr out of the way
152	lswx r5,0,rs // load xer bytes into r5-r12 (rs==r4)
153	stswx r5,0,r0 // store them
154	blr
155
156
157
158	// Forward, aligned loop. We use FPRs.
159
160	LLongFloat:
161	andi. w4,w2,7 // W4 <- #bytes to doubleword-align destination
162	sub rc,rc,w4 // adjust count for alignment
163	srwi r0,rc,5 // number of 32-byte chunks to xfer
164	rlwinm rc,rc,0,0x1F // mask down to leftover bytes
165	mtctr r0 // set up loop count
166	beq 1f // dest already doubleword aligned
167
168	// Doubleword align the destination.
169
170	mtxer w4 // byte count to xer
171	cmpwi r0,0 // any chunks to xfer?
172	lswx w1,0,rs // move w4 bytes to align dest
173	add rs,rs,w4
174	stswx w1,0,rd
175	add rd,rd,w4
176	beq- 2f // pathologic case, no chunks to xfer
177	1: // loop over 32-byte chunks
178	lfd f0,0(rs)
179	lfd f1,8(rs)
180	lfd f2,16(rs)
181	lfd f3,24(rs)
182	addi rs,rs,32
183	stfd f0,0(rd)
184	stfd f1,8(rd)
185	stfd f2,16(rd)
186	stfd f3,24(rd)
187	addi rd,rd,32
188	bdnz 1b
189	2: // rc = remaining bytes (0-31)
190	mtxer rc // set up count for string ops
191	mr r0,rd // move dest ptr out of the way
192	lswx r5,0,rs // load xer bytes into r5-r12 (rs==r4)
193	stswx r5,0,r0 // store them
194	blr
195
196
197	// Long, reverse moves.
198	// cr5 = beq if relatively word aligned
199
200	LLongReverse:
201	add rd,rd,rc // point to end of operands + 1
202	add rs,rs,rc
203	beq cr5,LReverseFloat // aligned operands so can use FPRs
204	srwi r0,rc,5 // get chunk count
205	rlwinm rc,rc,0,0x1F // mask down to leftover bytes
206	mtctr r0 // set up loop count
207	mtxer rc // set up for trailing bytes
208	1:
209	lwz w1,-4(rs)
210	lwz w2,-8(rs)
211	lwz w3,-12(rs)
212	lwz w4,-16(rs)
213	stw w1,-4(rd)
214	lwz w5,-20(rs)
215	stw w2,-8(rd)
216	lwz w6,-24(rs)
217	stw w3,-12(rd)
218	lwz w7,-28(rs)
219	stw w4,-16(rd)
220	lwzu w8,-32(rs)
221	stw w5,-20(rd)
222	stw w6,-24(rd)
223	stw w7,-28(rd)
224	stwu w8,-32(rd)
225	bdnz 1b
226
227	sub r4,rs,rc // point to 1st (leftmost) leftover byte (0..31)
228	sub r0,rd,rc // move dest ptr out of way
229	lswx r5,0,r4 // load xer bytes into r5-r12
230	stswx r5,0,r0 // store them
231	blr
232
233
234	// Long, reverse aligned moves. We use FPRs.
235
236	LReverseFloat:
237	andi. w4,rd,7 // W3 <- #bytes to doubleword-align destination
238	sub rc,rc,w4 // adjust count for alignment
239	srwi r0,rc,5 // number of 32-byte chunks to xfer
240	rlwinm rc,rc,0,0x1F // mask down to leftover bytes
241	mtctr r0 // set up loop count
242	beq 1f // dest already doubleword aligned
243
244	// Doubleword align the destination.
245
246	mtxer w4 // byte count to xer
247	cmpwi r0,0 // any chunks to xfer?
248	sub rs,rs,w4 // point to 1st bytes to xfer
249	sub rd,rd,w4
250	lswx w1,0,rs // move w3 bytes to align dest
251	stswx w1,0,rd
252	beq- 2f // pathologic case, no chunks to xfer
253	1:
254	lfd f0,-8(rs)
255	lfd f1,-16(rs)
256	lfd f2,-24(rs)
257	lfdu f3,-32(rs)
258	stfd f0,-8(rd)
259	stfd f1,-16(rd)
260	stfd f2,-24(rd)
261	stfdu f3,-32(rd)
262	bdnz 1b
263	2: // rc = remaining bytes (0-31)
264	mtxer rc // set up count for string ops
265	sub r4,rs,rc // point to 1st (leftmost) leftover byte (0..31)
266	sub r0,rd,rc // move dest ptr out of way
267	lswx r5,0,r4 // load xer bytes into r5-r12
268	stswx r5,0,r0 // store them
269	blr
270
271	COMMPAGE_DESCRIPTOR(bcopy_g3,_COMM_PAGE_BCOPY,0,k64Bit+kHasAltivec,0)