1/* NG2memcpy.S: Niagara-2 optimized memcpy.
2 *
3 * Copyright (C) 2007 David S. Miller (davem@davemloft.net)
4 */
5
6#ifdef __KERNEL__
7#include <asm/visasm.h>
8#include <asm/asi.h>
9#define GLOBAL_SPARE	%g7
10#else
11#define ASI_PNF 0x82
12#define ASI_BLK_P 0xf0
13#define ASI_BLK_INIT_QUAD_LDD_P 0xe2
14#define FPRS_FEF  0x04
15#ifdef MEMCPY_DEBUG
16#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs; \
17		     clr %g1; clr %g2; clr %g3; clr %g5; subcc %g0, %g0, %g0;
18#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
19#else
20#define VISEntryHalf rd %fprs, %o5; wr %g0, FPRS_FEF, %fprs
21#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
22#endif
23#define GLOBAL_SPARE	%g5
24#endif
25
26#ifndef STORE_ASI
27#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
28#define STORE_ASI	ASI_BLK_INIT_QUAD_LDD_P
29#else
30#define STORE_ASI	0x80		/* ASI_P */
31#endif
32#endif
33
34#ifndef EX_LD
35#define EX_LD(x)	x
36#endif
37
38#ifndef EX_ST
39#define EX_ST(x)	x
40#endif
41
42#ifndef EX_RETVAL
43#define EX_RETVAL(x)	x
44#endif
45
46#ifndef LOAD
47#define LOAD(type,addr,dest)	type [addr], dest
48#endif
49
50#ifndef LOAD_BLK
51#define LOAD_BLK(addr,dest)	ldda [addr] ASI_BLK_P, dest
52#endif
53
54#ifndef STORE
55#ifndef MEMCPY_DEBUG
56#define STORE(type,src,addr)	type src, [addr]
57#else
58#define STORE(type,src,addr)	type##a src, [addr] 0x80
59#endif
60#endif
61
62#ifndef STORE_BLK
63#define STORE_BLK(src,addr)	stda src, [addr] ASI_BLK_P
64#endif
65
66#ifndef STORE_INIT
67#define STORE_INIT(src,addr)	stxa src, [addr] STORE_ASI
68#endif
69
70#ifndef FUNC_NAME
71#define FUNC_NAME	NG2memcpy
72#endif
73
74#ifndef PREAMBLE
75#define PREAMBLE
76#endif
77
78#ifndef XCC
79#define XCC xcc
80#endif
81
82#define FREG_FROB(x0, x1, x2, x3, x4, x5, x6, x7, x8) \
83	faligndata	%x0, %x1, %f0; \
84	faligndata	%x1, %x2, %f2; \
85	faligndata	%x2, %x3, %f4; \
86	faligndata	%x3, %x4, %f6; \
87	faligndata	%x4, %x5, %f8; \
88	faligndata	%x5, %x6, %f10; \
89	faligndata	%x6, %x7, %f12; \
90	faligndata	%x7, %x8, %f14;
91
92#define FREG_MOVE_1(x0) \
93	fsrc2		%x0, %f0;
94#define FREG_MOVE_2(x0, x1) \
95	fsrc2		%x0, %f0; \
96	fsrc2		%x1, %f2;
97#define FREG_MOVE_3(x0, x1, x2) \
98	fsrc2		%x0, %f0; \
99	fsrc2		%x1, %f2; \
100	fsrc2		%x2, %f4;
101#define FREG_MOVE_4(x0, x1, x2, x3) \
102	fsrc2		%x0, %f0; \
103	fsrc2		%x1, %f2; \
104	fsrc2		%x2, %f4; \
105	fsrc2		%x3, %f6;
106#define FREG_MOVE_5(x0, x1, x2, x3, x4) \
107	fsrc2		%x0, %f0; \
108	fsrc2		%x1, %f2; \
109	fsrc2		%x2, %f4; \
110	fsrc2		%x3, %f6; \
111	fsrc2		%x4, %f8;
112#define FREG_MOVE_6(x0, x1, x2, x3, x4, x5) \
113	fsrc2		%x0, %f0; \
114	fsrc2		%x1, %f2; \
115	fsrc2		%x2, %f4; \
116	fsrc2		%x3, %f6; \
117	fsrc2		%x4, %f8; \
118	fsrc2		%x5, %f10;
119#define FREG_MOVE_7(x0, x1, x2, x3, x4, x5, x6) \
120	fsrc2		%x0, %f0; \
121	fsrc2		%x1, %f2; \
122	fsrc2		%x2, %f4; \
123	fsrc2		%x3, %f6; \
124	fsrc2		%x4, %f8; \
125	fsrc2		%x5, %f10; \
126	fsrc2		%x6, %f12;
127#define FREG_MOVE_8(x0, x1, x2, x3, x4, x5, x6, x7) \
128	fsrc2		%x0, %f0; \
129	fsrc2		%x1, %f2; \
130	fsrc2		%x2, %f4; \
131	fsrc2		%x3, %f6; \
132	fsrc2		%x4, %f8; \
133	fsrc2		%x5, %f10; \
134	fsrc2		%x6, %f12; \
135	fsrc2		%x7, %f14;
136#define FREG_LOAD_1(base, x0) \
137	EX_LD(LOAD(ldd, base + 0x00, %x0))
138#define FREG_LOAD_2(base, x0, x1) \
139	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
140	EX_LD(LOAD(ldd, base + 0x08, %x1));
141#define FREG_LOAD_3(base, x0, x1, x2) \
142	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
143	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
144	EX_LD(LOAD(ldd, base + 0x10, %x2));
145#define FREG_LOAD_4(base, x0, x1, x2, x3) \
146	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
147	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
148	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
149	EX_LD(LOAD(ldd, base + 0x18, %x3));
150#define FREG_LOAD_5(base, x0, x1, x2, x3, x4) \
151	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
152	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
153	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
154	EX_LD(LOAD(ldd, base + 0x18, %x3)); \
155	EX_LD(LOAD(ldd, base + 0x20, %x4));
156#define FREG_LOAD_6(base, x0, x1, x2, x3, x4, x5) \
157	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
158	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
159	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
160	EX_LD(LOAD(ldd, base + 0x18, %x3)); \
161	EX_LD(LOAD(ldd, base + 0x20, %x4)); \
162	EX_LD(LOAD(ldd, base + 0x28, %x5));
163#define FREG_LOAD_7(base, x0, x1, x2, x3, x4, x5, x6) \
164	EX_LD(LOAD(ldd, base + 0x00, %x0)); \
165	EX_LD(LOAD(ldd, base + 0x08, %x1)); \
166	EX_LD(LOAD(ldd, base + 0x10, %x2)); \
167	EX_LD(LOAD(ldd, base + 0x18, %x3)); \
168	EX_LD(LOAD(ldd, base + 0x20, %x4)); \
169	EX_LD(LOAD(ldd, base + 0x28, %x5)); \
170	EX_LD(LOAD(ldd, base + 0x30, %x6));
171
172	.register	%g2,#scratch
173	.register	%g3,#scratch
174
175	.text
176	.align		64
177
178	.globl	FUNC_NAME
179	.type	FUNC_NAME,#function
180FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
181	srlx		%o2, 31, %g2
182	cmp		%g2, 0
183	tne		%xcc, 5
184	PREAMBLE
185	mov		%o0, %o3
186	cmp		%o2, 0
187	be,pn		%XCC, 85f
188	 or		%o0, %o1, GLOBAL_SPARE
189	cmp		%o2, 16
190	blu,a,pn	%XCC, 80f
191	 or		GLOBAL_SPARE, %o2, GLOBAL_SPARE
192
193	/* 2 blocks (128 bytes) is the minimum we can do the block
194	 * copy with.  We need to ensure that we'll iterate at least
195	 * once in the block copy loop.  At worst we'll need to align
196	 * the destination to a 64-byte boundary which can chew up
197	 * to (64 - 1) bytes from the length before we perform the
198	 * block copy loop.
199	 *
200	 * However, the cut-off point, performance wise, is around
201	 * 4 64-byte blocks.
202	 */
203	cmp		%o2, (4 * 64)
204	blu,pt		%XCC, 75f
205	 andcc		GLOBAL_SPARE, 0x7, %g0
206
207	/* %o0:	dst
208	 * %o1:	src
209	 * %o2:	len  (known to be >= 128)
210	 *
211	 * The block copy loops can use %o4, %g2, %g3 as
212	 * temporaries while copying the data.  %o5 must
213	 * be preserved between VISEntryHalf and VISExitHalf
214	 */
215
216	LOAD(prefetch, %o1 + 0x000, #one_read)
217	LOAD(prefetch, %o1 + 0x040, #one_read)
218	LOAD(prefetch, %o1 + 0x080, #one_read)
219
220	/* Align destination on 64-byte boundary.  */
221	andcc		%o0, (64 - 1), %o4
222	be,pt		%XCC, 2f
223	 sub		%o4, 64, %o4
224	sub		%g0, %o4, %o4	! bytes to align dst
225	sub		%o2, %o4, %o2
2261:	subcc		%o4, 1, %o4
227	EX_LD(LOAD(ldub, %o1, %g1))
228	EX_ST(STORE(stb, %g1, %o0))
229	add		%o1, 1, %o1
230	bne,pt		%XCC, 1b
231	add		%o0, 1, %o0
232
2332:
234	/* Clobbers o5/g1/g2/g3/g7/icc/xcc.  We must preserve
235	 * o5 from here until we hit VISExitHalf.
236	 */
237	VISEntryHalf
238
239	membar		#Sync
240	alignaddr	%o1, %g0, %g0
241
242	add		%o1, (64 - 1), %o4
243	andn		%o4, (64 - 1), %o4
244	andn		%o2, (64 - 1), %g1
245	sub		%o2, %g1, %o2
246
247	and		%o1, (64 - 1), %g2
248	add		%o1, %g1, %o1
249	sub		%o0, %o4, %g3
250	brz,pt		%g2, 190f
251	 cmp		%g2, 32
252	blu,a		5f
253	 cmp		%g2, 16
254	cmp		%g2, 48
255	blu,a		4f
256	 cmp		%g2, 40
257	cmp		%g2, 56
258	blu		170f
259	 nop
260	ba,a,pt		%xcc, 180f
261
2624:	/* 32 <= low bits < 48 */
263	blu		150f
264	 nop
265	ba,a,pt		%xcc, 160f
2665:	/* 0 < low bits < 32 */
267	blu,a		6f
268	 cmp		%g2, 8
269	cmp		%g2, 24
270	blu		130f
271	 nop
272	ba,a,pt		%xcc, 140f
2736:	/* 0 < low bits < 16 */
274	bgeu		120f
275	 nop
276	/* fall through for 0 < low bits < 8 */
277110:	sub		%o4, 64, %g2
278	EX_LD(LOAD_BLK(%g2, %f0))
2791:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
280	EX_LD(LOAD_BLK(%o4, %f16))
281	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f14, f16)
282	EX_ST(STORE_BLK(%f0, %o4 + %g3))
283	FREG_MOVE_8(f16, f18, f20, f22, f24, f26, f28, f30)
284	subcc		%g1, 64, %g1
285	add		%o4, 64, %o4
286	bne,pt		%xcc, 1b
287	 LOAD(prefetch, %o4 + 64, #one_read)
288	ba,pt		%xcc, 195f
289	 nop
290
291120:	sub		%o4, 56, %g2
292	FREG_LOAD_7(%g2, f0, f2, f4, f6, f8, f10, f12)
2931:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
294	EX_LD(LOAD_BLK(%o4, %f16))
295	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f16, f18)
296	EX_ST(STORE_BLK(%f0, %o4 + %g3))
297	FREG_MOVE_7(f18, f20, f22, f24, f26, f28, f30)
298	subcc		%g1, 64, %g1
299	add		%o4, 64, %o4
300	bne,pt		%xcc, 1b
301	 LOAD(prefetch, %o4 + 64, #one_read)
302	ba,pt		%xcc, 195f
303	 nop
304
305130:	sub		%o4, 48, %g2
306	FREG_LOAD_6(%g2, f0, f2, f4, f6, f8, f10)
3071:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
308	EX_LD(LOAD_BLK(%o4, %f16))
309	FREG_FROB(f0, f2, f4, f6, f8, f10, f16, f18, f20)
310	EX_ST(STORE_BLK(%f0, %o4 + %g3))
311	FREG_MOVE_6(f20, f22, f24, f26, f28, f30)
312	subcc		%g1, 64, %g1
313	add		%o4, 64, %o4
314	bne,pt		%xcc, 1b
315	 LOAD(prefetch, %o4 + 64, #one_read)
316	ba,pt		%xcc, 195f
317	 nop
318
319140:	sub		%o4, 40, %g2
320	FREG_LOAD_5(%g2, f0, f2, f4, f6, f8)
3211:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
322	EX_LD(LOAD_BLK(%o4, %f16))
323	FREG_FROB(f0, f2, f4, f6, f8, f16, f18, f20, f22)
324	EX_ST(STORE_BLK(%f0, %o4 + %g3))
325	FREG_MOVE_5(f22, f24, f26, f28, f30)
326	subcc		%g1, 64, %g1
327	add		%o4, 64, %o4
328	bne,pt		%xcc, 1b
329	 LOAD(prefetch, %o4 + 64, #one_read)
330	ba,pt		%xcc, 195f
331	 nop
332
333150:	sub		%o4, 32, %g2
334	FREG_LOAD_4(%g2, f0, f2, f4, f6)
3351:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
336	EX_LD(LOAD_BLK(%o4, %f16))
337	FREG_FROB(f0, f2, f4, f6, f16, f18, f20, f22, f24)
338	EX_ST(STORE_BLK(%f0, %o4 + %g3))
339	FREG_MOVE_4(f24, f26, f28, f30)
340	subcc		%g1, 64, %g1
341	add		%o4, 64, %o4
342	bne,pt		%xcc, 1b
343	 LOAD(prefetch, %o4 + 64, #one_read)
344	ba,pt		%xcc, 195f
345	 nop
346
347160:	sub		%o4, 24, %g2
348	FREG_LOAD_3(%g2, f0, f2, f4)
3491:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
350	EX_LD(LOAD_BLK(%o4, %f16))
351	FREG_FROB(f0, f2, f4, f16, f18, f20, f22, f24, f26)
352	EX_ST(STORE_BLK(%f0, %o4 + %g3))
353	FREG_MOVE_3(f26, f28, f30)
354	subcc		%g1, 64, %g1
355	add		%o4, 64, %o4
356	bne,pt		%xcc, 1b
357	 LOAD(prefetch, %o4 + 64, #one_read)
358	ba,pt		%xcc, 195f
359	 nop
360
361170:	sub		%o4, 16, %g2
362	FREG_LOAD_2(%g2, f0, f2)
3631:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
364	EX_LD(LOAD_BLK(%o4, %f16))
365	FREG_FROB(f0, f2, f16, f18, f20, f22, f24, f26, f28)
366	EX_ST(STORE_BLK(%f0, %o4 + %g3))
367	FREG_MOVE_2(f28, f30)
368	subcc		%g1, 64, %g1
369	add		%o4, 64, %o4
370	bne,pt		%xcc, 1b
371	 LOAD(prefetch, %o4 + 64, #one_read)
372	ba,pt		%xcc, 195f
373	 nop
374
375180:	sub		%o4, 8, %g2
376	FREG_LOAD_1(%g2, f0)
3771:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
378	EX_LD(LOAD_BLK(%o4, %f16))
379	FREG_FROB(f0, f16, f18, f20, f22, f24, f26, f28, f30)
380	EX_ST(STORE_BLK(%f0, %o4 + %g3))
381	FREG_MOVE_1(f30)
382	subcc		%g1, 64, %g1
383	add		%o4, 64, %o4
384	bne,pt		%xcc, 1b
385	 LOAD(prefetch, %o4 + 64, #one_read)
386	ba,pt		%xcc, 195f
387	 nop
388
389190:
3901:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
391	subcc		%g1, 64, %g1
392	EX_LD(LOAD_BLK(%o4, %f0))
393	EX_ST(STORE_BLK(%f0, %o4 + %g3))
394	add		%o4, 64, %o4
395	bne,pt		%xcc, 1b
396	 LOAD(prefetch, %o4 + 64, #one_read)
397
398195:
399	add		%o4, %g3, %o0
400	membar		#Sync
401
402	VISExitHalf
403
404	/* %o2 contains any final bytes still needed to be copied
405	 * over. If anything is left, we copy it one byte at a time.
406	 */
407	brz,pt		%o2, 85f
408	 sub		%o0, %o1, GLOBAL_SPARE
409	ba,a,pt		%XCC, 90f
410
411	.align		64
41275: /* 16 < len <= 64 */
413	bne,pn		%XCC, 75f
414	 sub		%o0, %o1, GLOBAL_SPARE
415
41672:
417	andn		%o2, 0xf, %o4
418	and		%o2, 0xf, %o2
4191:	subcc		%o4, 0x10, %o4
420	EX_LD(LOAD(ldx, %o1, %o5))
421	add		%o1, 0x08, %o1
422	EX_LD(LOAD(ldx, %o1, %g1))
423	sub		%o1, 0x08, %o1
424	EX_ST(STORE(stx, %o5, %o1 + GLOBAL_SPARE))
425	add		%o1, 0x8, %o1
426	EX_ST(STORE(stx, %g1, %o1 + GLOBAL_SPARE))
427	bgu,pt		%XCC, 1b
428	 add		%o1, 0x8, %o1
42973:	andcc		%o2, 0x8, %g0
430	be,pt		%XCC, 1f
431	 nop
432	sub		%o2, 0x8, %o2
433	EX_LD(LOAD(ldx, %o1, %o5))
434	EX_ST(STORE(stx, %o5, %o1 + GLOBAL_SPARE))
435	add		%o1, 0x8, %o1
4361:	andcc		%o2, 0x4, %g0
437	be,pt		%XCC, 1f
438	 nop
439	sub		%o2, 0x4, %o2
440	EX_LD(LOAD(lduw, %o1, %o5))
441	EX_ST(STORE(stw, %o5, %o1 + GLOBAL_SPARE))
442	add		%o1, 0x4, %o1
4431:	cmp		%o2, 0
444	be,pt		%XCC, 85f
445	 nop
446	ba,pt		%xcc, 90f
447	 nop
448
44975:
450	andcc		%o0, 0x7, %g1
451	sub		%g1, 0x8, %g1
452	be,pn		%icc, 2f
453	 sub		%g0, %g1, %g1
454	sub		%o2, %g1, %o2
455
4561:	subcc		%g1, 1, %g1
457	EX_LD(LOAD(ldub, %o1, %o5))
458	EX_ST(STORE(stb, %o5, %o1 + GLOBAL_SPARE))
459	bgu,pt		%icc, 1b
460	 add		%o1, 1, %o1
461
4622:	add		%o1, GLOBAL_SPARE, %o0
463	andcc		%o1, 0x7, %g1
464	bne,pt		%icc, 8f
465	 sll		%g1, 3, %g1
466
467	cmp		%o2, 16
468	bgeu,pt		%icc, 72b
469	 nop
470	ba,a,pt		%xcc, 73b
471
4728:	mov		64, GLOBAL_SPARE
473	andn		%o1, 0x7, %o1
474	EX_LD(LOAD(ldx, %o1, %g2))
475	sub		GLOBAL_SPARE, %g1, GLOBAL_SPARE
476	andn		%o2, 0x7, %o4
477	sllx		%g2, %g1, %g2
4781:	add		%o1, 0x8, %o1
479	EX_LD(LOAD(ldx, %o1, %g3))
480	subcc		%o4, 0x8, %o4
481	srlx		%g3, GLOBAL_SPARE, %o5
482	or		%o5, %g2, %o5
483	EX_ST(STORE(stx, %o5, %o0))
484	add		%o0, 0x8, %o0
485	bgu,pt		%icc, 1b
486	 sllx		%g3, %g1, %g2
487
488	srl		%g1, 3, %g1
489	andcc		%o2, 0x7, %o2
490	be,pn		%icc, 85f
491	 add		%o1, %g1, %o1
492	ba,pt		%xcc, 90f
493	 sub		%o0, %o1, GLOBAL_SPARE
494
495	.align		64
49680: /* 0 < len <= 16 */
497	andcc		GLOBAL_SPARE, 0x3, %g0
498	bne,pn		%XCC, 90f
499	 sub		%o0, %o1, GLOBAL_SPARE
500
5011:
502	subcc		%o2, 4, %o2
503	EX_LD(LOAD(lduw, %o1, %g1))
504	EX_ST(STORE(stw, %g1, %o1 + GLOBAL_SPARE))
505	bgu,pt		%XCC, 1b
506	 add		%o1, 4, %o1
507
50885:	retl
509	 mov		EX_RETVAL(%o3), %o0
510
511	.align		32
51290:
513	subcc		%o2, 1, %o2
514	EX_LD(LOAD(ldub, %o1, %g1))
515	EX_ST(STORE(stb, %g1, %o1 + GLOBAL_SPARE))
516	bgu,pt		%XCC, 90b
517	 add		%o1, 1, %o1
518	retl
519	 mov		EX_RETVAL(%o3), %o0
520
521	.size		FUNC_NAME, .-FUNC_NAME
522