xref: /openbmc/linux/arch/sparc/lib/NG2memcpy.S (revision b34e08d5) 
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	alignaddr	%o1, %g0, %g0
240
241	add		%o1, (64 - 1), %o4
242	andn		%o4, (64 - 1), %o4
243	andn		%o2, (64 - 1), %g1
244	sub		%o2, %g1, %o2
245
246	and		%o1, (64 - 1), %g2
247	add		%o1, %g1, %o1
248	sub		%o0, %o4, %g3
249	brz,pt		%g2, 190f
250	 cmp		%g2, 32
251	blu,a		5f
252	 cmp		%g2, 16
253	cmp		%g2, 48
254	blu,a		4f
255	 cmp		%g2, 40
256	cmp		%g2, 56
257	blu		170f
258	 nop
259	ba,a,pt		%xcc, 180f
260
2614:	/* 32 <= low bits < 48 */
262	blu		150f
263	 nop
264	ba,a,pt		%xcc, 160f
2655:	/* 0 < low bits < 32 */
266	blu,a		6f
267	 cmp		%g2, 8
268	cmp		%g2, 24
269	blu		130f
270	 nop
271	ba,a,pt		%xcc, 140f
2726:	/* 0 < low bits < 16 */
273	bgeu		120f
274	 nop
275	/* fall through for 0 < low bits < 8 */
276110:	sub		%o4, 64, %g2
277	EX_LD(LOAD_BLK(%g2, %f0))
2781:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
279	EX_LD(LOAD_BLK(%o4, %f16))
280	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f14, f16)
281	EX_ST(STORE_BLK(%f0, %o4 + %g3))
282	FREG_MOVE_8(f16, f18, f20, f22, f24, f26, f28, f30)
283	subcc		%g1, 64, %g1
284	add		%o4, 64, %o4
285	bne,pt		%xcc, 1b
286	 LOAD(prefetch, %o4 + 64, #one_read)
287	ba,pt		%xcc, 195f
288	 nop
289
290120:	sub		%o4, 56, %g2
291	FREG_LOAD_7(%g2, f0, f2, f4, f6, f8, f10, f12)
2921:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
293	EX_LD(LOAD_BLK(%o4, %f16))
294	FREG_FROB(f0, f2, f4, f6, f8, f10, f12, f16, f18)
295	EX_ST(STORE_BLK(%f0, %o4 + %g3))
296	FREG_MOVE_7(f18, f20, f22, f24, f26, f28, f30)
297	subcc		%g1, 64, %g1
298	add		%o4, 64, %o4
299	bne,pt		%xcc, 1b
300	 LOAD(prefetch, %o4 + 64, #one_read)
301	ba,pt		%xcc, 195f
302	 nop
303
304130:	sub		%o4, 48, %g2
305	FREG_LOAD_6(%g2, f0, f2, f4, f6, f8, f10)
3061:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
307	EX_LD(LOAD_BLK(%o4, %f16))
308	FREG_FROB(f0, f2, f4, f6, f8, f10, f16, f18, f20)
309	EX_ST(STORE_BLK(%f0, %o4 + %g3))
310	FREG_MOVE_6(f20, f22, f24, f26, f28, f30)
311	subcc		%g1, 64, %g1
312	add		%o4, 64, %o4
313	bne,pt		%xcc, 1b
314	 LOAD(prefetch, %o4 + 64, #one_read)
315	ba,pt		%xcc, 195f
316	 nop
317
318140:	sub		%o4, 40, %g2
319	FREG_LOAD_5(%g2, f0, f2, f4, f6, f8)
3201:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
321	EX_LD(LOAD_BLK(%o4, %f16))
322	FREG_FROB(f0, f2, f4, f6, f8, f16, f18, f20, f22)
323	EX_ST(STORE_BLK(%f0, %o4 + %g3))
324	FREG_MOVE_5(f22, f24, f26, f28, f30)
325	subcc		%g1, 64, %g1
326	add		%o4, 64, %o4
327	bne,pt		%xcc, 1b
328	 LOAD(prefetch, %o4 + 64, #one_read)
329	ba,pt		%xcc, 195f
330	 nop
331
332150:	sub		%o4, 32, %g2
333	FREG_LOAD_4(%g2, f0, f2, f4, f6)
3341:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
335	EX_LD(LOAD_BLK(%o4, %f16))
336	FREG_FROB(f0, f2, f4, f6, f16, f18, f20, f22, f24)
337	EX_ST(STORE_BLK(%f0, %o4 + %g3))
338	FREG_MOVE_4(f24, f26, f28, f30)
339	subcc		%g1, 64, %g1
340	add		%o4, 64, %o4
341	bne,pt		%xcc, 1b
342	 LOAD(prefetch, %o4 + 64, #one_read)
343	ba,pt		%xcc, 195f
344	 nop
345
346160:	sub		%o4, 24, %g2
347	FREG_LOAD_3(%g2, f0, f2, f4)
3481:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
349	EX_LD(LOAD_BLK(%o4, %f16))
350	FREG_FROB(f0, f2, f4, f16, f18, f20, f22, f24, f26)
351	EX_ST(STORE_BLK(%f0, %o4 + %g3))
352	FREG_MOVE_3(f26, f28, f30)
353	subcc		%g1, 64, %g1
354	add		%o4, 64, %o4
355	bne,pt		%xcc, 1b
356	 LOAD(prefetch, %o4 + 64, #one_read)
357	ba,pt		%xcc, 195f
358	 nop
359
360170:	sub		%o4, 16, %g2
361	FREG_LOAD_2(%g2, f0, f2)
3621:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
363	EX_LD(LOAD_BLK(%o4, %f16))
364	FREG_FROB(f0, f2, f16, f18, f20, f22, f24, f26, f28)
365	EX_ST(STORE_BLK(%f0, %o4 + %g3))
366	FREG_MOVE_2(f28, f30)
367	subcc		%g1, 64, %g1
368	add		%o4, 64, %o4
369	bne,pt		%xcc, 1b
370	 LOAD(prefetch, %o4 + 64, #one_read)
371	ba,pt		%xcc, 195f
372	 nop
373
374180:	sub		%o4, 8, %g2
375	FREG_LOAD_1(%g2, f0)
3761:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
377	EX_LD(LOAD_BLK(%o4, %f16))
378	FREG_FROB(f0, f16, f18, f20, f22, f24, f26, f28, f30)
379	EX_ST(STORE_BLK(%f0, %o4 + %g3))
380	FREG_MOVE_1(f30)
381	subcc		%g1, 64, %g1
382	add		%o4, 64, %o4
383	bne,pt		%xcc, 1b
384	 LOAD(prefetch, %o4 + 64, #one_read)
385	ba,pt		%xcc, 195f
386	 nop
387
388190:
3891:	EX_ST(STORE_INIT(%g0, %o4 + %g3))
390	subcc		%g1, 64, %g1
391	EX_LD(LOAD_BLK(%o4, %f0))
392	EX_ST(STORE_BLK(%f0, %o4 + %g3))
393	add		%o4, 64, %o4
394	bne,pt		%xcc, 1b
395	 LOAD(prefetch, %o4 + 64, #one_read)
396
397195:
398	add		%o4, %g3, %o0
399	membar		#Sync
400
401	VISExitHalf
402
403	/* %o2 contains any final bytes still needed to be copied
404	 * over. If anything is left, we copy it one byte at a time.
405	 */
406	brz,pt		%o2, 85f
407	 sub		%o0, %o1, GLOBAL_SPARE
408	ba,a,pt		%XCC, 90f
409
410	.align		64
41175: /* 16 < len <= 64 */
412	bne,pn		%XCC, 75f
413	 sub		%o0, %o1, GLOBAL_SPARE
414
41572:
416	andn		%o2, 0xf, %o4
417	and		%o2, 0xf, %o2
4181:	subcc		%o4, 0x10, %o4
419	EX_LD(LOAD(ldx, %o1, %o5))
420	add		%o1, 0x08, %o1
421	EX_LD(LOAD(ldx, %o1, %g1))
422	sub		%o1, 0x08, %o1
423	EX_ST(STORE(stx, %o5, %o1 + GLOBAL_SPARE))
424	add		%o1, 0x8, %o1
425	EX_ST(STORE(stx, %g1, %o1 + GLOBAL_SPARE))
426	bgu,pt		%XCC, 1b
427	 add		%o1, 0x8, %o1
42873:	andcc		%o2, 0x8, %g0
429	be,pt		%XCC, 1f
430	 nop
431	sub		%o2, 0x8, %o2
432	EX_LD(LOAD(ldx, %o1, %o5))
433	EX_ST(STORE(stx, %o5, %o1 + GLOBAL_SPARE))
434	add		%o1, 0x8, %o1
4351:	andcc		%o2, 0x4, %g0
436	be,pt		%XCC, 1f
437	 nop
438	sub		%o2, 0x4, %o2
439	EX_LD(LOAD(lduw, %o1, %o5))
440	EX_ST(STORE(stw, %o5, %o1 + GLOBAL_SPARE))
441	add		%o1, 0x4, %o1
4421:	cmp		%o2, 0
443	be,pt		%XCC, 85f
444	 nop
445	ba,pt		%xcc, 90f
446	 nop
447
44875:
449	andcc		%o0, 0x7, %g1
450	sub		%g1, 0x8, %g1
451	be,pn		%icc, 2f
452	 sub		%g0, %g1, %g1
453	sub		%o2, %g1, %o2
454
4551:	subcc		%g1, 1, %g1
456	EX_LD(LOAD(ldub, %o1, %o5))
457	EX_ST(STORE(stb, %o5, %o1 + GLOBAL_SPARE))
458	bgu,pt		%icc, 1b
459	 add		%o1, 1, %o1
460
4612:	add		%o1, GLOBAL_SPARE, %o0
462	andcc		%o1, 0x7, %g1
463	bne,pt		%icc, 8f
464	 sll		%g1, 3, %g1
465
466	cmp		%o2, 16
467	bgeu,pt		%icc, 72b
468	 nop
469	ba,a,pt		%xcc, 73b
470
4718:	mov		64, GLOBAL_SPARE
472	andn		%o1, 0x7, %o1
473	EX_LD(LOAD(ldx, %o1, %g2))
474	sub		GLOBAL_SPARE, %g1, GLOBAL_SPARE
475	andn		%o2, 0x7, %o4
476	sllx		%g2, %g1, %g2
4771:	add		%o1, 0x8, %o1
478	EX_LD(LOAD(ldx, %o1, %g3))
479	subcc		%o4, 0x8, %o4
480	srlx		%g3, GLOBAL_SPARE, %o5
481	or		%o5, %g2, %o5
482	EX_ST(STORE(stx, %o5, %o0))
483	add		%o0, 0x8, %o0
484	bgu,pt		%icc, 1b
485	 sllx		%g3, %g1, %g2
486
487	srl		%g1, 3, %g1
488	andcc		%o2, 0x7, %o2
489	be,pn		%icc, 85f
490	 add		%o1, %g1, %o1
491	ba,pt		%xcc, 90f
492	 sub		%o0, %o1, GLOBAL_SPARE
493
494	.align		64
49580: /* 0 < len <= 16 */
496	andcc		GLOBAL_SPARE, 0x3, %g0
497	bne,pn		%XCC, 90f
498	 sub		%o0, %o1, GLOBAL_SPARE
499
5001:
501	subcc		%o2, 4, %o2
502	EX_LD(LOAD(lduw, %o1, %g1))
503	EX_ST(STORE(stw, %g1, %o1 + GLOBAL_SPARE))
504	bgu,pt		%XCC, 1b
505	 add		%o1, 4, %o1
506
50785:	retl
508	 mov		EX_RETVAL(%o3), %o0
509
510	.align		32
51190:
512	subcc		%o2, 1, %o2
513	EX_LD(LOAD(ldub, %o1, %g1))
514	EX_ST(STORE(stb, %g1, %o1 + GLOBAL_SPARE))
515	bgu,pt		%XCC, 90b
516	 add		%o1, 1, %o1
517	retl
518	 mov		EX_RETVAL(%o3), %o0
519
520	.size		FUNC_NAME, .-FUNC_NAME
521