xref: /openbmc/linux/arch/sparc/lib/NG4memcpy.S (revision 23c2b932) 
1/* NG4memcpy.S: Niagara-4 optimized memcpy.
2 *
3 * Copyright (C) 2012 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_BLK_INIT_QUAD_LDD_P 0xe2
12#define FPRS_FEF  0x04
13
14/* On T4 it is very expensive to access ASRs like %fprs and
15 * %asi, avoiding a read or a write can save ~50 cycles.
16 */
17#define FPU_ENTER			\
18	rd	%fprs, %o5;		\
19	andcc	%o5, FPRS_FEF, %g0;	\
20	be,a,pn	%icc, 999f;		\
21	 wr	%g0, FPRS_FEF, %fprs;	\
22	999:
23
24#ifdef MEMCPY_DEBUG
25#define VISEntryHalf FPU_ENTER; \
26		     clr %g1; clr %g2; clr %g3; clr %g5; subcc %g0, %g0, %g0;
27#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
28#else
29#define VISEntryHalf FPU_ENTER
30#define VISExitHalf and %o5, FPRS_FEF, %o5; wr %o5, 0x0, %fprs
31#endif
32
33#define GLOBAL_SPARE	%g5
34#endif
35
36#ifndef STORE_ASI
37#ifndef SIMULATE_NIAGARA_ON_NON_NIAGARA
38#define STORE_ASI	ASI_BLK_INIT_QUAD_LDD_P
39#else
40#define STORE_ASI	0x80		/* ASI_P */
41#endif
42#endif
43
44#if !defined(EX_LD) && !defined(EX_ST)
45#define NON_USER_COPY
46#endif
47
48#ifndef EX_LD
49#define EX_LD(x)	x
50#endif
51#ifndef EX_LD_FP
52#define EX_LD_FP(x)	x
53#endif
54
55#ifndef EX_ST
56#define EX_ST(x)	x
57#endif
58#ifndef EX_ST_FP
59#define EX_ST_FP(x)	x
60#endif
61
62#ifndef EX_RETVAL
63#define EX_RETVAL(x)	x
64#endif
65
66#ifndef LOAD
67#define LOAD(type,addr,dest)	type [addr], dest
68#endif
69
70#ifndef STORE
71#ifndef MEMCPY_DEBUG
72#define STORE(type,src,addr)	type src, [addr]
73#else
74#define STORE(type,src,addr)	type##a src, [addr] %asi
75#endif
76#endif
77
78#ifndef STORE_INIT
79#define STORE_INIT(src,addr)	stxa src, [addr] STORE_ASI
80#endif
81
82#ifndef FUNC_NAME
83#define FUNC_NAME	NG4memcpy
84#endif
85#ifndef PREAMBLE
86#define PREAMBLE
87#endif
88
89#ifndef XCC
90#define XCC xcc
91#endif
92
93	.register	%g2,#scratch
94	.register	%g3,#scratch
95
96	.text
97	.align		64
98
99	.globl	FUNC_NAME
100	.type	FUNC_NAME,#function
101FUNC_NAME:	/* %o0=dst, %o1=src, %o2=len */
102#ifdef MEMCPY_DEBUG
103	wr		%g0, 0x80, %asi
104#endif
105	srlx		%o2, 31, %g2
106	cmp		%g2, 0
107	tne		%XCC, 5
108	PREAMBLE
109	mov		%o0, %o3
110	brz,pn		%o2, .Lexit
111	 cmp		%o2, 3
112	ble,pn		%icc, .Ltiny
113	 cmp		%o2, 19
114	ble,pn		%icc, .Lsmall
115	 or		%o0, %o1, %g2
116	cmp		%o2, 128
117	bl,pn		%icc, .Lmedium
118	 nop
119
120.Llarge:/* len >= 0x80 */
121	/* First get dest 8 byte aligned.  */
122	sub		%g0, %o0, %g1
123	and		%g1, 0x7, %g1
124	brz,pt		%g1, 51f
125	 sub		%o2, %g1, %o2
126
1271:	EX_LD(LOAD(ldub, %o1 + 0x00, %g2))
128	add		%o1, 1, %o1
129	subcc		%g1, 1, %g1
130	add		%o0, 1, %o0
131	bne,pt		%icc, 1b
132	 EX_ST(STORE(stb, %g2, %o0 - 0x01))
133
13451:	LOAD(prefetch, %o1 + 0x040, #n_reads_strong)
135	LOAD(prefetch, %o1 + 0x080, #n_reads_strong)
136	LOAD(prefetch, %o1 + 0x0c0, #n_reads_strong)
137	LOAD(prefetch, %o1 + 0x100, #n_reads_strong)
138	LOAD(prefetch, %o1 + 0x140, #n_reads_strong)
139	LOAD(prefetch, %o1 + 0x180, #n_reads_strong)
140	LOAD(prefetch, %o1 + 0x1c0, #n_reads_strong)
141	LOAD(prefetch, %o1 + 0x200, #n_reads_strong)
142
143	/* Check if we can use the straight fully aligned
144	 * loop, or we require the alignaddr/faligndata variant.
145	 */
146	andcc		%o1, 0x7, %o5
147	bne,pn		%icc, .Llarge_src_unaligned
148	 sub		%g0, %o0, %g1
149
150	/* Legitimize the use of initializing stores by getting dest
151	 * to be 64-byte aligned.
152	 */
153	and		%g1, 0x3f, %g1
154	brz,pt		%g1, .Llarge_aligned
155	 sub		%o2, %g1, %o2
156
1571:	EX_LD(LOAD(ldx, %o1 + 0x00, %g2))
158	add		%o1, 8, %o1
159	subcc		%g1, 8, %g1
160	add		%o0, 8, %o0
161	bne,pt		%icc, 1b
162	 EX_ST(STORE(stx, %g2, %o0 - 0x08))
163
164.Llarge_aligned:
165	/* len >= 0x80 && src 8-byte aligned && dest 8-byte aligned */
166	andn		%o2, 0x3f, %o4
167	sub		%o2, %o4, %o2
168
1691:	EX_LD(LOAD(ldx, %o1 + 0x00, %g1))
170	add		%o1, 0x40, %o1
171	EX_LD(LOAD(ldx, %o1 - 0x38, %g2))
172	subcc		%o4, 0x40, %o4
173	EX_LD(LOAD(ldx, %o1 - 0x30, %g3))
174	EX_LD(LOAD(ldx, %o1 - 0x28, GLOBAL_SPARE))
175	EX_LD(LOAD(ldx, %o1 - 0x20, %o5))
176	EX_ST(STORE_INIT(%g1, %o0))
177	add		%o0, 0x08, %o0
178	EX_ST(STORE_INIT(%g2, %o0))
179	add		%o0, 0x08, %o0
180	EX_LD(LOAD(ldx, %o1 - 0x18, %g2))
181	EX_ST(STORE_INIT(%g3, %o0))
182	add		%o0, 0x08, %o0
183	EX_LD(LOAD(ldx, %o1 - 0x10, %g3))
184	EX_ST(STORE_INIT(GLOBAL_SPARE, %o0))
185	add		%o0, 0x08, %o0
186	EX_LD(LOAD(ldx, %o1 - 0x08, GLOBAL_SPARE))
187	EX_ST(STORE_INIT(%o5, %o0))
188	add		%o0, 0x08, %o0
189	EX_ST(STORE_INIT(%g2, %o0))
190	add		%o0, 0x08, %o0
191	EX_ST(STORE_INIT(%g3, %o0))
192	add		%o0, 0x08, %o0
193	EX_ST(STORE_INIT(GLOBAL_SPARE, %o0))
194	add		%o0, 0x08, %o0
195	bne,pt		%icc, 1b
196	 LOAD(prefetch, %o1 + 0x200, #n_reads_strong)
197
198	membar		#StoreLoad | #StoreStore
199
200	brz,pn		%o2, .Lexit
201	 cmp		%o2, 19
202	ble,pn		%icc, .Lsmall_unaligned
203	 nop
204	ba,a,pt		%icc, .Lmedium_noprefetch
205
206.Lexit:	retl
207	 mov		EX_RETVAL(%o3), %o0
208
209.Llarge_src_unaligned:
210#ifdef NON_USER_COPY
211	VISEntryHalfFast(.Lmedium_vis_entry_fail)
212#else
213	VISEntryHalf
214#endif
215	andn		%o2, 0x3f, %o4
216	sub		%o2, %o4, %o2
217	alignaddr	%o1, %g0, %g1
218	add		%o1, %o4, %o1
219	EX_LD_FP(LOAD(ldd, %g1 + 0x00, %f0))
2201:	EX_LD_FP(LOAD(ldd, %g1 + 0x08, %f2))
221	subcc		%o4, 0x40, %o4
222	EX_LD_FP(LOAD(ldd, %g1 + 0x10, %f4))
223	EX_LD_FP(LOAD(ldd, %g1 + 0x18, %f6))
224	EX_LD_FP(LOAD(ldd, %g1 + 0x20, %f8))
225	EX_LD_FP(LOAD(ldd, %g1 + 0x28, %f10))
226	EX_LD_FP(LOAD(ldd, %g1 + 0x30, %f12))
227	EX_LD_FP(LOAD(ldd, %g1 + 0x38, %f14))
228	faligndata	%f0, %f2, %f16
229	EX_LD_FP(LOAD(ldd, %g1 + 0x40, %f0))
230	faligndata	%f2, %f4, %f18
231	add		%g1, 0x40, %g1
232	faligndata	%f4, %f6, %f20
233	faligndata	%f6, %f8, %f22
234	faligndata	%f8, %f10, %f24
235	faligndata	%f10, %f12, %f26
236	faligndata	%f12, %f14, %f28
237	faligndata	%f14, %f0, %f30
238	EX_ST_FP(STORE(std, %f16, %o0 + 0x00))
239	EX_ST_FP(STORE(std, %f18, %o0 + 0x08))
240	EX_ST_FP(STORE(std, %f20, %o0 + 0x10))
241	EX_ST_FP(STORE(std, %f22, %o0 + 0x18))
242	EX_ST_FP(STORE(std, %f24, %o0 + 0x20))
243	EX_ST_FP(STORE(std, %f26, %o0 + 0x28))
244	EX_ST_FP(STORE(std, %f28, %o0 + 0x30))
245	EX_ST_FP(STORE(std, %f30, %o0 + 0x38))
246	add		%o0, 0x40, %o0
247	bne,pt		%icc, 1b
248	 LOAD(prefetch, %g1 + 0x200, #n_reads_strong)
249#ifdef NON_USER_COPY
250	VISExitHalfFast
251#else
252	VISExitHalf
253#endif
254	brz,pn		%o2, .Lexit
255	 cmp		%o2, 19
256	ble,pn		%icc, .Lsmall_unaligned
257	 nop
258	ba,a,pt		%icc, .Lmedium_unaligned
259
260#ifdef NON_USER_COPY
261.Lmedium_vis_entry_fail:
262	 or		%o0, %o1, %g2
263#endif
264.Lmedium:
265	LOAD(prefetch, %o1 + 0x40, #n_reads_strong)
266	andcc		%g2, 0x7, %g0
267	bne,pn		%icc, .Lmedium_unaligned
268	 nop
269.Lmedium_noprefetch:
270	andncc		%o2, 0x20 - 1, %o5
271	be,pn		%icc, 2f
272	 sub		%o2, %o5, %o2
2731:	EX_LD(LOAD(ldx, %o1 + 0x00, %g1))
274	EX_LD(LOAD(ldx, %o1 + 0x08, %g2))
275	EX_LD(LOAD(ldx, %o1 + 0x10, GLOBAL_SPARE))
276	EX_LD(LOAD(ldx, %o1 + 0x18, %o4))
277	add		%o1, 0x20, %o1
278	subcc		%o5, 0x20, %o5
279	EX_ST(STORE(stx, %g1, %o0 + 0x00))
280	EX_ST(STORE(stx, %g2, %o0 + 0x08))
281	EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x10))
282	EX_ST(STORE(stx, %o4, %o0 + 0x18))
283	bne,pt		%icc, 1b
284	 add		%o0, 0x20, %o0
2852:	andcc		%o2, 0x18, %o5
286	be,pt		%icc, 3f
287	 sub		%o2, %o5, %o2
2881:	EX_LD(LOAD(ldx, %o1 + 0x00, %g1))
289	add		%o1, 0x08, %o1
290	add		%o0, 0x08, %o0
291	subcc		%o5, 0x08, %o5
292	bne,pt		%icc, 1b
293	 EX_ST(STORE(stx, %g1, %o0 - 0x08))
2943:	brz,pt		%o2, .Lexit
295	 cmp		%o2, 0x04
296	bl,pn		%icc, .Ltiny
297	 nop
298	EX_LD(LOAD(lduw, %o1 + 0x00, %g1))
299	add		%o1, 0x04, %o1
300	add		%o0, 0x04, %o0
301	subcc		%o2, 0x04, %o2
302	bne,pn		%icc, .Ltiny
303	 EX_ST(STORE(stw, %g1, %o0 - 0x04))
304	ba,a,pt		%icc, .Lexit
305.Lmedium_unaligned:
306	/* First get dest 8 byte aligned.  */
307	sub		%g0, %o0, %g1
308	and		%g1, 0x7, %g1
309	brz,pt		%g1, 2f
310	 sub		%o2, %g1, %o2
311
3121:	EX_LD(LOAD(ldub, %o1 + 0x00, %g2))
313	add		%o1, 1, %o1
314	subcc		%g1, 1, %g1
315	add		%o0, 1, %o0
316	bne,pt		%icc, 1b
317	 EX_ST(STORE(stb, %g2, %o0 - 0x01))
3182:
319	and		%o1, 0x7, %g1
320	brz,pn		%g1, .Lmedium_noprefetch
321	 sll		%g1, 3, %g1
322	mov		64, %g2
323	sub		%g2, %g1, %g2
324	andn		%o1, 0x7, %o1
325	EX_LD(LOAD(ldx, %o1 + 0x00, %o4))
326	sllx		%o4, %g1, %o4
327	andn		%o2, 0x08 - 1, %o5
328	sub		%o2, %o5, %o2
3291:	EX_LD(LOAD(ldx, %o1 + 0x08, %g3))
330	add		%o1, 0x08, %o1
331	subcc		%o5, 0x08, %o5
332	srlx		%g3, %g2, GLOBAL_SPARE
333	or		GLOBAL_SPARE, %o4, GLOBAL_SPARE
334	EX_ST(STORE(stx, GLOBAL_SPARE, %o0 + 0x00))
335	add		%o0, 0x08, %o0
336	bne,pt		%icc, 1b
337	 sllx		%g3, %g1, %o4
338	srl		%g1, 3, %g1
339	add		%o1, %g1, %o1
340	brz,pn		%o2, .Lexit
341	 nop
342	ba,pt		%icc, .Lsmall_unaligned
343
344.Ltiny:
345	EX_LD(LOAD(ldub, %o1 + 0x00, %g1))
346	subcc		%o2, 1, %o2
347	be,pn		%icc, .Lexit
348	 EX_ST(STORE(stb, %g1, %o0 + 0x00))
349	EX_LD(LOAD(ldub, %o1 + 0x01, %g1))
350	subcc		%o2, 1, %o2
351	be,pn		%icc, .Lexit
352	 EX_ST(STORE(stb, %g1, %o0 + 0x01))
353	EX_LD(LOAD(ldub, %o1 + 0x02, %g1))
354	ba,pt		%icc, .Lexit
355	 EX_ST(STORE(stb, %g1, %o0 + 0x02))
356
357.Lsmall:
358	andcc		%g2, 0x3, %g0
359	bne,pn		%icc, .Lsmall_unaligned
360	 andn		%o2, 0x4 - 1, %o5
361	sub		%o2, %o5, %o2
3621:
363	EX_LD(LOAD(lduw, %o1 + 0x00, %g1))
364	add		%o1, 0x04, %o1
365	subcc		%o5, 0x04, %o5
366	add		%o0, 0x04, %o0
367	bne,pt		%icc, 1b
368	 EX_ST(STORE(stw, %g1, %o0 - 0x04))
369	brz,pt		%o2, .Lexit
370	 nop
371	ba,a,pt		%icc, .Ltiny
372
373.Lsmall_unaligned:
3741:	EX_LD(LOAD(ldub, %o1 + 0x00, %g1))
375	add		%o1, 1, %o1
376	add		%o0, 1, %o0
377	subcc		%o2, 1, %o2
378	bne,pt		%icc, 1b
379	 EX_ST(STORE(stb, %g1, %o0 - 0x01))
380	ba,a,pt		%icc, .Lexit
381	.size		FUNC_NAME, .-FUNC_NAME
382