1/*
2 * Copyright (c) 2013
3 *      MIPS Technologies, Inc., California.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
14 *    contributors may be used to endorse or promote products derived from
15 *    this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#ifdef __ANDROID__
31# include <private/bionic_asm.h>
32# define PREFETCH_STORE_HINT PREFETCH_HINT_PREPAREFORSTORE
33#elif _LIBC
34# include <sysdep.h>
35# include <regdef.h>
36# include <sys/asm.h>
37# define PREFETCH_STORE_HINT PREFETCH_HINT_PREPAREFORSTORE
38#elif _COMPILING_NEWLIB
39# include "machine/asm.h"
40# include "machine/regdef.h"
41# define PREFETCH_STORE_HINT PREFETCH_HINT_PREPAREFORSTORE
42#else
43# include <regdef.h>
44# include <sys/asm.h>
45#endif
46
47/* Check to see if the MIPS architecture we are compiling for supports
48   prefetching.  */
49
50#if (__mips == 4) || (__mips == 5) || (__mips == 32) || (__mips == 64)
51# ifndef DISABLE_PREFETCH
52#  define USE_PREFETCH
53# endif
54#endif
55
56#if defined(_MIPS_SIM) && ((_MIPS_SIM == _ABI64) || (_MIPS_SIM == _ABIN32))
57# ifndef DISABLE_DOUBLE
58#  define USE_DOUBLE
59# endif
60#endif
61
62#ifndef USE_DOUBLE
63# ifndef DISABLE_DOUBLE_ALIGN
64#  define DOUBLE_ALIGN
65# endif
66#endif
67
68/* Some asm.h files do not have the L macro definition.  */
69#ifndef L
70# if _MIPS_SIM == _ABIO32
71#  define L(label) $L ## label
72# else
73#  define L(label) .L ## label
74# endif
75#endif
76
77/* Some asm.h files do not have the PTR_ADDIU macro definition.  */
78#ifndef PTR_ADDIU
79# if _MIPS_SIM == _ABIO32
80#  define PTR_ADDIU	addiu
81# else
82#  define PTR_ADDIU	daddiu
83# endif
84#endif
85
86/* New R6 instructions that may not be in asm.h.  */
87#ifndef PTR_LSA
88# if _MIPS_SIM == _ABIO32
89#  define PTR_LSA        lsa
90# else
91#  define PTR_LSA        dlsa
92# endif
93#endif
94
95/* Using PREFETCH_HINT_PREPAREFORSTORE instead of PREFETCH_STORE
96   or PREFETCH_STORE_STREAMED offers a large performance advantage
97   but PREPAREFORSTORE has some special restrictions to consider.
98
99   Prefetch with the 'prepare for store' hint does not copy a memory
100   location into the cache, it just allocates a cache line and zeros
101   it out.  This means that if you do not write to the entire cache
102   line before writing it out to memory some data will get zero'ed out
103   when the cache line is written back to memory and data will be lost.
104
105   There are ifdef'ed sections of this memcpy to make sure that it does not
106   do prefetches on cache lines that are not going to be completely written.
107   This code is only needed and only used when PREFETCH_STORE_HINT is set to
108   PREFETCH_HINT_PREPAREFORSTORE.  This code assumes that cache lines are
109   less than MAX_PREFETCH_SIZE bytes and if the cache line is larger it will
110   not work correctly.  */
111
112#ifdef USE_PREFETCH
113# define PREFETCH_HINT_STORE		1
114# define PREFETCH_HINT_STORE_STREAMED	5
115# define PREFETCH_HINT_STORE_RETAINED	7
116# define PREFETCH_HINT_PREPAREFORSTORE	30
117
118/* If we have not picked out what hints to use at this point use the
119   standard load and store prefetch hints.  */
120# ifndef PREFETCH_STORE_HINT
121#  define PREFETCH_STORE_HINT PREFETCH_HINT_STORE
122# endif
123
124/* We double everything when USE_DOUBLE is true so we do 2 prefetches to
125   get 64 bytes in that case.  The assumption is that each individual
126   prefetch brings in 32 bytes.  */
127# ifdef USE_DOUBLE
128#  define PREFETCH_CHUNK 64
129#  define PREFETCH_FOR_STORE(chunk, reg) \
130    pref PREFETCH_STORE_HINT, (chunk)*64(reg); \
131    pref PREFETCH_STORE_HINT, ((chunk)*64)+32(reg)
132# else
133#  define PREFETCH_CHUNK 32
134#  define PREFETCH_FOR_STORE(chunk, reg) \
135    pref PREFETCH_STORE_HINT, (chunk)*32(reg)
136# endif
137
138/* MAX_PREFETCH_SIZE is the maximum size of a prefetch, it must not be less
139   than PREFETCH_CHUNK, the assumed size of each prefetch.  If the real size
140   of a prefetch is greater than MAX_PREFETCH_SIZE and the PREPAREFORSTORE
141   hint is used, the code will not work correctly.  If PREPAREFORSTORE is not
142   used than MAX_PREFETCH_SIZE does not matter.  */
143# define MAX_PREFETCH_SIZE 128
144/* PREFETCH_LIMIT is set based on the fact that we never use an offset greater
145   than 5 on a STORE prefetch and that a single prefetch can never be larger
146   than MAX_PREFETCH_SIZE.  We add the extra 32 when USE_DOUBLE is set because
147   we actually do two prefetches in that case, one 32 bytes after the other.  */
148# ifdef USE_DOUBLE
149#  define PREFETCH_LIMIT (5 * PREFETCH_CHUNK) + 32 + MAX_PREFETCH_SIZE
150# else
151#  define PREFETCH_LIMIT (5 * PREFETCH_CHUNK) + MAX_PREFETCH_SIZE
152# endif
153
154# if (PREFETCH_STORE_HINT == PREFETCH_HINT_PREPAREFORSTORE) \
155    && ((PREFETCH_CHUNK * 4) < MAX_PREFETCH_SIZE)
156/* We cannot handle this because the initial prefetches may fetch bytes that
157   are before the buffer being copied.  We start copies with an offset
158   of 4 so avoid this situation when using PREPAREFORSTORE.  */
159#  error "PREFETCH_CHUNK is too large and/or MAX_PREFETCH_SIZE is too small."
160# endif
161#else /* USE_PREFETCH not defined */
162# define PREFETCH_FOR_STORE(offset, reg)
163#endif
164
165#if __mips_isa_rev > 5
166# if (PREFETCH_STORE_HINT == PREFETCH_HINT_PREPAREFORSTORE)
167#  undef PREFETCH_STORE_HINT
168#  define PREFETCH_STORE_HINT PREFETCH_HINT_STORE_STREAMED
169# endif
170# define R6_CODE
171#endif
172
173/* Allow the routine to be named something else if desired.  */
174#ifndef MEMSET_NAME
175# define MEMSET_NAME memset
176#endif
177
178/* We load/store 64 bits at a time when USE_DOUBLE is true.
179   The C_ prefix stands for CHUNK and is used to avoid macro name
180   conflicts with system header files.  */
181
182#ifdef USE_DOUBLE
183# define C_ST	sd
184# if __MIPSEB
185#  define C_STHI	sdl	/* high part is left in big-endian	*/
186# else
187#  define C_STHI	sdr	/* high part is right in little-endian	*/
188# endif
189#else
190# define C_ST	sw
191# if __MIPSEB
192#  define C_STHI	swl	/* high part is left in big-endian	*/
193# else
194#  define C_STHI	swr	/* high part is right in little-endian	*/
195# endif
196#endif
197
198/* Bookkeeping values for 32 vs. 64 bit mode.  */
199#ifdef USE_DOUBLE
200# define NSIZE 8
201# define NSIZEMASK 0x3f
202# define NSIZEDMASK 0x7f
203#else
204# define NSIZE 4
205# define NSIZEMASK 0x1f
206# define NSIZEDMASK 0x3f
207#endif
208#define UNIT(unit) ((unit)*NSIZE)
209#define UNITM1(unit) (((unit)*NSIZE)-1)
210
211#ifdef __ANDROID__
212LEAF(MEMSET_NAME,0)
213#else
214LEAF(MEMSET_NAME)
215#endif
216
217	.set	nomips16
218	.set	noreorder
219/* If the size is less than 2*NSIZE (8 or 16), go to L(lastb).  Regardless of
220   size, copy dst pointer to v0 for the return value.  */
221	slti	t2,a2,(2 * NSIZE)
222	bne	t2,zero,L(lastb)
223	move	v0,a0
224
225/* If memset value is not zero, we copy it to all the bytes in a 32 or 64
226   bit word.  */
227	beq	a1,zero,L(set0)		/* If memset value is zero no smear  */
228	PTR_SUBU a3,zero,a0
229	nop
230
231	/* smear byte into 32 or 64 bit word */
232#if ((__mips == 64) || (__mips == 32)) && (__mips_isa_rev >= 2)
233# ifdef USE_DOUBLE
234	dins	a1, a1, 8, 8        /* Replicate fill byte into half-word.  */
235	dins	a1, a1, 16, 16      /* Replicate fill byte into word.       */
236	dins	a1, a1, 32, 32      /* Replicate fill byte into dbl word.   */
237# else
238	ins	a1, a1, 8, 8        /* Replicate fill byte into half-word.  */
239	ins	a1, a1, 16, 16      /* Replicate fill byte into word.       */
240# endif
241#else
242# ifdef USE_DOUBLE
243        and     a1,0xff
244	dsll	t2,a1,8
245	or	a1,t2
246	dsll	t2,a1,16
247	or	a1,t2
248	dsll	t2,a1,32
249	or	a1,t2
250# else
251        and     a1,0xff
252	sll	t2,a1,8
253	or	a1,t2
254	sll	t2,a1,16
255	or	a1,t2
256# endif
257#endif
258
259/* If the destination address is not aligned do a partial store to get it
260   aligned.  If it is already aligned just jump to L(aligned).  */
261L(set0):
262#ifndef R6_CODE
263	andi	t2,a3,(NSIZE-1)		/* word-unaligned address?          */
264	beq	t2,zero,L(aligned)	/* t2 is the unalignment count      */
265	PTR_SUBU a2,a2,t2
266	C_STHI	a1,0(a0)
267	PTR_ADDU a0,a0,t2
268#else /* R6_CODE */
269	andi	t2,a0,(NSIZE-1)
270	lapc	t9,L(atable)
271	PTR_LSA	t9,t2,t9,2
272	jrc	t9
273L(atable):
274	bc	L(aligned)
275# ifdef USE_DOUBLE
276	bc	L(lb7)
277	bc	L(lb6)
278	bc	L(lb5)
279	bc	L(lb4)
280# endif
281	bc	L(lb3)
282	bc	L(lb2)
283	bc	L(lb1)
284L(lb7):
285	sb	a1,6(a0)
286L(lb6):
287	sb	a1,5(a0)
288L(lb5):
289	sb	a1,4(a0)
290L(lb4):
291	sb	a1,3(a0)
292L(lb3):
293	sb	a1,2(a0)
294L(lb2):
295	sb	a1,1(a0)
296L(lb1):
297	sb	a1,0(a0)
298
299	li	t9,NSIZE
300	subu	t2,t9,t2
301	PTR_SUBU a2,a2,t2
302	PTR_ADDU a0,a0,t2
303#endif /* R6_CODE */
304
305L(aligned):
306/* If USE_DOUBLE is not set we may still want to align the data on a 16
307   byte boundry instead of an 8 byte boundry to maximize the opportunity
308   of proAptiv chips to do memory bonding (combining two sequential 4
309   byte stores into one 8 byte store).  We know there are at least 4 bytes
310   left to store or we would have jumped to L(lastb) earlier in the code.  */
311#ifdef DOUBLE_ALIGN
312	andi	t2,a3,4
313	beq	t2,zero,L(double_aligned)
314	PTR_SUBU a2,a2,t2
315	sw	a1,0(a0)
316	PTR_ADDU a0,a0,t2
317L(double_aligned):
318#endif
319
320/* Now the destination is aligned to (word or double word) aligned address
321   Set a2 to count how many bytes we have to copy after all the 64/128 byte
322   chunks are copied and a3 to the dest pointer after all the 64/128 byte
323   chunks have been copied.  We will loop, incrementing a0 until it equals
324   a3.  */
325	andi	t8,a2,NSIZEDMASK /* any whole 64-byte/128-byte chunks? */
326	beq	a2,t8,L(chkw)	 /* if a2==t8, no 64-byte/128-byte chunks */
327	PTR_SUBU a3,a2,t8	 /* subtract from a2 the reminder */
328	PTR_ADDU a3,a0,a3	 /* Now a3 is the final dst after loop */
329
330/* When in the loop we may prefetch with the 'prepare to store' hint,
331   in this case the a0+x should not be past the "t0-32" address.  This
332   means: for x=128 the last "safe" a0 address is "t0-160".  Alternatively,
333   for x=64 the last "safe" a0 address is "t0-96" In the current version we
334   will use "prefetch hint,128(a0)", so "t0-160" is the limit.  */
335#if defined(USE_PREFETCH) \
336    && (PREFETCH_STORE_HINT == PREFETCH_HINT_PREPAREFORSTORE)
337	PTR_ADDU t0,a0,a2		/* t0 is the "past the end" address */
338	PTR_SUBU t9,t0,PREFETCH_LIMIT	/* t9 is the "last safe pref" address */
339#endif
340#if defined(USE_PREFETCH) \
341    && (PREFETCH_STORE_HINT != PREFETCH_HINT_PREPAREFORSTORE)
342	PREFETCH_FOR_STORE (1, a0)
343	PREFETCH_FOR_STORE (2, a0)
344	PREFETCH_FOR_STORE (3, a0)
345#endif
346
347L(loop16w):
348#if defined(USE_PREFETCH) \
349    && (PREFETCH_STORE_HINT == PREFETCH_HINT_PREPAREFORSTORE)
350	sltu	v1,t9,a0		/* If a0 > t9 don't use next prefetch */
351	bgtz	v1,L(skip_pref)
352	nop
353#endif
354#ifndef R6_CODE
355	PREFETCH_FOR_STORE (4, a0)
356	PREFETCH_FOR_STORE (5, a0)
357#else
358	PREFETCH_FOR_STORE (2, a0)
359#endif
360L(skip_pref):
361	C_ST	a1,UNIT(0)(a0)
362	C_ST	a1,UNIT(1)(a0)
363	C_ST	a1,UNIT(2)(a0)
364	C_ST	a1,UNIT(3)(a0)
365	C_ST	a1,UNIT(4)(a0)
366	C_ST	a1,UNIT(5)(a0)
367	C_ST	a1,UNIT(6)(a0)
368	C_ST	a1,UNIT(7)(a0)
369	C_ST	a1,UNIT(8)(a0)
370	C_ST	a1,UNIT(9)(a0)
371	C_ST	a1,UNIT(10)(a0)
372	C_ST	a1,UNIT(11)(a0)
373	C_ST	a1,UNIT(12)(a0)
374	C_ST	a1,UNIT(13)(a0)
375	C_ST	a1,UNIT(14)(a0)
376	C_ST	a1,UNIT(15)(a0)
377	PTR_ADDIU a0,a0,UNIT(16)	/* adding 64/128 to dest */
378	bne	a0,a3,L(loop16w)
379	nop
380	move	a2,t8
381
382/* Here we have dest word-aligned but less than 64-bytes or 128 bytes to go.
383   Check for a 32(64) byte chunk and copy if if there is one.  Otherwise
384   jump down to L(chk1w) to handle the tail end of the copy.  */
385L(chkw):
386	andi	t8,a2,NSIZEMASK	/* is there a 32-byte/64-byte chunk.  */
387				/* the t8 is the reminder count past 32-bytes */
388	beq	a2,t8,L(chk1w)/* when a2==t8, no 32-byte chunk */
389	nop
390	C_ST	a1,UNIT(0)(a0)
391	C_ST	a1,UNIT(1)(a0)
392	C_ST	a1,UNIT(2)(a0)
393	C_ST	a1,UNIT(3)(a0)
394	C_ST	a1,UNIT(4)(a0)
395	C_ST	a1,UNIT(5)(a0)
396	C_ST	a1,UNIT(6)(a0)
397	C_ST	a1,UNIT(7)(a0)
398	PTR_ADDIU a0,a0,UNIT(8)
399
400/* Here we have less than 32(64) bytes to set.  Set up for a loop to
401   copy one word (or double word) at a time.  Set a2 to count how many
402   bytes we have to copy after all the word (or double word) chunks are
403   copied and a3 to the dest pointer after all the (d)word chunks have
404   been copied.  We will loop, incrementing a0 until a0 equals a3.  */
405L(chk1w):
406	andi	a2,t8,(NSIZE-1)	/* a2 is the reminder past one (d)word chunks */
407	beq	a2,t8,L(lastb)
408	PTR_SUBU a3,t8,a2	/* a3 is count of bytes in one (d)word chunks */
409	PTR_ADDU a3,a0,a3	/* a3 is the dst address after loop */
410
411/* copying in words (4-byte or 8 byte chunks) */
412L(wordCopy_loop):
413	PTR_ADDIU a0,a0,UNIT(1)
414	bne	a0,a3,L(wordCopy_loop)
415	C_ST	a1,UNIT(-1)(a0)
416
417/* Copy the last 8 (or 16) bytes */
418L(lastb):
419	blez	a2,L(leave)
420	PTR_ADDU a3,a0,a2       /* a3 is the last dst address */
421L(lastbloop):
422	PTR_ADDIU a0,a0,1
423	bne	a0,a3,L(lastbloop)
424	sb	a1,-1(a0)
425L(leave):
426	j	ra
427	nop
428
429	.set	at
430	.set	reorder
431END(MEMSET_NAME)
432#ifndef __ANDROID__
433# ifdef _LIBC
434libc_hidden_builtin_def (MEMSET_NAME)
435# endif
436#endif
437