1 /*
2   (C) Copyright 2001,2006,
3   International Business Machines Corporation,
4   Sony Computer Entertainment, Incorporated,
5   Toshiba Corporation,
6 
7   All rights reserved.
8 
9   Redistribution and use in source and binary forms, with or without
10   modification, are permitted provided that the following conditions are met:
11 
12     * Redistributions of source code must retain the above copyright notice,
13   this list of conditions and the following disclaimer.
14     * Redistributions in binary form must reproduce the above copyright
15   notice, this list of conditions and the following disclaimer in the
16   documentation and/or other materials provided with the distribution.
17     * Neither the names of the copyright holders nor the names of their
18   contributors may be used to endorse or promote products derived from this
19   software without specific prior written permission.
20 
21   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
22   AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23   IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24   ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
25   LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26   CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27   SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28   INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29   CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30   ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31   POSSIBILITY OF SUCH DAMAGE.
32 */
33 #include <picolibc.h>
34 
35 #include <spu_intrinsics.h>
36 #include <stddef.h>
37 #include "vec_literal.h"
38 
39 /* Copy n bytes from memory area src to memory area dest.
40  * Copying is performed as if the n characters pointed to
41  * by src are first copied into a temporary array that does
42  * not overlap the src and dest arrays. Then the n characters
43  * of the temporary array are copied into the destination
44  * array. The memmove subroutine returns a pointer to dest.
45  */
46 
memmove(void * __restrict__ dest,const void * __restrict__ src,size_t n)47 void * memmove(void * __restrict__ dest, const void * __restrict__ src, size_t n)
48 {
49   int adjust, delta;
50   unsigned int soffset1, soffset2, doffset1, doffset2;
51   vec_uchar16 *vSrc, *vDst;
52   vec_uchar16 sdata1, sdata2, sdata, ddata, shuffle;
53   vec_uchar16 mask, mask1, mask2, mask3, one = spu_splats((unsigned char)-1);
54 
55   soffset1  = (unsigned int)(src) & 15;
56   doffset1 = (unsigned int)(dest) & 15;
57   doffset2 = ((unsigned int)(dest) + n) & 15;
58 
59   /* Construct a series of masks used to data insert. The masks
60    * contains 0 bit when the destination word is unchanged, 1 when it
61    * must be replaced by source bits.
62    *
63    * mask1 = mask for leading unchanged bytes
64    * mask2 = mask for trailing unchange bytes
65    * mask3 = mask indicating the more than one qword is being changed.
66    */
67   mask  = one;
68   mask1 = spu_rlmaskqwbyte(mask, -doffset1);
69   mask2 = spu_slqwbyte(mask, 16-doffset2);
70   mask3 = (vec_uchar16)spu_cmpgt(spu_splats((unsigned int)(doffset1 + n)), 15);
71 
72   vDst = (vec_uchar16 *)(dest);
73 
74   delta  = (int)soffset1 - (int)doffset1;
75 
76   /* The follow check only works if the SPU addresses are not
77    * wrapped. No provisions have been made to correct for this
78    * limitation.
79    */
80   if (((unsigned int)dest - (unsigned int)src) >= (unsigned int)n) {
81     /* Forward copy. Perform a memcpy.
82      *
83      * Handle any leading destination partial quadwords as
84      * well a very short copy (ie, such that the n characters
85      * all reside in a single (destination) quadword.
86      */
87     vSrc = (vec_uchar16 *)(src);
88     vDst = (vec_uchar16 *)(dest);
89 
90     /* Handle any leading destination partial quadwords as
91      * well a very short copy (ie, such that the n characters
92      * all reside in a single (destination) quadword.
93      */
94     soffset1 = (unsigned int)(src) & 15;
95     doffset1 = (unsigned int)(dest) & 15;
96     doffset2 = ((unsigned int)(dest) + n) & 15;
97 
98     /* Compute a shuffle pattern used to align the source string
99      * with the alignment of the destination string.
100      */
101 
102     adjust = (int)spu_extract(spu_cmpgt(spu_promote(doffset1, 0), spu_promote(soffset1, 0)), 0);
103     delta  = (int)soffset1 - (int)doffset1;
104     delta += adjust & 16;
105 
106     shuffle = (vec_uchar16)spu_add((vec_uint4)spu_splats((unsigned char)delta),
107 				   VEC_LITERAL(vec_uint4, 0x00010203, 0x04050607, 0x08090A0B, 0x0C0D0E0F));
108 
109     vSrc += adjust;
110 
111     sdata1 = *vSrc++;
112     sdata2 = *vSrc++;
113 
114     ddata = *vDst;
115     sdata = spu_shuffle(sdata1, sdata2, shuffle);
116 
117     /* Construct a series of masks used to data insert. The masks
118      * contain 0 when the destination word is unchanged, 1 when it
119      * must be replaced by source bytes.
120      *
121      * mask1 = mask for leading unchanged bytes
122      * mask2 = mask for trailing unchange bytes
123      * mask3 = mask indicating the more than one qword is being changed.
124      */
125     mask  = one;
126     mask1 = spu_rlmaskqwbyte(mask, -doffset1);
127     mask2 = spu_slqwbyte(mask, 16-doffset2);
128     mask3 = (vec_uchar16)spu_cmpgt(spu_splats((unsigned int)(doffset1 + n)), 15);
129 
130     *vDst++ = spu_sel(ddata, sdata, spu_and(mask1, spu_or(mask2, mask3)));
131 
132     n += doffset1;
133 
134     /* Handle complete destination quadwords
135      */
136     while (n > 31) {
137       sdata1 = sdata2;
138       sdata2 = *vSrc++;
139       *vDst++ = spu_shuffle(sdata1, sdata2, shuffle);
140       n -= 16;
141     }
142 
143     /* Handle any trailing partial (destination) quadwords
144      */
145     mask = spu_and((vec_uchar16)spu_cmpgt(spu_splats((unsigned int)n), 16), mask2);
146     *vDst = spu_sel(*vDst, spu_shuffle(sdata2, *vSrc, shuffle), mask);
147 
148   } else {
149     /* Backward copy.
150      *
151      * Handle any leading destination partial quadwords as
152      * well a very short copy (ie, such that the n characters
153      * all reside in a single (destination) quadword.
154      */
155     vSrc = (vec_uchar16 *)((unsigned int)src  + n-1);
156     vDst = (vec_uchar16 *)((unsigned int)dest + n-1);
157 
158     /* Handle any leading destination partial quadwords as
159      * well a very short copy (ie, such that the n characters
160      * all reside in a single (destination) quadword.
161      */
162     soffset1 = (unsigned int)(src)  & 15;
163     soffset2 = (unsigned int)(vSrc) & 15;
164     doffset1 = (unsigned int)(dest) & 15;
165     doffset2 = (unsigned int)(vDst) & 15;
166 
167     /* Compute a shuffle pattern used to align the source string
168      * with the alignment of the destination string.
169      */
170     adjust = (int)spu_extract(spu_cmpgt(spu_promote(soffset2, 0), spu_promote(doffset2, 0)), 0);
171     delta  = (int)doffset2 - (int)soffset2;
172     delta += adjust & 16;
173 
174     shuffle = (vec_uchar16)spu_sub(VEC_LITERAL(vec_uint4, 0x10111213, 0x14151617, 0x18191A1B, 0x1C1D1E1F),
175 				   (vec_uint4)spu_splats((unsigned char)delta));
176 
177     vSrc -= adjust;
178 
179     sdata2 = *vSrc--;
180     sdata1 = *vSrc--;
181 
182     ddata = *vDst;
183     sdata = spu_shuffle(sdata1, sdata2, shuffle);
184 
185     /* Construct a series of masks used to data insert. The masks
186      * contain 0 when the destination word is unchanged, 1 when it
187      * must be replaced by source bytes.
188      *
189      * mask1 = mask for leading unchanged bytes
190      * mask2 = mask for trailing unchange bytes
191      * mask3 = mask indicating the more than one qword is being changed.
192      */
193     mask  = one;
194     mask1 = spu_rlmaskqwbyte(mask, -doffset1);
195     mask2 = spu_slqwbyte(mask, 15-doffset2);
196     mask3 = (vec_uchar16)spu_cmpgt(spu_splats((int)(doffset2 - n)), -2);
197 
198     *vDst-- = spu_sel(ddata, sdata, spu_and(mask2, spu_orc(mask1, mask3)));
199 
200     n -= doffset2 + 1;
201 
202     /* Handle complete destination quadwords
203      */
204     while ((int)n > 15) {
205       sdata2 = sdata1;
206       sdata1 = *vSrc--;
207       *vDst-- = spu_shuffle(sdata1, sdata2, shuffle);
208       n -= 16;
209     }
210 
211     /* Handle any trailing partial (destination) quadwords
212      */
213     mask = spu_and((vec_uchar16)spu_cmpgt(spu_splats((int)n), 0), mask1);
214     *vDst = spu_sel(*vDst, spu_shuffle(*vSrc, sdata1, shuffle), mask);
215   }
216   return (dest);
217 }
218 
219