1 /*
2 * Copyright 2015 Red Hat Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 *
23 * Author: Oded Gabbay <oded.gabbay@redhat.com>
24 */
25
26 /**
27 * @file
28 * POWER8 intrinsics portability header.
29 *
30 */
31
32 #ifndef U_PWR8_H_
33 #define U_PWR8_H_
34
35 #if defined(_ARCH_PWR8) && UTIL_ARCH_LITTLE_ENDIAN
36
37 #define VECTOR_ALIGN_16 __attribute__ ((__aligned__ (16)))
38
39 typedef VECTOR_ALIGN_16 vector unsigned char __m128i;
40
41 typedef VECTOR_ALIGN_16 union m128i {
42 __m128i m128i;
43 vector signed int m128si;
44 vector unsigned int m128ui;
45 ubyte ub[16];
46 ushort us[8];
47 int i[4];
48 uint ui[4];
49 } __m128i_union;
50
51 static inline __m128i
vec_set_epi32(int i3,int i2,int i1,int i0)52 vec_set_epi32 (int i3, int i2, int i1, int i0)
53 {
54 __m128i_union vdst;
55
56 #if UTIL_ARCH_LITTLE_ENDIAN
57 vdst.i[0] = i0;
58 vdst.i[1] = i1;
59 vdst.i[2] = i2;
60 vdst.i[3] = i3;
61 #else
62 vdst.i[3] = i0;
63 vdst.i[2] = i1;
64 vdst.i[1] = i2;
65 vdst.i[0] = i3;
66 #endif
67
68 return (__m128i) vdst.m128si;
69 }
70
71 static inline __m128i
vec_setr_epi32(int i0,int i1,int i2,int i3)72 vec_setr_epi32 (int i0, int i1, int i2, int i3)
73 {
74 return vec_set_epi32 (i3, i2, i1, i0);
75 }
76
77 static inline __m128i
vec_unpacklo_epi32(__m128i even,__m128i odd)78 vec_unpacklo_epi32 (__m128i even, __m128i odd)
79 {
80 static const __m128i perm_mask =
81 #if UTIL_ARCH_LITTLE_ENDIAN
82 { 0, 1, 2, 3, 16, 17, 18, 19, 4, 5, 6, 7, 20, 21, 22, 23};
83 #else
84 {24, 25, 26, 27, 8, 9, 10, 11, 28, 29, 30, 31, 12, 13, 14, 15};
85 #endif
86
87 return vec_perm (even, odd, perm_mask);
88 }
89
90 static inline __m128i
vec_unpackhi_epi32(__m128i even,__m128i odd)91 vec_unpackhi_epi32 (__m128i even, __m128i odd)
92 {
93 static const __m128i perm_mask =
94 #if UTIL_ARCH_LITTLE_ENDIAN
95 { 8, 9, 10, 11, 24, 25, 26, 27, 12, 13, 14, 15, 28, 29, 30, 31};
96 #else
97 {16, 17, 18, 19, 0, 1, 2, 3, 20, 21, 22, 23, 4, 5, 6, 7};
98 #endif
99
100 return vec_perm (even, odd, perm_mask);
101 }
102
103 static inline __m128i
vec_unpacklo_epi64(__m128i even,__m128i odd)104 vec_unpacklo_epi64 (__m128i even, __m128i odd)
105 {
106 static const __m128i perm_mask =
107 #if UTIL_ARCH_LITTLE_ENDIAN
108 { 0, 1, 2, 3, 4, 5, 6, 7, 16, 17, 18, 19, 20, 21, 22, 23};
109 #else
110 {24, 25, 26, 27, 28, 29, 30, 31, 8, 9, 10, 11, 12, 13, 14, 15};
111 #endif
112
113 return vec_perm (even, odd, perm_mask);
114 }
115
116 static inline __m128i
vec_unpackhi_epi64(__m128i even,__m128i odd)117 vec_unpackhi_epi64 (__m128i even, __m128i odd)
118 {
119 static const __m128i perm_mask =
120 #if UTIL_ARCH_LITTLE_ENDIAN
121 { 8, 9, 10, 11, 12, 13, 14, 15, 24, 25, 26, 27, 28, 29, 30, 31};
122 #else
123 {16, 17, 18, 19, 20, 21, 22, 23, 0, 1, 2, 3, 4, 5, 6, 7};
124 #endif
125
126 return vec_perm (even, odd, perm_mask);
127 }
128
129 static inline __m128i
vec_add_epi32(__m128i a,__m128i b)130 vec_add_epi32 (__m128i a, __m128i b)
131 {
132 return (__m128i) vec_add ((vector signed int) a, (vector signed int) b);
133 }
134
135 static inline __m128i
vec_sub_epi32(__m128i a,__m128i b)136 vec_sub_epi32 (__m128i a, __m128i b)
137 {
138 return (__m128i) vec_sub ((vector signed int) a, (vector signed int) b);
139 }
140
141 /* Call this function ONLY on POWER8 and newer platforms */
142 static inline __m128i
vec_mullo_epi32(__m128i a,__m128i b)143 vec_mullo_epi32 (__m128i a, __m128i b)
144 {
145 __m128i v;
146
147 __asm__(
148 "vmuluwm %0, %1, %2 \n"
149 : "=v" (v)
150 : "v" (a), "v" (b)
151 );
152
153 return v;
154 }
155
156 static inline __m128i
vec_andnot_si128(__m128i a,__m128i b)157 vec_andnot_si128 (__m128i a, __m128i b)
158 {
159 return vec_andc (b, a);
160 }
161
162 static inline void
transpose4_epi32(const __m128i * restrict a,const __m128i * restrict b,const __m128i * restrict c,const __m128i * restrict d,__m128i * restrict o,__m128i * restrict p,__m128i * restrict q,__m128i * restrict r)163 transpose4_epi32(const __m128i * restrict a,
164 const __m128i * restrict b,
165 const __m128i * restrict c,
166 const __m128i * restrict d,
167 __m128i * restrict o,
168 __m128i * restrict p,
169 __m128i * restrict q,
170 __m128i * restrict r)
171 {
172 __m128i t0 = vec_unpacklo_epi32(*a, *b);
173 __m128i t1 = vec_unpacklo_epi32(*c, *d);
174 __m128i t2 = vec_unpackhi_epi32(*a, *b);
175 __m128i t3 = vec_unpackhi_epi32(*c, *d);
176
177 *o = vec_unpacklo_epi64(t0, t1);
178 *p = vec_unpackhi_epi64(t0, t1);
179 *q = vec_unpacklo_epi64(t2, t3);
180 *r = vec_unpackhi_epi64(t2, t3);
181 }
182
183 static inline __m128i
vec_slli_epi32(__m128i vsrc,unsigned int count)184 vec_slli_epi32 (__m128i vsrc, unsigned int count)
185 {
186 __m128i_union vec_count;
187
188 if (count >= 32)
189 return (__m128i) vec_splats (0);
190 else if (count == 0)
191 return vsrc;
192
193 /* In VMX, all shift count fields must contain the same value */
194 vec_count.m128si = (vector signed int) vec_splats (count);
195 return (__m128i) vec_sl ((vector signed int) vsrc, vec_count.m128ui);
196 }
197
198 static inline __m128i
vec_srli_epi32(__m128i vsrc,unsigned int count)199 vec_srli_epi32 (__m128i vsrc, unsigned int count)
200 {
201 __m128i_union vec_count;
202
203 if (count >= 32)
204 return (__m128i) vec_splats (0);
205 else if (count == 0)
206 return vsrc;
207
208 /* In VMX, all shift count fields must contain the same value */
209 vec_count.m128si = (vector signed int) vec_splats (count);
210 return (__m128i) vec_sr ((vector signed int) vsrc, vec_count.m128ui);
211 }
212
213 static inline __m128i
vec_srai_epi32(__m128i vsrc,unsigned int count)214 vec_srai_epi32 (__m128i vsrc, unsigned int count)
215 {
216 __m128i_union vec_count;
217
218 if (count >= 32)
219 return (__m128i) vec_splats (0);
220 else if (count == 0)
221 return vsrc;
222
223 /* In VMX, all shift count fields must contain the same value */
224 vec_count.m128si = (vector signed int) vec_splats (count);
225 return (__m128i) vec_sra ((vector signed int) vsrc, vec_count.m128ui);
226 }
227
228 static inline __m128i
vec_cmpeq_epi32(__m128i a,__m128i b)229 vec_cmpeq_epi32 (__m128i a, __m128i b)
230 {
231 return (__m128i) vec_cmpeq ((vector signed int) a, (vector signed int) b);
232 }
233
234 static inline __m128i
vec_loadu_si128(const uint32_t * src)235 vec_loadu_si128 (const uint32_t* src)
236 {
237 __m128i_union vsrc;
238
239 #if UTIL_ARCH_LITTLE_ENDIAN
240
241 vsrc.m128ui = *((vector unsigned int *) src);
242
243 #else
244
245 __m128i vmask, tmp1, tmp2;
246
247 vmask = vec_lvsl(0, src);
248
249 tmp1 = (__m128i) vec_ld (0, src);
250 tmp2 = (__m128i) vec_ld (15, src);
251 vsrc.m128ui = (vector unsigned int) vec_perm (tmp1, tmp2, vmask);
252
253 #endif
254
255 return vsrc.m128i;
256 }
257
258 static inline __m128i
vec_load_si128(const uint32_t * src)259 vec_load_si128 (const uint32_t* src)
260 {
261 __m128i_union vsrc;
262
263 vsrc.m128ui = *((vector unsigned int *) src);
264
265 return vsrc.m128i;
266 }
267
268 static inline void
vec_store_si128(uint32_t * dest,__m128i vdata)269 vec_store_si128 (uint32_t* dest, __m128i vdata)
270 {
271 vec_st ((vector unsigned int) vdata, 0, dest);
272 }
273
274 /* Call this function ONLY on POWER8 and newer platforms */
275 static inline int
vec_movemask_epi8(__m128i vsrc)276 vec_movemask_epi8 (__m128i vsrc)
277 {
278 __m128i_union vtemp;
279 int result;
280
281 vtemp.m128i = vec_vgbbd(vsrc);
282
283 #if UTIL_ARCH_LITTLE_ENDIAN
284 result = vtemp.ub[15] << 8 | vtemp.ub[7];
285 #else
286 result = vtemp.ub[0] << 8 | vtemp.ub[8];
287 #endif
288
289 return result;
290 }
291
292 static inline __m128i
vec_packs_epi16(__m128i a,__m128i b)293 vec_packs_epi16 (__m128i a, __m128i b)
294 {
295 #if UTIL_ARCH_LITTLE_ENDIAN
296 return (__m128i) vec_packs ((vector signed short) a,
297 (vector signed short) b);
298 #else
299 return (__m128i) vec_packs ((vector signed short) b,
300 (vector signed short) a);
301 #endif
302 }
303
304 static inline __m128i
vec_packs_epi32(__m128i a,__m128i b)305 vec_packs_epi32 (__m128i a, __m128i b)
306 {
307 #if UTIL_ARCH_LITTLE_ENDIAN
308 return (__m128i) vec_packs ((vector signed int) a, (vector signed int) b);
309 #else
310 return (__m128i) vec_packs ((vector signed int) b, (vector signed int) a);
311 #endif
312 }
313
314 #endif /* _ARCH_PWR8 && UTIL_ARCH_LITTLE_ENDIAN */
315
316 #endif /* U_PWR8_H_ */
317