Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * simd.h
4 : * Support for platform-specific vector operations.
5 : *
6 : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : * src/include/port/simd.h
10 : *
11 : * NOTES
12 : * - VectorN in this file refers to a register where the element operands
13 : * are N bits wide. The vector width is platform-specific, so users that care
14 : * about that will need to inspect "sizeof(VectorN)".
15 : *
16 : *-------------------------------------------------------------------------
17 : */
18 : #ifndef SIMD_H
19 : #define SIMD_H
20 :
21 : #if (defined(__x86_64__) || defined(_M_AMD64))
22 : /*
23 : * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
24 : * that compilers targeting this architecture understand SSE2 intrinsics.
25 : *
26 : * We use emmintrin.h rather than the comprehensive header immintrin.h in
27 : * order to exclude extensions beyond SSE2. This is because MSVC, at least,
28 : * will allow the use of intrinsics that haven't been enabled at compile
29 : * time.
30 : */
31 : #include <emmintrin.h>
32 : #define USE_SSE2
33 : typedef __m128i Vector8;
34 : typedef __m128i Vector32;
35 :
36 : #elif defined(__aarch64__) && defined(__ARM_NEON)
37 : /*
38 : * We use the Neon instructions if the compiler provides access to them (as
39 : * indicated by __ARM_NEON) and we are on aarch64. While Neon support is
40 : * technically optional for aarch64, it appears that all available 64-bit
41 : * hardware does have it. Neon exists in some 32-bit hardware too, but we
42 : * could not realistically use it there without a run-time check, which seems
43 : * not worth the trouble for now.
44 : */
45 : #include <arm_neon.h>
46 : #define USE_NEON
47 : typedef uint8x16_t Vector8;
48 : typedef uint32x4_t Vector32;
49 :
50 : #else
51 : /*
52 : * If no SIMD instructions are available, we can in some cases emulate vector
53 : * operations using bitwise operations on unsigned integers. Note that many
54 : * of the functions in this file presently do not have non-SIMD
55 : * implementations. In particular, none of the functions involving Vector32
56 : * are implemented without SIMD since it's likely not worthwhile to represent
57 : * two 32-bit integers using a uint64.
58 : */
59 : #define USE_NO_SIMD
60 : typedef uint64 Vector8;
61 : #endif
62 :
63 : /* load/store operations */
64 : static inline void vector8_load(Vector8 *v, const uint8 *s);
65 : #ifndef USE_NO_SIMD
66 : static inline void vector32_load(Vector32 *v, const uint32 *s);
67 : #endif
68 :
69 : /* assignment operations */
70 : static inline Vector8 vector8_broadcast(const uint8 c);
71 : #ifndef USE_NO_SIMD
72 : static inline Vector32 vector32_broadcast(const uint32 c);
73 : #endif
74 :
75 : /* element-wise comparisons to a scalar */
76 : static inline bool vector8_has(const Vector8 v, const uint8 c);
77 : static inline bool vector8_has_zero(const Vector8 v);
78 : static inline bool vector8_has_le(const Vector8 v, const uint8 c);
79 : static inline bool vector8_is_highbit_set(const Vector8 v);
80 : #ifndef USE_NO_SIMD
81 : static inline bool vector32_is_highbit_set(const Vector32 v);
82 : static inline uint32 vector8_highbit_mask(const Vector8 v);
83 : #endif
84 :
85 : /* arithmetic operations */
86 : static inline Vector8 vector8_or(const Vector8 v1, const Vector8 v2);
87 : #ifndef USE_NO_SIMD
88 : static inline Vector32 vector32_or(const Vector32 v1, const Vector32 v2);
89 : static inline Vector8 vector8_ssub(const Vector8 v1, const Vector8 v2);
90 : #endif
91 :
92 : /*
93 : * comparisons between vectors
94 : *
95 : * Note: These return a vector rather than boolean, which is why we don't
96 : * have non-SIMD implementations.
97 : */
98 : #ifndef USE_NO_SIMD
99 : static inline Vector8 vector8_eq(const Vector8 v1, const Vector8 v2);
100 : static inline Vector8 vector8_min(const Vector8 v1, const Vector8 v2);
101 : static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
102 : #endif
103 :
104 : /*
105 : * Load a chunk of memory into the given vector.
106 : */
107 : static inline void
108 22817886 : vector8_load(Vector8 *v, const uint8 *s)
109 : {
110 : #if defined(USE_SSE2)
111 22817886 : *v = _mm_loadu_si128((const __m128i *) s);
112 : #elif defined(USE_NEON)
113 : *v = vld1q_u8(s);
114 : #else
115 : memcpy(v, s, sizeof(Vector8));
116 : #endif
117 22817886 : }
118 :
119 : #ifndef USE_NO_SIMD
120 : static inline void
121 272 : vector32_load(Vector32 *v, const uint32 *s)
122 : {
123 : #ifdef USE_SSE2
124 272 : *v = _mm_loadu_si128((const __m128i *) s);
125 : #elif defined(USE_NEON)
126 : *v = vld1q_u32(s);
127 : #endif
128 272 : }
129 : #endif /* ! USE_NO_SIMD */
130 :
131 : /*
132 : * Create a vector with all elements set to the same value.
133 : */
134 : static inline Vector8
135 20220584 : vector8_broadcast(const uint8 c)
136 : {
137 : #if defined(USE_SSE2)
138 40441168 : return _mm_set1_epi8(c);
139 : #elif defined(USE_NEON)
140 : return vdupq_n_u8(c);
141 : #else
142 : return ~UINT64CONST(0) / 0xFF * c;
143 : #endif
144 : }
145 :
146 : #ifndef USE_NO_SIMD
147 : static inline Vector32
148 10200708 : vector32_broadcast(const uint32 c)
149 : {
150 : #ifdef USE_SSE2
151 20401416 : return _mm_set1_epi32(c);
152 : #elif defined(USE_NEON)
153 : return vdupq_n_u32(c);
154 : #endif
155 : }
156 : #endif /* ! USE_NO_SIMD */
157 :
158 : /*
159 : * Return true if any elements in the vector are equal to the given scalar.
160 : */
161 : static inline bool
162 5117482 : vector8_has(const Vector8 v, const uint8 c)
163 : {
164 : bool result;
165 :
166 : /* pre-compute the result for assert checking */
167 : #ifdef USE_ASSERT_CHECKING
168 : bool assert_result = false;
169 :
170 : for (Size i = 0; i < sizeof(Vector8); i++)
171 : {
172 : if (((const uint8 *) &v)[i] == c)
173 : {
174 : assert_result = true;
175 : break;
176 : }
177 : }
178 : #endif /* USE_ASSERT_CHECKING */
179 :
180 : #if defined(USE_NO_SIMD)
181 : /* any bytes in v equal to c will evaluate to zero via XOR */
182 : result = vector8_has_zero(v ^ vector8_broadcast(c));
183 : #else
184 5117482 : result = vector8_is_highbit_set(vector8_eq(v, vector8_broadcast(c)));
185 : #endif
186 :
187 : Assert(assert_result == result);
188 5117482 : return result;
189 : }
190 :
191 : /*
192 : * Convenience function equivalent to vector8_has(v, 0)
193 : */
194 : static inline bool
195 485620 : vector8_has_zero(const Vector8 v)
196 : {
197 : #if defined(USE_NO_SIMD)
198 : /*
199 : * We cannot call vector8_has() here, because that would lead to a
200 : * circular definition.
201 : */
202 : return vector8_has_le(v, 0);
203 : #else
204 485620 : return vector8_has(v, 0);
205 : #endif
206 : }
207 :
208 : /*
209 : * Return true if any elements in the vector are less than or equal to the
210 : * given scalar.
211 : */
212 : static inline bool
213 485620 : vector8_has_le(const Vector8 v, const uint8 c)
214 : {
215 485620 : bool result = false;
216 :
217 : /* pre-compute the result for assert checking */
218 : #ifdef USE_ASSERT_CHECKING
219 : bool assert_result = false;
220 :
221 : for (Size i = 0; i < sizeof(Vector8); i++)
222 : {
223 : if (((const uint8 *) &v)[i] <= c)
224 : {
225 : assert_result = true;
226 : break;
227 : }
228 : }
229 : #endif /* USE_ASSERT_CHECKING */
230 :
231 : #if defined(USE_NO_SIMD)
232 :
233 : /*
234 : * To find bytes <= c, we can use bitwise operations to find bytes < c+1,
235 : * but it only works if c+1 <= 128 and if the highest bit in v is not set.
236 : * Adapted from
237 : * https://graphics.stanford.edu/~seander/bithacks.html#HasLessInWord
238 : */
239 : if ((int64) v >= 0 && c < 0x80)
240 : result = (v - vector8_broadcast(c + 1)) & ~v & vector8_broadcast(0x80);
241 : else
242 : {
243 : /* one byte at a time */
244 : for (Size i = 0; i < sizeof(Vector8); i++)
245 : {
246 : if (((const uint8 *) &v)[i] <= c)
247 : {
248 : result = true;
249 : break;
250 : }
251 : }
252 : }
253 : #else
254 :
255 : /*
256 : * Use saturating subtraction to find bytes <= c, which will present as
257 : * NUL bytes. This approach is a workaround for the lack of unsigned
258 : * comparison instructions on some architectures.
259 : */
260 485620 : result = vector8_has_zero(vector8_ssub(v, vector8_broadcast(c)));
261 : #endif
262 :
263 : Assert(assert_result == result);
264 485620 : return result;
265 : }
266 :
267 : /*
268 : * Return true if the high bit of any element is set
269 : */
270 : static inline bool
271 8001190 : vector8_is_highbit_set(const Vector8 v)
272 : {
273 : #ifdef USE_SSE2
274 8001190 : return _mm_movemask_epi8(v) != 0;
275 : #elif defined(USE_NEON)
276 : return vmaxvq_u8(v) > 0x7F;
277 : #else
278 : return v & vector8_broadcast(0x80);
279 : #endif
280 : }
281 :
282 : /*
283 : * Exactly like vector8_is_highbit_set except for the input type, so it
284 : * looks at each byte separately.
285 : *
286 : * XXX x86 uses the same underlying type for 8-bit, 16-bit, and 32-bit
287 : * integer elements, but Arm does not, hence the need for a separate
288 : * function. We could instead adopt the behavior of Arm's vmaxvq_u32(), i.e.
289 : * check each 32-bit element, but that would require an additional mask
290 : * operation on x86.
291 : */
292 : #ifndef USE_NO_SIMD
293 : static inline bool
294 68 : vector32_is_highbit_set(const Vector32 v)
295 : {
296 : #if defined(USE_NEON)
297 : return vector8_is_highbit_set((Vector8) v);
298 : #else
299 68 : return vector8_is_highbit_set(v);
300 : #endif
301 : }
302 : #endif /* ! USE_NO_SIMD */
303 :
304 : /*
305 : * Return a bitmask formed from the high-bit of each element.
306 : */
307 : #ifndef USE_NO_SIMD
308 : static inline uint32
309 11933124 : vector8_highbit_mask(const Vector8 v)
310 : {
311 : #ifdef USE_SSE2
312 11933124 : return (uint32) _mm_movemask_epi8(v);
313 : #elif defined(USE_NEON)
314 : /*
315 : * Note: It would be faster to use vget_lane_u64 and vshrn_n_u16, but that
316 : * returns a uint64, making it inconvenient to combine mask values from
317 : * multiple vectors.
318 : */
319 : static const uint8 mask[16] = {
320 : 1 << 0, 1 << 1, 1 << 2, 1 << 3,
321 : 1 << 4, 1 << 5, 1 << 6, 1 << 7,
322 : 1 << 0, 1 << 1, 1 << 2, 1 << 3,
323 : 1 << 4, 1 << 5, 1 << 6, 1 << 7,
324 : };
325 :
326 : uint8x16_t masked = vandq_u8(vld1q_u8(mask), (uint8x16_t) vshrq_n_s8((int8x16_t) v, 7));
327 : uint8x16_t maskedhi = vextq_u8(masked, masked, 8);
328 :
329 : return (uint32) vaddvq_u16((uint16x8_t) vzip1q_u8(masked, maskedhi));
330 : #endif
331 : }
332 : #endif /* ! USE_NO_SIMD */
333 :
334 : /*
335 : * Return the bitwise OR of the inputs
336 : */
337 : static inline Vector8
338 11534560 : vector8_or(const Vector8 v1, const Vector8 v2)
339 : {
340 : #ifdef USE_SSE2
341 11534560 : return _mm_or_si128(v1, v2);
342 : #elif defined(USE_NEON)
343 : return vorrq_u8(v1, v2);
344 : #else
345 : return v1 | v2;
346 : #endif
347 : }
348 :
349 : #ifndef USE_NO_SIMD
350 : static inline Vector32
351 204 : vector32_or(const Vector32 v1, const Vector32 v2)
352 : {
353 : #ifdef USE_SSE2
354 204 : return _mm_or_si128(v1, v2);
355 : #elif defined(USE_NEON)
356 : return vorrq_u32(v1, v2);
357 : #endif
358 : }
359 : #endif /* ! USE_NO_SIMD */
360 :
361 : /*
362 : * Return the result of subtracting the respective elements of the input
363 : * vectors using saturation (i.e., if the operation would yield a value less
364 : * than zero, zero is returned instead). For more information on saturation
365 : * arithmetic, see https://en.wikipedia.org/wiki/Saturation_arithmetic
366 : */
367 : #ifndef USE_NO_SIMD
368 : static inline Vector8
369 485620 : vector8_ssub(const Vector8 v1, const Vector8 v2)
370 : {
371 : #ifdef USE_SSE2
372 485620 : return _mm_subs_epu8(v1, v2);
373 : #elif defined(USE_NEON)
374 : return vqsubq_u8(v1, v2);
375 : #endif
376 : }
377 : #endif /* ! USE_NO_SIMD */
378 :
379 : /*
380 : * Return a vector with all bits set in each lane where the corresponding
381 : * lanes in the inputs are equal.
382 : */
383 : #ifndef USE_NO_SIMD
384 : static inline Vector8
385 22817886 : vector8_eq(const Vector8 v1, const Vector8 v2)
386 : {
387 : #ifdef USE_SSE2
388 22817886 : return _mm_cmpeq_epi8(v1, v2);
389 : #elif defined(USE_NEON)
390 : return vceqq_u8(v1, v2);
391 : #endif
392 : }
393 : #endif /* ! USE_NO_SIMD */
394 :
395 : #ifndef USE_NO_SIMD
396 : static inline Vector32
397 272 : vector32_eq(const Vector32 v1, const Vector32 v2)
398 : {
399 : #ifdef USE_SSE2
400 272 : return _mm_cmpeq_epi32(v1, v2);
401 : #elif defined(USE_NEON)
402 : return vceqq_u32(v1, v2);
403 : #endif
404 : }
405 : #endif /* ! USE_NO_SIMD */
406 :
407 : /*
408 : * Given two vectors, return a vector with the minimum element of each.
409 : */
410 : #ifndef USE_NO_SIMD
411 : static inline Vector8
412 209452 : vector8_min(const Vector8 v1, const Vector8 v2)
413 : {
414 : #ifdef USE_SSE2
415 209452 : return _mm_min_epu8(v1, v2);
416 : #elif defined(USE_NEON)
417 : return vminq_u8(v1, v2);
418 : #endif
419 : }
420 : #endif /* ! USE_NO_SIMD */
421 :
422 : #endif /* SIMD_H */
|
