[i386] Fold __builtin_ia32_shufpd to VEC_PERM_EXPR
[official-gcc.git] / gcc / vector-builder.h
blob9967daa6e4c993800f8046de57ae52ee21c793f5
1 /* A class for building vector constant patterns.
2 Copyright (C) 2017-2019 Free Software Foundation, Inc.
4 This file is part of GCC.
6 GCC is free software; you can redistribute it and/or modify it under
7 the terms of the GNU General Public License as published by the Free
8 Software Foundation; either version 3, or (at your option) any later
9 version.
11 GCC is distributed in the hope that it will be useful, but WITHOUT ANY
12 WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
14 for more details.
16 You should have received a copy of the GNU General Public License
17 along with GCC; see the file COPYING3. If not see
18 <http://www.gnu.org/licenses/>. */
20 #ifndef GCC_VECTOR_BUILDER_H
21 #define GCC_VECTOR_BUILDER_H
23 /* This class is a wrapper around auto_vec<T> for building vectors of T.
24 It aims to encode each vector as npatterns interleaved patterns,
25 where each pattern represents a sequence:
27 { BASE0, BASE1, BASE1 + STEP, BASE1 + STEP*2, BASE1 + STEP*3, ... }
29 The first three elements in each pattern provide enough information
30 to derive the other elements. If all patterns have a STEP of zero,
31 we only need to encode the first two elements in each pattern.
32 If BASE1 is also equal to BASE0 for all patterns, we only need to
33 encode the first element in each pattern. The number of encoded
34 elements per pattern is given by nelts_per_pattern.
36 The class can be used in two ways:
38 1. It can be used to build a full image of the vector, which is then
39 canonicalized by finalize (). In this case npatterns is initially
40 the number of elements in the vector and nelts_per_pattern is
41 initially 1.
43 2. It can be used to build a vector that already has a known encoding.
44 This is preferred since it is more efficient and copes with
45 variable-length vectors. finalize () then canonicalizes the encoding
46 to a simpler form if possible.
48 The derived class Derived provides this functionality for specific Ts.
49 Derived needs to provide the following interface:
51 bool equal_p (T elt1, T elt2) const;
53 Return true if elements ELT1 and ELT2 are equal.
55 bool allow_steps_p () const;
57 Return true if a stepped representation is OK. We don't allow
58 linear series for anything other than integers, to avoid problems
59 with rounding.
61 bool integral_p (T elt) const;
63 Return true if element ELT can be interpreted as an integer.
65 StepType step (T elt1, T elt2) const;
67 Return the value of element ELT2 minus the value of element ELT1,
68 given integral_p (ELT1) && integral_p (ELT2). There is no fixed
69 choice of StepType.
71 T apply_step (T base, unsigned int factor, StepType step) const;
73 Return a vector element with the value BASE + FACTOR * STEP.
75 bool can_elide_p (T elt) const;
77 Return true if we can drop element ELT, even if the retained
78 elements are different. This is provided for TREE_OVERFLOW
79 handling.
81 void note_representative (T *elt1_ptr, T elt2);
83 Record that ELT2 is being elided, given that ELT1_PTR points to
84 the last encoded element for the containing pattern. This is
85 again provided for TREE_OVERFLOW handling. */
87 template<typename T, typename Derived>
88 class vector_builder : public auto_vec<T, 32>
90 public:
91 vector_builder ();
93 poly_uint64 full_nelts () const { return m_full_nelts; }
94 unsigned int npatterns () const { return m_npatterns; }
95 unsigned int nelts_per_pattern () const { return m_nelts_per_pattern; }
96 unsigned int encoded_nelts () const;
97 bool encoded_full_vector_p () const;
98 T elt (unsigned int) const;
100 bool operator == (const Derived &) const;
101 bool operator != (const Derived &x) const { return !operator == (x); }
103 void finalize ();
105 protected:
106 void new_vector (poly_uint64, unsigned int, unsigned int);
107 void reshape (unsigned int, unsigned int);
108 bool repeating_sequence_p (unsigned int, unsigned int, unsigned int);
109 bool stepped_sequence_p (unsigned int, unsigned int, unsigned int);
110 bool try_npatterns (unsigned int);
112 private:
113 vector_builder (const vector_builder &);
114 vector_builder &operator= (const vector_builder &);
115 Derived *derived () { return static_cast<Derived *> (this); }
116 const Derived *derived () const;
118 poly_uint64 m_full_nelts;
119 unsigned int m_npatterns;
120 unsigned int m_nelts_per_pattern;
123 template<typename T, typename Derived>
124 inline const Derived *
125 vector_builder<T, Derived>::derived () const
127 return static_cast<const Derived *> (this);
130 template<typename T, typename Derived>
131 inline
132 vector_builder<T, Derived>::vector_builder ()
133 : m_full_nelts (0),
134 m_npatterns (0),
135 m_nelts_per_pattern (0)
138 /* Return the number of elements that are explicitly encoded. The vec
139 starts with these explicitly-encoded elements and may contain additional
140 elided elements. */
142 template<typename T, typename Derived>
143 inline unsigned int
144 vector_builder<T, Derived>::encoded_nelts () const
146 return m_npatterns * m_nelts_per_pattern;
149 /* Return true if every element of the vector is explicitly encoded. */
151 template<typename T, typename Derived>
152 inline bool
153 vector_builder<T, Derived>::encoded_full_vector_p () const
155 return known_eq (m_npatterns * m_nelts_per_pattern, m_full_nelts);
158 /* Start building a vector that has FULL_NELTS elements. Initially
159 encode it using NPATTERNS patterns with NELTS_PER_PATTERN each. */
161 template<typename T, typename Derived>
162 void
163 vector_builder<T, Derived>::new_vector (poly_uint64 full_nelts,
164 unsigned int npatterns,
165 unsigned int nelts_per_pattern)
167 m_full_nelts = full_nelts;
168 m_npatterns = npatterns;
169 m_nelts_per_pattern = nelts_per_pattern;
170 this->reserve (encoded_nelts ());
171 this->truncate (0);
174 /* Return true if this vector and OTHER have the same elements and
175 are encoded in the same way. */
177 template<typename T, typename Derived>
178 bool
179 vector_builder<T, Derived>::operator == (const Derived &other) const
181 if (maybe_ne (m_full_nelts, other.m_full_nelts)
182 || m_npatterns != other.m_npatterns
183 || m_nelts_per_pattern != other.m_nelts_per_pattern)
184 return false;
186 unsigned int nelts = encoded_nelts ();
187 for (unsigned int i = 0; i < nelts; ++i)
188 if (!derived ()->equal_p ((*this)[i], other[i]))
189 return false;
191 return true;
194 /* Return the value of vector element I, which might or might not be
195 encoded explicitly. */
197 template<typename T, typename Derived>
199 vector_builder<T, Derived>::elt (unsigned int i) const
201 /* This only makes sense if the encoding has been fully populated. */
202 gcc_checking_assert (encoded_nelts () <= this->length ());
204 /* First handle elements that are already present in the underlying
205 vector, regardless of whether they're part of the encoding or not. */
206 if (i < this->length ())
207 return (*this)[i];
209 /* Identify the pattern that contains element I and work out the index of
210 the last encoded element for that pattern. */
211 unsigned int pattern = i % m_npatterns;
212 unsigned int count = i / m_npatterns;
213 unsigned int final_i = encoded_nelts () - m_npatterns + pattern;
214 T final = (*this)[final_i];
216 /* If there are no steps, the final encoded value is the right one. */
217 if (m_nelts_per_pattern <= 2)
218 return final;
220 /* Otherwise work out the value from the last two encoded elements. */
221 T prev = (*this)[final_i - m_npatterns];
222 return derived ()->apply_step (final, count - 2,
223 derived ()->step (prev, final));
226 /* Change the encoding to NPATTERNS patterns of NELTS_PER_PATTERN each,
227 but without changing the underlying vector. */
229 template<typename T, typename Derived>
230 void
231 vector_builder<T, Derived>::reshape (unsigned int npatterns,
232 unsigned int nelts_per_pattern)
234 unsigned int old_encoded_nelts = encoded_nelts ();
235 unsigned int new_encoded_nelts = npatterns * nelts_per_pattern;
236 gcc_checking_assert (new_encoded_nelts <= old_encoded_nelts);
237 unsigned int next = new_encoded_nelts - npatterns;
238 for (unsigned int i = new_encoded_nelts; i < old_encoded_nelts; ++i)
240 derived ()->note_representative (&(*this)[next], (*this)[i]);
241 next += 1;
242 if (next == new_encoded_nelts)
243 next -= npatterns;
245 m_npatterns = npatterns;
246 m_nelts_per_pattern = nelts_per_pattern;
249 /* Return true if elements [START, END) contain a repeating sequence of
250 STEP elements. */
252 template<typename T, typename Derived>
253 bool
254 vector_builder<T, Derived>::repeating_sequence_p (unsigned int start,
255 unsigned int end,
256 unsigned int step)
258 for (unsigned int i = start; i < end - step; ++i)
259 if (!derived ()->equal_p ((*this)[i], (*this)[i + step]))
260 return false;
261 return true;
264 /* Return true if elements [START, END) contain STEP interleaved linear
265 series. */
267 template<typename T, typename Derived>
268 bool
269 vector_builder<T, Derived>::stepped_sequence_p (unsigned int start,
270 unsigned int end,
271 unsigned int step)
273 if (!derived ()->allow_steps_p ())
274 return false;
276 for (unsigned int i = start + step * 2; i < end; ++i)
278 T elt1 = (*this)[i - step * 2];
279 T elt2 = (*this)[i - step];
280 T elt3 = (*this)[i];
282 if (!derived ()->integral_p (elt1)
283 || !derived ()->integral_p (elt2)
284 || !derived ()->integral_p (elt3))
285 return false;
287 if (maybe_ne (derived ()->step (elt1, elt2),
288 derived ()->step (elt2, elt3)))
289 return false;
291 if (!derived ()->can_elide_p (elt3))
292 return false;
294 return true;
297 /* Try to change the number of encoded patterns to NPATTERNS, returning
298 true on success. */
300 template<typename T, typename Derived>
301 bool
302 vector_builder<T, Derived>::try_npatterns (unsigned int npatterns)
304 if (m_nelts_per_pattern == 1)
306 /* See whether NPATTERNS is valid with the current 1-element-per-pattern
307 encoding. */
308 if (repeating_sequence_p (0, encoded_nelts (), npatterns))
310 reshape (npatterns, 1);
311 return true;
314 /* We can only increase the number of elements per pattern if all
315 elements are still encoded explicitly. */
316 if (!encoded_full_vector_p ())
317 return false;
320 if (m_nelts_per_pattern <= 2)
322 /* See whether NPATTERNS is valid with a 2-element-per-pattern
323 encoding. */
324 if (repeating_sequence_p (npatterns, encoded_nelts (), npatterns))
326 reshape (npatterns, 2);
327 return true;
330 /* We can only increase the number of elements per pattern if all
331 elements are still encoded explicitly. */
332 if (!encoded_full_vector_p ())
333 return false;
336 if (m_nelts_per_pattern <= 3)
338 /* See whether we have NPATTERNS interleaved linear series,
339 giving a 3-element-per-pattern encoding. */
340 if (stepped_sequence_p (npatterns, encoded_nelts (), npatterns))
342 reshape (npatterns, 3);
343 return true;
345 return false;
348 gcc_unreachable ();
351 /* Replace the current encoding with the canonical form. */
353 template<typename T, typename Derived>
354 void
355 vector_builder<T, Derived>::finalize ()
357 /* The encoding requires the same number of elements to come from each
358 pattern. */
359 gcc_assert (multiple_p (m_full_nelts, m_npatterns));
361 /* Allow the caller to build more elements than necessary. For example,
362 it's often convenient to build a stepped vector from the natural
363 encoding of three elements even if the vector itself only has two. */
364 unsigned HOST_WIDE_INT const_full_nelts;
365 if (m_full_nelts.is_constant (&const_full_nelts)
366 && const_full_nelts <= encoded_nelts ())
368 m_npatterns = const_full_nelts;
369 m_nelts_per_pattern = 1;
372 /* Try to whittle down the number of elements per pattern. That is:
374 1. If we have stepped patterns whose steps are all 0, reduce the
375 number of elements per pattern from 3 to 2.
377 2. If we have background fill values that are the same as the
378 foreground values, reduce the number of elements per pattern
379 from 2 to 1. */
380 while (m_nelts_per_pattern > 1
381 && repeating_sequence_p (encoded_nelts () - m_npatterns * 2,
382 encoded_nelts (), m_npatterns))
383 /* The last two sequences of M_NPATTERNS elements are equal,
384 so remove the last one. */
385 reshape (m_npatterns, m_nelts_per_pattern - 1);
387 if (pow2p_hwi (m_npatterns))
389 /* Try to halve the number of patterns while doing so gives a
390 valid pattern. This approach is linear in the number of
391 elements, whereas searcing from 1 up would be O(n*log(n)).
393 Each halving step tries to keep the number of elements per pattern
394 the same. If that isn't possible, and if all elements are still
395 explicitly encoded, the halving step can instead increase the number
396 of elements per pattern.
398 E.g. for:
400 { 0, 2, 3, 4, 5, 6, 7, 8 } npatterns == 8 full_nelts == 8
402 we first realize that the second half of the sequence is not
403 equal to the first, so we cannot maintain 1 element per pattern
404 for npatterns == 4. Instead we halve the number of patterns
405 and double the number of elements per pattern, treating this
406 as a "foreground" { 0, 2, 3, 4 } against a "background" of
407 { 5, 6, 7, 8 | 5, 6, 7, 8 ... }:
409 { 0, 2, 3, 4 | 5, 6, 7, 8 } npatterns == 4
411 Next we realize that this is *not* a foreround of { 0, 2 }
412 against a background of { 3, 4 | 3, 4 ... }, so the only
413 remaining option for reducing the number of patterns is
414 to use a foreground of { 0, 2 } against a stepped background
415 of { 1, 2 | 3, 4 | 5, 6 ... }. This is valid because we still
416 haven't elided any elements:
418 { 0, 2 | 3, 4 | 5, 6 } npatterns == 2
420 This in turn can be reduced to a foreground of { 0 } against a
421 stepped background of { 1 | 2 | 3 ... }:
423 { 0 | 2 | 3 } npatterns == 1
425 This last step would not have been possible for:
427 { 0, 0 | 3, 4 | 5, 6 } npatterns == 2. */
428 while ((m_npatterns & 1) == 0 && try_npatterns (m_npatterns / 2))
429 continue;
431 /* Builders of arbitrary fixed-length vectors can use:
433 new_vector (x, x, 1)
435 so that every element is specified explicitly. Handle cases
436 that are actually wrapping series, like { 0, 1, 2, 3, 0, 1, 2, 3 }
437 would be for 2-bit elements. We'll have treated them as
438 duplicates in the loop above. */
439 if (m_nelts_per_pattern == 1
440 && m_full_nelts.is_constant (&const_full_nelts)
441 && this->length () >= const_full_nelts
442 && (m_npatterns & 3) == 0
443 && stepped_sequence_p (m_npatterns / 4, const_full_nelts,
444 m_npatterns / 4))
446 reshape (m_npatterns / 4, 3);
447 while ((m_npatterns & 1) == 0 && try_npatterns (m_npatterns / 2))
448 continue;
451 else
452 /* For the non-power-of-2 case, do a simple search up from 1. */
453 for (unsigned int i = 1; i <= m_npatterns / 2; ++i)
454 if (m_npatterns % i == 0 && try_npatterns (i))
455 break;
458 #endif