1 /*
  2  * Copyright (c) 1997, 2020, Oracle and/or its affiliates. All rights reserved.
  3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  4  *
  5  * This code is free software; you can redistribute it and/or modify it
  6  * under the terms of the GNU General Public License version 2 only, as
  7  * published by the Free Software Foundation.
  8  *
  9  * This code is distributed in the hope that it will be useful, but WITHOUT
 10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 12  * version 2 for more details (a copy is included in the LICENSE file that
 13  * accompanied this code).
 14  *
 15  * You should have received a copy of the GNU General Public License version
 16  * 2 along with this work; if not, write to the Free Software Foundation,
 17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 18  *
 19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 20  * or visit www.oracle.com if you need additional information or have any
 21  * questions.
 22  *
 23  */
 24 
 25 #ifndef SHARE_OPTO_REGMASK_HPP
 26 #define SHARE_OPTO_REGMASK_HPP
 27 
 28 #include "code/vmreg.hpp"
 29 #include "opto/optoreg.hpp"
 30 #include "utilities/count_leading_zeros.hpp"
 31 #include "utilities/count_trailing_zeros.hpp"
 32 
 33 class LRG;
 34 
 35 //-------------Non-zero bit search methods used by RegMask---------------------
 36 // Find lowest 1, undefined if empty/0
 37 static int find_lowest_bit(uint32_t mask) {
 38   return count_trailing_zeros(mask);
 39 }
 40 // Find highest 1, undefined if empty/0
 41 static int find_highest_bit(uint32_t mask) {
 42   return count_leading_zeros(mask) ^ 31;
 43 }
 44 
 45 //------------------------------RegMask----------------------------------------
 46 // The ADL file describes how to print the machine-specific registers, as well
 47 // as any notion of register classes.  We provide a register mask, which is
 48 // just a collection of Register numbers.
 49 
 50 // The ADLC defines 2 macros, RM_SIZE and FORALL_BODY.
 51 // RM_SIZE is the size of a register mask in words.
 52 // FORALL_BODY replicates a BODY macro once per word in the register mask.
 53 // The usage is somewhat clumsy and limited to the regmask.[h,c]pp files.
 54 // However, it means the ADLC can redefine the unroll macro and all loops
 55 // over register masks will be unrolled by the correct amount.
 56 
 57 class RegMask {
 58   union {
 59     double _dummy_force_double_alignment[RM_SIZE>>1];
 60     // Array of Register Mask bits.  This array is large enough to cover
 61     // all the machine registers and all parameters that need to be passed
 62     // on the stack (stack registers) up to some interesting limit.  Methods
 63     // that need more parameters will NOT be compiled.  On Intel, the limit
 64     // is something like 90+ parameters.
 65     int _A[RM_SIZE];
 66   };
 67   // The low and high water marks represents the lowest and highest word
 68   // that might contain set register mask bits, respectively. We guarantee
 69   // that there are no bits in words outside this range, but any word at
 70   // and between the two marks can still be 0.
 71   int _lwm;
 72   int _hwm;
 73 
 74   enum {
 75     _WordBits    = BitsPerInt,
 76     _LogWordBits = LogBitsPerInt,
 77     _RM_SIZE     = RM_SIZE   // local constant, imported, then hidden by #undef
 78   };
 79 
 80  public:
 81   enum { CHUNK_SIZE = RM_SIZE*_WordBits };
 82 
 83   // SlotsPerLong is 2, since slots are 32 bits and longs are 64 bits.
 84   // Also, consider the maximum alignment size for a normally allocated
 85   // value.  Since we allocate register pairs but not register quads (at
 86   // present), this alignment is SlotsPerLong (== 2).  A normally
 87   // aligned allocated register is either a single register, or a pair
 88   // of adjacent registers, the lower-numbered being even.
 89   // See also is_aligned_Pairs() below, and the padding added before
 90   // Matcher::_new_SP to keep allocated pairs aligned properly.
 91   // If we ever go to quad-word allocations, SlotsPerQuad will become
 92   // the controlling alignment constraint.  Note that this alignment
 93   // requirement is internal to the allocator, and independent of any
 94   // particular platform.
 95   enum { SlotsPerLong = 2,
 96          SlotsPerVecA = 8,
 97          SlotsPerVecS = 1,
 98          SlotsPerVecD = 2,
 99          SlotsPerVecX = 4,
100          SlotsPerVecY = 8,
101          SlotsPerVecZ = 16,
102          };
103 
104   // A constructor only used by the ADLC output.  All mask fields are filled
105   // in directly.  Calls to this look something like RM(1,2,3,4);
106   RegMask(
107 #   define BODY(I) int a##I,
108     FORALL_BODY
109 #   undef BODY
110     int dummy = 0) {
111 #   define BODY(I) _A[I] = a##I;
112     FORALL_BODY
113 #   undef BODY
114     _lwm = 0;
115     _hwm = RM_SIZE - 1;
116     while (_hwm > 0 && _A[_hwm] == 0) _hwm--;
117     while ((_lwm < _hwm) && _A[_lwm] == 0) _lwm++;
118     assert(valid_watermarks(), "post-condition");
119   }
120 
121   // Handy copying constructor
122   RegMask(RegMask *rm) {
123     _hwm = rm->_hwm;
124     _lwm = rm->_lwm;
125     for (int i = 0; i < RM_SIZE; i++) {
126       _A[i] = rm->_A[i];
127     }
128     assert(valid_watermarks(), "post-condition");
129   }
130 
131   // Construct an empty mask
132   RegMask() {
133     Clear();
134   }
135 
136   // Construct a mask with a single bit
137   RegMask(OptoReg::Name reg) {
138     Clear();
139     Insert(reg);
140   }
141 
142   // Check for register being in mask
143   int Member(OptoReg::Name reg) const {
144     assert(reg < CHUNK_SIZE, "");
145     return _A[reg>>_LogWordBits] & (1<<(reg&(_WordBits-1)));
146   }
147 
148   // The last bit in the register mask indicates that the mask should repeat
149   // indefinitely with ONE bits.  Returns TRUE if mask is infinite or
150   // unbounded in size.  Returns FALSE if mask is finite size.
151   int is_AllStack() const { return _A[RM_SIZE-1] >> (_WordBits-1); }
152 
153   // Work around an -xO3 optimization problme in WS6U1. The old way:
154   //   void set_AllStack() { _A[RM_SIZE-1] |= (1<<(_WordBits-1)); }
155   // will cause _A[RM_SIZE-1] to be clobbered, not updated when set_AllStack()
156   // follows an Insert() loop, like the one found in init_spill_mask(). Using
157   // Insert() instead works because the index into _A in computed instead of
158   // constant.  See bug 4665841.
159   void set_AllStack() { Insert(OptoReg::Name(CHUNK_SIZE-1)); }
160 
161   // Test for being a not-empty mask.
162   int is_NotEmpty() const {
163     assert(valid_watermarks(), "sanity");
164     int tmp = 0;
165     for (int i = _lwm; i <= _hwm; i++) {
166       tmp |= _A[i];
167     }
168     return tmp;
169   }
170 
171   // Find lowest-numbered register from mask, or BAD if mask is empty.
172   OptoReg::Name find_first_elem() const {
173     assert(valid_watermarks(), "sanity");
174     for (int i = _lwm; i <= _hwm; i++) {
175       int bits = _A[i];
176       if (bits) {
177         return OptoReg::Name((i<<_LogWordBits) + find_lowest_bit(bits));
178       }
179     }
180     return OptoReg::Name(OptoReg::Bad);
181   }
182 
183   // Get highest-numbered register from mask, or BAD if mask is empty.
184   OptoReg::Name find_last_elem() const {
185     assert(valid_watermarks(), "sanity");
186     for (int i = _hwm; i >= _lwm; i--) {
187       int bits = _A[i];
188       if (bits) {
189         return OptoReg::Name((i<<_LogWordBits) + find_highest_bit(bits));
190       }
191     }
192     return OptoReg::Name(OptoReg::Bad);
193   }
194 
195   // Clear out partial bits; leave only aligned adjacent bit pairs.
196   void clear_to_pairs();
197 
198 #ifdef ASSERT
199   // Verify watermarks are sane, i.e., within bounds and that no
200   // register words below or above the watermarks have bits set.
201   bool valid_watermarks() const {
202     assert(_hwm >= 0 && _hwm < RM_SIZE, "_hwm out of range: %d", _hwm);
203     assert(_lwm >= 0 && _lwm < RM_SIZE, "_lwm out of range: %d", _lwm);
204     for (int i = 0; i < _lwm; i++) {
205       assert(_A[i] == 0, "_lwm too high: %d regs at: %d", _lwm, i);
206     }
207     for (int i = _hwm + 1; i < RM_SIZE; i++) {
208       assert(_A[i] == 0, "_hwm too low: %d regs at: %d", _hwm, i);
209     }
210     return true;
211   }
212 #endif // !ASSERT
213 
214   // Test that the mask contains only aligned adjacent bit pairs
215   bool is_aligned_pairs() const;
216 
217   // mask is a pair of misaligned registers
218   bool is_misaligned_pair() const;
219   // Test for single register
220   bool is_bound1() const;
221   // Test for a single adjacent pair
222   bool is_bound_pair() const;
223   // Test for a single adjacent set of ideal register's size.
224   bool is_bound(uint ireg) const;
225 
226   // Check that whether given reg number with size is valid
227   // for current regmask, where reg is the highest number.
228   bool is_valid_reg(OptoReg::Name reg, const int size) const;
229 
230   // Find the lowest-numbered register set in the mask.  Return the
231   // HIGHEST register number in the set, or BAD if no sets.
232   // Assert that the mask contains only bit sets.
233   OptoReg::Name find_first_set(LRG &lrg, const int size) const;
234 
235   // Clear out partial bits; leave only aligned adjacent bit sets of size.
236   void clear_to_sets(const int size);
237   // Smear out partial bits to aligned adjacent bit sets.
238   void smear_to_sets(const int size);
239   // Test that the mask contains only aligned adjacent bit sets
240   bool is_aligned_sets(const int size) const;
241 
242   // Test for a single adjacent set
243   int is_bound_set(const int size) const;
244 
245   static bool is_vector(uint ireg);
246   static int num_registers(uint ireg);
247   static int num_registers(uint ireg, LRG &lrg);
248 
249   // Fast overlap test.  Non-zero if any registers in common.
250   int overlap(const RegMask &rm) const {
251     assert(valid_watermarks() && rm.valid_watermarks(), "sanity");
252     int hwm = MIN2(_hwm, rm._hwm);
253     int lwm = MAX2(_lwm, rm._lwm);
254     int result = 0;
255     for (int i = lwm; i <= hwm; i++) {
256       result |= _A[i] & rm._A[i];
257     }
258     return result;
259   }
260 
261   // Special test for register pressure based splitting
262   // UP means register only, Register plus stack, or stack only is DOWN
263   bool is_UP() const;
264 
265   // Clear a register mask
266   void Clear() {
267     _lwm = RM_SIZE - 1;
268     _hwm = 0;
269     memset(_A, 0, sizeof(int)*RM_SIZE);
270     assert(valid_watermarks(), "sanity");
271   }
272 
273   // Fill a register mask with 1's
274   void Set_All() {
275     _lwm = 0;
276     _hwm = RM_SIZE - 1;
277     memset(_A, 0xFF, sizeof(int)*RM_SIZE);
278     assert(valid_watermarks(), "sanity");
279   }
280 
281   // Insert register into mask
282   void Insert(OptoReg::Name reg) {
283     assert(reg < CHUNK_SIZE, "sanity");
284     assert(valid_watermarks(), "pre-condition");
285     int index = reg>>_LogWordBits;
286     if (index > _hwm) _hwm = index;
287     if (index < _lwm) _lwm = index;
288     _A[index] |= (1<<(reg&(_WordBits-1)));
289     assert(valid_watermarks(), "post-condition");
290   }
291 
292   // Remove register from mask
293   void Remove(OptoReg::Name reg) {
294     assert(reg < CHUNK_SIZE, "");
295     _A[reg>>_LogWordBits] &= ~(1<<(reg&(_WordBits-1)));
296   }
297 
298   // OR 'rm' into 'this'
299   void OR(const RegMask &rm) {
300     assert(valid_watermarks() && rm.valid_watermarks(), "sanity");
301     // OR widens the live range
302     if (_lwm > rm._lwm) _lwm = rm._lwm;
303     if (_hwm < rm._hwm) _hwm = rm._hwm;
304     for (int i = _lwm; i <= _hwm; i++) {
305       _A[i] |= rm._A[i];
306     }
307     assert(valid_watermarks(), "sanity");
308   }
309 
310   // AND 'rm' into 'this'
311   void AND(const RegMask &rm) {
312     assert(valid_watermarks() && rm.valid_watermarks(), "sanity");
313     // Do not evaluate words outside the current watermark range, as they are
314     // already zero and an &= would not change that
315     for (int i = _lwm; i <= _hwm; i++) {
316       _A[i] &= rm._A[i];
317     }
318     // Narrow the watermarks if &rm spans a narrower range.
319     // Update after to ensure non-overlapping words are zeroed out.
320     if (_lwm < rm._lwm) _lwm = rm._lwm;
321     if (_hwm > rm._hwm) _hwm = rm._hwm;
322   }
323 
324   // Subtract 'rm' from 'this'
325   void SUBTRACT(const RegMask &rm) {
326     assert(valid_watermarks() && rm.valid_watermarks(), "sanity");
327     int hwm = MIN2(_hwm, rm._hwm);
328     int lwm = MAX2(_lwm, rm._lwm);
329     for (int i = lwm; i <= hwm; i++) {
330       _A[i] &= ~rm._A[i];
331     }
332   }
333 
334   // Compute size of register mask: number of bits
335   uint Size() const;
336 
337 #ifndef PRODUCT
338   void print() const { dump(); }
339   void dump(outputStream *st = tty) const; // Print a mask
340 #endif
341 
342   static const RegMask Empty;   // Common empty mask
343 
344   static bool can_represent(OptoReg::Name reg) {
345     // NOTE: -1 in computation reflects the usage of the last
346     //       bit of the regmask as an infinite stack flag and
347     //       -7 is to keep mask aligned for largest value (VecZ).
348     return (int)reg < (int)(CHUNK_SIZE-1);
349   }
350   static bool can_represent_arg(OptoReg::Name reg) {
351     // NOTE: -SlotsPerVecZ in computation reflects the need
352     //       to keep mask aligned for largest value (VecZ).
353     return (int)reg < (int)(CHUNK_SIZE-SlotsPerVecZ);
354   }
355 };
356 
357 // Do not use this constant directly in client code!
358 #undef RM_SIZE
359 
360 #endif // SHARE_OPTO_REGMASK_HPP