summaryrefslogtreecommitdiff
path: root/src/intel/compiler/brw_ir_fs.h
blob: 95b069a2e024b07253f67301874e67574baf8bbc (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
/* -*- c++ -*- */
/*
 * Copyright © 2010-2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#ifndef BRW_IR_FS_H
#define BRW_IR_FS_H

#include "brw_shader.h"

class fs_inst;

class fs_reg : public backend_reg {
public:
   DECLARE_RALLOC_CXX_OPERATORS(fs_reg)

   void init();

   fs_reg();
   fs_reg(struct ::brw_reg reg);
   fs_reg(enum brw_reg_file file, int nr);
   fs_reg(enum brw_reg_file file, int nr, enum brw_reg_type type);

   bool equals(const fs_reg &r) const;
   bool negative_equals(const fs_reg &r) const;
   bool is_contiguous() const;

   /**
    * Return the size in bytes of a single logical component of the
    * register assuming the given execution width.
    */
   unsigned component_size(unsigned width) const;

   /** Register region horizontal stride */
   uint8_t stride;
};

static inline fs_reg
negate(fs_reg reg)
{
   assert(reg.file != IMM);
   reg.negate = !reg.negate;
   return reg;
}

static inline fs_reg
retype(fs_reg reg, enum brw_reg_type type)
{
   reg.type = type;
   return reg;
}

static inline fs_reg
byte_offset(fs_reg reg, unsigned delta)
{
   switch (reg.file) {
   case BAD_FILE:
      break;
   case VGRF:
   case ATTR:
   case UNIFORM:
      reg.offset += delta;
      break;
   case MRF: {
      const unsigned suboffset = reg.offset + delta;
      reg.nr += suboffset / REG_SIZE;
      reg.offset = suboffset % REG_SIZE;
      break;
   }
   case ARF:
   case FIXED_GRF: {
      const unsigned suboffset = reg.subnr + delta;
      reg.nr += suboffset / REG_SIZE;
      reg.subnr = suboffset % REG_SIZE;
      break;
   }
   case IMM:
   default:
      assert(delta == 0);
   }
   return reg;
}

static inline fs_reg
horiz_offset(const fs_reg &reg, unsigned delta)
{
   switch (reg.file) {
   case BAD_FILE:
   case UNIFORM:
   case IMM:
      /* These only have a single component that is implicitly splatted.  A
       * horizontal offset should be a harmless no-op.
       * XXX - Handle vector immediates correctly.
       */
      return reg;
   case VGRF:
   case MRF:
   case ATTR:
      return byte_offset(reg, delta * reg.stride * type_sz(reg.type));
   case ARF:
   case FIXED_GRF:
      if (reg.is_null()) {
         return reg;
      } else {
         const unsigned stride = reg.hstride ? 1 << (reg.hstride - 1) : 0;
         return byte_offset(reg, delta * stride * type_sz(reg.type));
      }
   }
   unreachable("Invalid register file");
}

static inline fs_reg
offset(fs_reg reg, unsigned width, unsigned delta)
{
   switch (reg.file) {
   case BAD_FILE:
      break;
   case ARF:
   case FIXED_GRF:
   case MRF:
   case VGRF:
   case ATTR:
   case UNIFORM:
      return byte_offset(reg, delta * reg.component_size(width));
   case IMM:
      assert(delta == 0);
   }
   return reg;
}

/**
 * Get the scalar channel of \p reg given by \p idx and replicate it to all
 * channels of the result.
 */
static inline fs_reg
component(fs_reg reg, unsigned idx)
{
   reg = horiz_offset(reg, idx);
   reg.stride = 0;
   return reg;
}

/**
 * Return an integer identifying the discrete address space a register is
 * contained in.  A register is by definition fully contained in the single
 * reg_space it belongs to, so two registers with different reg_space ids are
 * guaranteed not to overlap.  Most register files are a single reg_space of
 * its own, only the VGRF file is composed of multiple discrete address
 * spaces, one for each VGRF allocation.
 */
static inline uint32_t
reg_space(const fs_reg &r)
{
   return r.file << 16 | (r.file == VGRF ? r.nr : 0);
}

/**
 * Return the base offset in bytes of a register relative to the start of its
 * reg_space().
 */
static inline unsigned
reg_offset(const fs_reg &r)
{
   return (r.file == VGRF || r.file == IMM ? 0 : r.nr) *
          (r.file == UNIFORM ? 4 : REG_SIZE) + r.offset +
          (r.file == ARF || r.file == FIXED_GRF ? r.subnr : 0);
}

/**
 * Return the amount of padding in bytes left unused between individual
 * components of register \p r due to a (horizontal) stride value greater than
 * one, or zero if components are tightly packed in the register file.
 */
static inline unsigned
reg_padding(const fs_reg &r)
{
   const unsigned stride = ((r.file != ARF && r.file != FIXED_GRF) ? r.stride :
                            r.hstride == 0 ? 0 :
                            1 << (r.hstride - 1));
   return (MAX2(1, stride) - 1) * type_sz(r.type);
}

/**
 * Return whether the register region starting at \p r and spanning \p dr
 * bytes could potentially overlap the register region starting at \p s and
 * spanning \p ds bytes.
 */
static inline bool
regions_overlap(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds)
{
   if (r.file == MRF && (r.nr & BRW_MRF_COMPR4)) {
      fs_reg t = r;
      t.nr &= ~BRW_MRF_COMPR4;
      /* COMPR4 regions are translated by the hardware during decompression
       * into two separate half-regions 4 MRFs apart from each other.
       */
      return regions_overlap(t, dr / 2, s, ds) ||
             regions_overlap(byte_offset(t, 4 * REG_SIZE), dr / 2, s, ds);

   } else if (s.file == MRF && (s.nr & BRW_MRF_COMPR4)) {
      return regions_overlap(s, ds, r, dr);

   } else {
      return reg_space(r) == reg_space(s) &&
             !(reg_offset(r) + dr <= reg_offset(s) ||
               reg_offset(s) + ds <= reg_offset(r));
   }
}

/**
 * Check that the register region given by r [r.offset, r.offset + dr[
 * is fully contained inside the register region given by s
 * [s.offset, s.offset + ds[.
 */
static inline bool
region_contained_in(const fs_reg &r, unsigned dr, const fs_reg &s, unsigned ds)
{
   return reg_space(r) == reg_space(s) &&
          reg_offset(r) >= reg_offset(s) &&
          reg_offset(r) + dr <= reg_offset(s) + ds;
}

/**
 * Return whether the given register region is n-periodic, i.e. whether the
 * original region remains invariant after shifting it by \p n scalar
 * channels.
 */
static inline bool
is_periodic(const fs_reg &reg, unsigned n)
{
   if (reg.file == BAD_FILE || reg.is_null()) {
      return true;

   } else if (reg.file == IMM) {
      const unsigned period = (reg.type == BRW_REGISTER_TYPE_UV ||
                               reg.type == BRW_REGISTER_TYPE_V ? 8 :
                               reg.type == BRW_REGISTER_TYPE_VF ? 4 :
                               1);
      return n % period == 0;

   } else if (reg.file == ARF || reg.file == FIXED_GRF) {
      const unsigned period = (reg.hstride == 0 && reg.vstride == 0 ? 1 :
                               reg.vstride == 0 ? 1 << reg.width :
                               ~0);
      return n % period == 0;

   } else {
      return reg.stride == 0;
   }
}

static inline bool
is_uniform(const fs_reg &reg)
{
   return is_periodic(reg, 1);
}

/**
 * Get the specified 8-component quarter of a register.
 * XXX - Maybe come up with a less misleading name for this (e.g. quarter())?
 */
static inline fs_reg
half(const fs_reg &reg, unsigned idx)
{
   assert(idx < 2);
   return horiz_offset(reg, 8 * idx);
}

/**
 * Reinterpret each channel of register \p reg as a vector of values of the
 * given smaller type and take the i-th subcomponent from each.
 */
static inline fs_reg
subscript(fs_reg reg, brw_reg_type type, unsigned i)
{
   assert((i + 1) * type_sz(type) <= type_sz(reg.type));

   if (reg.file == ARF || reg.file == FIXED_GRF) {
      /* The stride is encoded inconsistently for fixed GRF and ARF registers
       * as the log2 of the actual vertical and horizontal strides.
       */
      const int delta = _mesa_logbase2(type_sz(reg.type)) -
                        _mesa_logbase2(type_sz(type));
      reg.hstride += (reg.hstride ? delta : 0);
      reg.vstride += (reg.vstride ? delta : 0);

   } else if (reg.file == IMM) {
      assert(reg.type == type);

   } else {
      reg.stride *= type_sz(reg.type) / type_sz(type);
   }

   return byte_offset(retype(reg, type), i * type_sz(type));
}

static inline fs_reg
horiz_stride(fs_reg reg, unsigned s)
{
   reg.stride *= s;
   return reg;
}

static const fs_reg reg_undef;

class fs_inst : public backend_instruction {
   fs_inst &operator=(const fs_inst &);

   void init(enum opcode opcode, uint8_t exec_width, const fs_reg &dst,
             const fs_reg *src, unsigned sources);

public:
   DECLARE_RALLOC_CXX_OPERATORS(fs_inst)

   fs_inst();
   fs_inst(enum opcode opcode, uint8_t exec_size);
   fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst);
   fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
           const fs_reg &src0);
   fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
           const fs_reg &src0, const fs_reg &src1);
   fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
           const fs_reg &src0, const fs_reg &src1, const fs_reg &src2);
   fs_inst(enum opcode opcode, uint8_t exec_size, const fs_reg &dst,
           const fs_reg src[], unsigned sources);
   fs_inst(const fs_inst &that);
   ~fs_inst();

   void resize_sources(uint8_t num_sources);

   bool equals(fs_inst *inst) const;
   bool is_send_from_grf() const;
   bool is_partial_write() const;
   bool is_copy_payload(const brw::simple_allocator &grf_alloc) const;
   unsigned components_read(unsigned i) const;
   unsigned size_read(int arg) const;
   bool can_do_source_mods(const struct gen_device_info *devinfo);
   bool can_do_cmod();
   bool can_change_types() const;
   bool has_source_and_destination_hazard() const;

   /**
    * Return the subset of flag registers read by the instruction as a bitset
    * with byte granularity.
    */
   unsigned flags_read(const gen_device_info *devinfo) const;

   /**
    * Return the subset of flag registers updated by the instruction (either
    * partially or fully) as a bitset with byte granularity.
    */
   unsigned flags_written() const;

   fs_reg dst;
   fs_reg *src;

   uint8_t sources; /**< Number of fs_reg sources. */

   bool last_rt:1;
   bool pi_noperspective:1;   /**< Pixel interpolator noperspective flag */
};

/**
 * Make the execution of \p inst dependent on the evaluation of a possibly
 * inverted predicate.
 */
static inline fs_inst *
set_predicate_inv(enum brw_predicate pred, bool inverse,
                  fs_inst *inst)
{
   inst->predicate = pred;
   inst->predicate_inverse = inverse;
   return inst;
}

/**
 * Make the execution of \p inst dependent on the evaluation of a predicate.
 */
static inline fs_inst *
set_predicate(enum brw_predicate pred, fs_inst *inst)
{
   return set_predicate_inv(pred, false, inst);
}

/**
 * Write the result of evaluating the condition given by \p mod to a flag
 * register.
 */
static inline fs_inst *
set_condmod(enum brw_conditional_mod mod, fs_inst *inst)
{
   inst->conditional_mod = mod;
   return inst;
}

/**
 * Clamp the result of \p inst to the saturation range of its destination
 * datatype.
 */
static inline fs_inst *
set_saturate(bool saturate, fs_inst *inst)
{
   inst->saturate = saturate;
   return inst;
}

/**
 * Return the number of dataflow registers written by the instruction (either
 * fully or partially) counted from 'floor(reg_offset(inst->dst) /
 * register_size)'.  The somewhat arbitrary register size unit is 4B for the
 * UNIFORM and IMM files and 32B for all other files.
 */
inline unsigned
regs_written(const fs_inst *inst)
{
   assert(inst->dst.file != UNIFORM && inst->dst.file != IMM);
   return DIV_ROUND_UP(reg_offset(inst->dst) % REG_SIZE +
                       inst->size_written -
                       MIN2(inst->size_written, reg_padding(inst->dst)),
                       REG_SIZE);
}

/**
 * Return the number of dataflow registers read by the instruction (either
 * fully or partially) counted from 'floor(reg_offset(inst->src[i]) /
 * register_size)'.  The somewhat arbitrary register size unit is 4B for the
 * UNIFORM and IMM files and 32B for all other files.
 */
inline unsigned
regs_read(const fs_inst *inst, unsigned i)
{
   const unsigned reg_size =
      inst->src[i].file == UNIFORM || inst->src[i].file == IMM ? 4 : REG_SIZE;
   return DIV_ROUND_UP(reg_offset(inst->src[i]) % reg_size +
                       inst->size_read(i) -
                       MIN2(inst->size_read(i), reg_padding(inst->src[i])),
                       reg_size);
}

static inline enum brw_reg_type
get_exec_type(const fs_inst *inst)
{
   brw_reg_type exec_type = BRW_REGISTER_TYPE_B;

   for (int i = 0; i < inst->sources; i++) {
      if (inst->src[i].file != BAD_FILE) {
         const brw_reg_type t = get_exec_type(inst->src[i].type);
         if (type_sz(t) > type_sz(exec_type))
            exec_type = t;
         else if (type_sz(t) == type_sz(exec_type) &&
                  brw_reg_type_is_floating_point(t))
            exec_type = t;
      }
   }

   if (exec_type == BRW_REGISTER_TYPE_B)
      exec_type = inst->dst.type;

   assert(exec_type != BRW_REGISTER_TYPE_B);

   return exec_type;
}

static inline unsigned
get_exec_type_size(const fs_inst *inst)
{
   return type_sz(get_exec_type(inst));
}

/**
 * Return whether the following regioning restriction applies to the specified
 * instruction.  From the Cherryview PRM Vol 7. "Register Region
 * Restrictions":
 *
 * "When source or destination datatype is 64b or operation is integer DWord
 *  multiply, regioning in Align1 must follow these rules:
 *
 *  1. Source and Destination horizontal stride must be aligned to the same qword.
 *  2. Regioning must ensure Src.Vstride = Src.Width * Src.Hstride.
 *  3. Source and Destination offset must be the same, except the case of
 *     scalar source."
 */
static inline bool
has_dst_aligned_region_restriction(const gen_device_info *devinfo,
                                   const fs_inst *inst)
{
   const brw_reg_type exec_type = get_exec_type(inst);
   const bool is_int_multiply = !brw_reg_type_is_floating_point(exec_type) &&
         (inst->opcode == BRW_OPCODE_MUL || inst->opcode == BRW_OPCODE_MAD);

   if (type_sz(inst->dst.type) > 4 || type_sz(exec_type) > 4 ||
       (type_sz(exec_type) == 4 && is_int_multiply))
      return devinfo->is_cherryview || gen_device_info_is_9lp(devinfo);
   else
      return false;
}

#endif