aboutsummaryrefslogtreecommitdiffstats
path: root/src/compiler/nir/nir_lower_vec_to_movs.c
blob: 8b24376b0a5befa4982c4b47187fa6f70113fe8f (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
/*
 * Copyright © 2014 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.
 *
 * Authors:
 *    Jason Ekstrand (jason@jlekstrand.net)
 *
 */

#include "nir.h"

/*
 * Implements a simple pass that lowers vecN instructions to a series of
 * moves with partial writes.
 */

static bool
src_matches_dest_reg(nir_dest *dest, nir_src *src)
{
   if (dest->is_ssa || src->is_ssa)
      return false;

   return (dest->reg.reg == src->reg.reg &&
           dest->reg.base_offset == src->reg.base_offset &&
           !dest->reg.indirect &&
           !src->reg.indirect);
}

/**
 * For a given starting writemask channel and corresponding source index in
 * the vec instruction, insert a MOV to the vec instruction's dest of all the
 * writemask channels that get read from the same src reg.
 *
 * Returns the writemask of our MOV, so the parent loop calling this knows
 * which ones have been processed.
 */
static unsigned
insert_mov(nir_alu_instr *vec, unsigned start_idx, nir_shader *shader)
{
   assert(start_idx < nir_op_infos[vec->op].num_inputs);

   nir_alu_instr *mov = nir_alu_instr_create(shader, nir_op_imov);
   nir_alu_src_copy(&mov->src[0], &vec->src[start_idx], mov);
   nir_alu_dest_copy(&mov->dest, &vec->dest, mov);

   mov->dest.write_mask = (1u << start_idx);
   mov->src[0].swizzle[start_idx] = vec->src[start_idx].swizzle[0];
   mov->src[0].negate = vec->src[start_idx].negate;
   mov->src[0].abs = vec->src[start_idx].abs;

   for (unsigned i = start_idx + 1; i < 4; i++) {
      if (!(vec->dest.write_mask & (1 << i)))
         continue;

      if (nir_srcs_equal(vec->src[i].src, vec->src[start_idx].src) &&
          vec->src[i].negate == vec->src[start_idx].negate &&
          vec->src[i].abs == vec->src[start_idx].abs) {
         mov->dest.write_mask |= (1 << i);
         mov->src[0].swizzle[i] = vec->src[i].swizzle[0];
      }
   }

   unsigned channels_handled = mov->dest.write_mask;

   /* In some situations (if the vecN is involved in a phi-web), we can end
    * up with a mov from a register to itself.  Some of those channels may end
    * up doing nothing and there's no reason to have them as part of the mov.
    */
   if (src_matches_dest_reg(&mov->dest.dest, &mov->src[0].src) &&
       !mov->src[0].abs && !mov->src[0].negate) {
      for (unsigned i = 0; i < 4; i++) {
         if (mov->src[0].swizzle[i] == i) {
            mov->dest.write_mask &= ~(1 << i);
         }
      }
   }

   /* Only emit the instruction if it actually does something */
   if (mov->dest.write_mask) {
      nir_instr_insert_before(&vec->instr, &mov->instr);
   } else {
      ralloc_free(mov);
   }

   return channels_handled;
}

static bool
has_replicated_dest(nir_alu_instr *alu)
{
   return alu->op == nir_op_fdot_replicated2 ||
          alu->op == nir_op_fdot_replicated3 ||
          alu->op == nir_op_fdot_replicated4 ||
          alu->op == nir_op_fdph_replicated;
}

/* Attempts to coalesce the "move" from the given source of the vec to the
 * destination of the instruction generating the value. If, for whatever
 * reason, we cannot coalesce the mmove, it does nothing and returns 0.  We
 * can then call insert_mov as normal.
 */
static unsigned
try_coalesce(nir_alu_instr *vec, unsigned start_idx)
{
   assert(start_idx < nir_op_infos[vec->op].num_inputs);

   /* We will only even try if the source is SSA */
   if (!vec->src[start_idx].src.is_ssa)
      return 0;

   assert(vec->src[start_idx].src.ssa);

   /* If we are going to do a reswizzle, then the vecN operation must be the
    * only use of the source value.  We also can't have any source modifiers.
    */
   nir_foreach_use(src, vec->src[start_idx].src.ssa) {
      if (src->parent_instr != &vec->instr)
         return 0;

      nir_alu_src *alu_src = exec_node_data(nir_alu_src, src, src);
      if (alu_src->abs || alu_src->negate)
         return 0;
   }

   if (!list_empty(&vec->src[start_idx].src.ssa->if_uses))
      return 0;

   if (vec->src[start_idx].src.ssa->parent_instr->type != nir_instr_type_alu)
      return 0;

   nir_alu_instr *src_alu =
      nir_instr_as_alu(vec->src[start_idx].src.ssa->parent_instr);

   if (has_replicated_dest(src_alu)) {
      /* The fdot instruction is special: It replicates its result to all
       * components.  This means that we can always rewrite its destination
       * and we don't need to swizzle anything.
       */
   } else {
      /* We only care about being able to re-swizzle the instruction if it is
       * something that we can reswizzle.  It must be per-component.  The one
       * exception to this is the fdotN instructions which implicitly splat
       * their result out to all channels.
       */
      if (nir_op_infos[src_alu->op].output_size != 0)
         return 0;

      /* If we are going to reswizzle the instruction, we can't have any
       * non-per-component sources either.
       */
      for (unsigned j = 0; j < nir_op_infos[src_alu->op].num_inputs; j++)
         if (nir_op_infos[src_alu->op].input_sizes[j] != 0)
            return 0;
   }

   /* Stash off all of the ALU instruction's swizzles. */
   uint8_t swizzles[4][4];
   for (unsigned j = 0; j < nir_op_infos[src_alu->op].num_inputs; j++)
      for (unsigned i = 0; i < 4; i++)
         swizzles[j][i] = src_alu->src[j].swizzle[i];

   unsigned write_mask = 0;
   for (unsigned i = start_idx; i < 4; i++) {
      if (!(vec->dest.write_mask & (1 << i)))
         continue;

      if (!vec->src[i].src.is_ssa ||
          vec->src[i].src.ssa != &src_alu->dest.dest.ssa)
         continue;

      /* At this point, the give vec source matchese up with the ALU
       * instruction so we can re-swizzle that component to match.
       */
      write_mask |= 1 << i;
      if (has_replicated_dest(src_alu)) {
         /* Since the destination is a single replicated value, we don't need
          * to do any reswizzling
          */
      } else {
         for (unsigned j = 0; j < nir_op_infos[src_alu->op].num_inputs; j++)
            src_alu->src[j].swizzle[i] = swizzles[j][vec->src[i].swizzle[0]];
      }

      /* Clear the no longer needed vec source */
      nir_instr_rewrite_src(&vec->instr, &vec->src[i].src, NIR_SRC_INIT);
   }

   nir_instr_rewrite_dest(&src_alu->instr, &src_alu->dest.dest, vec->dest.dest);
   src_alu->dest.write_mask = write_mask;

   return write_mask;
}

static bool
lower_vec_to_movs_block(nir_block *block, nir_function_impl *impl)
{
   bool progress = false;
   nir_shader *shader = impl->function->shader;

   nir_foreach_instr_safe(instr, block) {
      if (instr->type != nir_instr_type_alu)
         continue;

      nir_alu_instr *vec = nir_instr_as_alu(instr);

      switch (vec->op) {
      case nir_op_vec2:
      case nir_op_vec3:
      case nir_op_vec4:
         break;
      default:
         continue; /* The loop */
      }

      bool vec_had_ssa_dest = vec->dest.dest.is_ssa;
      if (vec->dest.dest.is_ssa) {
         /* Since we insert multiple MOVs, we have a register destination. */
         nir_register *reg = nir_local_reg_create(impl);
         reg->num_components = vec->dest.dest.ssa.num_components;
         reg->bit_size = vec->dest.dest.ssa.bit_size;

         nir_ssa_def_rewrite_uses(&vec->dest.dest.ssa, nir_src_for_reg(reg));

         nir_instr_rewrite_dest(&vec->instr, &vec->dest.dest,
                                nir_dest_for_reg(reg));
      }

      unsigned finished_write_mask = 0;

      /* First, emit a MOV for all the src channels that are in the
       * destination reg, in case other values we're populating in the dest
       * might overwrite them.
       */
      for (unsigned i = 0; i < 4; i++) {
         if (!(vec->dest.write_mask & (1 << i)))
            continue;

         if (src_matches_dest_reg(&vec->dest.dest, &vec->src[i].src)) {
            finished_write_mask |= insert_mov(vec, i, shader);
            break;
         }
      }

      /* Now, emit MOVs for all the other src channels. */
      for (unsigned i = 0; i < 4; i++) {
         if (!(vec->dest.write_mask & (1 << i)))
            continue;

         /* Coalescing moves the register writes from the vec up to the ALU
          * instruction in the source.  We can only do this if the original
          * vecN had an SSA destination.
          */
         if (vec_had_ssa_dest && !(finished_write_mask & (1 << i)))
            finished_write_mask |= try_coalesce(vec, i);

         if (!(finished_write_mask & (1 << i)))
            finished_write_mask |= insert_mov(vec, i, shader);
      }

      nir_instr_remove(&vec->instr);
      ralloc_free(vec);
      progress = true;
   }

   return progress;
}

static bool
nir_lower_vec_to_movs_impl(nir_function_impl *impl)
{
   bool progress = false;

   nir_foreach_block(block, impl) {
      progress |= lower_vec_to_movs_block(block, impl);
   }

   if (progress) {
      nir_metadata_preserve(impl, nir_metadata_block_index |
                                  nir_metadata_dominance);
   }

   return progress;
}

bool
nir_lower_vec_to_movs(nir_shader *shader)
{
   bool progress = false;

   nir_foreach_function(function, shader) {
      if (function->impl)
         progress = nir_lower_vec_to_movs_impl(function->impl) || progress;
   }

   return progress;
}