summaryrefslogtreecommitdiffstats
path: root/src/compiler/nir/nir_opt_constant_folding.c
blob: 8f040619bfaeeb5cc99de326e9ba8c29152a214e (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
/*
 * 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_constant_expressions.h"
#include <math.h>

/*
 * Implements SSA-based constant folding.
 */

struct constant_fold_state {
   void *mem_ctx;
   nir_function_impl *impl;
   bool progress;
};

static bool
constant_fold_alu_instr(nir_alu_instr *instr, void *mem_ctx)
{
   nir_const_value src[4];

   if (!instr->dest.dest.is_ssa)
      return false;

   /* In the case that any outputs/inputs have unsized types, then we need to
    * guess the bit-size. In this case, the validator ensures that all
    * bit-sizes match so we can just take the bit-size from first
    * output/input with an unsized type. If all the outputs/inputs are sized
    * then we don't need to guess the bit-size at all because the code we
    * generate for constant opcodes in this case already knows the sizes of
    * the types involved and does not need the provided bit-size for anything
    * (although it still requires to receive a valid bit-size).
    */
   unsigned bit_size = 0;
   if (!nir_alu_type_get_type_size(nir_op_infos[instr->op].output_type))
      bit_size = instr->dest.dest.ssa.bit_size;

   for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
      if (!instr->src[i].src.is_ssa)
         return false;

      if (bit_size == 0 &&
          !nir_alu_type_get_type_size(nir_op_infos[instr->op].input_sizes[i])) {
         bit_size = instr->src[i].src.ssa->bit_size;
      }

      nir_instr *src_instr = instr->src[i].src.ssa->parent_instr;

      if (src_instr->type != nir_instr_type_load_const)
         return false;
      nir_load_const_instr* load_const = nir_instr_as_load_const(src_instr);

      for (unsigned j = 0; j < nir_ssa_alu_instr_src_components(instr, i);
           j++) {
         if (load_const->def.bit_size == 64)
            src[i].u64[j] = load_const->value.u64[instr->src[i].swizzle[j]];
         else
            src[i].u32[j] = load_const->value.u32[instr->src[i].swizzle[j]];
      }

      /* We shouldn't have any source modifiers in the optimization loop. */
      assert(!instr->src[i].abs && !instr->src[i].negate);
   }

   if (bit_size == 0)
      bit_size = 32;

   /* We shouldn't have any saturate modifiers in the optimization loop. */
   assert(!instr->dest.saturate);

   nir_const_value dest =
      nir_eval_const_opcode(instr->op, instr->dest.dest.ssa.num_components,
                            bit_size, src);

   nir_load_const_instr *new_instr =
      nir_load_const_instr_create(mem_ctx,
                                  instr->dest.dest.ssa.num_components,
                                  instr->dest.dest.ssa.bit_size);

   new_instr->value = dest;

   nir_instr_insert_before(&instr->instr, &new_instr->instr);

   nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa,
                            nir_src_for_ssa(&new_instr->def));

   nir_instr_remove(&instr->instr);
   ralloc_free(instr);

   return true;
}

static bool
constant_fold_deref(nir_instr *instr, nir_deref_var *deref)
{
   bool progress = false;

   for (nir_deref *tail = deref->deref.child; tail; tail = tail->child) {
      if (tail->deref_type != nir_deref_type_array)
         continue;

      nir_deref_array *arr = nir_deref_as_array(tail);

      if (arr->deref_array_type == nir_deref_array_type_indirect &&
          arr->indirect.is_ssa &&
          arr->indirect.ssa->parent_instr->type == nir_instr_type_load_const) {
         nir_load_const_instr *indirect =
            nir_instr_as_load_const(arr->indirect.ssa->parent_instr);

         arr->base_offset += indirect->value.u32[0];

         /* Clear out the source */
         nir_instr_rewrite_src(instr, &arr->indirect, nir_src_for_ssa(NULL));

         arr->deref_array_type = nir_deref_array_type_direct;

         progress = true;
      }
   }

   return progress;
}

static bool
constant_fold_intrinsic_instr(nir_intrinsic_instr *instr)
{
   bool progress = false;

   unsigned num_vars = nir_intrinsic_infos[instr->intrinsic].num_variables;
   for (unsigned i = 0; i < num_vars; i++) {
      progress |= constant_fold_deref(&instr->instr, instr->variables[i]);
   }

   return progress;
}

static bool
constant_fold_tex_instr(nir_tex_instr *instr)
{
   bool progress = false;

   if (instr->texture)
      progress |= constant_fold_deref(&instr->instr, instr->texture);

   if (instr->sampler)
      progress |= constant_fold_deref(&instr->instr, instr->sampler);

   return progress;
}

static bool
constant_fold_block(nir_block *block, void *mem_ctx)
{
   bool progress = false;

   nir_foreach_instr_safe(instr, block) {
      switch (instr->type) {
      case nir_instr_type_alu:
         progress |= constant_fold_alu_instr(nir_instr_as_alu(instr), mem_ctx);
         break;
      case nir_instr_type_intrinsic:
         progress |=
            constant_fold_intrinsic_instr(nir_instr_as_intrinsic(instr));
         break;
      case nir_instr_type_tex:
         progress |= constant_fold_tex_instr(nir_instr_as_tex(instr));
         break;
      default:
         /* Don't know how to constant fold */
         break;
      }
   }

   return progress;
}

static bool
nir_opt_constant_folding_impl(nir_function_impl *impl)
{
   void *mem_ctx = ralloc_parent(impl);
   bool progress = false;

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

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

   return progress;
}

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

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

   return progress;
}