1/*
2 * Copyright © 2018 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24#include "nir.h"
25#include "nir_builder.h"
26
27static bool
28assert_ssa_def_is_not_1bit(nir_ssa_def *def, UNUSED void *unused)
29{
30   assert(def->bit_size > 1);
31   return true;
32}
33
34static bool
35rewrite_1bit_ssa_def_to_32bit(nir_ssa_def *def, void *_progress)
36{
37   bool *progress = _progress;
38   if (def->bit_size == 1) {
39      def->bit_size = 32;
40      *progress = true;
41   }
42   return true;
43}
44
45static bool
46lower_alu_instr(nir_builder *b, nir_alu_instr *alu)
47{
48   const nir_op_info *op_info = &nir_op_infos[alu->op];
49
50   b->cursor = nir_before_instr(&alu->instr);
51
52   /* Replacement SSA value */
53   nir_ssa_def *rep = NULL;
54   switch (alu->op) {
55   case nir_op_vec2:
56   case nir_op_vec3:
57   case nir_op_vec4:
58      /* These we expect to have booleans but the opcode doesn't change */
59      break;
60
61   case nir_op_b2f32: alu->op = nir_op_fmov; break;
62   case nir_op_b2i32: alu->op = nir_op_fmov; break;
63   case nir_op_f2b1:
64   case nir_op_i2b1:
65      rep = nir_sne(b, nir_ssa_for_alu_src(b, alu, 0),
66                       nir_imm_float(b, 0));
67      break;
68
69   case nir_op_flt: alu->op = nir_op_slt; break;
70   case nir_op_fge: alu->op = nir_op_sge; break;
71   case nir_op_feq: alu->op = nir_op_seq; break;
72   case nir_op_fne: alu->op = nir_op_sne; break;
73   case nir_op_ilt: alu->op = nir_op_slt; break;
74   case nir_op_ige: alu->op = nir_op_sge; break;
75   case nir_op_ieq: alu->op = nir_op_seq; break;
76   case nir_op_ine: alu->op = nir_op_sne; break;
77   case nir_op_ult: alu->op = nir_op_slt; break;
78   case nir_op_uge: alu->op = nir_op_sge; break;
79
80   case nir_op_ball_fequal2:  alu->op = nir_op_fall_equal2; break;
81   case nir_op_ball_fequal3:  alu->op = nir_op_fall_equal3; break;
82   case nir_op_ball_fequal4:  alu->op = nir_op_fall_equal4; break;
83   case nir_op_bany_fnequal2: alu->op = nir_op_fany_nequal2; break;
84   case nir_op_bany_fnequal3: alu->op = nir_op_fany_nequal3; break;
85   case nir_op_bany_fnequal4: alu->op = nir_op_fany_nequal4; break;
86   case nir_op_ball_iequal2:  alu->op = nir_op_fall_equal2; break;
87   case nir_op_ball_iequal3:  alu->op = nir_op_fall_equal3; break;
88   case nir_op_ball_iequal4:  alu->op = nir_op_fall_equal4; break;
89   case nir_op_bany_inequal2: alu->op = nir_op_fany_nequal2; break;
90   case nir_op_bany_inequal3: alu->op = nir_op_fany_nequal3; break;
91   case nir_op_bany_inequal4: alu->op = nir_op_fany_nequal4; break;
92
93   case nir_op_bcsel: alu->op = nir_op_fcsel; break;
94
95   case nir_op_imov: alu->op = nir_op_fmov; break;
96   case nir_op_iand: alu->op = nir_op_fmul; break;
97   case nir_op_ixor: alu->op = nir_op_sne; break;
98   case nir_op_ior: alu->op = nir_op_fmax; break;
99
100   case nir_op_inot:
101      rep = nir_seq(b, nir_ssa_for_alu_src(b, alu, 0),
102                       nir_imm_float(b, 0));
103      break;
104
105   default:
106      assert(alu->dest.dest.ssa.bit_size > 1);
107      for (unsigned i = 0; i < op_info->num_inputs; i++)
108         assert(alu->src[i].src.ssa->bit_size > 1);
109      return false;
110   }
111
112   if (rep) {
113      /* We've emitted a replacement instruction */
114      nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa, nir_src_for_ssa(rep));
115      nir_instr_remove(&alu->instr);
116   } else {
117      if (alu->dest.dest.ssa.bit_size == 1)
118         alu->dest.dest.ssa.bit_size = 32;
119   }
120
121   return true;
122}
123
124static bool
125nir_lower_bool_to_float_impl(nir_function_impl *impl)
126{
127   bool progress = false;
128
129   nir_builder b;
130   nir_builder_init(&b, impl);
131
132   nir_foreach_block(block, impl) {
133      nir_foreach_instr_safe(instr, block) {
134         switch (instr->type) {
135         case nir_instr_type_alu:
136            progress |= lower_alu_instr(&b, nir_instr_as_alu(instr));
137            break;
138
139         case nir_instr_type_load_const: {
140            nir_load_const_instr *load = nir_instr_as_load_const(instr);
141            if (load->def.bit_size == 1) {
142               nir_const_value *value = load->value;
143               for (unsigned i = 0; i < load->def.num_components; i++)
144                  load->value[i].f32 = value[i].b ? 1.0 : 0.0;
145               load->def.bit_size = 32;
146               progress = true;
147            }
148            break;
149         }
150
151         case nir_instr_type_intrinsic:
152         case nir_instr_type_ssa_undef:
153         case nir_instr_type_phi:
154         case nir_instr_type_tex:
155            nir_foreach_ssa_def(instr, rewrite_1bit_ssa_def_to_32bit,
156                                &progress);
157            break;
158
159         default:
160            nir_foreach_ssa_def(instr, assert_ssa_def_is_not_1bit, NULL);
161         }
162      }
163   }
164
165   if (progress) {
166      nir_metadata_preserve(impl, nir_metadata_block_index |
167                                  nir_metadata_dominance);
168   }
169
170   return progress;
171}
172
173bool
174nir_lower_bool_to_float(nir_shader *shader)
175{
176   bool progress = false;
177
178   nir_foreach_function(function, shader) {
179      if (function->impl && nir_lower_bool_to_float_impl(function->impl))
180         progress = true;
181   }
182
183   return progress;
184}
185