1/*
2 * Copyright © 2018 Intel Corporation
3 * Copyright © 2019 Vasily Khoruzhick <anarsoul@gmail.com>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25#include "nir.h"
26#include "nir_builder.h"
27
28static bool
29assert_ssa_def_is_not_int(nir_ssa_def *def, void *arg)
30{
31   ASSERTED BITSET_WORD *int_types = arg;
32   assert(!BITSET_TEST(int_types, def->index));
33   return true;
34}
35
36static bool
37lower_alu_instr(nir_builder *b, nir_alu_instr *alu)
38{
39   const nir_op_info *info = &nir_op_infos[alu->op];
40
41   bool is_bool_only = alu->dest.dest.ssa.bit_size == 1;
42   for (unsigned i = 0; i < info->num_inputs; i++) {
43      if (alu->src[i].src.ssa->bit_size != 1)
44         is_bool_only = false;
45   }
46
47   if (is_bool_only) {
48      /* avoid lowering integers ops are used for booleans (ieq,ine,etc) */
49      return false;
50   }
51
52   b->cursor = nir_before_instr(&alu->instr);
53
54   /* Replacement SSA value */
55   nir_ssa_def *rep = NULL;
56   switch (alu->op) {
57   case nir_op_mov:
58   case nir_op_vec2:
59   case nir_op_vec3:
60   case nir_op_vec4:
61   case nir_op_bcsel:
62      /* These we expect to have integers but the opcode doesn't change */
63      break;
64
65   case nir_op_b2i32: alu->op = nir_op_b2f32; break;
66   case nir_op_i2f32: alu->op = nir_op_mov; break;
67   case nir_op_u2f32: alu->op = nir_op_mov; break;
68   case nir_op_f2i32: alu->op = nir_op_ftrunc; break;
69   case nir_op_f2u32: alu->op = nir_op_ffloor; break;
70   case nir_op_i2b1: alu->op = nir_op_f2b1; break;
71
72   case nir_op_ilt: alu->op = nir_op_flt; break;
73   case nir_op_ige: alu->op = nir_op_fge; break;
74   case nir_op_ieq: alu->op = nir_op_feq; break;
75   case nir_op_ine: alu->op = nir_op_fneu; break;
76   case nir_op_ult: alu->op = nir_op_flt; break;
77   case nir_op_uge: alu->op = nir_op_fge; break;
78
79   case nir_op_iadd: alu->op = nir_op_fadd; break;
80   case nir_op_isub: alu->op = nir_op_fsub; break;
81   case nir_op_imul: alu->op = nir_op_fmul; break;
82   case nir_op_idiv:
83      rep = nir_ftrunc(b, nir_fdiv(b,
84                                   nir_ssa_for_alu_src(b, alu, 0),
85                                   nir_ssa_for_alu_src(b, alu, 1)));
86      break;
87   case nir_op_iabs: alu->op = nir_op_fabs; break;
88   case nir_op_ineg: alu->op = nir_op_fneg; break;
89   case nir_op_imax: alu->op = nir_op_fmax; break;
90   case nir_op_imin: alu->op = nir_op_fmin; break;
91   case nir_op_umax: alu->op = nir_op_fmax; break;
92   case nir_op_umin: alu->op = nir_op_fmin; break;
93
94   case nir_op_ball_iequal2:  alu->op = nir_op_ball_fequal2; break;
95   case nir_op_ball_iequal3:  alu->op = nir_op_ball_fequal3; break;
96   case nir_op_ball_iequal4:  alu->op = nir_op_ball_fequal4; break;
97   case nir_op_bany_inequal2: alu->op = nir_op_bany_fnequal2; break;
98   case nir_op_bany_inequal3: alu->op = nir_op_bany_fnequal3; break;
99   case nir_op_bany_inequal4: alu->op = nir_op_bany_fnequal4; break;
100
101   default:
102      assert(nir_alu_type_get_base_type(info->output_type) != nir_type_int &&
103             nir_alu_type_get_base_type(info->output_type) != nir_type_uint);
104      for (unsigned i = 0; i < info->num_inputs; i++) {
105         assert(nir_alu_type_get_base_type(info->input_types[i]) != nir_type_int &&
106                nir_alu_type_get_base_type(info->input_types[i]) != nir_type_uint);
107      }
108      return false;
109   }
110
111   if (rep) {
112      /* We've emitted a replacement instruction */
113      nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa, rep);
114      nir_instr_remove(&alu->instr);
115   }
116
117   return true;
118}
119
120static bool
121nir_lower_int_to_float_impl(nir_function_impl *impl)
122{
123   bool progress = false;
124   BITSET_WORD *float_types = NULL, *int_types = NULL;
125
126   nir_builder b;
127   nir_builder_init(&b, impl);
128
129   nir_index_ssa_defs(impl);
130   float_types = calloc(BITSET_WORDS(impl->ssa_alloc),
131                        sizeof(BITSET_WORD));
132   int_types = calloc(BITSET_WORDS(impl->ssa_alloc),
133                      sizeof(BITSET_WORD));
134   nir_gather_ssa_types(impl, float_types, int_types);
135
136   nir_foreach_block(block, impl) {
137      nir_foreach_instr_safe(instr, block) {
138         switch (instr->type) {
139         case nir_instr_type_alu:
140            progress |= lower_alu_instr(&b, nir_instr_as_alu(instr));
141            break;
142
143         case nir_instr_type_load_const: {
144            nir_load_const_instr *load = nir_instr_as_load_const(instr);
145            if (load->def.bit_size != 1 && BITSET_TEST(int_types, load->def.index)) {
146               for (unsigned i = 0; i < load->def.num_components; i++)
147                  load->value[i].f32 = load->value[i].i32;
148            }
149            break;
150         }
151
152         case nir_instr_type_intrinsic:
153         case nir_instr_type_ssa_undef:
154         case nir_instr_type_phi:
155         case nir_instr_type_tex:
156            break;
157
158         default:
159            nir_foreach_ssa_def(instr, assert_ssa_def_is_not_int, (void *)int_types);
160            break;
161         }
162      }
163   }
164
165   if (progress) {
166      nir_metadata_preserve(impl, nir_metadata_block_index |
167                                  nir_metadata_dominance);
168   } else {
169      nir_metadata_preserve(impl, nir_metadata_all);
170   }
171
172   free(float_types);
173   free(int_types);
174
175   return progress;
176}
177
178bool
179nir_lower_int_to_float(nir_shader *shader)
180{
181   bool progress = false;
182
183   nir_foreach_function(function, shader) {
184      if (function->impl && nir_lower_int_to_float_impl(function->impl))
185         progress = true;
186   }
187
188   return progress;
189}
190