1b8e80941Smrg/*
2b8e80941Smrg * Copyright © 2017 Intel Corporation
3b8e80941Smrg *
4b8e80941Smrg * Permission is hereby granted, free of charge, to any person obtaining a
5b8e80941Smrg * copy of this software and associated documentation files (the "Software"),
6b8e80941Smrg * to deal in the Software without restriction, including without limitation
7b8e80941Smrg * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8b8e80941Smrg * and/or sell copies of the Software, and to permit persons to whom the
9b8e80941Smrg * Software is furnished to do so, subject to the following conditions:
10b8e80941Smrg *
11b8e80941Smrg * The above copyright notice and this permission notice (including the next
12b8e80941Smrg * paragraph) shall be included in all copies or substantial portions of the
13b8e80941Smrg * Software.
14b8e80941Smrg *
15b8e80941Smrg * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16b8e80941Smrg * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17b8e80941Smrg * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18b8e80941Smrg * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19b8e80941Smrg * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20b8e80941Smrg * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21b8e80941Smrg * IN THE SOFTWARE.
22b8e80941Smrg */
23b8e80941Smrg
24b8e80941Smrg#include "nir.h"
25b8e80941Smrg#include "nir_builder.h"
26b8e80941Smrg
27b8e80941Smrg/**
28b8e80941Smrg * \file nir_opt_intrinsics.c
29b8e80941Smrg */
30b8e80941Smrg
31b8e80941Smrgstatic bool
32b8e80941Smrgopt_intrinsics_impl(nir_function_impl *impl,
33b8e80941Smrg                    const struct nir_shader_compiler_options *options)
34b8e80941Smrg{
35b8e80941Smrg   nir_builder b;
36b8e80941Smrg   nir_builder_init(&b, impl);
37b8e80941Smrg   bool progress = false;
38b8e80941Smrg
39b8e80941Smrg   nir_foreach_block(block, impl) {
40b8e80941Smrg      nir_foreach_instr_safe(instr, block) {
41b8e80941Smrg         if (instr->type != nir_instr_type_intrinsic)
42b8e80941Smrg            continue;
43b8e80941Smrg
44b8e80941Smrg         nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
45b8e80941Smrg         nir_ssa_def *replacement = NULL;
46b8e80941Smrg         b.cursor = nir_before_instr(instr);
47b8e80941Smrg
48b8e80941Smrg         switch (intrin->intrinsic) {
49b8e80941Smrg         case nir_intrinsic_vote_any:
50b8e80941Smrg         case nir_intrinsic_vote_all:
51b8e80941Smrg            if (nir_src_is_const(intrin->src[0]))
52b8e80941Smrg               replacement = nir_ssa_for_src(&b, intrin->src[0], 1);
53b8e80941Smrg            break;
54b8e80941Smrg         case nir_intrinsic_vote_feq:
55b8e80941Smrg         case nir_intrinsic_vote_ieq:
56b8e80941Smrg            if (nir_src_is_const(intrin->src[0]))
57b8e80941Smrg               replacement = nir_imm_true(&b);
58b8e80941Smrg            break;
59b8e80941Smrg         case nir_intrinsic_load_sample_mask_in:
60b8e80941Smrg            /* Transform:
61b8e80941Smrg             *   gl_SampleMaskIn == 0 ---> gl_HelperInvocation
62b8e80941Smrg             *   gl_SampleMaskIn != 0 ---> !gl_HelperInvocation
63b8e80941Smrg             */
64b8e80941Smrg            if (!options->optimize_sample_mask_in)
65b8e80941Smrg               continue;
66b8e80941Smrg
67b8e80941Smrg            nir_foreach_use_safe(use_src, &intrin->dest.ssa) {
68b8e80941Smrg               if (use_src->parent_instr->type == nir_instr_type_alu) {
69b8e80941Smrg                  nir_alu_instr *alu = nir_instr_as_alu(use_src->parent_instr);
70b8e80941Smrg
71b8e80941Smrg                  if (alu->op == nir_op_ieq ||
72b8e80941Smrg                      alu->op == nir_op_ine) {
73b8e80941Smrg                     /* Check for 0 in either operand. */
74b8e80941Smrg                     nir_const_value *const_val =
75b8e80941Smrg                         nir_src_as_const_value(alu->src[0].src);
76b8e80941Smrg                     if (!const_val)
77b8e80941Smrg                        const_val = nir_src_as_const_value(alu->src[1].src);
78b8e80941Smrg                     if (!const_val || const_val->i32 != 0)
79b8e80941Smrg                        continue;
80b8e80941Smrg
81b8e80941Smrg                     nir_ssa_def *new_expr = nir_load_helper_invocation(&b, 1);
82b8e80941Smrg
83b8e80941Smrg                     if (alu->op == nir_op_ine)
84b8e80941Smrg                        new_expr = nir_inot(&b, new_expr);
85b8e80941Smrg
86b8e80941Smrg                     nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa,
87b8e80941Smrg                                              nir_src_for_ssa(new_expr));
88b8e80941Smrg                     nir_instr_remove(&alu->instr);
89b8e80941Smrg                     continue;
90b8e80941Smrg                  }
91b8e80941Smrg               }
92b8e80941Smrg            }
93b8e80941Smrg            continue;
94b8e80941Smrg         default:
95b8e80941Smrg            break;
96b8e80941Smrg         }
97b8e80941Smrg
98b8e80941Smrg         if (!replacement)
99b8e80941Smrg            continue;
100b8e80941Smrg
101b8e80941Smrg         nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
102b8e80941Smrg                                  nir_src_for_ssa(replacement));
103b8e80941Smrg         nir_instr_remove(instr);
104b8e80941Smrg         progress = true;
105b8e80941Smrg      }
106b8e80941Smrg   }
107b8e80941Smrg
108b8e80941Smrg   return progress;
109b8e80941Smrg}
110b8e80941Smrg
111b8e80941Smrgbool
112b8e80941Smrgnir_opt_intrinsics(nir_shader *shader)
113b8e80941Smrg{
114b8e80941Smrg   bool progress = false;
115b8e80941Smrg
116b8e80941Smrg   nir_foreach_function(function, shader) {
117b8e80941Smrg      if (!function->impl)
118b8e80941Smrg         continue;
119b8e80941Smrg
120b8e80941Smrg      if (opt_intrinsics_impl(function->impl, shader->options)) {
121b8e80941Smrg         progress = true;
122b8e80941Smrg         nir_metadata_preserve(function->impl, nir_metadata_block_index |
123b8e80941Smrg                                               nir_metadata_dominance);
124b8e80941Smrg      }
125b8e80941Smrg   }
126b8e80941Smrg
127b8e80941Smrg   return progress;
128b8e80941Smrg}
129