1/*
2 * Copyright © 2013 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 "brw_fs.h"
25#include "brw_fs_live_variables.h"
26#include "brw_cfg.h"
27
28/** @file brw_fs_saturate_propagation.cpp
29 *
30 * Implements a pass that propagates the SAT modifier from a MOV.SAT into the
31 * instruction that produced the source of the MOV.SAT, thereby allowing the
32 * MOV's src and dst to be coalesced and the MOV removed.
33 *
34 * For instance,
35 *
36 *    ADD     tmp, src0, src1
37 *    MOV.SAT dst, tmp
38 *
39 * would be transformed into
40 *
41 *    ADD.SAT tmp, src0, src1
42 *    MOV     dst, tmp
43 */
44
45static bool
46opt_saturate_propagation_local(fs_visitor *v, bblock_t *block)
47{
48   bool progress = false;
49   int ip = block->end_ip + 1;
50
51   foreach_inst_in_block_reverse(fs_inst, inst, block) {
52      ip--;
53
54      if (inst->opcode != BRW_OPCODE_MOV ||
55          !inst->saturate ||
56          inst->dst.file != VGRF ||
57          inst->dst.type != inst->src[0].type ||
58          inst->src[0].file != VGRF ||
59          inst->src[0].abs)
60         continue;
61
62      int src_var = v->live_intervals->var_from_reg(inst->src[0]);
63      int src_end_ip = v->live_intervals->end[src_var];
64
65      bool interfered = false;
66      foreach_inst_in_block_reverse_starting_from(fs_inst, scan_inst, inst) {
67         if (scan_inst->exec_size == inst->exec_size &&
68             regions_overlap(scan_inst->dst, scan_inst->size_written,
69                             inst->src[0], inst->size_read(0))) {
70            if (scan_inst->is_partial_write() ||
71                (scan_inst->dst.type != inst->dst.type &&
72                 !scan_inst->can_change_types()))
73               break;
74
75            if (scan_inst->saturate) {
76               inst->saturate = false;
77               progress = true;
78            } else if (src_end_ip == ip || inst->dst.equals(inst->src[0])) {
79               if (scan_inst->can_do_saturate()) {
80                  if (scan_inst->dst.type != inst->dst.type) {
81                     scan_inst->dst.type = inst->dst.type;
82                     for (int i = 0; i < scan_inst->sources; i++) {
83                        scan_inst->src[i].type = inst->dst.type;
84                     }
85                  }
86
87                  if (inst->src[0].negate) {
88                     if (scan_inst->opcode == BRW_OPCODE_MUL) {
89                        scan_inst->src[0].negate = !scan_inst->src[0].negate;
90                        inst->src[0].negate = false;
91                     } else if (scan_inst->opcode == BRW_OPCODE_MAD) {
92                        for (int i = 0; i < 2; i++) {
93                           if (scan_inst->src[i].file == IMM) {
94                              brw_negate_immediate(scan_inst->src[i].type,
95                                                   &scan_inst->src[i].as_brw_reg());
96                           } else {
97                              scan_inst->src[i].negate = !scan_inst->src[i].negate;
98                           }
99                        }
100                        inst->src[0].negate = false;
101                     } else if (scan_inst->opcode == BRW_OPCODE_ADD) {
102                        if (scan_inst->src[1].file == IMM) {
103                           if (!brw_negate_immediate(scan_inst->src[1].type,
104                                                     &scan_inst->src[1].as_brw_reg())) {
105                              break;
106                           }
107                        } else {
108                           scan_inst->src[1].negate = !scan_inst->src[1].negate;
109                        }
110                        scan_inst->src[0].negate = !scan_inst->src[0].negate;
111                        inst->src[0].negate = false;
112                     } else {
113                        break;
114                     }
115                  }
116
117                  scan_inst->saturate = true;
118                  inst->saturate = false;
119                  progress = true;
120               }
121            }
122            break;
123         }
124         for (int i = 0; i < scan_inst->sources; i++) {
125            if (scan_inst->src[i].file == VGRF &&
126                scan_inst->src[i].nr == inst->src[0].nr &&
127                scan_inst->src[i].offset / REG_SIZE ==
128                 inst->src[0].offset / REG_SIZE) {
129               if (scan_inst->opcode != BRW_OPCODE_MOV ||
130                   !scan_inst->saturate ||
131                   scan_inst->src[0].abs ||
132                   scan_inst->src[0].negate ||
133                   scan_inst->src[0].abs != inst->src[0].abs ||
134                   scan_inst->src[0].negate != inst->src[0].negate) {
135                  interfered = true;
136                  break;
137               }
138            }
139         }
140
141         if (interfered)
142            break;
143      }
144   }
145
146   return progress;
147}
148
149bool
150fs_visitor::opt_saturate_propagation()
151{
152   bool progress = false;
153
154   calculate_live_intervals();
155
156   foreach_block (block, cfg) {
157      progress = opt_saturate_propagation_local(this, block) || progress;
158   }
159
160   /* Live intervals are still valid. */
161
162   return progress;
163}
164