1d7787adda8006506545256547d8d590a282487afEric Anholt/*
2d7787adda8006506545256547d8d590a282487afEric Anholt * Copyright © 2012 Intel Corporation
3d7787adda8006506545256547d8d590a282487afEric Anholt *
4d7787adda8006506545256547d8d590a282487afEric Anholt * Permission is hereby granted, free of charge, to any person obtaining a
5d7787adda8006506545256547d8d590a282487afEric Anholt * copy of this software and associated documentation files (the "Software"),
6d7787adda8006506545256547d8d590a282487afEric Anholt * to deal in the Software without restriction, including without limitation
7d7787adda8006506545256547d8d590a282487afEric Anholt * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8d7787adda8006506545256547d8d590a282487afEric Anholt * and/or sell copies of the Software, and to permit persons to whom the
9d7787adda8006506545256547d8d590a282487afEric Anholt * Software is furnished to do so, subject to the following conditions:
10d7787adda8006506545256547d8d590a282487afEric Anholt *
11d7787adda8006506545256547d8d590a282487afEric Anholt * The above copyright notice and this permission notice (including the next
12d7787adda8006506545256547d8d590a282487afEric Anholt * paragraph) shall be included in all copies or substantial portions of the
13d7787adda8006506545256547d8d590a282487afEric Anholt * Software.
14d7787adda8006506545256547d8d590a282487afEric Anholt *
15d7787adda8006506545256547d8d590a282487afEric Anholt * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16d7787adda8006506545256547d8d590a282487afEric Anholt * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17d7787adda8006506545256547d8d590a282487afEric Anholt * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18d7787adda8006506545256547d8d590a282487afEric Anholt * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19d7787adda8006506545256547d8d590a282487afEric Anholt * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20d7787adda8006506545256547d8d590a282487afEric Anholt * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21d7787adda8006506545256547d8d590a282487afEric Anholt * IN THE SOFTWARE.
22d7787adda8006506545256547d8d590a282487afEric Anholt */
23d7787adda8006506545256547d8d590a282487afEric Anholt
24d7787adda8006506545256547d8d590a282487afEric Anholt#include "brw_fs.h"
25d7787adda8006506545256547d8d590a282487afEric Anholt#include "brw_fs_cfg.h"
26d7787adda8006506545256547d8d590a282487afEric Anholt
27d7787adda8006506545256547d8d590a282487afEric Anholtnamespace { /* avoid conflict with opt_copy_propagation_elements */
28d7787adda8006506545256547d8d590a282487afEric Anholtstruct acp_entry : public exec_node {
29d7787adda8006506545256547d8d590a282487afEric Anholt   fs_reg dst;
30d7787adda8006506545256547d8d590a282487afEric Anholt   fs_reg src;
31d7787adda8006506545256547d8d590a282487afEric Anholt};
32d7787adda8006506545256547d8d590a282487afEric Anholt}
33d7787adda8006506545256547d8d590a282487afEric Anholt
34458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholtbool
35458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholtfs_visitor::try_copy_propagate(fs_inst *inst, int arg, acp_entry *entry)
36458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt{
37458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt   if (inst->src[arg].file != entry->dst.file ||
38458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt       inst->src[arg].reg != entry->dst.reg ||
39458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt       inst->src[arg].reg_offset != entry->dst.reg_offset) {
40458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt      return false;
41458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt   }
42458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt
430c4630bae001139dea42b78cd08157de4d90542bEric Anholt   /* See resolve_ud_negate() and comment in brw_fs_emit.cpp. */
440c4630bae001139dea42b78cd08157de4d90542bEric Anholt   if (inst->conditional_mod &&
450c4630bae001139dea42b78cd08157de4d90542bEric Anholt       inst->src[arg].type == BRW_REGISTER_TYPE_UD &&
460c4630bae001139dea42b78cd08157de4d90542bEric Anholt       entry->src.negate)
470c4630bae001139dea42b78cd08157de4d90542bEric Anholt      return false;
480c4630bae001139dea42b78cd08157de4d90542bEric Anholt
490c4630bae001139dea42b78cd08157de4d90542bEric Anholt   bool has_source_modifiers = entry->src.abs || entry->src.negate;
500c4630bae001139dea42b78cd08157de4d90542bEric Anholt
51dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt   if (intel->gen == 6 && inst->is_math() &&
52dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt       (has_source_modifiers || entry->src.file == UNIFORM))
530c4630bae001139dea42b78cd08157de4d90542bEric Anholt      return false;
540c4630bae001139dea42b78cd08157de4d90542bEric Anholt
55dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt   inst->src[arg].file = entry->src.file;
56458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt   inst->src[arg].reg = entry->src.reg;
57458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt   inst->src[arg].reg_offset = entry->src.reg_offset;
58458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt
590c4630bae001139dea42b78cd08157de4d90542bEric Anholt   if (!inst->src[arg].abs) {
600c4630bae001139dea42b78cd08157de4d90542bEric Anholt      inst->src[arg].abs = entry->src.abs;
610c4630bae001139dea42b78cd08157de4d90542bEric Anholt      inst->src[arg].negate ^= entry->src.negate;
620c4630bae001139dea42b78cd08157de4d90542bEric Anholt   }
630c4630bae001139dea42b78cd08157de4d90542bEric Anholt
64458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt   return true;
65458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt}
66458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt
67d7787adda8006506545256547d8d590a282487afEric Anholt/** @file brw_fs_copy_propagation.cpp
68d7787adda8006506545256547d8d590a282487afEric Anholt *
69d7787adda8006506545256547d8d590a282487afEric Anholt * Support for local copy propagation by walking the list of instructions
70d7787adda8006506545256547d8d590a282487afEric Anholt * and maintaining the ACP table of available copies for propagation.
71d7787adda8006506545256547d8d590a282487afEric Anholt *
72d7787adda8006506545256547d8d590a282487afEric Anholt * See Muchnik's Advanced Compiler Design and Implementation, section
73d7787adda8006506545256547d8d590a282487afEric Anholt * 12.5 (p356).
74d7787adda8006506545256547d8d590a282487afEric Anholt */
75d7787adda8006506545256547d8d590a282487afEric Anholt
76d7787adda8006506545256547d8d590a282487afEric Anholt/* Walks a basic block and does copy propagation on it using the acp
77d7787adda8006506545256547d8d590a282487afEric Anholt * list.
78d7787adda8006506545256547d8d590a282487afEric Anholt */
79d7787adda8006506545256547d8d590a282487afEric Anholtbool
80d7787adda8006506545256547d8d590a282487afEric Anholtfs_visitor::opt_copy_propagate_local(void *mem_ctx,
81d7787adda8006506545256547d8d590a282487afEric Anholt				     fs_bblock *block, exec_list *acp)
82d7787adda8006506545256547d8d590a282487afEric Anholt{
83d7787adda8006506545256547d8d590a282487afEric Anholt   bool progress = false;
84d7787adda8006506545256547d8d590a282487afEric Anholt
85d7787adda8006506545256547d8d590a282487afEric Anholt   for (fs_inst *inst = block->start;
86d7787adda8006506545256547d8d590a282487afEric Anholt	inst != block->end->next;
87d7787adda8006506545256547d8d590a282487afEric Anholt	inst = (fs_inst *)inst->next) {
88d7787adda8006506545256547d8d590a282487afEric Anholt
89d7787adda8006506545256547d8d590a282487afEric Anholt      /* Try propagating into this instruction. */
90d7787adda8006506545256547d8d590a282487afEric Anholt      foreach_list(entry_node, acp) {
91d7787adda8006506545256547d8d590a282487afEric Anholt	 acp_entry *entry = (acp_entry *)entry_node;
92d7787adda8006506545256547d8d590a282487afEric Anholt
93d7787adda8006506545256547d8d590a282487afEric Anholt	 for (int i = 0; i < 3; i++) {
94458f7f014139deb48a4cf0a9e6bdca3a57d24208Eric Anholt	    if (try_copy_propagate(inst, i, entry))
95d7787adda8006506545256547d8d590a282487afEric Anholt	       progress = true;
96d7787adda8006506545256547d8d590a282487afEric Anholt	 }
97d7787adda8006506545256547d8d590a282487afEric Anholt      }
98d7787adda8006506545256547d8d590a282487afEric Anholt
99d7787adda8006506545256547d8d590a282487afEric Anholt      /* kill the destination from the ACP */
100d7787adda8006506545256547d8d590a282487afEric Anholt      if (inst->dst.file == GRF) {
101d7787adda8006506545256547d8d590a282487afEric Anholt	 foreach_list_safe(entry_node, acp) {
102d7787adda8006506545256547d8d590a282487afEric Anholt	    acp_entry *entry = (acp_entry *)entry_node;
103d7787adda8006506545256547d8d590a282487afEric Anholt
104a454f8ec6df9334df42249be910cc2d57d913bffEric Anholt	    if (inst->overwrites_reg(entry->dst) ||
105a454f8ec6df9334df42249be910cc2d57d913bffEric Anholt                inst->overwrites_reg(entry->src)) {
106d7787adda8006506545256547d8d590a282487afEric Anholt	       entry->remove();
107d7787adda8006506545256547d8d590a282487afEric Anholt	    }
108d7787adda8006506545256547d8d590a282487afEric Anholt	 }
109d7787adda8006506545256547d8d590a282487afEric Anholt      }
110d7787adda8006506545256547d8d590a282487afEric Anholt
111d7787adda8006506545256547d8d590a282487afEric Anholt      /* If this instruction is a raw copy, add it to the ACP. */
112d7787adda8006506545256547d8d590a282487afEric Anholt      if (inst->opcode == BRW_OPCODE_MOV &&
113d7787adda8006506545256547d8d590a282487afEric Anholt	  inst->dst.file == GRF &&
114dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt	  ((inst->src[0].file == GRF &&
115dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt	    (inst->src[0].reg != inst->dst.reg ||
116dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt	     inst->src[0].reg_offset != inst->dst.reg_offset)) ||
117dd4282e38fd92c081875da6bce0b2345bd472532Eric Anholt	   inst->src[0].file == UNIFORM) &&
118d7787adda8006506545256547d8d590a282487afEric Anholt	  inst->src[0].type == inst->dst.type &&
119d7787adda8006506545256547d8d590a282487afEric Anholt	  !inst->saturate &&
120d7787adda8006506545256547d8d590a282487afEric Anholt	  !inst->predicated &&
121d7787adda8006506545256547d8d590a282487afEric Anholt	  !inst->force_uncompressed &&
122d7787adda8006506545256547d8d590a282487afEric Anholt	  !inst->force_sechalf &&
1230c4630bae001139dea42b78cd08157de4d90542bEric Anholt	  inst->src[0].smear == -1) {
124d7787adda8006506545256547d8d590a282487afEric Anholt	 acp_entry *entry = ralloc(mem_ctx, acp_entry);
125d7787adda8006506545256547d8d590a282487afEric Anholt	 entry->dst = inst->dst;
126d7787adda8006506545256547d8d590a282487afEric Anholt	 entry->src = inst->src[0];
127d7787adda8006506545256547d8d590a282487afEric Anholt	 acp->push_tail(entry);
128d7787adda8006506545256547d8d590a282487afEric Anholt      }
129d7787adda8006506545256547d8d590a282487afEric Anholt   }
130d7787adda8006506545256547d8d590a282487afEric Anholt
131d7787adda8006506545256547d8d590a282487afEric Anholt   return progress;
132d7787adda8006506545256547d8d590a282487afEric Anholt}
133d7787adda8006506545256547d8d590a282487afEric Anholt
134d7787adda8006506545256547d8d590a282487afEric Anholtbool
135d7787adda8006506545256547d8d590a282487afEric Anholtfs_visitor::opt_copy_propagate()
136d7787adda8006506545256547d8d590a282487afEric Anholt{
137d7787adda8006506545256547d8d590a282487afEric Anholt   bool progress = false;
138d7787adda8006506545256547d8d590a282487afEric Anholt   void *mem_ctx = ralloc_context(this->mem_ctx);
139d7787adda8006506545256547d8d590a282487afEric Anholt
140d7787adda8006506545256547d8d590a282487afEric Anholt   fs_cfg cfg(this);
141d7787adda8006506545256547d8d590a282487afEric Anholt
142d7787adda8006506545256547d8d590a282487afEric Anholt   for (int b = 0; b < cfg.num_blocks; b++) {
143d7787adda8006506545256547d8d590a282487afEric Anholt      fs_bblock *block = cfg.blocks[b];
144d7787adda8006506545256547d8d590a282487afEric Anholt      exec_list acp;
145d7787adda8006506545256547d8d590a282487afEric Anholt
146d7787adda8006506545256547d8d590a282487afEric Anholt      progress = opt_copy_propagate_local(mem_ctx, block, &acp) || progress;
147d7787adda8006506545256547d8d590a282487afEric Anholt   }
148d7787adda8006506545256547d8d590a282487afEric Anholt
149d7787adda8006506545256547d8d590a282487afEric Anholt   ralloc_free(mem_ctx);
150d7787adda8006506545256547d8d590a282487afEric Anholt
1511e28f55ab7909496d93ab1b552faad17453c10acEric Anholt   if (progress)
1521e28f55ab7909496d93ab1b552faad17453c10acEric Anholt      live_intervals_valid = false;
1531e28f55ab7909496d93ab1b552faad17453c10acEric Anholt
154d7787adda8006506545256547d8d590a282487afEric Anholt   return progress;
155d7787adda8006506545256547d8d590a282487afEric Anholt}
156