2 * Copyright © 2012, 2013, 2014 Intel Corporation
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:
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
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
29 /** @file brw_vec4_cse.cpp
31 * Support for local common subexpression elimination.
33 * See Muchnick's Advanced Compiler Design and Implementation, section
38 struct aeb_entry
: public exec_node
{
39 /** The instruction that generates the expression value. */
40 vec4_instruction
*generator
;
42 /** The temporary where the value is stored. */
48 is_expression(const vec4_instruction
*const inst
)
50 switch (inst
->opcode
) {
73 case SHADER_OPCODE_RCP
:
74 case SHADER_OPCODE_RSQ
:
75 case SHADER_OPCODE_SQRT
:
76 case SHADER_OPCODE_EXP2
:
77 case SHADER_OPCODE_LOG2
:
78 case SHADER_OPCODE_POW
:
79 case SHADER_OPCODE_INT_QUOTIENT
:
80 case SHADER_OPCODE_INT_REMAINDER
:
81 case SHADER_OPCODE_SIN
:
82 case SHADER_OPCODE_COS
:
83 return inst
->mlen
== 0;
90 is_expression_commutative(enum opcode op
)
105 operands_match(enum opcode op
, src_reg
*xs
, src_reg
*ys
)
107 if (!is_expression_commutative(op
)) {
108 return xs
[0].equals(ys
[0]) && xs
[1].equals(ys
[1]) && xs
[2].equals(ys
[2]);
110 return (xs
[0].equals(ys
[0]) && xs
[1].equals(ys
[1])) ||
111 (xs
[1].equals(ys
[0]) && xs
[0].equals(ys
[1]));
116 instructions_match(vec4_instruction
*a
, vec4_instruction
*b
)
118 return a
->opcode
== b
->opcode
&&
119 a
->saturate
== b
->saturate
&&
120 a
->conditional_mod
== b
->conditional_mod
&&
121 a
->dst
.type
== b
->dst
.type
&&
122 a
->dst
.writemask
== b
->dst
.writemask
&&
123 operands_match(a
->opcode
, a
->src
, b
->src
);
127 vec4_visitor::opt_cse_local(bblock_t
*block
)
129 bool progress
= false;
132 void *cse_ctx
= ralloc_context(NULL
);
134 int ip
= block
->start_ip
;
135 foreach_inst_in_block (vec4_instruction
, inst
, block
) {
136 /* Skip some cases. */
137 if (is_expression(inst
) && !inst
->predicate
&& inst
->mlen
== 0 &&
138 (inst
->dst
.file
!= HW_REG
|| inst
->dst
.is_null()))
142 foreach_in_list_use_after(aeb_entry
, entry
, &aeb
) {
143 /* Match current instruction's expression against those in AEB. */
144 if (instructions_match(inst
, entry
->generator
)) {
152 /* Our first sighting of this expression. Create an entry. */
153 aeb_entry
*entry
= ralloc(cse_ctx
, aeb_entry
);
154 entry
->tmp
= src_reg(); /* file will be BAD_FILE */
155 entry
->generator
= inst
;
156 aeb
.push_tail(entry
);
158 /* This is at least our second sighting of this expression.
159 * If we don't have a temporary already, make one.
161 bool no_existing_temp
= entry
->tmp
.file
== BAD_FILE
;
162 if (no_existing_temp
&& !entry
->generator
->dst
.is_null()) {
163 entry
->tmp
= src_reg(this, glsl_type::float_type
);
164 entry
->tmp
.type
= inst
->dst
.type
;
165 entry
->tmp
.swizzle
= BRW_SWIZZLE_XYZW
;
167 vec4_instruction
*copy
= MOV(entry
->generator
->dst
, entry
->tmp
);
168 entry
->generator
->insert_after(copy
);
169 entry
->generator
->dst
= dst_reg(entry
->tmp
);
173 if (!inst
->dst
.is_null()) {
174 assert(inst
->dst
.type
== entry
->tmp
.type
);
175 vec4_instruction
*copy
= MOV(inst
->dst
, entry
->tmp
);
176 copy
->force_writemask_all
= inst
->force_writemask_all
;
177 inst
->insert_before(copy
);
180 /* Set our iterator so that next time through the loop inst->next
181 * will get the instruction in the basic block after the one we've
184 vec4_instruction
*prev
= (vec4_instruction
*)inst
->prev
;
188 /* Appending an instruction may have changed our bblock end. */
189 if (inst
== block
->end
) {
197 foreach_in_list_safe(aeb_entry
, entry
, &aeb
) {
198 /* Kill all AEB entries that write a different value to or read from
199 * the flag register if we just wrote it.
201 if (inst
->writes_flag()) {
202 if (entry
->generator
->reads_flag() ||
203 (entry
->generator
->writes_flag() &&
204 !instructions_match(inst
, entry
->generator
))) {
211 for (int i
= 0; i
< 3; i
++) {
212 src_reg
*src
= &entry
->generator
->src
[i
];
214 /* Kill all AEB entries that use the destination we just
217 if (inst
->dst
.file
== entry
->generator
->src
[i
].file
&&
218 inst
->dst
.reg
== entry
->generator
->src
[i
].reg
) {
224 /* Kill any AEB entries using registers that don't get reused any
225 * more -- a sure sign they'll fail operands_match().
227 int last_reg_use
= MAX2(MAX2(virtual_grf_end
[src
->reg
* 4 + 0],
228 virtual_grf_end
[src
->reg
* 4 + 1]),
229 MAX2(virtual_grf_end
[src
->reg
* 4 + 2],
230 virtual_grf_end
[src
->reg
* 4 + 3]));
231 if (src
->file
== GRF
&& last_reg_use
< ip
) {
242 ralloc_free(cse_ctx
);
248 vec4_visitor::opt_cse()
250 bool progress
= false;
252 calculate_live_intervals();
254 for (int b
= 0; b
< cfg
->num_blocks
; b
++) {
255 bblock_t
*block
= cfg
->blocks
[b
];
257 progress
= opt_cse_local(block
) || progress
;
261 invalidate_live_intervals();