i965/nir: Sort uniforms direct-first and use two different uniform registers
[mesa.git] / src / mesa / drivers / dri / i965 / brw_vec4_cse.cpp
1 /*
2 * Copyright © 2012, 2013, 2014 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_vec4.h"
25 #include "brw_cfg.h"
26
27 using namespace brw;
28
29 /** @file brw_vec4_cse.cpp
30 *
31 * Support for local common subexpression elimination.
32 *
33 * See Muchnick's Advanced Compiler Design and Implementation, section
34 * 13.1 (p378).
35 */
36
37 namespace {
38 struct aeb_entry : public exec_node {
39 /** The instruction that generates the expression value. */
40 vec4_instruction *generator;
41
42 /** The temporary where the value is stored. */
43 src_reg tmp;
44 };
45 }
46
47 static bool
48 is_expression(const vec4_instruction *const inst)
49 {
50 switch (inst->opcode) {
51 case BRW_OPCODE_MOV:
52 case BRW_OPCODE_SEL:
53 case BRW_OPCODE_NOT:
54 case BRW_OPCODE_AND:
55 case BRW_OPCODE_OR:
56 case BRW_OPCODE_XOR:
57 case BRW_OPCODE_SHR:
58 case BRW_OPCODE_SHL:
59 case BRW_OPCODE_ASR:
60 case BRW_OPCODE_CMP:
61 case BRW_OPCODE_CMPN:
62 case BRW_OPCODE_ADD:
63 case BRW_OPCODE_MUL:
64 case BRW_OPCODE_FRC:
65 case BRW_OPCODE_RNDU:
66 case BRW_OPCODE_RNDD:
67 case BRW_OPCODE_RNDE:
68 case BRW_OPCODE_RNDZ:
69 case BRW_OPCODE_LINE:
70 case BRW_OPCODE_PLN:
71 case BRW_OPCODE_MAD:
72 case BRW_OPCODE_LRP:
73 case VEC4_OPCODE_UNPACK_UNIFORM:
74 return true;
75 case SHADER_OPCODE_RCP:
76 case SHADER_OPCODE_RSQ:
77 case SHADER_OPCODE_SQRT:
78 case SHADER_OPCODE_EXP2:
79 case SHADER_OPCODE_LOG2:
80 case SHADER_OPCODE_POW:
81 case SHADER_OPCODE_INT_QUOTIENT:
82 case SHADER_OPCODE_INT_REMAINDER:
83 case SHADER_OPCODE_SIN:
84 case SHADER_OPCODE_COS:
85 return inst->mlen == 0;
86 default:
87 return false;
88 }
89 }
90
91 static bool
92 operands_match(const vec4_instruction *a, const vec4_instruction *b)
93 {
94 const src_reg *xs = a->src;
95 const src_reg *ys = b->src;
96
97 if (a->opcode == BRW_OPCODE_MAD) {
98 return xs[0].equals(ys[0]) &&
99 ((xs[1].equals(ys[1]) && xs[2].equals(ys[2])) ||
100 (xs[2].equals(ys[1]) && xs[1].equals(ys[2])));
101 } else if (!a->is_commutative()) {
102 return xs[0].equals(ys[0]) && xs[1].equals(ys[1]) && xs[2].equals(ys[2]);
103 } else {
104 return (xs[0].equals(ys[0]) && xs[1].equals(ys[1])) ||
105 (xs[1].equals(ys[0]) && xs[0].equals(ys[1]));
106 }
107 }
108
109 static bool
110 instructions_match(vec4_instruction *a, vec4_instruction *b)
111 {
112 return a->opcode == b->opcode &&
113 a->saturate == b->saturate &&
114 a->conditional_mod == b->conditional_mod &&
115 a->dst.type == b->dst.type &&
116 a->dst.writemask == b->dst.writemask &&
117 operands_match(a, b);
118 }
119
120 bool
121 vec4_visitor::opt_cse_local(bblock_t *block)
122 {
123 bool progress = false;
124 exec_list aeb;
125
126 void *cse_ctx = ralloc_context(NULL);
127
128 int ip = block->start_ip;
129 foreach_inst_in_block (vec4_instruction, inst, block) {
130 /* Skip some cases. */
131 if (is_expression(inst) && !inst->predicate && inst->mlen == 0 &&
132 (inst->dst.file != HW_REG || inst->dst.is_null()))
133 {
134 bool found = false;
135
136 foreach_in_list_use_after(aeb_entry, entry, &aeb) {
137 /* Match current instruction's expression against those in AEB. */
138 if (!(entry->generator->dst.is_null() && !inst->dst.is_null()) &&
139 instructions_match(inst, entry->generator)) {
140 found = true;
141 progress = true;
142 break;
143 }
144 }
145
146 if (!found) {
147 if (inst->opcode != BRW_OPCODE_MOV ||
148 (inst->opcode == BRW_OPCODE_MOV &&
149 inst->src[0].file == IMM &&
150 inst->src[0].type == BRW_REGISTER_TYPE_VF)) {
151 /* Our first sighting of this expression. Create an entry. */
152 aeb_entry *entry = ralloc(cse_ctx, aeb_entry);
153 entry->tmp = src_reg(); /* file will be BAD_FILE */
154 entry->generator = inst;
155 aeb.push_tail(entry);
156 }
157 } else {
158 /* This is at least our second sighting of this expression.
159 * If we don't have a temporary already, make one.
160 */
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;
166
167 vec4_instruction *copy = MOV(entry->generator->dst, entry->tmp);
168 entry->generator->insert_after(block, copy);
169 entry->generator->dst = dst_reg(entry->tmp);
170 }
171
172 /* dest <- temp */
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(block, copy);
178 }
179
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
182 * removed.
183 */
184 vec4_instruction *prev = (vec4_instruction *)inst->prev;
185
186 inst->remove(block);
187 inst = prev;
188 }
189 }
190
191 foreach_in_list_safe(aeb_entry, entry, &aeb) {
192 /* Kill all AEB entries that write a different value to or read from
193 * the flag register if we just wrote it.
194 */
195 if (inst->writes_flag()) {
196 if (entry->generator->reads_flag() ||
197 (entry->generator->writes_flag() &&
198 !instructions_match(inst, entry->generator))) {
199 entry->remove();
200 ralloc_free(entry);
201 continue;
202 }
203 }
204
205 for (int i = 0; i < 3; i++) {
206 src_reg *src = &entry->generator->src[i];
207
208 /* Kill all AEB entries that use the destination we just
209 * overwrote.
210 */
211 if (inst->dst.file == entry->generator->src[i].file &&
212 inst->dst.reg == entry->generator->src[i].reg) {
213 entry->remove();
214 ralloc_free(entry);
215 break;
216 }
217
218 /* Kill any AEB entries using registers that don't get reused any
219 * more -- a sure sign they'll fail operands_match().
220 */
221 if (src->file == GRF) {
222 assert((unsigned)(src->reg * 4 + 3) < (alloc.count * 4));
223
224 int last_reg_use = MAX2(MAX2(virtual_grf_end[src->reg * 4 + 0],
225 virtual_grf_end[src->reg * 4 + 1]),
226 MAX2(virtual_grf_end[src->reg * 4 + 2],
227 virtual_grf_end[src->reg * 4 + 3]));
228 if (last_reg_use < ip) {
229 entry->remove();
230 ralloc_free(entry);
231 break;
232 }
233 }
234 }
235 }
236
237 ip++;
238 }
239
240 ralloc_free(cse_ctx);
241
242 return progress;
243 }
244
245 bool
246 vec4_visitor::opt_cse()
247 {
248 bool progress = false;
249
250 calculate_live_intervals();
251
252 foreach_block (block, cfg) {
253 progress = opt_cse_local(block) || progress;
254 }
255
256 if (progress)
257 invalidate_live_intervals();
258
259 return progress;
260 }