1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
|
/*
* Copyright © 2011 Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
extern "C" {
#include "main/macros.h"
#include "program/register_allocate.h"
} /* extern "C" */
#include "brw_vec4.h"
#include "glsl/ir_print_visitor.h"
using namespace brw;
namespace brw {
static void
assign(int *reg_hw_locations, reg *reg)
{
if (reg->file == GRF) {
reg->reg = reg_hw_locations[reg->reg];
}
}
void
vec4_visitor::reg_allocate_trivial()
{
int hw_reg_mapping[this->virtual_grf_count];
bool virtual_grf_used[this->virtual_grf_count];
int i;
int next;
/* Calculate which virtual GRFs are actually in use after whatever
* optimization passes have occurred.
*/
for (int i = 0; i < this->virtual_grf_count; i++) {
virtual_grf_used[i] = false;
}
foreach_iter(exec_list_iterator, iter, this->instructions) {
vec4_instruction *inst = (vec4_instruction *)iter.get();
if (inst->dst.file == GRF)
virtual_grf_used[inst->dst.reg] = true;
for (int i = 0; i < 3; i++) {
if (inst->src[i].file == GRF)
virtual_grf_used[inst->src[i].reg] = true;
}
}
hw_reg_mapping[0] = this->first_non_payload_grf;
next = hw_reg_mapping[0] + this->virtual_grf_sizes[0];
for (i = 1; i < this->virtual_grf_count; i++) {
if (virtual_grf_used[i]) {
hw_reg_mapping[i] = next;
next += this->virtual_grf_sizes[i];
}
}
prog_data->total_grf = next;
foreach_iter(exec_list_iterator, iter, this->instructions) {
vec4_instruction *inst = (vec4_instruction *)iter.get();
assign(hw_reg_mapping, &inst->dst);
assign(hw_reg_mapping, &inst->src[0]);
assign(hw_reg_mapping, &inst->src[1]);
assign(hw_reg_mapping, &inst->src[2]);
}
if (prog_data->total_grf > BRW_MAX_GRF) {
fail("Ran out of regs on trivial allocator (%d/%d)\n",
prog_data->total_grf, BRW_MAX_GRF);
}
}
static void
brw_alloc_reg_set_for_classes(struct brw_context *brw,
int *class_sizes,
int class_count,
int base_reg_count)
{
/* Compute the total number of registers across all classes. */
int ra_reg_count = 0;
for (int i = 0; i < class_count; i++) {
ra_reg_count += base_reg_count - (class_sizes[i] - 1);
}
ralloc_free(brw->vs.ra_reg_to_grf);
brw->vs.ra_reg_to_grf = ralloc_array(brw, uint8_t, ra_reg_count);
ralloc_free(brw->vs.regs);
brw->vs.regs = ra_alloc_reg_set(ra_reg_count);
ralloc_free(brw->vs.classes);
brw->vs.classes = ralloc_array(brw, int, class_count + 1);
/* Now, add the registers to their classes, and add the conflicts
* between them and the base GRF registers (and also each other).
*/
int reg = 0;
for (int i = 0; i < class_count; i++) {
int class_reg_count = base_reg_count - (class_sizes[i] - 1);
brw->vs.classes[i] = ra_alloc_reg_class(brw->vs.regs);
for (int j = 0; j < class_reg_count; j++) {
ra_class_add_reg(brw->vs.regs, brw->vs.classes[i], reg);
brw->vs.ra_reg_to_grf[reg] = j;
for (int base_reg = j;
base_reg < j + class_sizes[i];
base_reg++) {
ra_add_transitive_reg_conflict(brw->vs.regs, base_reg, reg);
}
reg++;
}
}
assert(reg == ra_reg_count);
ra_set_finalize(brw->vs.regs);
}
void
vec4_visitor::reg_allocate()
{
int hw_reg_mapping[virtual_grf_count];
int first_assigned_grf = this->first_non_payload_grf;
int base_reg_count = BRW_MAX_GRF - first_assigned_grf;
int class_sizes[base_reg_count];
int class_count = 0;
/* Using the trivial allocator can be useful in debugging undefined
* register access as a result of broken optimization passes.
*/
if (0) {
reg_allocate_trivial();
return;
}
calculate_live_intervals();
/* Set up the register classes.
*
* The base registers store a vec4. However, we'll need larger
* storage for arrays, structures, and matrices, which will be sets
* of contiguous registers.
*/
class_sizes[class_count++] = 1;
for (int r = 0; r < virtual_grf_count; r++) {
int i;
for (i = 0; i < class_count; i++) {
if (class_sizes[i] == this->virtual_grf_sizes[r])
break;
}
if (i == class_count) {
if (this->virtual_grf_sizes[r] >= base_reg_count) {
fail("Object too large to register allocate.\n");
}
class_sizes[class_count++] = this->virtual_grf_sizes[r];
}
}
brw_alloc_reg_set_for_classes(brw, class_sizes, class_count, base_reg_count);
struct ra_graph *g = ra_alloc_interference_graph(brw->vs.regs,
virtual_grf_count);
for (int i = 0; i < virtual_grf_count; i++) {
for (int c = 0; c < class_count; c++) {
if (class_sizes[c] == this->virtual_grf_sizes[i]) {
ra_set_node_class(g, i, brw->vs.classes[c]);
break;
}
}
for (int j = 0; j < i; j++) {
if (virtual_grf_interferes(i, j)) {
ra_add_node_interference(g, i, j);
}
}
}
if (!ra_allocate_no_spills(g)) {
ralloc_free(g);
fail("No register spilling support yet\n");
return;
}
/* Get the chosen virtual registers for each node, and map virtual
* regs in the register classes back down to real hardware reg
* numbers.
*/
prog_data->total_grf = first_assigned_grf;
for (int i = 0; i < virtual_grf_count; i++) {
int reg = ra_get_node_reg(g, i);
hw_reg_mapping[i] = first_assigned_grf + brw->vs.ra_reg_to_grf[reg];
prog_data->total_grf = MAX2(prog_data->total_grf,
hw_reg_mapping[i] + virtual_grf_sizes[i]);
}
foreach_list(node, &this->instructions) {
vec4_instruction *inst = (vec4_instruction *)node;
assign(hw_reg_mapping, &inst->dst);
assign(hw_reg_mapping, &inst->src[0]);
assign(hw_reg_mapping, &inst->src[1]);
assign(hw_reg_mapping, &inst->src[2]);
}
ralloc_free(g);
}
} /* namespace brw */
|
