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
|
/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
/*
* Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
*
* 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.
*
* Authors:
* Rob Clark <robclark@freedesktop.org>
*/
#include "util/u_math.h"
#include "ir3.h"
/*
* Instruction Depth:
*
* Calculates weighted instruction depth, ie. the sum of # of needed
* instructions plus delay slots back to original input (ie INPUT or
* CONST). That is to say, an instructions depth is:
*
* depth(instr) {
* d = 0;
* // for each src register:
* foreach (src in instr->regs[1..n])
* d = max(d, delayslots(src->instr, n) + depth(src->instr));
* return d + 1;
* }
*
* After an instruction's depth is calculated, it is inserted into the
* blocks depth sorted list, which is used by the scheduling pass.
*/
/* calculate required # of delay slots between the instruction that
* assigns a value and the one that consumes
*/
int ir3_delayslots(struct ir3_instruction *assigner,
struct ir3_instruction *consumer, unsigned n)
{
/* worst case is cat1-3 (alu) -> cat4/5 needing 6 cycles, normal
* alu -> alu needs 3 cycles, cat4 -> alu and texture fetch
* handled with sync bits
*/
if (is_meta(assigner))
return 0;
if (writes_addr(assigner))
return 6;
/* handled via sync flags: */
if (is_sfu(assigner) || is_tex(assigner))
return 0;
/* assigner must be alu: */
if (is_flow(consumer) || is_sfu(consumer) || is_tex(consumer)) {
return 6;
} else if ((consumer->category == 3) &&
is_mad(consumer->opc) && (n == 2)) {
/* special case, 3rd src to cat3 not required on first cycle */
return 1;
} else {
return 3;
}
}
static void insert_by_depth(struct ir3_instruction *instr)
{
struct ir3_block *block = instr->block;
struct ir3_instruction *n = block->head;
struct ir3_instruction *p = NULL;
while (n && (n != instr) && (n->depth > instr->depth)) {
p = n;
n = n->next;
}
instr->next = n;
if (p)
p->next = instr;
else
block->head = instr;
}
static void ir3_instr_depth(struct ir3_instruction *instr)
{
unsigned i;
/* if we've already visited this instruction, bail now: */
if (ir3_instr_check_mark(instr))
return;
instr->depth = 0;
for (i = 1; i < instr->regs_count; i++) {
struct ir3_register *src = instr->regs[i];
if (src->flags & IR3_REG_SSA) {
unsigned sd;
/* visit child to compute it's depth: */
ir3_instr_depth(src->instr);
sd = ir3_delayslots(src->instr, instr, i-1) +
src->instr->depth;
instr->depth = MAX2(instr->depth, sd);
}
}
/* meta-instructions don't add cycles, other than PHI.. which
* might translate to a real instruction..
*
* well, not entirely true, fan-in/out, etc might need to need
* to generate some extra mov's in edge cases, etc.. probably
* we might want to do depth calculation considering the worst
* case for these??
*/
if (!is_meta(instr))
instr->depth++;
insert_by_depth(instr);
}
void ir3_block_depth(struct ir3_block *block)
{
unsigned i;
block->head = NULL;
ir3_clear_mark(block->shader);
for (i = 0; i < block->noutputs; i++)
if (block->outputs[i])
ir3_instr_depth(block->outputs[i]);
/* mark un-used instructions: */
for (i = 0; i < block->shader->instrs_count; i++) {
struct ir3_instruction *instr = block->shader->instrs[i];
/* just consider instructions within this block: */
if (instr->block != block)
continue;
if (!ir3_instr_check_mark(instr))
instr->depth = DEPTH_UNUSED;
}
/* cleanup unused inputs: */
for (i = 0; i < block->ninputs; i++) {
struct ir3_instruction *in = block->inputs[i];
if (in && (in->depth == DEPTH_UNUSED))
block->inputs[i] = NULL;
}
}
|
