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
|
/* -*- 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 "pipe/p_shader_tokens.h"
#include "util/u_math.h"
#include "freedreno_util.h"
#include "ir3.h"
/*
* Legalize:
*
* We currently require that scheduling ensures that we have enough nop's
* in all the right places. The legalize step mostly handles fixing up
* instruction flags ((ss)/(sy)/(ei)), and collapses sequences of nop's
* into fewer nop's w/ rpt flag.
*/
struct ir3_legalize_ctx {
struct ir3_block *block;
bool has_samp;
int max_bary;
};
static void legalize(struct ir3_legalize_ctx *ctx)
{
struct ir3_block *block = ctx->block;
struct ir3_instruction *n;
struct ir3 *shader = block->shader;
struct ir3_instruction *end =
ir3_instr_create(block, 0, OPC_END);
struct ir3_instruction *last_input = NULL;
struct ir3_instruction *last_rel = NULL;
regmask_t needs_ss_war; /* write after read */
regmask_t needs_ss;
regmask_t needs_sy;
regmask_init(&needs_ss_war);
regmask_init(&needs_ss);
regmask_init(&needs_sy);
shader->instrs_count = 0;
for (n = block->head; n; n = n->next) {
struct ir3_register *reg;
unsigned i;
if (is_meta(n))
continue;
if (is_input(n)) {
struct ir3_register *inloc = n->regs[1];
assert(inloc->flags & IR3_REG_IMMED);
ctx->max_bary = MAX2(ctx->max_bary, inloc->iim_val);
}
for (i = 1; i < n->regs_count; i++) {
reg = n->regs[i];
if (reg_gpr(reg)) {
/* TODO: we probably only need (ss) for alu
* instr consuming sfu result.. need to make
* some tests for both this and (sy)..
*/
if (regmask_get(&needs_ss, reg)) {
n->flags |= IR3_INSTR_SS;
regmask_init(&needs_ss);
}
if (regmask_get(&needs_sy, reg)) {
n->flags |= IR3_INSTR_SY;
regmask_init(&needs_sy);
}
}
/* TODO: is it valid to have address reg loaded from a
* relative src (ie. mova a0, c<a0.x+4>)? If so, the
* last_rel check below should be moved ahead of this:
*/
if (reg->flags & IR3_REG_RELATIV)
last_rel = n;
}
if (n->regs_count > 0) {
reg = n->regs[0];
if (regmask_get(&needs_ss_war, reg)) {
n->flags |= IR3_INSTR_SS;
regmask_init(&needs_ss_war); // ??? I assume?
}
if (last_rel && (reg->num == regid(REG_A0, 0))) {
last_rel->flags |= IR3_INSTR_UL;
last_rel = NULL;
}
}
/* cat5+ does not have an (ss) bit, if needed we need to
* insert a nop to carry the sync flag. Would be kinda
* clever if we were aware of this during scheduling, but
* this should be a pretty rare case:
*/
if ((n->flags & IR3_INSTR_SS) && (n->category >= 5)) {
struct ir3_instruction *nop;
nop = ir3_instr_create(block, 0, OPC_NOP);
nop->flags |= IR3_INSTR_SS;
n->flags &= ~IR3_INSTR_SS;
}
/* need to be able to set (ss) on first instruction: */
if ((shader->instrs_count == 0) && (n->category >= 5))
ir3_instr_create(block, 0, OPC_NOP);
if (is_nop(n) && shader->instrs_count) {
struct ir3_instruction *last =
shader->instrs[shader->instrs_count-1];
if (is_nop(last) && (last->repeat < 5)) {
last->repeat++;
last->flags |= n->flags;
continue;
}
}
shader->instrs[shader->instrs_count++] = n;
if (is_sfu(n))
regmask_set(&needs_ss, n->regs[0]);
if (is_tex(n)) {
/* this ends up being the # of samp instructions.. but that
* is ok, everything else only cares whether it is zero or
* not. We do this here, rather than when we encounter a
* SAMP decl, because (especially in binning pass shader)
* the samp instruction(s) could get eliminated if the
* result is not used.
*/
ctx->has_samp = true;
regmask_set(&needs_sy, n->regs[0]);
}
/* both tex/sfu appear to not always immediately consume
* their src register(s):
*/
if (is_tex(n) || is_sfu(n)) {
for (i = 1; i < n->regs_count; i++) {
reg = n->regs[i];
if (reg_gpr(reg))
regmask_set(&needs_ss_war, reg);
}
}
if (is_input(n))
last_input = n;
}
if (last_input)
last_input->regs[0]->flags |= IR3_REG_EI;
if (last_rel)
last_rel->flags |= IR3_INSTR_UL;
shader->instrs[shader->instrs_count++] = end;
shader->instrs[0]->flags |= IR3_INSTR_SS | IR3_INSTR_SY;
}
void ir3_block_legalize(struct ir3_block *block,
bool *has_samp, int *max_bary)
{
struct ir3_legalize_ctx ctx = {
.block = block,
.max_bary = -1,
};
legalize(&ctx);
*has_samp = ctx.has_samp;
*max_bary = ctx.max_bary;
}
|
