2
* Copyright © 2014 Broadcom
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
25
* @file v3d_opt_copy_propagation.c
27
* This implements simple copy propagation for VIR without control flow.
29
* For each temp, it keeps a qreg of which source it was MOVed from, if it
30
* was. If we see that used later, we can just reuse the source value, since
31
* we know we don't have control flow, and we have SSA for our values so
32
* there's no killing to worry about.
35
#include "v3d_compiler.h"
38
is_copy_mov(struct qinst *inst)
43
if (inst->qpu.type != V3D_QPU_INSTR_TYPE_ALU ||
44
(inst->qpu.alu.mul.op != V3D_QPU_M_FMOV &&
45
inst->qpu.alu.mul.op != V3D_QPU_M_MOV)) {
49
if (inst->dst.file != QFILE_TEMP)
52
if (inst->src[0].file != QFILE_TEMP)
55
if (inst->qpu.alu.add.output_pack != V3D_QPU_PACK_NONE ||
56
inst->qpu.alu.mul.output_pack != V3D_QPU_PACK_NONE) {
60
if (inst->qpu.flags.ac != V3D_QPU_COND_NONE ||
61
inst->qpu.flags.mc != V3D_QPU_COND_NONE) {
65
switch (inst->src[0].file) {
67
/* No copy propagating from R3/R4/R5 -- the MOVs from those
68
* are there to register allocate values produced into R3/4/5
69
* to other regs (though hopefully r3/4/5).
71
switch (inst->src[0].index) {
72
case V3D_QPU_WADDR_R3:
73
case V3D_QPU_WADDR_R4:
74
case V3D_QPU_WADDR_R5:
82
switch (inst->src[0].index) {
86
/* MOVs from rf0/1/2 are only to track the live
87
* intervals for W/centroid W/Z.
101
vir_has_unpack(struct qinst *inst, int chan)
103
assert(chan == 0 || chan == 1);
105
if (vir_is_add(inst)) {
107
return inst->qpu.alu.add.a_unpack != V3D_QPU_UNPACK_NONE;
109
return inst->qpu.alu.add.b_unpack != V3D_QPU_UNPACK_NONE;
112
return inst->qpu.alu.mul.a_unpack != V3D_QPU_UNPACK_NONE;
114
return inst->qpu.alu.mul.b_unpack != V3D_QPU_UNPACK_NONE;
119
try_copy_prop(struct v3d_compile *c, struct qinst *inst, struct qinst **movs)
122
bool progress = false;
124
for (int i = 0; i < vir_get_nsrc(inst); i++) {
125
if (inst->src[i].file != QFILE_TEMP)
128
/* We have two ways of finding MOVs we can copy propagate
129
* from. One is if it's an SSA def: then we can reuse it from
130
* any block in the program, as long as its source is also an
131
* SSA def. Alternatively, if it's in the "movs" array
132
* tracked within the block, then we know the sources for it
133
* haven't been changed since we saw the instruction within
136
struct qinst *mov = movs[inst->src[i].index];
138
if (!is_copy_mov(c->defs[inst->src[i].index]))
140
mov = c->defs[inst->src[i].index];
142
if (mov->src[0].file == QFILE_TEMP &&
143
!c->defs[mov->src[0].index])
147
if (vir_has_unpack(mov, 0)) {
148
/* Make sure that the meaning of the unpack
149
* would be the same between the two
152
if (v3d_qpu_unpacks_f32(&inst->qpu) !=
153
v3d_qpu_unpacks_f32(&mov->qpu) ||
154
v3d_qpu_unpacks_f16(&inst->qpu) !=
155
v3d_qpu_unpacks_f16(&mov->qpu)) {
159
/* No composing the unpacks. */
160
if (vir_has_unpack(inst, i))
163
/* these ops can't represent abs. */
164
if (mov->qpu.alu.mul.a_unpack == V3D_QPU_UNPACK_ABS) {
165
switch (inst->qpu.alu.add.op) {
166
case V3D_QPU_A_VFPACK:
167
case V3D_QPU_A_FROUND:
168
case V3D_QPU_A_FTRUNC:
169
case V3D_QPU_A_FFLOOR:
170
case V3D_QPU_A_FCEIL:
173
case V3D_QPU_A_FTOIN:
174
case V3D_QPU_A_FTOIZ:
175
case V3D_QPU_A_FTOUZ:
185
fprintf(stderr, "Copy propagate: ");
186
vir_dump_inst(c, inst);
187
fprintf(stderr, "\n");
190
inst->src[i] = mov->src[0];
191
if (vir_has_unpack(mov, 0)) {
192
enum v3d_qpu_input_unpack unpack = mov->qpu.alu.mul.a_unpack;
194
vir_set_unpack(inst, i, unpack);
198
fprintf(stderr, "to: ");
199
vir_dump_inst(c, inst);
200
fprintf(stderr, "\n");
210
apply_kills(struct v3d_compile *c, struct qinst **movs, struct qinst *inst)
212
if (inst->dst.file != QFILE_TEMP)
215
for (int i = 0; i < c->num_temps; i++) {
217
(movs[i]->dst.index == inst->dst.index ||
218
(movs[i]->src[0].file == QFILE_TEMP &&
219
movs[i]->src[0].index == inst->dst.index))) {
226
vir_opt_copy_propagate(struct v3d_compile *c)
228
bool progress = false;
231
movs = ralloc_array(c, struct qinst *, c->num_temps);
235
vir_for_each_block(block, c) {
236
/* The MOVs array tracks only available movs within the
239
memset(movs, 0, sizeof(struct qinst *) * c->num_temps);
241
c->cur_block = block;
242
vir_for_each_inst(inst, block) {
244
progress = try_copy_prop(c, inst, movs) || progress;
246
apply_kills(c, movs, inst);
248
if (is_copy_mov(inst))
249
movs[inst->dst.index] = inst;