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
|
/*
* Copyright 2011 Christoph Bumiller
* 2014 Red Hat Inc.
*
* 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 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.
*/
#include "codegen/nv50_ir.h"
#include "codegen/nv50_ir_build_util.h"
#include "codegen/nv50_ir_target_nvc0.h"
#include "codegen/nv50_ir_lowering_gm107.h"
#include <limits>
namespace nv50_ir {
#define QOP_ADD 0
#define QOP_SUBR 1
#define QOP_SUB 2
#define QOP_MOV2 3
// UL UR LL LR
#define QUADOP(q, r, s, t) \
((QOP_##q << 6) | (QOP_##r << 4) | \
(QOP_##s << 2) | (QOP_##t << 0))
bool
GM107LoweringPass::handleManualTXD(TexInstruction *i)
{
static const uint8_t qOps[4][2] =
{
{ QUADOP(MOV2, ADD, MOV2, ADD), QUADOP(MOV2, MOV2, ADD, ADD) }, // l0
{ QUADOP(SUBR, MOV2, SUBR, MOV2), QUADOP(MOV2, MOV2, ADD, ADD) }, // l1
{ QUADOP(MOV2, ADD, MOV2, ADD), QUADOP(SUBR, SUBR, MOV2, MOV2) }, // l2
{ QUADOP(SUBR, MOV2, SUBR, MOV2), QUADOP(SUBR, SUBR, MOV2, MOV2) }, // l3
};
Value *def[4][4];
Value *crd[3];
Value *tmp;
Instruction *tex, *add;
Value *zero = bld.loadImm(bld.getSSA(), 0);
int l, c;
const int dim = i->tex.target.getDim();
const int array = i->tex.target.isArray();
i->op = OP_TEX; // no need to clone dPdx/dPdy later
for (c = 0; c < dim; ++c)
crd[c] = bld.getScratch();
tmp = bld.getScratch();
for (l = 0; l < 4; ++l) {
// mov coordinates from lane l to all lanes
bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
for (c = 0; c < dim; ++c) {
bld.mkOp2(OP_SHFL, TYPE_F32, crd[c], i->getSrc(c + array), bld.mkImm(l));
add = bld.mkOp2(OP_QUADOP, TYPE_F32, crd[c], crd[c], zero);
add->subOp = 0x00;
add->lanes = 1; /* abused for .ndv */
}
// add dPdx from lane l to lanes dx
for (c = 0; c < dim; ++c) {
bld.mkOp2(OP_SHFL, TYPE_F32, tmp, i->dPdx[c].get(), bld.mkImm(l));
add = bld.mkOp2(OP_QUADOP, TYPE_F32, crd[c], tmp, crd[c]);
add->subOp = qOps[l][0];
add->lanes = 1; /* abused for .ndv */
}
// add dPdy from lane l to lanes dy
for (c = 0; c < dim; ++c) {
bld.mkOp2(OP_SHFL, TYPE_F32, tmp, i->dPdy[c].get(), bld.mkImm(l));
add = bld.mkOp2(OP_QUADOP, TYPE_F32, crd[c], tmp, crd[c]);
add->subOp = qOps[l][1];
add->lanes = 1; /* abused for .ndv */
}
// texture
bld.insert(tex = cloneForward(func, i));
for (c = 0; c < dim; ++c)
tex->setSrc(c + array, crd[c]);
bld.mkOp(OP_QUADPOP, TYPE_NONE, NULL);
// save results
for (c = 0; i->defExists(c); ++c) {
Instruction *mov;
def[c][l] = bld.getSSA();
mov = bld.mkMov(def[c][l], tex->getDef(c));
mov->fixed = 1;
mov->lanes = 1 << l;
}
}
for (c = 0; i->defExists(c); ++c) {
Instruction *u = bld.mkOp(OP_UNION, TYPE_U32, i->getDef(c));
for (l = 0; l < 4; ++l)
u->setSrc(l, def[c][l]);
}
i->bb->remove(i);
return true;
}
bool
GM107LoweringPass::handleDFDX(Instruction *insn)
{
Instruction *shfl;
int qop = 0, xid = 0;
switch (insn->op) {
case OP_DFDX:
qop = QUADOP(SUB, SUBR, SUB, SUBR);
xid = 1;
break;
case OP_DFDY:
qop = QUADOP(SUB, SUB, SUBR, SUBR);
xid = 2;
break;
default:
assert(!"invalid dfdx opcode");
break;
}
shfl = bld.mkOp2(OP_SHFL, TYPE_F32, bld.getScratch(),
insn->getSrc(0), bld.mkImm(xid));
shfl->subOp = NV50_IR_SUBOP_SHFL_BFLY;
insn->op = OP_QUADOP;
insn->subOp = qop;
insn->lanes = 0; /* abused for !.ndv */
insn->setSrc(1, insn->getSrc(0));
insn->setSrc(0, shfl->getDef(0));
return true;
}
bool
GM107LoweringPass::handlePFETCH(Instruction *i)
{
Value *tmp0 = bld.getScratch();
Value *tmp1 = bld.getScratch();
Value *tmp2 = bld.getScratch();
bld.mkOp1(OP_RDSV, TYPE_U32, tmp0, bld.mkSysVal(SV_INVOCATION_INFO, 0));
bld.mkOp2(OP_SHR , TYPE_U32, tmp1, tmp0, bld.mkImm(16));
bld.mkOp2(OP_AND , TYPE_U32, tmp0, tmp0, bld.mkImm(0xff));
bld.mkOp2(OP_AND , TYPE_U32, tmp1, tmp1, bld.mkImm(0xff));
if (i->getSrc(1))
bld.mkOp2(OP_ADD , TYPE_U32, tmp2, i->getSrc(0), i->getSrc(1));
else
bld.mkOp1(OP_MOV , TYPE_U32, tmp2, i->getSrc(0));
bld.mkOp3(OP_MAD , TYPE_U32, tmp0, tmp0, tmp1, tmp2);
i->setSrc(0, tmp0);
i->setSrc(1, NULL);
return true;
}
bool
GM107LoweringPass::handlePOPCNT(Instruction *i)
{
Value *tmp = bld.mkOp2v(OP_AND, i->sType, bld.getScratch(),
i->getSrc(0), i->getSrc(1));
i->setSrc(0, tmp);
i->setSrc(1, NULL);
return true;
}
//
// - add quadop dance for texturing
// - put FP outputs in GPRs
// - convert instruction sequences
//
bool
GM107LoweringPass::visit(Instruction *i)
{
bld.setPosition(i, false);
if (i->cc != CC_ALWAYS)
checkPredicate(i);
switch (i->op) {
case OP_PFETCH:
return handlePFETCH(i);
case OP_DFDX:
case OP_DFDY:
return handleDFDX(i);
case OP_POPCNT:
return handlePOPCNT(i);
default:
return NVC0LoweringPass::visit(i);
}
}
} // namespace nv50_ir
|
