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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
|
/*
* Copyright (c) 2015 Advanced Micro Devices, Inc.
* All rights reserved.
*
* For use for simulation and test purposes only
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
* Author: John Kalamatianos
*/
#include "gpu-compute/vector_register_file.hh"
#include <string>
#include "base/misc.hh"
#include "gpu-compute/compute_unit.hh"
#include "gpu-compute/gpu_dyn_inst.hh"
#include "gpu-compute/shader.hh"
#include "gpu-compute/simple_pool_manager.hh"
#include "gpu-compute/wavefront.hh"
#include "params/VectorRegisterFile.hh"
VectorRegisterFile::VectorRegisterFile(const VectorRegisterFileParams *p)
: SimObject(p),
manager(new SimplePoolManager(p->min_alloc, p->num_regs_per_simd)),
simdId(p->simd_id), numRegsPerSimd(p->num_regs_per_simd),
vgprState(new VecRegisterState())
{
fatal_if(numRegsPerSimd % 2, "VRF size is illegal\n");
fatal_if(simdId < 0, "Illegal SIMD id for VRF");
fatal_if(numRegsPerSimd % p->min_alloc, "Min VGPR region allocation is not "
"multiple of VRF size\n");
busy.clear();
busy.resize(numRegsPerSimd, 0);
nxtBusy.clear();
nxtBusy.resize(numRegsPerSimd, 0);
vgprState->init(numRegsPerSimd, p->wfSize);
}
void
VectorRegisterFile::setParent(ComputeUnit *_computeUnit)
{
computeUnit = _computeUnit;
vgprState->setParent(computeUnit);
}
uint8_t
VectorRegisterFile::regNxtBusy(int idx, uint32_t operandSize) const
{
uint8_t status = nxtBusy.at(idx);
if (operandSize > 4) {
status = status | (nxtBusy.at((idx + 1) % numRegs()));
}
return status;
}
uint8_t
VectorRegisterFile::regBusy(int idx, uint32_t operandSize) const
{
uint8_t status = busy.at(idx);
if (operandSize > 4) {
status = status | (busy.at((idx + 1) % numRegs()));
}
return status;
}
void
VectorRegisterFile::preMarkReg(int regIdx, uint32_t operandSize, uint8_t value)
{
nxtBusy.at(regIdx) = value;
if (operandSize > 4) {
nxtBusy.at((regIdx + 1) % numRegs()) = value;
}
}
void
VectorRegisterFile::markReg(int regIdx, uint32_t operandSize, uint8_t value)
{
busy.at(regIdx) = value;
if (operandSize > 4) {
busy.at((regIdx + 1) % numRegs()) = value;
}
}
bool
VectorRegisterFile::operandsReady(Wavefront *w, GPUDynInstPtr ii) const
{
for (int i = 0; i < ii->getNumOperands(); ++i) {
if (ii->isVectorRegister(i)) {
uint32_t vgprIdx = ii->getRegisterIndex(i);
uint32_t pVgpr = w->remap(vgprIdx, ii->getOperandSize(i), 1);
if (regBusy(pVgpr, ii->getOperandSize(i)) == 1) {
if (ii->isDstOperand(i)) {
w->numTimesBlockedDueWAXDependencies++;
} else if (ii->isSrcOperand(i)) {
w->numTimesBlockedDueRAWDependencies++;
}
return false;
}
if (regNxtBusy(pVgpr, ii->getOperandSize(i)) == 1) {
if (ii->isDstOperand(i)) {
w->numTimesBlockedDueWAXDependencies++;
} else if (ii->isSrcOperand(i)) {
w->numTimesBlockedDueRAWDependencies++;
}
return false;
}
}
}
return true;
}
void
VectorRegisterFile::exec(GPUDynInstPtr ii, Wavefront *w)
{
bool loadInstr = ii->isLoad();
bool atomicInstr = ii->isAtomic() || ii->isMemFence();
bool loadNoArgInstr = loadInstr && !ii->isArgLoad();
// iterate over all register destination operands
for (int i = 0; i < ii->getNumOperands(); ++i) {
if (ii->isVectorRegister(i) && ii->isDstOperand(i)) {
uint32_t physReg = w->remap(ii->getRegisterIndex(i),
ii->getOperandSize(i), 1);
// mark the destination vector register as busy
markReg(physReg, ii->getOperandSize(i), 1);
// clear the in-flight status of the destination vector register
preMarkReg(physReg, ii->getOperandSize(i), 0);
// FIXME: if we ever model correct timing behavior
// for load argument instructions then we should not
// set the destination register as busy now but when
// the data returns. Loads and Atomics should free
// their destination registers when the data returns,
// not now
if (!atomicInstr && !loadNoArgInstr) {
uint32_t pipeLen = ii->getOperandSize(i) <= 4 ?
computeUnit->spBypassLength() :
computeUnit->dpBypassLength();
// schedule an event for marking the register as ready
computeUnit->registerEvent(w->simdId, physReg,
ii->getOperandSize(i),
computeUnit->shader->tick_cnt +
computeUnit->shader->ticks(pipeLen),
0);
}
}
}
}
int
VectorRegisterFile::exec(uint64_t dynamic_id, Wavefront *w,
std::vector<uint32_t> ®Vec, uint32_t operandSize,
uint64_t timestamp)
{
int delay = 0;
panic_if(regVec.size() <= 0, "Illegal VGPR vector size=%d\n",
regVec.size());
for (int i = 0; i < regVec.size(); ++i) {
// mark the destination VGPR as free when the timestamp expires
computeUnit->registerEvent(w->simdId, regVec[i], operandSize,
computeUnit->shader->tick_cnt + timestamp +
computeUnit->shader->ticks(delay), 0);
}
return delay;
}
void
VectorRegisterFile::updateResources(Wavefront *w, GPUDynInstPtr ii)
{
// iterate over all register destination operands
for (int i = 0; i < ii->getNumOperands(); ++i) {
if (ii->isVectorRegister(i) && ii->isDstOperand(i)) {
uint32_t physReg = w->remap(ii->getRegisterIndex(i),
ii->getOperandSize(i), 1);
// set the in-flight status of the destination vector register
preMarkReg(physReg, ii->getOperandSize(i), 1);
}
}
}
bool
VectorRegisterFile::vrfOperandAccessReady(uint64_t dynamic_id, Wavefront *w,
GPUDynInstPtr ii,
VrfAccessType accessType)
{
bool ready = true;
return ready;
}
bool
VectorRegisterFile::vrfOperandAccessReady(Wavefront *w, GPUDynInstPtr ii,
VrfAccessType accessType)
{
bool ready = true;
return ready;
}
VectorRegisterFile*
VectorRegisterFileParams::create()
{
return new VectorRegisterFile(this);
}
|