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
|
/*
* Copyright (c) 2014-2015 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.
*
* Authors: John Kalamatianos,
* Sooraj Puthoor
*/
#include "gpu-compute/global_memory_pipeline.hh"
#include "debug/GPUMem.hh"
#include "debug/GPUReg.hh"
#include "gpu-compute/compute_unit.hh"
#include "gpu-compute/gpu_dyn_inst.hh"
#include "gpu-compute/shader.hh"
#include "gpu-compute/vector_register_file.hh"
#include "gpu-compute/wavefront.hh"
GlobalMemPipeline::GlobalMemPipeline(const ComputeUnitParams* p) :
computeUnit(nullptr), gmQueueSize(p->global_mem_queue_size),
outOfOrderDataDelivery(p->out_of_order_data_delivery), inflightStores(0),
inflightLoads(0)
{
}
void
GlobalMemPipeline::init(ComputeUnit *cu)
{
computeUnit = cu;
globalMemSize = computeUnit->shader->globalMemSize;
_name = computeUnit->name() + ".GlobalMemPipeline";
}
void
GlobalMemPipeline::exec()
{
// apply any returned global memory operations
GPUDynInstPtr m = getNextReadyResp();
bool accessVrf = true;
Wavefront *w = nullptr;
// check the VRF to see if the operands of a load (or load component
// of an atomic) are accessible
if ((m) && (m->isLoad() || m->isAtomicRet())) {
w = m->wavefront();
accessVrf =
w->computeUnit->vrf[w->simdId]->
vrfOperandAccessReady(m->seqNum(), w, m, VrfAccessType::WRITE);
}
if (m && m->latency.rdy() && computeUnit->glbMemToVrfBus.rdy() &&
accessVrf && m->statusBitVector == VectorMask(0) &&
(computeUnit->shader->coissue_return ||
computeUnit->wfWait.at(m->pipeId).rdy())) {
w = m->wavefront();
m->completeAcc(m);
completeRequest(m);
// Decrement outstanding register count
computeUnit->shader->ScheduleAdd(&w->outstandingReqs, m->time, -1);
if (m->isStore() || m->isAtomic()) {
computeUnit->shader->ScheduleAdd(&w->outstandingReqsWrGm,
m->time, -1);
}
if (m->isLoad() || m->isAtomic()) {
computeUnit->shader->ScheduleAdd(&w->outstandingReqsRdGm,
m->time, -1);
}
// Mark write bus busy for appropriate amount of time
computeUnit->glbMemToVrfBus.set(m->time);
if (!computeUnit->shader->coissue_return)
w->computeUnit->wfWait.at(m->pipeId).set(m->time);
}
// If pipeline has executed a global memory instruction
// execute global memory packets and issue global
// memory packets to DTLB
if (!gmIssuedRequests.empty()) {
GPUDynInstPtr mp = gmIssuedRequests.front();
if (mp->isLoad() || mp->isAtomic()) {
if (inflightLoads >= gmQueueSize) {
return;
} else {
++inflightLoads;
}
} else if (mp->isStore()) {
if (inflightStores >= gmQueueSize) {
return;
} else {
++inflightStores;
}
}
mp->initiateAcc(mp);
if (!outOfOrderDataDelivery && !mp->isMemFence()) {
/**
* if we are not in out-of-order data delivery mode
* then we keep the responses sorted in program order.
* in order to do so we must reserve an entry in the
* resp buffer before we issue the request to the mem
* system. mem fence requests will not be stored here
* because once they are issued from the GM pipeline,
* they do not send any response back to it.
*/
gmOrderedRespBuffer.insert(std::make_pair(mp->seqNum(),
std::make_pair(mp, false)));
}
gmIssuedRequests.pop();
DPRINTF(GPUMem, "CU%d: WF[%d][%d] Popping 0 mem_op = \n",
computeUnit->cu_id, mp->simdId, mp->wfSlotId);
}
}
GPUDynInstPtr
GlobalMemPipeline::getNextReadyResp()
{
if (outOfOrderDataDelivery) {
if (!gmReturnedLoads.empty()) {
return gmReturnedLoads.front();
} else if (!gmReturnedStores.empty()) {
return gmReturnedStores.front();
}
} else {
if (!gmOrderedRespBuffer.empty()) {
auto mem_req = gmOrderedRespBuffer.begin();
if (mem_req->second.second) {
return mem_req->second.first;
}
}
}
return nullptr;
}
void
GlobalMemPipeline::completeRequest(GPUDynInstPtr gpuDynInst)
{
if (gpuDynInst->isLoad() || gpuDynInst->isAtomic()) {
assert(inflightLoads > 0);
--inflightLoads;
} else if (gpuDynInst->isStore()) {
assert(inflightStores > 0);
--inflightStores;
}
if (outOfOrderDataDelivery) {
if (gpuDynInst->isLoad() || gpuDynInst->isAtomic()) {
assert(!gmReturnedLoads.empty());
gmReturnedLoads.pop();
} else if (gpuDynInst->isStore()) {
assert(!gmReturnedStores.empty());
gmReturnedStores.pop();
}
} else {
// we should only pop the oldest requst, and it
// should be marked as done if we are here
assert(gmOrderedRespBuffer.begin()->first == gpuDynInst->seqNum());
assert(gmOrderedRespBuffer.begin()->second.first == gpuDynInst);
assert(gmOrderedRespBuffer.begin()->second.second);
// remove this instruction from the buffer by its
// unique seq ID
gmOrderedRespBuffer.erase(gpuDynInst->seqNum());
}
}
void
GlobalMemPipeline::issueRequest(GPUDynInstPtr gpuDynInst)
{
gmIssuedRequests.push(gpuDynInst);
}
void
GlobalMemPipeline::handleResponse(GPUDynInstPtr gpuDynInst)
{
if (outOfOrderDataDelivery) {
if (gpuDynInst->isLoad() || gpuDynInst->isAtomic()) {
assert(isGMLdRespFIFOWrRdy());
gmReturnedLoads.push(gpuDynInst);
} else {
assert(isGMStRespFIFOWrRdy());
gmReturnedStores.push(gpuDynInst);
}
} else {
auto mem_req = gmOrderedRespBuffer.find(gpuDynInst->seqNum());
// if we are getting a response for this mem request,
// then it ought to already be in the ordered response
// buffer
assert(mem_req != gmOrderedRespBuffer.end());
mem_req->second.second = true;
}
}
void
GlobalMemPipeline::regStats()
{
loadVrfBankConflictCycles
.name(name() + ".load_vrf_bank_conflict_cycles")
.desc("total number of cycles GM data are delayed before updating "
"the VRF")
;
}
|
