derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

gpu-compute: enable kernel-end WB functionality

Browse Source 
Change-Id: Ib17e1d700586d1aa04d408e7b924270f0de82efe
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29938
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Xianwei Zhang <xianwei.zhang@amd.com>
This commit is contained in:
Xianwei Zhang
2018-06-28 02:34:41 -04:00
committed by Anthony Gutierrez
parent 07fcbf16fc
commit 024f978cff
3 changed files with 27 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
src/gpu-compute/compute_unit.cc
View File

@@ -1218,23 +1218,25 @@ ComputeUnit::injectGlobalMemFence(GPUDynInstPtr gpuDynInst,
schedule(mem_req_event, curTick() + req_tick_latency);
} else {
assert(gpuDynInst->isEndOfKernel());
// kernel end release must be enabled
assert(shader->impl_kern_end_rel);
assert(gpuDynInst->isEndOfKernel());
req->setCacheCoherenceFlags(Request::RELEASE);
req->setReqInstSeqNum(gpuDynInst->seqNum());
req->setFlags(Request::KERNEL);
pkt = new Packet(req, MemCmd::MemSyncReq);
pkt->pushSenderState(
new ComputeUnit::DataPort::SenderState(gpuDynInst, 0, nullptr));
req->setCacheCoherenceFlags(Request::WB_L2);
req->setReqInstSeqNum(gpuDynInst->seqNum());
req->setFlags(Request::KERNEL);
pkt = new Packet(req, MemCmd::MemSyncReq);
pkt->pushSenderState(
new ComputeUnit::DataPort::SenderState(gpuDynInst, 0, nullptr));
EventFunctionWrapper *mem_req_event =
memPort[0]->createMemReqEvent(pkt);
EventFunctionWrapper *mem_req_event =
memPort[0]->createMemReqEvent(pkt);
DPRINTF(GPUPort, "CU%d: WF[%d][%d]: index %d, addr %#x scheduling "
"a release\n", cu_id, gpuDynInst->simdId,
gpuDynInst->wfSlotId, 0, pkt->req->getPaddr());
DPRINTF(GPUPort, "CU%d: WF[%d][%d]: index %d, addr %#x scheduling "
"a release\n", cu_id, gpuDynInst->simdId,
gpuDynInst->wfSlotId, 0, pkt->req->getPaddr());
schedule(mem_req_event, curTick() + req_tick_latency);
schedule(mem_req_event, curTick() + req_tick_latency);
}
} else {
gpuDynInst->setRequestFlags(req);