cwpearson/astaroth
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Some concurrency optimizations for 3D blocking

Browse Source
This commit is contained in:
jpekkila
2020-01-20 17:08:23 +02:00
parent 6d4f696e60
commit 765ce9a573
1 changed files with 32 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
src/core/device.cc
View File

@@ -690,19 +690,25 @@ acCreatePackedDataHost(const int3 dims)
}
static void
acTransferPackedDataToHost(const PackedData ddata, PackedData* hdata)
acTransferPackedDataToHost(const Device device, const cudaStream_t stream, const PackedData ddata,
PackedData* hdata)
{
cudaSetDevice(device->id);
const size_t bytes = ddata.dims.x * ddata.dims.y * ddata.dims.z * sizeof(ddata.data[0]) *
NUM_VTXBUF_HANDLES;
ERRCHK_CUDA_ALWAYS(cudaMemcpy(hdata->data, ddata.data, bytes, cudaMemcpyDeviceToHost));
ERRCHK_CUDA(cudaMemcpyAsync(hdata->data, ddata.data, bytes, cudaMemcpyDeviceToHost, stream));
}
static void
acTransferPackedDataToDevice(const PackedData hdata, PackedData* ddata)
acTransferPackedDataToDevice(const Device device, const cudaStream_t stream, const PackedData hdata,
PackedData* ddata)
{
cudaSetDevice(device->id);
const size_t bytes = hdata.dims.x * hdata.dims.y * hdata.dims.z * sizeof(hdata.data[0]) *
NUM_VTXBUF_HANDLES;
ERRCHK_CUDA_ALWAYS(cudaMemcpy(ddata->data, hdata.data, bytes, cudaMemcpyHostToDevice));
ERRCHK_CUDA(cudaMemcpyAsync(ddata->data, hdata.data, bytes, cudaMemcpyHostToDevice, stream));
}
static AcResult
@@ -930,7 +936,8 @@ static AcResult acDeviceCommunicateBlocksMPI(const Device device, //
// Host ////////////////////////////////////////////////
PackedData src_host = acCreatePackedDataHost(dims);
PackedData dst_host = acCreatePackedDataHost(dims);
acTransferPackedDataToHost(src, &src_host);
acTransferPackedDataToHost(device, device->streams[STREAM_DEFAULT], src,
&src_host);
acDeviceSynchronizeStream(device, STREAM_ALL);
MPI_Barrier(MPI_COMM_WORLD);
////////////////////////////////////////////////////////
@@ -948,7 +955,8 @@ static AcResult acDeviceCommunicateBlocksMPI(const Device device, //
MPI_Wait(&recv_req, MPI_STATUS_IGNORE);
// Host ////////////////////////////////////////////////
acTransferPackedDataToDevice(dst_host, &dst);
acTransferPackedDataToDevice(device, device->streams[STREAM_DEFAULT],
dst_host, &dst);
acDeviceSynchronizeStream(device, STREAM_ALL);
acDestroyPackedDataHost(&src_host);
acDestroyPackedDataHost(&dst_host);
@@ -1187,33 +1195,38 @@ static void
acPackCommData(const Device device, const int3* a0s, const size_t count, CommData* data)
{
cudaSetDevice(device->id);
cudaStream_t stream = device->streams[STREAM_DEFAULT];
cudaStream_t streams[count];
for (size_t i = 0; i < count; ++i)
cudaStreamCreate(&streams[i]);
for (size_t i = 0; i < count; ++i)
acKernelPackData(stream, device->vba, a0s[i], data->srcs[i]);
acKernelPackData(streams[i], device->vba, a0s[i], data->srcs[i]);
// HOST
acDeviceSynchronizeStream(device, STREAM_DEFAULT);
for (size_t i = 0; i < count; ++i)
acTransferPackedDataToHost(data->srcs[i], &data->srcs_host[i]);
//////
acTransferPackedDataToHost(device, streams[i], data->srcs[i], &data->srcs_host[i]);
for (size_t i = 0; i < count; ++i)
cudaStreamDestroy(streams[i]);
}
static void
acUnpackCommData(const Device device, const int3* b0s, const size_t count, CommData* data)
{
cudaSetDevice(device->id);
cudaStream_t stream = device->streams[STREAM_DEFAULT];
// HOST
cudaStream_t streams[count];
for (size_t i = 0; i < count; ++i)
acTransferPackedDataToDevice(data->dsts_host[i], &data->dsts[i]);
acDeviceSynchronizeStream(device, STREAM_ALL);
//////
cudaStreamCreate(&streams[i]);
for (size_t i = 0; i < count; ++i)
acKernelUnpackData(stream, data->dsts[i], b0s[i], device->vba);
acTransferPackedDataToDevice(device, streams[i], data->dsts_host[i], &data->dsts[i]);
for (size_t i = 0; i < count; ++i)
acKernelUnpackData(streams[i], data->dsts[i], b0s[i], device->vba);
for (size_t i = 0; i < count; ++i)
cudaStreamDestroy(streams[i]);
}
static AcResult