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

Added the optimized implementation of acNodeIntegrate where boundconds are done before integration instead of after

Browse Source
This commit is contained in:
jpekkila
2019-08-05 20:10:13 +03:00
parent 8df49370c8
commit b73c2675e8
3 changed files with 128 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
include/astaroth_defines.h
View File

@@ -104,7 +104,27 @@ typedef enum {
NUM_REDUCTION_TYPES
} ReductionType;
typedef enum { STREAM_DEFAULT, NUM_STREAM_TYPES } Stream;
typedef enum {
STREAM_DEFAULT,
STREAM_0,
STREAM_1,
STREAM_2,
STREAM_3,
STREAM_4,
STREAM_5,
STREAM_6,
STREAM_7,
STREAM_8,
STREAM_9,
STREAM_10,
STREAM_11,
STREAM_12,
STREAM_13,
STREAM_14,
STREAM_15,
STREAM_16,
NUM_STREAM_TYPES
} Stream;
#define STREAM_ALL (NUM_STREAM_TYPES)
#define AC_GEN_ID(X) X

3
src/core/astaroth.cu
View File

@@ -73,7 +73,8 @@ acStore(AcMesh* host_mesh)
AcResult
acIntegrate(const AcReal dt)
{
return acNodeIntegrate(nodes[0], dt);
acNodeIntegrate(nodes[0], dt);
return acBoundcondStep();
}
AcResult

124
src/core/node.cu
View File

@@ -478,25 +478,6 @@ acNodeIntegrateSubstep(const Node node, const Stream stream, const int isubstep,
return AC_SUCCESS;
}
AcResult
acNodeIntegrate(const Node node, const AcReal dt)
{
acNodeSynchronizeStream(node, STREAM_ALL);
for (int isubstep = 0; isubstep < 3; ++isubstep) {
acNodePeriodicBoundconds(node, STREAM_DEFAULT);
acNodeSynchronizeStream(node, STREAM_DEFAULT); // DEBUG
const int3 start = (int3){NGHOST, NGHOST, NGHOST};
const int3 end = start + node->grid.n;
acNodeIntegrateSubstep(node, STREAM_DEFAULT, isubstep, start, end, dt);
acNodeSwapBuffers(node);
acNodeSynchronizeStream(node, STREAM_DEFAULT); // DEBUG
}
acNodePeriodicBoundconds(node, STREAM_DEFAULT); // DEBUG
acNodeSynchronizeStream(node, STREAM_ALL);
return AC_SUCCESS;
}
static AcResult
local_boundcondstep(const Node node, const Stream stream, const VertexBufferHandle vtxbuf)
{
@@ -550,6 +531,111 @@ global_boundcondstep(const Node node, const Stream stream, const VertexBufferHan
return AC_SUCCESS;
}
AcResult
acNodeIntegrate(const Node node, const AcReal dt)
{
acNodeSynchronizeStream(node, STREAM_ALL);
// xxx|OOO OOOOOOOOO OOO|xxx
// ^ ^ ^ ^
// n0 n1 n2 n3
const int3 n0 = (int3){NGHOST, NGHOST, NGHOST};
const int3 n1 = (int3){2 * NGHOST, 2 * NGHOST, 2 * NGHOST};
const int3 n2 = node->grid.n;
const int3 n3 = n0 + node->grid.n;
for (int isubstep = 0; isubstep < 3; ++isubstep) {
acNodeSynchronizeStream(node, STREAM_ALL);
for (int vtxbuf = 0; vtxbuf < NUM_VTXBUF_HANDLES; ++vtxbuf) {
local_boundcondstep(node, (Stream)vtxbuf, (VertexBufferHandle)vtxbuf);
}
acNodeSynchronizeStream(node, STREAM_ALL);
// Inner inner
for (int i = 0; i < node->num_devices; ++i) {
const int3 m1 = n1 + (int3){0, 0, i * node->subgrid.n.z};
const int3 m2 = m1 + node->subgrid.n - (int3){2 * NGHOST, 2 * NGHOST, 2 * NGHOST};
acNodeIntegrateSubstep(node, STREAM_16, isubstep, m1, m2, dt);
}
for (int vtxbuf = 0; vtxbuf < NUM_VTXBUF_HANDLES; ++vtxbuf) {
const int num_vertices = node->subgrid.m.x * node->subgrid.m.y * NGHOST;
for (int device_id = 0; device_id < node->num_devices; ++device_id) {
// ...|ooooxxx|... -> xxx|ooooooo|...
{
const int3 src = (int3){0, 0, node->subgrid.n.z};
const int3 dst = (int3){0, 0, 0};
acDeviceTransferVertexBufferWithOffset(
node->devices[device_id], (Stream)vtxbuf, (VertexBufferHandle)vtxbuf, src,
dst, num_vertices, node->devices[(device_id + 1) % node->num_devices]);
}
// ...|ooooooo|xxx <- ...|xxxoooo|...
{
const int3 src = (int3){0, 0, NGHOST};
const int3 dst = (int3){0, 0, NGHOST + node->subgrid.n.z};
acDeviceTransferVertexBufferWithOffset(
node->devices[device_id], (Stream)vtxbuf, (VertexBufferHandle)vtxbuf, src,
dst, num_vertices,
node->devices[(device_id - 1 + node->num_devices) % node->num_devices]);
}
}
}
for (int vtxbuf = 0; vtxbuf < 2 * NUM_VTXBUF_HANDLES; ++vtxbuf) {
acNodeSynchronizeStream(node, (Stream)vtxbuf);
}
// Inner outer
for (int i = 0; i < node->num_devices - 1; ++i) {
const int3 m1 = n1 + (int3){0, 0, (i + 1) * node->subgrid.n.z - 2 * NGHOST};
const int3 m2 = m1 + (int3){node->subgrid.n.x - 2 * NGHOST,
node->subgrid.n.y - 2 * NGHOST, 2 * NGHOST};
acNodeIntegrateSubstep(node, STREAM_0, isubstep, m1, m2, dt);
}
// Outer
// Front
{
const int3 m1 = (int3){n0.x, n0.y, n0.z};
const int3 m2 = (int3){n3.x, n3.y, n1.z};
acNodeIntegrateSubstep(node, STREAM_1, isubstep, m1, m2, dt);
}
// Back
{
const int3 m1 = (int3){n0.x, n0.y, n2.z};
const int3 m2 = (int3){n3.x, n3.y, n3.z};
acNodeIntegrateSubstep(node, STREAM_2, isubstep, m1, m2, dt);
}
// Top
{
const int3 m1 = (int3){n0.x, n0.y, n1.z};
const int3 m2 = (int3){n3.x, n1.y, n2.z};
acNodeIntegrateSubstep(node, STREAM_3, isubstep, m1, m2, dt);
}
// Bottom
{
const int3 m1 = (int3){n0.x, n2.y, n1.z};
const int3 m2 = (int3){n3.x, n3.y, n2.z};
acNodeIntegrateSubstep(node, STREAM_4, isubstep, m1, m2, dt);
}
// Left
{
const int3 m1 = (int3){n0.x, n1.y, n1.z};
const int3 m2 = (int3){n1.x, n2.y, n2.z};
acNodeIntegrateSubstep(node, STREAM_5, isubstep, m1, m2, dt);
}
// Right
{
const int3 m1 = (int3){n2.x, n1.y, n1.z};
const int3 m2 = (int3){n3.x, n2.y, n2.z};
acNodeIntegrateSubstep(node, STREAM_6, isubstep, m1, m2, dt);
}
acNodeSwapBuffers(node);
}
acNodeSynchronizeStream(node, STREAM_ALL);
return AC_SUCCESS;
}
AcResult
acNodePeriodicBoundcondStep(const Node node, const Stream stream,
const VertexBufferHandle vtxbuf_handle)