Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Remove manual memory management in LB code #4108

Merged
merged 17 commits into from
Jan 29, 2021
Merged

Remove manual memory management in LB code #4108

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
3b92f40
core: Remove unused Utils::Mpi::all_gatherv()
jngrad Jan 26, 2021
dc0c997
core: Remove unused Utils::realloc()
jngrad Jan 26, 2021
63f8c96
core: Use static memory allocation
jngrad Jan 26, 2021
436ce07
core: Remove a global variable in LB GPU code
jngrad Jan 26, 2021
619f938
core: Use dynamic memory allocation
jngrad Jan 27, 2021
59504ea
core: Fix memory leak
jngrad Jan 27, 2021
504b155
core: Simplify halo constructors
jngrad Jan 27, 2021
bf4ff5e
core: Remove Fieldtype pointer typedef for clarity
jngrad Jan 27, 2021
1f42f5a
core: Remove unnecessary macro
jngrad Jan 27, 2021
8850a09
core: Use dynamic memory allocation
jngrad Jan 27, 2021
9ebeb43
core: Cleanup comments
jngrad Jan 27, 2021
e4a4bc0
core: Rename struct
jngrad Jan 28, 2021
73b74f4
core: Use smart pointers
jngrad Jan 28, 2021
ab18bed
core: Change global variable type to bool
jngrad Jan 28, 2021
d036f62
lb: removed some C style code.
KaiSzuttor Jan 29, 2021
62d97d1
core: Fix regression and clang-tidy warnings
jngrad Jan 29, 2021
619725f
Merge branch 'python' into remove_manual_memory_lb
jngrad Jan 29, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
170 changes: 61 additions & 109 deletions src/core/grid_based_algorithms/halo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,91 +33,43 @@
#include "halo.hpp"

#include <utils/Vector.hpp>
#include <utils/memory.hpp>

#include <cstdlib>
#include <cstring>
#include <memory>

/** Primitive fieldtypes and their initializers */
struct _Fieldtype fieldtype_double = {0, nullptr, nullptr, sizeof(double), 0,
0, 0, false, nullptr};

void halo_create_field_vector(int vblocks, int vstride, int vskip,
Fieldtype oldtype, Fieldtype *const newtype) {

Fieldtype ntype = *newtype = (Fieldtype)Utils::malloc(sizeof(*ntype));

ntype->subtype = oldtype;
ntype->vflag = true;

ntype->vblocks = vblocks;
ntype->vstride = vstride;
ntype->vskip = vskip;

ntype->extent = oldtype->extent * ((vblocks - 1) * vskip + vstride);

int count = ntype->count = oldtype->count;
ntype->lengths = (int *)Utils::malloc(count * 2 * sizeof(int));
ntype->disps = (int *)((char *)ntype->lengths + count * sizeof(int));

for (int i = 0; i < count; i++) {
ntype->disps[i] = oldtype->disps[i];
ntype->lengths[i] = oldtype->lengths[i];
}
}

void halo_create_field_hvector(int vblocks, int vstride, int vskip,
Fieldtype oldtype, Fieldtype *const newtype) {

Fieldtype ntype = *newtype = (Fieldtype)Utils::malloc(sizeof(*ntype));

ntype->subtype = oldtype;
ntype->vflag = false;

ntype->vblocks = vblocks;
ntype->vstride = vstride;
ntype->vskip = vskip;

ntype->extent = oldtype->extent * vstride + (vblocks - 1) * vskip;

int const count = ntype->count = oldtype->count;
ntype->lengths = (int *)Utils::malloc(count * 2 * sizeof(int));
ntype->disps = (int *)((char *)ntype->lengths + count * sizeof(int));

for (int i = 0; i < count; i++) {
ntype->disps[i] = oldtype->disps[i];
ntype->lengths[i] = oldtype->lengths[i];
}
}
/** Predefined fieldtype for double-precision LB */
static std::shared_ptr<FieldType> fieldtype_double =
std::make_shared<FieldType>(static_cast<int>(sizeof(double)));

/** Set halo region to a given value
* @param[out] dest pointer to the halo buffer
* @param value integer value to write into the halo buffer
* @param type halo field layout description
*/
void halo_dtset(char *dest, int value, Fieldtype type) {
void halo_dtset(char *dest, int value, std::shared_ptr<FieldType> type) {
auto const vblocks = type->vblocks;
auto const vstride = type->vstride;
auto const vskip = type->vskip;
auto const count = type->count;
int const *const lens = type->lengths;
int const *const disps = type->disps;
auto const &lens = type->lengths;
auto const &disps = type->disps;
auto const extent = type->extent;
auto const block_size = static_cast<long>(vskip) * static_cast<long>(extent);

for (int i = 0; i < vblocks; i++) {
for (int j = 0; j < vstride; j++) {
for (int k = 0; k < count; k++)
for (std::size_t k = 0; k < disps.size(); k++)
memset(dest + disps[k], value, lens[k]);
}
dest += block_size;
}
}

void halo_dtcopy(char *r_buffer, char *s_buffer, int count, Fieldtype type);
void halo_dtcopy(char *r_buffer, char *s_buffer, int count,
std::shared_ptr<FieldType> type);

void halo_copy_vector(char *r_buffer, char *s_buffer, int count, Fieldtype type,
bool vflag) {
void halo_copy_vector(char *r_buffer, char *s_buffer, int count,
std::shared_ptr<FieldType> type, bool vflag) {

auto const vblocks = type->vblocks;
auto const vstride = type->vstride;
Expand All @@ -136,13 +88,14 @@ void halo_copy_vector(char *r_buffer, char *s_buffer, int count, Fieldtype type,
}
}

/** Copy lattice data with layout described by fieldtype.
/** Copy lattice data with layout described by @p type.
* @param r_buffer data destination
* @param s_buffer data source
* @param count amount of data to copy
* @param type field layout type
*/
void halo_dtcopy(char *r_buffer, char *s_buffer, int count, Fieldtype type) {
void halo_dtcopy(char *r_buffer, char *s_buffer, int count,
std::shared_ptr<FieldType> type) {

if (type->subtype) {
halo_copy_vector(r_buffer, s_buffer, count, type, type->vflag);
Expand All @@ -162,31 +115,30 @@ void halo_dtcopy(char *r_buffer, char *s_buffer, int count, Fieldtype type) {
}
}

void prepare_halo_communication(HaloCommunicator *const hc,
Lattice const *const lattice,
Fieldtype fieldtype, MPI_Datatype datatype,
void prepare_halo_communication(HaloCommunicator &hc, const Lattice &lattice,
MPI_Datatype datatype,
const Utils::Vector3i &local_node_grid) {

const auto grid = lattice->grid;
const auto period = lattice->halo_grid;
const auto &grid = lattice.grid;
const auto &period = lattice.halo_grid;

for (int n = 0; n < hc->num; n++) {
MPI_Type_free(&(hc->halo_info[n].datatype));
for (int n = 0; n < hc.num; n++) {
MPI_Type_free(&(hc.halo_info[n].datatype));
}

int const num = 2 * 3; /* two communications in each space direction */
hc->num = num;
hc->halo_info.resize(num);
hc.num = num;
hc.halo_info.resize(num);

auto const extent = static_cast<long>(fieldtype->extent);
auto const extent = static_cast<long>(fieldtype_double->extent);

auto const node_neighbors = calc_node_neighbors(comm_cart);

int cnt = 0;
for (int dir = 0; dir < 3; dir++) {
for (int lr = 0; lr < 2; lr++) {

HaloInfo *hinfo = &(hc->halo_info[cnt]);
HaloInfo &hinfo = hc.halo_info[cnt];

int nblocks = 1;
for (int k = dir + 1; k < 3; k++) {
Expand All @@ -203,107 +155,107 @@ void prepare_halo_communication(HaloCommunicator *const hc,

if (lr == 0) {
/* send to left, recv from right */
hinfo->s_offset = extent * static_cast<long>(stride * 1);
hinfo->r_offset = extent * static_cast<long>(stride * (grid[dir] + 1));
hinfo.s_offset = extent * static_cast<long>(stride * 1);
hinfo.r_offset = extent * static_cast<long>(stride * (grid[dir] + 1));
} else {
/* send to right, recv from left */
hinfo->s_offset = extent * static_cast<long>(stride * grid[dir]);
hinfo->r_offset = extent * static_cast<long>(stride * 0);
hinfo.s_offset = extent * static_cast<long>(stride * grid[dir]);
hinfo.r_offset = extent * static_cast<long>(stride * 0);
}

hinfo->source_node = node_neighbors[2 * dir + 1 - lr];
hinfo->dest_node = node_neighbors[2 * dir + lr];
hinfo.source_node = node_neighbors[2 * dir + 1 - lr];
hinfo.dest_node = node_neighbors[2 * dir + lr];

halo_create_field_vector(nblocks, stride, skip, fieldtype,
&hinfo->fieldtype);
hinfo.fieldtype = std::make_shared<FieldType>(nblocks, stride, skip, true,
fieldtype_double);

MPI_Type_vector(nblocks, stride, skip, datatype, &hinfo->datatype);
MPI_Type_commit(&hinfo->datatype);
MPI_Type_vector(nblocks, stride, skip, datatype, &hinfo.datatype);
MPI_Type_commit(&hinfo.datatype);

if (!box_geo.periodic(dir) &&
(local_geo.boundary()[2 * dir + lr] != 0 ||
local_geo.boundary()[2 * dir + 1 - lr] != 0)) {
if (local_node_grid[dir] == 1) {
hinfo->type = HALO_OPEN;
hinfo.type = HALO_OPEN;
} else if (lr == 0) {
if (local_geo.boundary()[2 * dir + lr] == 1) {
hinfo->type = HALO_RECV;
hinfo.type = HALO_RECV;
} else {
hinfo->type = HALO_SEND;
hinfo.type = HALO_SEND;
}
} else {
if (local_geo.boundary()[2 * dir + lr] == -1) {
hinfo->type = HALO_RECV;
hinfo.type = HALO_RECV;
} else {
hinfo->type = HALO_SEND;
hinfo.type = HALO_SEND;
}
}
} else {
if (local_node_grid[dir] == 1) {
hc->halo_info[cnt].type = HALO_LOCL;
hc.halo_info[cnt].type = HALO_LOCL;
} else {
hc->halo_info[cnt].type = HALO_SENDRECV;
hc.halo_info[cnt].type = HALO_SENDRECV;
}
}
cnt++;
}
}
}

void release_halo_communication(HaloCommunicator *const hc) {
for (int n = 0; n < hc->num; n++) {
MPI_Type_free(&(hc->halo_info[n].datatype));
void release_halo_communication(HaloCommunicator &hc) {
for (int n = 0; n < hc.num; n++) {
MPI_Type_free(&(hc.halo_info[n].datatype));
}
}

void halo_communication(HaloCommunicator const *const hc, char *const base) {
void halo_communication(const HaloCommunicator &hc, char *const base) {

Fieldtype fieldtype;
std::shared_ptr<FieldType> fieldtype;
MPI_Datatype datatype;
MPI_Request request;
MPI_Status status;

for (int n = 0; n < hc->num; n++) {
for (int n = 0; n < hc.num; n++) {
int s_node, r_node;
int comm_type = hc->halo_info[n].type;
char *s_buffer = (char *)base + hc->halo_info[n].s_offset;
char *r_buffer = (char *)base + hc->halo_info[n].r_offset;
int comm_type = hc.halo_info[n].type;
char *s_buffer = (char *)base + hc.halo_info[n].s_offset;
char *r_buffer = (char *)base + hc.halo_info[n].r_offset;

switch (comm_type) {

case HALO_LOCL:
fieldtype = hc->halo_info[n].fieldtype;
fieldtype = hc.halo_info[n].fieldtype;
halo_dtcopy(r_buffer, s_buffer, 1, fieldtype);
break;

case HALO_SENDRECV:
datatype = hc->halo_info[n].datatype;
s_node = hc->halo_info[n].source_node;
r_node = hc->halo_info[n].dest_node;
datatype = hc.halo_info[n].datatype;
s_node = hc.halo_info[n].source_node;
r_node = hc.halo_info[n].dest_node;
MPI_Sendrecv(s_buffer, 1, datatype, r_node, REQ_HALO_SPREAD, r_buffer, 1,
datatype, s_node, REQ_HALO_SPREAD, comm_cart, &status);
break;

case HALO_SEND:
datatype = hc->halo_info[n].datatype;
fieldtype = hc->halo_info[n].fieldtype;
r_node = hc->halo_info[n].dest_node;
datatype = hc.halo_info[n].datatype;
fieldtype = hc.halo_info[n].fieldtype;
r_node = hc.halo_info[n].dest_node;
MPI_Isend(s_buffer, 1, datatype, r_node, REQ_HALO_SPREAD, comm_cart,
&request);
halo_dtset(r_buffer, 0, fieldtype);
MPI_Wait(&request, &status);
break;

case HALO_RECV:
datatype = hc->halo_info[n].datatype;
s_node = hc->halo_info[n].source_node;
datatype = hc.halo_info[n].datatype;
s_node = hc.halo_info[n].source_node;
MPI_Irecv(r_buffer, 1, datatype, s_node, REQ_HALO_SPREAD, comm_cart,
&request);
MPI_Wait(&request, &status);
break;

case HALO_OPEN:
fieldtype = hc->halo_info[n].fieldtype;
fieldtype = hc.halo_info[n].fieldtype;
/** \todo this does not work for the n_i - \<n_i\> */
halo_dtset(r_buffer, 0, fieldtype);
break;
Expand Down
Loading