bda
/
gpu-mpi-tests
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Compare commits

base: bda:b2ed53adc1421359bee0d9f5304f1a6ad93ce54b
Branches Tags
bda:main
...
compare: bda:8e43edc21e3e49a256ff1665a5cf52e528d377b6
Branches Tags
bda:main

3 Commits
b2ed53adc1 ... 8e43edc21e

Author SHA1 Message Date
Bryce Allen 8e43edc21e use milliseconds for timing
3 years ago
Bryce Allen 44d72ad2bb switch exchange/stencil dim on sycl example
3 years ago
Bryce Allen 1d779fd510 update comment
3 years ago
2 changed files with 109 additions and 93 deletions
Show Stats

7
mpi_stencil2d_gt.cc
Unescape View File

@ -112,8 +112,7 @@ void set_rank_device(int n_ranks, int rank)
gt::backend::clib::device_set(device);
}
// exchange in non-contiguous second dimension, staging into contiguous buffers
// on device
// exchange in first dimension, staging into contiguous buffers on device
void boundary_exchange_x(MPI_Comm comm, int world_size, int rank,
gt::gtensor_device<double, 2>& d_z, int n_bnd,
bool stage_host = false)
@ -362,8 +361,8 @@ int main(int argc, char** argv)
d_dzdx_numeric = stencil2d_1d_5(d_z, stencil5) * scale;
gt::synchronize();
}
printf("%d/%d exchange time %0.8f\n", world_rank, world_size,
total_time / n_iter);
printf("%d/%d exchange time %0.8f ms\n", world_rank, world_size,
total_time / n_iter * 1000);
gt::copy(d_dzdx_numeric, h_dzdx_numeric);

195
mpi_stencil2d_sycl.cc
Unescape View File

@ -14,6 +14,7 @@
* by default.
*/
#include <cassert>
#include <cmath>
#include <mpi.h>
#include <stdio.h>
@ -22,6 +23,16 @@
#include "sycl/sycl.hpp"
inline void check(const char* fname, int line, int mpi_rval)
{
if (mpi_rval != MPI_SUCCESS) {
printf("%s:%d error %d\n", fname, line, mpi_rval);
exit(2);
}
}
#define CHECK(x) check(__FILE__, __LINE__, (x))
static constexpr double stencil5[] = {1.0 / 12.0, -2.0 / 3.0, 0.0, 2.0 / 3.0,
-1.0 / 12.0};
@ -38,65 +49,66 @@ constexpr int idx2(int n, int row, int col)
*
* Returns sycl event, async with respect to host.
*/
auto stencil2d_1d_5(sycl::queue& q, int nrows, int ncols, double* out2d,
auto stencil2d_1d_5(sycl::queue& q, int out_nrows, int out_ncols, double* out2d,
const double* in2d, double scale)
{
// Note: swap index order; SYCL is row-major oriented, and this example
// is col-major
auto range = sycl::range<2>(ncols, nrows);
int in_nrows = out_nrows + 4;
auto range = sycl::range<2>(out_ncols, out_nrows);
auto e = q.submit([&](sycl::handler& cgh) {
cgh.parallel_for(range, [=](sycl::item<2> item) {
int row = item.get_id(1);
int col = item.get_id(0);
int base_idx = idx2(nrows, row, col);
int stride = nrows;
out2d[base_idx] = (stencil5[0] * in2d[base_idx + 0 * stride] +
stencil5[1] * in2d[base_idx + 1 * stride] +
stencil5[2] * in2d[base_idx + 2 * stride] +
stencil5[3] * in2d[base_idx + 3 * stride] +
stencil5[4] * in2d[base_idx + 4 * stride]) *
scale;
int out_idx = idx2(out_nrows, row, col);
int in_base_idx = idx2(in_nrows, row, col);
out2d[out_idx] = (stencil5[0] * in2d[in_base_idx + 0] +
stencil5[1] * in2d[in_base_idx + 1] +
stencil5[2] * in2d[in_base_idx + 2] +
stencil5[3] * in2d[in_base_idx + 3] +
stencil5[4] * in2d[in_base_idx + 4]) *
scale;
});
});
return e;
}
/*
* Copy slice of second (non-contiguous) dimension of in array into contiguous
* buffer out. In has dimension nrows x ncols, buf has dimension nrows x (end
* -start + 1).
* Copy slice of first dimension of in array into contiguous
* buffer out. In has dimension nrows x ncols, buf has dimension (end - start) x
* ncols.
*/
auto buf_from_view(sycl::queue& q, int nrows, double* buf, double* in,
int start, int end)
auto buf_from_view(sycl::queue& q, int ncols, int buf_nrows, double* buf,
int in_nrows, double* in, int start, int end)
{
assert(buf_nrows >= end - start);
// Note: reverse index order because SYCL is row-major
auto range = sycl::range<2>(end - start, nrows);
auto range = sycl::range<2>(ncols, end - start);
auto e = q.submit([&](sycl::handler& cgh) {
cgh.parallel_for(range, [=](sycl::item<2> item) {
int row = item.get_id(1);
int col = item.get_id(0);
buf[idx2(nrows, row, col)] = in[idx2(nrows, row, start + col)];
buf[idx2(buf_nrows, row, col)] = in[idx2(in_nrows, start + row, col)];
});
});
return e;
}
/*
* Copy contiguous buffer into second (non-contiguous) dimension of array as a
* slice. Out has dimension nrows x ncols, buf has dimension nrows x (end -
* start + 1). ncols is not actually needed or used, just need start to end
* to be valid indexes.
* Copy contiguous buffer into first dimension of array as a slice. Out has
* dimension nrows x ncols, buf has dimension (end - start) * ncols.
*/
auto buf_to_view(sycl::queue& q, int nrows, double* out, double* buf, int start,
int end)
auto buf_to_view(sycl::queue& q, int ncols, int out_nrows, double* out,
int buf_nrows, double* buf, int start, int end)
{
assert(buf_nrows >= end - start);
// Note: reverse index order because SYCL is row-major
auto range = sycl::range<2>(end - start, nrows);
auto range = sycl::range<2>(ncols, end - start);
auto e = q.submit([&](sycl::handler& cgh) {
cgh.parallel_for(range, [=](sycl::item<2> item) {
int row = item.get_id(1);
int col = item.get_id(0);
out[idx2(nrows, row, start + col)] = buf[idx2(nrows, row, col)];
out[idx2(out_nrows, start + row, col)] = buf[idx2(buf_nrows, row, col)];
});
});
return e;
@ -104,41 +116,43 @@ auto buf_to_view(sycl::queue& q, int nrows, double* out, double* buf, int start,
void test_buf_view(sycl::queue& q, const int n)
{
double* data = sycl::malloc_host<double>(n * n, q);
double* buf = sycl::malloc_host<double>(n * 2, q);
double* buf2 = sycl::malloc_host<double>(n * 2, q);
for (int i = 0; i < n; i++) {
for (int j = 0; j < n; j++) {
data[j + i * n] = i + j / 1000.0;
int n_bnd = 2;
int n_with_ghost = n + 2 * n_bnd;
double* data = sycl::malloc_host<double>(n_with_ghost * n, q);
double* buf = sycl::malloc_host<double>(n_bnd * n, q);
double* buf2 = sycl::malloc_host<double>(n_bnd * n, q);
for (int j = 0; j < n; j++) {
for (int i = 0; i < n_with_ghost; i++) {
data[idx2(n_with_ghost, i, j)] = (i - n_bnd) + j / 1000.0;
}
buf2[i] = 100.0 + i;
buf2[i + n] = 100.0 + i + 0.1;
buf2[idx2(n_bnd, 0, j)] = 100.0 + j;
buf2[idx2(n_bnd, 1, j)] = 100.0 + j + 0.1;
}
for (int i = 0; i < n; i++) {
for (int j = 0; j < n; j++) {
printf("data[%d, %d] = %f\n", j, i, data[j + i * n]);
for (int j = 0; j < n; j++) {
for (int i = 0; i < n; i++) {
printf("data[%d, %d] = %f\n", i, j, data[idx2(n_with_ghost, i, j)]);
}
}
for (int i = 0; i < 2; i++) {
for (int j = 0; j < n; j++) {
printf("buf2[%d, %d] = %f\n", j, i, buf2[j + i * n]);
for (int j = 0; j < n; j++) {
for (int i = 0; i < n_bnd; i++) {
printf("buf2[%d, %d] = %f\n", i, j, buf2[idx2(n_bnd, i, j)]);
}
}
buf_from_view(q, n, buf, data, 0, 2).wait();
for (int i = 0; i < 2; i++) {
for (int j = 0; j < n; j++) {
printf("buf[%d, %d] = %f\n", j, i, buf[j + i * n]);
buf_from_view(q, n, n_bnd, buf, n_with_ghost, data, 0, n_bnd).wait();
for (int j = 0; j < n; j++) {
for (int i = 0; i < n_bnd; i++) {
printf("buf[%d, %d] = %f\n", i, j, buf[idx2(n_bnd, i, j)]);
}
}
buf_to_view(q, n, data, buf2, n - 2, n).wait();
buf_to_view(q, n, n_with_ghost, data, n_bnd, buf2, n - n_bnd, n).wait();
for (int i = 0; i < n; i++) {
for (int j = 0; j < n; j++) {
printf("data[%d, %d] = %f\n", j, i, data[j + i * n]);
for (int j = 0; j < n; j++) {
for (int i = 0; i < n; i++) {
printf("data[%d, %d] = %f\n", i, j, data[idx2(n_with_ghost, i, j)]);
}
}
}
@ -194,13 +208,13 @@ sycl::queue get_rank_queue(int n_ranks, int rank)
cl::sycl::property::queue::in_order()};
}
// exchange in non-contiguous second dimension, staging into contiguous buffers
// on device
void boundary_exchange_y(MPI_Comm comm, int world_size, int rank,
// exchange in first dimension, staging into contiguous buffers on device
void boundary_exchange_x(MPI_Comm comm, int world_size, int rank,
sycl::queue& q, int n_global, int n_local, int n_bnd,
double* d_z, bool stage_host = false)
{
int buf_size = n_global * n_bnd;
int n_local_with_ghost = n_local + 2 * n_bnd;
int buf_size = n_bnd * n_global;
static double* sbuf_l = nullptr;
static double* sbuf_r = nullptr;
static double* rbuf_l = nullptr;
@ -234,7 +248,8 @@ void boundary_exchange_y(MPI_Comm comm, int world_size, int rank,
if (rank_l >= 0) {
// printf("rank_l = %d\n", rank_l); fflush(nullptr);
// sbuf_l = d_z.view(_all, _s(n_bnd, 2 * n_bnd));
auto e = buf_from_view(q, n_global, sbuf_l, d_z, n_bnd, 2 * n_bnd);
auto e = buf_from_view(q, n_global, n_bnd, sbuf_l, n_local_with_ghost, d_z,
n_bnd, 2 * n_bnd);
if (stage_host) {
q.copy(sbuf_l, h_sbuf_l, buf_size, e);
/*
@ -251,7 +266,8 @@ void boundary_exchange_y(MPI_Comm comm, int world_size, int rank,
if (rank_r < world_size) {
// printf("rank_r = %d\n", rank_r); fflush(nullptr);
// sbuf_r = d_z.view(_all, _s(-2 * n_bnd, -n_bnd));
auto e = buf_from_view(q, n_global, sbuf_r, d_z, n_local, n_local + n_bnd);
auto e = buf_from_view(q, n_global, n_bnd, sbuf_r, n_local_with_ghost, d_z,
n_local, n_local + n_bnd);
if (stage_host) {
q.copy(sbuf_r, h_sbuf_r, buf_size, e);
/*
@ -331,7 +347,7 @@ void boundary_exchange_y(MPI_Comm comm, int world_size, int rank,
q.copy(h_rbuf_l, rbuf_l, buf_size);
}
// d_z.view(_all, _s(0, n_bnd)) = rbuf_l;
buf_to_view(q, n_global, d_z, rbuf_l, 0, n_bnd);
buf_to_view(q, n_global, n_local_with_ghost, d_z, n_bnd, rbuf_l, 0, n_bnd);
}
if (rank_r < world_size) {
mpi_rval = MPI_Waitall(2, req_r, MPI_STATUSES_IGNORE);
@ -351,7 +367,8 @@ void boundary_exchange_y(MPI_Comm comm, int world_size, int rank,
q.copy(h_rbuf_r, rbuf_r, buf_size);
}
// d_z.view(_all, _s(-n_bnd, _)) = rbuf_r;
buf_to_view(q, n_global, d_z, rbuf_r, n_local + n_bnd, n_local + 2 * n_bnd);
buf_to_view(q, n_global, n_local_with_ghost, d_z, n_bnd, rbuf_r,
n_local + n_bnd, n_local + 2 * n_bnd);
}
q.wait();
@ -412,53 +429,53 @@ int main(int argc, char** argv)
printf("stage_host = %d\n", stage_host);
}
int z_size = n_global * n_local_with_ghost;
int dzdy_size = n_global * n_local;
int z_size = n_local_with_ghost * n_global;
int dzdx_size = n_local * n_global;
double* h_z = sycl::malloc_host<double>(z_size, q);
double* d_z = sycl::malloc_device<double>(z_size, q);
double* h_dzdy_numeric = sycl::malloc_host<double>(dzdy_size, q);
double* h_dzdy_actual = sycl::malloc_host<double>(dzdy_size, q);
double* d_dzdy_numeric = sycl::malloc_device<double>(dzdy_size, q);
double* h_dzdx_numeric = sycl::malloc_host<double>(dzdx_size, q);
double* h_dzdx_actual = sycl::malloc_host<double>(dzdx_size, q);
double* d_dzdx_numeric = sycl::malloc_device<double>(dzdx_size, q);
double lx = 8;
double dx = lx / n_global;
double lx_local = lx / world_size;
double scale = n_global / lx;
auto fn = [](double x, double y) { return x * x + y * y; };
auto fn_dzdy = [](double x, double y) { return 2 * x; };
auto fn = [](double x, double y) { return x * x * x + y * y; };
auto fn_dzdx = [](double x, double y) { return 3 * x * x; };
struct timespec start, end;
double iter_time = 0.0;
double total_time = 0.0;
double x_start = world_rank * lx_local;
for (int i = 0; i < n_local; i++) {
double xtmp = x_start + i * dx;
for (int j = 0; j < n_global; j++) {
double ytmp = j * dx;
h_z[idx2(n_global, j, i + n_bnd)] = fn(xtmp, ytmp);
h_dzdy_actual[idx2(n_global, j, i)] = fn_dzdy(xtmp, ytmp);
for (int j = 0; j < n_global; j++) {
double ytmp = j * dx;
for (int i = 0; i < n_local; i++) {
double xtmp = x_start + i * dx;
h_z[idx2(n_local_with_ghost, i + n_bnd, j)] = fn(xtmp, ytmp);
h_dzdx_actual[idx2(n_local, i, j)] = fn_dzdx(xtmp, ytmp);
}
}
// fill boundary points on ends
if (world_rank == 0) {
for (int i = 0; i < n_bnd; i++) {
double xtmp = (i - n_bnd) * dx;
for (int j = 0; j < n_global; j++) {
double ytmp = j * dx;
h_z[idx2(n_global, j, i)] = fn(xtmp, ytmp);
for (int j = 0; j < n_global; j++) {
double ytmp = j * dx;
for (int i = 0; i < n_bnd; i++) {
double xtmp = (i - n_bnd) * dx;
h_z[idx2(n_local_with_ghost, i, j)] = fn(xtmp, ytmp);
}
}
}
if (world_rank == world_size - 1) {
for (int i = 0; i < n_bnd; i++) {
double xtmp = lx + i * dx;
for (int j = 0; j < n_global; j++) {
double ytmp = j * dx;
h_z[idx2(n_global, j, n_bnd + n_local + i)] = fn(xtmp, ytmp);
for (int j = 0; j < n_global; j++) {
double ytmp = j * dx;
for (int i = 0; i < n_bnd; i++) {
double xtmp = lx + i * dx;
h_z[idx2(n_local_with_ghost, n_bnd + n_local + i, j)] = fn(xtmp, ytmp);
}
}
}
@ -478,7 +495,7 @@ int main(int argc, char** argv)
for (int i = 0; i < n_warmup + n_iter; i++) {
clock_gettime(CLOCK_MONOTONIC, &start);
boundary_exchange_y(MPI_COMM_WORLD, world_size, world_rank, q, n_global,
boundary_exchange_x(MPI_COMM_WORLD, world_size, world_rank, q, n_global,
n_local, n_bnd, d_z, stage_host);
clock_gettime(CLOCK_MONOTONIC, &end);
iter_time =
@ -489,28 +506,28 @@ int main(int argc, char** argv)
}
// do some calculation, to try to more closely simulate what happens in GENE
auto e = stencil2d_1d_5(q, n_global, n_local, d_dzdy_numeric, d_z, scale);
auto e = stencil2d_1d_5(q, n_local, n_global, d_dzdx_numeric, d_z, scale);
e.wait();
}
printf("%d/%d exchange time %0.8f\n", world_rank, world_size,
total_time / n_iter);
printf("%d/%d exchange time %0.8f ms\n", world_rank, world_size,
total_time / n_iter * 1000);
q.copy(d_dzdy_numeric, h_dzdy_numeric, dzdy_size).wait();
q.copy(d_dzdx_numeric, h_dzdx_numeric, dzdx_size).wait();
/*
for (int i = 0; i < 5; i++) {
int idx = idx2(n_global, 8, i);
printf("%d la %f\n%d ln %f\n", world_rank, h_dzdy_actual[idx], world_rank,
h_dzdy_numeric[idx]);
printf("%d la %f\n%d ln %f\n", world_rank, h_dzdx_actual[idx], world_rank,
h_dzdx_numeric[idx]);
}
for (int i = 0; i < 5; i++) {
int idx = idx2(n_global, 8, n_local - 1 - i);
printf("%d ra %f\n%d rn %f\n", world_rank, h_dzdy_actual[idx], world_rank,
h_dzdy_numeric[idx]);
printf("%d ra %f\n%d rn %f\n", world_rank, h_dzdx_actual[idx], world_rank,
h_dzdx_numeric[idx]);
}
*/
double err_norm = diff_norm(q, dzdy_size, h_dzdy_numeric, h_dzdy_actual);
double err_norm = diff_norm(q, dzdx_size, h_dzdx_numeric, h_dzdx_actual);
printf("%d/%d [%d:0x%08x] err_norm = %.8f\n", world_rank, world_size,
device_id, vendor_id, err_norm);

Write Preview
Loading…
Cancel
Save