Files
Aurelien Bouteiller 8837414042 Cleanup/wg init (#260)
* remove wg_init and wg_finalize from functional tests

* Remove wg_init and wg_finalize from examples

* deprecate wg_init/finalize

* Updated docs

* Typo in documentation

---------

Co-authored-by: Yiltan <yiltan@amd.com>

[ROCm/rocshmem commit: 6e7277b544]
2025-10-07 14:34:18 -04:00

156 γραμμές
5.0 KiB
C++

/******************************************************************************
* Copyright (c) Advanced Micro Devices, Inc. All rights reserved.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*****************************************************************************/
/*
* First find your offload target, and if xnack is enabled/disabled using
rocminfo | grep amdgcn
* It should output a string like so:
"Name: amdgcn-amd-amdhsa--gfx942:sramecc+:xnack-"
* This lists the offload taret (gfx942) and that xnack is disabled (xnack-).
* Therefore, we need to specify --offload-arch=gfx942:xnack- to our link and compile commands.
* Please modify the compile and link commands to suit your system
* To compile:
hipcc -c -fgpu-rdc -x hip rocshmem_allreduce_test.cc \
--offload-arch=<target>:<xnack> \
-I/opt/rocm/include \
-I$ROCSHMEM_INSTALL_DIR/include \
-I$OPENMPI_UCX_INSTALL_DIR/include/
* To link:
hipcc -fgpu-rdc --hip-link rocshmem_allreduce_test.o -o rocshmem_allreduce_test \
--offload-arch=<target>:<xnack> \
$ROCSHMEM_INSTALL_DIR/lib/librocshmem.a \
$OPENMPI_UCX_INSTALL_DIR/lib/libmpi.so \
-L/opt/rocm/lib -lamdhip64 -lhsa-runtime64
* To run:
mpirun -np 8 -x ROCSHMEM_MAX_NUM_CONTEXTS=2 ./rocshmem_allreduce_test
*/
#include <rocshmem/rocshmem.hpp>
#include "util.h"
using namespace rocshmem;
__global__ void allreduce_test(int *source, int *dest, size_t nelem,
rocshmem_team_t team) {
__shared__ rocshmem_ctx_t ctx;
int64_t ctx_type = 0;
rocshmem_wg_ctx_create(ctx_type, &ctx);
int num_pes = rocshmem_ctx_n_pes(ctx);
rocshmem_ctx_int_sum_reduce_wg(ctx, team, dest, source, nelem);
rocshmem_ctx_quiet(ctx);
__syncthreads();
rocshmem_wg_ctx_destroy(&ctx);
}
static void init_sendbuf (int *source, int nelem, int my_pe)
{
for (int i = 0; i < nelem; i++) {
source[i] = my_pe + i%9;
}
}
static bool check_recvbuf(int *dest, int nelem, int my_pe, int npes)
{
bool res=true;
int expected = npes * (npes -1) / 2;
for (int i = 0; i < nelem; i++) {
int result = expected + npes * (i%9);
if (dest[i] != result) {
res = false;
#ifdef VERBOSE
printf("recvbuf[%d] = %d expected %d \n", i, dest[i], result);
#endif
}
}
return res;
}
#define MAX_ELEM 256
int main (int argc, char **argv)
{
int nelem = MAX_ELEM;
if (argc > 1) {
nelem = atoi(argv[1]);
}
CHECK_HIP(hipSetDevice(get_launcher_local_rank()));
rocshmem_init();
int my_pe = rocshmem_my_pe();
int npes = rocshmem_n_pes();
int *source = (int *)rocshmem_malloc(nelem * sizeof(int));
int *dest = (int *)rocshmem_malloc(nelem * sizeof(int));
if (NULL == source || NULL == dest) {
std::cout << "Error allocating memory from symmetric heap" << std::endl;
std::cout << "source: " << source << ", dest: " << dest << ", size: "
<< sizeof(int) * nelem << std::endl;
rocshmem_global_exit(1);
}
init_sendbuf(source, nelem, my_pe);
for (int i=0; i<nelem; i++) {
dest[i] = -1;
}
rocshmem_team_t team_reduce_world_dup;
team_reduce_world_dup = ROCSHMEM_TEAM_INVALID;
rocshmem_team_split_strided(ROCSHMEM_TEAM_WORLD, 0, 1, npes, nullptr, 0,
&team_reduce_world_dup);
CHECK_HIP(hipDeviceSynchronize());
int threadsPerBlock=256;
allreduce_test<<<dim3(1), dim3(threadsPerBlock), 0, 0>>>(source, dest,
nelem, team_reduce_world_dup);
CHECK_HIP(hipDeviceSynchronize());
bool pass = check_recvbuf(dest, nelem, my_pe, npes);
printf("Test %s \t nelem %d %s\n", argv[0], nelem, pass ? "[PASS]" : "[FAIL]");
rocshmem_free(source);
rocshmem_free(dest);
rocshmem_finalize();
return 0;
}