diff options
Diffstat (limited to 'BS/host/app.c')
| -rw-r--r-- | BS/host/app.c | 25 |
1 files changed, 21 insertions, 4 deletions
diff --git a/BS/host/app.c b/BS/host/app.c index 6d8db6a..85eb883 100644 --- a/BS/host/app.c +++ b/BS/host/app.c @@ -18,10 +18,8 @@ #include <dpu_probe.h> #endif -#if WITH_DPUINFO #include <dpu_management.h> #include <dpu_target_macros.h> -#endif #define XSTR(x) STR(x) #define STR(x) #x @@ -89,6 +87,8 @@ int main(int argc, char **argv) { // Timer declaration Timer timer; + int numa_node_rank = -2; + // Allocate DPUs and load binary #if !WITH_ALLOC_OVERHEAD DPU_ASSERT(dpu_alloc(NR_DPUS, NULL, &dpu_set)); @@ -162,6 +162,23 @@ int main(int argc, char **argv) { assert(nr_of_dpus == NR_DPUS); #endif + // int prev_rank_id = -1; + int rank_id = -1; + DPU_FOREACH (dpu_set, dpu) { + rank_id = dpu_get_rank_id(dpu_get_rank(dpu_from_set(dpu))) & DPU_TARGET_MASK; + if ((numa_node_rank != -2) && numa_node_rank != dpu_get_rank_numa_node(dpu_get_rank(dpu_from_set(dpu)))) { + numa_node_rank = -1; + } else { + numa_node_rank = dpu_get_rank_numa_node(dpu_get_rank(dpu_from_set(dpu))); + } + /* + if (rank_id != prev_rank_id) { + printf("/dev/dpu_rank%d @ NUMA node %d\n", rank_id, numa_node_rank); + prev_rank_id = rank_id; + } + */ + } + // Compute host solution if(rep >= p.n_warmup) { start(&timer, 2, 0); @@ -283,8 +300,8 @@ int main(int argc, char **argv) { if (rep >= p.n_warmup) { printf("[::] BS-UPMEM | n_dpus=%d n_ranks=%d n_tasklets=%d e_type=%s block_size_B=%d n_elements=%lu", NR_DPUS, nr_of_ranks, NR_TASKLETS, XSTR(DTYPE), BLOCK_SIZE, input_size); - printf(" with_alloc_overhead=%d b_with_load_overhead=%d b_with_free_overhead=%d ", - WITH_ALLOC_OVERHEAD, WITH_LOAD_OVERHEAD, WITH_FREE_OVERHEAD); + printf(" with_alloc_overhead=%d b_with_load_overhead=%d b_with_free_overhead=%d numa_node_rank=%d ", + WITH_ALLOC_OVERHEAD, WITH_LOAD_OVERHEAD, WITH_FREE_OVERHEAD, numa_node_rank); printf("| latency_alloc_us=%f latency_load_us=%f latency_cpu_us=%f latency_write_us=%f latency_kernel_us=%f latency_read_us=%f latency_free_us=%f", timer.time[0], timer.time[1], |