summaryrefslogtreecommitdiff
path: root/Microbenchmarks
diff options
context:
space:
mode:
authorBirte Kristina Friesel <birte.friesel@uos.de>2023-12-15 15:07:17 +0100
committerBirte Kristina Friesel <birte.friesel@uos.de>2023-12-15 15:07:17 +0100
commit3341ff57ab4464a4c3d15504df228ddc779c9fd4 (patch)
treebbf88be536ab8a90c976bee82915b3fd8ae0ec93 /Microbenchmarks
parent1e4425961d2a8ab1ec2504276dd2029a9092385c (diff)
MRAM-Latency: correctly calculate and label throughput
Diffstat (limited to 'Microbenchmarks')
-rw-r--r--Microbenchmarks/MRAM-Latency/host/app.c10
1 files changed, 5 insertions, 5 deletions
diff --git a/Microbenchmarks/MRAM-Latency/host/app.c b/Microbenchmarks/MRAM-Latency/host/app.c
index 3b5c530..3f8b1d2 100644
--- a/Microbenchmarks/MRAM-Latency/host/app.c
+++ b/Microbenchmarks/MRAM-Latency/host/app.c
@@ -143,14 +143,14 @@ int main(int argc, char **argv) {
printf("[::] DMA UPMEM | n_dpus=%d n_ranks=%d n_tasklets=%d e_type=%s n_elements=%u block_size_B=%d"
" | latency_mram_read_us=%f latency_mram_write_us=%f"
" throughput_dpu_mram_read_MBps=%f throughput_dpu_mram_write_MBps=%f"
- " throughput_mram_read_MBps=%f throughput_mram_write_MBps=%f\n",
+ " throughput_tasklet_mram_read_MBps=%f throughput_tasklet_mram_write_MBps=%f\n",
nr_of_dpus, nr_of_ranks, NR_TASKLETS, XSTR(T), input_size_dpu, BLOCK_SIZE,
((double)result.r_cycles * 1e6 / clocks_per_sec) / result.count,
((double)result.w_cycles * 1e6 / clocks_per_sec) / result.count,
- input_size * sizeof(T) / ((double)result.r_cycles * 1e6 / clocks_per_sec),
- input_size * sizeof(T) / ((double)result.w_cycles * 1e6 / clocks_per_sec),
- input_size * sizeof(T) / ((double)result.r_cycles * 1e6 * NR_TASKLETS / clocks_per_sec),
- input_size * sizeof(T) / ((double)result.w_cycles * 1e6 * NR_TASKLETS / clocks_per_sec));
+ input_size_dpu * sizeof(T) / ((double)result.r_cycles * 1e6 / clocks_per_sec),
+ input_size_dpu * sizeof(T) / ((double)result.w_cycles * 1e6 / clocks_per_sec),
+ input_size_dpu * sizeof(T) / ((double)result.r_cycles * 1e6 * NR_TASKLETS / clocks_per_sec),
+ input_size_dpu * sizeof(T) / ((double)result.w_cycles * 1e6 * NR_TASKLETS / clocks_per_sec));
}
#endif
i++;