Clean up and split demand_paging_speed.c from paging_speed.c

This also removes the --no-seq flag from paging_speed.c, which
never worked.

Cleanups include:
- Copyright notices
- Better usage messages & code comments
- More similar structure in the CPU-side benchmarks
- Random disk buffer initialization in demand_paging_speed.c
- Output comments detailing data source
- Stripping of unused or deprecated codepaths

At this point, the mapping of benchmarks to RTSS'22 plots is:
- paging_speed: Fig. 4
- demand_paging_speed: Fig. 10
- directio_paging_speed: Fig. 10
- gpu_paging_speed: Fig. 10
- gpu_paging_overhead_speed: Fig. 11
- gpu_paging_evil_task: Figs. 12 and 13

1 files changed, 4 insertions, 0 deletions

diff --git a/gpu_paging_overhead_speed.cu b/gpu_paging_overhead_speed.cu
index c7a0f3a..0cf364e 100644
--- a/gpu_paging_overhead_speed.cu
+++ b/gpu_paging_overhead_speed.cu
@@ -1,3 +1,6 @@
+/**
+ * Copyright 2022 Joshua Bakita
+ */
 #include <stdio.h>
 #include <cuda.h>
 #include <curand_kernel.h> // curandState_t and curand
@@ -78,6 +81,7 @@ __global__ void count_zero(char* buf, uint64_t buf_len) {
         atomicAdd_block((unsigned long long int*)&gpu_res, (unsigned long long int)num_zero);
 }
 
+// Subtract first parameter from second parameter. Return as nanoseconds.
 long time_diff_ns(struct timespec start, struct timespec stop) {
         return (s2ns(stop.tv_sec) + stop.tv_nsec) - (s2ns(start.tv_sec) + start.tv_nsec);
 }