Comments and check for successful cleanup in mon_cross_ctx_copies

1 files changed, 11 insertions, 5 deletions

diff --git a/copy_experiments/mon_cross_ctx_copies.cu b/copy_experiments/mon_cross_ctx_copies.cu
index dfccdfc..d9749e0 100644
--- a/copy_experiments/mon_cross_ctx_copies.cu
+++ b/copy_experiments/mon_cross_ctx_copies.cu
@@ -3,7 +3,13 @@
  * (by default) and logs the times that pages are copied to the GPU. Supports
  * monitoring copy progress from either the GPU or the CPU. All output times are
  * converted to ns.
- * WARNING: By default, assumes that GPU and CPU clocks tick at the same rate.
+ * Some assumptions:
+ * - GPU clock ticks at a constant rate
+ * - GPU clock ticks even while GPU is idle
+ * - Different contexts view the same underlying GPU clock
+
+WARNING: By default, assumes that GPU and CPU clocks tick at the same rate.
+ *
  */
 #include "copy_testbench.h"
 #include "../task_host_utilities.cu" // Just for GetHostDeviceTimeOffset()
@@ -46,7 +52,7 @@ static const int CPU_MON_DIVISOR = 50;
 static const int GPU_MON_THREADS = 32;
 // If delay start is enabled on any thread, how great is the delay?
 static const struct timespec DELAY_TIME = {0, 50*1000l*1000l}; // 50ms (use 30ms for GV100)
-// Should each thread doing GPU monitoring check for GPU clock skew?
+// Should each GPU-copy-monitoring thread check for GPU clock skew?
 static const bool SKEW_CHECK = false;
 
 // Size of each copy to the GPU, in bytes
@@ -204,9 +210,9 @@ void* copy_thread(void* args_raw) {
                 fprintf(stderr, "GPU clock ran at rate %f of CPU clock (%f ticks GPU vs %f ticks CPU)\n", dev_diff/host_diff, dev_diff, host_diff);
         }
 
-        cudaFreeHost(pinned_hostmem);
-        cudaFree(devtimes);
-        cudaFree(devmem);
+        SAFE(cudaFreeHost(pinned_hostmem));
+        SAFE(cudaFree(devtimes));
+        SAFE(cudaFree(devmem));
 
         return times;
 }