Always synchronize GPU and CPU time in the current process

Performing the synchronization in a child process was fragile,
high-overhead, and unnecessary for our use cases.

2 files changed, 8 insertions, 5 deletions

diff --git a/copy_experiments/mon_cross_ctx_copies.cu b/copy_experiments/mon_cross_ctx_copies.cu
index 2912aec..e836179 100644
--- a/copy_experiments/mon_cross_ctx_copies.cu
+++ b/copy_experiments/mon_cross_ctx_copies.cu
@@ -1,4 +1,4 @@
-/* Copyright 2022 Joshua Bakita
+/* Copyright 2024 Joshua Bakita
  * Starts two copies of configurable size in seperate contexts at the same time
  * (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
@@ -12,7 +12,7 @@ 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()
+#include "../task_host_utilities.cu"
 #include <stdio.h>
 #include <stdlib.h>
 #include <stdint.h>
diff --git a/preemption_logger.cu b/preemption_logger.cu
index d93600e..89b348a 100644
--- a/preemption_logger.cu
+++ b/preemption_logger.cu
@@ -1,4 +1,4 @@
-/* Copyright 2023 Joshua Bakita
+/* Copyright 2024 Joshua Bakita
  * Tool that tracks preemptions as discontinuities in GPU time and logs them.
  *
  * The logged intervals are output to standard out as a CSV, with each row
@@ -92,8 +92,11 @@ int main(int argc, char **argv) {
         if (!skip_conversion) {
                 // Note that this appears to almost always be 1.0 on recent GPUs
                 d2h_scale = GetGPUTimerScale(0);
-                // The following uses a subprocess for clock synchronization. Sketch?
-                GetHostDeviceTimeOffset(0, &host_s, &dev_ns);
+                InternalReadGPUNanoseconds(0, &host_s, &dev_ns);
+                if (host_s == 0 && !dev_ns) {
+                        fprintf(stderr, "Unable to synchronize time with the GPU. Aborting...\n");
+                        return 1;
+                }
         }
 
         // Initialize GPU and CPU memory to store the intervals of execution