#0: Update pgm_dispatch_golden.json

Since this golden file was created, the performance of
one_processors_all_cores_1_rta and all_processors_all_cores_1_rta seems to have
improved, so update the baseline file from a newer CI run.

The kernel preloading change slightly increased the size of kernel binaries,
since it added a loop that spun waiting for the go message. This slighlty hurt
the performance of some tests, since the increased size pushed them over some
thresholds (that were near powers of two in size).

Also increase the threshold to 5, since all the test flakes have been less than
5% off, and update the comparison script to show the percent difference.

tests/tt_metal/tt_metal/perf_microbenchmark/dispatch/compare_pgm_dispatch_perf_ci.py

@@ -20,7 +20,7 @@
     )
 )
 
-THRESHOLD = 4
+THRESHOLD_PCT = 5
 
 parser = argparse.ArgumentParser(description="Compare benchmark JSON to golden")
 parser.add_argument("json", help="JSON file to compare", type=argparse.FileType("r"))
@@ -59,12 +59,15 @@
 
     golden_time = benchmark["IterationTime"] * 1000000
     result_time = result["IterationTime"] * 1000000
-    if result_time / golden_time > (1 + THRESHOLD / 100):
-        print(f"Error:Test {name} expected value {golden_time:.2f}us but got {result_time:.2f}us")
+    result_diff_pct = result_time / golden_time * 100 - 100
+    if result_diff_pct > THRESHOLD_PCT:
+        print(
+            f"Error:Test {name} expected value {golden_time:.2f}us but got {result_time:.2f}us ({result_diff_pct:.2f}% worse)"
+        )
         exit_code = 1
-    if golden_time / result_time > (1 + THRESHOLD / 100):
+    if result_diff_pct < -THRESHOLD_PCT:
         print(
-            f"Consider adjusting baselines. Test {name} got value {result_time:.2f}us but expected {golden_time:.2f}us."
+            f"Consider adjusting baselines. Test {name} got value {result_time:.2f}us but expected {golden_time:.2f}us ({-result_diff_pct:.2f}% better)."
         )
 
 for name in result_benchmarks: