[TSan] Update test values

[vlad.tsyrklevich]

Summary

This test was broken by the tail duplication logic being changed in
r311139, update the test values and add a note about how to properly run
a benchmark to verify that the values are safe to update.

[kcc]

This sucks. If we have so many pops in the hot function, the test essentially becomes useless.
Does this function really need to many pops? Isn't it a pessimization?

[vlad.tsyrklevich]

The increase in pops is because of function epilogue duplication, a single execution of the function should still see the same number of pops. That said, it is significantly less useful to be tracking the total number of pops rather than accounting for the control-flow somehow.

compiler-rt/trunk/lib/tsan/check_analyze.sh

 #!/bin/bash
 #
 # Script that checks that critical functions in TSan runtime have correct number
 # of push/pop/rsp instructions to verify that runtime is efficient enough.
+#
+# This test can fail when backend code generation changes the output for various
+# tsan interceptors. When such a change happens, you can ensure that the
+# performance has not regressed by running the following benchmarks before and
+# after the breaking change to verify that the values in this file are safe to
+# update:
+# ./projects/compiler-rt/lib/tsan/tests/rtl/TsanRtlTest
+#   --gtest_also_run_disabled_tests --gtest_filter=DISABLED_BENCH.Mop*
 
 set -u
 
 if [[ "$#" != 1 ]]; then
   echo "Usage: $0 /path/to/binary/built/with/tsan"
   exit 1
 fi
 
for f in write1 write2 write4 write8; do
   check $f rsp 1
   check $f push 2
   check $f pop 12
 done
 
 for f in read1 read2 read4 read8; do
   check $f rsp 1
   check $f push 3
-  check $f pop 3
+  check $f pop 18
 done
 
 for f in func_entry func_exit; do
   check $f rsp 0
   check $f push 0
   check $f pop 0
   check $f call 1  # TraceSwitch()
 done
 
 echo LGTM