[tools/perf] Workaround for counter overflow on different devices

It turns out different devices have different limits for power rails
counters, so we can't compute the exact drain value if we don't know
the device. This CL replaces the exact value with an estimation
in that case.

Change-Id: I8b4f1f63a6b9716a119f5f0a8f001c7801e10eb5
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2236480Reviewed-by: Eric Seckler <eseckler@chromium.org>
Commit-Queue: Mikhail Khokhlov <khokhlov@google.com>
Cr-Commit-Position: refs/heads/master@{#776555}

tools/perf/core/tbmv3/metrics/power_rails_metric.sql

@@ -25,8 +25,10 @@ INSERT INTO power_counters VALUES
 -- Convert power counter data into table of events, where each event has
 -- start timestamp, duration and the average power drain during its duration
 -- in Watts.
--- Note that power counters wrap around at 2^30, we take that into account
--- when computing average drain between counters.
+-- Note that power counters wrap around at different values on different
+-- devices. When that happens, we ignore the value before overflow, and only
+-- take into account the value after it. This underestimates the actual power
+-- drain between those counters.
 CREATE VIEW drain_in_watts AS
 SELECT
     name,
@@ -35,7 +37,7 @@ SELECT
     CASE
         WHEN LEAD(value) OVER (PARTITION BY track_id ORDER BY ts) >= value
             THEN (LEAD(value) OVER (PARTITION BY track_id ORDER BY ts) - value)
-        ELSE (1<<30 + LEAD(value) OVER (PARTITION BY track_id ORDER BY ts) - value)
+        ELSE LEAD(value) OVER (PARTITION BY track_id ORDER BY ts)
     END /
        (LEAD(ts) OVER (PARTITION BY track_id ORDER BY ts) - ts) * 1e3
     AS drain_w