Profiler: Show used local memory.

[maleadt]

cc @thomasfaingnaert

[codecov]

Codecov Report

Attention: 7 lines in your changes are missing coverage. Please review.

Comparison is base (8dbcb2a) 72.02% compared to head (30f29cf) 71.99%.
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2124      +/-   ##
==========================================
- Coverage   72.02%   71.99%   -0.04%     
==========================================
  Files         158      158              
  Lines       14133    14145      +12     
==========================================
+ Hits        10179    10183       +4     
- Misses       3954     3962       +8     

Files	Coverage Δ
src/profile.jl	80.52% <63.15%> (-1.54%)	⬇️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[maleadt]

Example output in the interactive profiler with trace=true:

Device-side activity: GPU was busy for 1.19 µs (0.01% of the trace)
┌──────┬─────────┬───────────────────────────────────┬───────────────────────┬─
│   ID │   Start │                        Shared Mem │             Local Mem │ ...
├──────┼─────────┼───────────────────────────────────┼───────────────────────┼─
│ 4670 │ 7.96 ms │ 256 bytes static, 0 bytes dynamic │ 0 bytes / 164.188 MiB │ ...
└──────┴─────────┴───────────────────────────────────┴───────────────────────┴─

Normally shmem is hidden if both are 0, and lmem is hidden if the per thread is 0. Both are shown with raw=true, which is what I used here to demonstrate.