[bug] Fix test_fibonacci

[k-ye]

This is probably more of a workaround. I don't think the IR optimizer should change the semantics of the for loop. That means we need to preserve the top-level if if the compile time condition is true, or transform it to a serial kernel, like this PR does..

Maybe we should also have a dedicated method to instruct that the for loop is meant to be executed inside a serial kernel. Right now this for _ in range(1): seems hacky..

Related issue = #1393 (comment)

[Click here for the format server]

---

[archibate]

Cool hack.. Btw I'm wondering once we come to real function support, i.e. non-inline, will the for loop inside function being parallelized?

[k-ye]

will the for loop inside function being parallelized?

Hmm, i don't think so? Why?

BTW, did this also break on OpenGL?

[archibate]

Hmm, i don't think so? Why?

Just to confirm...

BTW, did this also break on OpenGL?

Don't know, my laptop is of OpenGL 3.3 which doesn't support compute shader..

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@b709fb6). Click here to learn what that means.
The diff coverage is n/a.

@@            Coverage Diff            @@
##             master    #1414   +/-   ##
=========================================
  Coverage          ?   66.76%           
=========================================
  Files             ?       37           
  Lines             ?     5206           
  Branches          ?      934           
=========================================
  Hits              ?     3476           
  Misses            ?     1570           
  Partials          ?      160           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b709fb6...1dd441e. Read the comment docs.

[xumingkuan]

IIRC some other unit tests also use for _ in range(1): to make the loop serial. It makes sense to me using only for (but not if) to make the loop serial for now.