[IPython] [opengl] Make OpenGL kernel print work in IPython & IDLE

[archibate]

Related issue = depends #1222

[Click here for the format server]

---

[codecov]

Codecov Report

Merging #1303 into master will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1303   +/-   ##
=======================================
  Coverage   85.50%   85.50%           
=======================================
  Files          19       19           
  Lines        3352     3353    +1     
  Branches      622      622           
=======================================
+ Hits         2866     2867    +1     
  Misses        356      356           
  Partials      130      130           

Impacted Files	Coverage Δ
python/taichi/lang/impl.py	90.09% <100.00%> (+0.03%)	⬆️

---

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 8b7078b...9bc605c. Read the comment docs.

[k-ye]

Cool!

[archibate]

Note that this only fix for the OpenGL backend. Please unify the CPU+CUDA (@yuanming-hu) and Metal (@k-ye) behavior to PythonPrintBuffer later.

[yuanming-hu]

Cool!

I think it makes sense to always call ti.sync after kernels with print. When people are printing (e.g. for debugging purposes), they usually don't really care about performance.

[archibate]

Cool!

I think it makes sense to always call ti.sync after kernels with print. When people are printing (e.g. for debugging purposes), they usually don't really care about performance.

But how do I hook kernel invocations? According to your comments in Kernel.__call__, I'm afraid detecting if self.has_print will harm non-print kernel's performance.

[yuanming-hu]

Cool!
I think it makes sense to always call ti.sync after kernels with print. When people are printing (e.g. for debugging purposes), they usually don't really care about performance.

But how do I hook kernel invocations? According to your comments in Kernel.__call__, I'm afraid detecting if self.has_print will harm non-print kernel's performance.

We can do that in the C++ part.

[archibate]

Cool!
I think it makes sense to always call ti.sync after kernels with print. When people are printing (e.g. for debugging purposes), they usually don't really care about performance.

But how do I hook kernel invocations? According to your comments in Kernel.__call__, I'm afraid detecting if self.has_print will harm non-print kernel's performance.

We can do that in the C++ part.

No, we have to perform a Python print to make IDLE/IPython get the string to their GUI, instead of console stdout.
That's why I need the print(ti.core.pop_print_buffer()) to pop C++ prints into Python prints.