[CLI] Add TI_ADVANCED_OPTIMIZATION config env var

[archibate]

Related issue = #... (if any)

• [Click here for the format server]
• [Click here for coverage report]

---

Btw. is that possible to move advanced_optimization to CompilerConfig?

[yuanming-hu]

Thanks! Originally the flag was introduced as an ad-hoc solution to long compilation time, but now it seems that we have to live with it for long... Yes, in the future we should make it part of CompileConfig. Maybe we should change the name to allow_expensive_optimizations?

[archibate]

IMO the current name is good enough :)
@xumingkuan seems the obstacle of moving advanced_optimization to CompilerConfig is that we don't pass a CompilerConfig var to all transforms/passes? I understand the difficulty, but I think ultimately we should make all transfroms a class method of Compiler, who has a member variable of CompilerConfig?

[codecov]

Codecov Report

Merging #1212 into master will decrease coverage by 0.02%.
The diff coverage is 33.33%.

@@            Coverage Diff             @@
##           master    #1212      +/-   ##
==========================================
- Coverage   67.10%   67.08%   -0.03%     
==========================================
  Files          35       35              
  Lines        4831     4834       +3     
  Branches      885      886       +1     
==========================================
+ Hits         3242     3243       +1     
- Misses       1414     1415       +1     
- Partials      175      176       +1     

Impacted Files	Coverage Δ
python/taichi/lang/__init__.py	71.81% <33.33%> (-0.35%)	⬇️

---

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 bec9bdb...3875fb7. Read the comment docs.

[xumingkuan]

LGTM!

[xumingkuan]

IMO the current name is good enough :)
@xumingkuan seems the obstacle of moving advanced_optimization to CompilerConfig is that we don't pass a CompilerConfig var to all transforms/passes? I understand the difficulty, but I think ultimately we should make all transfroms a class method of Compiler, who has a member variable of CompilerConfig?

Please checkout #988 and https://github.com/taichi-dev/taichi/pull/1058/files#diff-60ee40c377644d6a6a4ff0cedad49110R182: we can obtain the CompileConfig in all passes by root->get_kernel()->program.config now.

[archibate]

IMO the current name is good enough :)
@xumingkuan seems the obstacle of moving advanced_optimization to CompilerConfig is that we don't pass a CompilerConfig var to all transforms/passes? I understand the difficulty, but I think ultimately we should make all transfroms a class method of Compiler, who has a member variable of CompilerConfig?

Please checkout #988 and https://github.com/taichi-dev/taichi/pull/1058/files#diff-60ee40c377644d6a6a4ff0cedad49110R182: we can obtain the CompileConfig in all passes by root->get_kernel()->program.config now.

So, should I refactor all advanced_optimization to root->get_kernel()->program.config.advanced_optimization in this PR?

[xumingkuan]

IMO the current name is good enough :)
@xumingkuan seems the obstacle of moving advanced_optimization to CompilerConfig is that we don't pass a CompilerConfig var to all transforms/passes? I understand the difficulty, but I think ultimately we should make all transfroms a class method of Compiler, who has a member variable of CompilerConfig?

Please checkout #988 and https://github.com/taichi-dev/taichi/pull/1058/files#diff-60ee40c377644d6a6a4ff0cedad49110R182: we can obtain the CompileConfig in all passes by root->get_kernel()->program.config now.

So, should I refactor all advanced_optimization to root->get_kernel()->program.config.advanced_optimization in this PR?

I think that should be in another PR.

[archibate]

Great! Let's merge this then,