[TIR] Semantic extension for CSE Pass #10544

yuanfz98 · 2022-03-09T10:28:08Z

Hello,

This is a PR for #10211, aimed to add semantic extension for CSE Pass. Should be an easy PR.

For rewriting the test case using TVMScript, if no one is now working on it, I will later refactor the rest part.

Thanks for your time for your code review!

tqchen · 2022-03-09T14:10:00Z

src/tir/transforms/common_subexpr_elim_tools.cc

@@ -727,7 +728,10 @@ bool EquivalentTerms(const PrimExpr& a, const PrimExpr& b) {
  // For now, we just check the syntactic equality, but that could later become a semantic test,
  // for instance identifying computations modulo commutativity (like x+y and y+x), or modulo
  // associativity (like (x+y)+z and x+(y+z)), etc.
-  return EqualTerms(a, b);
+  arith::Analyzer analyser;


Thanks @yuanfz98 , One thing to note here is that running recursively simplification on every subexpression would result in a possibly quadratic complexity wrt to the expression size, so we would want to use it with care, perhaps only triger with limited expression length to avoid long compilation time.

In this particular case, directly running a simplification pass before the common subexpr elim would have a same effect.

Yes I agree, I will close this PR. I will follow the ongoing development of pass dependencies and come back to this feature after. Thanks for your reply !
cc @FranckQC @zhiics

…emantic extension for CSE Pass apache#10544

…orm - avoids comparison of terms (#11574) The CSE pass had been designed for potentially allowing comparisons (and commonings) of equivalent terms (like (x+y)+z and x+(y+z)), where **the notion of being equivalent was customizable, and no assumption was made about it**. That means that the implementation of the equivalence test function `EquivalentTerms()` - which was at the moment just calling the syntactical equality test `EqualTerms()` - could be replaced later by a cleverer equality test. However, having such a generic way of comparing elements meant that in the function `SyntacticToSemanticComputations()`, where we were going from a hashtable of syntactical entities to what I called a vector of "semantical entites" (which are just canonical forms/representants of classes of equivalence of terms), **the only way was to compare each pair**. That resulted in a quadratic behavior of this function, but there was no way around it as in order to merge equivalent entities into their class of equivalence, we had to compare them. **This PR essentially does the following:** - When computing the classes of equivalences of terms (therefore transforming a ComputationTable (i.e. a hashtable) into a vector of classes of equivalence) : **instead of comparing each pair of terms, relies on a normalization procedure to obtain a normal form for each of them**. That transforms a small part of the algorithm that was quadratic to n.logn. However, it's difficult to see improvements in practice, in particular for average sized programs, as that part was a "small" quadratic to a "big" n.logn (finding things in a hash-table, copying it to a vector, etc). It was probably going from a complexity of ~O(((n²-n)/2) + n.logn) to a complexity of ~O(3n + n.logn), so potential gains would only be expected for very large programs. - Completely gives the user the possibility to turn ON/OFF the semantical comparisons of terms. It is turned OFF by default (as it's quite longer to compile with it ON, unsurprisingly), which means that by default, the equivalence coincides with the (syntactical) equality of terms. As the pass was written with the possibility to do these additional commonings (like (x+y)+z and x+(y+z)), it was a good time to fully plug that completely, up to the Python user who can now turn that ON if he wants to. But again, it is OFF by default, so no real change on that. To run it ON, simply do: `with tvm.transform.PassContext(config={'tir.enable_equiv_terms_in_cse_tir':True}):` before calling `build()` - When this boolean is set to ON, it uses a simple implementation of the normalization function with equivalences that uses `arith::Analyzer::Simplify` as noted by in #10544 . Note that this is not a real normalization procedure as it is incomplete (i.e., it is not guarantee to converge to the normal form), but it is correct, and it works well with most properties : associativity of +, distributivity of * on +, etc. - Clarifies and enhance the test base for the pass. In particular, it adds the tests that were written in #10544 but which did not make it through. - Also add the test ( https://github.com/AndrewZhaoLuo/TVM-Sandbox/blob/19284ddbd6bb28af61c0c2aa8bb334c5c53731a7/tir/test_inconsistent_tir_lowering.py#L1 ) demonstrating the (older) non-deterministic lowering and put it into a proper test, as I found it useful for making sure that this does not happen again. It has been copied from #10663 and only slightly adapted (in particular for doing the comparison of hashes automatically instead of printing them and relying on a human to compare them).

…orm - avoids comparison of terms (apache#11574) The CSE pass had been designed for potentially allowing comparisons (and commonings) of equivalent terms (like (x+y)+z and x+(y+z)), where **the notion of being equivalent was customizable, and no assumption was made about it**. That means that the implementation of the equivalence test function `EquivalentTerms()` - which was at the moment just calling the syntactical equality test `EqualTerms()` - could be replaced later by a cleverer equality test. However, having such a generic way of comparing elements meant that in the function `SyntacticToSemanticComputations()`, where we were going from a hashtable of syntactical entities to what I called a vector of "semantical entites" (which are just canonical forms/representants of classes of equivalence of terms), **the only way was to compare each pair**. That resulted in a quadratic behavior of this function, but there was no way around it as in order to merge equivalent entities into their class of equivalence, we had to compare them. **This PR essentially does the following:** - When computing the classes of equivalences of terms (therefore transforming a ComputationTable (i.e. a hashtable) into a vector of classes of equivalence) : **instead of comparing each pair of terms, relies on a normalization procedure to obtain a normal form for each of them**. That transforms a small part of the algorithm that was quadratic to n.logn. However, it's difficult to see improvements in practice, in particular for average sized programs, as that part was a "small" quadratic to a "big" n.logn (finding things in a hash-table, copying it to a vector, etc). It was probably going from a complexity of ~O(((n²-n)/2) + n.logn) to a complexity of ~O(3n + n.logn), so potential gains would only be expected for very large programs. - Completely gives the user the possibility to turn ON/OFF the semantical comparisons of terms. It is turned OFF by default (as it's quite longer to compile with it ON, unsurprisingly), which means that by default, the equivalence coincides with the (syntactical) equality of terms. As the pass was written with the possibility to do these additional commonings (like (x+y)+z and x+(y+z)), it was a good time to fully plug that completely, up to the Python user who can now turn that ON if he wants to. But again, it is OFF by default, so no real change on that. To run it ON, simply do: `with tvm.transform.PassContext(config={'tir.enable_equiv_terms_in_cse_tir':True}):` before calling `build()` - When this boolean is set to ON, it uses a simple implementation of the normalization function with equivalences that uses `arith::Analyzer::Simplify` as noted by in apache#10544 . Note that this is not a real normalization procedure as it is incomplete (i.e., it is not guarantee to converge to the normal form), but it is correct, and it works well with most properties : associativity of +, distributivity of * on +, etc. - Clarifies and enhance the test base for the pass. In particular, it adds the tests that were written in apache#10544 but which did not make it through. - Also add the test ( https://github.com/AndrewZhaoLuo/TVM-Sandbox/blob/19284ddbd6bb28af61c0c2aa8bb334c5c53731a7/tir/test_inconsistent_tir_lowering.py#L1 ) demonstrating the (older) non-deterministic lowering and put it into a proper test, as I found it useful for making sure that this does not happen again. It has been copied from apache#10663 and only slightly adapted (in particular for doing the comparison of hashes automatically instead of printing them and relying on a human to compare them).

yuanfz98 added 6 commits March 9, 2022 10:35

commit

d767e28

black format

cd13186

rm prints

6a794b6

format

e14af51

format

9a75aca

Merge branch 'main' into cse_semantic_extension_fix

5f9c830

yuanfz98 requested review from junrushao, vinx13, tqchen, kparzysz-quic, ZihengJiang, masahi, Hzfengsy, comaniac, jroesch, areusch, yzhliu, merrymercy and icemelon as code owners March 9, 2022 10:28

masahi approved these changes Mar 9, 2022

View reviewed changes

tqchen requested changes Mar 9, 2022

View reviewed changes

yuanfz98 closed this Mar 9, 2022

FranckQC mentioned this pull request May 26, 2022

[TIR] Avoid all-pairs comparison in subexpr elimination #11423

Closed

FranckQC added a commit to FranckQC/tvm that referenced this pull request Jun 4, 2022

Tests for equivalence added, copied from @yuanfz98 in his PR: [TIR] S…

b3ab028

…emantic extension for CSE Pass apache#10544

FranckQC mentioned this pull request Jun 4, 2022

[TIR] CSE pass : Restrict the equivalence to be decided by a normal form - avoids comparison of terms #11574

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[TIR] Semantic extension for CSE Pass #10544

[TIR] Semantic extension for CSE Pass #10544

yuanfz98 commented Mar 9, 2022

tqchen Mar 9, 2022

yuanfz98 Mar 9, 2022

[TIR] Semantic extension for CSE Pass #10544

[TIR] Semantic extension for CSE Pass #10544

Conversation

yuanfz98 commented Mar 9, 2022

tqchen Mar 9, 2022

Choose a reason for hiding this comment

yuanfz98 Mar 9, 2022

Choose a reason for hiding this comment