Feature/lospn2std opt select by mhalk · Pull Request #61 · esa-tu-darmstadt/spn-compiler

mhalk · 2021-06-28T11:47:34Z

Optimization feature which will replace certain Categorical and Histogram leaves with Select operations.

LoSPN Categorical leaves are lowered to SelectOp if they consist of two probabilities.

- ADDED support for marginals when lowering from lospn to cpu - ADDED/FIXED support for log-computations - FIXED argument types of SPNSelectLeaf

… log-space.

- Test if lospn.select is replaced by llvm.select (scalar and vectorized)

sommerlukas

Mostly looks good, but the lowering tests need improvement:

The tests should be more focused on the select and contain less other operations, so it is easier to identify the actual interesting part.
The tests should cover a scenario where the input type to the leaf is different from it's result type, e.g. taking i32 as input, but producing f64.

For the remaining nits, please see the inline comments.

sommerlukas · 2021-07-29T09:17:56Z

+                                               op.supportMarginalAttr());
+    return success();
+  }
+  return failure();


If possible, provide a message on failure by returning rewriter.notifyMatchFailure(...)

ACK, an explanation is now provided, regarding the rewrite conditions.

sommerlukas · 2021-07-29T10:57:03Z

+      return success();
+    }
+  }
+  return failure();


If possible, provide a message on failure by returning rewriter.notifyMatchFailure(...)

ACK, an explanation is now provided, regarding the rewrite conditions.

…m LS.

mhalk · 2021-08-02T12:59:38Z

Thanks for your review.

The tests should be more focused on the select and contain less other operations, so it is easier to identify the actual interesting part.

The tests are now very concentrated on these instructions, I think they are far better now (and far easier to understand).

The tests should cover a scenario where the input type to the leaf is different from it's result type, e.g. taking i32 as input, but producing f64.

I thought that would be covered by select-replacement-histogram.mlir
See lines L14 and L41, respectively.

sommerlukas

Thanks for the update. I found one minor issue in the code, see inline comment.

Other than that, we should also make sure that the optimization is also covered by the Python tests that also check the result of the computation.

Please make sure that

not all Categorical or Histogram in the existing are replaced by the new optimization
for every configuration (vectorization, GPU, ...) at least one Categorical and Histogram is replaced and check that the result is computed correctly.

sommerlukas · 2021-08-02T16:32:02Z

+    // Convert from floating-point input to integer value if necessary.
+    // This conversion is also possible in vectorized mode.
+    auto intVectorTy = VectorType::get(vectorType.getShape(), inputTy);
+    thresholdVec = rewriter.create<FPToSIOp>(op->getLoc(), thresholdVec, intVectorTy);


Why convert a constant value? thresholdVec is vector of constant values, so it should not be created as vector of floats and then converted, but rather should be created as a vector integers in the first place.

Absolutely right, it'll be fixed.

mhalk · 2021-08-02T17:55:10Z

Thanks for the update. I found one minor issue in the code, see inline comment.

Other than that, we should also make sure that the optimization is also covered by the Python tests that also check the result of the computation.

ACK.

Please make sure that

* not all Categorical or Histogram in the existing are replaced by the new optimization

* for every configuration (vectorization, GPU, ...) at least one Categorical and Histogram is replaced and check that the result is computed correctly.

But all "2 probability/bucket Categoricals/Histograms" should be replaced, right?
So "not all" means, there should be corresponding (Cat/Hist-)leaves in the SPN with a count != 2?
Shall I add new tests or adapt existing ones?

…f convert).

sommerlukas · 2021-08-03T07:19:15Z

But all "2 probability/bucket Categoricals/Histograms" should be replaced, right?

Yes.

So "not all" means, there should be corresponding (Cat/Hist-)leaves in the SPN with a count != 2?

Correct, there should be Cat/Hist that are not eligible for the transformation.

Shall I add new tests or adapt existing ones?

Please adapt the existing ones, so they are not eligible for the transformation and add new tests for the transformation.

- Now, they will not(!) be eligible for "select-transformation". (New tests will be added in the next commit.)

- Added corresponding scalar and vectorized codegen - Adapted regression tests - Added Python computation tests

mhalk added 4 commits May 31, 2021 19:14

[WIP] Added optimization which replaces specific Categoricals by Select.

01f3383

LoSPN Categorical leaves are lowered to SelectOp if they consist of two probabilities.

[WIP] Added support for Histogram w.r.t. Select-canonicalization.

d315a35

- ADDED support for marginals when lowering from lospn to cpu - ADDED/FIXED support for log-computations - FIXED argument types of SPNSelectLeaf

[WIP] Added stubs for vectorized transformation of SPNSelectLeaf.

b59fb0c

Merge branch 'develop' into feature/lospn2std-opt-select

fe3193a

mhalk requested a review from sommerlukas June 28, 2021 11:47

mhalk self-assigned this Jun 28, 2021

mhalk added the enhancement New feature or request label Jun 28, 2021

mhalk added 5 commits June 28, 2021 22:07

[WIP] Implemented first iteration of vectorized SPNSelectleaf pattern.

9cb2ed0

[WIP] [FIX] SPNSelectLeaf should now emit the correct type when using…

1e5f758

… log-space.

[WIP] [FIX] Removed faulty/unnecessary log-space check.

68cf270

[RC] Added tests for lospn.select transformations.

9045d7f

- Test if lospn.select is replaced by llvm.select (scalar and vectorized)

Merge branch 'develop' into feature/lospn2std-opt-select

2382714

sommerlukas requested changes Jul 29, 2021

View reviewed changes

mhalk added 2 commits August 2, 2021 14:24

[FIX] Reduced / concentrated tests on specific instructions.

01840f3

[FIX] Incorporated remaining feedback (excluding LoSPNOps.cpp/.h) fro…

7f2f782

…m LS.

mhalk requested a review from sommerlukas August 2, 2021 13:04

sommerlukas requested changes Aug 2, 2021

View reviewed changes

[FIX] Create correct element-type for the threshold vector (instead o…

89ff05b

…f convert).

mhalk added 3 commits August 5, 2021 13:50

[FIX] Adapted Python tests.

47beb84

- Now, they will not(!) be eligible for "select-transformation". (New tests will be added in the next commit.)

Implemented avoidance of out-of-bounds access.

6ebc4e0

- Added corresponding scalar and vectorized codegen - Adapted regression tests - Added Python computation tests

Merge branch 'develop' into feature/lospn2std-opt-select

7a82eb8

csvtuda mentioned this pull request Nov 18, 2021

Feature/slp vectorization #69

Merged

Uh oh!

Conversation

mhalk commented Jun 28, 2021

Uh oh!

sommerlukas left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

sommerlukas Jul 29, 2021

Choose a reason for hiding this comment

Uh oh!

mhalk Aug 2, 2021

Choose a reason for hiding this comment

Uh oh!

Uh oh!

sommerlukas Jul 29, 2021

Choose a reason for hiding this comment

Uh oh!

mhalk Aug 2, 2021

Choose a reason for hiding this comment

Uh oh!

mhalk commented Aug 2, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

sommerlukas left a comment

Choose a reason for hiding this comment

Uh oh!

sommerlukas Aug 2, 2021

Choose a reason for hiding this comment

Uh oh!

mhalk Aug 2, 2021

Choose a reason for hiding this comment

Uh oh!

mhalk commented Aug 2, 2021

Uh oh!

sommerlukas commented Aug 3, 2021

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

mhalk commented Aug 2, 2021 •

edited

Loading