Fix some precompute transformation algorithm bugs that arose

[weiya711]

Fix a bug in the algorithm so that the precompute transformation can only occur on cascaded ForallNodes without any WhereNodes inside the forall->stmt. The current test DISABLED_tile_dotProduct_1 tests this case but is currently disabled because of the issue stated in the test's comments/FIXME.

Also fix a minor bug where expected and A were flipped in test/tests-workspaces.cpp when calling ASSERT_TENSOR_EQ(expected, A).

Also added in some disabled precompute tests that should work but currently do not due to certain new bugs found in the algorithm. I would like to add these tests in to the regression disabled and track re-enabling them using Git issues.

[weiya711]

Run on GPU is pending, will link to a passing Action

[stephenchouca]

Do you know why exactly the arguments to ASSERT_TENSOR_EQ needed to be flipped? Isn't that function supposed to be commutative since it's just checking for equality?

[weiya711]

Do you know why exactly the arguments to ASSERT_TENSOR_EQ needed to be flipped? Isn't that function supposed to be commutative since it's just checking for equality?
@stephenchouca

Although functionally they are equivalent and the code is still correct, when there is an error message and expected and A are switched, something like this gets printed:

Value of: equals(expected, actual)
  Actual: false
Expected: true
Google Test trace:
/Users/oliviahsu/Files/research/taco2/taco/test/test.cpp:37:   actual: expected () (; ):
[3.5739e+08]
/Users/oliviahsu/Files/research/taco2/taco/test/test.cpp:36: expected: A () (; ):
[3.24845e+07]

which I found super annoying and confusing so I decided to bulk switch them just for the test file that I was working on. Mainly it's the mismatch between actual: expected and expected: A

[stephenchouca]

Do you know why exactly the arguments to ASSERT_TENSOR_EQ needed to be flipped? Isn't that function supposed to be commutative since it's just checking for equality?
@stephenchouca

Although functionally they are equivalent and the code is still correct, when there is an error message and expected and A are switched, something like this gets printed:

Value of: equals(expected, actual)
  Actual: false
Expected: true
Google Test trace:
/Users/oliviahsu/Files/research/taco2/taco/test/test.cpp:37:   actual: expected () (; ):
[3.5739e+08]
/Users/oliviahsu/Files/research/taco2/taco/test/test.cpp:36: expected: A () (; ):
[3.24845e+07]

which I found super annoying and confusing so I decided to bulk switch them just for the test file that I was working on. Mainly it's the mismatch between actual: expected and expected: A

Ah I see, that makes sense.