User Details
- User Since
- Feb 10 2022, 3:19 PM (93 w, 6 d)
May 2 2023
Submitted.
Sep 14 2022
It works great. Thanks!
Jun 22 2022
Recover the missing r.
Thanks for the review. BTW, how can I test the changes on the cmake and bazel files? I usually run "ninja check-mlir" for testing.
- Expose hasOnlyScalarElementwiseOp as a global function.
- Specify a dependency to FuncDialect in the CMake and Bazel files.
Jun 15 2022
Make the commit title more explicit.
rebase to fix a build error for bazel
Jun 14 2022
update the python pass manager test
I reverted the commit because there is a test failure for the python pass manager. Looking into it.
Jun 13 2022
The debian failure is caused by "OSError: [Errno 28] No space left on device".
Update to use a json object instead of an array
Jun 9 2022
The build failed because there is no space available.
Jun 7 2022
It would be nice if you don't mind reverting.
Hi, just for heads-up, this change caused many failures in IREE tests including compilation errors and accuracy errors.
Apr 22 2022
Thanks Mahesh. Much cleaner!
address the comment
Apr 20 2022
LGTM. Thanks!
Apr 13 2022
Apr 12 2022
Apr 8 2022
Adding more comments.
Apr 7 2022
Adding minor comments. LGTM.
Apr 6 2022
Adding more comments.
Apr 5 2022
I left a few comments mostly related to naming and comments. I will review the rest later.
Mar 31 2022
Mar 30 2022
Rebased onto origin/main
Thanks Tobias for the info. I believe if the padding is added as an optimization for vectorization. We need to make this transformation happens before it.
If we see more cases that need a fine control, we can add a control function to this pass. Thanks!
Revert an unnecessary change on a comment.
Thanks for the review! Addressed all the TODO items.
Address the comments
Mar 29 2022
Mar 25 2022
Replace getAsValues with getValueOrCreateConstantIndexOp.
Mar 24 2022
Mar 21 2022
The plan is changed because
- I made a big code change to support matmul cases, and
- Mahesh pointed me to the tiling implementation, which works for any Linalg operations. I need to take a look at the code and understand how it works for all without any restrictions on the linalg operation, and
- Apply the same or similar approach to my implementation.
Mar 18 2022
Address Mahesh's comments
Mar 17 2022
merge and rebase
Mar 16 2022
check projected permutation for inputs
Mar 15 2022
Mar 4 2022
LGTM. Thanks
Feb 28 2022
Fix the bazel build error and rewrite the commit message.
Reopening: The push to origin/main caused a build break due to a cyclic header inclusion between Tensor.h and Transforms.h. The issue didn't occur in my local machine, but it broke the testing system.
Merged to origin/main.
Feb 26 2022
Addressed Uday's comments.
Thanks Uday for the comments. Please take another look.
Feb 25 2022
Address Mahesh's comments
The code is updated with the control function approach. Please take a look.
Add control function support
Feb 22 2022
Merged into the LLVM main.
Feb 18 2022
Introduce resizeSplat instead of using reshape since reshape is not suppposed
to change the number of elements.
Feb 16 2022
There is already reshape(), but it didn't support a shape with a different number of element even for a splat type.
I relaxed the assert condition to allow a type with a different number of elements.
Support a different shape for reshape for a splat type.
Use the attribute operand instead of matching.
Feb 15 2022
Right. It does not handle the splat case yet.
I have added two more unit tests.
add two more unit tests
Feb 14 2022
Fix a missing change for IterTy and ElemTy
rebase onto main
Resolve the patch error