Merge pull request triton-lang#9 from openai/keren/todo

[TODO] Add a bunch of TODOs

TODO.md

@@ -2,10 +2,27 @@
 
 ## cleanups
 
+* Hard coded alignment, don't why it exists and why it is that number
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Analysis/Allocation.cpp#L165
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Analysis/Allocation.cpp#L176
+* Shared memory buffer sizes should be determined before graph coloring
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Analysis/Allocation.cpp#L493
+* Don't use any LLVM dialects in TritonGPU conversions
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Analysis/Membar.cpp#L110
+* We should have a more general barrier op
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Analysis/Membar.cpp#L180
+* tt.load's verifier has the `allowTensorPointerType` flag, which skips checking encoding if ptr is a tensor pointer. This is a hack and should be improved because it is unclear and violating the encoding is not good idea.
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Dialect/Triton/IR/Traits.cpp#L10
 * Don't call arith to LLVM conversion from MLIR
 * linearize/delinearize helper have been duplicated (most likely due to layering problems). This should be merged
 
 ## bug fixes
 
 * Currently lowering of InsertSliceAsyncV2Op doesn't work if the mask is not a scalar even though the op semantic allows it.
-* smem base is being passed to the patterns. This seems broken when using function calls.
+* smem base is being passed to the patterns. This seems broken when using function calls.
+* Revisit the constraint for RewriteTensorPointer pass
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/lib/Dialect/TritonGPU/Transforms/RewriteTensorPointer.cpp#L644
+* The IR output of `make_tensor_ptr` is wrong. `!tt.ptr` is ignored in the output.
+https://github.com/openai/triton-hopper/blob/1ada046fdaef13f94dc7e2f6e6d0966e5d1efe92/include/triton/Dialect/Triton/IR/TritonOps.td#L562
+* We rely on the `cuda-python` package currently, which prevents us from building triton on any node without CUDA installed. We should invoke TMA related functions in our thin CUDA wrapper.
+https://github.com/openai/triton-hopper/blob/b6a6b32b0ee79e93247d20c95f15fd75039a40b9/python/triton/compiler/utils.py#L3