Fix Automerger for Branch-21.12 from branch-21.10
#9285
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Removes `-g` from the compile commands generated by distutils to compile Cython files. This will make our container images, conda packages, and python wheels smaller.
Signed-off-by: Jordan Jacobelli <[email protected]>
Fixes: rapidsai#9234 - [x] This PR introduces optimizations to `sort_index` when there is an already sorted `Index` object and avoids sorting them and performing a `take` operation on them. This **alleviates** a lot of **memory pressure** and has **a 3x to 6x speed up.** On a T4 GPU: `This PR`: ```python In [1]: import cudf In [2]: df = cudf.DataFrame({'a':[1, 2, 3]*100000000, 'b':['a', 'b', 'c']*100000000, 'c':[0.0, 0.12, 10.12]*100000000}) In [3]: %timeit df.sort_index() 174 ms ± 368 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) ``` `branch-21.10`: Won't fit into memory and will error :( on T4 as it tries to perform argsort on an already sorted index. `THIS PR`: ```python In [1]: import cudf In [2]: df = cudf.DataFrame({'a':[1, 2, 3]*10000000, 'b':['a', 'b', 'c']*10000000, 'c':[0.0, 0.12, 10.12]*10000000}) In [3]: %timeit df.sort_index(ascending=False) 69.1 ms ± 221 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) In [4]: %timeit df.sort_index() 15.2 ms ± 213 µs per loop (mean ± std. dev. of 7 runs, 100 loops each) In [5]: df_reversed = df[::-1] In [6]: %timeit df_reversed.sort_index() 72.6 ms ± 433 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) In [7]: %timeit df_reversed.sort_index(ascending=False) 24.1 ms ± 423 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) ``` `branch-21.10`: ```python In [1]: import cudf In [2]: df = cudf.DataFrame({'a':[1, 2, 3]*10000000, 'b':['a', 'b', 'c']*10000000, 'c':[0.0, 0.12, 10.12]*10000000}) In [3]: %timeit df.sort_index(ascending=False) 71.6 ms ± 141 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) In [4]: %timeit df.sort_index() 71.7 ms ± 189 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) In [5]: df_reversed = df[::-1] In [6]: %timeit df_reversed.sort_index() 69.1 ms ± 201 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) In [7]: %timeit df_reversed.sort_index(ascending=False) 69 ms ± 127 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) ``` - [x] Also expands params to `Series.sort_index` and refactored the common implementation to `Frame._sort_index`. Authors: - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - Michael Wang (https://github.com/isVoid) - Benjamin Zaitlen (https://github.com/quasiben) URL: rapidsai#9238
This PR fixes the `gather` API for structs columns when the input is a sliced column. Previously, `gather` calls `child_begin()` and `child_end()` to access the children column so if the input structs column is sliced then the output is incorrect. This closes rapidsai#9213, and is blocked by rapidsai#9194 due to conflict work. Authors: - Nghia Truong (https://github.com/ttnghia) Approvers: - MithunR (https://github.com/mythrocks) - Mark Harris (https://github.com/harrism) URL: rapidsai#9218
When rapidsai#9030 was merged it incorrectly resolved `get_cucollections.cmake` to use features of `rapids_cpm_find` but still call `CPMFindPackage`. This corrects the issues by properly calling `rapids_cpm_find`. Authors: - Robert Maynard (https://github.com/robertmaynard) Approvers: - Keith Kraus (https://github.com/kkraus14) - Mark Harris (https://github.com/harrism) URL: rapidsai#9189
libcudf doesn't expose zlib in the public facing API, and therefore C++ consumers don't need to also link / include zlib. Authors: - Robert Maynard (https://github.com/robertmaynard) - Keith Kraus (https://github.com/kkraus14) Approvers: - Keith Kraus (https://github.com/kkraus14) - Mark Harris (https://github.com/harrism) URL: rapidsai#9204
Only run imports tests on x86_64
Provides the Python/Cython bindings for rapidsai#8702 multibyte_split. This PR depends on rapidsai#8702 being merged first. Closes rapidsai#8557 Authors: - Jeremy Dyer (https://github.com/jdye64) - Christopher Harris (https://github.com/cwharris) Approvers: - https://github.com/nvdbaranec - Vyas Ramasubramani (https://github.com/vyasr) - GALI PREM SAGAR (https://github.com/galipremsagar) URL: rapidsai#8998
Issue rapidsai#9205 Authors: - Devavret Makkar (https://github.com/devavret) Approvers: - Mike Wilson (https://github.com/hyperbolic2346) - Vukasin Milovanovic (https://github.com/vuule) URL: rapidsai#9235
Temporary workaround for `arm64` Importing cudf on arm64 CPU only nodes is currently not working due to a difference in reported gpu devices between arm64 and amd64 Authors: - Jordan Jacobelli (https://github.com/Ethyling) Approvers: - Ray Douglass (https://github.com/raydouglass) URL: rapidsai#9252
Fixes rapidsai#8905. Attempting groupby aggregations with `LIST` keys leads to silent failures and bad results. For instance, attempting hash-based `groupby` aggregations with `LIST` keys only fails on DEBUG builds, thus: ``` /home/myth/dev/cudf/2/cpp/include/cudf/table/row_operators.cuh:447: unsigned int cudf: :element_hasher_with_seed<hash_function, has_nulls>::operator()(cudf::column_device_view, signed in t) const [with T = cudf::list_view; void *<anonymous> = (void *)nullptr; hash_function = default_ha sh; __nv_bool has_nulls = false]: block: [0,0,0], thread: [0,0,0] Assertion `false && "Unsupported type in hash."` failed. ``` In RELEASE builds, a copy of the input `LIST` column is returned, causing each output row to be interpreted as its own group. This commit adds an explicit failure for unsupported groupby key types, i.e. those that don't support equality comparisons (like `LIST`). Authors: - MithunR (https://github.com/mythrocks) Approvers: - Nghia Truong (https://github.com/ttnghia) - Robert Maynard (https://github.com/robertmaynard) - Jake Hemstad (https://github.com/jrhemstad) URL: rapidsai#9227
Fixes: rapidsai#9254 This PR fixes `deserialize` in `cudf.MultiIndex` so that there is no data-corruption happening when there are duplicate names. Authors: - GALI PREM SAGAR (https://github.com/galipremsagar) Approvers: - Vyas Ramasubramani (https://github.com/vyasr) URL: rapidsai#9258
This PR add support for struct type into the existing `drop_list_duplicates` API. This is the first time a nested type is supported in this function. Some more code cleanup has also been done. To be clear: Only structs of basic types and structs of structs are supported. Structs of lists are not, due to their complex nature. Closes rapidsai#8972. Blocked by rapidsai#9218 (it is merged). Authors: - Nghia Truong (https://github.com/ttnghia) Approvers: - Vyas Ramasubramani (https://github.com/vyasr) - https://github.com/nvdbaranec - Mark Harris (https://github.com/harrism) URL: rapidsai#9202
…apidsai#9263) Closes rapidsai#9156 This PR simplifies the parameters when calling thrust::reduce_by_key for the argmin/argmax aggregations in cudf::groupby. The illegalMemoryAccess found in rapidsai#9156 was due to invalid data being passed from thrust::reduce_by_key through to the BinaryPredicate function as documented in NVIDIA/thrust#1525 The invalid data being passed is only a real issue for strings columns where the device pointer was neither nullptr nor a valid address. The new logic provides only size_type values to thrust::reduce_by_key so invalid values can only be out-of-bounds for the input column which is easily checked before retrieving the string_view objects within the ArgMin and ArgMax operators. This the same as rapidsai#9244 but based on 21.10 Authors: - David Wendt (https://github.com/davidwendt) Approvers: - Devavret Makkar (https://github.com/devavret) - Nghia Truong (https://github.com/ttnghia) - Robert Maynard (https://github.com/robertmaynard) URL: rapidsai#9263
…lean (rapidsai#9192) Currently, we map boolean type to `pa.int8` because the bitwidth of cudf boolean mismatches that in arrow. However the implication of this mapping is subtle and may cause unwanted result such as: ```python >>> cudf.StructDtype({ "a": np.bool_, "b": np.int8, }) StructDtype({'a': dtype('int8'), 'b': dtype('int8')}) ``` This PR changes the mapping back to `pa.bool_`, and use explicit type handling when we are dealing with type conversion to arrow. Authors: - Michael Wang (https://github.com/isVoid) Approvers: - https://github.com/brandon-b-miller - H. Thomson Comer (https://github.com/thomcom) URL: rapidsai#9192
Fixes a Java column vector leak in TableTest#testParquetWriteMap. Authors: - Jason Lowe (https://github.com/jlowe) Approvers: - Robert (Bobby) Evans (https://github.com/revans2) URL: rapidsai#9271
Forward-merge `branch-21.08` into `branch-21.10`
This changes the calls in java/cudf to check for an empty input and return an empty result instead of crashing. Fixes rapidsai#9253 Authors: - Mike Wilson (https://github.com/hyperbolic2346) Approvers: - Jason Lowe (https://github.com/jlowe) URL: rapidsai#9262
Issue rapidsai#9205 depends on rapidsai#9235 Authors: - Devavret Makkar (https://github.com/devavret) Approvers: - Vukasin Milovanovic (https://github.com/vuule) - Elias Stehle (https://github.com/elstehle) - https://github.com/nvdbaranec - Mike Wilson (https://github.com/hyperbolic2346) URL: rapidsai#9242
Closes rapidsai#8660 Per discussions in thread rapidsai#8872 , this PR adds a struct-accessor member function to provide a lateral view to a struct type series. Example: ```python >>> import cudf, dask_cudf as dgd >>> ds = dgd.from_cudf(cudf.Series( ... [{'a': 42, 'b': 'str1', 'c': [-1]}, ... {'a': 0, 'b': 'str2', 'c': [400, 500]}, ... {'a': 7, 'b': '', 'c': []}]), npartitions=2) >>> ds.struct.explode().compute() a b c 0 42 str1 [-1] 1 0 str2 [400, 500] 2 7 [] ``` Authors: - Michael Wang (https://github.com/isVoid) Approvers: - Richard (Rick) Zamora (https://github.com/rjzamora) URL: rapidsai#9086
…view (rapidsai#9185) Fixes rapidsai#9140 Added `shallow_hash(column_view)` Added unit tests It computes hash values based on the shallow states of `column_view`: type, size, data pointer, null_mask pointer, offset, and the hash value of the children. `null_count` is not used since it is a cached value and it may vary based on contents of `null_mask`, and may be pre-computed or not. Fixes rapidsai#9139 Added `is_shallow_equivalent(column_view, column_view)` ~shallow_equal~ Added unit tests It compares two column_views based on the shallow states of column_view: type, size, data pointer, null_mask pointer, offset, and the column_view of the children. null_count is not used since it is a cached value and it may vary based on contents of null_mask, and may be pre-computed or not. Authors: - Karthikeyan (https://github.com/karthikeyann) Approvers: - Mark Harris (https://github.com/harrism) - Vyas Ramasubramani (https://github.com/vyasr) - Jake Hemstad (https://github.com/jrhemstad) - David Wendt (https://github.com/davidwendt) URL: rapidsai#9185
This PR strips the pyarrow-NativeFile component out of rapidsai#9225 (since those changes are not yet stable). I feel that it is reasonable to start by merging these fsspec-specific optimizations for 21.10, because they are stable and already result in a significant performance boost over the existing approach to remote storage. I still think it is very important that we eventually plumb NativeFile support into python (cudf and dask_cudf), but we will likely need to target 21.12 for that improvement. Authors: - Richard (Rick) Zamora (https://github.com/rjzamora) Approvers: - Ashwin Srinath (https://github.com/shwina) - Benjamin Zaitlen (https://github.com/quasiben) URL: rapidsai#9265
Fixes rapidsai#7830, rapidsai#8443 Features: - Use the new table metadata type that matches the table hierarchy, `table_input_metadata`. - Support struct columns in the writer. Changes: - Null masks are encoded as aligned rowgroups to avoid invalid bits when the number of encoded rows is not divisible by 8 (except for the last rowgroup in each stripe). This also affects list columns. The issue is equivalent to rapidsai#6763 (boolean columns only). - Added pushdown masks that are used to determine which child elements should not be encoded, including null mask bits. - Use pushdown masks for rowgroup alignment, null mask encoding and value encoding. - Separated the null mask encoding from value encoding - can be further moved to a separate kernel call. Breaking because the table metadata type has changed. Authors: - Vukasin Milovanovic (https://github.com/vuule) - Jason Lowe (https://github.com/jlowe) Approvers: - Robert Maynard (https://github.com/robertmaynard) - AJ Schmidt (https://github.com/ajschmidt8) - Robert (Bobby) Evans (https://github.com/revans2) - Vyas Ramasubramani (https://github.com/vyasr) - Devavret Makkar (https://github.com/devavret) - Ram (Ramakrishna Prabhu) (https://github.com/rgsl888prabhu) URL: rapidsai#9025
…9276) Authors: - Devavret Makkar (https://github.com/devavret) Approvers: - Robert Maynard (https://github.com/robertmaynard) - Keith Kraus (https://github.com/kkraus14) URL: rapidsai#9276
Aligns the function signature for `cudf.DataFrame.apply` with that of `pandas.DataFrame.apply`. This is needed so that dask can build on a common `apply` interface between backends among other reasons. Authors: - https://github.com/brandon-b-miller Approvers: - GALI PREM SAGAR (https://github.com/galipremsagar) URL: rapidsai#9275
github-actions
bot
added
CMake
galipremsagar
added
non-breaking
ajschmidt8
approved these changes
Sep 23, 2021
Codecov Report
@@ Coverage Diff @@
## branch-21.12 #9285 +/- ##
===============================================
Coverage ? 10.79%
===============================================
Files ? 116
Lines ? 18869
Branches ? 0
===============================================
Hits ? 2036
Misses ? 16833
Partials ? 0 Continue to review full report at Codecov.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR resolves conflicts to allow auto-merger from
branch-21.10tobranch-21.12: #9274