Upgrade thrust version to 1.15 #9912
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
Compile times between 1.12 and 1.15 are comparable.
libcudf library size with 1.15 is < 1MB smaller compared to 1.12
Runtime Performance results:
```
Time CPU Time Old Time New CPU Old CPU New
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
COMPILED_BINARYOP/BM_compiled_binaryop<double, double, double, cudf::binary_operator::MOD>/100000/manual_time +0.0519 +0.0268 10 10 27 28
COMPILED_BINARYOP/BM_compiled_binaryop<int32_t, int64_t, double, cudf::binary_operator::PMOD>/10000/manual_time +0.0851 +0.0295 9 10 28 29
COMPILED_BINARYOP/BM_compiled_binaryop<int, int, int, cudf::binary_operator::SHIFT_LEFT>/10000/manual_time -0.0971 -0.0218 6 5 25 24
COMPILED_BINARYOP/BM_compiled_binaryop<double, int8_t, bool, cudf::binary_operator::LOGICAL_AND>/10000/manual_time -0.0717 -0.0196 7 6 26 25
COMPILED_BINARYOP/BM_compiled_binaryop<decimal32, decimal32, decimal32, cudf::binary_operator::NULL_MAX>/10000/manual_time +0.0748 +0.0191 7 8 26 27
ReductionDictionary/float_max/10000/manual_time -0.0714 -0.0512 31498 29249 50665 48071
ReductionScan/int16_nulls/100000/manual_time -0.1073 -0.0562 17598 15710 35995 33974
Concatenate/BM_concatenate<int64_t, true>/4096/64/manual_time -0.1211 -0.0844 0 0 0 0
OrcWrite/floats_file_output/31/0/32/1/0/manual_time +0.0227 +0.0576 84 86 78 83
CopyIfElse/int16/4096/manual_time +0.0648 +0.0262 0 0 0 0
CopyIfElse/float64/262144/manual_time +0.0609 +0.0192 0 0 0 0
CopyIfElse/int16_no_nulls/4096/manual_time +0.1355 +0.0219 0 0 0 0
CopyIfElse/int16_no_nulls/32768/manual_time +0.1582 +0.0395 0 0 0 0
CopyIfElse/uint32_no_nulls/4096/manual_time +0.1300 +0.0213 0 0 0 0
CopyIfElse/uint32_no_nulls/32768/manual_time +0.0500 +0.0155 0 0 0 0
CopyIfElse/float64_no_nulls/4096/manual_time +0.0583 +0.0083 0 0 0 0
TypeDispatcher/fp64_bandwidth_host/1/1024/1/manual_time +0.1010 +0.0236 3291 3623 22245 22771
TypeDispatcher/fp64_bandwidth_host/2/1024/1/manual_time +0.1157 +0.0309 4888 5453 23798 24533
TypeDispatcher/fp64_bandwidth_host/4/1024/1/manual_time +0.0901 +0.0324 8053 8779 26780 27647
TypeDispatcher/fp64_bandwidth_host/1/2048/1/manual_time +0.1191 +0.0246 3335 3732 22378 22928
TypeDispatcher/fp64_bandwidth_host/2/2048/1/manual_time +0.1349 +0.0321 4890 5550 23839 24604
TypeDispatcher/fp64_bandwidth_host/4/2048/1/manual_time +0.0622 +0.0177 8820 9369 27717 28207
TypeDispatcher/fp64_bandwidth_host/1/4096/1/manual_time +0.0951 +0.0229 3387 3709 22375 22888
TypeDispatcher/fp64_bandwidth_host/2/4096/1/manual_time +0.0507 +0.0125 5509 5788 24491 24796
TypeDispatcher/fp64_bandwidth_device/4/1024/1/manual_time +0.0601 +0.0266 10775 11422 29432 30215
TypeDispatcher/fp64_bandwidth_device/1/2048/1/manual_time +0.0618 +0.0290 9258 9831 27967 28778
TypeDispatcher/fp64_bandwidth_device/2/2048/1/manual_time +0.0516 +0.0218 9773 10277 28453 29073
TypeDispatcher/fp64_bandwidth_device/1/4096/1/manual_time +0.0653 +0.0259 9265 9869 27956 28681
TypeDispatcher/fp64_bandwidth_no/1/1024/1/manual_time +0.0868 +0.0275 3441 3739 22367 22983
TypeDispatcher/fp64_bandwidth_no/2/1024/1/manual_time +0.1635 +0.0396 3690 4293 22609 23503
TypeDispatcher/fp64_bandwidth_no/4/1024/1/manual_time +0.1144 +0.0316 4523 5040 23445 24186
TypeDispatcher/fp64_bandwidth_no/1/2048/1/manual_time +0.1469 +0.0413 3497 4011 22382 23306
TypeDispatcher/fp64_bandwidth_no/2/2048/1/manual_time +0.0550 +0.0216 4097 4323 23012 23509
TypeDispatcher/fp64_bandwidth_no/2/4096/1/manual_time +0.0510 +0.0151 4418 4643 23447 23800
TypeDispatcher/fp64_bandwidth_no/8/4096/1/manual_time +0.1093 +0.0235 6787 7528 26000 26612
TypeDispatcher/fp64_bandwidth_no/4/8192/1/manual_time +0.0608 +0.0071 5435 5766 24657 24832
SetNullmask/SetNullMaskKernel/1048576/manual_time -0.0949 -0.0022 3800 3440 22750 22699
StringExtract/four/32768/32/manual_time +0.0578 +0.0570 1 1 1 1
StringExtract/eight/4096/32/manual_time -0.0973 -0.0969 5 4 5 4
StringExtract/eight/4096/128/manual_time -0.0882 -0.0879 5 5 5 5
StringExtract/eight/4096/512/manual_time -0.0780 -0.0777 5 5 5 5
StringExtract/eight/4096/2048/manual_time -0.0703 -0.0701 5 5 6 5
StringExtract/eight/4096/8192/manual_time -0.0718 -0.0716 7 6 7 6
UrlDecode/BM_url_decode<10>/10000000/100/manual_time +0.0576 +0.0577 37 39 37 39
UrlDecode/BM_url_decode<50>/100000000/10/manual_time +0.0599 +0.0599 140 148 140 148
Shift/shift_half_nullable_out/1048576/manual_time -0.0813 -0.0745 0 0 0 0
Sort<false>/unstable_no_nulls/1024/8/manual_time -0.2900 -0.2832 1 1 1 1
Sort<false>/unstable_no_nulls/4096/8/manual_time -0.3819 -0.3739 1 1 1 1
Sort<false>/unstable_no_nulls/32768/8/manual_time -0.2767 -0.2714 1 1 1 1
Sort<false>/unstable_no_nulls/262144/8/manual_time -0.1866 -0.1840 1 1 1 1
Sort<true>/stable_no_nulls/1024/8/manual_time -0.2920 -0.2852 1 1 1 1
Sort<true>/stable_no_nulls/4096/8/manual_time -0.3818 -0.3737 1 1 1 1
Sort<true>/stable_no_nulls/32768/8/manual_time -0.2777 -0.2722 1 1 1 1
Sort<true>/stable_no_nulls/262144/8/manual_time -0.1886 -0.1859 1 1 1 1
Sort<false>/unstable/1024/8/manual_time -0.2835 -0.2778 1 1 1 1
Sort<false>/unstable/4096/8/manual_time -0.3425 -0.3365 1 1 1 1
Sort<false>/unstable/32768/8/manual_time -0.2179 -0.2149 1 1 1 1
Sort<false>/unstable/262144/8/manual_time -0.0930 -0.0922 2 2 2 2
Sort<true>/stable/1024/8/manual_time -0.2874 -0.2817 1 1 1 1
Sort<true>/stable/4096/8/manual_time -0.3490 -0.3429 1 1 1 1
Sort<true>/stable/32768/8/manual_time -0.2232 -0.2202 1 1 1 1
Sort<true>/stable/262144/8/manual_time -0.0964 -0.0955 2 2 2 2
Search/Table/4/1000000/manual_time +0.0796 +0.0560 4 4 4 4
Gather/double_coalesce_x/8192/1/manual_time +0.0751 +0.0182 8503 9141 27316 27812
Gather/double_coalesce_x/16384/1/manual_time +0.0879 +0.0246 8641 9400 27247 27916
Gather/double_coalesce_x/8192/2/manual_time +0.0635 +0.0286 16209 17238 34700 35692
Gather/double_coalesce_x/1024/4/manual_time +0.0517 +0.0308 30234 31797 49033 50543
Gather/double_coalesce_x/4096/4/manual_time +0.0568 +0.0327 30768 32517 49409 51026
Gather/double_coalesce_x/2048/8/manual_time +0.0533 +0.0376 59436 62601 78176 81117
Gather/double_coalesce_o/16384/1/manual_time +0.0525 +0.0138 8921 9389 27447 27824
ReplaceClamp/float_no_nulls/10000/manual_time +0.0502 +0.0309 41663 43753 60740 62616
AST/BM_ast_transform<int32_t, TreeType::IMBALANCED_LEFT, true, true>/100000000/1/manual_time +0.0581 +0.0579 8 9 8 9
```
github-actions
bot
added
CMake
|
Here are the significant changes to performance that the benchmark found: I am going to re-run the |
robertmaynard
added
improvement
Codecov Report
@@ Coverage Diff @@
## branch-22.02 #9912 +/- ##
================================================
- Coverage 10.49% 10.40% -0.09%
================================================
Files 119 119
Lines 20305 20503 +198
================================================
+ Hits 2130 2134 +4
- Misses 18175 18369 +194
Continue to review full report at Codecov.
|
|
After re-running the
|
|
I agree that your repeated benchmarks look fine. Two questions: 1) Are the negative standard deviations just typos? 2) Are any of the speedups you originally reported meaningful, or are those also just statistical noise? |
|
|
After running the sort benchmark for ten iterations we see consistent improvements in the following:
|
vyasr
approved these changes
Jan 4, 2022
|
@gpucibot merge |
rapids-bot bot
pushed a commit
that referenced
this pull request
Apr 5, 2022
PR #10489 updated from Thrust 1.15 to Thrust 1.16. However, this appears to be causing conflicts with other repositories -- [cuSpatial](rapidsai/cuspatial#511 (comment)) and cuGraph have reported issues where their builds are finding Thrust 1.16 from libcudf instead of Thrust 1.15 which is [currently pinned by rapids-cmake](https://github.com/rapidsai/rapids-cmake/blob/06a657281cdd83781e49afcdbb39abc491eeab17/rapids-cmake/cpm/versions.json#L26). This PR is intended to unblock local builds and CI builds for other RAPIDS packages until we are able to identify the root cause (which may be due to CMake include path orderingsrapids-cmake). Last time Thrust was updated, [rapids-cmake was updated](rapidsai/rapids-cmake#138) one day before [libcudf was updated](#9912). That may explain why we didn't notice this problem with the 1.15 update. The plan I currently have in mind is: 1. Merge this PR to roll back libcudf to Thrust 1.15 (and revert the patch for Thrust 1.16 [10577](#10577)). This will hopefully unblock CI for cugraph and cuspatial. 2. Try to work out whatever issues with CMake / include paths may exist. 3. Prepare all rapids-cmake repos for Thrust 1.16 compatibility. I've [done this for RMM already](rapidsai/rmm#1011), and I am working on [PR 4675](rapidsai/cuml#4675) to cuML now. I am planning to make the same fixes for `#include`s in cuCollections, raft, cuSpatial, and cuGraph so they will be compatible with Thrust 1.16. 4. Try to upgrade libcudf to Thrust 1.16 again (and re-apply the updated patch). If (2) has been resolved, I hope we won't see any issues in other RAPIDS libraries 5. Upgrade rapids-cmake to Thrust 1.16. Authors: - Bradley Dice (https://github.com/bdice) Approvers: - Vyas Ramasubramani (https://github.com/vyasr) - Mark Harris (https://github.com/harrism) URL: #10586
abellina
pushed a commit
to abellina/cudf
that referenced
this pull request
Apr 11, 2022
PR rapidsai#10489 updated from Thrust 1.15 to Thrust 1.16. However, this appears to be causing conflicts with other repositories -- [cuSpatial](rapidsai/cuspatial#511 (comment)) and cuGraph have reported issues where their builds are finding Thrust 1.16 from libcudf instead of Thrust 1.15 which is [currently pinned by rapids-cmake](https://github.com/rapidsai/rapids-cmake/blob/06a657281cdd83781e49afcdbb39abc491eeab17/rapids-cmake/cpm/versions.json#L26). This PR is intended to unblock local builds and CI builds for other RAPIDS packages until we are able to identify the root cause (which may be due to CMake include path orderingsrapids-cmake). Last time Thrust was updated, [rapids-cmake was updated](rapidsai/rapids-cmake#138) one day before [libcudf was updated](rapidsai#9912). That may explain why we didn't notice this problem with the 1.15 update. The plan I currently have in mind is: 1. Merge this PR to roll back libcudf to Thrust 1.15 (and revert the patch for Thrust 1.16 [10577](rapidsai#10577)). This will hopefully unblock CI for cugraph and cuspatial. 2. Try to work out whatever issues with CMake / include paths may exist. 3. Prepare all rapids-cmake repos for Thrust 1.16 compatibility. I've [done this for RMM already](rapidsai/rmm#1011), and I am working on [PR 4675](rapidsai/cuml#4675) to cuML now. I am planning to make the same fixes for `#include`s in cuCollections, raft, cuSpatial, and cuGraph so they will be compatible with Thrust 1.16. 4. Try to upgrade libcudf to Thrust 1.16 again (and re-apply the updated patch). If (2) has been resolved, I hope we won't see any issues in other RAPIDS libraries 5. Upgrade rapids-cmake to Thrust 1.16. Authors: - Bradley Dice (https://github.com/bdice) Approvers: - Vyas Ramasubramani (https://github.com/vyasr) - Mark Harris (https://github.com/harrism) URL: rapidsai#10586
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.
Compile times between 1.12 and 1.15 are comparable.
libcudf library size with 1.15 is < 1MB smaller compared to 1.12