| 1094740 | jax authors | 27 January 2021, 04:47:13 UTC | Merge pull request #5528 from google:update-pypi PiperOrigin-RevId: 354012572 | 27 January 2021, 04:47:13 UTC |
| a9e5fab | jax authors | 27 January 2021, 04:43:41 UTC | Merge pull request #5526 from google:issue5522 PiperOrigin-RevId: 354012448 | 27 January 2021, 04:43:41 UTC |
| d2ae949 | Matthew Johnson | 27 January 2021, 04:19:09 UTC | update version and changelog for pypi | 27 January 2021, 04:19:09 UTC |
| ff66c55 | Matthew Johnson | 27 January 2021, 01:25:22 UTC | add axis_env argument to make_jaxpr fixes #5522 | 27 January 2021, 04:11:05 UTC |
| 1fd1faa | jax authors | 26 January 2021, 23:00:32 UTC | Merge pull request #5523 from sharadmv:custom-gradient-tuple PiperOrigin-RevId: 353957596 | 26 January 2021, 23:00:32 UTC |
| 6061b09 | Sharad Vikram | 26 January 2021, 20:39:35 UTC | Allow `jax.custom_gradient` to return vjp with singleton return value | 26 January 2021, 21:40:23 UTC |
| 814c4ad | jax authors | 26 January 2021, 20:56:05 UTC | Merge pull request #5490 from google:xeinsum PiperOrigin-RevId: 353929996 | 26 January 2021, 20:56:05 UTC |
| e2140bf | jax authors | 26 January 2021, 20:40:50 UTC | Merge pull request #5426 from jakevdp:thinking-in-jax PiperOrigin-RevId: 353926472 | 26 January 2021, 20:40:50 UTC |
| eec2dbc | jax authors | 26 January 2021, 20:37:24 UTC | Merge pull request #5512 from google:closure-convert-docs PiperOrigin-RevId: 353926248 | 26 January 2021, 20:37:24 UTC |
| 737c162 | Matthew Johnson | 26 January 2021, 18:59:22 UTC | relax test tolerance on tpu | 26 January 2021, 20:35:43 UTC |
| 967f3ac | Jake VanderPlas | 26 January 2021, 20:08:37 UTC | Add thinking_in_jax.ipynb | 26 January 2021, 20:08:37 UTC |
| 93c61e4 | Roy Frostig | 26 January 2021, 17:44:48 UTC | import closure_convert at top module level | 26 January 2021, 17:44:48 UTC |
| f4cf710 | George Necula | 26 January 2021, 11:36:07 UTC | Add support for sharded_jit conversion to jax2tf. PiperOrigin-RevId: 353834475 | 26 January 2021, 11:36:33 UTC |
| 237f3ee | jax authors | 26 January 2021, 03:56:16 UTC | Merge pull request #5509 from danieldjohnson:all_gather_to_unbatched PiperOrigin-RevId: 353784844 | 26 January 2021, 03:56:16 UTC |
| a8fc8b3 | jax authors | 26 January 2021, 03:30:07 UTC | Merge pull request #5511 from lukaszlew:patch-2 PiperOrigin-RevId: 353781688 | 26 January 2021, 03:30:07 UTC |
| 2e19b1d | jax authors | 26 January 2021, 03:21:58 UTC | Merge pull request #5383 from zhangqiaorjc:jax_cpp PiperOrigin-RevId: 353780826 | 26 January 2021, 03:21:58 UTC |
| 7e15524 | Qiao Zhang | 12 January 2021, 22:02:32 UTC | Add example code to save JAX program and run using C++ runtime. | 26 January 2021, 03:12:26 UTC |
| 4adc436 | Roy Frostig | 26 January 2021, 01:42:46 UTC | include closure_convert in generated docs | 26 January 2021, 01:42:46 UTC |
| 9dccf56 | Łukasz Lew | 26 January 2021, 01:16:29 UTC | Clarify tracking Clarify tracing a bit and use wording that does not suggest that JAX executed python program. | 26 January 2021, 01:16:29 UTC |
| 15b95e3 | Daniel Johnson | 25 January 2021, 22:47:50 UTC | Use np.shape instead of assuming argument has a shape attr | 25 January 2021, 23:11:38 UTC |
| 84f55f8 | jax authors | 25 January 2021, 22:38:28 UTC | Merge pull request #5508 from jakevdp:fix-sphinx PiperOrigin-RevId: 353733345 | 25 January 2021, 22:38:28 UTC |
| c6a1bba | Daniel Johnson | 25 January 2021, 22:27:39 UTC | Add evaluation rule for all_gather. This should only be called when an all_gather runs on arguments that are not batch tracers, for instance when all_gather-ing a constant. | 25 January 2021, 22:27:39 UTC |
| cfe934c | Jake VanderPlas | 25 January 2021, 18:46:58 UTC | Fix some doc build warnings | 25 January 2021, 22:08:57 UTC |
| 7865043 | Daniel Johnson | 25 January 2021, 21:52:38 UTC | Improve batched collective rule for all_gather_p When an all_gather references a vmapped axis, there is a particularly simple way of implementing it: simply "forget" that the axis was mapped, and return the full array. Conveniently, this doesn't require any explicit broadcasting, and makes it possible to use out_axes=None with the results. | 25 January 2021, 21:52:38 UTC |
| f4b5ff9 | jax authors | 25 January 2021, 21:51:12 UTC | Merge pull request #5488 from jakevdp:x64-contextmanager PiperOrigin-RevId: 353722592 | 25 January 2021, 21:51:12 UTC |
| b545461 | Jake VanderPlas | 25 January 2021, 21:23:15 UTC | Add experimental context manager to enable/disable X64 mode | 25 January 2021, 21:23:15 UTC |
| 136358c | jax authors | 25 January 2021, 18:03:17 UTC | Merge pull request #5480 from pschuh:wrap-outs PiperOrigin-RevId: 353670528 | 25 January 2021, 18:03:17 UTC |
| 13b921b | jax authors | 25 January 2021, 15:16:25 UTC | Merge pull request #5465 from apaszke:xmap-replica-nesting PiperOrigin-RevId: 353640436 | 25 January 2021, 15:16:25 UTC |
| 137321f | Adam Paszke | 19 January 2021, 14:18:53 UTC | Restore support for nested xmaps when using the (standard) replica lowering Fortunately this wasn't too difficult, as most of the code was already there. The biggest issues were a lack of axis name substitution and inability to handle multiple resource assignments for a single logical axis. I'm planning to add a more comprehensive test suite for this, but I'd like to wait for the pdot test PR (#5459) to land first. It contains a bunch of utilities that would come in handy. | 25 January 2021, 13:39:49 UTC |
| 802c773 | jax authors | 25 January 2021, 03:58:49 UTC | Merge pull request #5502 from jackd:testNorm-fix PiperOrigin-RevId: 353565580 | 25 January 2021, 03:58:49 UTC |
| f02f775 | jax authors | 25 January 2021, 03:55:15 UTC | Merge pull request #5500 from terhorst:poisson-cdf PiperOrigin-RevId: 353565572 | 25 January 2021, 03:55:15 UTC |
| f288cf4 | Dominic Jack | 25 January 2021, 02:56:11 UTC | fixed testNorm | 25 January 2021, 02:56:11 UTC |
| 1524b82 | Jonathan Terhorst | 24 January 2021, 16:15:31 UTC | add support for scipy.stats.poisson.cdf | 24 January 2021, 16:15:31 UTC |
| b0f5ef4 | jax authors | 23 January 2021, 22:44:30 UTC | Merge pull request #5492 from google:revive-revive-leak-checker PiperOrigin-RevId: 353440404 | 23 January 2021, 22:44:30 UTC |
| a7bfebe | Matthew Johnson | 23 January 2021, 22:17:22 UTC | improve leak checker flag description | 23 January 2021, 22:17:22 UTC |
| 9787894 | Matthew Johnson | 22 January 2021, 05:29:09 UTC | refactor batching transform logic, fix leak checks See PR description in #5492 for details. Co-authored-by: Peter Hawkins <phawkins@google.com> | 23 January 2021, 04:17:03 UTC |
| 203af45 | Matthew Johnson | 20 January 2021, 02:38:53 UTC | revive the leak checker, as a debug mode Co-authored-by: James Bradbury <jekbradbury@google.com> | 23 January 2021, 02:31:00 UTC |
| bc3cd12 | jax authors | 22 January 2021, 19:34:43 UTC | Merge pull request #5495 from hawkinsp:tokens PiperOrigin-RevId: 353283976 | 22 January 2021, 19:34:43 UTC |
| dd34d48 | Peter Hawkins | 22 January 2021, 15:57:33 UTC | Fix exception when tokens are used in AD. | 22 January 2021, 16:00:31 UTC |
| 9ccfc9f | jax authors | 22 January 2021, 00:53:06 UTC | Merge pull request #5478 from jakevdp:fix-docstrings PiperOrigin-RevId: 353133338 | 22 January 2021, 00:53:06 UTC |
| 7c67ec1 | jax authors | 21 January 2021, 23:24:31 UTC | Merge pull request #5477 from google:more-escaped-tracer-error-message-tweaks PiperOrigin-RevId: 353115571 | 21 January 2021, 23:24:31 UTC |
| 6d2f832 | Matthew Johnson | 21 January 2021, 19:09:59 UTC | add xeinsum, an einsum for xmap (& einsum easter egg) Co-authored-by: Adam Paszke <apaszke@google.com> | 21 January 2021, 22:47:35 UTC |
| 4a7b7df | jax authors | 21 January 2021, 22:38:16 UTC | Merge pull request #5491 from google:pdot-tests3 PiperOrigin-RevId: 353105977 | 21 January 2021, 22:38:16 UTC |
| c02d804 | Matthew Johnson | 18 January 2021, 18:02:24 UTC | add systematic pdot tests, utility functions Run lots of tests with e.g. ``` env JAX_NUM_GENERATED_CASES=1000 python tests/xmap_test.py PDotTests ``` | 21 January 2021, 22:06:30 UTC |
| 7925a3a | jax authors | 21 January 2021, 18:00:01 UTC | Merge pull request #5486 from jakevdp:finfo PiperOrigin-RevId: 353042636 | 21 January 2021, 18:00:01 UTC |
| 60a87fd | jax authors | 21 January 2021, 17:14:34 UTC | Merge pull request #5419 from jakevdp:int-power PiperOrigin-RevId: 353033127 | 21 January 2021, 17:14:34 UTC |
| f726709 | jax authors | 21 January 2021, 16:11:34 UTC | Merge pull request #5446 from gsp-27:master PiperOrigin-RevId: 353021985 | 21 January 2021, 16:11:34 UTC |
| fc258c3 | Jake VanderPlas | 21 January 2021, 16:05:36 UTC | Make jnp.finfo a class and support bfloat16 | 21 January 2021, 16:05:36 UTC |
| 988949f | Peter Hawkins | 21 January 2021, 14:56:00 UTC | Move jax.bzl into correct location. Unbreaks build. PiperOrigin-RevId: 353009344 | 21 January 2021, 14:56:23 UTC |
| 32de6ff | Jean-Baptiste Lespiau | 21 January 2021, 09:14:56 UTC | Replace None with an object NoSharding. This is to make the change to a C++ ShardingSpec easier. See also https://github.com/google/jax/pull/5444 PiperOrigin-RevId: 352965689 | 21 January 2021, 09:15:25 UTC |
| a900620 | jax authors | 21 January 2021, 05:15:05 UTC | Merge pull request #5482 from google:contributing-squash PiperOrigin-RevId: 352937260 | 21 January 2021, 05:15:05 UTC |
| df273c0 | Parker Schuh | 20 January 2021, 21:26:13 UTC | Refactor how the result processing code wraps the indexing information. Convert `array_result_handler` and `avals_to_results_handler` to use functools.partial and a callable class so that the sharding information can be introspected instead of opaque lambdas. | 21 January 2021, 03:01:04 UTC |
| 186c973 | Gaurav Pathak | 21 January 2021, 02:47:18 UTC | Added test for python scalar | 21 January 2021, 02:47:18 UTC |
| 160dfd3 | Peter Hawkins | 21 January 2021, 01:35:30 UTC | Revert import path changes to examples/ and benchmarks/ PiperOrigin-RevId: 352911869 | 21 January 2021, 01:35:55 UTC |
| 297250d | Roy Frostig | 21 January 2021, 00:32:21 UTC | add a request for PR-squashing in contributing.md | 21 January 2021, 00:37:57 UTC |
| ffa05d1 | jax authors | 20 January 2021, 23:47:03 UTC | Merge pull request #5481 from jakevdp:doc-formatting PiperOrigin-RevId: 352892385 | 20 January 2021, 23:47:03 UTC |
| 84e91d5 | Matthew Johnson | 20 January 2021, 18:56:13 UTC | add transformed fun src info to escaped tracer err This change adds to the error message when we hit an escaped tracer. In particular, it adds source info for the function that was transformed. This change currently only applies to escaped `DynamicJaxprTracer`s (arising from `jit`, `pmap`, `scan`, and other staging functions) and not other traces. A natural follow-up would be to attach this information to other traces. Co-authored-by: Lena Martens <lenamartens@google.com> | 20 January 2021, 23:30:37 UTC |
| a0b12bb | Jake VanderPlas | 20 January 2021, 22:38:19 UTC | DOC: fix minor formatting issues | 20 January 2021, 22:38:19 UTC |
| 9bdc2ec | Peter Hawkins | 20 January 2021, 22:05:54 UTC | Consolidate build macros into a single jax.bzl file. PiperOrigin-RevId: 352871429 | 20 January 2021, 22:06:22 UTC |
| fe9a3e7 | Jake VanderPlas | 20 January 2021, 21:40:26 UTC | util._wraps: correctly handle initial line indentation | 20 January 2021, 21:40:26 UTC |
| 929a684 | Peter Hawkins | 20 January 2021, 20:43:00 UTC | Small cleanups to dependency structure. PiperOrigin-RevId: 352853244 | 20 January 2021, 20:43:28 UTC |
| e217a7b | jax authors | 20 January 2021, 19:43:47 UTC | Merge pull request #5473 from hawkinsp:docs2 PiperOrigin-RevId: 352839545 | 20 January 2021, 19:43:47 UTC |
| 2991b04 | Jake VanderPlas | 14 January 2021, 19:11:04 UTC | jnp.power: use integer_power path in all applicable cases | 20 January 2021, 17:21:57 UTC |
| 62e89cb | jax authors | 20 January 2021, 16:45:06 UTC | Merge pull request #5213 from malmaud:changelist/346683050 PiperOrigin-RevId: 352800932 | 20 January 2021, 16:45:06 UTC |
| 8420ee2 | jax authors | 20 January 2021, 16:21:02 UTC | Merge pull request #5472 from hawkinsp:build PiperOrigin-RevId: 352797345 | 20 January 2021, 16:21:02 UTC |
| 4fdd029 | Peter Hawkins | 20 January 2021, 14:59:35 UTC | Clarify the CuDNN versions expected by jaxlib wheels. | 20 January 2021, 15:00:19 UTC |
| d585638 | jax authors | 20 January 2021, 14:46:44 UTC | Merge pull request #5470 from inailuig:fix-rocm-build PiperOrigin-RevId: 352783431 | 20 January 2021, 14:46:44 UTC |
| 3bec4b3 | Peter Hawkins | 20 January 2021, 14:38:55 UTC | Fix ABSL build dependencies of //jaxlib:handle_pool | 20 January 2021, 14:38:55 UTC |
| 5ced12f | jax authors | 20 January 2021, 11:03:40 UTC | Merge pull request #5455 from chr1sj0nes:changelist/351545649 PiperOrigin-RevId: 352756095 | 20 January 2021, 11:03:40 UTC |
| 57405b0 | Clemens Giuliani | 20 January 2021, 09:49:29 UTC | fix building on ROCm | 20 January 2021, 09:50:04 UTC |
| 619bfd5 | jax authors | 20 January 2021, 05:05:18 UTC | Merge pull request #5468 from google:issue5440 PiperOrigin-RevId: 352715439 | 20 January 2021, 05:05:18 UTC |
| 0f70451 | Matthew Johnson | 20 January 2021, 03:08:23 UTC | add BatchTrace.process_custom_vjp_call It was an oversight not to include this! Notice we have BatchTrace.process_custom_jvp_call. In fact, we can use the same function! We just needed the simplest possible post-process-call which just peels and packages. fixes #5440 Co-authored-by: Roy Frostig <frostig@google.com> | 20 January 2021, 03:08:23 UTC |
| 34e798f | jax authors | 20 January 2021, 01:01:47 UTC | Merge pull request #5466 from jakevdp:fix-integer-pow PiperOrigin-RevId: 352685155 | 20 January 2021, 01:01:47 UTC |
| 0a89fc8 | Jake VanderPlas | 19 January 2021, 23:42:40 UTC | integer_pow: fix jvp rule for y=0 | 19 January 2021, 23:42:40 UTC |
| 7acf521 | Jake VanderPlas | 19 January 2021, 23:42:18 UTC | integer_pow: fix translation rule for y=0 | 19 January 2021, 23:42:18 UTC |
| d641e6c | Matthew Johnson | 19 January 2021, 23:01:30 UTC | Revert part of #5319: don't use `traceback_util.path_starts` in `source_info_util.user_frame`. The issue was that it incurs filesystem accesses while we're lowering to XLA, which can cause a significant slowdown. PiperOrigin-RevId: 352662434 | 19 January 2021, 23:01:52 UTC |
| 9076008 | Jake VanderPlas | 19 January 2021, 19:36:39 UTC | lax.integer_pow(): always bind the primitive | 19 January 2021, 19:36:39 UTC |
| c0c4843 | Jonathan Malmaud | 10 December 2020, 02:29:40 UTC | Add support for 'preferred_element_type' keyword arg in `dot` and `dot_general`. XLA recently added support for this parameter to xops.DotGeneral. It's an optional parameter that controls the accumulation type used by the dot operation. This is useful for eg quantized ANNs, where you might want to do matrix multiples with int8 tensors and get back an int32 tensor instead of an int8 tensor that suffers from severe overflow. Note it's not sufficient in this case to cast the inputs to 'dot' to int32 beforehand and rely on the default output dtype inference, since backend devices might have an accelerated path for int8*int8->int32 matmuls and we want that explicitly represented in the XLA. Note because XLA still doesn't support integer dots on the CPU backend, that use case can't tested with a CPU-only test at the moment. | 19 January 2021, 18:56:46 UTC |
| 2d3f33d | Gaurav Pathak | 19 January 2021, 14:49:28 UTC | fix float32 | 19 January 2021, 14:49:28 UTC |
| 9260f2c | Gaurav Pathak | 19 January 2021, 14:12:11 UTC | regex check for each test case | 19 January 2021, 14:12:11 UTC |
| b4900ee | jax authors | 19 January 2021, 12:16:28 UTC | Merge pull request #5453 from apaszke:xmap-vmap PiperOrigin-RevId: 352537923 | 19 January 2021, 12:16:28 UTC |
| 4b48c7f | Chris Jones | 13 January 2021, 10:33:03 UTC | Use XLA AllGather op for GPU (attempt 2). This is an expansion of the first, rolled-back attempt (https://github.com/google/jax/pull/5260), this time with auto-diff and batching rules that some users are relying on. My benchmarks suggest a speed-up of ~2-2.5x for larger inputs. | 19 January 2021, 11:16:25 UTC |
| 0b57a16 | jax authors | 19 January 2021, 05:02:35 UTC | Merge pull request #5462 from google:error-message-tweak2 PiperOrigin-RevId: 352488512 | 19 January 2021, 05:02:35 UTC |
| 47f7cd4 | Matthew Johnson | 19 January 2021, 04:37:12 UTC | avoid printing double periods in error messages | 19 January 2021, 04:37:12 UTC |
| bd9ac93 | Gaurav Pathak | 19 January 2021, 04:21:33 UTC | Added changes requested | 19 January 2021, 04:21:33 UTC |
| 772a6da | jax authors | 19 January 2021, 03:56:19 UTC | Merge pull request #5319 from google:source-line-info-on-escaped-tracer-errors PiperOrigin-RevId: 352482449 | 19 January 2021, 03:56:19 UTC |
| 886b26f | Matthew Johnson | 05 January 2021, 22:52:54 UTC | add source line info to more escaped tracer errors This extra source info is still only on jaxpr staging tracers, but those seem to be the most common culprits. I moved the `_line_info` attribute to the base Tracer class in core.py in anticipation of populating it for more traces than just DynamicJaxprTrace, but I'll leave that extension to follow-up. I adapted the main escaped tracer error messages in core.py, and also slightly generalized and debugged source_info_util functions (thanks for explaining the path prefix bug, @froystig !). | 19 January 2021, 03:00:04 UTC |
| c287180 | Adam Paszke | 18 January 2021, 18:40:46 UTC | Update jax/experimental/maps.py Co-authored-by: Matthew Johnson <mattjj@google.com> | 18 January 2021, 18:40:46 UTC |
| 6d2b307 | Adam Paszke | 18 January 2021, 10:42:06 UTC | Make it possible to vmap xmapped functions Or perhaps more importantly make it possible to nest xmaps that don't specify any `axis_resources`. The math is a little tricky, so I've added a fairly strong test that enumerates a wide range of potential ways of interleaving vmapped and xmapped axes in both inputs and the output. Thanks to that, I've actually caught one very subtle bug in the dynamic tracing rule for xmap (sorting by dimension names instead of positional axes). | 18 January 2021, 15:35:47 UTC |
| eeb5c42 | jax authors | 18 January 2021, 14:12:07 UTC | Merge pull request #5452 from gnecula:tf_limit2 PiperOrigin-RevId: 352400540 | 18 January 2021, 14:12:07 UTC |
| d77a941 | George Necula | 18 January 2021, 13:19:01 UTC | [jax2tf] Reflect in the limitations that add is now implemented for uint32 in TF | 18 January 2021, 13:19:01 UTC |
| 040d268 | jax authors | 18 January 2021, 13:03:27 UTC | Merge pull request #5451 from gnecula:tf_jit_compile PiperOrigin-RevId: 352392391 | 18 January 2021, 13:03:27 UTC |
| 04628c9 | jax authors | 18 January 2021, 12:41:50 UTC | Merge pull request #5435 from apaszke:xmap-parameterized PiperOrigin-RevId: 352389890 | 18 January 2021, 12:41:50 UTC |
| 6d2b976 | George Necula | 18 January 2021, 12:41:42 UTC | [jax2tf] Start using jit_compile instead of the deprecated experimental_compile | 18 January 2021, 12:41:42 UTC |
| 449c2bc | jax authors | 18 January 2021, 11:44:38 UTC | Merge pull request #4804 from marcvanzee:patch-1 PiperOrigin-RevId: 352382502 | 18 January 2021, 11:44:38 UTC |
| 6bf6349 | George Necula | 18 January 2021, 11:37:55 UTC | Copybara import of the project: -- 781492e0120ec915f9fdc83479884908f59d113d by George Necula <gcnecula@gmail.com>: [jax2tf] Update limitations Some bugs were fixed on the TF-side, and we can remove some limitations. COPYBARA_INTEGRATE_REVIEW=https://github.com/google/jax/pull/5449 from gnecula:jax2tf_limit 781492e0120ec915f9fdc83479884908f59d113d PiperOrigin-RevId: 352381535 | 18 January 2021, 11:38:22 UTC |
| a1e182c | Marc van Zee | 18 January 2021, 09:29:16 UTC | Updates comment | 18 January 2021, 10:39:37 UTC |
| 166e129 | Adam Paszke | 11 December 2020, 11:57:22 UTC | Clean up xmap tests Move `pdot` tests to a separate class. Automatically run all regular xmap tests with both the `pmap`-style lowering and `sharded_jit`-style lowering. | 18 January 2021, 10:21:32 UTC |
| 67b5af9 | George Necula | 18 January 2021, 09:59:25 UTC | Copybara import of the project: -- 9be685946252edc67c2c28261b100b9aee68614a by George Necula <gcnecula@gmail.com>: Change the translation rule for lax.nextafter_p to ensure broadcasting during translation. Previously, this was the only binary arithmetic primitive that did not have broadcasting during translation. Trying to use it with non-equal shapes resulted in the error: ``` RuntimeError: Internal: RET_CHECK failure (external/org_tensorflow/tensorflow/compiler/xla/client/xla_builder.cc:748) non_scalar_shape.value().dimensions() == shape->dimensions() Unimplemented implicit broadcast.: This is a bug in JAX's shape-checking rules; please report it! ``` COPYBARA_INTEGRATE_REVIEW=https://github.com/google/jax/pull/5448 from gnecula:nextafter 9be685946252edc67c2c28261b100b9aee68614a PiperOrigin-RevId: 352367039 | 18 January 2021, 09:59:55 UTC |
| 4dd50b8 | Gaurav Pathak | 18 January 2021, 02:16:28 UTC | removed whitespace causing flake failing | 18 January 2021, 02:16:28 UTC |
