mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-21 07:19:46 +08:00
Code Issues Packages Projects Releases Wiki Activity
10,580 Commits 11 Branches 89 Tags 162 MiB
c5019f722b
Commit Graph

2155 Commits

Author SHA1 Message Date
Michael Tesch
c5019f722b Use pade for matrix exponential also for complex values. 2019-05-08 17:04:55 +02:00
Rasmus Larsen
e92486b8c3 Merged in rmlarsen/eigen (pull request PR-643) 
Make Eigen build with cuda 10 and clang.

Approved-by: Justin Lebar <justin.lebar@gmail.com>
 2019-05-20 17:02:39 +00:00
Eugene Zhulenev
01654d97fa Prevent potential division by zero in TensorExecutor 2019-05-17 14:02:25 -07:00
Eugene Zhulenev
96a276803c Always evaluate Tensor expressions with broadcasting via tiled evaluation code path 2019-05-16 16:15:45 -07:00
Rasmus Munk Larsen
ab0a30e429 Make Eigen build with cuda 10 and clang. 2019-05-15 13:32:15 -07:00
Rasmus Munk Larsen
e5ac8cbd7a A) fix deadlocks in thread pool caused by EventCount 
This fixed 2 deadlocks caused by sloppiness in the EventCount logic.
Both most likely were introduced by cl/236729920 which includes the new EventCount algorithm:
01da8caf00

bug #1 (Prewait):
Prewait must not consume existing signals.
Consider the following scenario.
There are 2 thread pool threads (1 and 2) and 1 external thread (3). RunQueue is empty.
Thread 1 checks the queue, calls Prewait, checks RunQueue again and now is going to call CommitWait.
Thread 2 checks the queue and now is going to call Prewait.
Thread 3 submits 2 tasks, EventCount signals is set to 1 because only 1 waiter is registered the second signal is discarded).
Now thread 2 resumes and calls Prewait and takes away the signal.
Thread 1 resumes and calls CommitWait, there are no pending signals anymore, so it blocks.
As the result we have 2 tasks, but only 1 thread is running.

bug #2 (CancelWait):
CancelWait must not take away a signal if it's not sure that the signal was meant for this thread.
When one thread blocks and another submits a new task concurrently, the EventCount protocol guarantees only the following properties (similar to the Dekker's algorithm):
(a) the registered waiter notices presence of the new task and does not block
(b) the signaler notices presence of the waiters and wakes it
(c) both the waiter notices presence of the new task and signaler notices presence of the waiter
[it's only that both of them do not notice each other must not be possible, because it would lead to a deadlock]
CancelWait is called for cases (a) and (c). For case (c) it is OK to take the notification signal away, but it's not OK for (a) because nobody queued a signals for us and we take away a signal meant for somebody else.
Consider:
Thread 1 calls Prewait, checks RunQueue, it's empty, now it's going to call CommitWait.
Thread 3 submits 2 tasks, EventCount signals is set to 1 because only 1 waiter is registered the second signal is discarded).
Thread 2 calls Prewait, checks RunQueue, discovers the tasks, calls CancelWait and consumes the pending signal (meant for thread 1).
Now Thread 1 resumes and calls CommitWait, since there are no signals it blocks.
As the result we have 2 tasks, but only 1 thread is running.

Both deadlocks are only a problem if the tasks require parallelism. Most computational tasks do not require parallelism, i.e. a single thread will run task 1, finish it and then dequeue and run task 2.

This fix undoes some of the sloppiness in the EventCount that was meant to reduce CPU consumption by idle threads, because we now have more threads running in these corner cases. But we still don't have pthread_yield's and maybe the strictness introduced by this change will actually help to reduce tail latency because we will have threads running when we actually need them running.



B) fix deadlock in thread pool caused by RunQueue

This fixed a deadlock caused by sloppiness in the RunQueue logic.
Most likely this was introduced with the non-blocking thread pool.
The deadlock only affects workloads that require parallelism.
Most computational tasks don't require parallelism.

PopBack must not fail spuriously. If it does, it can effectively lead to single thread consuming several wake up signals.
Consider 2 worker threads are blocked.
External thread submits a task. One of the threads is woken.
It tries to steal the task, but fails due to a spurious failure in PopBack (external thread submits another task and holds the lock).
The thread executes blocking protocol again (it won't block because NonEmptyQueueIndex is precise and the thread will discover pending work, but it has called PrepareWait).
Now external thread submits another task and signals EventCount again.
The signal is consumed by the first thread again. But now we have 2 tasks pending but only 1 worker thread running.

It may be possible to fix this in a different way: make EventCount::CancelWait forward wakeup signal to a blocked thread rather then consuming it. But this looks more complex and I am not 100% that it will fix the bug.
It's also possible to have 2 versions of PopBack: one will do try_to_lock and another won't. Then worker threads could first opportunistically check all queues with try_to_lock, and only use the blocking version before blocking. But let's first fix the bug with the simpler change.
 2019-05-08 10:16:46 -07:00
Christoph Hertzberg
e6667a7060 Fix stupid shadow-warnings (with old clang versions) 2019-05-07 18:32:19 +02:00
Christoph Hertzberg
e54dc24d62 Restore C++03 compatibility 2019-05-07 18:30:44 +02:00
Rasmus Larsen
ac50afaffa Merged in ezhulenev/eigen-01 (pull request PR-633) 
Check if gpu_assert was overridden in TensorGpuHipCudaDefines
 2019-04-29 16:29:35 +00:00
Eugene Zhulenev
01d7e6ee9b Check if gpu_assert was overridden in TensorGpuHipCudaDefines 2019-04-25 11:19:17 -07:00
Eugene Zhulenev
8ead5bb3d8 Fix doxygen warnings to enable statis code analysis 2019-04-24 12:42:28 -07:00
Rasmus Munk Larsen
144ca33321 Remove deprecation annotation from typedef Eigen::Index Index, as it would generate too many build warnings. 2019-04-24 08:50:07 -07:00
Eugene Zhulenev
a7b7f3ca8a Add missing EIGEN_DEPRECATED annotations to deprecated functions and fix few other doxygen warnings 2019-04-23 17:23:19 -07:00
Anuj Rawat
8c7a6feb8e Adding lowlevel APIs for optimized RHS packet load in TensorFlow 
SpatialConvolution

Low-level APIs are added in order to optimized packet load in gemm_pack_rhs
in TensorFlow SpatialConvolution. The optimization is for scenario when a
packet is split across 2 adjacent columns. In this case we read it as two
'partial' packets and then merge these into 1. Currently this only works for
Packet16f (AVX512) and Packet8f (AVX2). We plan to add this for other
packet types (such as Packet8d) also.

This optimization shows significant speedup in SpatialConvolution with
certain parameters. Some examples are below.

Benchmark parameters are specified as:
Batch size, Input dim, Depth, Num of filters, Filter dim

Speedup numbers are specified for number of threads 1, 2, 4, 8, 16.

AVX512:

Parameters                  | Speedup (Num of threads: 1, 2, 4, 8, 16)
----------------------------|------------------------------------------
128,   24x24,  3, 64,   5x5 |2.18X, 2.13X, 1.73X, 1.64X, 1.66X
128,   24x24,  1, 64,   8x8 |2.00X, 1.98X, 1.93X, 1.91X, 1.91X
 32,   24x24,  3, 64,   5x5 |2.26X, 2.14X, 2.17X, 2.22X, 2.33X
128,   24x24,  3, 64,   3x3 |1.51X, 1.45X, 1.45X, 1.67X, 1.57X
 32,   14x14, 24, 64,   5x5 |1.21X, 1.19X, 1.16X, 1.70X, 1.17X
128, 128x128,  3, 96, 11x11 |2.17X, 2.18X, 2.19X, 2.20X, 2.18X

AVX2:

Parameters                  | Speedup (Num of threads: 1, 2, 4, 8, 16)
----------------------------|------------------------------------------
128,   24x24,  3, 64,   5x5 | 1.66X, 1.65X, 1.61X, 1.56X, 1.49X
 32,   24x24,  3, 64,   5x5 | 1.71X, 1.63X, 1.77X, 1.58X, 1.68X
128,   24x24,  1, 64,   5x5 | 1.44X, 1.40X, 1.38X, 1.37X, 1.33X
128,   24x24,  3, 64,   3x3 | 1.68X, 1.63X, 1.58X, 1.56X, 1.62X
128, 128x128,  3, 96, 11x11 | 1.36X, 1.36X, 1.37X, 1.37X, 1.37X

In the higher level benchmark cifar10, we observe a runtime improvement
of around 6% for AVX512 on Intel Skylake server (8 cores).

On lower level PackRhs micro-benchmarks specified in TensorFlow
tensorflow/core/kernels/eigen_spatial_convolutions_test.cc, we observe
the following runtime numbers:

AVX512:

Parameters                                                     | Runtime without patch (ns) | Runtime with patch (ns) | Speedup
---------------------------------------------------------------|----------------------------|-------------------------|---------
BM_RHS_NAME(PackRhs, 128, 24, 24, 3, 64, 5, 5, 1, 1, 256, 56)  |  41350                     | 15073                   | 2.74X
BM_RHS_NAME(PackRhs, 32, 64, 64, 32, 64, 5, 5, 1, 1, 256, 56)  |   7277                     |  7341                   | 0.99X
BM_RHS_NAME(PackRhs, 32, 64, 64, 32, 64, 5, 5, 2, 2, 256, 56)  |   8675                     |  8681                   | 1.00X
BM_RHS_NAME(PackRhs, 32, 64, 64, 30, 64, 5, 5, 1, 1, 256, 56)  |  24155                     | 16079                   | 1.50X
BM_RHS_NAME(PackRhs, 32, 64, 64, 30, 64, 5, 5, 2, 2, 256, 56)  |  25052                     | 17152                   | 1.46X
BM_RHS_NAME(PackRhs, 32, 256, 256, 4, 16, 8, 8, 1, 1, 256, 56) |  18269                     | 18345                   | 1.00X
BM_RHS_NAME(PackRhs, 32, 256, 256, 4, 16, 8, 8, 2, 4, 256, 56) |  19468                     | 19872                   | 0.98X
BM_RHS_NAME(PackRhs, 32, 64, 64, 4, 16, 3, 3, 1, 1, 36, 432)   | 156060                     | 42432                   | 3.68X
BM_RHS_NAME(PackRhs, 32, 64, 64, 4, 16, 3, 3, 2, 2, 36, 432)   | 132701                     | 36944                   | 3.59X

AVX2:

Parameters                                                     | Runtime without patch (ns) | Runtime with patch (ns) | Speedup
---------------------------------------------------------------|----------------------------|-------------------------|---------
BM_RHS_NAME(PackRhs, 128, 24, 24, 3, 64, 5, 5, 1, 1, 256, 56)  | 26233                      | 12393                   | 2.12X
BM_RHS_NAME(PackRhs, 32, 64, 64, 32, 64, 5, 5, 1, 1, 256, 56)  |  6091                      |  6062                   | 1.00X
BM_RHS_NAME(PackRhs, 32, 64, 64, 32, 64, 5, 5, 2, 2, 256, 56)  |  7427                      |  7408                   | 1.00X
BM_RHS_NAME(PackRhs, 32, 64, 64, 30, 64, 5, 5, 1, 1, 256, 56)  | 23453                      | 20826                   | 1.13X
BM_RHS_NAME(PackRhs, 32, 64, 64, 30, 64, 5, 5, 2, 2, 256, 56)  | 23167                      | 22091                   | 1.09X
BM_RHS_NAME(PackRhs, 32, 256, 256, 4, 16, 8, 8, 1, 1, 256, 56) | 23422                      | 23682                   | 0.99X
BM_RHS_NAME(PackRhs, 32, 256, 256, 4, 16, 8, 8, 2, 4, 256, 56) | 23165                      | 23663                   | 0.98X
BM_RHS_NAME(PackRhs, 32, 64, 64, 4, 16, 3, 3, 1, 1, 36, 432)   | 72689                      | 44969                   | 1.62X
BM_RHS_NAME(PackRhs, 32, 64, 64, 4, 16, 3, 3, 2, 2, 36, 432)   | 61732                      | 39779                   | 1.55X

All benchmarks on Intel Skylake server with 8 cores.
 2019-04-20 06:46:43 +00:00
Rasmus Munk Larsen
039ee52125 Tweak cost model for tensor contraction when parallelizing over the inner dimension. 
https://bitbucket.org/snippets/rmlarsen/MexxLo
 2019-04-12 13:35:10 -07:00
Jonathon Koyle
9a3f06d836 Update TheadPoolDevice example to include ThreadPool creation and passing pointer into constructor. 2019-04-10 10:02:33 -06:00
Deven Desai
66a885b61e adding EIGEN_DEVICE_FUNC to the recently added TensorContractionKernel constructor. Not having the EIGEN_DEVICE_FUNC attribute on it was leading to compiler errors when compiling Eigen in the ROCm/HIP path 2019-04-08 13:45:08 +00:00
Eugene Zhulenev
629ddebd15 Add missing semicolon 2019-04-02 15:04:26 -07:00
Eugene Zhulenev
4e2f6de1a8 Add support for custom packed Lhs/Rhs blocks in tensor contractions 2019-04-01 11:47:31 -07:00
Deven Desai
2dbea5510f Merged eigen/eigen into default 2019-03-19 16:52:38 -04:00
David Tellenbach
bd9c2ae3fd Fix include guard comments 2019-03-15 15:29:17 +01:00
Eugene Zhulenev
001f10e3c9 Fix segfaults with cuda compilation 2019-03-11 09:43:33 -07:00
Eugene Zhulenev
899c16fa2c Fix a bug in TensorGenerator for 1d tensors 2019-03-11 09:42:01 -07:00
Eugene Zhulenev
0f8bfff23d Fix a data race in NonBlockingThreadPool 2019-03-11 09:38:44 -07:00
Gael Guennebaud
2df4f00246 Change license from LGPL to MPL2 with agreement from David Harmon. 2019-03-07 18:17:10 +01:00
Rasmus Munk Larsen
3c3f639fe2 Merge. 2019-03-06 11:54:30 -08:00
Rasmus Munk Larsen
f4ec8edea8 Add macro EIGEN_AVOID_THREAD_LOCAL to make it possible to manually disable the use of thread_local. 2019-03-06 11:52:04 -08:00
Rasmus Munk Larsen
41cdc370d0 Fix placement of "#if defined(EIGEN_GPUCC)" guard region. 
Found with -Wundefined-func-template.

Author: tkoeppe@google.com
 2019-03-06 11:42:22 -08:00
Rasmus Munk Larsen
cc407c9d4d Fix placement of "#if defined(EIGEN_GPUCC)" guard region. 
Found with -Wundefined-func-template.

Author: tkoeppe@google.com
 2019-03-06 11:40:06 -08:00
Eugene Zhulenev
1bc2a0a57c Add missing return to NonBlockingThreadPool::LocalSteal 2019-03-06 10:49:49 -08:00
Eugene Zhulenev
4e4dcd9026 Remove redundant steal loop 2019-03-06 10:39:07 -08:00
Eugene Zhulenev
25abaa2e41 Check that inner block dimension is continuous 2019-03-05 17:34:35 -08:00
Eugene Zhulenev
5d9a6686ed Block evaluation for TensorGeneratorOp 2019-03-05 16:35:21 -08:00
Eugene Zhulenev
a407e022e6 Tune tensor contraction threadpool heuristics 2019-03-05 14:19:59 -08:00
Eugene Zhulenev
56c6373f82 Add an extra check for the RunQueue size estimate 2019-03-05 11:51:26 -08:00
Eugene Zhulenev
efb5080d31 Do not initialize invalid fast_strides in TensorGeneratorOp 2019-03-04 16:58:49 -08:00
Eugene Zhulenev
b95941e5c2 Add tiled evaluation for TensorForcedEvalOp 2019-03-04 16:02:22 -08:00
Eugene Zhulenev
694084ecbd Use fast divisors in TensorGeneratorOp 2019-03-04 11:10:21 -08:00
Rasmus Munk Larsen
cf4a1c81fa Fix specialization for conjugate on non-complex types in TensorBase.h. 2019-03-01 14:21:09 -08:00
Rasmus Munk Larsen
6560692c67 Improve EventCount used by the non-blocking threadpool. 
The current algorithm requires threads to commit/cancel waiting in order
they called Prewait. Spinning caused by that serialization can consume
lots of CPU time on some workloads. Restructure the algorithm to not
require that serialization and remove spin waits from Commit/CancelWait.
Note: this reduces max number of threads from 2^16 to 2^14 to leave
more space for ABA counter (which is now 22 bits).
Implementation details are explained in comments.
 2019-02-22 13:56:26 -08:00
Gael Guennebaud
9ac1634fdf Fix conversion warnings 2019-02-19 21:59:53 +01:00
Rasmus Munk Larsen
071629a440 Fix incorrect value of NumDimensions in TensorContraction traits. 
Reported here: #1671
 2019-02-19 10:49:54 -08:00
Rasmus Larsen
efeabee445 Merged in ezhulenev/eigen-01 (pull request PR-590) 
Do not generate no-op cast() and conjugate() expressions
 2019-02-14 21:16:12 +00:00
Eugene Zhulenev
7b837559a7 Fix signed-unsigned return in RuqQueue 2019-02-14 10:40:21 -08:00
Eugene Zhulenev
f0d42d2265 Fix signed-unsigned comparison warning in RunQueue 2019-02-14 10:27:28 -08:00
Eugene Zhulenev
106ba7bb1a Do not generate no-op cast() and conjugate() expressions 2019-02-14 09:51:51 -08:00
Eugene Zhulenev
8c2f30c790 Speedup Tensor ThreadPool RunQueu::Empty() 2019-02-13 10:20:53 -08:00
Eugene Zhulenev
21eb97d3e0 Add PacketConv implementation for non-vectorizable src expressions 2019-02-08 15:47:25 -08:00
Eugene Zhulenev
1e36166ed1 Optimize TensorConversion evaluator: do not convert same type 2019-02-08 15:13:24 -08:00
Steven Peters
953ca5ba2f Spline.h: fix spelling "spang" -> "span" 2019-02-08 06:23:24 +00:00