Repository: systemml
Updated Branches:
refs/heads/master 3ca053535 -> 8df0697e0
[MINOR] [SYSTEMML-445] GPU bugfix for metadata checking
This commit contains two fixes:
- First fix ensures that the method allocating output matrix (for eg:
getSparseMatrixOutputForGPUInstruction,
http://git-wip-us.apache.org/repos/asf/systemml/blob/628ffad1/src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java
--
diff --git
a/src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java
[SYSTEMML-1793] Support matrix range indexing on GPU
- This commit supports matrix range indexing (i.e. right indexing) without
requiring sparse to dense conversion of inputs. Note: this PR only supports
dense output.
- Also, added RightIndexingTests in gpu package.
Closes #637.
Project:
http://git-wip-us.apache.org/repos/asf/systemml/blob/628ffad1/src/main/java/org/apache/sysml/hops/IndexingOp.java
--
diff --git a/src/main/java/org/apache/sysml/hops/IndexingOp.java
Repository: systemml
Updated Branches:
refs/heads/master 8fb74b158 -> 628ffad1b
http://git-wip-us.apache.org/repos/asf/systemml/blob/628ffad1/src/main/java/org/apache/sysml/utils/GPUStatistics.java
--
diff --git
http://git-wip-us.apache.org/repos/asf/systemml/blob/628ffad1/src/main/cpp/kernels/SystemML.ptx
--
diff --git a/src/main/cpp/kernels/SystemML.ptx
b/src/main/cpp/kernels/SystemML.ptx
index 7778317..f6ba15a 100644
---
Repository: systemml
Updated Branches:
refs/heads/master c145dbe3f -> f5de13e7b
[MINOR] Added scikit learn as dependency for systemml pip package
- Without this fix, running a simple mlcontext example after `pip
install systemml` will result in an error.
Project:
Repository: systemml
Updated Branches:
refs/heads/master 9a8421d4c -> bd1946a3d
[MINOR] Allow users to set explain level of mllearn algorithms
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/bd1946a3
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 3a7f38e58 -> 259814e6c
[MINOR] [SYSTEMML-540] Reset the _cachedParams to avoid incorrect sizes
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/259814e6
Tree:
Repository: systemml
Updated Branches:
refs/heads/master a9c14b02b -> 4bc1fea87
[MINOR] Fixed incorrect memory estimates in Caffe2DML summary for a network
with separate label and features data layer
- Also added a warn message when the user tries to run SystemML with less
than local
Repository: systemml
Updated Branches:
refs/heads/master 5df6ab6dd -> 62b5725d8
[SYSTEMML-1630] Allow user to specify custom path to BLAS libraries
- This feature is useful in the cloud environment where the user doesnot
have sudo permission or where setting environment variables such as
Repository: systemml
Updated Branches:
refs/heads/master 4ea1d5324 -> b9831f31a
[MINOR] Turn on the parallel flush of jvm stdout by default.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/b9831f31
Tree:
Repository: systemml
Updated Branches:
refs/heads/master b9831f31a -> 4416b5e51
[MINOR] [DOC] Updated the deep learning documentation
- Scale the input features for training lenet using Keras.
- Added table to highlight differences between three APIs.
Project:
Repository: systemml
Updated Branches:
refs/heads/gh-pages b2e1e18f1 -> 9873b3f7e
[MINOR] [DOC] Updated the deep learning documentation
- Scale the input features for training lenet using Keras.
- Added table to highlight differences between three APIs.
Project:
http://git-wip-us.apache.org/repos/asf/systemml/blob/3ca91e68/src/main/cpp/kernels/SystemML.ptx
--
diff --git a/src/main/cpp/kernels/SystemML.ptx
b/src/main/cpp/kernels/SystemML.ptx
index d382fc5..c990f27 100644
---
[SYSTEMML-540] Support sigmoid function on GPU
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/3ca91e68
Tree: http://git-wip-us.apache.org/repos/asf/systemml/tree/3ca91e68
Diff:
Repository: systemml
Updated Branches:
refs/heads/master 8c9af37d1 -> 696fe3f6d
[HOTFIX] Fix the loading logic of SystemML jars by associating the right
jar files with the matching classloader
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/gh-pages 9873b3f7e -> a82eaec8f
[MINOR] Doc svd builtin function in dml ref
Closes #704.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/a82eaec8
Tree:
Repository: systemml
Updated Branches:
refs/heads/master db378759e -> 3fa8d3793
[MINOR] Bugfix in mllearn's SVM class
- The PR #687 introduced a bug in mllearn SVM class as the scala
wrappers were not updated after adding the command-line parameter
scoring_only.
Project:
Repository: systemml
Updated Branches:
refs/heads/master 7b44380e4 -> 70c7952b9
[SYSTEMML-540] Add support for RNN and LSTM in Caffe2DML and Keras2DML
Closes #707.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/gh-pages 88e70806f -> 1b7f88abd
[MINOR] [DOC] Updated the native BLAS documentation
- Updated the documentation to suggest using a released version of
OpenBLAS.
- Also corrected the documentation to use 'sysml.native.blas' rather
than
[SYSTEMML-540] Remove caffe dependency from Keras2DML
- Also, added documentation to explain the three deep learning APIs.
Closes #702.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/1b7f88ab
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 2e0c72f0e -> 3b4656f5e
[MINOR] [DOC] Updated the native BLAS documentation
- Updated the documentation to suggest using a released version of
OpenBLAS.
- Also corrected the documentation to use 'sysml.native.blas' rather
than
Repository: systemml
Updated Branches:
refs/heads/master 3da574684 -> de69afdc8
[MINOR] Merge extra GPU and extra DNN flags into single
"sysml.stats.finegrained" flag to simplify usage
Closes #701.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master 903cb -> 8c11b5d82
[HOTFIX] Disable transposed matrix multiplication optimization as
cusparseDcsrmm2 failures.
- Also, added minor fixes for GPU tests.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master 118e3c0f6 -> 06d5bb073
[SYSTEMML-540] Avoid redundant computation of cudnnPoolingForward in
max_pool_backward
- If the max_pool is invoked in the forward pass, then its output can be
reused by the max_pool_backward rather than
Repository: systemml
Updated Branches:
refs/heads/master 8f4ecdce2 -> 881caa9ba
[SYSTEMML-540] Include the memory requirement of each layer in the summary
table of Caffe2DML
- This helps the user to estimate the batch size she should set for
optimal performance.
Project:
[SYSTEMML-1969] Support single-precision operations on GPU backend
- Since single-precision operations are faster on most GPU, we should allow our
users to perform the instructions on GPU in single precision.
- The GPU backend has been refactored to support arbitrary precision.
- This feature
http://git-wip-us.apache.org/repos/asf/systemml/blob/abbffc55/src/main/java/org/apache/sysml/api/DMLScript.java
--
diff --git a/src/main/java/org/apache/sysml/api/DMLScript.java
b/src/main/java/org/apache/sysml/api/DMLScript.java
Repository: systemml
Updated Branches:
refs/heads/master abbffc55e -> d3917effd
[SYSTEMML-446] Bugfix for GPU sparse right indexing with empty output
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/d3917eff
Tree:
Repository: systemml
Updated Branches:
refs/heads/master d3917effd -> f04067466
[MINOR] Enable single precision GPU tests
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/f0406746
Tree:
Repository: systemml
Updated Branches:
refs/heads/master f04067466 -> 591a0f775
[SYSTEMML-446] Minimize the chances of eviction for right indexing
operation
Indexing is only supported on GPU if:
1. the input is of type matrix AND
2. the input is less than 2GB.
The second condition is added
Repository: systemml
Updated Branches:
refs/heads/master 323dd72a8 -> 6de8f051d
[SYSTEMML-446] [SYSTEMML-702] Updated the sparse matrix multiplication to
minimize sparse-to-dense as well as dense-to-sparse conversion
1. The goal of this PR is not to improve performance (for example: by
Repository: systemml
Updated Branches:
refs/heads/master 2896f3316 -> d916ba5bd
[SYSTEMML-540] Added a rewrite to support a common tensor operation (sum over
channels)
- Added a rewrite to convert out = rowSums(matrix(colSums(A), rows=C, cols=HW))
to out = channel_sums(A) when nrow(A) > 1
Repository: systemml
Updated Branches:
refs/heads/master 1e5984cca -> e4220e3bc
[SYSTEMML-445] Added builtin functions for efficient computation of
lstm_backward function
- The current implementation treats lstm and lstm_backward as stateless
function for simplicity. We can revisit this
Repository: systemml
Updated Branches:
refs/heads/master a62b65c8f -> 72fd8fda3
[SYSTEMML-445] Upgraded CUDA and CuDNN version
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/72fd8fda
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 72fd8fda3 -> 7350a0c6d
[SYSTEMML-540] Remove unnecessary variables from batch_norm2d layer
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/7350a0c6
Tree:
Repository: systemml
Updated Branches:
refs/heads/master af9cc8a90 -> 4d3216678
http://git-wip-us.apache.org/repos/asf/systemml/blob/4d321667/src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java
--
diff --git
http://git-wip-us.apache.org/repos/asf/systemml/blob/4d321667/src/main/cpp/kernels/SystemML.ptx
--
diff --git a/src/main/cpp/kernels/SystemML.ptx
b/src/main/cpp/kernels/SystemML.ptx
index 5f72887..1865e18 100644
---
[SYSTEMML-445] Refactored GPU Memory Manager
- Several bugfixes found during recent experiments with ResNet200.
- Added align_memory eviction policy.
- Added GPU usage documentation.
- Refactored the GPU Memory Manager into distinct components.
Closes #774.
Project:
Repository: systemml
Updated Branches:
refs/heads/master c9739d788 -> 3c519e739
[SYSTEMML-445] Bugfix for compilation error on Mac OS and PowerPC arch.
Closes #775.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master 9861d7a3c -> 86f0e3f70
[SYSTEMML-445] Added CP implementation for batch_norm2d and
batch_norm2d_backward implementation.
- This feature is required for NN tests.
- The current version of batch_norm2d_backward only supports dense image
[SYSTEMML-445] Added builtin functions for efficient computation of batch
normalization and lstm layers.
- Following builtin functions are added: lstm, batch_norm2d and
batch_norm2d_backward.
- The DML language documentation and the NN layers are also updated.
- Since the builtin function for
Repository: systemml
Updated Branches:
refs/heads/master cba082eb1 -> 276065f93
http://git-wip-us.apache.org/repos/asf/systemml/blob/276065f9/src/main/java/org/apache/sysml/runtime/controlprogram/caching/CacheableData.java
--
[MINOR] Fix outdated language documentation (missing builtins), part 1
This patch updates the dml language documentation as follows:
* Fix removeEmpty mark down
* Extended boolean operators for scalars and matrices
* Extended cbind/rbind for nary inputs
* Add missing ifelse builtin function
*
[SYSTEMML-445] Added builtin functions for efficient computation of batch
normalization and lstm layers.
- Following builtin functions are added: lstm, batch_norm2d and
batch_norm2d_backward.
- The DML language documentation and the NN layers are also updated.
- Since the builtin function for
[SYSTEMML-445] Refactored GPU Memory Manager
- Several bugfixes found during recent experiments with ResNet200.
- Added align_memory eviction policy.
- Added GPU usage documentation.
- Refactored the GPU Memory Manager into distinct components.
Closes #774.
Project:
Repository: systemml
Updated Branches:
refs/heads/gh-pages 84c429c01 -> f5ae0596d
[MINOR] update deprecated brew commands for installing Java and Spark.
Closes #761.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
http://git-wip-us.apache.org/repos/asf/systemml/blob/8e3c6f8b/src/main/cpp/kernels/SystemML.ptx
--
diff --git a/src/main/cpp/kernels/SystemML.ptx
b/src/main/cpp/kernels/SystemML.ptx
index c990f27..5f72887 100644
---
[SYSTEMML-540] Add support for sparse filter dense image conv2d
- Also, supported skipping convolution operations for empty image and filters.
- Disabled sparse native convolution operations.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master e0006a272 -> 8e3c6f8b8
http://git-wip-us.apache.org/repos/asf/systemml/blob/8e3c6f8b/src/main/java/org/apache/sysml/runtime/instructions/gpu/GPUInstruction.java
--
diff
Repository: systemml
Updated Branches:
refs/heads/master 8e3c6f8b8 -> 269aef35f
[MINOR] Limit the size of intermediate im2col matrix to
MAX_WORKSPACE_LIMIT_BYTES
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master 269aef35f -> f702c03be
[SYSTEMML-540] Bugfix in GPU axpy instruction due to recent ternary changes
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/f702c03b
Repository: systemml
Updated Branches:
refs/heads/gh-pages 573427fb5 -> 936846406
[SYSTEMML-540] Added an assert builtin function
- Assert function halts the execution of DML program if the boolean argument
doesnot evaluate to TRUE.
- Like stop, assert is not supported inside a parfor.
-
Repository: systemml
Updated Branches:
refs/heads/master f702c03be -> 0529350a3
[SYSTEMML-540] Added an assert builtin function
- Assert function halts the execution of DML program if the boolean argument
doesnot evaluate to TRUE.
- Like stop, assert is not supported inside a parfor.
-
Repository: systemml
Updated Branches:
refs/heads/master 0529350a3 -> b9b273d87
[SYSTEMML-540] Added support for GPU relu, scalar min and max operations
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/b9b273d8
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 47973a905 -> 6af1df01d
[SYSTEMML-540] Low-rank fully connected layer
This layer has three advantages over the affine layer:
- It has significantly lower memory requirement than affine layer making it
ideal for devices such as GPUs.
-
Repository: systemml
Updated Branches:
refs/heads/gh-pages a46f1fb68 -> 573427fb5
[SYSTEMML-540] Add support for RNN and LSTM in Caffe2DML and Keras2DML
Closes #707.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
[SYSTEMML-540] Added optimizer support in Keras2DML
- Also, updated the documentation.
- Added a controlled error when batch size is not multiple of training
data points in lstm.
- Added perform_one_hot_encoding flag to deal with non-label data.
- Bug fix for EuclideanLoss layer in Caffe2DML.
-
Repository: systemml
Updated Branches:
refs/heads/master 45eec2d25 -> 9dc354ac2
[SYSTEMML-540] Added optimizer support in Keras2DML
- Also, updated the documentation.
- Added a controlled error when batch size is not multiple of training
data points in lstm.
- Added perform_one_hot_encoding
Repository: systemml
Updated Branches:
refs/heads/master 87d7fee73 -> 9b270d61a
[SYSTEMML-445] Added two-step strategy to deal with potential fragmentation on
GPU
Closes #669.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master 68b93c75f -> cf31ed2ab
[SYSTEMML-540] Improved the script generated by Caffe2DML in allreduce setting
- Allowed users to pass parfor parameters for generated allreduce scripts.
- Incorporated initial += support.
- Cleaned up functions
Repository: systemml
Updated Branches:
refs/heads/master cf31ed2ab -> 87d7fee73
[SYSTEMML-2004] Covariance Kernels
Closes #719.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/87d7fee7
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 0ba165cdd -> 9c3057a34
[SYSTEMML-445] Support newer as well as older Keras API
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/9c3057a3
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 5da8132ea -> ad5275932
[SYSTEMML-445] Disable ternary aggregate rewrite on GPU backend
- This issue will be revisited when we add tack+ and tak+ kernels.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master ad5275932 -> 416ebc02a
[SYSTEMML-445] Added load_keras_weights flag in Keras2DML to avoid transfering
randomly initialized weights
- By default, load_keras_weights is set to False. Hence, the weights will
be transferred to SystemML
Repository: systemml
Updated Branches:
refs/heads/master 9970fd814 -> f69047ea4
[SYSTEMML-445] Update algorithm selection logic on GPU for
conv2d_backward_data
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/f69047ea
Repository: systemml
Updated Branches:
refs/heads/master 9b270d61a -> 91f6fb572
[SYSTEMML-445] Display time taken for eviction as well as two-staged
fragmentation in the statistics.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master f69047ea4 -> 5da8132ea
[SYSTEMML-445] Guard JCudaKernels with sysml.gpu.sync.postProcess flag
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/5da8132e
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 92ee2cbf8 -> 8b054804e
[SYSTEMML-445] Bugfix in Caffe2DML/Keras2DML's concat layer for sentence CNN
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/8b054804
Repository: systemml
Updated Branches:
refs/heads/master 8f8743c81 -> 180c4f281
[SYSTEMML-445] Added support for sysml.gpu.memory.util.factor property
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/180c4f28
Tree:
Repository: systemml
Updated Branches:
refs/heads/master aed66df13 -> bdf27084b
[SYSTEMML-445] Bugfix and GPU shadow buffer for single precision
- Added shadow buffer logic for using CPU memory as temporary memory for
evicted matrices. This avoids unnecessary overhead of float to double, nnz
Repository: systemml
Updated Branches:
refs/heads/master b366c0f89 -> 13baec95c
[SYSTEMML-445] Added support for unified memory allocator for GPU backend
- The allocator can be configured using the property
'sysml.gpu.memory.allocator'.
- Minor bugfix for setting configuration properties
Repository: systemml
Updated Branches:
refs/heads/master f35cb6005 -> 7fb38b9b0
[SYSTEMML-445] Turn off batchnorm rewrite for mode="train"
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/7fb38b9b
Tree:
Repository: systemml
Updated Branches:
refs/heads/gh-pages 135cf394c -> ddce20266
[MINOR] Fixed failing GPU tests and updated the documentation.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/ddce2026
Tree:
Repository: systemml
Updated Branches:
refs/heads/master f296f8f51 -> 64110f31d
[MINOR] Fixed failing GPU tests and updated the documentation.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/64110f31
Tree:
Repository: systemml
Updated Branches:
refs/heads/master b7f569bd0 -> 5ca8706e9
[SYSTEMML-445] Added rshape operator for the GPU backend
- This leads to 1.2x speedup for ResNet200 with batch size of 32 by reducing
the number of host-to-device transfers.
- Also, added GPU tests for this
Repository: systemml
Updated Branches:
refs/heads/master 5ca8706e9 -> 04bc667f3
[SYSTEMML-445] Added SGD Nesterov update operator via rewrite for the GPU
backend
- This leads to 10-15% speedup for ResNet200 with batch size of 32.
- Also, added GPU tests for this operator.
Project:
Repository: systemml
Updated Branches:
refs/heads/master 11f0291d7 -> a11933002
[SYSTEMML-445] Refactored the shadow buffer and added documentation for newly
added features
- Refactored the shadow buffer logic from GPUObject to ShadowBuffer class for
maintenance.
- Added an additional timer
Repository: systemml
Updated Branches:
refs/heads/master 9e7ee19a4 -> be465dd65
Updates standalone jar dependencies with commons-lang3
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/be465dd6
Tree:
http://git-wip-us.apache.org/repos/asf/systemml/blob/ae268a9e/src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java
--
diff --git
a/src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixCUDA.java
[SYSTEMML-1325] Cleanup static variables in DMLScript
We use ThreadLocal DMLOptions and DMLConfig instead of static variables in
DMLScript class. It allows different JMLC instances (or MLCContext instances)
to run with different options (such as with GPU, with CPU, etc).
Closes #832.
[MINOR] Refactored the Python code to match PEP8 guidelines and bugfix in
Keras2DML for Python 3.5
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/9e7ee19a
Tree:
Repository: systemml
Updated Branches:
refs/heads/master ddc3672f1 -> 9e7ee19a4
http://git-wip-us.apache.org/repos/asf/systemml/blob/9e7ee19a/src/main/python/systemml/mllearn/keras2caffe.py
--
diff --git
http://git-wip-us.apache.org/repos/asf/systemml/blob/9e7ee19a/src/main/python/systemml/mlcontext.py
--
diff --git a/src/main/python/systemml/mlcontext.py
b/src/main/python/systemml/mlcontext.py
index c945b2b..9956815 100644
---
Repository: systemml
Updated Branches:
refs/heads/master 5aadb4b22 -> a6bca8851
[SYSTEMML-445] Added a rewrite for batch normalization train
- This PR fuses a batch normalization train pattern into a FunctionOp. The
method batchNormTrain in RewriteGPUSpecificOps performs the fusing.
- This
Repository: systemml
Updated Branches:
refs/heads/master 81419ae6a -> 0f36780a8
http://git-wip-us.apache.org/repos/asf/systemml/blob/0f36780a/src/main/java/org/apache/sysml/runtime/instructions/gpu/DnnGPUInstruction.java
--
[SYSTEMML-445] Removed batch_norm builtin functions
- Removed batch_norm builtin functions to exploit codegen in CP.
- Added rewrites for compiling efficient CuDNN operators.
- Added rewrites for SGD update operations.
- To simplify adding new GPU rewrites, added HopDagPatternMatcher that allows
Repository: systemml
Updated Branches:
refs/heads/gh-pages 14049d257 -> d38bf4ee9
[SYSTEMML-445] Removed batch_norm builtin functions
- Removed batch_norm builtin functions to exploit codegen in CP.
- Added rewrites for compiling efficient CuDNN operators.
- Added rewrites for SGD update
Repository: systemml
Updated Branches:
refs/heads/master 0f36780a8 -> ab251f6ee
[MINOR] Fixed javadoc errors
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/ab251f6e
Tree:
Repository: systemml
Updated Branches:
refs/heads/master ab251f6ee -> 2fc26b3dc
[MINOR] Allow non-literal values in parameterized built-in functions
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit: http://git-wip-us.apache.org/repos/asf/systemml/commit/2fc26b3d
Tree:
Repository: systemml
Updated Branches:
refs/heads/master 86b3090ba -> b56612f02
[SYSTEMML-445] Fixed a memory leak in GPU lstm builtin function and also
added developer utility to debug such bugs in the future.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master b56612f02 -> 3bad7e7a3
[MINOR] Profile memory use in JMLC execution
This PR adds utilities to profile memory use during execution in JMLC.
Specifically, the following changes were made:
1. Added options setStatistics() and
Repository: systemml
Updated Branches:
refs/heads/master 997eb2aa2 -> 47973a905
[SYSTEMML-540] Bugfix for fused ReLU-maxpooling and ReLU-maxpooling
backward operators
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master 08aae51d5 -> 997eb2aa2
[SYSTEMML-1992] Implemenation of Mode finding for Gaussian Process
Closes #711.
Project: http://git-wip-us.apache.org/repos/asf/systemml/repo
Commit:
Repository: systemml
Updated Branches:
refs/heads/master d16cc7cf8 -> 54a11eed3
http://git-wip-us.apache.org/repos/asf/systemml/blob/54a11eed/src/main/java/org/apache/sysml/runtime/matrix/data/LibMatrixDNNPooling.java
--
diff
[SYSTEMML-1872] Added average pooling and upsampling layers
- Added avg_pool and avg_pool_backward builtin functions.
- The above builtin functions are supported on both CPU and GPU.
- Also, added compare_backend tests for the above function to compare the
results of CP operator with that of
Repository: systemml
Updated Branches:
refs/heads/gh-pages 36e9f1f21 -> 781fa43dc
[SYSTEMML-1872] Added average pooling and upsampling layers
- Added avg_pool and avg_pool_backward builtin functions.
- The above builtin functions are supported on both CPU and GPU.
- Also, added
Repository: systemml
Updated Branches:
refs/heads/master 54a11eed3 -> 8ffa3d158
[SYSTEMML-445] Allow users to pass the file paths to the binary blocked, csv
and ijv datasets to mllearn classes.
- This allows the advance users who already have data materialized in
binary blocked formats to
Repository: systemml
Updated Branches:
refs/heads/master 1fa8e126f -> bdf42c068
[SYSTEMML-445] Cleanup GPU memory management
- Simplified GPU memory management (allocate/deallocate/evict) to debug any
OOM-related issues.
- Also, streamlined fragmentation related code into malloc and prepared
201 - 300 of 395 matches
Mail list logo